diff options
214 files changed, 12077 insertions, 1525 deletions
@@ -12,6 +12,9 @@ # See the License for the specific language governing permissions and # limitations under the License. +# Build with C++17. +build --cxxopt=-std=c++17 + # Display the current git revision in the info block. build --stamp --workspace_status_command tools/workspace_status.sh @@ -23,7 +23,18 @@ go_path( "//runsc", # Packages that are not dependencies of //runsc. + "//pkg/sentry/kernel/memevent", + "//pkg/tcpip/adapters/gonet", "//pkg/tcpip/link/channel", + "//pkg/tcpip/link/muxed", + "//pkg/tcpip/link/sharedmem", + "//pkg/tcpip/link/sharedmem/pipe", + "//pkg/tcpip/link/sharedmem/queue", + "//pkg/tcpip/link/tun", + "//pkg/tcpip/link/waitable", + "//pkg/tcpip/sample/tun_tcp_connect", + "//pkg/tcpip/sample/tun_tcp_echo", + "//pkg/tcpip/transport/tcpconntrack", ], ) @@ -48,9 +48,10 @@ Make sure the following dependencies are installed: * Linux 4.14.77+ ([older linux][old-linux]) * [git][git] -* [Bazel][bazel] 0.28.0+ +* [Bazel][bazel] 1.2+ * [Python][python] * [Docker version 17.09.0 or greater][docker] +* C++ toolchain supporting C++17 (GCC 7+, Clang 5+) * Gold linker (e.g. `binutils-gold` package on Ubuntu) ### Building @@ -106,6 +106,45 @@ load("@rules_pkg//:deps.bzl", "rules_pkg_dependencies") rules_pkg_dependencies() +# Container rules. +http_archive( + name = "io_bazel_rules_docker", + sha256 = "14ac30773fdb393ddec90e158c9ec7ebb3f8a4fd533ec2abbfd8789ad81a284b", + strip_prefix = "rules_docker-0.12.1", + urls = ["https://github.com/bazelbuild/rules_docker/releases/download/v0.12.1/rules_docker-v0.12.1.tar.gz"], +) + +load( + "@io_bazel_rules_docker//repositories:repositories.bzl", + container_repositories = "repositories", +) + +container_repositories() + +load("@io_bazel_rules_docker//repositories:deps.bzl", container_deps = "deps") + +container_deps() + +load( + "@io_bazel_rules_docker//container:container.bzl", + "container_pull", +) + +# This container is built from the Dockerfile in test/iptables/runner. +container_pull( + name = "iptables-test", + registry = "gcr.io", + repository = "gvisor-presubmit/iptables-test", + digest = "sha256:a137d692a2eb9fc7bf95c5f4a568da090e2c31098e93634421ed88f3a3f1db65", +) + +load( + "@io_bazel_rules_docker//go:image.bzl", + _go_image_repos = "repositories", +) + +_go_image_repos() + # External repositories, in sorted order. go_repository( name = "com_github_cenkalti_backoff", diff --git a/benchmarks/harness/machine.py b/benchmarks/harness/machine.py index 2166d040a..66b719b63 100644 --- a/benchmarks/harness/machine.py +++ b/benchmarks/harness/machine.py @@ -11,7 +11,22 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -"""Machine abstraction. This is the primary API for benchmarks.""" +"""Machine abstraction passed to benchmarks to run docker containers. + +Abstraction for interacting with test machines. Machines are produced +by Machine producers and represent a local or remote machine. Benchmark +methods in /benchmarks/suite are passed the required number of machines in order +to run the benchmark. Machines contain methods to run commands via bash, +possibly over ssh. Machines also hold a connection to the docker UNIX socket +to run contianers. + + Typical usage example: + + machine = Machine() + machine.run(cmd) + machine.pull(path) + container = machine.container() +""" import logging import re @@ -28,12 +43,16 @@ from benchmarks.harness import ssh_connection from benchmarks.harness import tunnel_dispatcher -class Machine: +class Machine(object): """The machine object is the primary object for benchmarks. Machine objects are passed to each metric function call and benchmarks use machines to access real connections to those machines. + + Attributes: + _name: Name as a string """ + _name = "" def run(self, cmd: str) -> Tuple[str, str]: """Convenience method for running a bash command on a machine object. @@ -90,11 +109,15 @@ class Machine: def sleep(self, amount: float): """Sleeps the given amount of time.""" - raise NotImplementedError + time.sleep(amount) + + def __str__(self): + return self._name class MockMachine(Machine): """A mocked machine.""" + _name = "mock" def run(self, cmd: str) -> Tuple[str, str]: return "", "" @@ -119,15 +142,18 @@ def get_address(machine: Machine) -> str: class LocalMachine(Machine): - """The local machine.""" + """The local machine. + + Attributes: + _name: Name as a string + _docker_client: a pythonic connection to to the local dockerd unix socket. + See: https://github.com/docker/docker-py + """ def __init__(self, name): self._name = name self._docker_client = docker.from_env() - def __str__(self): - return self._name - def run(self, cmd: str) -> Tuple[str, str]: process = subprocess.Popen( cmd.split(" "), stdout=subprocess.PIPE, stderr=subprocess.PIPE) @@ -155,7 +181,17 @@ class LocalMachine(Machine): class RemoteMachine(Machine): - """Remote machine accessible via an SSH connection.""" + """Remote machine accessible via an SSH connection. + + Attributes: + _name: Name as a string + _ssh_connection: a paramiko backed ssh connection which can be used to run + commands on this machine + _tunnel: a python wrapper around a port forwarded ssh connection between a + local unix socket and the remote machine's dockerd unix socket. + _docker_client: a pythonic wrapper backed by the _tunnel. Allows sending + docker commands: see https://github.com/docker/docker-py + """ def __init__(self, name, **kwargs): self._name = name @@ -164,9 +200,6 @@ class RemoteMachine(Machine): self._tunnel.connect() self._docker_client = self._tunnel.get_docker_client() - def __str__(self): - return self._name - def run(self, cmd: str) -> Tuple[str, str]: return self._ssh_connection.run(cmd) diff --git a/benchmarks/harness/machine_producers/BUILD b/benchmarks/harness/machine_producers/BUILD index 5b2228e01..a48da02a1 100644 --- a/benchmarks/harness/machine_producers/BUILD +++ b/benchmarks/harness/machine_producers/BUILD @@ -33,3 +33,8 @@ py_library( requirement("PyYAML", False), ], ) + +py_library( + name = "gcloud_mock_recorder", + srcs = ["gcloud_mock_recorder.py"], +) diff --git a/benchmarks/harness/machine_producers/gcloud_mock_recorder.py b/benchmarks/harness/machine_producers/gcloud_mock_recorder.py new file mode 100644 index 000000000..fd9837a37 --- /dev/null +++ b/benchmarks/harness/machine_producers/gcloud_mock_recorder.py @@ -0,0 +1,97 @@ +# python3 +# Copyright 2019 Google LLC +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +"""A recorder and replay for testing the GCloudProducer. + +MockPrinter and MockReader handle printing and reading mock data for the +purposes of testing. MockPrinter is passed to GCloudProducer objects. The user +can then run scenarios and record them for playback in tests later. + +MockReader is passed to MockGcloudProducer objects and handles reading the +previously recorded mock data. + +It is left to the user to check if data printed is properly redacted for their +own use. The intended usecase for this class is data coming from gcloud +commands, which will contain public IPs and other instance data. + +The data format is json and printed/read from the ./test_data directory. The +data is the output of subprocess.CompletedProcess objects in json format. + + Typical usage example: + + recorder = MockPrinter() + producer = GCloudProducer(args, recorder) + machines = producer.get_machines(1) + with open("my_file.json") as fd: + recorder.write_out(fd) + + reader = MockReader(filename) + producer = MockGcloudProducer(args, mock) + machines = producer.get_machines(1) + assert len(machines) == 1 +""" + +import io +import json +import subprocess + + +class MockPrinter(object): + """Handles printing Mock data for MockGcloudProducer. + + Attributes: + _records: list of json object records for printing + """ + + def __init__(self): + self._records = [] + + def record(self, entry: subprocess.CompletedProcess): + """Records data and strips out ip addresses.""" + + record = { + "args": entry.args, + "stdout": entry.stdout.decode("utf-8"), + "returncode": str(entry.returncode) + } + self._records.append(record) + + def write_out(self, fd: io.FileIO): + """Prints out the data into the given filepath.""" + fd.write(json.dumps(self._records, indent=4)) + + +class MockReader(object): + """Handles reading Mock data for MockGcloudProducer. + + Attributes: + _records: List[json] records read from the passed in file. + """ + + def __init__(self, filepath: str): + with open(filepath, "rb") as file: + self._records = json.loads(file.read()) + self._i = 0 + + def __iter__(self): + return self + + def __next__(self, args) -> subprocess.CompletedProcess: + """Returns the next record as a CompletedProcess.""" + if self._i < len(self._records): + record = self._records[self._i] + stdout = record["stdout"].encode("ascii") + returncode = int(record["returncode"]) + return subprocess.CompletedProcess( + args=args, returncode=returncode, stdout=stdout, stderr=b"") + raise StopIteration() diff --git a/benchmarks/harness/tunnel_dispatcher.py b/benchmarks/harness/tunnel_dispatcher.py index 8dfe2862a..c56fd022a 100644 --- a/benchmarks/harness/tunnel_dispatcher.py +++ b/benchmarks/harness/tunnel_dispatcher.py @@ -11,7 +11,25 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -"""Tunnel handles setting up connections to remote machines.""" +"""Tunnel handles setting up connections to remote machines. + +Tunnel dispatcher is a wrapper around the connection from a local UNIX socket +and a remote UNIX socket via SSH with port forwarding. This is done to +initialize the pythonic dockerpy client to run containers on the remote host by +connecting to /var/run/docker.sock (where Docker is listening). Tunnel +dispatcher sets up the local UNIX socket and calls the `ssh` command as a +subprocess, and holds a reference to that subprocess. It manages clean-up on +exit as best it can by killing the ssh subprocess and deleting the local UNIX +socket,stored in /tmp for easy cleanup in most systems if this fails. + + Typical usage example: + + t = Tunnel(name, **kwargs) + t.connect() + client = t.get_docker_client() # + client.containers.run("ubuntu", "echo hello world") + +""" import os import tempfile @@ -21,31 +39,53 @@ import docker import pexpect SSH_TUNNEL_COMMAND = """ssh - -o GlobalKnownHostsFile=/dev/null - -o UserKnownHostsFile=/dev/null - -o StrictHostKeyChecking=no - -nNT -L {filename}:/var/run/docker.sock - -i {key_path} - {username}@{hostname}""" + -o GlobalKnownHostsFile=/dev/null + -o UserKnownHostsFile=/dev/null + -o StrictHostKeyChecking=no + -o IdentitiesOnly=yes + -nNT -L {filename}:/var/run/docker.sock + -i {key_path} + {username}@{hostname}""" -class Tunnel: +class Tunnel(object): """The tunnel object represents the tunnel via ssh. This connects a local unix domain socket with a remote socket. + + Attributes: + _filename: a temporary name of the UNIX socket prefixed by the name + argument. + _hostname: the IP or resolvable hostname of the remote host. + _username: the username of the ssh_key used to run ssh. + _key_path: path to a valid key. + _key_password: optional password to the ssh key in _key_path + _process: holds reference to the ssh subprocess created. + + Returns: + The new minimum port. + + Raises: + ConnectionError: If no available port is found. """ - def __init__(self, name, hostname: str, username: str, key_path: str, + def __init__(self, + name: str, + hostname: str, + username: str, + key_path: str, + key_password: str = "", **kwargs): self._filename = tempfile.NamedTemporaryFile(prefix=name).name self._hostname = hostname self._username = username self._key_path = key_path + self._key_password = key_password self._kwargs = kwargs self._process = None def connect(self): - """Connects the SSH tunnel.""" + """Connects the SSH tunnel and stores the subprocess reference in _process.""" cmd = SSH_TUNNEL_COMMAND.format( filename=self._filename, key_path=self._key_path, @@ -54,9 +94,9 @@ class Tunnel: self._process = pexpect.spawn(cmd, timeout=10) # If given a password, assume we'll be asked for it. - if "key_password" in self._kwargs: + if self._key_password: self._process.expect(["Enter passphrase for key .*: "]) - self._process.sendline(self._kwargs["key_password"]) + self._process.sendline(self._key_password) while True: # Wait for the tunnel to appear. @@ -71,7 +111,7 @@ class Tunnel: return self._filename def get_docker_client(self): - """Returns a docker client for this Tunne0l.""" + """Returns a docker client for this Tunnel.""" return docker.DockerClient(base_url="unix:/" + self._filename) def __del__(self): diff --git a/benchmarks/runner/BUILD b/benchmarks/runner/BUILD index a3941da42..de24824cc 100644 --- a/benchmarks/runner/BUILD +++ b/benchmarks/runner/BUILD @@ -34,7 +34,10 @@ py_test( name = "runner_test", srcs = ["runner_test.py"], python_version = "PY3", - tags = ["local"], + tags = [ + "local", + "manual", + ], deps = [ ":runner", requirement("click", True), diff --git a/benchmarks/workloads/node_template/package-lock.json b/benchmarks/workloads/node_template/package-lock.json index 1653597a1..580e68aa5 100644 --- a/benchmarks/workloads/node_template/package-lock.json +++ b/benchmarks/workloads/node_template/package-lock.json @@ -233,9 +233,9 @@ "integrity": "sha1-6qM9bd16zo9/b+DJygRA5wZzix4=" }, "lodash": { - "version": "4.17.11", - "resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.11.tgz", - "integrity": "sha512-cQKh8igo5QUhZ7lg38DYWAxMvjSAKG0A8wGSVimP07SIUEK2UO+arSRKbRZWtelMtN5V0Hkwh5ryOto/SshYIg==" + "version": "4.17.15", + "resolved": "https://registry.npmjs.org/lodash/-/lodash-4.17.15.tgz", + "integrity": "sha512-8xOcRHvCjnocdS5cpwXQXVzmmh5e5+saE2QGoeQmbKmRS6J3VQppPOIt0MnmE+4xlZoumy0GPG0D0MVIQbNA1A==" }, "media-typer": { "version": "0.3.0", @@ -364,6 +364,16 @@ } } }, + "redis-commands": { + "version": "1.5.0", + "resolved": "https://registry.npmjs.org/redis-commands/-/redis-commands-1.5.0.tgz", + "integrity": "sha512-6KxamqpZ468MeQC3bkWmCB1fp56XL64D4Kf0zJSwDZbVLLm7KFkoIcHrgRvQ+sk8dnhySs7+yBg94yIkAK7aJg==" + }, + "redis-parser": { + "version": "2.6.0", + "resolved": "https://registry.npmjs.org/redis-parser/-/redis-parser-2.6.0.tgz", + "integrity": "sha1-Uu0J2srBCPGmMcB+m2mUHnoZUEs=" + }, "safe-buffer": { "version": "5.1.2", "resolved": "https://registry.npmjs.org/safe-buffer/-/safe-buffer-5.1.2.tgz", @@ -1,21 +1,21 @@ module gvisor.dev/gvisor -go 1.12 +go 1.13 require ( - github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 - github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 - github.com/golang/mock v1.3.1 - github.com/golang/protobuf v1.3.1 - github.com/google/btree v1.0.0 - github.com/google/go-cmp v0.2.0 - github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 - github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d - github.com/kr/pty v1.1.1 - github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 - github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 - github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e - github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 - golang.org/x/net v0.0.0-20190311183353-d8887717615a - golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a + github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 + github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 + github.com/golang/mock v1.3.1 + github.com/golang/protobuf v1.3.1 + github.com/google/btree v1.0.0 + github.com/google/go-cmp v0.2.0 + github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 + github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d + github.com/kr/pty v1.1.1 + github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 + github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 + github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e + github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 + golang.org/x/net v0.0.0-20190311183353-d8887717615a + golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a ) diff --git a/kokoro/iptables_tests.cfg b/kokoro/iptables_tests.cfg new file mode 100644 index 000000000..7af20629a --- /dev/null +++ b/kokoro/iptables_tests.cfg @@ -0,0 +1,10 @@ +build_file: "repo/scripts/iptables_test.sh" + +action { + define_artifacts { + regex: "**/sponge_log.xml" + regex: "**/sponge_log.log" + regex: "**/outputs.zip" + regex: "**/runsc_logs_*.tar.gz" + } +} diff --git a/kokoro/kythe/generate_xrefs.cfg b/kokoro/kythe/generate_xrefs.cfg new file mode 100644 index 000000000..03e65c54e --- /dev/null +++ b/kokoro/kythe/generate_xrefs.cfg @@ -0,0 +1,28 @@ +build_file: "gvisor/kokoro/kythe/generate_xrefs.sh" + +before_action { + fetch_keystore { + keystore_resource { + keystore_config_id: 73898 + keyname: "kokoro-rbe-service-account" + } + } +} + +bazel_setting { + project_id: "gvisor-rbe" + local_execution: false + auth_credential: { + keystore_config_id: 73898 + keyname: "kokoro-rbe-service-account" + } + bes_backend_address: "buildeventservice.googleapis.com" + foundry_backend_address: "remotebuildexecution.googleapis.com" + upsalite_frontend_address: "https://source.cloud.google.com" +} + +action { + define_artifacts { + regex: "*.kzip" + } +} diff --git a/kokoro/kythe/generate_xrefs.sh b/kokoro/kythe/generate_xrefs.sh new file mode 100644 index 000000000..799467a34 --- /dev/null +++ b/kokoro/kythe/generate_xrefs.sh @@ -0,0 +1,54 @@ +#!/bin/bash + +# Copyright 2019 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +set -ex + +# Install the latest version of Bazel. The default on Kokoro images is out of +# date. +if command -v use_bazel.sh >/dev/null; then + use_bazel.sh latest +fi +bazel version + +python3 -V + +readonly KYTHE_VERSION='v0.0.37' +readonly WORKDIR="$(mktemp -d)" +readonly KYTHE_DIR="${WORKDIR}/kythe-${KYTHE_VERSION}" +if [[ -n "$KOKORO_GIT_COMMIT" ]]; then + readonly KZIP_FILENAME="${KOKORO_ARTIFACTS_DIR}/${KOKORO_GIT_COMMIT}.kzip" +else + readonly KZIP_FILENAME="$(git rev-parse HEAD).kzip" +fi + +wget -q -O "${WORKDIR}/kythe.tar.gz" \ + "https://github.com/kythe/kythe/releases/download/${KYTHE_VERSION}/kythe-${KYTHE_VERSION}.tar.gz" +tar --no-same-owner -xzf "${WORKDIR}/kythe.tar.gz" --directory "$WORKDIR" + +if [[ -n "$KOKORO_ARTIFACTS_DIR" ]]; then + cd "${KOKORO_ARTIFACTS_DIR}/github/gvisor" +fi +bazel \ + --bazelrc="${KYTHE_DIR}/extractors.bazelrc" \ + build \ + --override_repository kythe_release="${KYTHE_DIR}" \ + --define=kythe_corpus=gvisor.dev \ + --cxxopt=-std=c++17 \ + //... + +"${KYTHE_DIR}/tools/kzip" merge \ + --output "$KZIP_FILENAME" \ + $(find -L bazel-out/*/extra_actions/ -name '*.kzip') diff --git a/kokoro/ubuntu1604/10_core.sh b/kokoro/ubuntu1604/10_core.sh index e87a6eee8..46dda6bb1 100755 --- a/kokoro/ubuntu1604/10_core.sh +++ b/kokoro/ubuntu1604/10_core.sh @@ -21,8 +21,8 @@ apt-get update && apt-get -y install make git-core build-essential linux-headers # Install a recent go toolchain. if ! [[ -d /usr/local/go ]]; then - wget https://dl.google.com/go/go1.12.linux-amd64.tar.gz - tar -xvf go1.12.linux-amd64.tar.gz + wget https://dl.google.com/go/go1.13.5.linux-amd64.tar.gz + tar -xvf go1.13.5.linux-amd64.tar.gz mv go /usr/local fi diff --git a/kokoro/ubuntu1604/README.md b/kokoro/ubuntu1604/README.md new file mode 100644 index 000000000..64f913b9a --- /dev/null +++ b/kokoro/ubuntu1604/README.md @@ -0,0 +1,34 @@ +## Image Update + +After making changes to files in the directory, you must run the following +commands to update the image Kokoro uses: + +```shell +gcloud config set project gvisor-kokoro-testing +third_party/gvisor/kokoro/ubuntu1604/build.sh +third_party/gvisor/kokoro/ubuntu1804/build.sh +``` + +Note: the command above will change your default project for `gcloud`. Run +`gcloud config set project` again to revert back to your default project. + +Note: Files in `third_party/gvisor/kokoro/ubuntu1804/` as symlinks to +`ubuntu1604`, therefore both images must be updated. + +After the script finishes, the last few lines of the output will container the +image name. If the output was lost, you can run `build.sh` again to print the +image name. + +``` +NAME PROJECT FAMILY DEPRECATED STATUS +image-6777fa4666a968c8 gvisor-kokoro-testing READY ++ cleanup ++ gcloud compute instances delete --quiet build-tlfrdv +Deleted [https://www.googleapis.com/compute/v1/projects/gvisor-kokoro-testing/zones/us-central1-f/instances/build-tlfrdv]. +``` + +To setup Kokoro to use the new image, copy the image names to their +corresponding file below: + +* //devtools/kokoro/config/gcp/gvisor/ubuntu1604.gcl +* //devtools/kokoro/config/gcp/gvisor/ubuntu1804.gcl diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 51774c6b6..9553f164d 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -57,6 +57,7 @@ go_library( "uio.go", "utsname.go", "wait.go", + "xattr.go", ], importpath = "gvisor.dev/gvisor/pkg/abi/linux", visibility = ["//visibility:public"], diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go index f78315ebf..6663a199c 100644 --- a/pkg/abi/linux/fcntl.go +++ b/pkg/abi/linux/fcntl.go @@ -16,15 +16,17 @@ package linux // Commands from linux/fcntl.h. const ( - F_DUPFD = 0x0 - F_GETFD = 0x1 - F_SETFD = 0x2 - F_GETFL = 0x3 - F_SETFL = 0x4 - F_SETLK = 0x6 - F_SETLKW = 0x7 - F_SETOWN = 0x8 - F_GETOWN = 0x9 + F_DUPFD = 0 + F_GETFD = 1 + F_SETFD = 2 + F_GETFL = 3 + F_SETFL = 4 + F_SETLK = 6 + F_SETLKW = 7 + F_SETOWN = 8 + F_GETOWN = 9 + F_SETOWN_EX = 15 + F_GETOWN_EX = 16 F_DUPFD_CLOEXEC = 1024 + 6 F_SETPIPE_SZ = 1024 + 7 F_GETPIPE_SZ = 1024 + 8 @@ -32,9 +34,9 @@ const ( // Commands for F_SETLK. const ( - F_RDLCK = 0x0 - F_WRLCK = 0x1 - F_UNLCK = 0x2 + F_RDLCK = 0 + F_WRLCK = 1 + F_UNLCK = 2 ) // Flags for fcntl. @@ -42,7 +44,7 @@ const ( FD_CLOEXEC = 00000001 ) -// Lock structure for F_SETLK. +// Flock is the lock structure for F_SETLK. type Flock struct { Type int16 Whence int16 @@ -52,3 +54,16 @@ type Flock struct { Pid int32 _ [4]byte } + +// Flags for F_SETOWN_EX and F_GETOWN_EX. +const ( + F_OWNER_TID = 0 + F_OWNER_PID = 1 + F_OWNER_PGRP = 2 +) + +// FOwnerEx is the owner structure for F_SETOWN_EX and F_GETOWN_EX. +type FOwnerEx struct { + Type int32 + PID int32 +} diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index c9ee098f4..16791d03e 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -144,9 +144,13 @@ const ( ModeCharacterDevice = S_IFCHR ModeNamedPipe = S_IFIFO - ModeSetUID = 04000 - ModeSetGID = 02000 - ModeSticky = 01000 + S_ISUID = 04000 + S_ISGID = 02000 + S_ISVTX = 01000 + + ModeSetUID = S_ISUID + ModeSetGID = S_ISGID + ModeSticky = S_ISVTX ModeUserAll = 0700 ModeUserRead = 0400 @@ -282,6 +286,29 @@ func (m FileMode) String() string { return strings.Join(s, "|") } +// DirentType maps file types to dirent types appropriate for (struct +// dirent)::d_type. +func (m FileMode) DirentType() uint8 { + switch m.FileType() { + case ModeSocket: + return DT_SOCK + case ModeSymlink: + return DT_LNK + case ModeRegular: + return DT_REG + case ModeBlockDevice: + return DT_BLK + case ModeDirectory: + return DT_DIR + case ModeCharacterDevice: + return DT_CHR + case ModeNamedPipe: + return DT_FIFO + default: + return DT_UNKNOWN + } +} + var modeExtraBits = abi.FlagSet{ { Flag: ModeSetUID, diff --git a/pkg/abi/linux/fs.go b/pkg/abi/linux/fs.go index b416e3472..2c652baa2 100644 --- a/pkg/abi/linux/fs.go +++ b/pkg/abi/linux/fs.go @@ -92,3 +92,10 @@ const ( SYNC_FILE_RANGE_WRITE = 2 SYNC_FILE_RANGE_WAIT_AFTER = 4 ) + +// Flag argument to renameat2(2), from include/uapi/linux/fs.h. +const ( + RENAME_NOREPLACE = (1 << 0) // Don't overwrite target. + RENAME_EXCHANGE = (1 << 1) // Exchange src and dst. + RENAME_WHITEOUT = (1 << 2) // Whiteout src. +) diff --git a/pkg/abi/linux/xattr.go b/pkg/abi/linux/xattr.go new file mode 100644 index 000000000..a3b6406fa --- /dev/null +++ b/pkg/abi/linux/xattr.go @@ -0,0 +1,27 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// Constants for extended attributes. +const ( + XATTR_NAME_MAX = 255 + XATTR_SIZE_MAX = 65536 + + XATTR_CREATE = 1 + XATTR_REPLACE = 2 + + XATTR_USER_PREFIX = "user." + XATTR_USER_PREFIX_LEN = len(XATTR_USER_PREFIX) +) diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index 51869c7d6..b9582c07f 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -257,7 +257,6 @@ func CanOpen(mode FileMode) bool { // handle implements handler.handle. func (t *Tlopen) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -294,7 +293,6 @@ func (t *Tlopen) handle(cs *connState) message { return syscall.EINVAL } - // Do the open. osFile, qid, ioUnit, err = ref.file.Open(t.Flags) return err }); err != nil { @@ -311,12 +309,10 @@ func (t *Tlopen) handle(cs *connState) message { } func (t *Tlcreate) do(cs *connState, uid UID) (*Rlcreate, error) { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return nil, err } - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return nil, syscall.EBADF @@ -390,12 +386,10 @@ func (t *Tsymlink) handle(cs *connState) message { } func (t *Tsymlink) do(cs *connState, uid UID) (*Rsymlink, error) { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return nil, err } - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return nil, syscall.EBADF @@ -426,19 +420,16 @@ func (t *Tsymlink) do(cs *connState, uid UID) (*Rsymlink, error) { // handle implements handler.handle. func (t *Tlink) handle(cs *connState) message { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return newErr(err) } - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return newErr(syscall.EBADF) } defer ref.DecRef() - // Lookup the other FID. refTarget, ok := cs.LookupFID(t.Target) if !ok { return newErr(syscall.EBADF) @@ -467,7 +458,6 @@ func (t *Tlink) handle(cs *connState) message { // handle implements handler.handle. func (t *Trenameat) handle(cs *connState) message { - // Don't allow complex names. if err := checkSafeName(t.OldName); err != nil { return newErr(err) } @@ -475,14 +465,12 @@ func (t *Trenameat) handle(cs *connState) message { return newErr(err) } - // Lookup the FID. ref, ok := cs.LookupFID(t.OldDirectory) if !ok { return newErr(syscall.EBADF) } defer ref.DecRef() - // Lookup the other FID. refTarget, ok := cs.LookupFID(t.NewDirectory) if !ok { return newErr(syscall.EBADF) @@ -523,12 +511,10 @@ func (t *Trenameat) handle(cs *connState) message { // handle implements handler.handle. func (t *Tunlinkat) handle(cs *connState) message { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return newErr(err) } - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return newErr(syscall.EBADF) @@ -577,19 +563,16 @@ func (t *Tunlinkat) handle(cs *connState) message { // handle implements handler.handle. func (t *Trename) handle(cs *connState) message { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return newErr(err) } - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) } defer ref.DecRef() - // Lookup the target. refTarget, ok := cs.LookupFID(t.Directory) if !ok { return newErr(syscall.EBADF) @@ -641,7 +624,6 @@ func (t *Trename) handle(cs *connState) message { // handle implements handler.handle. func (t *Treadlink) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -669,7 +651,6 @@ func (t *Treadlink) handle(cs *connState) message { // handle implements handler.handle. func (t *Tread) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -708,7 +689,6 @@ func (t *Tread) handle(cs *connState) message { // handle implements handler.handle. func (t *Twrite) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -747,12 +727,10 @@ func (t *Tmknod) handle(cs *connState) message { } func (t *Tmknod) do(cs *connState, uid UID) (*Rmknod, error) { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return nil, err } - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return nil, syscall.EBADF @@ -791,12 +769,10 @@ func (t *Tmkdir) handle(cs *connState) message { } func (t *Tmkdir) do(cs *connState, uid UID) (*Rmkdir, error) { - // Don't allow complex names. if err := checkSafeName(t.Name); err != nil { return nil, err } - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return nil, syscall.EBADF @@ -827,7 +803,6 @@ func (t *Tmkdir) do(cs *connState, uid UID) (*Rmkdir, error) { // handle implements handler.handle. func (t *Tgetattr) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -856,7 +831,6 @@ func (t *Tgetattr) handle(cs *connState) message { // handle implements handler.handle. func (t *Tsetattr) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -883,7 +857,6 @@ func (t *Tsetattr) handle(cs *connState) message { // handle implements handler.handle. func (t *Tallocate) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -917,7 +890,6 @@ func (t *Tallocate) handle(cs *connState) message { // handle implements handler.handle. func (t *Txattrwalk) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -930,7 +902,6 @@ func (t *Txattrwalk) handle(cs *connState) message { // handle implements handler.handle. func (t *Txattrcreate) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -943,7 +914,6 @@ func (t *Txattrcreate) handle(cs *connState) message { // handle implements handler.handle. func (t *Treaddir) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.Directory) if !ok { return newErr(syscall.EBADF) @@ -977,7 +947,6 @@ func (t *Treaddir) handle(cs *connState) message { // handle implements handler.handle. func (t *Tfsync) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -1001,7 +970,6 @@ func (t *Tfsync) handle(cs *connState) message { // handle implements handler.handle. func (t *Tstatfs) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -1192,7 +1160,6 @@ func doWalk(cs *connState, ref *fidRef, names []string, getattr bool) (qids []QI // handle implements handler.handle. func (t *Twalk) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -1213,7 +1180,6 @@ func (t *Twalk) handle(cs *connState) message { // handle implements handler.handle. func (t *Twalkgetattr) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -1270,7 +1236,6 @@ func (t *Tumknod) handle(cs *connState) message { // handle implements handler.handle. func (t *Tlconnect) handle(cs *connState) message { - // Lookup the FID. ref, ok := cs.LookupFID(t.FID) if !ok { return newErr(syscall.EBADF) @@ -1303,7 +1268,6 @@ func (t *Tchannel) handle(cs *connState) message { return newErr(err) } - // Lookup the given channel. ch := cs.lookupChannel(t.ID) if ch == nil { return newErr(syscall.ENOSYS) diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 9e7db8b30..67daa6c24 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -305,7 +305,7 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { buf := binary.Marshal(nil, usermem.ByteOrder, c.ptraceGetRegs()) return c.Native(uintptr(usermem.ByteOrder.Uint64(buf[addr:]))), nil } - // TODO(b/34088053): debug registers + // Note: x86 debug registers are missing. return c.Native(0), nil } @@ -320,6 +320,6 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error { _, err := c.PtraceSetRegs(bytes.NewBuffer(buf)) return err } - // TODO(b/34088053): debug registers + // Note: x86 debug registers are missing. return nil } diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 1f78d54a2..e1f2fea60 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -22,6 +22,7 @@ import ( "sync" "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/urpc" ) @@ -56,6 +57,9 @@ type Profile struct { // traceFile is the current execution trace output file. traceFile *fd.FD + + // Kernel is the kernel under profile. + Kernel *kernel.Kernel } // StartCPUProfile is an RPC stub which starts recording the CPU profile in a @@ -147,6 +151,9 @@ func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error { return err } + // Ensure all trace contexts are registered. + p.Kernel.RebuildTraceContexts() + p.traceFile = output return nil } @@ -158,9 +165,15 @@ func (p *Profile) StopTrace(_, _ *struct{}) error { defer p.mu.Unlock() if p.traceFile == nil { - return errors.New("Execution tracing not start") + return errors.New("Execution tracing not started") } + // Similarly to the case above, if tasks have not ended traces, we will + // lose information. Thus we need to rebuild the tasks in order to have + // complete information. This will not lose information if multiple + // traces are overlapping. + p.Kernel.RebuildTraceContexts() + trace.Stop() p.traceFile.Close() p.traceFile = nil diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index c35faeb4c..ced51c66c 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -268,14 +268,17 @@ func (proc *Proc) Ps(args *PsArgs, out *string) error { } // Process contains information about a single process in a Sandbox. -// TODO(b/117881927): Implement TTY field. type Process struct { UID auth.KUID `json:"uid"` PID kernel.ThreadID `json:"pid"` // Parent PID - PPID kernel.ThreadID `json:"ppid"` + PPID kernel.ThreadID `json:"ppid"` + Threads []kernel.ThreadID `json:"threads"` // Processor utilization C int32 `json:"c"` + // TTY name of the process. Will be of the form "pts/N" if there is a + // TTY, or "?" if there is not. + TTY string `json:"tty"` // Start time STime string `json:"stime"` // CPU time @@ -285,18 +288,19 @@ type Process struct { } // ProcessListToTable prints a table with the following format: -// UID PID PPID C STIME TIME CMD -// 0 1 0 0 14:04 505262ns tail +// UID PID PPID C TTY STIME TIME CMD +// 0 1 0 0 pty/4 14:04 505262ns tail func ProcessListToTable(pl []*Process) string { var buf bytes.Buffer tw := tabwriter.NewWriter(&buf, 10, 1, 3, ' ', 0) - fmt.Fprint(tw, "UID\tPID\tPPID\tC\tSTIME\tTIME\tCMD") + fmt.Fprint(tw, "UID\tPID\tPPID\tC\tTTY\tSTIME\tTIME\tCMD") for _, d := range pl { - fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s", + fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s\t%s", d.UID, d.PID, d.PPID, d.C, + d.TTY, d.STime, d.Time, d.Cmd) @@ -307,7 +311,7 @@ func ProcessListToTable(pl []*Process) string { // ProcessListToJSON will return the JSON representation of ps. func ProcessListToJSON(pl []*Process) (string, error) { - b, err := json.Marshal(pl) + b, err := json.MarshalIndent(pl, "", " ") if err != nil { return "", fmt.Errorf("couldn't marshal process list %v: %v", pl, err) } @@ -334,7 +338,9 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ts := k.TaskSet() now := k.RealtimeClock().Now() for _, tg := range ts.Root.ThreadGroups() { - pid := tg.PIDNamespace().IDOfThreadGroup(tg) + pidns := tg.PIDNamespace() + pid := pidns.IDOfThreadGroup(tg) + // If tg has already been reaped ignore it. if pid == 0 { continue @@ -345,16 +351,19 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ppid := kernel.ThreadID(0) if p := tg.Leader().Parent(); p != nil { - ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup()) + ppid = pidns.IDOfThreadGroup(p.ThreadGroup()) } + threads := tg.MemberIDs(pidns) *out = append(*out, &Process{ - UID: tg.Leader().Credentials().EffectiveKUID, - PID: pid, - PPID: ppid, - STime: formatStartTime(now, tg.Leader().StartTime()), - C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), - Time: tg.CPUStats().SysTime.String(), - Cmd: tg.Leader().Name(), + UID: tg.Leader().Credentials().EffectiveKUID, + PID: pid, + PPID: ppid, + Threads: threads, + STime: formatStartTime(now, tg.Leader().StartTime()), + C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), + Time: tg.CPUStats().SysTime.String(), + Cmd: tg.Leader().Name(), + TTY: ttyName(tg.TTY()), }) } sort.Slice(*out, func(i, j int) bool { return (*out)[i].PID < (*out)[j].PID }) @@ -395,3 +404,10 @@ func percentCPU(stats usage.CPUStats, startTime, now ktime.Time) int32 { } return int32(percentCPU) } + +func ttyName(tty *kernel.TTY) string { + if tty == nil { + return "?" + } + return fmt.Sprintf("pts/%d", tty.Index) +} diff --git a/pkg/sentry/control/proc_test.go b/pkg/sentry/control/proc_test.go index d8ada2694..0a88459b2 100644 --- a/pkg/sentry/control/proc_test.go +++ b/pkg/sentry/control/proc_test.go @@ -34,7 +34,7 @@ func TestProcessListTable(t *testing.T) { }{ { pl: []*Process{}, - expected: "UID PID PPID C STIME TIME CMD", + expected: "UID PID PPID C TTY STIME TIME CMD", }, { pl: []*Process{ @@ -43,6 +43,7 @@ func TestProcessListTable(t *testing.T) { PID: 0, PPID: 0, C: 0, + TTY: "?", STime: "0", Time: "0", Cmd: "zero", @@ -52,14 +53,15 @@ func TestProcessListTable(t *testing.T) { PID: 1, PPID: 1, C: 1, + TTY: "pts/4", STime: "1", Time: "1", Cmd: "one", }, }, - expected: `UID PID PPID C STIME TIME CMD -0 0 0 0 0 0 zero -1 1 1 1 1 1 one`, + expected: `UID PID PPID C TTY STIME TIME CMD +0 0 0 0 ? 0 0 zero +1 1 1 1 pts/4 1 1 one`, }, } diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index 8c17603f8..c09f3b71c 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -234,6 +234,8 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, if err != nil { return nil, err } + // We're not going to use newFile after return. + defer newFile.close(ctx) // Stabilize the endpoint map while creation is in progress. unlock := i.session().endpoints.lock() @@ -254,7 +256,6 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, // Get the attributes of the file to create inode key. qid, mask, attr, err := getattr(ctx, newFile) if err != nil { - newFile.close(ctx) return nil, err } @@ -270,7 +271,6 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, // cloned and re-opened multiple times after creation. _, unopened, err := i.fileState.file.walk(ctx, []string{name}) if err != nil { - newFile.close(ctx) return nil, err } diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 0da608548..4e358a46a 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -143,9 +143,9 @@ type session struct { // socket files. This allows unix domain sockets to be used with paths that // belong to a gofer. // - // TODO(b/77154739): there are few possible races with someone stat'ing the - // file and another deleting it concurrently, where the file will not be - // reported as socket file. + // TODO(gvisor.dev/issue/1200): there are few possible races with someone + // stat'ing the file and another deleting it concurrently, where the file + // will not be reported as socket file. endpoints *endpointMaps `state:"wait"` } diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 2d43dff1d..91e2fde2f 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -270,6 +270,14 @@ func (i *Inode) Getxattr(name string) (string, error) { return i.InodeOperations.Getxattr(i, name) } +// Setxattr calls i.InodeOperations.Setxattr with i as the Inode. +func (i *Inode) Setxattr(name, value string) error { + if i.overlay != nil { + return overlaySetxattr(i.overlay, name, value) + } + return i.InodeOperations.Setxattr(i, name, value) +} + // Listxattr calls i.InodeOperations.Listxattr with i as the Inode. func (i *Inode) Listxattr() (map[string]struct{}, error) { if i.overlay != nil { diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index a09147080..63a991beb 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -552,6 +552,11 @@ func overlayGetxattr(o *overlayEntry, name string) (string, error) { return s, err } +// TODO(b/146028302): Support setxattr for overlayfs. +func overlaySetxattr(o *overlayEntry, name, value string) error { + return syserror.EOPNOTSUPP +} + func overlayListxattr(o *overlayEntry) (map[string]struct{}, error) { o.copyMu.RLock() defer o.copyMu.RUnlock() diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 2a598149d..0e46c5fb7 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -67,29 +67,28 @@ type taskDir struct { var _ fs.InodeOperations = (*taskDir)(nil) // newTaskDir creates a new proc task entry. -func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, showSubtasks bool) *fs.Inode { +func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode { contents := map[string]*fs.Inode{ - "auxv": newAuxvec(t, msrc), - "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), - "comm": newComm(t, msrc), - "environ": newExecArgInode(t, msrc, environExecArg), - "exe": newExe(t, msrc), - "fd": newFdDir(t, msrc), - "fdinfo": newFdInfoDir(t, msrc), - "gid_map": newGIDMap(t, msrc), - // FIXME(b/123511468): create the correct io file for threads. - "io": newIO(t, msrc), + "auxv": newAuxvec(t, msrc), + "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), + "comm": newComm(t, msrc), + "environ": newExecArgInode(t, msrc, environExecArg), + "exe": newExe(t, msrc), + "fd": newFdDir(t, msrc), + "fdinfo": newFdInfoDir(t, msrc), + "gid_map": newGIDMap(t, msrc), + "io": newIO(t, msrc, isThreadGroup), "maps": newMaps(t, msrc), "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), "ns": newNamespaceDir(t, msrc), "smaps": newSmaps(t, msrc), - "stat": newTaskStat(t, msrc, showSubtasks, p.pidns), + "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns), "statm": newStatm(t, msrc), "status": newStatus(t, msrc, p.pidns), "uid_map": newUIDMap(t, msrc), } - if showSubtasks { + if isThreadGroup { contents["task"] = p.newSubtasks(t, msrc) } if len(p.cgroupControllers) > 0 { @@ -619,8 +618,11 @@ type ioData struct { ioUsage } -func newIO(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) +func newIO(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode { + if isThreadGroup { + return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) + } + return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index ff8138820..917f90cc0 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -53,8 +53,8 @@ func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal d: d, n: n, ld: newLineDiscipline(termios), - masterKTTY: &kernel.TTY{}, - slaveKTTY: &kernel.TTY{}, + masterKTTY: &kernel.TTY{Index: n}, + slaveKTTY: &kernel.TTY{Index: n}, } t.EnableLeakCheck("tty.Terminal") return &t diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 7ccff8b0d..880b7bcd3 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -38,6 +38,7 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/fd", + "//pkg/fspath", "//pkg/log", "//pkg/sentry/arch", "//pkg/sentry/context", diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 94cd74095..177ce2cb9 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -81,7 +81,11 @@ func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vf ctx := contexttest.Context(b) creds := auth.CredentialsFromContext(ctx) - if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}); err != nil { + if err := vfsfs.MountAt(ctx, creds, imagePath, pop, "extfs", &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + InternalData: int(f.Fd()), + }, + }); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } return func() { diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 307e4d68c..e9f756732 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -147,55 +147,54 @@ func TestSeek(t *testing.T) { t.Fatalf("vfsfs.OpenAt failed: %v", err) } - if n, err := fd.Impl().Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { + if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { t.Errorf("expected seek position 0, got %d and error %v", n, err) } - stat, err := fd.Impl().Stat(ctx, vfs.StatOptions{}) + stat, err := fd.Stat(ctx, vfs.StatOptions{}) if err != nil { t.Errorf("fd.stat failed for file %s in image %s: %v", test.path, test.image, err) } // We should be able to seek beyond the end of file. size := int64(stat.Size) - if n, err := fd.Impl().Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { + if n, err := fd.Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } - if n, err := fd.Impl().Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { + if n, err := fd.Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+3, n, err) } // Make sure negative offsets work with SEEK_CUR. - if n, err := fd.Impl().Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { + if n, err := fd.Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } // Make sure SEEK_END works with regular files. - switch fd.Impl().(type) { - case *regularFileFD: + if _, ok := fd.Impl().(*regularFileFD); ok { // Seek back to 0. - if n, err := fd.Impl().Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { + if n, err := fd.Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", 0, n, err) } // Seek forward beyond EOF. - if n, err := fd.Impl().Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { + if n, err := fd.Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } } @@ -456,7 +455,7 @@ func TestRead(t *testing.T) { want := make([]byte, 1) for { n, err := f.Read(want) - fd.Impl().Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) + fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) if diff := cmp.Diff(got, want); diff != "" { t.Errorf("file data mismatch (-want +got):\n%s", diff) @@ -464,7 +463,7 @@ func TestRead(t *testing.T) { // Make sure there is no more file data left after getting EOF. if n == 0 || err == io.EOF { - if n, _ := fd.Impl().Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { + if n, _ := fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { t.Errorf("extra unexpected file data in file %s in image %s", test.absPath, test.image) } @@ -574,7 +573,7 @@ func TestIterDirents(t *testing.T) { } cb := &iterDirentsCb{} - if err = fd.Impl().IterDirents(ctx, cb); err != nil { + if err = fd.IterDirents(ctx, cb); err != nil { t.Fatalf("dir fd.IterDirents() failed: %v", err) } diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 2d15e8aaf..e7aa3b41b 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -20,6 +20,7 @@ import ( "sync" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -441,3 +442,10 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return syserror.EROFS } + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD new file mode 100644 index 000000000..52596c090 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -0,0 +1,60 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "slot_list", + out = "slot_list.go", + package = "kernfs", + prefix = "slot", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*slot", + "Linker": "*slot", + }, +) + +go_library( + name = "kernfs", + srcs = [ + "dynamic_bytes_file.go", + "fd_impl_util.go", + "filesystem.go", + "inode_impl_util.go", + "kernfs.go", + "slot_list.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs", + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/log", + "//pkg/refs", + "//pkg/sentry/context", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/memmap", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) + +go_test( + name = "kernfs_test", + size = "small", + srcs = ["kernfs_test.go"], + deps = [ + ":kernfs", + "//pkg/abi/linux", + "//pkg/sentry/context", + "//pkg/sentry/context/contexttest", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + "@com_github_google_go-cmp//cmp:go_default_library", + ], +) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go new file mode 100644 index 000000000..30c06baf0 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -0,0 +1,131 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// DynamicBytesFile implements kernfs.Inode and represents a read-only +// file whose contents are backed by a vfs.DynamicBytesSource. +// +// Must be initialized with Init before first use. +type DynamicBytesFile struct { + InodeAttrs + InodeNoopRefCount + InodeNotDirectory + InodeNotSymlink + + data vfs.DynamicBytesSource +} + +// Init intializes a dynamic bytes file. +func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource) { + f.InodeAttrs.Init(creds, ino, linux.ModeRegular|0444) + f.data = data +} + +// Open implements Inode.Open. +func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &DynamicBytesFD{} + fd.Init(rp.Mount(), vfsd, f.data, flags) + return &fd.vfsfd, nil +} + +// SetStat implements Inode.SetStat. +func (f *DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + // DynamicBytesFiles are immutable. + return syserror.EPERM +} + +// DynamicBytesFD implements vfs.FileDescriptionImpl for an FD backed by a +// DynamicBytesFile. +// +// Must be initialized with Init before first use. +type DynamicBytesFD struct { + vfs.FileDescriptionDefaultImpl + vfs.DynamicBytesFileDescriptionImpl + + vfsfd vfs.FileDescription + inode Inode + flags uint32 +} + +// Init initializes a DynamicBytesFD. +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) { + m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + fd.flags = flags + fd.inode = d.Impl().(*Dentry).inode + fd.SetDataSource(data) + fd.vfsfd.Init(fd, m, d) +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *DynamicBytesFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.Seek(ctx, offset, whence) +} + +// Read implmenets vfs.FileDescriptionImpl.Read. +func (fd *DynamicBytesFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.Read(ctx, dst, opts) +} + +// PRead implmenets vfs.FileDescriptionImpl.PRead. +func (fd *DynamicBytesFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.PRead(ctx, dst, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *DynamicBytesFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return fd.FileDescriptionDefaultImpl.Write(ctx, src, opts) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return fd.FileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *DynamicBytesFD) Release() {} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := fd.vfsfd.VirtualDentry().Mount().Filesystem() + return fd.inode.Stat(fs), nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { + // DynamicBytesFiles are immutable. + return syserror.EPERM +} + +// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. +func (fd *DynamicBytesFD) StatusFlags(ctx context.Context) (uint32, error) { + return fd.flags, nil +} + +// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. +func (fd *DynamicBytesFD) SetStatusFlags(ctx context.Context, flags uint32) error { + // None of the flags settable by fcntl(F_SETFL) are supported, so this is a + // no-op. + return nil +} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go new file mode 100644 index 000000000..d6c18937a --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -0,0 +1,207 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// GenericDirectoryFD implements vfs.FileDescriptionImpl for a generic directory +// inode that uses OrderChildren to track child nodes. GenericDirectoryFD is not +// compatible with dynamic directories. +// +// Note that GenericDirectoryFD holds a lock over OrderedChildren while calling +// IterDirents callback. The IterDirents callback therefore cannot hash or +// unhash children, or recursively call IterDirents on the same underlying +// inode. +// +// Must be initialize with Init before first use. +type GenericDirectoryFD struct { + vfs.FileDescriptionDefaultImpl + vfs.DirectoryFileDescriptionDefaultImpl + + vfsfd vfs.FileDescription + children *OrderedChildren + flags uint32 + off int64 +} + +// Init initializes a GenericDirectoryFD. +func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) { + m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + fd.children = children + fd.flags = flags + fd.vfsfd.Init(fd, m, d) +} + +// VFSFileDescription returns a pointer to the vfs.FileDescription representing +// this object. +func (fd *GenericDirectoryFD) VFSFileDescription() *vfs.FileDescription { + return &fd.vfsfd +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *GenericDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return fd.FileDescriptionDefaultImpl.ConfigureMMap(ctx, opts) +} + +// Read implmenets vfs.FileDescriptionImpl.Read. +func (fd *GenericDirectoryFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.Read(ctx, dst, opts) +} + +// PRead implmenets vfs.FileDescriptionImpl.PRead. +func (fd *GenericDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.PRead(ctx, dst, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *GenericDirectoryFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.Write(ctx, src, opts) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) +} + +// Release implements vfs.FileDecriptionImpl.Release. +func (fd *GenericDirectoryFD) Release() {} + +func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { + return fd.vfsfd.VirtualDentry().Mount().Filesystem() +} + +func (fd *GenericDirectoryFD) inode() Inode { + return fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode +} + +// IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds +// o.mu when calling cb. +func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + vfsFS := fd.filesystem() + fs := vfsFS.Impl().(*Filesystem) + vfsd := fd.vfsfd.VirtualDentry().Dentry() + + fs.mu.Lock() + defer fs.mu.Unlock() + + // Handle ".". + if fd.off == 0 { + stat := fd.inode().Stat(vfsFS) + dirent := vfs.Dirent{ + Name: ".", + Type: linux.DT_DIR, + Ino: stat.Ino, + NextOff: 1, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + // Handle "..". + if fd.off == 1 { + parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode + stat := parentInode.Stat(vfsFS) + dirent := vfs.Dirent{ + Name: "..", + Type: linux.FileMode(stat.Mode).DirentType(), + Ino: stat.Ino, + NextOff: 2, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + // Handle static children. + fd.children.mu.RLock() + defer fd.children.mu.RUnlock() + // fd.off accounts for "." and "..", but fd.children do not track + // these. + childIdx := fd.off - 2 + for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { + inode := it.Dentry.Impl().(*Dentry).inode + stat := inode.Stat(vfsFS) + dirent := vfs.Dirent{ + Name: it.Name, + Type: linux.FileMode(stat.Mode).DirentType(), + Ino: stat.Ino, + NextOff: fd.off + 1, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + return nil +} + +// Seek implements vfs.FileDecriptionImpl.Seek. +func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fs := fd.filesystem().Impl().(*Filesystem) + fs.mu.Lock() + defer fs.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + // Use offset as given. + case linux.SEEK_CUR: + offset += fd.off + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. +func (fd *GenericDirectoryFD) StatusFlags(ctx context.Context) (uint32, error) { + return fd.flags, nil +} + +// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. +func (fd *GenericDirectoryFD) SetStatusFlags(ctx context.Context, flags uint32) error { + // None of the flags settable by fcntl(F_SETFL) are supported, so this is a + // no-op. + return nil +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := fd.filesystem() + inode := fd.inode() + return inode.Stat(fs), nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + fs := fd.filesystem() + inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode + return inode.SetStat(fs, opts) +} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go new file mode 100644 index 000000000..db486b6c1 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -0,0 +1,691 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This file implements vfs.FilesystemImpl for kernfs. + +package kernfs + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// stepExistingLocked resolves rp.Component() in parent directory vfsd. +// +// stepExistingLocked is loosely analogous to fs/namei.c:walk_component(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). +// +// Postcondition: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) (*vfs.Dentry, error) { + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, syserror.ENOTDIR + } + // Directory searchable? + if err := d.inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } +afterSymlink: + d.dirMu.Lock() + nextVFSD, err := rp.ResolveComponent(vfsd) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + if nextVFSD != nil { + // Cached dentry exists, revalidate. + next := nextVFSD.Impl().(*Dentry) + if !next.inode.Valid(ctx) { + d.dirMu.Lock() + rp.VirtualFilesystem().ForceDeleteDentry(nextVFSD) + d.dirMu.Unlock() + fs.deferDecRef(nextVFSD) // Reference from Lookup. + nextVFSD = nil + } + } + if nextVFSD == nil { + // Dentry isn't cached; it either doesn't exist or failed + // revalidation. Attempt to resolve it via Lookup. + name := rp.Component() + var err error + nextVFSD, err = d.inode.Lookup(ctx, name) + // Reference on nextVFSD dropped by a corresponding Valid. + if err != nil { + return nil, err + } + d.InsertChild(name, nextVFSD) + } + next := nextVFSD.Impl().(*Dentry) + + // Resolve any symlink at current path component. + if rp.ShouldFollowSymlink() && d.isSymlink() { + // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks". + target, err := next.inode.Readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + goto afterSymlink + + } + rp.Advance() + return nextVFSD, nil +} + +// walkExistingLocked resolves rp to an existing file. +// +// walkExistingLocked is loosely analogous to Linux's +// fs/namei.c:path_lookupat(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. +// +// Postconditions: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { + vfsd := rp.Start() + for !rp.Done() { + var err error + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + if err != nil { + return nil, nil, err + } + } + d := vfsd.Impl().(*Dentry) + if rp.MustBeDir() && !d.isDir() { + return nil, nil, syserror.ENOTDIR + } + return vfsd, d.inode, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory. It does not check that the returned directory is +// searchable by the provider of rp. +// +// walkParentDirLocked is loosely analogous to Linux's +// fs/namei.c:path_parentat(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). +// +// Postconditions: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { + vfsd := rp.Start() + for !rp.Final() { + var err error + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + if err != nil { + return nil, nil, err + } + } + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, nil, syserror.ENOTDIR + } + return vfsd, d.inode, nil +} + +// checkCreateLocked checks that a file named rp.Component() may be created in +// directory parentVFSD, then returns rp.Component(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. parentInode +// == parentVFSD.Impl().(*Dentry).Inode. isDir(parentInode) == true. +func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return "", err + } + pc := rp.Component() + if pc == "." || pc == ".." { + return "", syserror.EEXIST + } + childVFSD, err := rp.ResolveChild(parentVFSD, pc) + if err != nil { + return "", err + } + if childVFSD != nil { + return "", syserror.EEXIST + } + if parentVFSD.IsDisowned() { + return "", syserror.ENOENT + } + return pc, nil +} + +// checkDeleteLocked checks that the file represented by vfsd may be deleted. +// +// Preconditions: Filesystem.mu must be locked for at least reading. +func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { + parentVFSD := vfsd.Parent() + if parentVFSD == nil { + return syserror.EBUSY + } + if parentVFSD.IsDisowned() { + return syserror.ENOENT + } + if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + return nil +} + +// checkRenameLocked checks that a rename operation may be performed on the +// target dentry across the given set of parent directories. The target dentry +// may be nil. +// +// Precondition: isDir(dstInode) == true. +func checkRenameLocked(creds *auth.Credentials, src, dstDir *vfs.Dentry, dstInode Inode) error { + srcDir := src.Parent() + if srcDir == nil { + return syserror.EBUSY + } + if srcDir.IsDisowned() { + return syserror.ENOENT + } + if dstDir.IsDisowned() { + return syserror.ENOENT + } + // Check for creation permissions on dst dir. + if err := dstInode.CheckPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + + return nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *Filesystem) Release() { +} + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *Filesystem) Sync(ctx context.Context) error { + // All filesystem state is in-memory. + return nil +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + if err != nil { + return nil, err + } + + if opts.CheckSearchable { + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + } + vfsd.IncRef() // Ownership transferred to caller. + return vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + + d := vd.Dentry().Impl().(*Dentry) + if d.isDir() { + return syserror.EPERM + } + + child, err := parentInode.NewLink(ctx, pc, d.inode) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + child, err := parentInode.NewDir(ctx, pc, opts) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + new, err := parentInode.NewNode(ctx, pc, opts) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, new) + return nil +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // Filter out flags that are not supported by kernfs. O_DIRECTORY and + // O_NOFOLLOW have no effect here (they're handled by VFS by setting + // appropriate bits in rp), but are returned by + // FileDescriptionImpl.StatusFlags(). + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW + ats := vfs.AccessTypesForOpenFlags(opts.Flags) + + // Do not create new file. + if opts.Flags&linux.O_CREAT == 0 { + fs.mu.RLock() + defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + if err != nil { + return nil, err + } + if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return inode.Open(rp, vfsd, opts.Flags) + } + + // May create new file. + mustCreate := opts.Flags&linux.O_EXCL != 0 + vfsd := rp.Start() + inode := vfsd.Impl().(*Dentry).inode + fs.mu.Lock() + defer fs.mu.Unlock() + if rp.Done() { + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + if mustCreate { + return nil, syserror.EEXIST + } + if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return inode.Open(rp, vfsd, opts.Flags) + } +afterTrailingSymlink: + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + // Reject attempts to open directories with O_CREAT. + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + pc := rp.Component() + if pc == "." || pc == ".." { + return nil, syserror.EISDIR + } + // Determine whether or not we need to create a file. + childVFSD, err := rp.ResolveChild(parentVFSD, pc) + if err != nil { + return nil, err + } + if childVFSD == nil { + // Already checked for searchability above; now check for writability. + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return nil, err + } + defer rp.Mount().EndWrite() + // Create and open the child. + child, err := parentInode.NewFile(ctx, pc, opts) + if err != nil { + return nil, err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return child.Impl().(*Dentry).inode.Open(rp, child, opts.Flags) + } + // Open existing file or follow symlink. + if mustCreate { + return nil, syserror.EEXIST + } + childDentry := childVFSD.Impl().(*Dentry) + childInode := childDentry.inode + if rp.ShouldFollowSymlink() { + if childDentry.isSymlink() { + target, err := childInode.Readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + // rp.Final() may no longer be true since we now need to resolve the + // symlink target. + goto afterTrailingSymlink + } + } + if err := childInode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return childInode.Open(rp, childVFSD, opts.Flags) +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + fs.mu.RLock() + d, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return "", err + } + if !d.Impl().(*Dentry).isSymlink() { + return "", syserror.EINVAL + } + return inode.Readlink(ctx) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { + noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 + exchange := opts.Flags&linux.RENAME_EXCHANGE != 0 + whiteout := opts.Flags&linux.RENAME_WHITEOUT != 0 + if exchange && (noReplace || whiteout) { + // Can't specify RENAME_NOREPLACE or RENAME_WHITEOUT with RENAME_EXCHANGE. + return syserror.EINVAL + } + if exchange || whiteout { + // Exchange and Whiteout flags are not supported on kernfs. + return syserror.EINVAL + } + + fs.mu.Lock() + defer fs.mu.Lock() + + mnt := rp.Mount() + if mnt != vd.Mount() { + return syserror.EXDEV + } + + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + + srcVFSD := vd.Dentry() + srcDirVFSD := srcVFSD.Parent() + + // Can we remove the src dentry? + if err := checkDeleteLocked(rp, srcVFSD); err != nil { + return err + } + + // Can we create the dst dentry? + var dstVFSD *vfs.Dentry + pc, err := checkCreateLocked(rp, dstDirVFSD, dstDirInode) + switch err { + case nil: + // Ok, continue with rename as replacement. + case syserror.EEXIST: + if noReplace { + // Won't overwrite existing node since RENAME_NOREPLACE was requested. + return syserror.EEXIST + } + dstVFSD, err = rp.ResolveChild(dstDirVFSD, pc) + if err != nil { + panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) + } + default: + return err + } + + mntns := vfs.MountNamespaceFromContext(ctx) + virtfs := rp.VirtualFilesystem() + + srcDirDentry := srcDirVFSD.Impl().(*Dentry) + dstDirDentry := dstDirVFSD.Impl().(*Dentry) + + // We can't deadlock here due to lock ordering because we're protected from + // concurrent renames by fs.mu held for writing. + srcDirDentry.dirMu.Lock() + defer srcDirDentry.dirMu.Unlock() + dstDirDentry.dirMu.Lock() + defer dstDirDentry.dirMu.Unlock() + + if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { + return err + } + srcDirInode := srcDirDentry.inode + replaced, err := srcDirInode.Rename(ctx, srcVFSD.Name(), pc, srcVFSD, dstDirVFSD) + if err != nil { + virtfs.AbortRenameDentry(srcVFSD, dstVFSD) + return err + } + virtfs.CommitRenameReplaceDentry(srcVFSD, dstDirVFSD, pc, replaced) + return nil +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + if err := checkDeleteLocked(rp, vfsd); err != nil { + return err + } + if !vfsd.Impl().(*Dentry).isDir() { + return syserror.ENOTDIR + } + if inode.HasChildren() { + return syserror.ENOTEMPTY + } + virtfs := rp.VirtualFilesystem() + parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry.dirMu.Lock() + defer parentDentry.dirMu.Unlock() + if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + return err + } + if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { + virtfs.AbortDeleteDentry(vfsd) + return err + } + virtfs.CommitDeleteDentry(vfsd) + return nil +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + fs.mu.RLock() + _, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return err + } + if opts.Stat.Mask == 0 { + return nil + } + return inode.SetStat(fs.VFSFilesystem(), opts) +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + fs.mu.RLock() + _, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return linux.Statx{}, err + } + return inode.Stat(fs.VFSFilesystem()), nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return linux.Statfs{}, err + } + // TODO: actually implement statfs + return linux.Statfs{}, syserror.ENOSYS +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + child, err := parentInode.NewSymlink(ctx, pc, target) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + vfsd, _, err := fs.walkExistingLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + if err := checkDeleteLocked(rp, vfsd); err != nil { + return err + } + if vfsd.Impl().(*Dentry).isDir() { + return syserror.EISDIR + } + virtfs := rp.VirtualFilesystem() + parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry.dirMu.Lock() + defer parentDentry.dirMu.Unlock() + if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + return err + } + if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { + virtfs.AbortDeleteDentry(vfsd) + return err + } + virtfs.CommitDeleteDentry(vfsd) + return nil +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *Filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go new file mode 100644 index 000000000..7b45b702a --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -0,0 +1,492 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "fmt" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// InodeNoopRefCount partially implements the Inode interface, specifically the +// inodeRefs sub interface. InodeNoopRefCount implements a simple reference +// count for inodes, performing no extra actions when references are obtained or +// released. This is suitable for simple file inodes that don't reference any +// resources. +type InodeNoopRefCount struct { +} + +// IncRef implements Inode.IncRef. +func (n *InodeNoopRefCount) IncRef() { +} + +// DecRef implements Inode.DecRef. +func (n *InodeNoopRefCount) DecRef() { +} + +// TryIncRef implements Inode.TryIncRef. +func (n *InodeNoopRefCount) TryIncRef() bool { + return true +} + +// Destroy implements Inode.Destroy. +func (n *InodeNoopRefCount) Destroy() { +} + +// InodeDirectoryNoNewChildren partially implements the Inode interface. +// InodeDirectoryNoNewChildren represents a directory inode which does not +// support creation of new children. +type InodeDirectoryNoNewChildren struct{} + +// NewFile implements Inode.NewFile. +func (*InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewDir implements Inode.NewDir. +func (*InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewLink implements Inode.NewLink. +func (*InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewSymlink implements Inode.NewSymlink. +func (*InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewNode implements Inode.NewNode. +func (*InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// InodeNotDirectory partially implements the Inode interface, specifically the +// inodeDirectory and inodeDynamicDirectory sub interfaces. Inodes that do not +// represent directories can embed this to provide no-op implementations for +// directory-related functions. +type InodeNotDirectory struct { +} + +// HasChildren implements Inode.HasChildren. +func (*InodeNotDirectory) HasChildren() bool { + return false +} + +// NewFile implements Inode.NewFile. +func (*InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { + panic("NewFile called on non-directory inode") +} + +// NewDir implements Inode.NewDir. +func (*InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { + panic("NewDir called on non-directory inode") +} + +// NewLink implements Inode.NewLinkink. +func (*InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { + panic("NewLink called on non-directory inode") +} + +// NewSymlink implements Inode.NewSymlink. +func (*InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + panic("NewSymlink called on non-directory inode") +} + +// NewNode implements Inode.NewNode. +func (*InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + panic("NewNode called on non-directory inode") +} + +// Unlink implements Inode.Unlink. +func (*InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { + panic("Unlink called on non-directory inode") +} + +// RmDir implements Inode.RmDir. +func (*InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { + panic("RmDir called on non-directory inode") +} + +// Rename implements Inode.Rename. +func (*InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { + panic("Rename called on non-directory inode") +} + +// Lookup implements Inode.Lookup. +func (*InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + panic("Lookup called on non-directory inode") +} + +// Valid implements Inode.Valid. +func (*InodeNotDirectory) Valid(context.Context) bool { + return true +} + +// InodeNoDynamicLookup partially implements the Inode interface, specifically +// the inodeDynamicLookup sub interface. Directory inodes that do not support +// dymanic entries (i.e. entries that are not "hashed" into the +// vfs.Dentry.children) can embed this to provide no-op implementations for +// functions related to dynamic entries. +type InodeNoDynamicLookup struct{} + +// Lookup implements Inode.Lookup. +func (*InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + return nil, syserror.ENOENT +} + +// Valid implements Inode.Valid. +func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool { + return true +} + +// InodeNotSymlink partially implements the Inode interface, specifically the +// inodeSymlink sub interface. All inodes that are not symlinks may embed this +// to return the appropriate errors from symlink-related functions. +type InodeNotSymlink struct{} + +// Readlink implements Inode.Readlink. +func (*InodeNotSymlink) Readlink(context.Context) (string, error) { + return "", syserror.EINVAL +} + +// InodeAttrs partially implements the Inode interface, specifically the +// inodeMetadata sub interface. InodeAttrs provides functionality related to +// inode attributes. +// +// Must be initialized by Init prior to first use. +type InodeAttrs struct { + ino uint64 + mode uint32 + uid uint32 + gid uint32 + nlink uint32 +} + +// Init initializes this InodeAttrs. +func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMode) { + if mode.FileType() == 0 { + panic(fmt.Sprintf("No file type specified in 'mode' for InodeAttrs.Init(): mode=0%o", mode)) + } + + nlink := uint32(1) + if mode.FileType() == linux.ModeDirectory { + nlink = 2 + } + atomic.StoreUint64(&a.ino, ino) + atomic.StoreUint32(&a.mode, uint32(mode)) + atomic.StoreUint32(&a.uid, uint32(creds.EffectiveKUID)) + atomic.StoreUint32(&a.gid, uint32(creds.EffectiveKGID)) + atomic.StoreUint32(&a.nlink, nlink) +} + +// Mode implements Inode.Mode. +func (a *InodeAttrs) Mode() linux.FileMode { + return linux.FileMode(atomic.LoadUint32(&a.mode)) +} + +// Stat partially implements Inode.Stat. Note that this function doesn't provide +// all the stat fields, and the embedder should consider extending the result +// with filesystem-specific fields. +func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx { + var stat linux.Statx + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK + stat.Ino = atomic.LoadUint64(&a.ino) + stat.Mode = uint16(a.Mode()) + stat.UID = atomic.LoadUint32(&a.uid) + stat.GID = atomic.LoadUint32(&a.gid) + stat.Nlink = atomic.LoadUint32(&a.nlink) + + // TODO: Implement other stat fields like timestamps. + + return stat +} + +// SetStat implements Inode.SetStat. +func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { + stat := opts.Stat + if stat.Mask&linux.STATX_MODE != 0 { + for { + old := atomic.LoadUint32(&a.mode) + new := old | uint32(stat.Mode & ^uint16(linux.S_IFMT)) + if swapped := atomic.CompareAndSwapUint32(&a.mode, old, new); swapped { + break + } + } + } + + if stat.Mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&a.uid, stat.UID) + } + if stat.Mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&a.gid, stat.GID) + } + + // Note that not all fields are modifiable. For example, the file type and + // inode numbers are immutable after node creation. + + // TODO: Implement other stat fields like timestamps. + + return nil +} + +// CheckPermissions implements Inode.CheckPermissions. +func (a *InodeAttrs) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + mode := a.Mode() + return vfs.GenericCheckPermissions( + creds, + ats, + mode.FileType() == linux.ModeDirectory, + uint16(mode), + auth.KUID(atomic.LoadUint32(&a.uid)), + auth.KGID(atomic.LoadUint32(&a.gid)), + ) +} + +// IncLinks implements Inode.IncLinks. +func (a *InodeAttrs) IncLinks(n uint32) { + if atomic.AddUint32(&a.nlink, n) <= n { + panic("InodeLink.IncLinks called with no existing links") + } +} + +// DecLinks implements Inode.DecLinks. +func (a *InodeAttrs) DecLinks() { + if nlink := atomic.AddUint32(&a.nlink, ^uint32(0)); nlink == ^uint32(0) { + // Negative overflow + panic("Inode.DecLinks called at 0 links") + } +} + +type slot struct { + Name string + Dentry *vfs.Dentry + slotEntry +} + +// OrderedChildrenOptions contains initialization options for OrderedChildren. +type OrderedChildrenOptions struct { + // Writable indicates whether vfs.FilesystemImpl methods implemented by + // OrderedChildren may modify the tracked children. This applies to + // operations related to rename, unlink and rmdir. If an OrderedChildren is + // not writable, these operations all fail with EPERM. + Writable bool +} + +// OrderedChildren partially implements the Inode interface. OrderedChildren can +// be embedded in directory inodes to keep track of the children in the +// directory, and can then be used to implement a generic directory FD -- see +// GenericDirectoryFD. OrderedChildren is not compatible with dynamic +// directories. +// +// Must be initialize with Init before first use. +type OrderedChildren struct { + refs.AtomicRefCount + + // Can children be modified by user syscalls? It set to false, interface + // methods that would modify the children return EPERM. Immutable. + writable bool + + mu sync.RWMutex + order slotList + set map[string]*slot +} + +// Init initializes an OrderedChildren. +func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { + o.writable = opts.Writable + o.set = make(map[string]*slot) +} + +// DecRef implements Inode.DecRef. +func (o *OrderedChildren) DecRef() { + o.AtomicRefCount.DecRefWithDestructor(o.Destroy) +} + +// Destroy cleans up resources referenced by this OrderedChildren. +func (o *OrderedChildren) Destroy() { + o.mu.Lock() + defer o.mu.Unlock() + o.order.Reset() + o.set = nil +} + +// Populate inserts children into this OrderedChildren, and d's dentry +// cache. Populate returns the number of directories inserted, which the caller +// may use to update the link count for the parent directory. +// +// Precondition: d.Impl() must be a kernfs Dentry. d must represent a directory +// inode. children must not contain any conflicting entries already in o. +func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint32 { + var links uint32 + for name, child := range children { + if child.isDir() { + links++ + } + if err := o.Insert(name, child.VFSDentry()); err != nil { + panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) + } + d.InsertChild(name, child.VFSDentry()) + } + return links +} + +// HasChildren implements Inode.HasChildren. +func (o *OrderedChildren) HasChildren() bool { + o.mu.RLock() + defer o.mu.RUnlock() + return len(o.set) > 0 +} + +// Insert inserts child into o. This ignores the writability of o, as this is +// not part of the vfs.FilesystemImpl interface, and is a lower-level operation. +func (o *OrderedChildren) Insert(name string, child *vfs.Dentry) error { + o.mu.Lock() + defer o.mu.Unlock() + if _, ok := o.set[name]; ok { + return syserror.EEXIST + } + s := &slot{ + Name: name, + Dentry: child, + } + o.order.PushBack(s) + o.set[name] = s + return nil +} + +// Precondition: caller must hold o.mu for writing. +func (o *OrderedChildren) removeLocked(name string) { + if s, ok := o.set[name]; ok { + delete(o.set, name) + o.order.Remove(s) + } +} + +// Precondition: caller must hold o.mu for writing. +func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs.Dentry { + if s, ok := o.set[name]; ok { + // Existing slot with given name, simply replace the dentry. + var old *vfs.Dentry + old, s.Dentry = s.Dentry, new + return old + } + + // No existing slot with given name, create and hash new slot. + s := &slot{ + Name: name, + Dentry: new, + } + o.order.PushBack(s) + o.set[name] = s + return nil +} + +// Precondition: caller must hold o.mu for reading or writing. +func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) error { + s, ok := o.set[name] + if !ok { + return syserror.ENOENT + } + if s.Dentry != child { + panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! OrderedChild: %+v, vfs: %+v", s.Dentry, child)) + } + return nil +} + +// Unlink implements Inode.Unlink. +func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { + if !o.writable { + return syserror.EPERM + } + o.mu.Lock() + defer o.mu.Unlock() + if err := o.checkExistingLocked(name, child); err != nil { + return err + } + o.removeLocked(name) + return nil +} + +// Rmdir implements Inode.Rmdir. +func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { + // We're not responsible for checking that child is a directory, that it's + // empty, or updating any link counts; so this is the same as unlink. + return o.Unlink(ctx, name, child) +} + +type renameAcrossDifferentImplementationsError struct{} + +func (renameAcrossDifferentImplementationsError) Error() string { + return "rename across inodes with different implementations" +} + +// Rename implements Inode.Rename. +// +// Precondition: Rename may only be called across two directory inodes with +// identical implementations of Rename. Practically, this means filesystems that +// implement Rename by embedding OrderedChildren for any directory +// implementation must use OrderedChildren for all directory implementations +// that will support Rename. +// +// Postcondition: reference on any replaced dentry transferred to caller. +func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (*vfs.Dentry, error) { + dst, ok := dstDir.Impl().(*Dentry).inode.(interface{}).(*OrderedChildren) + if !ok { + return nil, renameAcrossDifferentImplementationsError{} + } + if !o.writable || !dst.writable { + return nil, syserror.EPERM + } + // Note: There's a potential deadlock below if concurrent calls to Rename + // refer to the same src and dst directories in reverse. We avoid any + // ordering issues because the caller is required to serialize concurrent + // calls to Rename in accordance with the interface declaration. + o.mu.Lock() + defer o.mu.Unlock() + if dst != o { + dst.mu.Lock() + defer dst.mu.Unlock() + } + if err := o.checkExistingLocked(oldname, child); err != nil { + return nil, err + } + replaced := dst.replaceChildLocked(newname, child) + return replaced, nil +} + +// nthLocked returns an iterator to the nth child tracked by this object. The +// iterator is valid until the caller releases o.mu. Returns nil if the +// requested index falls out of bounds. +// +// Preconditon: Caller must hold o.mu for reading. +func (o *OrderedChildren) nthLocked(i int64) *slot { + for it := o.order.Front(); it != nil && i >= 0; it = it.Next() { + if i == 0 { + return it + } + i-- + } + return nil +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go new file mode 100644 index 000000000..bb01c3d01 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -0,0 +1,405 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package kernfs provides the tools to implement inode-based filesystems. +// Kernfs has two main features: +// +// 1. The Inode interface, which maps VFS2's path-based filesystem operations to +// specific filesystem nodes. Kernfs uses the Inode interface to provide a +// blanket implementation for the vfs.FilesystemImpl. Kernfs also serves as +// the synchronization mechanism for all filesystem operations by holding a +// filesystem-wide lock across all operations. +// +// 2. Various utility types which provide generic implementations for various +// parts of the Inode and vfs.FileDescription interfaces. Client filesystems +// based on kernfs can embed the appropriate set of these to avoid having to +// reimplement common filesystem operations. See inode_impl_util.go and +// fd_impl_util.go. +// +// Reference Model: +// +// Kernfs dentries represents named pointers to inodes. Dentries and inode have +// independent lifetimes and reference counts. A child dentry unconditionally +// holds a reference on its parent directory's dentry. A dentry also holds a +// reference on the inode it points to. Multiple dentries can point to the same +// inode (for example, in the case of hardlinks). File descriptors hold a +// reference to the dentry they're opened on. +// +// Dentries are guaranteed to exist while holding Filesystem.mu for +// reading. Dropping dentries require holding Filesystem.mu for writing. To +// queue dentries for destruction from a read critical section, see +// Filesystem.deferDecRef. +// +// Lock ordering: +// +// kernfs.Filesystem.mu +// kernfs.Dentry.dirMu +// vfs.VirtualFilesystem.mountMu +// vfs.Dentry.mu +// kernfs.Filesystem.droppedDentriesMu +// (inode implementation locks, if any) +package kernfs + +import ( + "fmt" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory +// filesystem. Concrete implementations are expected to embed this in their own +// Filesystem type. +type Filesystem struct { + vfsfs vfs.Filesystem + + droppedDentriesMu sync.Mutex + + // droppedDentries is a list of dentries waiting to be DecRef()ed. This is + // used to defer dentry destruction until mu can be acquired for + // writing. Protected by droppedDentriesMu. + droppedDentries []*vfs.Dentry + + // mu synchronizes the lifetime of Dentries on this filesystem. Holding it + // for reading guarantees continued existence of any resolved dentries, but + // the dentry tree may be modified. + // + // Kernfs dentries can only be DecRef()ed while holding mu for writing. For + // example: + // + // fs.mu.Lock() + // defer fs.mu.Unlock() + // ... + // dentry1.DecRef() + // defer dentry2.DecRef() // Ok, will run before Unlock. + // + // If discarding dentries in a read context, use Filesystem.deferDecRef. For + // example: + // + // fs.mu.RLock() + // fs.mu.processDeferredDecRefs() + // defer fs.mu.RUnlock() + // ... + // fs.deferDecRef(dentry) + mu sync.RWMutex + + // nextInoMinusOne is used to to allocate inode numbers on this + // filesystem. Must be accessed by atomic operations. + nextInoMinusOne uint64 +} + +// deferDecRef defers dropping a dentry ref until the next call to +// processDeferredDecRefs{,Locked}. See comment on Filesystem.mu. +// +// Precondition: d must not already be pending destruction. +func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { + fs.droppedDentriesMu.Lock() + fs.droppedDentries = append(fs.droppedDentries, d) + fs.droppedDentriesMu.Unlock() +} + +// processDeferredDecRefs calls vfs.Dentry.DecRef on all dentries in the +// droppedDentries list. See comment on Filesystem.mu. +func (fs *Filesystem) processDeferredDecRefs() { + fs.mu.Lock() + fs.processDeferredDecRefsLocked() + fs.mu.Unlock() +} + +// Precondition: fs.mu must be held for writing. +func (fs *Filesystem) processDeferredDecRefsLocked() { + fs.droppedDentriesMu.Lock() + for _, d := range fs.droppedDentries { + d.DecRef() + } + fs.droppedDentries = fs.droppedDentries[:0] // Keep slice memory for reuse. + fs.droppedDentriesMu.Unlock() +} + +// Init initializes a kernfs filesystem. This should be called from during +// vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding +// kernfs. +func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem) { + fs.vfsfs.Init(vfsObj, fs) +} + +// VFSFilesystem returns the generic vfs filesystem object. +func (fs *Filesystem) VFSFilesystem() *vfs.Filesystem { + return &fs.vfsfs +} + +// NextIno allocates a new inode number on this filesystem. +func (fs *Filesystem) NextIno() uint64 { + return atomic.AddUint64(&fs.nextInoMinusOne, 1) +} + +// These consts are used in the Dentry.flags field. +const ( + // Dentry points to a directory inode. + dflagsIsDir = 1 << iota + + // Dentry points to a symlink inode. + dflagsIsSymlink +) + +// Dentry implements vfs.DentryImpl. +// +// A kernfs dentry is similar to a dentry in a traditional filesystem: it's a +// named reference to an inode. A dentry generally lives as long as it's part of +// a mounted filesystem tree. Kernfs doesn't cache dentries once all references +// to them are removed. Dentries hold a single reference to the inode they point +// to, and child dentries hold a reference on their parent. +// +// Must be initialized by Init prior to first use. +type Dentry struct { + refs.AtomicRefCount + + vfsd vfs.Dentry + inode Inode + + refs uint64 + + // flags caches useful information about the dentry from the inode. See the + // dflags* consts above. Must be accessed by atomic ops. + flags uint32 + + // dirMu protects vfsd.children for directory dentries. + dirMu sync.Mutex +} + +// Init initializes this dentry. +// +// Precondition: Caller must hold a reference on inode. +// +// Postcondition: Caller's reference on inode is transferred to the dentry. +func (d *Dentry) Init(inode Inode) { + d.vfsd.Init(d) + d.inode = inode + ftype := inode.Mode().FileType() + if ftype == linux.ModeDirectory { + d.flags |= dflagsIsDir + } + if ftype == linux.ModeSymlink { + d.flags |= dflagsIsSymlink + } +} + +// VFSDentry returns the generic vfs dentry for this kernfs dentry. +func (d *Dentry) VFSDentry() *vfs.Dentry { + return &d.vfsd +} + +// isDir checks whether the dentry points to a directory inode. +func (d *Dentry) isDir() bool { + return atomic.LoadUint32(&d.flags)&dflagsIsDir != 0 +} + +// isSymlink checks whether the dentry points to a symlink inode. +func (d *Dentry) isSymlink() bool { + return atomic.LoadUint32(&d.flags)&dflagsIsSymlink != 0 +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *Dentry) DecRef() { + d.AtomicRefCount.DecRefWithDestructor(d.destroy) +} + +// Precondition: Dentry must be removed from VFS' dentry cache. +func (d *Dentry) destroy() { + d.inode.DecRef() // IncRef from Init. + d.inode = nil + if parent := d.vfsd.Parent(); parent != nil { + parent.DecRef() // IncRef from Dentry.InsertChild. + } +} + +// InsertChild inserts child into the vfs dentry cache with the given name under +// this dentry. This does not update the directory inode, so calling this on +// it's own isn't sufficient to insert a child into a directory. InsertChild +// updates the link count on d if required. +// +// Precondition: d must represent a directory inode. +func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { + if !d.isDir() { + panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) + } + vfsDentry := d.VFSDentry() + vfsDentry.IncRef() // DecRef in child's Dentry.destroy. + d.dirMu.Lock() + vfsDentry.InsertChild(child, name) + d.dirMu.Unlock() +} + +// The Inode interface maps filesystem-level operations that operate on paths to +// equivalent operations on specific filesystem nodes. +// +// The interface methods are groups into logical categories as sub interfaces +// below. Generally, an implementation for each sub interface can be provided by +// embedding an appropriate type from inode_impl_utils.go. The sub interfaces +// are purely organizational. Methods declared directly in the main interface +// have no generic implementations, and should be explicitly provided by the +// client filesystem. +// +// Generally, implementations are not responsible for tasks that are common to +// all filesystems. These include: +// +// - Checking that dentries passed to methods are of the appropriate file type. +// - Checking permissions. +// - Updating link and reference counts. +// +// Specific responsibilities of implementations are documented below. +type Inode interface { + // Methods related to reference counting. A generic implementation is + // provided by InodeNoopRefCount. These methods are generally called by the + // equivalent Dentry methods. + inodeRefs + + // Methods related to node metadata. A generic implementation is provided by + // InodeAttrs. + inodeMetadata + + // Method for inodes that represent symlink. InodeNotSymlink provides a + // blanket implementation for all non-symlink inodes. + inodeSymlink + + // Method for inodes that represent directories. InodeNotDirectory provides + // a blanket implementation for all non-directory inodes. + inodeDirectory + + // Method for inodes that represent dynamic directories and their + // children. InodeNoDynamicLookup provides a blanket implementation for all + // non-dynamic-directory inodes. + inodeDynamicLookup + + // Open creates a file description for the filesystem object represented by + // this inode. The returned file description should hold a reference on the + // inode for its lifetime. + // + // Precondition: !rp.Done(). vfsd.Impl() must be a kernfs Dentry. + Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) +} + +type inodeRefs interface { + IncRef() + DecRef() + TryIncRef() bool + // Destroy is called when the inode reaches zero references. Destroy release + // all resources (references) on objects referenced by the inode, including + // any child dentries. + Destroy() +} + +type inodeMetadata interface { + // CheckPermissions checks that creds may access this inode for the + // requested access type, per the the rules of + // fs/namei.c:generic_permission(). + CheckPermissions(creds *auth.Credentials, atx vfs.AccessTypes) error + + // Mode returns the (struct stat)::st_mode value for this inode. This is + // separated from Stat for performance. + Mode() linux.FileMode + + // Stat returns the metadata for this inode. This corresponds to + // vfs.FilesystemImpl.StatAt. + Stat(fs *vfs.Filesystem) linux.Statx + + // SetStat updates the metadata for this inode. This corresponds to + // vfs.FilesystemImpl.SetStatAt. + SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error +} + +// Precondition: All methods in this interface may only be called on directory +// inodes. +type inodeDirectory interface { + // The New{File,Dir,Node,Symlink} methods below should return a new inode + // hashed into this inode. + // + // These inode constructors are inode-level operations rather than + // filesystem-level operations to allow client filesystems to mix different + // implementations based on the new node's location in the + // filesystem. + + // HasChildren returns true if the directory inode has any children. + HasChildren() bool + + // NewFile creates a new regular file inode. + NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) + + // NewDir creates a new directory inode. + NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) + + // NewLink creates a new hardlink to a specified inode in this + // directory. Implementations should create a new kernfs Dentry pointing to + // target, and update target's link count. + NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) + + // NewSymlink creates a new symbolic link inode. + NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) + + // NewNode creates a new filesystem node for a mknod syscall. + NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) + + // Unlink removes a child dentry from this directory inode. + Unlink(ctx context.Context, name string, child *vfs.Dentry) error + + // RmDir removes an empty child directory from this directory + // inode. Implementations must update the parent directory's link count, + // if required. Implementations are not responsible for checking that child + // is a directory, checking for an empty directory. + RmDir(ctx context.Context, name string, child *vfs.Dentry) error + + // Rename is called on the source directory containing an inode being + // renamed. child should point to the resolved child in the source + // directory. If Rename replaces a dentry in the destination directory, it + // should return the replaced dentry or nil otherwise. + // + // Precondition: Caller must serialize concurrent calls to Rename. + Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (replaced *vfs.Dentry, err error) +} + +type inodeDynamicLookup interface { + // Lookup should return an appropriate dentry if name should resolve to a + // child of this dynamic directory inode. This gives the directory an + // opportunity on every lookup to resolve additional entries that aren't + // hashed into the directory. This is only called when the inode is a + // directory. If the inode is not a directory, or if the directory only + // contains a static set of children, the implementer can unconditionally + // return an appropriate error (ENOTDIR and ENOENT respectively). + // + // The child returned by Lookup will be hashed into the VFS dentry tree. Its + // lifetime can be controlled by the filesystem implementation with an + // appropriate implementation of Valid. + // + // Lookup returns the child with an extra reference and the caller owns this + // reference. + Lookup(ctx context.Context, name string) (*vfs.Dentry, error) + + // Valid should return true if this inode is still valid, or needs to + // be resolved again by a call to Lookup. + Valid(ctx context.Context) bool +} + +type inodeSymlink interface { + // Readlink resolves the target of a symbolic link. If an inode is not a + // symlink, the implementation should return EINVAL. + Readlink(ctx context.Context) (string, error) +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go new file mode 100644 index 000000000..f78bb7b04 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -0,0 +1,423 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs_test + +import ( + "bytes" + "fmt" + "io" + "runtime" + "sync" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const defaultMode linux.FileMode = 01777 +const staticFileContent = "This is sample content for a static test file." + +// RootDentryFn is a generator function for creating the root dentry of a test +// filesystem. See newTestSystem. +type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry + +// TestSystem represents the context for a single test. +type TestSystem struct { + t *testing.T + ctx context.Context + creds *auth.Credentials + vfs *vfs.VirtualFilesystem + mns *vfs.MountNamespace + root vfs.VirtualDentry +} + +// newTestSystem sets up a minimal environment for running a test, including an +// instance of a test filesystem. Tests can control the contents of the +// filesystem by providing an appropriate rootFn, which should return a +// pre-populated root dentry. +func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { + ctx := contexttest.Context(t) + creds := auth.CredentialsFromContext(ctx) + v := vfs.New() + v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}) + mns, err := v.NewMountNamespace(ctx, creds, "", "testfs", &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("Failed to create testfs root mount: %v", err) + } + + s := &TestSystem{ + t: t, + ctx: ctx, + creds: creds, + vfs: v, + mns: mns, + root: mns.Root(), + } + runtime.SetFinalizer(s, func(s *TestSystem) { s.root.DecRef() }) + return s +} + +// PathOpAtRoot constructs a vfs.PathOperation for a path from the +// root of the test filesystem. +// +// Precondition: path should be relative path. +func (s *TestSystem) PathOpAtRoot(path string) vfs.PathOperation { + return vfs.PathOperation{ + Root: s.root, + Start: s.root, + Pathname: path, + } +} + +// GetDentryOrDie attempts to resolve a dentry referred to by the +// provided path operation. If unsuccessful, the test fails. +func (s *TestSystem) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { + vd, err := s.vfs.GetDentryAt(s.ctx, s.creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) + } + return vd +} + +func (s *TestSystem) ReadToEnd(fd *vfs.FileDescription) (string, error) { + buf := make([]byte, usermem.PageSize) + bufIOSeq := usermem.BytesIOSequence(buf) + opts := vfs.ReadOptions{} + + var content bytes.Buffer + for { + n, err := fd.Impl().Read(s.ctx, bufIOSeq, opts) + if n == 0 || err != nil { + if err == io.EOF { + err = nil + } + return content.String(), err + } + content.Write(buf[:n]) + } +} + +type fsType struct { + rootFn RootDentryFn +} + +type filesystem struct { + kernfs.Filesystem +} + +type file struct { + kernfs.DynamicBytesFile + content string +} + +func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { + f := &file{} + f.content = content + f.DynamicBytesFile.Init(creds, fs.NextIno(), f) + + d := &kernfs.Dentry{} + d.Init(f) + return d +} + +func (f *file) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%s", f.content) + return nil +} + +type attrs struct { + kernfs.InodeAttrs +} + +func (a *attrs) SetStat(fs *vfs.Filesystem, opt vfs.SetStatOptions) error { + return syserror.EPERM +} + +type readonlyDir struct { + attrs + kernfs.InodeNotSymlink + kernfs.InodeNoDynamicLookup + kernfs.InodeDirectoryNoNewChildren + + kernfs.OrderedChildren + dentry kernfs.Dentry +} + +func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &readonlyDir{} + dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + dir.dentry.Init(dir) + + dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) + + return &dir.dentry +} + +func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +type dir struct { + attrs + kernfs.InodeNotSymlink + kernfs.InodeNoDynamicLookup + + fs *filesystem + dentry kernfs.Dentry + kernfs.OrderedChildren +} + +func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &dir{} + dir.fs = fs + dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) + dir.dentry.Init(dir) + + dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) + + return &dir.dentry +} + +func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { + creds := auth.CredentialsFromContext(ctx) + dir := d.fs.newDir(creds, opts.Mode, nil) + dirVFSD := dir.VFSDentry() + if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { + dir.DecRef() + return nil, err + } + d.IncLinks(1) + return dirVFSD, nil +} + +func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { + creds := auth.CredentialsFromContext(ctx) + f := d.fs.newFile(creds, "") + fVFSD := f.VFSDentry() + if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { + f.DecRef() + return nil, err + } + return fVFSD, nil +} + +func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (fst *fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + fs := &filesystem{} + fs.Init(vfsObj) + root := fst.rootFn(creds, fs) + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// -------------------- Remainder of the file are test cases -------------------- + +func TestBasic(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "file1": fs.newFile(creds, staticFileContent), + }) + }) + sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() +} + +func TestMkdirGetDentry(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir1": fs.newDir(creds, 0755, nil), + }) + }) + + pop := sys.PathOpAtRoot("dir1/a new directory") + if err := sys.vfs.MkdirAt(sys.ctx, sys.creds, &pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { + t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) + } + sys.GetDentryOrDie(pop).DecRef() +} + +func TestReadStaticFile(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "file1": fs.newFile(creds, staticFileContent), + }) + }) + + pop := sys.PathOpAtRoot("file1") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + content, err := sys.ReadToEnd(fd) + if err != nil { + sys.t.Fatalf("Read failed: %v", err) + } + if diff := cmp.Diff(staticFileContent, content); diff != "" { + sys.t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) + } +} + +func TestCreateNewFileInStaticDir(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir1": fs.newDir(creds, 0755, nil), + }) + }) + + pop := sys.PathOpAtRoot("dir1/newfile") + opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, opts) + if err != nil { + sys.t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) + } + + // Close the file. The file should persist. + fd.DecRef() + + fd, err = sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) + } + fd.DecRef() +} + +// direntCollector provides an implementation for vfs.IterDirentsCallback for +// testing. It simply iterates to the end of a given directory FD and collects +// all dirents emitted by the callback. +type direntCollector struct { + mu sync.Mutex + dirents map[string]vfs.Dirent +} + +// Handle implements vfs.IterDirentsCallback.Handle. +func (d *direntCollector) Handle(dirent vfs.Dirent) bool { + d.mu.Lock() + if d.dirents == nil { + d.dirents = make(map[string]vfs.Dirent) + } + d.dirents[dirent.Name] = dirent + d.mu.Unlock() + return true +} + +// count returns the number of dirents currently in the collector. +func (d *direntCollector) count() int { + d.mu.Lock() + defer d.mu.Unlock() + return len(d.dirents) +} + +// contains checks whether the collector has a dirent with the given name and +// type. +func (d *direntCollector) contains(name string, typ uint8) error { + d.mu.Lock() + defer d.mu.Unlock() + dirent, ok := d.dirents[name] + if !ok { + return fmt.Errorf("No dirent named %q found", name) + } + if dirent.Type != typ { + return fmt.Errorf("Dirent named %q found, but was expecting type %d, got: %+v", name, typ, dirent) + } + return nil +} + +func TestDirFDReadWrite(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, nil) + }) + + pop := sys.PathOpAtRoot("/") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + // Read/Write should fail for directory FDs. + if _, err := fd.Read(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { + sys.t.Fatalf("Read for directory FD failed with unexpected error: %v", err) + } + if _, err := fd.Write(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EISDIR { + sys.t.Fatalf("Wrire for directory FD failed with unexpected error: %v", err) + } +} + +func TestDirFDIterDirents(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + // Fill root with nodes backed by various inode implementations. + "dir1": fs.newReadonlyDir(creds, 0755, nil), + "dir2": fs.newDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir3": fs.newDir(creds, 0755, nil), + }), + "file1": fs.newFile(creds, staticFileContent), + }) + }) + + pop := sys.PathOpAtRoot("/") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + collector := &direntCollector{} + if err := fd.IterDirents(sys.ctx, collector); err != nil { + sys.t.Fatalf("IterDirent failed: %v", err) + } + + // Root directory should contain ".", ".." and 3 children: + if collector.count() != 5 { + sys.t.Fatalf("IterDirent returned too many dirents") + } + for _, dirName := range []string{".", "..", "dir1", "dir2"} { + if err := collector.contains(dirName, linux.DT_DIR); err != nil { + sys.t.Fatalf("IterDirent had unexpected results: %v", err) + } + } + if err := collector.contains("file1", linux.DT_REG); err != nil { + sys.t.Fatalf("IterDirent had unexpected results: %v", err) + } + +} diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index bc5c0b591..0cc751eb8 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -32,6 +32,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/fspath", "//pkg/sentry/arch", "//pkg/sentry/context", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index ea6417ce7..4a7a94a52 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -394,7 +394,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { } defer mountPoint.DecRef() // Create and mount the submount. - if err := vfsObj.NewMount(ctx, creds, "", &pop, "memfs", &vfs.GetFilesystemOptions{}); err != nil { + if err := vfsObj.MountAt(ctx, creds, "", &pop, "memfs", &vfs.MountOptions{}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } filePathBuilder.WriteString(mountPointName) diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index 08a9cb8ef..1f2a5122a 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -19,6 +19,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -582,3 +583,10 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error inode.decLinksLocked() return nil } + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go index a3a870571..5bf527c80 100644 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -194,7 +194,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { readData := make([]byte, 1) dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) if err != syserror.ErrWouldBlock { t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) } @@ -207,7 +207,7 @@ func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { writeData := []byte(msg) src := usermem.BytesIOSequence(writeData) - bytesWritten, err := fd.Impl().Write(ctx, src, vfs.WriteOptions{}) + bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) if err != nil { t.Fatalf("error writing to pipe %q: %v", fileName, err) } @@ -220,7 +220,7 @@ func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { readData := make([]byte, len(msg)) dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) if err != nil { t.Fatalf("error reading from pipe %q: %v", fileName, err) } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 28ba950bd..bd3fb4c03 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -841,9 +841,11 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, AbstractSocketNamespace: args.AbstractSocketNamespace, ContainerID: args.ContainerID, } - if _, err := k.tasks.NewTask(config); err != nil { + t, err := k.tasks.NewTask(config) + if err != nil { return nil, 0, err } + t.traceExecEvent(tc) // Simulate exec for tracing. // Success. tgid := k.tasks.Root.IDOfThreadGroup(tg) @@ -1118,6 +1120,22 @@ func (k *Kernel) SendContainerSignal(cid string, info *arch.SignalInfo) error { return lastErr } +// RebuildTraceContexts rebuilds the trace context for all tasks. +// +// Unfortunately, if these are built while tracing is not enabled, then we will +// not have meaningful trace data. Rebuilding here ensures that we can do so +// after tracing has been enabled. +func (k *Kernel) RebuildTraceContexts() { + k.extMu.Lock() + defer k.extMu.Unlock() + k.tasks.mu.RLock() + defer k.tasks.mu.RUnlock() + + for t, tid := range k.tasks.Root.tids { + t.rebuildTraceContext(tid) + } +} + // FeatureSet returns the FeatureSet. func (k *Kernel) FeatureSet() *cpuid.FeatureSet { return k.featureSet diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index 93fe68a3e..de9617e9d 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -302,7 +302,7 @@ func (s *Set) SetVal(ctx context.Context, num int32, val int16, creds *auth.Cred return syserror.ERANGE } - // TODO(b/29354920): Clear undo entries in all processes + // TODO(gvisor.dev/issue/137): Clear undo entries in all processes. sem.value = val sem.pid = pid s.changeTime = ktime.NowFromContext(ctx) @@ -336,7 +336,7 @@ func (s *Set) SetValAll(ctx context.Context, vals []uint16, creds *auth.Credenti for i, val := range vals { sem := &s.sems[i] - // TODO(b/29354920): Clear undo entries in all processes + // TODO(gvisor.dev/issue/137): Clear undo entries in all processes. sem.value = int16(val) sem.pid = pid sem.wakeWaiters() @@ -481,7 +481,7 @@ func (s *Set) executeOps(ctx context.Context, ops []linux.Sembuf, pid int32) (ch } // All operations succeeded, apply them. - // TODO(b/29354920): handle undo operations. + // TODO(gvisor.dev/issue/137): handle undo operations. for i, v := range tmpVals { s.sems[i].value = v s.sems[i].wakeWaiters() diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 220fa73a2..2fdee0282 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -339,6 +339,14 @@ func (s *SyscallTable) Lookup(sysno uintptr) SyscallFn { return nil } +// LookupName looks up a syscall name. +func (s *SyscallTable) LookupName(sysno uintptr) string { + if sc, ok := s.Table[sysno]; ok { + return sc.Name + } + return fmt.Sprintf("sys_%d", sysno) // Unlikely. +} + // LookupEmulate looks up an emulation syscall number. func (s *SyscallTable) LookupEmulate(addr usermem.Addr) (uintptr, bool) { sysno, ok := s.Emulate[addr] diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 80c8e5464..ab0c6c4aa 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -15,6 +15,8 @@ package kernel import ( + gocontext "context" + "runtime/trace" "sync" "sync/atomic" @@ -390,7 +392,14 @@ type Task struct { // logPrefix is a string containing the task's thread ID in the root PID // namespace, and is prepended to log messages emitted by Task.Infof etc. - logPrefix atomic.Value `state:".(string)"` + logPrefix atomic.Value `state:"nosave"` + + // traceContext and traceTask are both used for tracing, and are + // updated along with the logPrefix in updateInfoLocked. + // + // These are exclusive to the task goroutine. + traceContext gocontext.Context `state:"nosave"` + traceTask *trace.Task `state:"nosave"` // creds is the task's credentials. // @@ -528,14 +537,6 @@ func (t *Task) loadPtraceTracer(tracer *Task) { t.ptraceTracer.Store(tracer) } -func (t *Task) saveLogPrefix() string { - return t.logPrefix.Load().(string) -} - -func (t *Task) loadLogPrefix(prefix string) { - t.logPrefix.Store(prefix) -} - func (t *Task) saveSyscallFilters() []bpf.Program { if f := t.syscallFilters.Load(); f != nil { return f.([]bpf.Program) @@ -549,6 +550,7 @@ func (t *Task) loadSyscallFilters(filters []bpf.Program) { // afterLoad is invoked by stateify. func (t *Task) afterLoad() { + t.updateInfoLocked() t.interruptChan = make(chan struct{}, 1) t.gosched.State = TaskGoroutineNonexistent if t.stop != nil { diff --git a/pkg/sentry/kernel/task_block.go b/pkg/sentry/kernel/task_block.go index dd69939f9..4a4a69ee2 100644 --- a/pkg/sentry/kernel/task_block.go +++ b/pkg/sentry/kernel/task_block.go @@ -16,6 +16,7 @@ package kernel import ( "runtime" + "runtime/trace" "time" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -133,19 +134,24 @@ func (t *Task) block(C <-chan struct{}, timerChan <-chan struct{}) error { runtime.Gosched() } + region := trace.StartRegion(t.traceContext, blockRegion) select { case <-C: + region.End() t.SleepFinish(true) + // Woken by event. return nil case <-interrupt: + region.End() t.SleepFinish(false) // Return the indicated error on interrupt. return syserror.ErrInterrupted case <-timerChan: - // We've timed out. + region.End() t.SleepFinish(true) + // We've timed out. return syserror.ETIMEDOUT } } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 0916fd658..3eadfedb4 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -299,6 +299,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { // nt that it must receive before its task goroutine starts running. tid := nt.k.tasks.Root.IDOfTask(nt) defer nt.Start(tid) + t.traceCloneEvent(tid) // "If fork/clone and execve are allowed by @prog, any child processes will // be constrained to the same filters and system call ABI as the parent." - diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 17a089b90..90a6190f1 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -129,6 +129,7 @@ type runSyscallAfterExecStop struct { } func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { + t.traceExecEvent(r.tc) t.tg.pidns.owner.mu.Lock() t.tg.execing = nil if t.killed() { @@ -253,7 +254,7 @@ func (t *Task) promoteLocked() { t.tg.leader = t t.Infof("Becoming TID %d (in root PID namespace)", t.tg.pidns.owner.Root.tids[t]) - t.updateLogPrefixLocked() + t.updateInfoLocked() // Reap the original leader. If it has a tracer, detach it instead of // waiting for it to acknowledge the original leader's death. oldLeader.exitParentNotified = true diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 535f03e50..435761e5a 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -236,6 +236,7 @@ func (*runExit) execute(t *Task) taskRunState { type runExitMain struct{} func (*runExitMain) execute(t *Task) taskRunState { + t.traceExitEvent() lastExiter := t.exitThreadGroup() // If the task has a cleartid, and the thread group wasn't killed by a diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index a29e9b9eb..0fb3661de 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -16,6 +16,7 @@ package kernel import ( "fmt" + "runtime/trace" "sort" "gvisor.dev/gvisor/pkg/log" @@ -127,11 +128,88 @@ func (t *Task) debugDumpStack() { } } -// updateLogPrefix updates the task's cached log prefix to reflect its -// current thread ID. +// trace definitions. +// +// Note that all region names are prefixed by ':' in order to ensure that they +// are lexically ordered before all system calls, which use the naked system +// call name (e.g. "read") for maximum clarity. +const ( + traceCategory = "task" + runRegion = ":run" + blockRegion = ":block" + cpuidRegion = ":cpuid" + faultRegion = ":fault" +) + +// updateInfoLocked updates the task's cached log prefix and tracing +// information to reflect its current thread ID. // // Preconditions: The task's owning TaskSet.mu must be locked. -func (t *Task) updateLogPrefixLocked() { +func (t *Task) updateInfoLocked() { // Use the task's TID in the root PID namespace for logging. - t.logPrefix.Store(fmt.Sprintf("[% 4d] ", t.tg.pidns.owner.Root.tids[t])) + tid := t.tg.pidns.owner.Root.tids[t] + t.logPrefix.Store(fmt.Sprintf("[% 4d] ", tid)) + t.rebuildTraceContext(tid) +} + +// rebuildTraceContext rebuilds the trace context. +// +// Precondition: the passed tid must be the tid in the root namespace. +func (t *Task) rebuildTraceContext(tid ThreadID) { + // Re-initialize the trace context. + if t.traceTask != nil { + t.traceTask.End() + } + + // Note that we define the "task type" to be the dynamic TID. This does + // not align perfectly with the documentation for "tasks" in the + // tracing package. Tasks may be assumed to be bounded by analysis + // tools. However, if we just use a generic "task" type here, then the + // "user-defined tasks" page on the tracing dashboard becomes nearly + // unusable, as it loads all traces from all tasks. + // + // We can assume that the number of tasks in the system is not + // arbitrarily large (in general it won't be, especially for cases + // where we're collecting a brief profile), so using the TID is a + // reasonable compromise in this case. + t.traceContext, t.traceTask = trace.NewTask(t, fmt.Sprintf("tid:%d", tid)) +} + +// traceCloneEvent is called when a new task is spawned. +// +// ntid must be the new task's ThreadID in the root namespace. +func (t *Task) traceCloneEvent(ntid ThreadID) { + if !trace.IsEnabled() { + return + } + trace.Logf(t.traceContext, traceCategory, "spawn: %d", ntid) +} + +// traceExitEvent is called when a task exits. +func (t *Task) traceExitEvent() { + if !trace.IsEnabled() { + return + } + trace.Logf(t.traceContext, traceCategory, "exit status: 0x%x", t.exitStatus.Status()) +} + +// traceExecEvent is called when a task calls exec. +func (t *Task) traceExecEvent(tc *TaskContext) { + if !trace.IsEnabled() { + return + } + d := tc.MemoryManager.Executable() + if d == nil { + trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") + return + } + defer d.DecRef() + root := t.fsContext.RootDirectory() + if root == nil { + trace.Logf(t.traceContext, traceCategory, "exec: << no root directory >>") + return + } + defer root.DecRef() + n, _ := d.FullName(root) + trace.Logf(t.traceContext, traceCategory, "exec: %s", n) } diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index c92266c59..d97f8c189 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -17,6 +17,7 @@ package kernel import ( "bytes" "runtime" + "runtime/trace" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -205,9 +206,11 @@ func (*runApp) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.RUnlock() } + region := trace.StartRegion(t.traceContext, runRegion) t.accountTaskGoroutineEnter(TaskGoroutineRunningApp) info, at, err := t.p.Switch(t.MemoryManager().AddressSpace(), t.Arch(), t.rseqCPU) t.accountTaskGoroutineLeave(TaskGoroutineRunningApp) + region.End() if clearSinglestep { t.Arch().ClearSingleStep() @@ -225,6 +228,7 @@ func (*runApp) execute(t *Task) taskRunState { case platform.ErrContextSignalCPUID: // Is this a CPUID instruction? + region := trace.StartRegion(t.traceContext, cpuidRegion) expected := arch.CPUIDInstruction[:] found := make([]byte, len(expected)) _, err := t.CopyIn(usermem.Addr(t.Arch().IP()), &found) @@ -232,10 +236,12 @@ func (*runApp) execute(t *Task) taskRunState { // Skip the cpuid instruction. t.Arch().CPUIDEmulate(t) t.Arch().SetIP(t.Arch().IP() + uintptr(len(expected))) + region.End() // Resume execution. return (*runApp)(nil) } + region.End() // Not an actual CPUID, but required copy-in. // The instruction at the given RIP was not a CPUID, and we // fallthrough to the default signal deliver behavior below. @@ -251,8 +257,10 @@ func (*runApp) execute(t *Task) taskRunState { // an application-generated signal and we should continue execution // normally. if at.Any() { + region := trace.StartRegion(t.traceContext, faultRegion) addr := usermem.Addr(info.Addr()) err := t.MemoryManager().HandleUserFault(t, addr, at, usermem.Addr(t.Arch().Stack())) + region.End() if err == nil { // The fault was handled appropriately. // We can resume running the application. @@ -260,6 +268,12 @@ func (*runApp) execute(t *Task) taskRunState { } // Is this a vsyscall that we need emulate? + // + // Note that we don't track vsyscalls as part of a + // specific trace region. This is because regions don't + // stack, and the actual system call will count as a + // region. We should be able to easily identify + // vsyscalls by having a <fault><syscall> pair. if at.Execute { if sysno, ok := t.tc.st.LookupEmulate(addr); ok { return t.doVsyscall(addr, sysno) diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index ae6fc4025..3522a4ae5 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -154,10 +154,10 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { // Below this point, newTask is expected not to fail (there is no rollback // of assignTIDsLocked or any of the following). - // Logging on t's behalf will panic if t.logPrefix hasn't been initialized. - // This is the earliest point at which we can do so (since t now has thread - // IDs). - t.updateLogPrefixLocked() + // Logging on t's behalf will panic if t.logPrefix hasn't been + // initialized. This is the earliest point at which we can do so + // (since t now has thread IDs). + t.updateInfoLocked() if cfg.InheritParent != nil { t.parent = cfg.InheritParent.parent diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index b543d536a..3180f5560 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -17,6 +17,7 @@ package kernel import ( "fmt" "os" + "runtime/trace" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -160,6 +161,10 @@ func (t *Task) executeSyscall(sysno uintptr, args arch.SyscallArguments) (rval u ctrl = ctrlStopAndReinvokeSyscall } else { fn := s.Lookup(sysno) + var region *trace.Region // Only non-nil if tracing == true. + if trace.IsEnabled() { + region = trace.StartRegion(t.traceContext, s.LookupName(sysno)) + } if fn != nil { // Call our syscall implementation. rval, ctrl, err = fn(t, args) @@ -167,6 +172,9 @@ func (t *Task) executeSyscall(sysno uintptr, args arch.SyscallArguments) (rval u // Use the missing function if not found. rval, err = t.SyscallTable().Missing(t, sysno, args) } + if region != nil { + region.End() + } } if bits.IsOn32(fe, ExternalAfterEnable) && (s.ExternalFilterAfter == nil || s.ExternalFilterAfter(t, sysno, args)) { diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go index 34f84487a..048de26dc 100644 --- a/pkg/sentry/kernel/tty.go +++ b/pkg/sentry/kernel/tty.go @@ -21,8 +21,19 @@ import "sync" // // +stateify savable type TTY struct { + // Index is the terminal index. It is immutable. + Index uint32 + mu sync.Mutex `state:"nosave"` // tg is protected by mu. tg *ThreadGroup } + +// TTY returns the thread group's controlling terminal. If nil, there is no +// controlling terminal. +func (tg *ThreadGroup) TTY() *TTY { + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() + return tg.tty +} diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index c2c3ec06e..6299a3e2f 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -408,6 +408,8 @@ func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, info el start = vaddr } if vaddr < end { + // NOTE(b/37474556): Linux allows out-of-order + // segments, in violation of the spec. ctx.Infof("PT_LOAD headers out-of-order. %#x < %#x", vaddr, end) return loadedELF{}, syserror.ENOEXEC } diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD index 3ef84245b..112794e9c 100644 --- a/pkg/sentry/memmap/BUILD +++ b/pkg/sentry/memmap/BUILD @@ -41,7 +41,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", - "//pkg/refs", "//pkg/sentry/context", "//pkg/sentry/platform", "//pkg/sentry/usermem", diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index 03b99aaea..16a722a13 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -18,7 +18,6 @@ package memmap import ( "fmt" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -235,8 +234,11 @@ type InvalidateOpts struct { // coincidental; fs.File implements MappingIdentity, and some // fs.InodeOperations implement Mappable.) type MappingIdentity interface { - // MappingIdentity is reference-counted. - refs.RefCounter + // IncRef increments the MappingIdentity's reference count. + IncRef() + + // DecRef decrements the MappingIdentity's reference count. + DecRef() // MappedName returns the application-visible name shown in // /proc/[pid]/maps. diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 6803d488c..5c52d4007 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -12,18 +12,26 @@ go_library( "bluepill_amd64.go", "bluepill_amd64.s", "bluepill_amd64_unsafe.go", + "bluepill_arm64.go", + "bluepill_arm64.s", + "bluepill_arm64_unsafe.go", "bluepill_fault.go", "bluepill_unsafe.go", "context.go", - "filters.go", + "filters_amd64.go", + "filters_arm64.go", "kvm.go", "kvm_amd64.go", "kvm_amd64_unsafe.go", + "kvm_arm64.go", + "kvm_arm64_unsafe.go", "kvm_const.go", + "kvm_const_arm64.go", "machine.go", "machine_amd64.go", "machine_amd64_unsafe.go", "machine_arm64.go", + "machine_arm64_unsafe.go", "machine_unsafe.go", "physical_map.go", "virtual_map.go", diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 043de51b3..30dbb74d6 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -20,6 +20,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" ) @@ -36,6 +37,18 @@ func sighandler() func dieTrampoline() var ( + // bounceSignal is the signal used for bouncing KVM. + // + // We use SIGCHLD because it is not masked by the runtime, and + // it will be ignored properly by other parts of the kernel. + bounceSignal = syscall.SIGCHLD + + // bounceSignalMask has only bounceSignal set. + bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) + + // bounce is the interrupt vector used to return to the kernel. + bounce = uint32(ring0.VirtualizationException) + // savedHandler is a pointer to the previous handler. // // This is called by bluepillHandler. @@ -45,6 +58,13 @@ var ( dieTrampolineAddr uintptr ) +// redpill invokes a syscall with -1. +// +//go:nosplit +func redpill() { + syscall.RawSyscall(^uintptr(0), 0, 0, 0) +} + // dieHandler is called by dieTrampoline. // //go:nosplit @@ -73,8 +93,8 @@ func (c *vCPU) die(context *arch.SignalContext64, msg string) { func init() { // Install the handler. - if err := safecopy.ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { - panic(fmt.Sprintf("Unable to set handler for signal %d: %v", syscall.SIGSEGV, err)) + if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { + panic(fmt.Sprintf("Unable to set handler for signal %d: %v", bluepillSignal, err)) } // Extract the address for the trampoline. diff --git a/pkg/sentry/platform/kvm/bluepill_amd64.go b/pkg/sentry/platform/kvm/bluepill_amd64.go index 421c88220..133c2203d 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64.go @@ -24,26 +24,10 @@ import ( ) var ( - // bounceSignal is the signal used for bouncing KVM. - // - // We use SIGCHLD because it is not masked by the runtime, and - // it will be ignored properly by other parts of the kernel. - bounceSignal = syscall.SIGCHLD - - // bounceSignalMask has only bounceSignal set. - bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) - - // bounce is the interrupt vector used to return to the kernel. - bounce = uint32(ring0.VirtualizationException) + // The action for bluepillSignal is changed by sigaction(). + bluepillSignal = syscall.SIGSEGV ) -// redpill on amd64 invokes a syscall with -1. -// -//go:nosplit -func redpill() { - syscall.RawSyscall(^uintptr(0), 0, 0, 0) -} - // bluepillArchEnter is called during bluepillEnter. // //go:nosplit diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index 9d8af143e..a63a6a071 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -23,13 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) -// bluepillArchContext returns the arch-specific context. -// -//go:nosplit -func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { - return &((*arch.UContext64)(context).MContext) -} - // dieArchSetup initializes the state for dieTrampoline. // // The amd64 dieTrampoline requires the vCPU to be set in BX, and the last RIP diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go new file mode 100644 index 000000000..552341721 --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -0,0 +1,79 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" +) + +var ( + // The action for bluepillSignal is changed by sigaction(). + bluepillSignal = syscall.SIGILL +) + +// bluepillArchEnter is called during bluepillEnter. +// +//go:nosplit +func bluepillArchEnter(context *arch.SignalContext64) (c *vCPU) { + c = vCPUPtr(uintptr(context.Regs[8])) + regs := c.CPU.Registers() + regs.Regs = context.Regs + regs.Sp = context.Sp + regs.Pc = context.Pc + regs.Pstate = context.Pstate + regs.Pstate &^= uint64(ring0.KernelFlagsClear) + regs.Pstate |= ring0.KernelFlagsSet + return +} + +// bluepillArchExit is called during bluepillEnter. +// +//go:nosplit +func bluepillArchExit(c *vCPU, context *arch.SignalContext64) { + regs := c.CPU.Registers() + context.Regs = regs.Regs + context.Sp = regs.Sp + context.Pc = regs.Pc + context.Pstate = regs.Pstate + context.Pstate &^= uint64(ring0.UserFlagsClear) + context.Pstate |= ring0.UserFlagsSet +} + +// KernelSyscall handles kernel syscalls. +// +//go:nosplit +func (c *vCPU) KernelSyscall() { + regs := c.Registers() + if regs.Regs[8] != ^uint64(0) { + regs.Pc -= 4 // Rewind. + } + ring0.Halt() +} + +// KernelException handles kernel exceptions. +// +//go:nosplit +func (c *vCPU) KernelException(vector ring0.Vector) { + regs := c.Registers() + if vector == ring0.Vector(bounce) { + regs.Pc = 0 + } + ring0.Halt() +} diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.s b/pkg/sentry/platform/kvm/bluepill_arm64.s new file mode 100644 index 000000000..c61700892 --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64.s @@ -0,0 +1,87 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "textflag.h" + +// VCPU_CPU is the location of the CPU in the vCPU struct. +// +// This is guaranteed to be zero. +#define VCPU_CPU 0x0 + +// CPU_SELF is the self reference in ring0's percpu. +// +// This is guaranteed to be zero. +#define CPU_SELF 0x0 + +// Context offsets. +// +// Only limited use of the context is done in the assembly stub below, most is +// done in the Go handlers. +#define SIGINFO_SIGNO 0x0 +#define CONTEXT_PC 0x1B8 +#define CONTEXT_R0 0xB8 + +// See bluepill.go. +TEXT ·bluepill(SB),NOSPLIT,$0 +begin: + MOVD vcpu+0(FP), R8 + MOVD $VCPU_CPU(R8), R9 + ORR $0xffff000000000000, R9, R9 + // Trigger sigill. + // In ring0.Start(), the value of R8 will be stored into tpidr_el1. + // When the context was loaded into vcpu successfully, + // we will check if the value of R10 and R9 are the same. + WORD $0xd538d08a // MRS TPIDR_EL1, R10 +check_vcpu: + CMP R10, R9 + BEQ right_vCPU +wrong_vcpu: + CALL ·redpill(SB) + B begin +right_vCPU: + RET + +// sighandler: see bluepill.go for documentation. +// +// The arguments are the following: +// +// R0 - The signal number. +// R1 - Pointer to siginfo_t structure. +// R2 - Pointer to ucontext structure. +// +TEXT ·sighandler(SB),NOSPLIT,$0 + // si_signo should be sigill. + MOVD SIGINFO_SIGNO(R1), R7 + CMPW $4, R7 + BNE fallback + + MOVD CONTEXT_PC(R2), R7 + CMPW $0, R7 + BEQ fallback + + MOVD R2, 8(RSP) + BL ·bluepillHandler(SB) // Call the handler. + + RET + +fallback: + // Jump to the previous signal handler. + MOVD ·savedHandler(SB), R7 + B (R7) + +// dieTrampoline: see bluepill.go, bluepill_arm64_unsafe.go for documentation. +TEXT ·dieTrampoline(SB),NOSPLIT,$0 + // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. + MOVD R9, 8(RSP) + BL ·dieHandler(SB) diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go new file mode 100644 index 000000000..e5fac0d6a --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" +) + +//go:nosplit +func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { + // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. +} diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index ca011ef78..9add7c944 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -23,6 +23,8 @@ import ( "sync/atomic" "syscall" "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) //go:linkname throw runtime.throw @@ -49,6 +51,13 @@ func uintptrValue(addr *byte) uintptr { return (uintptr)(unsafe.Pointer(addr)) } +// bluepillArchContext returns the UContext64. +// +//go:nosplit +func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { + return &((*arch.UContext64)(context).MContext) +} + // bluepillHandler is called from the signal stub. // // The world may be stopped while this is executing, and it executes on the diff --git a/pkg/sentry/platform/kvm/filters.go b/pkg/sentry/platform/kvm/filters_amd64.go index 7d949f1dd..7d949f1dd 100644 --- a/pkg/sentry/platform/kvm/filters.go +++ b/pkg/sentry/platform/kvm/filters_amd64.go diff --git a/pkg/sentry/platform/kvm/filters_arm64.go b/pkg/sentry/platform/kvm/filters_arm64.go new file mode 100644 index 000000000..9245d07c2 --- /dev/null +++ b/pkg/sentry/platform/kvm/filters_arm64.go @@ -0,0 +1,32 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/seccomp" +) + +// SyscallFilters returns syscalls made exclusively by the KVM platform. +func (*KVM) SyscallFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + syscall.SYS_IOCTL: {}, + syscall.SYS_MMAP: {}, + syscall.SYS_RT_SIGSUSPEND: {}, + syscall.SYS_RT_SIGTIMEDWAIT: {}, + 0xffffffffffffffff: {}, // KVM uses syscall -1 to transition to host. + } +} diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index ee4cd2f4d..f2c2c059e 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -21,7 +21,6 @@ import ( "sync" "syscall" - "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" @@ -56,9 +55,7 @@ func New(deviceFile *os.File) (*KVM, error) { // Ensure global initialization is done. globalOnce.Do(func() { - physicalInit() - globalErr = updateSystemValues(int(fd)) - ring0.Init(cpuid.HostFeatureSet()) + globalErr = updateGlobalOnce(int(fd)) }) if globalErr != nil { return nil, globalErr diff --git a/pkg/sentry/platform/kvm/kvm_amd64.go b/pkg/sentry/platform/kvm/kvm_amd64.go index 5d8ef4761..c5a6f9c7d 100644 --- a/pkg/sentry/platform/kvm/kvm_amd64.go +++ b/pkg/sentry/platform/kvm/kvm_amd64.go @@ -17,6 +17,7 @@ package kvm import ( + "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) @@ -211,3 +212,11 @@ type cpuidEntries struct { _ uint32 entries [_KVM_NR_CPUID_ENTRIES]cpuidEntry } + +// updateGlobalOnce does global initialization. It has to be called only once. +func updateGlobalOnce(fd int) error { + physicalInit() + err := updateSystemValues(int(fd)) + ring0.Init(cpuid.HostFeatureSet()) + return err +} diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go new file mode 100644 index 000000000..2319c86d3 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -0,0 +1,83 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "syscall" +) + +// userMemoryRegion is a region of physical memory. +// +// This mirrors kvm_memory_region. +type userMemoryRegion struct { + slot uint32 + flags uint32 + guestPhysAddr uint64 + memorySize uint64 + userspaceAddr uint64 +} + +type kvmOneReg struct { + id uint64 + addr uint64 +} + +const KVM_NR_SPSR = 5 + +type userFpsimdState struct { + vregs [64]uint64 + fpsr uint32 + fpcr uint32 + reserved [2]uint32 +} + +type userRegs struct { + Regs syscall.PtraceRegs + sp_el1 uint64 + elr_el1 uint64 + spsr [KVM_NR_SPSR]uint64 + fpRegs userFpsimdState +} + +// runData is the run structure. This may be mapped for synchronous register +// access (although that doesn't appear to be supported by my kernel at least). +// +// This mirrors kvm_run. +type runData struct { + requestInterruptWindow uint8 + _ [7]uint8 + + exitReason uint32 + readyForInterruptInjection uint8 + ifFlag uint8 + _ [2]uint8 + + cr8 uint64 + apicBase uint64 + + // This is the union data for exits. Interpretation depends entirely on + // the exitReason above (see vCPU code for more information). + data [32]uint64 +} + +// updateGlobalOnce does global initialization. It has to be called only once. +func updateGlobalOnce(fd int) error { + physicalInit() + err := updateSystemValues(int(fd)) + updateVectorTable() + return err +} diff --git a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go new file mode 100644 index 000000000..6531bae1d --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go @@ -0,0 +1,39 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "fmt" + "syscall" +) + +var ( + runDataSize int +) + +func updateSystemValues(fd int) error { + // Extract the mmap size. + sz, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(fd), _KVM_GET_VCPU_MMAP_SIZE, 0) + if errno != 0 { + return fmt.Errorf("getting VCPU mmap size: %v", errno) + } + // Save the data. + runDataSize = int(sz) + + // Success. + return nil +} diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 766131d60..1d5c77ff4 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -49,11 +49,13 @@ const ( _KVM_EXIT_SHUTDOWN = 0x8 _KVM_EXIT_FAIL_ENTRY = 0x9 _KVM_EXIT_INTERNAL_ERROR = 0x11 + _KVM_EXIT_SYSTEM_EVENT = 0x18 ) // KVM capability options. const ( - _KVM_CAP_MAX_VCPUS = 0x42 + _KVM_CAP_MAX_VCPUS = 0x42 + _KVM_CAP_ARM_VM_IPA_SIZE = 0xa5 ) // KVM limits. diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go new file mode 100644 index 000000000..5a74c6e36 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -0,0 +1,132 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +// KVM ioctls for Arm64. +const ( + _KVM_GET_ONE_REG = 0x4010aeab + _KVM_SET_ONE_REG = 0x4010aeac + + _KVM_ARM_PREFERRED_TARGET = 0x8020aeaf + _KVM_ARM_VCPU_INIT = 0x4020aeae + _KVM_ARM64_REGS_PSTATE = 0x6030000000100042 + _KVM_ARM64_REGS_SP_EL1 = 0x6030000000100044 + _KVM_ARM64_REGS_R0 = 0x6030000000100000 + _KVM_ARM64_REGS_R1 = 0x6030000000100002 + _KVM_ARM64_REGS_R2 = 0x6030000000100004 + _KVM_ARM64_REGS_R3 = 0x6030000000100006 + _KVM_ARM64_REGS_R8 = 0x6030000000100010 + _KVM_ARM64_REGS_R18 = 0x6030000000100024 + _KVM_ARM64_REGS_PC = 0x6030000000100040 + _KVM_ARM64_REGS_MAIR_EL1 = 0x603000000013c510 + _KVM_ARM64_REGS_TCR_EL1 = 0x603000000013c102 + _KVM_ARM64_REGS_TTBR0_EL1 = 0x603000000013c100 + _KVM_ARM64_REGS_TTBR1_EL1 = 0x603000000013c101 + _KVM_ARM64_REGS_SCTLR_EL1 = 0x603000000013c080 + _KVM_ARM64_REGS_CPACR_EL1 = 0x603000000013c082 + _KVM_ARM64_REGS_VBAR_EL1 = 0x603000000013c600 +) + +// Arm64: Architectural Feature Access Control Register EL1. +const ( + _FPEN_NOTRAP = 0x3 + _FPEN_SHIFT = 0x20 +) + +// Arm64: System Control Register EL1. +const ( + _SCTLR_M = 1 << 0 + _SCTLR_C = 1 << 2 + _SCTLR_I = 1 << 12 +) + +// Arm64: Translation Control Register EL1. +const ( + _TCR_IPS_40BITS = 2 << 32 // PA=40 + _TCR_IPS_48BITS = 5 << 32 // PA=48 + + _TCR_T0SZ_OFFSET = 0 + _TCR_T1SZ_OFFSET = 16 + _TCR_IRGN0_SHIFT = 8 + _TCR_IRGN1_SHIFT = 24 + _TCR_ORGN0_SHIFT = 10 + _TCR_ORGN1_SHIFT = 26 + _TCR_SH0_SHIFT = 12 + _TCR_SH1_SHIFT = 28 + _TCR_TG0_SHIFT = 14 + _TCR_TG1_SHIFT = 30 + + _TCR_T0SZ_VA48 = 64 - 48 // VA=48 + _TCR_T1SZ_VA48 = 64 - 48 // VA=48 + + _TCR_ASID16 = 1 << 36 + _TCR_TBI0 = 1 << 37 + + _TCR_TXSZ_VA48 = (_TCR_T0SZ_VA48 << _TCR_T0SZ_OFFSET) | (_TCR_T1SZ_VA48 << _TCR_T1SZ_OFFSET) + + _TCR_TG0_4K = 0 << _TCR_TG0_SHIFT // 4K + _TCR_TG0_64K = 1 << _TCR_TG0_SHIFT // 64K + + _TCR_TG1_4K = 2 << _TCR_TG1_SHIFT + + _TCR_TG_FLAGS = _TCR_TG0_4K | _TCR_TG1_4K + + _TCR_IRGN0_WBWA = 1 << _TCR_IRGN0_SHIFT + _TCR_IRGN1_WBWA = 1 << _TCR_IRGN1_SHIFT + _TCR_IRGN_WBWA = _TCR_IRGN0_WBWA | _TCR_IRGN1_WBWA + + _TCR_ORGN0_WBWA = 1 << _TCR_ORGN0_SHIFT + _TCR_ORGN1_WBWA = 1 << _TCR_ORGN1_SHIFT + + _TCR_ORGN_WBWA = _TCR_ORGN0_WBWA | _TCR_ORGN1_WBWA + + _TCR_SHARED = (3 << _TCR_SH0_SHIFT) | (3 << _TCR_SH1_SHIFT) + + _TCR_CACHE_FLAGS = _TCR_IRGN_WBWA | _TCR_ORGN_WBWA +) + +// Arm64: Memory Attribute Indirection Register EL1. +const ( + _MT_DEVICE_nGnRnE = 0 + _MT_DEVICE_nGnRE = 1 + _MT_DEVICE_GRE = 2 + _MT_NORMAL_NC = 3 + _MT_NORMAL = 4 + _MT_NORMAL_WT = 5 + _MT_EL1_INIT = (0 << _MT_DEVICE_nGnRnE) | (0x4 << _MT_DEVICE_nGnRE * 8) | (0xc << _MT_DEVICE_GRE * 8) | (0x44 << _MT_NORMAL_NC * 8) | (0xff << _MT_NORMAL * 8) | (0xbb << _MT_NORMAL_WT * 8) +) + +const ( + _KVM_ARM_VCPU_POWER_OFF = 0 // CPU is started in OFF state + _KVM_ARM_VCPU_PSCI_0_2 = 2 // CPU uses PSCI v0.2 +) + +// Arm64: Exception Syndrome Register EL1. +const ( + _ESR_ELx_FSC = 0x3F + + _ESR_SEGV_MAPERR_L0 = 0x4 + _ESR_SEGV_MAPERR_L1 = 0x5 + _ESR_SEGV_MAPERR_L2 = 0x6 + _ESR_SEGV_MAPERR_L3 = 0x7 + + _ESR_SEGV_ACCERR_L1 = 0x9 + _ESR_SEGV_ACCERR_L2 = 0xa + _ESR_SEGV_ACCERR_L3 = 0xb + + _ESR_SEGV_PEMERR_L1 = 0xd + _ESR_SEGV_PEMERR_L2 = 0xe + _ESR_SEGV_PEMERR_L3 = 0xf +) diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 61227cafb..7156c245f 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -135,3 +135,43 @@ func (c *vCPU) setSignalMask() error { } return nil } + +// setUserRegisters sets user registers in the vCPU. +func (c *vCPU) setUserRegisters(uregs *userRegs) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_REGS, + uintptr(unsafe.Pointer(uregs))); errno != 0 { + return fmt.Errorf("error setting user registers: %v", errno) + } + return nil +} + +// getUserRegisters reloads user registers in the vCPU. +// +// This is safe to call from a nosplit context. +// +//go:nosplit +func (c *vCPU) getUserRegisters(uregs *userRegs) syscall.Errno { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_REGS, + uintptr(unsafe.Pointer(uregs))); errno != 0 { + return errno + } + return 0 +} + +// setSystemRegisters sets system registers. +func (c *vCPU) setSystemRegisters(sregs *systemRegs) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SREGS, + uintptr(unsafe.Pointer(sregs))); errno != 0 { + return fmt.Errorf("error setting system registers: %v", errno) + } + return nil +} diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index b7e2cfb9d..7ae47f291 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -12,8 +12,38 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package kvm +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +type vCPUArchState struct { + // PCIDs is the set of PCIDs for this vCPU. + // + // This starts above fixedKernelPCID. + PCIDs *pagetables.PCIDs +} + +const ( + // fixedKernelPCID is a fixed kernel PCID used for the kernel page + // tables. We must start allocating user PCIDs above this in order to + // avoid any conflict (see below). + fixedKernelPCID = 1 + + // poolPCIDs is the number of PCIDs to record in the database. As this + // grows, assignment can take longer, since it is a simple linear scan. + // Beyond a relatively small number, there are likely few perform + // benefits, since the TLB has likely long since lost any translations + // from more than a few PCIDs past. + poolPCIDs = 8 +) + // Get all read-only physicalRegions. func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { var rdonlyRegions []region @@ -59,3 +89,95 @@ func availableRegionsForSetMem() (phyRegions []physicalRegion) { return phyRegions } + +// dropPageTables drops cached page table entries. +func (m *machine) dropPageTables(pt *pagetables.PageTables) { + m.mu.Lock() + defer m.mu.Unlock() + + // Clear from all PCIDs. + for _, c := range m.vCPUs { + c.PCIDs.Drop(pt) + } +} + +// nonCanonical generates a canonical address return. +// +//go:nosplit +func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + *info = arch.SignalInfo{ + Signo: signal, + Code: arch.SignalInfoKernel, + } + info.SetAddr(addr) // Include address. + return usermem.NoAccess, platform.ErrContextSignal +} + +// fault generates an appropriate fault return. +// +//go:nosplit +func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + faultAddr := c.GetFaultAddr() + code, user := c.ErrorCode() + + // Reset the pointed SignalInfo. + *info = arch.SignalInfo{Signo: signal} + info.SetAddr(uint64(faultAddr)) + + read := true + write := false + execute := true + + ret := code & _ESR_ELx_FSC + switch ret { + case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3: + info.Code = 1 //SEGV_MAPERR + read = false + write = true + execute = false + case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3: + info.Code = 2 // SEGV_ACCERR. + read = true + write = false + execute = false + default: + info.Code = 2 + } + + if !user { + read = true + write = false + execute = true + + } + accessType := usermem.AccessType{ + Read: read, + Write: write, + Execute: execute, + } + + return accessType, platform.ErrContextSignal +} + +// retryInGuest runs the given function in guest mode. +// +// If the function does not complete in guest mode (due to execution of a +// system call due to a GC stall, for example), then it will be retried. The +// given function must be idempotent as a result of the retry mechanism. +func (m *machine) retryInGuest(fn func()) { + c := m.Get() + defer m.Put(c) + for { + c.ClearErrorCode() // See below. + bluepill(c) // Force guest mode. + fn() // Execute the given function. + _, user := c.ErrorCode() + if user { + // If user is set, then we haven't bailed back to host + // mode via a kernel exception or system call. We + // consider the full function to have executed in guest + // mode and we can return. + break + } + } +} diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go new file mode 100644 index 000000000..3f2f97a6b --- /dev/null +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -0,0 +1,362 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "fmt" + "reflect" + "sync/atomic" + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// setMemoryRegion initializes a region. +// +// This may be called from bluepillHandler, and therefore returns an errno +// directly (instead of wrapping in an error) to avoid allocations. +// +//go:nosplit +func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr) syscall.Errno { + userRegion := userMemoryRegion{ + slot: uint32(slot), + flags: 0, + guestPhysAddr: uint64(physical), + memorySize: uint64(length), + userspaceAddr: uint64(virtual), + } + + // Set the region. + _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(m.fd), + _KVM_SET_USER_MEMORY_REGION, + uintptr(unsafe.Pointer(&userRegion))) + return errno +} + +type kvmVcpuInit struct { + target uint32 + features [7]uint32 +} + +var vcpuInit kvmVcpuInit + +// initArchState initializes architecture-specific state. +func (m *machine) initArchState() error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(m.fd), + _KVM_ARM_PREFERRED_TARGET, + uintptr(unsafe.Pointer(&vcpuInit))); errno != 0 { + panic(fmt.Sprintf("error setting KVM_ARM_PREFERRED_TARGET failed: %v", errno)) + } + return nil +} + +func getPageWithReflect(p uintptr) []byte { + return (*(*[0xFFFFFF]byte)(unsafe.Pointer(p & ^uintptr(syscall.Getpagesize()-1))))[:syscall.Getpagesize()] +} + +// Work around: move ring0.Vectors() into a specific address with 11-bits alignment. +// +// According to the design documentation of Arm64, +// the start address of exception vector table should be 11-bits aligned. +// Please see the code in linux kernel as reference: arch/arm64/kernel/entry.S +// But, we can't align a function's start address to a specific address by using golang. +// We have raised this question in golang community: +// https://groups.google.com/forum/m/#!topic/golang-dev/RPj90l5x86I +// This function will be removed when golang supports this feature. +// +// There are 2 jobs were implemented in this function: +// 1, move the start address of exception vector table into the specific address. +// 2, modify the offset of each instruction. +func updateVectorTable() { + fromLocation := reflect.ValueOf(ring0.Vectors).Pointer() + offset := fromLocation & (1<<11 - 1) + if offset != 0 { + offset = 1<<11 - offset + } + + toLocation := fromLocation + offset + page := getPageWithReflect(toLocation) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { + panic(err) + } + + page = getPageWithReflect(toLocation + 4096) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { + panic(err) + } + + // Move exception-vector-table into the specific address. + var entry *uint32 + var entryFrom *uint32 + for i := 1; i <= 0x800; i++ { + entry = (*uint32)(unsafe.Pointer(toLocation + 0x800 - uintptr(i))) + entryFrom = (*uint32)(unsafe.Pointer(fromLocation + 0x800 - uintptr(i))) + *entry = *entryFrom + } + + // The offset from the address of each unconditionally branch is changed. + // We should modify the offset of each instruction. + nums := []uint32{0x0, 0x80, 0x100, 0x180, 0x200, 0x280, 0x300, 0x380, 0x400, 0x480, 0x500, 0x580, 0x600, 0x680, 0x700, 0x780} + for _, num := range nums { + entry = (*uint32)(unsafe.Pointer(toLocation + uintptr(num))) + *entry = *entry - (uint32)(offset/4) + } + + page = getPageWithReflect(toLocation) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { + panic(err) + } + + page = getPageWithReflect(toLocation + 4096) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { + panic(err) + } +} + +// initArchState initializes architecture-specific state. +func (c *vCPU) initArchState() error { + var ( + reg kvmOneReg + data uint64 + regGet kvmOneReg + dataGet uint64 + ) + + reg.addr = uint64(reflect.ValueOf(&data).Pointer()) + regGet.addr = uint64(reflect.ValueOf(&dataGet).Pointer()) + + vcpuInit.features[0] |= (1 << _KVM_ARM_VCPU_PSCI_0_2) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_ARM_VCPU_INIT, + uintptr(unsafe.Pointer(&vcpuInit))); errno != 0 { + panic(fmt.Sprintf("error setting KVM_ARM_VCPU_INIT failed: %v", errno)) + } + + // cpacr_el1 + reg.id = _KVM_ARM64_REGS_CPACR_EL1 + data = (_FPEN_NOTRAP << _FPEN_SHIFT) + if err := c.setOneRegister(®); err != nil { + return err + } + + // sctlr_el1 + regGet.id = _KVM_ARM64_REGS_SCTLR_EL1 + if err := c.getOneRegister(®Get); err != nil { + return err + } + + dataGet |= (_SCTLR_M | _SCTLR_C | _SCTLR_I) + data = dataGet + reg.id = _KVM_ARM64_REGS_SCTLR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // tcr_el1 + data = _TCR_TXSZ_VA48 | _TCR_CACHE_FLAGS | _TCR_SHARED | _TCR_TG_FLAGS | _TCR_ASID16 | _TCR_IPS_40BITS + reg.id = _KVM_ARM64_REGS_TCR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // mair_el1 + data = _MT_EL1_INIT + reg.id = _KVM_ARM64_REGS_MAIR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // ttbr0_el1 + data = c.machine.kernel.PageTables.TTBR0_EL1(false, 0) + + reg.id = _KVM_ARM64_REGS_TTBR0_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + c.SetTtbr0Kvm(uintptr(data)) + + // ttbr1_el1 + data = c.machine.kernel.PageTables.TTBR1_EL1(false, 0) + + reg.id = _KVM_ARM64_REGS_TTBR1_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // sp_el1 + data = c.CPU.StackTop() + reg.id = _KVM_ARM64_REGS_SP_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // pc + reg.id = _KVM_ARM64_REGS_PC + data = uint64(reflect.ValueOf(ring0.Start).Pointer()) + if err := c.setOneRegister(®); err != nil { + return err + } + + // r8 + reg.id = _KVM_ARM64_REGS_R8 + data = uint64(reflect.ValueOf(&c.CPU).Pointer()) + if err := c.setOneRegister(®); err != nil { + return err + } + + // vbar_el1 + reg.id = _KVM_ARM64_REGS_VBAR_EL1 + + fromLocation := reflect.ValueOf(ring0.Vectors).Pointer() + offset := fromLocation & (1<<11 - 1) + if offset != 0 { + offset = 1<<11 - offset + } + + toLocation := fromLocation + offset + data = uint64(ring0.KernelStartAddress | toLocation) + if err := c.setOneRegister(®); err != nil { + return err + } + + data = ring0.PsrDefaultSet | ring0.KernelFlagsSet + reg.id = _KVM_ARM64_REGS_PSTATE + if err := c.setOneRegister(®); err != nil { + return err + } + + return nil +} + +//go:nosplit +func (c *vCPU) loadSegments(tid uint64) { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // Get TLS from tpidr_el0. + atomic.StoreUint64(&c.tid, tid) +} + +func (c *vCPU) setOneRegister(reg *kvmOneReg) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_ONE_REG, + uintptr(unsafe.Pointer(reg))); errno != 0 { + return fmt.Errorf("error setting one register: %v", errno) + } + return nil +} + +func (c *vCPU) getOneRegister(reg *kvmOneReg) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_ONE_REG, + uintptr(unsafe.Pointer(reg))); errno != 0 { + return fmt.Errorf("error setting one register: %v", errno) + } + return nil +} + +// setCPUID sets the CPUID to be used by the guest. +func (c *vCPU) setCPUID() error { + return nil +} + +// setSystemTime sets the TSC for the vCPU. +func (c *vCPU) setSystemTime() error { + return nil +} + +// setSignalMask sets the vCPU signal mask. +// +// This must be called prior to running the vCPU. +func (c *vCPU) setSignalMask() error { + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) + data.mask2 = ^uint32(bounceSignalMask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + return fmt.Errorf("error setting signal mask: %v", errno) + } + + return nil +} + +// SwitchToUser unpacks architectural-details. +func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) (usermem.AccessType, error) { + // Check for canonical addresses. + if regs := switchOpts.Registers; !ring0.IsCanonical(regs.Pc) { + return nonCanonical(regs.Pc, int32(syscall.SIGSEGV), info) + } else if !ring0.IsCanonical(regs.Sp) { + return nonCanonical(regs.Sp, int32(syscall.SIGBUS), info) + } + + var vector ring0.Vector + ttbr0App := switchOpts.PageTables.TTBR0_EL1(false, 0) + c.SetTtbr0App(uintptr(ttbr0App)) + + // TODO(gvisor.dev/issue/1238): full context-switch supporting for Arm64. + // The Arm64 user-mode execution state consists of: + // x0-x30 + // PC, SP, PSTATE + // V0-V31: 32 128-bit registers for floating point, and simd + // FPSR + // TPIDR_EL0, used for TLS + appRegs := switchOpts.Registers + c.SetAppAddr(ring0.KernelStartAddress | uintptr(unsafe.Pointer(appRegs))) + + entersyscall() + bluepill(c) + vector = c.CPU.SwitchToUser(switchOpts) + exitsyscall() + + switch vector { + case ring0.Syscall: + // Fast path: system call executed. + return usermem.NoAccess, nil + + case ring0.PageFault: + return c.fault(int32(syscall.SIGSEGV), info) + case 0xaa: + return usermem.NoAccess, nil + default: + return usermem.NoAccess, platform.ErrContextSignal + } + +} diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index ed9433311..f04be2ab5 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -87,46 +87,6 @@ func unmapRunData(r *runData) error { return nil } -// setUserRegisters sets user registers in the vCPU. -func (c *vCPU) setUserRegisters(uregs *userRegs) error { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_REGS, - uintptr(unsafe.Pointer(uregs))); errno != 0 { - return fmt.Errorf("error setting user registers: %v", errno) - } - return nil -} - -// getUserRegisters reloads user registers in the vCPU. -// -// This is safe to call from a nosplit context. -// -//go:nosplit -func (c *vCPU) getUserRegisters(uregs *userRegs) syscall.Errno { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_GET_REGS, - uintptr(unsafe.Pointer(uregs))); errno != 0 { - return errno - } - return 0 -} - -// setSystemRegisters sets system registers. -func (c *vCPU) setSystemRegisters(sregs *systemRegs) error { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SREGS, - uintptr(unsafe.Pointer(sregs))); errno != 0 { - return fmt.Errorf("error setting system registers: %v", errno) - } - return nil -} - // atomicAddressSpace is an atomic address space pointer. type atomicAddressSpace struct { pointer unsafe.Pointer diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index 4649a94a7..a55cff507 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -21,6 +21,8 @@ import ( "strings" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -143,3 +145,43 @@ func (t *thread) adjustInitRegsRip() { func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.R15 = uint64(ppid) } + +// patchSignalInfo patches the signal info to account for hitting the seccomp +// filters from vsyscall emulation, specified below. We allow for SIGSYS as a +// synchronous trap, but patch the structure to appear like a SIGSEGV with the +// Rip as the faulting address. +// +// Note that this should only be called after verifying that the signalInfo has +// been generated by the kernel. +func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { + if linux.Signal(signalInfo.Signo) == linux.SIGSYS { + signalInfo.Signo = int32(linux.SIGSEGV) + + // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered + // with the si_call_addr field pointing to the current RIP. This field + // aligns with the si_addr field for a SIGSEGV, so we don't need to touch + // anything there. We do need to unwind emulation however, so we set the + // instruction pointer to the faulting value, and "unpop" the stack. + regs.Rip = signalInfo.Addr() + regs.Rsp -= 8 + } +} + +// enableCpuidFault enable cpuid-faulting; this may fail on older kernels or hardware, +// so we just disregard the result. Host CPUID will be enabled. +func enableCpuidFault() { + syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) +} + +// appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. +// Ref attachedThread() for more detail. +func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { + return append(rules, seccomp.RuleSet{ + Rules: seccomp.SyscallRules{ + syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ + {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, + }, + }, + Action: linux.SECCOMP_RET_ALLOW, + }) +} diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index bec884ba5..aed34e7ee 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -17,8 +17,12 @@ package ptrace import ( + "fmt" + "strings" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -37,7 +41,7 @@ const ( // resetSysemuRegs sets up emulation registers. // // This should be called prior to calling sysemu. -func (s *subprocess) resetSysemuRegs(regs *syscall.PtraceRegs) { +func (t *thread) resetSysemuRegs(regs *syscall.PtraceRegs) { } // createSyscallRegs sets up syscall registers. @@ -124,3 +128,34 @@ func (t *thread) adjustInitRegsRip() { func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.Regs[7] = uint64(ppid) } + +// patchSignalInfo patches the signal info to account for hitting the seccomp +// filters from vsyscall emulation, specified below. We allow for SIGSYS as a +// synchronous trap, but patch the structure to appear like a SIGSEGV with the +// Rip as the faulting address. +// +// Note that this should only be called after verifying that the signalInfo has +// been generated by the kernel. +func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { + if linux.Signal(signalInfo.Signo) == linux.SIGSYS { + signalInfo.Signo = int32(linux.SIGSEGV) + + // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered + // with the si_call_addr field pointing to the current RIP. This field + // aligns with the si_addr field for a SIGSEGV, so we don't need to touch + // anything there. We do need to unwind emulation however, so we set the + // instruction pointer to the faulting value, and "unpop" the stack. + regs.Pc = signalInfo.Addr() + regs.Sp -= 8 + } +} + +// Noop on arm64. +func enableCpuidFault() { +} + +// appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. +// Ref attachedThread() for more detail. +func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { + return rules +} diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index 3782d4332..cf13ea5e4 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -20,6 +20,7 @@ import ( "fmt" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/procid" @@ -77,27 +78,6 @@ func probeSeccomp() bool { } } -// patchSignalInfo patches the signal info to account for hitting the seccomp -// filters from vsyscall emulation, specified below. We allow for SIGSYS as a -// synchronous trap, but patch the structure to appear like a SIGSEGV with the -// Rip as the faulting address. -// -// Note that this should only be called after verifying that the signalInfo has -// been generated by the kernel. -func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { - if linux.Signal(signalInfo.Signo) == linux.SIGSYS { - signalInfo.Signo = int32(linux.SIGSEGV) - - // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered - // with the si_call_addr field pointing to the current RIP. This field - // aligns with the si_addr field for a SIGSEGV, so we don't need to touch - // anything there. We do need to unwind emulation however, so we set the - // instruction pointer to the faulting value, and "unpop" the stack. - regs.Rip = signalInfo.Addr() - regs.Rsp -= 8 - } -} - // createStub creates a fresh stub processes. // // Precondition: the runtime OS thread must be locked. @@ -149,7 +129,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro Rules: seccomp.SyscallRules{ syscall.SYS_GETTIMEOFDAY: {}, syscall.SYS_TIME: {}, - 309: {}, // SYS_GETCPU. + unix.SYS_GETCPU: {}, // SYS_GETCPU was not defined in package syscall on amd64. }, Action: linux.SECCOMP_RET_TRAP, Vsyscall: true, @@ -173,10 +153,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro // For the initial process creation. syscall.SYS_WAIT4: {}, - syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ - {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, - }, - syscall.SYS_EXIT: {}, + syscall.SYS_EXIT: {}, // For the stub prctl dance (all). syscall.SYS_PRCTL: []seccomp.Rule{ @@ -196,6 +173,8 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro }, Action: linux.SECCOMP_RET_ALLOW, }) + + rules = appendArchSeccompRules(rules) } instrs, err := seccomp.BuildProgram(rules, defaultAction) if err != nil { @@ -267,9 +246,8 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro syscall.RawSyscall(syscall.SYS_EXIT, uintptr(errno), 0, 0) } - // Enable cpuid-faulting; this may fail on older kernels or hardware, - // so we just disregard the result. Host CPUID will be enabled. - syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) + // Enable cpuid-faulting. + enableCpuidFault() // Call the stub; should not return. stubCall(stubStart, ppid) diff --git a/pkg/sentry/sighandling/sighandling.go b/pkg/sentry/sighandling/sighandling.go index 2f65db70b..ba1f9043d 100644 --- a/pkg/sentry/sighandling/sighandling.go +++ b/pkg/sentry/sighandling/sighandling.go @@ -16,7 +16,6 @@ package sighandling import ( - "fmt" "os" "os/signal" "reflect" @@ -31,37 +30,25 @@ const numSignals = 32 // handleSignals listens for incoming signals and calls the given handler // function. // -// It starts when the start channel is closed, stops when the stop channel -// is closed, and closes done once it will no longer deliver signals to k. -func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), start, stop, done chan struct{}) { +// It stops when the stop channel is closed. The done channel is closed once it +// will no longer deliver signals to k. +func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), stop, done chan struct{}) { // Build a select case. - sc := []reflect.SelectCase{{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(start)}} + sc := []reflect.SelectCase{{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(stop)}} for _, sigchan := range sigchans { sc = append(sc, reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(sigchan)}) } - started := false for { // Wait for a notification. index, _, ok := reflect.Select(sc) - // Was it the start / stop channel? + // Was it the stop channel? if index == 0 { if !ok { - if !started { - // start channel; start forwarding and - // swap this case for the stop channel - // to select stop requests. - started = true - sc[0] = reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(stop)} - } else { - // stop channel; stop forwarding and - // clear this case so it is never - // selected again. - started = false - close(done) - sc[0].Chan = reflect.Value{} - } + // Stop forwarding and notify that it's done. + close(done) + return } continue } @@ -73,44 +60,17 @@ func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), start, // Otherwise, it was a signal on channel N. Index 0 represents the stop // channel, so index N represents the channel for signal N. - signal := linux.Signal(index) - - if !started { - // Kernel cannot receive signals, either because it is - // not ready yet or is shutting down. - // - // Kill ourselves if this signal would have killed the - // process before PrepareForwarding was called. i.e., all - // _SigKill signals; see Go - // src/runtime/sigtab_linux_generic.go. - // - // Otherwise ignore the signal. - // - // TODO(b/114489875): Drop in Go 1.12, which uses tgkill - // in runtime.raise. - switch signal { - case linux.SIGHUP, linux.SIGINT, linux.SIGTERM: - dieFromSignal(signal) - panic(fmt.Sprintf("Failed to die from signal %d", signal)) - default: - continue - } - } - - // Pass the signal to the handler. - handler(signal) + handler(linux.Signal(index)) } } -// PrepareHandler ensures that synchronous signals are passed to the given -// handler function and returns a callback that starts signal delivery, which -// itself returns a callback that stops signal handling. +// StartSignalForwarding ensures that synchronous signals are passed to the +// given handler function and returns a callback that stops signal delivery. // // Note that this function permanently takes over signal handling. After the // stop callback, signals revert to the default Go runtime behavior, which // cannot be overridden with external calls to signal.Notify. -func PrepareHandler(handler func(linux.Signal)) func() func() { - start := make(chan struct{}) +func StartSignalForwarding(handler func(linux.Signal)) func() { stop := make(chan struct{}) done := make(chan struct{}) @@ -128,13 +88,10 @@ func PrepareHandler(handler func(linux.Signal)) func() func() { signal.Notify(sigchan, syscall.Signal(sig)) } // Start up our listener. - go handleSignals(sigchans, handler, start, stop, done) // S/R-SAFE: synchronized by Kernel.extMu. + go handleSignals(sigchans, handler, stop, done) // S/R-SAFE: synchronized by Kernel.extMu. - return func() func() { - close(start) - return func() { - close(stop) - <-done - } + return func() { + close(stop) + <-done } } diff --git a/pkg/sentry/sighandling/sighandling_unsafe.go b/pkg/sentry/sighandling/sighandling_unsafe.go index c303435d5..1ebe22d34 100644 --- a/pkg/sentry/sighandling/sighandling_unsafe.go +++ b/pkg/sentry/sighandling/sighandling_unsafe.go @@ -15,8 +15,6 @@ package sighandling import ( - "fmt" - "runtime" "syscall" "unsafe" @@ -48,27 +46,3 @@ func IgnoreChildStop() error { return nil } - -// dieFromSignal kills the current process with sig. -// -// Preconditions: The default action of sig is termination. -func dieFromSignal(sig linux.Signal) { - runtime.LockOSThread() - defer runtime.UnlockOSThread() - - sa := sigaction{handler: linux.SIG_DFL} - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, linux.SignalSetSize, 0, 0); e != 0 { - panic(fmt.Sprintf("rt_sigaction failed: %v", e)) - } - - set := linux.MakeSignalSet(sig) - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGPROCMASK, linux.SIG_UNBLOCK, uintptr(unsafe.Pointer(&set)), 0, linux.SignalSetSize, 0, 0); e != 0 { - panic(fmt.Sprintf("rt_sigprocmask failed: %v", e)) - } - - if err := syscall.Tgkill(syscall.Getpid(), syscall.Gettid(), syscall.Signal(sig)); err != nil { - panic(fmt.Sprintf("tgkill failed: %v", err)) - } - - panic("failed to die") -} diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 782a3cb92..af1a4e95f 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -195,15 +195,15 @@ func putCmsg(buf []byte, flags int, msgType uint32, align uint, data []int32) ([ // the available space, we must align down. // // align must be >= 4 and each data int32 is 4 bytes. The length of the - // header is already aligned, so if we align to the with of the data there + // header is already aligned, so if we align to the width of the data there // are two cases: // 1. The aligned length is less than the length of the header. The // unaligned length was also less than the length of the header, so we // can't write anything. // 2. The aligned length is greater than or equal to the length of the - // header. We can write the header plus zero or more datas. We can't write - // a partial int32, so the length of the message will be - // min(aligned length, header + datas). + // header. We can write the header plus zero or more bytes of data. We can't + // write a partial int32, so the length of the message will be + // min(aligned length, header + data). if space < linux.SizeOfControlMessageHeader { flags |= linux.MSG_CTRUNC return buf, flags @@ -240,12 +240,12 @@ func putCmsgStruct(buf []byte, msgLevel, msgType uint32, align uint, data interf buf = binary.Marshal(buf, usermem.ByteOrder, data) - // Check if we went over. + // If the control message data brought us over capacity, omit it. if cap(buf) != cap(ob) { return hdrBuf } - // Fix up length. + // Update control message length to include data. putUint64(ob, uint64(len(buf)-len(ob))) return alignSlice(buf, align) @@ -348,43 +348,62 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { ) } -func addSpaceForCmsg(cmsgDataLen int, buf []byte) []byte { - newBuf := make([]byte, 0, len(buf)+linux.SizeOfControlMessageHeader+cmsgDataLen) - return append(newBuf, buf...) -} - -// PackControlMessages converts the given ControlMessages struct into a buffer. +// PackControlMessages packs control messages into the given buffer. +// // We skip control messages specific to Unix domain sockets. -func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages) []byte { - var buf []byte - // The use of t.Arch().Width() is analogous to Linux's use of sizeof(long) in - // CMSG_ALIGN. - width := t.Arch().Width() - +// +// Note that some control messages may be truncated if they do not fit under +// the capacity of buf. +func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byte) []byte { if cmsgs.IP.HasTimestamp { - buf = addSpaceForCmsg(int(width), buf) buf = PackTimestamp(t, cmsgs.IP.Timestamp, buf) } if cmsgs.IP.HasInq { // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageInq, width), buf) buf = PackInq(t, cmsgs.IP.Inq, buf) } if cmsgs.IP.HasTOS { - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTOS, width), buf) buf = PackTOS(t, cmsgs.IP.TOS, buf) } if cmsgs.IP.HasTClass { - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTClass, width), buf) buf = PackTClass(t, cmsgs.IP.TClass, buf) } return buf } +// cmsgSpace is equivalent to CMSG_SPACE in Linux. +func cmsgSpace(t *kernel.Task, dataLen int) int { + return linux.SizeOfControlMessageHeader + AlignUp(dataLen, t.Arch().Width()) +} + +// CmsgsSpace returns the number of bytes needed to fit the control messages +// represented in cmsgs. +func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int { + space := 0 + + if cmsgs.IP.HasTimestamp { + space += cmsgSpace(t, linux.SizeOfTimeval) + } + + if cmsgs.IP.HasInq { + space += cmsgSpace(t, linux.SizeOfControlMessageInq) + } + + if cmsgs.IP.HasTOS { + space += cmsgSpace(t, linux.SizeOfControlMessageTOS) + } + + if cmsgs.IP.HasTClass { + space += cmsgSpace(t, linux.SizeOfControlMessageTClass) + } + + return space +} + // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index a8c152b54..c957b0f1d 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -45,7 +45,7 @@ const ( sizeofSockaddr = syscall.SizeofSockaddrInet6 // sizeof(sockaddr_in6) > sizeof(sockaddr_in) // maxControlLen is the maximum size of a control message buffer used in a - // recvmsg syscall. + // recvmsg or sendmsg syscall. maxControlLen = 1024 ) @@ -412,9 +412,12 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags msg.Namelen = uint32(len(senderAddrBuf)) } if controlLen > 0 { - controlBuf = make([]byte, maxControlLen) + if controlLen > maxControlLen { + controlLen = maxControlLen + } + controlBuf = make([]byte, controlLen) msg.Control = &controlBuf[0] - msg.Controllen = maxControlLen + msg.Controllen = controlLen } n, err := recvmsg(s.fd, &msg, sysflags) if err != nil { @@ -489,7 +492,14 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return 0, syserr.ErrInvalidArgument } - controlBuf := control.PackControlMessages(t, controlMessages) + space := uint64(control.CmsgsSpace(t, controlMessages)) + if space > maxControlLen { + space = maxControlLen + } + controlBuf := make([]byte, 0, space) + // PackControlMessages will append up to space bytes to controlBuf. + controlBuf = control.PackControlMessages(t, controlMessages, controlBuf) + sendmsgFromBlocks := safemem.WriterFunc(func(srcs safemem.BlockSeq) (uint64, error) { // Refuse to do anything if any part of src.Addrs was unusable. if uint64(src.NumBytes()) != srcs.NumBytes() { diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d92399efd..140851c17 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -151,6 +151,8 @@ var Metrics = tcpip.Stats{ PassiveConnectionOpenings: mustCreateMetric("/netstack/tcp/passive_connection_openings", "Number of connections opened successfully via Listen."), CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in either ESTABLISHED or CLOSE-WAIT state now."), EstablishedResets: mustCreateMetric("/netstack/tcp/established_resets", "Number of times TCP connections have made a direct transition to the CLOSED state from either the ESTABLISHED state or the CLOSE-WAIT state"), + EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "number of times established TCP connections made a transition to CLOSED state."), + EstablishedTimedout: mustCreateMetric("/netstack/tcp/established_timedout", "Number of times an established connection was reset because of keep-alive time out."), ListenOverflowSynDrop: mustCreateMetric("/netstack/tcp/listen_overflow_syn_drop", "Number of times the listen queue overflowed and a SYN was dropped."), ListenOverflowAckDrop: mustCreateMetric("/netstack/tcp/listen_overflow_ack_drop", "Number of times the listen queue overflowed and the final ACK in the handshake was dropped."), ListenOverflowSynCookieSent: mustCreateMetric("/netstack/tcp/listen_overflow_syn_cookie_sent", "Number of times a SYN cookie was sent."), @@ -324,7 +326,7 @@ func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, } family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { + if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported } @@ -1125,6 +1127,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_USER_TIMEOUT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPUserTimeoutOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Millisecond), nil + case linux.TCP_INFO: var v tcpip.TCPInfoOption if err := ep.GetSockOpt(&v); err != nil { @@ -1561,6 +1575,17 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveIntervalOption(time.Second * time.Duration(v)))) + case linux.TCP_USER_TIMEOUT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + return syserr.ErrInvalidArgument + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPUserTimeoutOption(time.Millisecond * time.Duration(v)))) + case linux.TCP_CONGESTION: v := tcpip.CongestionControlOption(optVal) if err := ep.SetSockOpt(v); err != nil { diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 8c250c325..2389a9cdb 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -43,6 +43,11 @@ type ControlMessages struct { IP tcpip.ControlMessages } +// Release releases Unix domain socket credentials and rights. +func (c *ControlMessages) Release() { + c.Unix.Release() +} + // Socket is the interface containing socket syscalls used by the syscall layer // to redirect them to the appropriate implementation. type Socket interface { diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 1aaae8487..885758054 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -118,6 +118,9 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { func extractPath(sockaddr []byte) (string, *syserr.Error) { addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) if err != nil { + if err == syserr.ErrAddressFamilyNotSupported { + err = syserr.ErrInvalidArgument + } return "", err } diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 72ebf766d..d46421199 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -14,6 +14,7 @@ go_library( "open.go", "poll.go", "ptrace.go", + "select.go", "signal.go", "socket.go", "strace.go", diff --git a/pkg/sentry/strace/linux64.go b/pkg/sentry/strace/linux64.go index 5d57b75af..e603f858f 100644 --- a/pkg/sentry/strace/linux64.go +++ b/pkg/sentry/strace/linux64.go @@ -40,7 +40,7 @@ var linuxAMD64 = SyscallMap{ 20: makeSyscallInfo("writev", FD, WriteIOVec, Hex), 21: makeSyscallInfo("access", Path, Oct), 22: makeSyscallInfo("pipe", PipeFDs), - 23: makeSyscallInfo("select", Hex, Hex, Hex, Hex, Timeval), + 23: makeSyscallInfo("select", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timeval), 24: makeSyscallInfo("sched_yield"), 25: makeSyscallInfo("mremap", Hex, Hex, Hex, Hex, Hex), 26: makeSyscallInfo("msync", Hex, Hex, Hex), @@ -287,7 +287,7 @@ var linuxAMD64 = SyscallMap{ 267: makeSyscallInfo("readlinkat", FD, Path, ReadBuffer, Hex), 268: makeSyscallInfo("fchmodat", FD, Path, Mode), 269: makeSyscallInfo("faccessat", FD, Path, Oct, Hex), - 270: makeSyscallInfo("pselect6", Hex, Hex, Hex, Hex, Hex, Hex), + 270: makeSyscallInfo("pselect6", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timespec, SigSet), 271: makeSyscallInfo("ppoll", PollFDs, Hex, Timespec, SigSet, Hex), 272: makeSyscallInfo("unshare", CloneFlags), 273: makeSyscallInfo("set_robust_list", Hex, Hex), @@ -335,5 +335,33 @@ var linuxAMD64 = SyscallMap{ 315: makeSyscallInfo("sched_getattr", Hex, Hex, Hex), 316: makeSyscallInfo("renameat2", FD, Path, Hex, Path, Hex), 317: makeSyscallInfo("seccomp", Hex, Hex, Hex), + 318: makeSyscallInfo("getrandom", Hex, Hex, Hex), + 319: makeSyscallInfo("memfd_create", Path, Hex), // Not quite a path, but close. + 320: makeSyscallInfo("kexec_file_load", FD, FD, Hex, Hex, Hex), + 321: makeSyscallInfo("bpf", Hex, Hex, Hex), + 322: makeSyscallInfo("execveat", FD, Path, ExecveStringVector, ExecveStringVector, Hex), + 323: makeSyscallInfo("userfaultfd", Hex), + 324: makeSyscallInfo("membarrier", Hex, Hex), + 325: makeSyscallInfo("mlock2", Hex, Hex, Hex), + 326: makeSyscallInfo("copy_file_range", FD, Hex, FD, Hex, Hex, Hex), + 327: makeSyscallInfo("preadv2", FD, ReadIOVec, Hex, Hex, Hex), + 328: makeSyscallInfo("pwritev2", FD, WriteIOVec, Hex, Hex, Hex), + 329: makeSyscallInfo("pkey_mprotect", Hex, Hex, Hex, Hex), + 330: makeSyscallInfo("pkey_alloc", Hex, Hex), + 331: makeSyscallInfo("pkey_free", Hex), 332: makeSyscallInfo("statx", FD, Path, Hex, Hex, Hex), + 333: makeSyscallInfo("io_pgetevents", Hex, Hex, Hex, Hex, Timespec, SigSet), + 334: makeSyscallInfo("rseq", Hex, Hex, Hex, Hex), + 424: makeSyscallInfo("pidfd_send_signal", FD, Signal, Hex, Hex), + 425: makeSyscallInfo("io_uring_setup", Hex, Hex), + 426: makeSyscallInfo("io_uring_enter", FD, Hex, Hex, Hex, SigSet, Hex), + 427: makeSyscallInfo("io_uring_register", FD, Hex, Hex, Hex), + 428: makeSyscallInfo("open_tree", FD, Path, Hex), + 429: makeSyscallInfo("move_mount", FD, Path, FD, Path, Hex), + 430: makeSyscallInfo("fsopen", Path, Hex), // Not quite a path, but close. + 431: makeSyscallInfo("fsconfig", FD, Hex, Hex, Hex, Hex), + 432: makeSyscallInfo("fsmount", FD, Hex, Hex), + 433: makeSyscallInfo("fspick", FD, Path, Hex), + 434: makeSyscallInfo("pidfd_open", Hex, Hex), + 435: makeSyscallInfo("clone3", Hex, Hex), } diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go new file mode 100644 index 000000000..dea309fda --- /dev/null +++ b/pkg/sentry/strace/select.go @@ -0,0 +1,53 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package strace + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +func fdsFromSet(t *kernel.Task, set []byte) []int { + var fds []int + // Append n if the n-th bit is 1. + for i, v := range set { + for j := 0; j < 8; j++ { + if (v>>j)&1 == 1 { + fds = append(fds, i*8+j) + } + } + } + return fds +} + +func fdSet(t *kernel.Task, nfds int, addr usermem.Addr) string { + if addr == 0 { + return "null" + } + + // Calculate the size of the fd set (one bit per fd). + nBytes := (nfds + 7) / 8 + nBitsInLastPartialByte := nfds % 8 + + set, err := linux.CopyInFDSet(t, addr, nBytes, nBitsInLastPartialByte) + if err != nil { + return fmt.Sprintf("%#x (error decoding fdset: %s)", addr, err) + } + + return fmt.Sprintf("%#x %v", addr, fdsFromSet(t, set)) +} diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 311389547..629c1f308 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -439,6 +439,8 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo output = append(output, capData(t, args[arg-1].Pointer(), args[arg].Pointer())) case PollFDs: output = append(output, pollFDs(t, args[arg].Pointer(), uint(args[arg+1].Uint()), false)) + case SelectFDSet: + output = append(output, fdSet(t, int(args[0].Int()), args[arg].Pointer())) case Oct: output = append(output, "0o"+strconv.FormatUint(args[arg].Uint64(), 8)) case Hex: diff --git a/pkg/sentry/strace/syscalls.go b/pkg/sentry/strace/syscalls.go index 3c389d375..e5d486c4e 100644 --- a/pkg/sentry/strace/syscalls.go +++ b/pkg/sentry/strace/syscalls.go @@ -206,6 +206,10 @@ const ( // PollFDs is an array of struct pollfd. The number of entries in the // array is in the next argument. PollFDs + + // SelectFDSet is an fd_set argument in select(2)/pselect(2). The number of + // fds represented must be the first argument. + SelectFDSet ) // defaultFormat is the syscall argument format to use if the actual format is diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 4c0bf96e4..6766ba587 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -49,6 +49,7 @@ go_library( "sys_tls.go", "sys_utsname.go", "sys_write.go", + "sys_xattr.go", "timespec.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/syscalls/linux", diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 81e4f93a6..272ae9991 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,10 +228,10 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 188: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 191: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), @@ -260,7 +260,7 @@ var AMD64 = &kernel.SyscallTable{ 217: syscalls.Supported("getdents64", Getdents64), 218: syscalls.Supported("set_tid_address", SetTidAddress), 219: syscalls.Supported("restart_syscall", RestartSyscall), - 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), 222: syscalls.Supported("timer_create", TimerCreate), 223: syscalls.Supported("timer_settime", TimerSettime), @@ -367,11 +367,31 @@ var AMD64 = &kernel.SyscallTable{ 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - // Syscalls after 325 are "backports" from versions of Linux after 4.4. + // Syscalls implemented after 325 are "backports" from versions + // of Linux after 4.4. 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 327: syscalls.Supported("preadv2", Preadv2), 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 329: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 330: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 331: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 332: syscalls.Supported("statx", Statx), + 333: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 334: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), }, Emulate: map[usermem.Addr]uintptr{ diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index f1dd4b0c0..3b584eed9 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -41,10 +41,10 @@ var ARM64 = &kernel.SyscallTable{ 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 5: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 8: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), @@ -224,7 +224,7 @@ var ARM64 = &kernel.SyscallTable{ 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) 190: syscalls.Supported("semget", Semget), 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), - 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), 195: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), @@ -302,7 +302,26 @@ var ARM64 = &kernel.SyscallTable{ 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 286: syscalls.Supported("preadv2", Preadv2), 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 288: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 289: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 290: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 291: syscalls.Supported("statx", Statx), + 292: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 293: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), }, Emulate: map[usermem.Addr]uintptr{}, diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 3b9181002..9bc2445a5 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -840,25 +840,42 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(newfd), nil, nil } -func fGetOwn(t *kernel.Task, file *fs.File) int32 { +func fGetOwnEx(t *kernel.Task, file *fs.File) linux.FOwnerEx { ma := file.Async(nil) if ma == nil { - return 0 + return linux.FOwnerEx{} } a := ma.(*fasync.FileAsync) ot, otg, opg := a.Owner() switch { case ot != nil: - return int32(t.PIDNamespace().IDOfTask(ot)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_TID, + PID: int32(t.PIDNamespace().IDOfTask(ot)), + } case otg != nil: - return int32(t.PIDNamespace().IDOfThreadGroup(otg)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_PID, + PID: int32(t.PIDNamespace().IDOfThreadGroup(otg)), + } case opg != nil: - return int32(-t.PIDNamespace().IDOfProcessGroup(opg)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_PGRP, + PID: int32(t.PIDNamespace().IDOfProcessGroup(opg)), + } default: - return 0 + return linux.FOwnerEx{} } } +func fGetOwn(t *kernel.Task, file *fs.File) int32 { + owner := fGetOwnEx(t, file) + if owner.Type == linux.F_OWNER_PGRP { + return -owner.PID + } + return owner.PID +} + // fSetOwn sets the file's owner with the semantics of F_SETOWN in Linux. // // If who is positive, it represents a PID. If negative, it represents a PGID. @@ -901,11 +918,13 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) + return 0, nil, nil case linux.F_GETFL: return uintptr(file.Flags().ToLinux()), nil, nil case linux.F_SETFL: flags := uint(args[2].Uint()) file.SetFlags(linuxToFlags(flags).Settable()) + return 0, nil, nil case linux.F_SETLK, linux.F_SETLKW: // In Linux the file system can choose to provide lock operations for an inode. // Normally pipe and socket types lack lock operations. We diverge and use a heavy @@ -1008,6 +1027,44 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_SETOWN: fSetOwn(t, file, args[2].Int()) return 0, nil, nil + case linux.F_GETOWN_EX: + addr := args[2].Pointer() + owner := fGetOwnEx(t, file) + _, err := t.CopyOut(addr, &owner) + return 0, nil, err + case linux.F_SETOWN_EX: + addr := args[2].Pointer() + var owner linux.FOwnerEx + n, err := t.CopyIn(addr, &owner) + if err != nil { + return 0, nil, err + } + a := file.Async(fasync.New).(*fasync.FileAsync) + switch owner.Type { + case linux.F_OWNER_TID: + task := t.PIDNamespace().TaskWithID(kernel.ThreadID(owner.PID)) + if task == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerTask(t, task) + return uintptr(n), nil, nil + case linux.F_OWNER_PID: + tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID)) + if tg == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerThreadGroup(t, tg) + return uintptr(n), nil, nil + case linux.F_OWNER_PGRP: + pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID)) + if pg == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerProcessGroup(t, pg) + return uintptr(n), nil, nil + default: + return 0, nil, syserror.EINVAL + } case linux.F_GET_SEALS: val, err := tmpfs.GetSeals(file.Dirent.Inode) return uintptr(val), nil, err @@ -1035,7 +1092,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Everything else is not yet supported. return 0, nil, syserror.EINVAL } - return 0, nil, nil } const ( diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 7a13beac2..2b2df989a 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -197,53 +197,51 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) return remainingTimeout, n, err } +// CopyInFDSet copies an fd set from select(2)/pselect(2). +func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialByte int) ([]byte, error) { + set := make([]byte, nBytes) + + if addr != 0 { + if _, err := t.CopyIn(addr, &set); err != nil { + return nil, err + } + // If we only use part of the last byte, mask out the extraneous bits. + // + // N.B. This only works on little-endian architectures. + if nBitsInLastPartialByte != 0 { + set[nBytes-1] &^= byte(0xff) << nBitsInLastPartialByte + } + } + return set, nil +} + func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Addr, timeout time.Duration) (uintptr, error) { if nfds < 0 || nfds > fileCap { return 0, syserror.EINVAL } - // Capture all the provided input vectors. - // - // N.B. This only works on little-endian architectures. - byteCount := (nfds + 7) / 8 - - bitsInLastPartialByte := uint(nfds % 8) - r := make([]byte, byteCount) - w := make([]byte, byteCount) - e := make([]byte, byteCount) + // Calculate the size of the fd sets (one bit per fd). + nBytes := (nfds + 7) / 8 + nBitsInLastPartialByte := nfds % 8 - if readFDs != 0 { - if _, err := t.CopyIn(readFDs, &r); err != nil { - return 0, err - } - // Mask out bits above nfds. - if bitsInLastPartialByte != 0 { - r[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + // Capture all the provided input vectors. + r, err := CopyInFDSet(t, readFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } - - if writeFDs != 0 { - if _, err := t.CopyIn(writeFDs, &w); err != nil { - return 0, err - } - if bitsInLastPartialByte != 0 { - w[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + w, err := CopyInFDSet(t, writeFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } - - if exceptFDs != 0 { - if _, err := t.CopyIn(exceptFDs, &e); err != nil { - return 0, err - } - if bitsInLastPartialByte != 0 { - e[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + e, err := CopyInFDSet(t, exceptFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } // Count how many FDs are actually being requested so that we can build // a PollFD array. fdCount := 0 - for i := 0; i < byteCount; i++ { + for i := 0; i < nBytes; i++ { v := r[i] | w[i] | e[i] for v != 0 { v &= (v - 1) @@ -254,7 +252,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Build the PollFD array. pfd := make([]linux.PollFD, 0, fdCount) var fd int32 - for i := 0; i < byteCount; i++ { + for i := 0; i < nBytes; i++ { rV, wV, eV := r[i], w[i], e[i] v := rV | wV | eV m := byte(1) @@ -295,8 +293,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add } // Do the syscall, then count the number of bits set. - _, _, err := pollBlock(t, pfd, timeout) - if err != nil { + if _, _, err = pollBlock(t, pfd, timeout); err != nil { return 0, syserror.ConvertIntr(err, syserror.EINTR) } diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index d8acae063..4b5aafcc0 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -764,7 +764,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Unix.Release() + cms.Release() } if int(msg.Flags) != mflags { @@ -784,32 +784,16 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Unix.Release() + defer cms.Release() controlData := make([]byte, 0, msg.ControlLen) + controlData = control.PackControlMessages(t, cms, controlData) if cr, ok := s.(transport.Credentialer); ok && cr.Passcred() { creds, _ := cms.Unix.Credentials.(control.SCMCredentials) controlData, mflags = control.PackCredentials(t, creds, controlData, mflags) } - if cms.IP.HasTimestamp { - controlData = control.PackTimestamp(t, cms.IP.Timestamp, controlData) - } - - if cms.IP.HasInq { - // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. - controlData = control.PackInq(t, cms.IP.Inq, controlData) - } - - if cms.IP.HasTOS { - controlData = control.PackTOS(t, cms.IP.TOS, controlData) - } - - if cms.IP.HasTClass { - controlData = control.PackTClass(t, cms.IP.TClass, controlData) - } - if cms.Unix.Rights != nil { controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) } @@ -885,7 +869,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Unix.Release() + cm.Release() if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -1071,7 +1055,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Unix.Release() + controlMessages.Release() } return uintptr(n), err } diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go new file mode 100644 index 000000000..97d9a65ea --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -0,0 +1,169 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +import ( + "strings" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Getxattr implements linux syscall getxattr(2). +func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + valueLen := 0 + err = fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + value, err := getxattr(t, d, dirPath, nameAddr) + if err != nil { + return err + } + + valueLen = len(value) + if size == 0 { + return nil + } + if size > linux.XATTR_SIZE_MAX { + size = linux.XATTR_SIZE_MAX + } + if valueLen > int(size) { + return syserror.ERANGE + } + + _, err = t.CopyOutBytes(valueAddr, []byte(value)) + return err + }) + if err != nil { + return 0, nil, err + } + return uintptr(valueLen), nil, nil +} + +// getxattr implements getxattr from the given *fs.Dirent. +func getxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr) (string, error) { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return "", syserror.ENOTDIR + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { + return "", err + } + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return "", err + } + + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return "", syserror.EOPNOTSUPP + } + + return d.Inode.Getxattr(name) +} + +// Setxattr implements linux syscall setxattr(2). +func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + flags := args[4].Uint() + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return 0, nil, syserror.EINVAL + } + + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + return setxattr(t, d, dirPath, nameAddr, valueAddr, size, flags) + }) +} + +// setxattr implements setxattr from the given *fs.Dirent. +func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint, flags uint32) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { + return err + } + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return err + } + + if size > linux.XATTR_SIZE_MAX { + return syserror.E2BIG + } + buf := make([]byte, size) + if _, err = t.CopyInBytes(valueAddr, buf); err != nil { + return err + } + value := string(buf) + + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + + return d.Inode.Setxattr(name, value) +} + +func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { + name, err := t.CopyInString(nameAddr, linux.XATTR_NAME_MAX+1) + if err != nil { + if err == syserror.ENAMETOOLONG { + return "", syserror.ERANGE + } + return "", err + } + if len(name) == 0 { + return "", syserror.ERANGE + } + return name, nil +} + +func checkXattrPermissions(t *kernel.Task, i *fs.Inode, perms fs.PermMask) error { + // Restrict xattrs to regular files and directories. + // + // In Linux, this restriction technically only applies to xattrs in the + // "user.*" namespace, but we don't allow any other xattr prefixes anyway. + if !fs.IsRegular(i.StableAttr) && !fs.IsDir(i.StableAttr) { + if perms.Write { + return syserror.EPERM + } + return syserror.ENODATA + } + + return i.CheckPermission(t, perms) +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 74a325309..e3e554b88 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -17,9 +17,9 @@ go_library( "mount.go", "mount_unsafe.go", "options.go", + "pathname.go", "permissions.go", "resolving_path.go", - "syscalls.go", "testutil.go", "vfs.go", ], diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index 32cf9151b..705194ebc 100644 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go @@ -24,6 +24,9 @@ type contextID int const ( // CtxMountNamespace is a Context.Value key for a MountNamespace. CtxMountNamespace contextID = iota + + // CtxRoot is a Context.Value key for a VFS root. + CtxRoot ) // MountNamespaceFromContext returns the MountNamespace used by ctx. It does @@ -35,3 +38,13 @@ func MountNamespaceFromContext(ctx context.Context) *MountNamespace { } return nil } + +// RootFromContext returns the VFS root used by ctx. It takes a reference on +// the returned VirtualDentry. If ctx does not have a specific VFS root, +// RootFromContext returns a zero-value VirtualDentry. +func RootFromContext(ctx context.Context) VirtualDentry { + if v := ctx.Value(CtxRoot); v != nil { + return v.(VirtualDentry) + } + return VirtualDentry{} +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 34007eb57..6575afd16 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -241,3 +241,140 @@ type IterDirentsCallback interface { // called. Handle(dirent Dirent) bool } + +// OnClose is called when a file descriptor representing the FileDescription is +// closed. Returning a non-nil error should not prevent the file descriptor +// from being closed. +func (fd *FileDescription) OnClose(ctx context.Context) error { + return fd.impl.OnClose(ctx) +} + +// StatusFlags returns file description status flags, as for fcntl(F_GETFL). +func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) { + flags, err := fd.impl.StatusFlags(ctx) + flags |= linux.O_LARGEFILE + return flags, err +} + +// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL). +func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { + return fd.impl.SetStatusFlags(ctx, flags) +} + +// Stat returns metadata for the file represented by fd. +func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { + return fd.impl.Stat(ctx, opts) +} + +// SetStat updates metadata for the file represented by fd. +func (fd *FileDescription) SetStat(ctx context.Context, opts SetStatOptions) error { + return fd.impl.SetStat(ctx, opts) +} + +// StatFS returns metadata for the filesystem containing the file represented +// by fd. +func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { + return fd.impl.StatFS(ctx) +} + +// PRead reads from the file represented by fd into dst, starting at the given +// offset, and returns the number of bytes read. PRead is permitted to return +// partial reads with a nil error. +func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { + return fd.impl.PRead(ctx, dst, offset, opts) +} + +// Read is similar to PRead, but does not specify an offset. +func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { + return fd.impl.Read(ctx, dst, opts) +} + +// PWrite writes src to the file represented by fd, starting at the given +// offset, and returns the number of bytes written. PWrite is permitted to +// return partial writes with a nil error. +func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + return fd.impl.PWrite(ctx, src, offset, opts) +} + +// Write is similar to PWrite, but does not specify an offset. +func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + return fd.impl.Write(ctx, src, opts) +} + +// IterDirents invokes cb on each entry in the directory represented by fd. If +// IterDirents has been called since the last call to Seek, it continues +// iteration from the end of the last call. +func (fd *FileDescription) IterDirents(ctx context.Context, cb IterDirentsCallback) error { + return fd.impl.IterDirents(ctx, cb) +} + +// Seek changes fd's offset (assuming one exists) and returns its new value. +func (fd *FileDescription) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return fd.impl.Seek(ctx, offset, whence) +} + +// Sync has the semantics of fsync(2). +func (fd *FileDescription) Sync(ctx context.Context) error { + return fd.impl.Sync(ctx) +} + +// ConfigureMMap mutates opts to implement mmap(2) for the file represented by +// fd. +func (fd *FileDescription) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return fd.impl.ConfigureMMap(ctx, opts) +} + +// Ioctl implements the ioctl(2) syscall. +func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return fd.impl.Ioctl(ctx, uio, args) +} + +// SyncFS instructs the filesystem containing fd to execute the semantics of +// syncfs(2). +func (fd *FileDescription) SyncFS(ctx context.Context) error { + return fd.vd.mount.fs.impl.Sync(ctx) +} + +// MappedName implements memmap.MappingIdentity.MappedName. +func (fd *FileDescription) MappedName(ctx context.Context) string { + vfsroot := RootFromContext(ctx) + s, _ := fd.vd.mount.vfs.PathnameWithDeleted(ctx, vfsroot, fd.vd) + if vfsroot.Ok() { + vfsroot.DecRef() + } + return s +} + +// DeviceID implements memmap.MappingIdentity.DeviceID. +func (fd *FileDescription) DeviceID() uint64 { + stat, err := fd.impl.Stat(context.Background(), StatOptions{ + // There is no STATX_DEV; we assume that Stat will return it if it's + // available regardless of mask. + Mask: 0, + // fs/proc/task_mmu.c:show_map_vma() just reads inode::i_sb->s_dev + // directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil { + return 0 + } + return uint64(linux.MakeDeviceID(uint16(stat.DevMajor), stat.DevMinor)) +} + +// InodeID implements memmap.MappingIdentity.InodeID. +func (fd *FileDescription) InodeID() uint64 { + stat, err := fd.impl.Stat(context.Background(), StatOptions{ + Mask: linux.STATX_INO, + // fs/proc/task_mmu.c:show_map_vma() just reads inode::i_ino directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil || stat.Mask&linux.STATX_INO == 0 { + return 0 + } + return stat.Ino +} + +// Msync implements memmap.MappingIdentity.Msync. +func (fd *FileDescription) Msync(ctx context.Context, mr memmap.MappableRange) error { + return fd.impl.Sync(ctx) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 4fbad7840..aae023254 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -252,3 +252,12 @@ func (fd *DynamicBytesFileDescriptionImpl) Seek(ctx context.Context, offset int6 fd.off = offset return offset, nil } + +// GenericConfigureMMap may be used by most implementations of +// FileDescriptionImpl.ConfigureMMap. +func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.MMapOpts) error { + opts.Mappable = m + opts.MappingIdentity = fd + fd.IncRef() + return nil +} diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index a5561dcbe..ac7799296 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -103,7 +103,7 @@ func TestGenCountFD(t *testing.T) { // The first read causes Generate to be called to fill the FD's buffer. buf := make([]byte, 2) ioseq := usermem.BytesIOSequence(buf) - n, err := fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err := fd.Read(ctx, ioseq, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("first Read: got (%d, %v), wanted (1, nil or EOF)", n, err) } @@ -112,17 +112,17 @@ func TestGenCountFD(t *testing.T) { } // A second read without seeking is still at EOF. - n, err = fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err = fd.Read(ctx, ioseq, ReadOptions{}) if n != 0 || err != io.EOF { t.Fatalf("second Read: got (%d, %v), wanted (0, EOF)", n, err) } // Seeking to the beginning of the file causes it to be regenerated. - n, err = fd.Impl().Seek(ctx, 0, linux.SEEK_SET) + n, err = fd.Seek(ctx, 0, linux.SEEK_SET) if n != 0 || err != nil { t.Fatalf("Seek: got (%d, %v), wanted (0, nil)", n, err) } - n, err = fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err = fd.Read(ctx, ioseq, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("Read after Seek: got (%d, %v), wanted (1, nil or EOF)", n, err) } @@ -131,7 +131,7 @@ func TestGenCountFD(t *testing.T) { } // PRead at the beginning of the file also causes it to be regenerated. - n, err = fd.Impl().PRead(ctx, ioseq, 0, ReadOptions{}) + n, err = fd.PRead(ctx, ioseq, 0, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("PRead: got (%d, %v), wanted (1, nil or EOF)", n, err) } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 76ff8cf51..8011eba3f 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" ) @@ -47,6 +48,9 @@ func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, impl FilesystemImpl) { fs.refs = 1 fs.vfs = vfsObj fs.impl = impl + vfsObj.filesystemsMu.Lock() + vfsObj.filesystems[fs] = struct{}{} + vfsObj.filesystemsMu.Unlock() } // VirtualFilesystem returns the containing VirtualFilesystem. @@ -66,9 +70,28 @@ func (fs *Filesystem) IncRef() { } } +// TryIncRef increments fs' reference count and returns true. If fs' reference +// count is zero, TryIncRef does nothing and returns false. +// +// TryIncRef does not require that a reference is held on fs. +func (fs *Filesystem) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&fs.refs) + if refs <= 0 { + return false + } + if atomic.CompareAndSwapInt64(&fs.refs, refs, refs+1) { + return true + } + } +} + // DecRef decrements fs' reference count. func (fs *Filesystem) DecRef() { if refs := atomic.AddInt64(&fs.refs, -1); refs == 0 { + fs.vfs.filesystemsMu.Lock() + delete(fs.vfs.filesystems, fs) + fs.vfs.filesystemsMu.Unlock() fs.impl.Release() } else if refs < 0 { panic("Filesystem.decRef() called without holding a reference") @@ -163,5 +186,56 @@ type FilesystemImpl interface { // UnlinkAt removes the non-directory file at rp. UnlinkAt(ctx context.Context, rp *ResolvingPath) error - // TODO: d_path(); extended attributes; inotify_add_watch(); bind() + // PrependPath prepends a path from vd to vd.Mount().Root() to b. + // + // If vfsroot.Ok(), it is the contextual VFS root; if it is encountered + // before vd.Mount().Root(), PrependPath should stop prepending path + // components and return a PrependPathAtVFSRootError. + // + // If traversal of vd.Dentry()'s ancestors encounters an independent + // ("root") Dentry that is not vd.Mount().Root() (i.e. vd.Dentry() is not a + // descendant of vd.Mount().Root()), PrependPath should stop prepending + // path components and return a PrependPathAtNonMountRootError. + // + // Filesystems for which Dentries do not have meaningful paths may prepend + // an arbitrary descriptive string to b and then return a + // PrependPathSyntheticError. + // + // Most implementations can acquire the appropriate locks to ensure that + // Dentry.Name() and Dentry.Parent() are fixed for vd.Dentry() and all of + // its ancestors, then call GenericPrependPath. + // + // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. + PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error + + // TODO: extended attributes; inotify_add_watch(); bind() +} + +// PrependPathAtVFSRootError is returned by implementations of +// FilesystemImpl.PrependPath() when they encounter the contextual VFS root. +type PrependPathAtVFSRootError struct{} + +// Error implements error.Error. +func (PrependPathAtVFSRootError) Error() string { + return "vfs.FilesystemImpl.PrependPath() reached VFS root" +} + +// PrependPathAtNonMountRootError is returned by implementations of +// FilesystemImpl.PrependPath() when they encounter an independent ancestor +// Dentry that is not the Mount root. +type PrependPathAtNonMountRootError struct{} + +// Error implements error.Error. +func (PrependPathAtNonMountRootError) Error() string { + return "vfs.FilesystemImpl.PrependPath() reached root other than Mount root" +} + +// PrependPathSyntheticError is returned by implementations of +// FilesystemImpl.PrependPath() for which prepended names do not represent real +// paths. +type PrependPathSyntheticError struct{} + +// Error implements error.Error. +func (PrependPathSyntheticError) Error() string { + return "vfs.FilesystemImpl.PrependPath() prepended synthetic name" } diff --git a/pkg/sentry/vfs/filesystem_impl_util.go b/pkg/sentry/vfs/filesystem_impl_util.go index 465e610e0..7315a588e 100644 --- a/pkg/sentry/vfs/filesystem_impl_util.go +++ b/pkg/sentry/vfs/filesystem_impl_util.go @@ -16,6 +16,8 @@ package vfs import ( "strings" + + "gvisor.dev/gvisor/pkg/fspath" ) // GenericParseMountOptions parses a comma-separated list of options of the @@ -41,3 +43,27 @@ func GenericParseMountOptions(str string) map[string]string { } return m } + +// GenericPrependPath may be used by implementations of +// FilesystemImpl.PrependPath() for which a single statically-determined lock +// or set of locks is sufficient to ensure its preconditions (as opposed to +// e.g. per-Dentry locks). +// +// Preconditions: Dentry.Name() and Dentry.Parent() must be held constant for +// vd.Dentry() and all of its ancestors. +func GenericPrependPath(vfsroot, vd VirtualDentry, b *fspath.Builder) error { + mnt, d := vd.mount, vd.dentry + for { + if mnt == vfsroot.mount && d == vfsroot.dentry { + return PrependPathAtVFSRootError{} + } + if d == mnt.root { + return nil + } + if d.parent == nil { + return PrependPathAtNonMountRootError{} + } + b.PrependComponent(d.name) + d = d.parent + } +} diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 1c3b2e987..ec23ab0dd 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -18,6 +18,7 @@ import ( "math" "sync/atomic" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" @@ -133,13 +134,13 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth return mntns, nil } -// NewMount creates and mounts a Filesystem configured by the given arguments. -func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *GetFilesystemOptions) error { +// MountAt creates and mounts a Filesystem configured by the given arguments. +func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { fsType := vfs.getFilesystemType(fsTypeName) if fsType == nil { return syserror.ENODEV } - fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, *opts) + fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions) if err != nil { return err } @@ -207,6 +208,68 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti return nil } +// UmountAt removes the Mount at the given path. +func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *UmountOptions) error { + if opts.Flags&^(linux.MNT_FORCE|linux.MNT_DETACH) != 0 { + return syserror.EINVAL + } + + // MNT_FORCE is currently unimplemented except for the permission check. + if opts.Flags&linux.MNT_FORCE != 0 && creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) { + return syserror.EPERM + } + + vd, err := vfs.GetDentryAt(ctx, creds, pop, &GetDentryOptions{}) + if err != nil { + return err + } + defer vd.DecRef() + if vd.dentry != vd.mount.root { + return syserror.EINVAL + } + vfs.mountMu.Lock() + if mntns := MountNamespaceFromContext(ctx); mntns != nil && mntns != vd.mount.ns { + vfs.mountMu.Unlock() + return syserror.EINVAL + } + + // TODO(jamieliu): Linux special-cases umount of the caller's root, which + // we don't implement yet (we'll just fail it since the caller holds a + // reference on it). + + vfs.mounts.seq.BeginWrite() + if opts.Flags&linux.MNT_DETACH == 0 { + if len(vd.mount.children) != 0 { + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + return syserror.EBUSY + } + // We are holding a reference on vd.mount. + expectedRefs := int64(1) + if !vd.mount.umounted { + expectedRefs = 2 + } + if atomic.LoadInt64(&vd.mount.refs)&^math.MinInt64 != expectedRefs { // mask out MSB + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + return syserror.EBUSY + } + } + vdsToDecRef, mountsToDecRef := vfs.umountRecursiveLocked(vd.mount, &umountRecursiveOptions{ + eager: opts.Flags&linux.MNT_DETACH == 0, + disconnectHierarchy: true, + }, nil, nil) + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + for _, vd := range vdsToDecRef { + vd.DecRef() + } + for _, mnt := range mountsToDecRef { + mnt.DecRef() + } + return nil +} + type umountRecursiveOptions struct { // If eager is true, ensure that future calls to Mount.tryIncMountedRef() // on umounted mounts fail. diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 3aa73d911..3ecbc8fc1 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -46,6 +46,12 @@ type MknodOptions struct { DevMinor uint32 } +// MountOptions contains options to VirtualFilesystem.MountAt(). +type MountOptions struct { + // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). + GetFilesystemOptions GetFilesystemOptions +} + // OpenOptions contains options to VirtualFilesystem.OpenAt() and // FilesystemImpl.OpenAt(). type OpenOptions struct { @@ -114,6 +120,12 @@ type StatOptions struct { Sync uint32 } +// UmountOptions contains options to VirtualFilesystem.UmountAt(). +type UmountOptions struct { + // Flags contains flags as specified for umount2(2). + Flags uint32 +} + // WriteOptions contains options to FileDescription.PWrite(), // FileDescriptionImpl.PWrite(), FileDescription.Write(), and // FileDescriptionImpl.Write(). diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go new file mode 100644 index 000000000..8e155654f --- /dev/null +++ b/pkg/sentry/vfs/pathname.go @@ -0,0 +1,153 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/syserror" +) + +var fspathBuilderPool = sync.Pool{ + New: func() interface{} { + return &fspath.Builder{} + }, +} + +func getFSPathBuilder() *fspath.Builder { + return fspathBuilderPool.Get().(*fspath.Builder) +} + +func putFSPathBuilder(b *fspath.Builder) { + // No methods can be called on b after b.String(), so reset it to its zero + // value (as returned by fspathBuilderPool.New) instead. + *b = fspath.Builder{} + fspathBuilderPool.Put(b) +} + +// PathnameWithDeleted returns an absolute pathname to vd, consistent with +// Linux's d_path(). In particular, if vd.Dentry() has been disowned, +// PathnameWithDeleted appends " (deleted)" to the returned pathname. +func (vfs *VirtualFilesystem) PathnameWithDeleted(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { + b := getFSPathBuilder() + defer putFSPathBuilder(b) + haveRef := false + defer func() { + if haveRef { + vd.DecRef() + } + }() + + origD := vd.dentry +loop: + for { + err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b) + switch err.(type) { + case nil: + if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { + // GenericPrependPath() will have returned + // PrependPathAtVFSRootError in this case since it checks + // against vfsroot before mnt.root, but other implementations + // of FilesystemImpl.PrependPath() may return nil instead. + break loop + } + nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + if !nextVD.Ok() { + break loop + } + if haveRef { + vd.DecRef() + } + vd = nextVD + haveRef = true + // continue loop + case PrependPathSyntheticError: + // Skip prepending "/" and appending " (deleted)". + return b.String(), nil + case PrependPathAtVFSRootError, PrependPathAtNonMountRootError: + break loop + default: + return "", err + } + } + b.PrependByte('/') + if origD.IsDisowned() { + b.AppendString(" (deleted)") + } + return b.String(), nil +} + +// PathnameForGetcwd returns an absolute pathname to vd, consistent with +// Linux's sys_getcwd(). +func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { + if vd.dentry.IsDisowned() { + return "", syserror.ENOENT + } + + b := getFSPathBuilder() + defer putFSPathBuilder(b) + haveRef := false + defer func() { + if haveRef { + vd.DecRef() + } + }() + unreachable := false +loop: + for { + err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b) + switch err.(type) { + case nil: + if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { + break loop + } + nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + if !nextVD.Ok() { + unreachable = true + break loop + } + if haveRef { + vd.DecRef() + } + vd = nextVD + haveRef = true + case PrependPathAtVFSRootError: + break loop + case PrependPathAtNonMountRootError, PrependPathSyntheticError: + unreachable = true + break loop + default: + return "", err + } + } + b.PrependByte('/') + if unreachable { + b.PrependString("(unreachable)") + } + return b.String(), nil +} + +// As of this writing, we do not have equivalents to: +// +// - d_absolute_path(), which returns EINVAL if (effectively) any call to +// FilesystemImpl.PrependPath() would return PrependPathAtNonMountRootError. +// +// - dentry_path(), which does not walk up mounts (and only returns the path +// relative to Filesystem root), but also appends "//deleted" for disowned +// Dentries. +// +// These should be added as necessary. diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f8e74355c..f1edb0680 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -119,3 +119,65 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { return false } } + +// CheckSetStat checks that creds has permission to change the metadata of a +// file with the given permissions, UID, and GID as specified by stat, subject +// to the rules of Linux's fs/attr.c:setattr_prepare(). +func CheckSetStat(creds *auth.Credentials, stat *linux.Statx, mode uint16, kuid auth.KUID, kgid auth.KGID) error { + if stat.Mask&linux.STATX_MODE != 0 { + if !CanActAsOwner(creds, kuid) { + return syserror.EPERM + } + // TODO(b/30815691): "If the calling process is not privileged (Linux: + // does not have the CAP_FSETID capability), and the group of the file + // does not match the effective group ID of the process or one of its + // supplementary group IDs, the S_ISGID bit will be turned off, but + // this will not cause an error to be returned." - chmod(2) + } + if stat.Mask&linux.STATX_UID != 0 { + if !((creds.EffectiveKUID == kuid && auth.KUID(stat.UID) == kuid) || + HasCapabilityOnFile(creds, linux.CAP_CHOWN, kuid, kgid)) { + return syserror.EPERM + } + } + if stat.Mask&linux.STATX_GID != 0 { + if !((creds.EffectiveKUID == kuid && creds.InGroup(auth.KGID(stat.GID))) || + HasCapabilityOnFile(creds, linux.CAP_CHOWN, kuid, kgid)) { + return syserror.EPERM + } + } + if stat.Mask&(linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME) != 0 { + if !CanActAsOwner(creds, kuid) { + if (stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec != linux.UTIME_NOW) || + (stat.Mask&linux.STATX_MTIME != 0 && stat.Mtime.Nsec != linux.UTIME_NOW) || + (stat.Mask&linux.STATX_CTIME != 0 && stat.Ctime.Nsec != linux.UTIME_NOW) { + return syserror.EPERM + } + // isDir is irrelevant in the following call to + // GenericCheckPermissions since ats == MayWrite means that + // CAP_DAC_READ_SEARCH does not apply, and CAP_DAC_OVERRIDE + // applies, regardless of isDir. + if err := GenericCheckPermissions(creds, MayWrite, false /* isDir */, mode, kuid, kgid); err != nil { + return err + } + } + } + return nil +} + +// CanActAsOwner returns true if creds can act as the owner of a file with the +// given owning UID, consistent with Linux's +// fs/inode.c:inode_owner_or_capable(). +func CanActAsOwner(creds *auth.Credentials, kuid auth.KUID) bool { + if creds.EffectiveKUID == kuid { + return true + } + return creds.HasCapability(linux.CAP_FOWNER) && creds.UserNamespace.MapFromKUID(kuid).Ok() +} + +// HasCapabilityOnFile returns true if creds has the given capability with +// respect to a file with the given owning UID and GID, consistent with Linux's +// kernel/capability.c:capable_wrt_inode_uidgid(). +func HasCapabilityOnFile(creds *auth.Credentials, cp linux.Capability, kuid auth.KUID, kgid auth.KGID) bool { + return creds.HasCapability(cp) && creds.UserNamespace.MapFromKUID(kuid).Ok() && creds.UserNamespace.MapFromKGID(kgid).Ok() +} diff --git a/pkg/sentry/vfs/syscalls.go b/pkg/sentry/vfs/syscalls.go deleted file mode 100644 index 436151afa..000000000 --- a/pkg/sentry/vfs/syscalls.go +++ /dev/null @@ -1,237 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/syserror" -) - -// PathOperation specifies the path operated on by a VFS method. -// -// PathOperation is passed to VFS methods by pointer to reduce memory copying: -// it's somewhat large and should never escape. (Options structs are passed by -// pointer to VFS and FileDescription methods for the same reason.) -type PathOperation struct { - // Root is the VFS root. References on Root are borrowed from the provider - // of the PathOperation. - // - // Invariants: Root.Ok(). - Root VirtualDentry - - // Start is the starting point for the path traversal. References on Start - // are borrowed from the provider of the PathOperation (i.e. the caller of - // the VFS method to which the PathOperation was passed). - // - // Invariants: Start.Ok(). If Pathname.Absolute, then Start == Root. - Start VirtualDentry - - // Path is the pathname traversed by this operation. - Pathname string - - // If FollowFinalSymlink is true, and the Dentry traversed by the final - // path component represents a symbolic link, the symbolic link should be - // followed. - FollowFinalSymlink bool -} - -// GetDentryAt returns a VirtualDentry representing the given path, at which a -// file must exist. A reference is taken on the returned VirtualDentry. -func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return VirtualDentry{}, err - } - for { - d, err := rp.mount.fs.impl.GetDentryAt(ctx, rp, *opts) - if err == nil { - vd := VirtualDentry{ - mount: rp.mount, - dentry: d, - } - rp.mount.IncRef() - vfs.putResolvingPath(rp) - return vd, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return VirtualDentry{}, err - } - } -} - -// MkdirAt creates a directory at the given path. -func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MkdirOptions) error { - // "Under Linux, apart from the permission bits, the S_ISVTX mode bit is - // also honored." - mkdir(2) - opts.Mode &= 01777 - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } - for { - err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return err - } - } -} - -// MknodAt creates a file of the given mode at the given path. It returns an -// error from the syserror package. -func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil - } - for { - if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { - vfs.putResolvingPath(rp) - return nil - } - // Handle mount traversals. - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return err - } - } -} - -// OpenAt returns a FileDescription providing access to the file at the given -// path. A reference is taken on the returned FileDescription. -func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { - // Remove: - // - // - O_LARGEFILE, which we always report in FileDescription status flags - // since only 64-bit architectures are supported at this time. - // - // - O_CLOEXEC, which affects file descriptors and therefore must be - // handled outside of VFS. - // - // - Unknown flags. - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE - // Linux's __O_SYNC (which we call linux.O_SYNC) implies O_DSYNC. - if opts.Flags&linux.O_SYNC != 0 { - opts.Flags |= linux.O_DSYNC - } - // Linux's __O_TMPFILE (which we call linux.O_TMPFILE) must be specified - // with O_DIRECTORY and a writable access mode (to ensure that it fails on - // filesystem implementations that do not support it). - if opts.Flags&linux.O_TMPFILE != 0 { - if opts.Flags&linux.O_DIRECTORY == 0 { - return nil, syserror.EINVAL - } - if opts.Flags&linux.O_CREAT != 0 { - return nil, syserror.EINVAL - } - if opts.Flags&linux.O_ACCMODE == linux.O_RDONLY { - return nil, syserror.EINVAL - } - } - // O_PATH causes most other flags to be ignored. - if opts.Flags&linux.O_PATH != 0 { - opts.Flags &= linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_PATH - } - // "On Linux, the following bits are also honored in mode: [S_ISUID, - // S_ISGID, S_ISVTX]" - open(2) - opts.Mode &= 07777 - - if opts.Flags&linux.O_NOFOLLOW != 0 { - pop.FollowFinalSymlink = false - } - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil, err - } - if opts.Flags&linux.O_DIRECTORY != 0 { - rp.mustBeDir = true - rp.mustBeDirOrig = true - } - for { - fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return fd, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return nil, err - } - } -} - -// StatAt returns metadata for the file at the given path. -func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *StatOptions) (linux.Statx, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return linux.Statx{}, err - } - for { - stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return stat, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return linux.Statx{}, err - } - } -} - -// StatusFlags returns file description status flags. -func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) { - flags, err := fd.impl.StatusFlags(ctx) - flags |= linux.O_LARGEFILE - return flags, err -} - -// SetStatusFlags sets file description status flags. -func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { - return fd.impl.SetStatusFlags(ctx, flags) -} - -// TODO: -// -// - VFS.SyncAllFilesystems() for sync(2) -// -// - Something for syncfs(2) -// -// - VFS.LinkAt() -// -// - VFS.ReadlinkAt() -// -// - VFS.RenameAt() -// -// - VFS.RmdirAt() -// -// - VFS.SetStatAt() -// -// - VFS.StatFSAt() -// -// - VFS.SymlinkAt() -// -// - VFS.UmountAt() -// -// - VFS.UnlinkAt() -// -// - FileDescription.(almost everything) diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index 593144cb7..7a1d9e383 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -15,7 +15,10 @@ package vfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" @@ -114,6 +117,12 @@ func (fs *FDTestFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) err return syserror.EPERM } +// PrependPath implements FilesystemImpl.PrependPath. +func (fs *FDTestFilesystem) PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error { + b.PrependComponent(fmt.Sprintf("vfs.fdTestDentry:%p", vd.dentry.impl.(*fdTestDentry))) + return PrependPathSyntheticError{} +} + type fdTestDentry struct { vfsd Dentry } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index f0cd3ffe5..7262b0d0a 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -20,6 +20,7 @@ // VirtualFilesystem.mountMu // Dentry.mu // Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry +// VirtualFilesystem.filesystemsMu // VirtualFilesystem.fsTypesMu // // Locking Dentry.mu in multiple Dentries requires holding @@ -28,6 +29,11 @@ package vfs import ( "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/syserror" ) // A VirtualFilesystem (VFS for short) combines Filesystems in trees of Mounts. @@ -67,6 +73,11 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // filesystems contains all Filesystems. filesystems is protected by + // filesystemsMu. + filesystemsMu sync.Mutex + filesystems map[*Filesystem]struct{} + // fsTypes contains all FilesystemTypes that are usable in the // VirtualFilesystem. fsTypes is protected by fsTypesMu. fsTypesMu sync.RWMutex @@ -77,12 +88,379 @@ type VirtualFilesystem struct { func New() *VirtualFilesystem { vfs := &VirtualFilesystem{ mountpoints: make(map[*Dentry]map[*Mount]struct{}), + filesystems: make(map[*Filesystem]struct{}), fsTypes: make(map[string]FilesystemType), } vfs.mounts.Init() return vfs } +// PathOperation specifies the path operated on by a VFS method. +// +// PathOperation is passed to VFS methods by pointer to reduce memory copying: +// it's somewhat large and should never escape. (Options structs are passed by +// pointer to VFS and FileDescription methods for the same reason.) +type PathOperation struct { + // Root is the VFS root. References on Root are borrowed from the provider + // of the PathOperation. + // + // Invariants: Root.Ok(). + Root VirtualDentry + + // Start is the starting point for the path traversal. References on Start + // are borrowed from the provider of the PathOperation (i.e. the caller of + // the VFS method to which the PathOperation was passed). + // + // Invariants: Start.Ok(). If Pathname.Absolute, then Start == Root. + Start VirtualDentry + + // Path is the pathname traversed by this operation. + Pathname string + + // If FollowFinalSymlink is true, and the Dentry traversed by the final + // path component represents a symbolic link, the symbolic link should be + // followed. + FollowFinalSymlink bool +} + +// GetDentryAt returns a VirtualDentry representing the given path, at which a +// file must exist. A reference is taken on the returned VirtualDentry. +func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return VirtualDentry{}, err + } + for { + d, err := rp.mount.fs.impl.GetDentryAt(ctx, rp, *opts) + if err == nil { + vd := VirtualDentry{ + mount: rp.mount, + dentry: d, + } + rp.mount.IncRef() + vfs.putResolvingPath(rp) + return vd, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return VirtualDentry{}, err + } + } +} + +// LinkAt creates a hard link at newpop representing the existing file at +// oldpop. +func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation) error { + oldVD, err := vfs.GetDentryAt(ctx, creds, oldpop, &GetDentryOptions{}) + if err != nil { + return err + } + rp, err := vfs.getResolvingPath(creds, newpop) + if err != nil { + oldVD.DecRef() + return err + } + for { + err := rp.mount.fs.impl.LinkAt(ctx, rp, oldVD) + if err == nil { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return err + } + } +} + +// MkdirAt creates a directory at the given path. +func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MkdirOptions) error { + // "Under Linux, apart from the permission bits, the S_ISVTX mode bit is + // also honored." - mkdir(2) + opts.Mode &= 0777 | linux.S_ISVTX + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// MknodAt creates a file of the given mode at the given path. It returns an +// error from the syserror package. +func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil + } + for { + if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { + vfs.putResolvingPath(rp) + return nil + } + // Handle mount traversals. + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// OpenAt returns a FileDescription providing access to the file at the given +// path. A reference is taken on the returned FileDescription. +func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { + // Remove: + // + // - O_LARGEFILE, which we always report in FileDescription status flags + // since only 64-bit architectures are supported at this time. + // + // - O_CLOEXEC, which affects file descriptors and therefore must be + // handled outside of VFS. + // + // - Unknown flags. + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE + // Linux's __O_SYNC (which we call linux.O_SYNC) implies O_DSYNC. + if opts.Flags&linux.O_SYNC != 0 { + opts.Flags |= linux.O_DSYNC + } + // Linux's __O_TMPFILE (which we call linux.O_TMPFILE) must be specified + // with O_DIRECTORY and a writable access mode (to ensure that it fails on + // filesystem implementations that do not support it). + if opts.Flags&linux.O_TMPFILE != 0 { + if opts.Flags&linux.O_DIRECTORY == 0 { + return nil, syserror.EINVAL + } + if opts.Flags&linux.O_CREAT != 0 { + return nil, syserror.EINVAL + } + if opts.Flags&linux.O_ACCMODE == linux.O_RDONLY { + return nil, syserror.EINVAL + } + } + // O_PATH causes most other flags to be ignored. + if opts.Flags&linux.O_PATH != 0 { + opts.Flags &= linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_PATH + } + // "On Linux, the following bits are also honored in mode: [S_ISUID, + // S_ISGID, S_ISVTX]" - open(2) + opts.Mode &= 0777 | linux.S_ISUID | linux.S_ISGID | linux.S_ISVTX + + if opts.Flags&linux.O_NOFOLLOW != 0 { + pop.FollowFinalSymlink = false + } + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil, err + } + if opts.Flags&linux.O_DIRECTORY != 0 { + rp.mustBeDir = true + rp.mustBeDirOrig = true + } + for { + fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return fd, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return nil, err + } + } +} + +// ReadlinkAt returns the target of the symbolic link at the given path. +func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (string, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return "", err + } + for { + target, err := rp.mount.fs.impl.ReadlinkAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return target, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return "", err + } + } +} + +// RenameAt renames the file at oldpop to newpop. +func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation, opts *RenameOptions) error { + oldVD, err := vfs.GetDentryAt(ctx, creds, oldpop, &GetDentryOptions{}) + if err != nil { + return err + } + rp, err := vfs.getResolvingPath(creds, newpop) + if err != nil { + oldVD.DecRef() + return err + } + for { + err := rp.mount.fs.impl.RenameAt(ctx, rp, oldVD, *opts) + if err == nil { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return err + } + } +} + +// RmdirAt removes the directory at the given path. +func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.RmdirAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// SetStatAt changes metadata for the file at the given path. +func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *SetStatOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.SetStatAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// StatAt returns metadata for the file at the given path. +func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *StatOptions) (linux.Statx, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return linux.Statx{}, err + } + for { + stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return stat, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return linux.Statx{}, err + } + } +} + +// StatFSAt returns metadata for the filesystem containing the file at the +// given path. +func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (linux.Statfs, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return linux.Statfs{}, err + } + for { + statfs, err := rp.mount.fs.impl.StatFSAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return statfs, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return linux.Statfs{}, err + } + } +} + +// SymlinkAt creates a symbolic link at the given path with the given target. +func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, target string) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.SymlinkAt(ctx, rp, target) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// UnlinkAt deletes the non-directory file at the given path. +func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.UnlinkAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// SyncAllFilesystems has the semantics of Linux's sync(2). +func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { + fss := make(map[*Filesystem]struct{}) + vfs.filesystemsMu.Lock() + for fs := range vfs.filesystems { + if !fs.TryIncRef() { + continue + } + fss[fs] = struct{}{} + } + vfs.filesystemsMu.Unlock() + var retErr error + for fs := range fss { + if err := fs.impl.Sync(ctx); err != nil && retErr == nil { + retErr = err + } + fs.DecRef() + } + return retErr +} + // A VirtualDentry represents a node in a VFS tree, by combining a Dentry // (which represents a node in a Filesystem's tree) and a Mount (which // represents the Filesystem's position in a VFS mount tree). diff --git a/pkg/syncutil/BUILD b/pkg/syncutil/BUILD index b06a90bef..cb1f41628 100644 --- a/pkg/syncutil/BUILD +++ b/pkg/syncutil/BUILD @@ -31,8 +31,6 @@ go_template( go_library( name = "syncutil", srcs = [ - "downgradable_rwmutex_1_12_unsafe.go", - "downgradable_rwmutex_1_13_unsafe.go", "downgradable_rwmutex_unsafe.go", "memmove_unsafe.go", "norace_unsafe.go", diff --git a/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go deleted file mode 100644 index 7c6336e62..000000000 --- a/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go +++ /dev/null @@ -1,21 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.13 - -// TODO(b/133868570): Delete once Go 1.12 is no longer supported. - -package syncutil - -import _ "unsafe" - -//go:linkname runtimeSemrelease112 sync.runtime_Semrelease -func runtimeSemrelease112(s *uint32, handoff bool) - -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) { - // 'skipframes' is only available starting from 1.13. - runtimeSemrelease112(s, handoff) -} diff --git a/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go deleted file mode 100644 index 3c3673119..000000000 --- a/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.13 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -package syncutil - -import _ "unsafe" - -//go:linkname runtimeSemrelease sync.runtime_Semrelease -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) diff --git a/pkg/syncutil/downgradable_rwmutex_unsafe.go b/pkg/syncutil/downgradable_rwmutex_unsafe.go index 07feca402..51e11555d 100644 --- a/pkg/syncutil/downgradable_rwmutex_unsafe.go +++ b/pkg/syncutil/downgradable_rwmutex_unsafe.go @@ -3,7 +3,7 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// +build go1.12 +// +build go1.13 // +build !go1.15 // Check go:linkname function signatures when updating Go version. @@ -27,6 +27,9 @@ import ( //go:linkname runtimeSemacquire sync.runtime_Semacquire func runtimeSemacquire(s *uint32) +//go:linkname runtimeSemrelease sync.runtime_Semrelease +func runtimeSemrelease(s *uint32, handoff bool, skipframes int) + // DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock // method. type DowngradableRWMutex struct { diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index a3485b35c..f1d837196 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -38,12 +38,15 @@ go_test( size = "small", srcs = [ "checksum_test.go", + "ipv6_test.go", "ipversion_test.go", "tcp_test.go", ], deps = [ ":header", + "//pkg/tcpip", "//pkg/tcpip/buffer", + "@com_github_google_go-cmp//cmp:go_default_library", ], ) @@ -55,5 +58,8 @@ go_test( "ndp_test.go", ], embed = [":header"], - deps = ["//pkg/tcpip"], + deps = [ + "//pkg/tcpip", + "@com_github_google_go-cmp//cmp:go_default_library", + ], ) diff --git a/pkg/tcpip/header/ipv6.go b/pkg/tcpip/header/ipv6.go index 0caa51c1e..fc671e439 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -90,6 +90,18 @@ const ( // IPv6Any is the non-routable IPv6 "any" meta address. It is also // known as the unspecified address. IPv6Any tcpip.Address = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" + + // IIDSize is the size of an interface identifier (IID), in bytes, as + // defined by RFC 4291 section 2.5.1. + IIDSize = 8 + + // IIDOffsetInIPv6Address is the offset, in bytes, from the start + // of an IPv6 address to the beginning of the interface identifier + // (IID) for auto-generated addresses. That is, all bytes before + // the IIDOffsetInIPv6Address-th byte are the prefix bytes, and all + // bytes including and after the IIDOffsetInIPv6Address-th byte are + // for the IID. + IIDOffsetInIPv6Address = 8 ) // IPv6EmptySubnet is the empty IPv6 subnet. It may also be known as the @@ -266,27 +278,43 @@ func SolicitedNodeAddr(addr tcpip.Address) tcpip.Address { return solicitedNodeMulticastPrefix + addr[len(addr)-3:] } +// EthernetAdddressToModifiedEUI64IntoBuf populates buf with a modified EUI-64 +// from a 48-bit Ethernet/MAC address, as per RFC 4291 section 2.5.1. +// +// buf MUST be at least 8 bytes. +func EthernetAdddressToModifiedEUI64IntoBuf(linkAddr tcpip.LinkAddress, buf []byte) { + buf[0] = linkAddr[0] ^ 2 + buf[1] = linkAddr[1] + buf[2] = linkAddr[2] + buf[3] = 0xFF + buf[4] = 0xFE + buf[5] = linkAddr[3] + buf[6] = linkAddr[4] + buf[7] = linkAddr[5] +} + +// EthernetAddressToModifiedEUI64 computes a modified EUI-64 from a 48-bit +// Ethernet/MAC address, as per RFC 4291 section 2.5.1. +func EthernetAddressToModifiedEUI64(linkAddr tcpip.LinkAddress) [IIDSize]byte { + var buf [IIDSize]byte + EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, buf[:]) + return buf +} + // LinkLocalAddr computes the default IPv6 link-local address from a link-layer // (MAC) address. func LinkLocalAddr(linkAddr tcpip.LinkAddress) tcpip.Address { - // Convert a 48-bit MAC to an EUI-64 and then prepend the link-local - // header, FE80::. + // Convert a 48-bit MAC to a modified EUI-64 and then prepend the + // link-local header, FE80::. // // The conversion is very nearly: // aa:bb:cc:dd:ee:ff => FE80::Aabb:ccFF:FEdd:eeff // Note the capital A. The conversion aa->Aa involves a bit flip. - lladdrb := [16]byte{ - 0: 0xFE, - 1: 0x80, - 8: linkAddr[0] ^ 2, - 9: linkAddr[1], - 10: linkAddr[2], - 11: 0xFF, - 12: 0xFE, - 13: linkAddr[3], - 14: linkAddr[4], - 15: linkAddr[5], + lladdrb := [IPv6AddressSize]byte{ + 0: 0xFE, + 1: 0x80, } + EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, lladdrb[IIDOffsetInIPv6Address:]) return tcpip.Address(lladdrb[:]) } diff --git a/pkg/tcpip/header/ipv6_test.go b/pkg/tcpip/header/ipv6_test.go new file mode 100644 index 000000000..42c5c6fc1 --- /dev/null +++ b/pkg/tcpip/header/ipv6_test.go @@ -0,0 +1,45 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package header_test + +import ( + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +const linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + +func TestEthernetAdddressToModifiedEUI64(t *testing.T) { + expectedIID := [header.IIDSize]byte{0, 2, 3, 255, 254, 4, 5, 6} + + if diff := cmp.Diff(expectedIID, header.EthernetAddressToModifiedEUI64(linkAddr)); diff != "" { + t.Errorf("EthernetAddressToModifiedEUI64(%s) mismatch (-want +got):\n%s", linkAddr, diff) + } + + var buf [header.IIDSize]byte + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, buf[:]) + if diff := cmp.Diff(expectedIID, buf); diff != "" { + t.Errorf("EthernetAddressToModifiedEUI64IntoBuf(%s, _) mismatch (-want +got):\n%s", linkAddr, diff) + } +} + +func TestLinkLocalAddr(t *testing.T) { + if got, want := header.LinkLocalAddr(linkAddr), tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x02\x03\xff\xfe\x04\x05\x06"); got != want { + t.Errorf("got LinkLocalAddr(%s) = %s, want = %s", linkAddr, got, want) + } +} diff --git a/pkg/tcpip/header/ndp_options.go b/pkg/tcpip/header/ndp_options.go index 1ca6199ef..06e0bace2 100644 --- a/pkg/tcpip/header/ndp_options.go +++ b/pkg/tcpip/header/ndp_options.go @@ -17,6 +17,7 @@ package header import ( "encoding/binary" "errors" + "math" "time" "gvisor.dev/gvisor/pkg/tcpip" @@ -85,6 +86,23 @@ const ( // within an NDPPrefixInformation. ndpPrefixInformationPrefixOffset = 14 + // NDPRecursiveDNSServerOptionType is the type of the Recursive DNS + // Server option, as per RFC 8106 section 5.1. + NDPRecursiveDNSServerOptionType = 25 + + // ndpRecursiveDNSServerLifetimeOffset is the start of the 4-byte + // Lifetime field within an NDPRecursiveDNSServer. + ndpRecursiveDNSServerLifetimeOffset = 2 + + // ndpRecursiveDNSServerAddressesOffset is the start of the addresses + // for IPv6 Recursive DNS Servers within an NDPRecursiveDNSServer. + ndpRecursiveDNSServerAddressesOffset = 6 + + // minNDPRecursiveDNSServerLength is the minimum NDP Recursive DNS + // Server option's length field value when it contains at least one + // IPv6 address. + minNDPRecursiveDNSServerLength = 3 + // lengthByteUnits is the multiplier factor for the Length field of an // NDP option. That is, the length field for NDP options is in units of // 8 octets, as per RFC 4861 section 4.6. @@ -92,13 +110,13 @@ const ( ) var ( - // NDPPrefixInformationInfiniteLifetime is a value that represents - // infinity for the Valid and Preferred Lifetime fields in a NDP Prefix - // Information option. Its value is (2^32 - 1)s = 4294967295s + // NDPInfiniteLifetime is a value that represents infinity for the + // 4-byte lifetime fields found in various NDP options. Its value is + // (2^32 - 1)s = 4294967295s. // // This is a variable instead of a constant so that tests can change // this value to a smaller value. It should only be modified by tests. - NDPPrefixInformationInfiniteLifetime = time.Second * 4294967295 + NDPInfiniteLifetime = time.Second * math.MaxUint32 ) // NDPOptionIterator is an iterator of NDPOption. @@ -118,6 +136,7 @@ var ( ErrNDPOptBufExhausted = errors.New("Buffer unexpectedly exhausted") ErrNDPOptZeroLength = errors.New("NDP option has zero-valued Length field") ErrNDPOptMalformedBody = errors.New("NDP option has a malformed body") + ErrNDPInvalidLength = errors.New("NDP option's Length value is invalid as per relevant RFC") ) // Next returns the next element in the backing NDPOptions, or true if we are @@ -182,6 +201,22 @@ func (i *NDPOptionIterator) Next() (NDPOption, bool, error) { } return NDPPrefixInformation(body), false, nil + + case NDPRecursiveDNSServerOptionType: + // RFC 8106 section 5.3.1 outlines that the RDNSS option + // must have a minimum length of 3 so it contains at + // least one IPv6 address. + if l < minNDPRecursiveDNSServerLength { + return nil, true, ErrNDPInvalidLength + } + + opt := NDPRecursiveDNSServer(body) + if len(opt.Addresses()) == 0 { + return nil, true, ErrNDPOptMalformedBody + } + + return opt, false, nil + default: // We do not yet recognize the option, just skip for // now. This is okay because RFC 4861 allows us to @@ -434,7 +469,7 @@ func (o NDPPrefixInformation) AutonomousAddressConfigurationFlag() bool { // // Note, a value of 0 implies the prefix should not be considered as on-link, // and a value of infinity/forever is represented by -// NDPPrefixInformationInfiniteLifetime. +// NDPInfiniteLifetime. func (o NDPPrefixInformation) ValidLifetime() time.Duration { // The field is the time in seconds, as per RFC 4861 section 4.6.2. return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpPrefixInformationValidLifetimeOffset:])) @@ -447,7 +482,7 @@ func (o NDPPrefixInformation) ValidLifetime() time.Duration { // // Note, a value of 0 implies that addresses generated from the prefix should // no longer remain preferred, and a value of infinity is represented by -// NDPPrefixInformationInfiniteLifetime. +// NDPInfiniteLifetime. // // Also note that the value of this field MUST NOT exceed the Valid Lifetime // field to avoid preferring addresses that are no longer valid, for the @@ -476,3 +511,79 @@ func (o NDPPrefixInformation) Subnet() tcpip.Subnet { } return addrWithPrefix.Subnet() } + +// NDPRecursiveDNSServer is the NDP Recursive DNS Server option, as defined by +// RFC 8106 section 5.1. +// +// To make sure that the option meets its minimum length and does not end in the +// middle of a DNS server's IPv6 address, the length of a valid +// NDPRecursiveDNSServer must meet the following constraint: +// (Length - ndpRecursiveDNSServerAddressesOffset) % IPv6AddressSize == 0 +type NDPRecursiveDNSServer []byte + +// Type returns the type of an NDP Recursive DNS Server option. +// +// Type implements NDPOption.Type. +func (NDPRecursiveDNSServer) Type() uint8 { + return NDPRecursiveDNSServerOptionType +} + +// Length implements NDPOption.Length. +func (o NDPRecursiveDNSServer) Length() int { + return len(o) +} + +// serializeInto implements NDPOption.serializeInto. +func (o NDPRecursiveDNSServer) serializeInto(b []byte) int { + used := copy(b, o) + + // Zero out the reserved bytes that are before the Lifetime field. + for i := 0; i < ndpRecursiveDNSServerLifetimeOffset; i++ { + b[i] = 0 + } + + return used +} + +// Lifetime returns the length of time that the DNS server addresses +// in this option may be used for name resolution. +// +// Note, a value of 0 implies the addresses should no longer be used, +// and a value of infinity/forever is represented by NDPInfiniteLifetime. +// +// Lifetime may panic if o does not have enough bytes to hold the Lifetime +// field. +func (o NDPRecursiveDNSServer) Lifetime() time.Duration { + // The field is the time in seconds, as per RFC 8106 section 5.1. + return time.Second * time.Duration(binary.BigEndian.Uint32(o[ndpRecursiveDNSServerLifetimeOffset:])) +} + +// Addresses returns the recursive DNS server IPv6 addresses that may be +// used for name resolution. +// +// Note, some of the addresses returned MAY be link-local addresses. +// +// Addresses may panic if o does not hold valid IPv6 addresses. +func (o NDPRecursiveDNSServer) Addresses() []tcpip.Address { + l := len(o) + if l < ndpRecursiveDNSServerAddressesOffset { + return nil + } + + l -= ndpRecursiveDNSServerAddressesOffset + if l%IPv6AddressSize != 0 { + return nil + } + + buf := o[ndpRecursiveDNSServerAddressesOffset:] + var addrs []tcpip.Address + for len(buf) > 0 { + addr := tcpip.Address(buf[:IPv6AddressSize]) + if !IsV6UnicastAddress(addr) { + return nil + } + addrs = append(addrs, addr) + buf = buf[IPv6AddressSize:] + } + return addrs +} diff --git a/pkg/tcpip/header/ndp_test.go b/pkg/tcpip/header/ndp_test.go index ad6daafcd..2c439d70c 100644 --- a/pkg/tcpip/header/ndp_test.go +++ b/pkg/tcpip/header/ndp_test.go @@ -19,6 +19,7 @@ import ( "testing" "time" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -369,6 +370,175 @@ func TestNDPPrefixInformationOption(t *testing.T) { } } +func TestNDPRecursiveDNSServerOptionSerialize(t *testing.T) { + b := []byte{ + 9, 8, + 1, 2, 4, 8, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + } + targetBuf := []byte{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1} + expected := []byte{ + 25, 3, 0, 0, + 1, 2, 4, 8, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + } + opts := NDPOptions(targetBuf) + serializer := NDPOptionsSerializer{ + NDPRecursiveDNSServer(b), + } + if got, want := opts.Serialize(serializer), len(expected); got != want { + t.Errorf("got Serialize = %d, want = %d", got, want) + } + if !bytes.Equal(targetBuf, expected) { + t.Fatalf("got targetBuf = %x, want = %x", targetBuf, expected) + } + + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got := next.Type(); got != NDPRecursiveDNSServerOptionType { + t.Errorf("got Type = %d, want = %d", got, NDPRecursiveDNSServerOptionType) + } + + opt, ok := next.(NDPRecursiveDNSServer) + if !ok { + t.Fatalf("next (type = %T) cannot be casted to an NDPRecursiveDNSServer", next) + } + if got := opt.Type(); got != 25 { + t.Errorf("got Type = %d, want = 31", got) + } + if got := opt.Length(); got != 22 { + t.Errorf("got Length = %d, want = 22", got) + } + if got, want := opt.Lifetime(), 16909320*time.Second; got != want { + t.Errorf("got Lifetime = %s, want = %s", got, want) + } + want := []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + } + if got := opt.Addresses(); !cmp.Equal(got, want) { + t.Errorf("got Addresses = %v, want = %v", got, want) + } + + // Iterator should not return anything else. + next, done, err = it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } +} + +func TestNDPRecursiveDNSServerOption(t *testing.T) { + tests := []struct { + name string + buf []byte + lifetime time.Duration + addrs []tcpip.Address + }{ + { + "Valid1Addr", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + }, + }, + { + "Valid2Addr", + []byte{ + 25, 5, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 16, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x10", + }, + }, + { + "Valid3Addr", + []byte{ + 25, 7, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 16, + 17, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 17, + }, + 0, + []tcpip.Address{ + "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x10", + "\x11\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x11", + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := NDPOptions(test.buf) + it, err := opts.Iter(true) + if err != nil { + t.Fatalf("got Iter = (_, %s), want = (_, nil)", err) + } + + // Iterator should get our option. + next, done, err := it.Next() + if err != nil { + t.Fatalf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if done { + t.Fatal("got Next = (_, true, _), want = (_, false, _)") + } + if got := next.Type(); got != NDPRecursiveDNSServerOptionType { + t.Fatalf("got Type %= %d, want = %d", got, NDPRecursiveDNSServerOptionType) + } + + opt, ok := next.(NDPRecursiveDNSServer) + if !ok { + t.Fatalf("next (type = %T) cannot be casted to an NDPRecursiveDNSServer", next) + } + if got := opt.Lifetime(); got != test.lifetime { + t.Errorf("got Lifetime = %d, want = %d", got, test.lifetime) + } + if got := opt.Addresses(); !cmp.Equal(got, test.addrs) { + t.Errorf("got Addresses = %v, want = %v", got, test.addrs) + } + + // Iterator should not return anything else. + next, done, err = it.Next() + if err != nil { + t.Errorf("got Next = (_, _, %s), want = (_, _, nil)", err) + } + if !done { + t.Error("got Next = (_, false, _), want = (_, true, _)") + } + if next != nil { + t.Errorf("got Next = (%x, _, _), want = (nil, _, _)", next) + } + }) + } +} + // TestNDPOptionsIterCheck tests that Iter will return false if the NDPOptions // the iterator was returned for is malformed. func TestNDPOptionsIterCheck(t *testing.T) { @@ -473,6 +643,51 @@ func TestNDPOptionsIterCheck(t *testing.T) { }, nil, }, + { + "InvalidRecursiveDNSServerCutsOffAddress", + []byte{ + 25, 4, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + 0, 1, 2, 3, 4, 5, 6, 7, + }, + ErrNDPOptMalformedBody, + }, + { + "InvalidRecursiveDNSServerInvalidLengthField", + []byte{ + 25, 2, 0, 0, + 0, 0, 0, 0, + 0, 1, 2, 3, 4, 5, 6, 7, 8, + }, + ErrNDPInvalidLength, + }, + { + "RecursiveDNSServerTooSmall", + []byte{ + 25, 1, 0, 0, + 0, 0, 0, + }, + ErrNDPOptBufExhausted, + }, + { + "RecursiveDNSServerMulticast", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + }, + ErrNDPOptMalformedBody, + }, + { + "RecursiveDNSServerUnspecified", + []byte{ + 25, 3, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + }, + ErrNDPOptMalformedBody, + }, } for _, test := range tests { diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD index 4839f0a65..e156b01f6 100644 --- a/pkg/tcpip/ports/BUILD +++ b/pkg/tcpip/ports/BUILD @@ -1,5 +1,5 @@ -load("//tools/go_stateify:defs.bzl", "go_library") load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 30cea8996..6c5e19e8f 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -41,6 +41,30 @@ type portDescriptor struct { port uint16 } +// Flags represents the type of port reservation. +// +// +stateify savable +type Flags struct { + // MostRecent represents UDP SO_REUSEADDR. + MostRecent bool + + // LoadBalanced indicates SO_REUSEPORT. + // + // LoadBalanced takes precidence over MostRecent. + LoadBalanced bool +} + +func (f Flags) bits() reuseFlag { + var rf reuseFlag + if f.MostRecent { + rf |= mostRecentFlag + } + if f.LoadBalanced { + rf |= loadBalancedFlag + } + return rf +} + // PortManager manages allocating, reserving and releasing ports. type PortManager struct { mu sync.RWMutex @@ -54,9 +78,59 @@ type PortManager struct { hint uint32 } +type reuseFlag int + +const ( + mostRecentFlag reuseFlag = 1 << iota + loadBalancedFlag + nextFlag + + flagMask = nextFlag - 1 +) + type portNode struct { - reuse bool - refs int + // refs stores the count for each possible flag combination. + refs [nextFlag]int +} + +func (p portNode) totalRefs() int { + var total int + for _, r := range p.refs { + total += r + } + return total +} + +// flagRefs returns the number of references with all specified flags. +func (p portNode) flagRefs(flags reuseFlag) int { + var total int + for i, r := range p.refs { + if reuseFlag(i)&flags == flags { + total += r + } + } + return total +} + +// allRefsHave returns if all references have all specified flags. +func (p portNode) allRefsHave(flags reuseFlag) bool { + for i, r := range p.refs { + if reuseFlag(i)&flags == flags && r > 0 { + return false + } + } + return true +} + +// intersectionRefs returns the set of flags shared by all references. +func (p portNode) intersectionRefs() reuseFlag { + intersection := flagMask + for i, r := range p.refs { + if r > 0 { + intersection &= reuseFlag(i) + } + } + return intersection } // deviceNode is never empty. When it has no elements, it is removed from the @@ -66,30 +140,44 @@ type deviceNode map[tcpip.NICID]portNode // isAvailable checks whether binding is possible by device. If not binding to a // device, check against all portNodes. If binding to a specific device, check // against the unspecified device and the provided device. -func (d deviceNode) isAvailable(reuse bool, bindToDevice tcpip.NICID) bool { +// +// If either of the port reuse flags is enabled on any of the nodes, all nodes +// sharing a port must share at least one reuse flag. This matches Linux's +// behavior. +func (d deviceNode) isAvailable(flags Flags, bindToDevice tcpip.NICID) bool { + flagBits := flags.bits() if bindToDevice == 0 { // Trying to binding all devices. - if !reuse { + if flagBits == 0 { // Can't bind because the (addr,port) is already bound. return false } + intersection := flagMask for _, p := range d { - if !p.reuse { - // Can't bind because the (addr,port) was previously bound without reuse. + i := p.intersectionRefs() + intersection &= i + if intersection&flagBits == 0 { + // Can't bind because the (addr,port) was + // previously bound without reuse. return false } } return true } + intersection := flagMask + if p, ok := d[0]; ok { - if !reuse || !p.reuse { + intersection = p.intersectionRefs() + if intersection&flagBits == 0 { return false } } if p, ok := d[bindToDevice]; ok { - if !reuse || !p.reuse { + i := p.intersectionRefs() + intersection &= i + if intersection&flagBits == 0 { return false } } @@ -103,12 +191,12 @@ type bindAddresses map[tcpip.Address]deviceNode // isAvailable checks whether an IP address is available to bind to. If the // address is the "any" address, check all other addresses. Otherwise, just // check against the "any" address and the provided address. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice tcpip.NICID) bool { +func (b bindAddresses) isAvailable(addr tcpip.Address, flags Flags, bindToDevice tcpip.NICID) bool { if addr == anyIPAddress { // If binding to the "any" address then check that there are no conflicts // with all addresses. for _, d := range b { - if !d.isAvailable(reuse, bindToDevice) { + if !d.isAvailable(flags, bindToDevice) { return false } } @@ -117,14 +205,14 @@ func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice // Check that there is no conflict with the "any" address. if d, ok := b[anyIPAddress]; ok { - if !d.isAvailable(reuse, bindToDevice) { + if !d.isAvailable(flags, bindToDevice) { return false } } // Check that this is no conflict with the provided address. if d, ok := b[addr]; ok { - if !d.isAvailable(reuse, bindToDevice) { + if !d.isAvailable(flags, bindToDevice) { return false } } @@ -190,17 +278,17 @@ func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p ui } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) + return s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, reuse, bindToDevice) { + if !addrs.isAvailable(addr, flags, bindToDevice) { return false } } @@ -212,14 +300,14 @@ func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumb // reserved by another endpoint. If port is zero, ReservePort will search for // an unreserved ephemeral port and reserve it, returning its value in the // "port" return value. -func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, reuse, bindToDevice) { + if !s.reserveSpecificPort(networks, transport, addr, port, flags, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -227,15 +315,16 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), nil + return s.reserveSpecificPort(networks, transport, addr, p, flags, bindToDevice), nil }) } // reserveSpecificPort tries to reserve the given port on all given protocols. -func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, flags, bindToDevice) { return false } + flagBits := flags.bits() // Reserve port on all network protocols. for _, network := range networks { @@ -250,12 +339,9 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber d = make(deviceNode) m[addr] = d } - if n, ok := d[bindToDevice]; ok { - n.refs++ - d[bindToDevice] = n - } else { - d[bindToDevice] = portNode{reuse: reuse, refs: 1} - } + n := d[bindToDevice] + n.refs[flagBits]++ + d[bindToDevice] = n } return true @@ -263,10 +349,12 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber // ReleasePort releases the reservation on a port/IP combination so that it can // be reserved by other endpoints. -func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, bindToDevice tcpip.NICID) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, flags Flags, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() + flagBits := flags.bits() + for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { @@ -278,9 +366,9 @@ func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transp if !ok { continue } - n.refs-- + n.refs[flagBits]-- d[bindToDevice] = n - if n.refs == 0 { + if n.refs == [nextFlag]int{} { delete(d, bindToDevice) } if len(d) == 0 { diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 19f4833fc..d6969d050 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -33,7 +33,7 @@ type portReserveTestAction struct { port uint16 ip tcpip.Address want *tcpip.Error - reuse bool + flags Flags release bool device tcpip.NICID } @@ -50,7 +50,7 @@ func TestPortReservation(t *testing.T) { {port: 80, ip: fakeIPAddress1, want: nil}, /* N.B. Order of tests matters! */ {port: 80, ip: anyIPAddress, want: tcpip.ErrPortInUse}, - {port: 80, ip: fakeIPAddress, want: tcpip.ErrPortInUse, reuse: true}, + {port: 80, ip: fakeIPAddress, want: tcpip.ErrPortInUse, flags: Flags{LoadBalanced: true}}, }, }, { @@ -61,7 +61,7 @@ func TestPortReservation(t *testing.T) { /* release fakeIPAddress, but anyIPAddress is still inuse */ {port: 22, ip: fakeIPAddress, release: true}, {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, - {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse, reuse: true}, + {port: 22, ip: fakeIPAddress, want: tcpip.ErrPortInUse, flags: Flags{LoadBalanced: true}}, /* Release port 22 from any IP address, then try to reserve fake IP address on 22 */ {port: 22, ip: anyIPAddress, want: nil, release: true}, {port: 22, ip: fakeIPAddress, want: nil}, @@ -71,36 +71,36 @@ func TestPortReservation(t *testing.T) { actions: []portReserveTestAction{ {port: 00, ip: fakeIPAddress, want: nil}, {port: 00, ip: fakeIPAddress, want: nil}, - {port: 00, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 00, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { tname: "bind to ip with reuseport", actions: []portReserveTestAction{ - {port: 25, ip: fakeIPAddress, reuse: true, want: nil}, - {port: 25, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 25, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 25, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, - {port: 25, ip: fakeIPAddress, reuse: false, want: tcpip.ErrPortInUse}, - {port: 25, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 25, ip: fakeIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 25, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 25, ip: anyIPAddress, reuse: true, want: nil}, + {port: 25, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { tname: "bind to inaddr any with reuseport", actions: []portReserveTestAction{ - {port: 24, ip: anyIPAddress, reuse: true, want: nil}, - {port: 24, ip: anyIPAddress, reuse: true, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, - {port: 24, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, release: true, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, release: true, want: nil}, - {port: 24, ip: anyIPAddress, release: true}, - {port: 24, ip: anyIPAddress, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, release: true}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: tcpip.ErrPortInUse}, - {port: 24, ip: anyIPAddress, release: true}, - {port: 24, ip: anyIPAddress, reuse: false, want: nil}, + {port: 24, ip: anyIPAddress, flags: Flags{LoadBalanced: true}, release: true}, + {port: 24, ip: anyIPAddress, flags: Flags{}, want: nil}, }, }, { tname: "bind twice with device fails", @@ -125,88 +125,152 @@ func TestPortReservation(t *testing.T) { actions: []portReserveTestAction{ {port: 24, ip: fakeIPAddress, want: nil}, {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, }, }, { tname: "bind with device", actions: []portReserveTestAction{ {port: 24, ip: fakeIPAddress, device: 123, want: nil}, {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 789, want: nil}, {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, }, }, { - tname: "bind with reuse", + tname: "bind with reuseport", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, }, }, { - tname: "binding with reuse and device", + tname: "binding with reuseport and device", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, }, }, { - tname: "mixing reuse and not reuse by binding to device", + tname: "mixing reuseport and not reuseport by binding to device", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 456, want: nil}, - {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 999, want: nil}, }, }, { - tname: "can't bind to 0 after mixing reuse and not reuse", + tname: "can't bind to 0 after mixing reuseport and not reuseport", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, {port: 24, ip: fakeIPAddress, device: 456, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, }, }, { tname: "bind and release", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, - {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: tcpip.ErrPortInUse}, - {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, flags: Flags{LoadBalanced: true}, want: nil}, // Release the bind to device 0 and try again. - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil, release: true}, - {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: nil}, }, }, { - tname: "bind twice with reuse once", + tname: "bind twice with reuseport once", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, - {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, }, }, { tname: "release an unreserved device", actions: []portReserveTestAction{ - {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, - {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{}, want: nil}, // The below don't exist. - {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil, release: true}, - {port: 9999, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, flags: Flags{}, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil, release: true}, // Release all. - {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, - {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, flags: Flags{}, want: nil, release: true}, + }, + }, { + tname: "bind with reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{MostRecent: true}, want: nil}, + }, + }, { + tname: "bind twice with reuseaddr once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, flags: Flags{}, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr and reuseport, and then reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport, and then reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseaddr and reuseport twice, and then reuseaddr", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr and reuseport twice, and then reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + }, + }, { + tname: "bind with reuseaddr, and then reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuseport, and then reuseaddr and reuseport", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, flags: Flags{LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true, LoadBalanced: true}, want: nil}, + {port: 24, ip: fakeIPAddress, flags: Flags{MostRecent: true}, want: tcpip.ErrPortInUse}, }, }, } { @@ -216,12 +280,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.flags, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %+v, %d) = %v, want %v", test.ip, test.port, test.flags, test.device, err, test.want) } if test.port == 0 && (gotPort == 0 || gotPort < FirstEphemeral) { t.Fatalf("ReservePort(.., .., .., 0) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) diff --git a/pkg/tcpip/sample/tun_tcp_connect/BUILD b/pkg/tcpip/sample/tun_tcp_connect/BUILD index a57752a7c..d7496fde6 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/BUILD +++ b/pkg/tcpip/sample/tun_tcp_connect/BUILD @@ -5,6 +5,7 @@ package(licenses = ["notice"]) go_binary( name = "tun_tcp_connect", srcs = ["main.go"], + visibility = ["//:sandbox"], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/sample/tun_tcp_echo/BUILD b/pkg/tcpip/sample/tun_tcp_echo/BUILD index dad8ef399..875561566 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/BUILD +++ b/pkg/tcpip/sample/tun_tcp_echo/BUILD @@ -5,6 +5,7 @@ package(licenses = ["notice"]) go_binary( name = "tun_tcp_echo", srcs = ["main.go"], + visibility = ["//:sandbox"], deps = [ "//pkg/tcpip", "//pkg/tcpip/link/fdbased", diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index cfdd0496e..27bd02e76 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -58,6 +58,14 @@ const ( // Default = true. defaultDiscoverOnLinkPrefixes = true + // defaultAutoGenGlobalAddresses is the default configuration for + // whether or not to generate global IPv6 addresses in response to + // receiving a new Prefix Information option with its Autonomous + // Address AutoConfiguration flag set, as a host. + // + // Default = true. + defaultAutoGenGlobalAddresses = true + // minimumRetransmitTimer is the minimum amount of time to wait between // sending NDP Neighbor solicitation messages. Note, RFC 4861 does // not impose a minimum Retransmit Timer, but we do here to make sure @@ -87,6 +95,24 @@ const ( // // Max = 10. MaxDiscoveredOnLinkPrefixes = 10 + + // validPrefixLenForAutoGen is the expected prefix length that an + // address can be generated for. Must be 64 bits as the interface + // identifier (IID) is 64 bits and an IPv6 address is 128 bits, so + // 128 - 64 = 64. + validPrefixLenForAutoGen = 64 +) + +var ( + // MinPrefixInformationValidLifetimeForUpdate is the minimum Valid + // Lifetime to update the valid lifetime of a generated address by + // SLAAC. + // + // This is exported as a variable (instead of a constant) so tests + // can update it to a smaller value. + // + // Min = 2hrs. + MinPrefixInformationValidLifetimeForUpdate = 2 * time.Hour ) // NDPDispatcher is the interface integrators of netstack must implement to @@ -139,6 +165,33 @@ type NDPDispatcher interface { // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) []tcpip.Route + + // OnAutoGenAddress will be called when a new prefix with its + // autonomous address-configuration flag set has been received and SLAAC + // has been performed. Implementations may prevent the stack from + // assigning the address to the NIC by returning false. + // + // This function is not permitted to block indefinitely. It must not + // call functions on the stack itself. + OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool + + // OnAutoGenAddressInvalidated will be called when an auto-generated + // address (as part of SLAAC) has been invalidated. + // + // This function is not permitted to block indefinitely. It must not + // call functions on the stack itself. + OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) + + // OnRecursiveDNSServerOption will be called when an NDP option with + // recursive DNS servers has been received. Note, addrs may contain + // link-local addresses. + // + // It is up to the caller to use the DNS Servers only for their valid + // lifetime. OnRecursiveDNSServerOption may be called for new or + // already known DNS servers. If called with known DNS servers, their + // valid lifetimes must be refreshed to lifetime (it may be increased, + // decreased, or completely invalidated when lifetime = 0). + OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) } // NDPConfigurations is the NDP configurations for the netstack. @@ -168,6 +221,17 @@ type NDPConfigurations struct { // will be discovered from Router Advertisements' Prefix Information // option. This configuration is ignored if HandleRAs is false. DiscoverOnLinkPrefixes bool + + // AutoGenGlobalAddresses determines whether or not global IPv6 + // addresses will be generated for a NIC in response to receiving a new + // Prefix Information option with its Autonomous Address + // AutoConfiguration flag set, as a host, as per RFC 4862 (SLAAC). + // + // Note, if an address was already generated for some unique prefix, as + // part of SLAAC, this option does not affect whether or not the + // lifetime(s) of the generated address changes; this option only + // affects the generation of new addresses as part of SLAAC. + AutoGenGlobalAddresses bool } // DefaultNDPConfigurations returns an NDPConfigurations populated with @@ -179,6 +243,7 @@ func DefaultNDPConfigurations() NDPConfigurations { HandleRAs: defaultHandleRAs, DiscoverDefaultRouters: defaultDiscoverDefaultRouters, DiscoverOnLinkPrefixes: defaultDiscoverOnLinkPrefixes, + AutoGenGlobalAddresses: defaultAutoGenGlobalAddresses, } } @@ -210,6 +275,9 @@ type ndpState struct { // The on-link prefixes discovered through Router Advertisements' Prefix // Information option. onLinkPrefixes map[tcpip.Subnet]onLinkPrefixState + + // The addresses generated by SLAAC. + autoGenAddresses map[tcpip.Address]autoGenAddressState } // dadState holds the Duplicate Address Detection timer and channel to signal @@ -270,6 +338,32 @@ type onLinkPrefixState struct { doNotInvalidate *bool } +// autoGenAddressState holds data associated with an address generated via +// SLAAC. +type autoGenAddressState struct { + invalidationTimer *time.Timer + + // Used to signal the timer not to invalidate the SLAAC address (A) in + // a race condition (T1 is a goroutine that handles a PI for A and T2 + // is the goroutine that handles A's invalidation timer firing): + // T1: Receive a new PI for A + // T1: Obtain the NIC's lock before processing the PI + // T2: A's invalidation timer fires, and gets blocked on obtaining the + // NIC's lock + // T1: Refreshes/extends A's lifetime & releases NIC's lock + // T2: Obtains NIC's lock & invalidates A immediately + // + // To resolve this, T1 will check to see if the timer already fired, and + // inform the timer using doNotInvalidate to not invalidate A, so that + // once T2 obtains the lock, it will see that it is set to true and do + // nothing further. + doNotInvalidate *bool + + // Nonzero only when the address is not valid forever (invalidationTimer + // is not nil). + validUntil time.Time +} + // startDuplicateAddressDetection performs Duplicate Address Detection. // // This function must only be called by IPv6 addresses that are currently @@ -534,19 +628,21 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { // we do not check the iterator for errors on calls to Next. it, _ := ra.Options().Iter(false) for opt, done, _ := it.Next(); !done; opt, done, _ = it.Next() { - switch opt.Type() { - case header.NDPPrefixInformationType: - if !ndp.configs.DiscoverOnLinkPrefixes { + switch opt := opt.(type) { + case header.NDPRecursiveDNSServer: + if ndp.nic.stack.ndpDisp == nil { continue } - pi := opt.(header.NDPPrefixInformation) + ndp.nic.stack.ndpDisp.OnRecursiveDNSServerOption(ndp.nic.ID(), opt.Addresses(), opt.Lifetime()) - prefix := pi.Subnet() + case header.NDPPrefixInformation: + prefix := opt.Subnet() // Is the prefix a link-local? if header.IsV6LinkLocalAddress(prefix.ID()) { - // ...Yes, skip as per RFC 4861 section 6.3.4. + // ...Yes, skip as per RFC 4861 section 6.3.4, + // and RFC 4862 section 5.5.3.b (for SLAAC). continue } @@ -557,82 +653,13 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { continue } - if !pi.OnLinkFlag() { - // Not on-link so don't "discover" it as an - // on-link prefix. - continue - } - - prefixState, ok := ndp.onLinkPrefixes[prefix] - vl := pi.ValidLifetime() - switch { - case !ok && vl == 0: - // Don't know about this prefix but has a zero - // valid lifetime, so just ignore. - continue - - case !ok && vl != 0: - // This is a new on-link prefix we are - // discovering. - // - // Only remember it if we currently know about - // less than MaxDiscoveredOnLinkPrefixes on-link - // prefixes. - if len(ndp.onLinkPrefixes) < MaxDiscoveredOnLinkPrefixes { - ndp.rememberOnLinkPrefix(prefix, vl) - } - continue - - case ok && vl == 0: - // We know about the on-link prefix, but it is - // no longer to be considered on-link, so - // invalidate it. - ndp.invalidateOnLinkPrefix(prefix) - continue - } - - // This is an already discovered on-link prefix with a - // new non-zero valid lifetime. - // Update the invalidation timer. - timer := prefixState.invalidationTimer - - if timer == nil && vl >= header.NDPPrefixInformationInfiniteLifetime { - // Had infinite valid lifetime before and - // continues to have an invalid lifetime. Do - // nothing further. - continue - } - - if timer != nil && !timer.Stop() { - // If we reach this point, then we know the - // timer already fired after we took the NIC - // lock. Inform the timer to not invalidate - // the prefix once it obtains the lock as we - // just got a new PI that refeshes its lifetime - // to a non-zero value. See - // onLinkPrefixState.doNotInvalidate for more - // details. - *prefixState.doNotInvalidate = true + if opt.OnLinkFlag() { + ndp.handleOnLinkPrefixInformation(opt) } - if vl >= header.NDPPrefixInformationInfiniteLifetime { - // Prefix is now valid forever so we don't need - // an invalidation timer. - prefixState.invalidationTimer = nil - ndp.onLinkPrefixes[prefix] = prefixState - continue + if opt.AutonomousAddressConfigurationFlag() { + ndp.handleAutonomousPrefixInformation(opt) } - - if timer != nil { - // We already have a timer so just reset it to - // expire after the new valid lifetime. - timer.Reset(vl) - continue - } - - // We do not have a timer so just create a new one. - prefixState.invalidationTimer = ndp.prefixInvalidationCallback(prefix, vl, prefixState.doNotInvalidate) - ndp.onLinkPrefixes[prefix] = prefixState } // TODO(b/141556115): Do (MTU) Parameter Discovery. @@ -734,7 +761,7 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) var timer *time.Timer // Only create a timer if the lifetime is not infinite. - if l < header.NDPPrefixInformationInfiniteLifetime { + if l < header.NDPInfiniteLifetime { timer = ndp.prefixInvalidationCallback(prefix, l, &doNotInvalidate) } @@ -795,3 +822,345 @@ func (ndp *ndpState) prefixInvalidationCallback(prefix tcpip.Subnet, vl time.Dur ndp.invalidateOnLinkPrefix(prefix) }) } + +// handleOnLinkPrefixInformation handles a Prefix Information option with +// its on-link flag set, as per RFC 4861 section 6.3.4. +// +// handleOnLinkPrefixInformation assumes that the prefix this pi is for is +// not the link-local prefix and the on-link flag is set. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformation) { + prefix := pi.Subnet() + prefixState, ok := ndp.onLinkPrefixes[prefix] + vl := pi.ValidLifetime() + + if !ok && vl == 0 { + // Don't know about this prefix but it has a zero valid + // lifetime, so just ignore. + return + } + + if !ok && vl != 0 { + // This is a new on-link prefix we are discovering + // + // Only remember it if we currently know about less than + // MaxDiscoveredOnLinkPrefixes on-link prefixes. + if ndp.configs.DiscoverOnLinkPrefixes && len(ndp.onLinkPrefixes) < MaxDiscoveredOnLinkPrefixes { + ndp.rememberOnLinkPrefix(prefix, vl) + } + return + } + + if ok && vl == 0 { + // We know about the on-link prefix, but it is + // no longer to be considered on-link, so + // invalidate it. + ndp.invalidateOnLinkPrefix(prefix) + return + } + + // This is an already discovered on-link prefix with a + // new non-zero valid lifetime. + // Update the invalidation timer. + timer := prefixState.invalidationTimer + + if timer == nil && vl >= header.NDPInfiniteLifetime { + // Had infinite valid lifetime before and + // continues to have an invalid lifetime. Do + // nothing further. + return + } + + if timer != nil && !timer.Stop() { + // If we reach this point, then we know the timer alread fired + // after we took the NIC lock. Inform the timer to not + // invalidate the prefix once it obtains the lock as we just + // got a new PI that refreshes its lifetime to a non-zero value. + // See onLinkPrefixState.doNotInvalidate for more details. + *prefixState.doNotInvalidate = true + } + + if vl >= header.NDPInfiniteLifetime { + // Prefix is now valid forever so we don't need + // an invalidation timer. + prefixState.invalidationTimer = nil + ndp.onLinkPrefixes[prefix] = prefixState + return + } + + if timer != nil { + // We already have a timer so just reset it to + // expire after the new valid lifetime. + timer.Reset(vl) + return + } + + // We do not have a timer so just create a new one. + prefixState.invalidationTimer = ndp.prefixInvalidationCallback(prefix, vl, prefixState.doNotInvalidate) + ndp.onLinkPrefixes[prefix] = prefixState +} + +// handleAutonomousPrefixInformation handles a Prefix Information option with +// its autonomous flag set, as per RFC 4862 section 5.5.3. +// +// handleAutonomousPrefixInformation assumes that the prefix this pi is for is +// not the link-local prefix and the autonomous flag is set. +// +// The NIC that ndp belongs to and its associated stack MUST be locked. +func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) { + vl := pi.ValidLifetime() + pl := pi.PreferredLifetime() + + // If the preferred lifetime is greater than the valid lifetime, + // silently ignore the Prefix Information option, as per RFC 4862 + // section 5.5.3.c. + if pl > vl { + return + } + + prefix := pi.Subnet() + + // Check if we already have an auto-generated address for prefix. + for _, ref := range ndp.nic.endpoints { + if ref.protocol != header.IPv6ProtocolNumber { + continue + } + + if ref.configType != slaac { + continue + } + + addr := ref.ep.ID().LocalAddress + refAddrWithPrefix := tcpip.AddressWithPrefix{Address: addr, PrefixLen: ref.ep.PrefixLen()} + if refAddrWithPrefix.Subnet() != prefix { + continue + } + + // + // At this point, we know we are refreshing a SLAAC generated + // IPv6 address with the prefix, prefix. Do the work as outlined + // by RFC 4862 section 5.5.3.e. + // + + addrState, ok := ndp.autoGenAddresses[addr] + if !ok { + panic(fmt.Sprintf("must have an autoGenAddressess entry for the SLAAC generated IPv6 address %s", addr)) + } + + // TODO(b/143713887): Handle deprecating auto-generated address + // after the preferred lifetime. + + // As per RFC 4862 section 5.5.3.e, the valid lifetime of the + // address generated by SLAAC is as follows: + // + // 1) If the received Valid Lifetime is greater than 2 hours or + // greater than RemainingLifetime, set the valid lifetime of + // the address to the advertised Valid Lifetime. + // + // 2) If RemainingLifetime is less than or equal to 2 hours, + // ignore the advertised Valid Lifetime. + // + // 3) Otherwise, reset the valid lifetime of the address to 2 + // hours. + + // Handle the infinite valid lifetime separately as we do not + // keep a timer in this case. + if vl >= header.NDPInfiniteLifetime { + if addrState.invalidationTimer != nil { + // Valid lifetime was finite before, but now it + // is valid forever. + if !addrState.invalidationTimer.Stop() { + *addrState.doNotInvalidate = true + } + addrState.invalidationTimer = nil + addrState.validUntil = time.Time{} + ndp.autoGenAddresses[addr] = addrState + } + + return + } + + var effectiveVl time.Duration + var rl time.Duration + + // If the address was originally set to be valid forever, + // assume the remaining time to be the maximum possible value. + if addrState.invalidationTimer == nil { + rl = header.NDPInfiniteLifetime + } else { + rl = time.Until(addrState.validUntil) + } + + if vl > MinPrefixInformationValidLifetimeForUpdate || vl > rl { + effectiveVl = vl + } else if rl <= MinPrefixInformationValidLifetimeForUpdate { + ndp.autoGenAddresses[addr] = addrState + return + } else { + effectiveVl = MinPrefixInformationValidLifetimeForUpdate + } + + if addrState.invalidationTimer == nil { + addrState.invalidationTimer = ndp.autoGenAddrInvalidationTimer(addr, effectiveVl, addrState.doNotInvalidate) + } else { + if !addrState.invalidationTimer.Stop() { + *addrState.doNotInvalidate = true + } + addrState.invalidationTimer.Reset(effectiveVl) + } + + addrState.validUntil = time.Now().Add(effectiveVl) + ndp.autoGenAddresses[addr] = addrState + return + } + + // We do not already have an address within the prefix, prefix. Do the + // work as outlined by RFC 4862 section 5.5.3.d if n is configured + // to auto-generated global addresses by SLAAC. + + // Are we configured to auto-generate new global addresses? + if !ndp.configs.AutoGenGlobalAddresses { + return + } + + // If we do not already have an address for this prefix and the valid + // lifetime is 0, no need to do anything further, as per RFC 4862 + // section 5.5.3.d. + if vl == 0 { + return + } + + // Make sure the prefix is valid (as far as its length is concerned) to + // generate a valid IPv6 address from an interface identifier (IID), as + // per RFC 4862 sectiion 5.5.3.d. + if prefix.Prefix() != validPrefixLenForAutoGen { + return + } + + // Only attempt to generate an interface-specific IID if we have a valid + // link address. + // + // TODO(b/141011931): Validate a LinkEndpoint's link address + // (provided by LinkEndpoint.LinkAddress) before reaching this + // point. + linkAddr := ndp.nic.linkEP.LinkAddress() + if !header.IsValidUnicastEthernetAddress(linkAddr) { + return + } + + // Generate an address within prefix from the modified EUI-64 of ndp's + // NIC's Ethernet MAC address. + addrBytes := make([]byte, header.IPv6AddressSize) + copy(addrBytes[:header.IIDOffsetInIPv6Address], prefix.ID()[:header.IIDOffsetInIPv6Address]) + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:]) + addr := tcpip.Address(addrBytes) + addrWithPrefix := tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: validPrefixLenForAutoGen, + } + + // If the nic already has this address, do nothing further. + if ndp.nic.hasPermanentAddrLocked(addr) { + return + } + + // Inform the integrator that we have a new SLAAC address. + if ndp.nic.stack.ndpDisp == nil { + return + } + if !ndp.nic.stack.ndpDisp.OnAutoGenAddress(ndp.nic.ID(), addrWithPrefix) { + // Informed by the integrator not to add the address. + return + } + + if _, err := ndp.nic.addAddressLocked(tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: addrWithPrefix, + }, FirstPrimaryEndpoint, permanent, slaac); err != nil { + panic(err) + } + + // Setup the timers to deprecate and invalidate this newly generated + // address. + + // TODO(b/143713887): Handle deprecating auto-generated addresses + // after the preferred lifetime. + + var doNotInvalidate bool + var vTimer *time.Timer + if vl < header.NDPInfiniteLifetime { + vTimer = ndp.autoGenAddrInvalidationTimer(addr, vl, &doNotInvalidate) + } + + ndp.autoGenAddresses[addr] = autoGenAddressState{ + invalidationTimer: vTimer, + doNotInvalidate: &doNotInvalidate, + validUntil: time.Now().Add(vl), + } +} + +// invalidateAutoGenAddress invalidates an auto-generated address. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) invalidateAutoGenAddress(addr tcpip.Address) { + if !ndp.cleanupAutoGenAddrResourcesAndNotify(addr) { + return + } + + ndp.nic.removePermanentAddressLocked(addr) +} + +// cleanupAutoGenAddrResourcesAndNotify cleans up an invalidated auto-generated +// address's resources from ndp. If the stack has an NDP dispatcher, it will +// be notified that addr has been invalidated. +// +// Returns true if ndp had resources for addr to cleanup. +// +// The NIC that ndp belongs to MUST be locked. +func (ndp *ndpState) cleanupAutoGenAddrResourcesAndNotify(addr tcpip.Address) bool { + state, ok := ndp.autoGenAddresses[addr] + + if !ok { + return false + } + + if state.invalidationTimer != nil { + state.invalidationTimer.Stop() + state.invalidationTimer = nil + *state.doNotInvalidate = true + } + + state.doNotInvalidate = nil + + delete(ndp.autoGenAddresses, addr) + + if ndp.nic.stack.ndpDisp != nil { + ndp.nic.stack.ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: validPrefixLenForAutoGen, + }) + } + + return true +} + +// autoGenAddrInvalidationTimer returns a new invalidation timer for an +// auto-generated address that fires after vl. +// +// doNotInvalidate is used to inform the timer when it fires at the same time +// that an auto-generated address's valid lifetime gets refreshed. See +// autoGenAddrState.doNotInvalidate for more details. +func (ndp *ndpState) autoGenAddrInvalidationTimer(addr tcpip.Address, vl time.Duration, doNotInvalidate *bool) *time.Timer { + return time.AfterFunc(vl, func() { + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + + if *doNotInvalidate { + *doNotInvalidate = false + return + } + + ndp.invalidateAutoGenAddress(addr) + }) +} diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 5b901f947..d8e7ce67e 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -38,7 +38,7 @@ const ( linkAddr1 = "\x02\x02\x03\x04\x05\x06" linkAddr2 = "\x02\x02\x03\x04\x05\x07" linkAddr3 = "\x02\x02\x03\x04\x05\x08" - defaultTimeout = 250 * time.Millisecond + defaultTimeout = 100 * time.Millisecond ) var ( @@ -47,6 +47,31 @@ var ( llAddr3 = header.LinkLocalAddr(linkAddr3) ) +// prefixSubnetAddr returns a prefix (Address + Length), the prefix's equivalent +// tcpip.Subnet, and an address where the lower half of the address is composed +// of the EUI-64 of linkAddr if it is a valid unicast ethernet address. +func prefixSubnetAddr(offset uint8, linkAddr tcpip.LinkAddress) (tcpip.AddressWithPrefix, tcpip.Subnet, tcpip.AddressWithPrefix) { + prefixBytes := []byte{1, 2, 3, 4, 5, 6, 7, 8 + offset, 0, 0, 0, 0, 0, 0, 0, 0} + prefix := tcpip.AddressWithPrefix{ + Address: tcpip.Address(prefixBytes), + PrefixLen: 64, + } + + subnet := prefix.Subnet() + + var addr tcpip.AddressWithPrefix + if header.IsValidUnicastEthernetAddress(linkAddr) { + addrBytes := []byte(subnet.ID()) + header.EthernetAdddressToModifiedEUI64IntoBuf(linkAddr, addrBytes[header.IIDOffsetInIPv6Address:]) + addr = tcpip.AddressWithPrefix{ + Address: tcpip.Address(addrBytes), + PrefixLen: 64, + } + } + + return prefix, subnet, addr +} + // TestDADDisabled tests that an address successfully resolves immediately // when DAD is not enabled (the default for an empty stack.Options). func TestDADDisabled(t *testing.T) { @@ -103,6 +128,29 @@ type ndpPrefixEvent struct { discovered bool } +type ndpAutoGenAddrEventType int + +const ( + newAddr ndpAutoGenAddrEventType = iota + invalidatedAddr +) + +type ndpAutoGenAddrEvent struct { + nicID tcpip.NICID + addr tcpip.AddressWithPrefix + eventType ndpAutoGenAddrEventType +} + +type ndpRDNSS struct { + addrs []tcpip.Address + lifetime time.Duration +} + +type ndpRDNSSEvent struct { + nicID tcpip.NICID + rdnss ndpRDNSS +} + var _ stack.NDPDispatcher = (*ndpDispatcher)(nil) // ndpDispatcher implements NDPDispatcher so tests can know when various NDP @@ -113,6 +161,8 @@ type ndpDispatcher struct { rememberRouter bool prefixC chan ndpPrefixEvent rememberPrefix bool + autoGenAddrC chan ndpAutoGenAddrEvent + rdnssC chan ndpRDNSSEvent routeTable []tcpip.Route } @@ -211,7 +261,7 @@ func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpi } } - rt := make([]tcpip.Route, 0) + var rt []tcpip.Route exclude := tcpip.Route{ Destination: prefix, NIC: nicID, @@ -226,6 +276,40 @@ func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpi return rt } +func (n *ndpDispatcher) OnAutoGenAddress(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) bool { + if n.autoGenAddrC != nil { + n.autoGenAddrC <- ndpAutoGenAddrEvent{ + nicID, + addr, + newAddr, + } + } + return true +} + +func (n *ndpDispatcher) OnAutoGenAddressInvalidated(nicID tcpip.NICID, addr tcpip.AddressWithPrefix) { + if n.autoGenAddrC != nil { + n.autoGenAddrC <- ndpAutoGenAddrEvent{ + nicID, + addr, + invalidatedAddr, + } + } +} + +// Implements stack.NDPDispatcher.OnRecursiveDNSServerOption. +func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) { + if n.rdnssC != nil { + n.rdnssC <- ndpRDNSSEvent{ + nicID, + ndpRDNSS{ + addrs, + lifetime, + }, + } + } +} + // TestDADResolve tests that an address successfully resolves after performing // DAD for various values of DupAddrDetectTransmits and RetransmitTimer. // Included in the subtests is a test to make sure that an invalid @@ -247,6 +331,8 @@ func TestDADResolve(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent), } @@ -781,16 +867,33 @@ func raBuf(ip tcpip.Address, rl uint16) tcpip.PacketBuffer { // // Note, raBufWithPI does not populate any of the RA fields other than the // Router Lifetime. -func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink bool, vl uint32) tcpip.PacketBuffer { +func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) tcpip.PacketBuffer { flags := uint8(0) if onLink { - flags |= 128 + // The OnLink flag is the 7th bit in the flags byte. + flags |= 1 << 7 + } + if auto { + // The Address Auto-Configuration flag is the 6th bit in the + // flags byte. + flags |= 1 << 6 } + // A valid header.NDPPrefixInformation must be 30 bytes. buf := [30]byte{} + // The first byte in a header.NDPPrefixInformation is the Prefix Length + // field. buf[0] = uint8(prefix.PrefixLen) + // The 2nd byte within a header.NDPPrefixInformation is the Flags field. buf[1] = flags + // The Valid Lifetime field starts after the 2nd byte within a + // header.NDPPrefixInformation. binary.BigEndian.PutUint32(buf[2:], vl) + // The Preferred Lifetime field starts after the 6th byte within a + // header.NDPPrefixInformation. + binary.BigEndian.PutUint32(buf[6:], pl) + // The Prefix Address field starts after the 14th byte within a + // header.NDPPrefixInformation. copy(buf[14:], prefix.Address) return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{ header.NDPPrefixInformation(buf[:]), @@ -800,6 +903,8 @@ func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, on // TestNoRouterDiscovery tests that router discovery will not be performed if // configured not to. func TestNoRouterDiscovery(t *testing.T) { + t.Parallel() + // Being configured to discover routers means handle and // discover are set to true and forwarding is set to false. // This tests all possible combinations of the configurations, @@ -812,6 +917,8 @@ func TestNoRouterDiscovery(t *testing.T) { forwarding := i&4 == 0 t.Run(fmt.Sprintf("HandleRAs(%t), DiscoverDefaultRouters(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 10), } @@ -844,6 +951,8 @@ func TestNoRouterDiscovery(t *testing.T) { // TestRouterDiscoveryDispatcherNoRemember tests that the stack does not // remember a discovered router when the dispatcher asks it not to. func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 10), } @@ -909,6 +1018,8 @@ func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { } func TestRouterDiscovery(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 10), rememberRouter: true, @@ -1040,6 +1151,8 @@ func TestRouterDiscovery(t *testing.T) { // TestRouterDiscoveryMaxRouters tests that only // stack.MaxDiscoveredDefaultRouters discovered routers are remembered. func TestRouterDiscoveryMaxRouters(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 10), rememberRouter: true, @@ -1104,6 +1217,8 @@ func TestRouterDiscoveryMaxRouters(t *testing.T) { // TestNoPrefixDiscovery tests that prefix discovery will not be performed if // configured not to. func TestNoPrefixDiscovery(t *testing.T) { + t.Parallel() + prefix := tcpip.AddressWithPrefix{ Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"), PrefixLen: 64, @@ -1121,6 +1236,8 @@ func TestNoPrefixDiscovery(t *testing.T) { forwarding := i&4 == 0 t.Run(fmt.Sprintf("HandleRAs(%t), DiscoverOnLinkPrefixes(%t), Forwarding(%t)", handle, discover, forwarding), func(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ prefixC: make(chan ndpPrefixEvent, 10), } @@ -1140,7 +1257,7 @@ func TestNoPrefixDiscovery(t *testing.T) { } // Rx an RA with prefix with non-zero lifetime. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, 10)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 10, 0)) select { case <-ndpDisp.prefixC: @@ -1154,11 +1271,9 @@ func TestNoPrefixDiscovery(t *testing.T) { // TestPrefixDiscoveryDispatcherNoRemember tests that the stack does not // remember a discovered on-link prefix when the dispatcher asks it not to. func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { - prefix := tcpip.AddressWithPrefix{ - Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"), - PrefixLen: 64, - } - subnet := prefix.Subnet() + t.Parallel() + + prefix, subnet, _ := prefixSubnetAddr(0, "") ndpDisp := ndpDispatcher{ prefixC: make(chan ndpPrefixEvent, 10), @@ -1189,7 +1304,7 @@ func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { // Rx an RA with prefix with a short lifetime. const lifetime = 1 - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, lifetime)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, lifetime, 0)) select { case r := <-ndpDisp.prefixC: if r.nicID != 1 { @@ -1226,21 +1341,11 @@ func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { } func TestPrefixDiscovery(t *testing.T) { - prefix1 := tcpip.AddressWithPrefix{ - Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x00"), - PrefixLen: 64, - } - prefix2 := tcpip.AddressWithPrefix{ - Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x09\x00\x00\x00\x00\x00\x00\x00\x00"), - PrefixLen: 64, - } - prefix3 := tcpip.AddressWithPrefix{ - Address: tcpip.Address("\x01\x02\x03\x04\x05\x06\x07\x09\x0a\x00\x00\x00\x00\x00\x00\x00"), - PrefixLen: 72, - } - subnet1 := prefix1.Subnet() - subnet2 := prefix2.Subnet() - subnet3 := prefix3.Subnet() + t.Parallel() + + prefix1, subnet1, _ := prefixSubnetAddr(0, "") + prefix2, subnet2, _ := prefixSubnetAddr(1, "") + prefix3, subnet3, _ := prefixSubnetAddr(2, "") ndpDisp := ndpDispatcher{ prefixC: make(chan ndpPrefixEvent, 10), @@ -1281,7 +1386,7 @@ func TestPrefixDiscovery(t *testing.T) { // Receive an RA with prefix1 in an NDP Prefix Information option (PI) // with zero valid lifetime. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, 0)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0)) select { case <-ndpDisp.prefixC: t.Fatal("unexpectedly discovered a prefix with 0 lifetime") @@ -1290,7 +1395,7 @@ func TestPrefixDiscovery(t *testing.T) { // Receive an RA with prefix1 in an NDP Prefix Information option (PI) // with non-zero lifetime. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, 100)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 100, 0)) waitForEvent(subnet1, true, defaultTimeout) // Should have added a device route for subnet1 through the nic. @@ -1299,7 +1404,7 @@ func TestPrefixDiscovery(t *testing.T) { } // Receive an RA with prefix2 in a PI. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, 100)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, 100, 0)) waitForEvent(subnet2, true, defaultTimeout) // Should have added a device route for subnet2 through the nic. @@ -1308,7 +1413,7 @@ func TestPrefixDiscovery(t *testing.T) { } // Receive an RA with prefix3 in a PI. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, 100)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 100, 0)) waitForEvent(subnet3, true, defaultTimeout) // Should have added a device route for subnet3 through the nic. @@ -1317,7 +1422,7 @@ func TestPrefixDiscovery(t *testing.T) { } // Receive an RA with prefix1 in a PI with lifetime = 0. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, 0)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, false, 0, 0)) waitForEvent(subnet1, false, defaultTimeout) // Should have removed the device route for subnet1 through the nic. @@ -1327,7 +1432,7 @@ func TestPrefixDiscovery(t *testing.T) { // Receive an RA with prefix2 in a PI with lesser lifetime. lifetime := uint32(2) - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, lifetime)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, false, lifetime, 0)) select { case <-ndpDisp.prefixC: t.Fatal("unexpectedly received prefix event when updating lifetime") @@ -1349,7 +1454,7 @@ func TestPrefixDiscovery(t *testing.T) { } // Receive RA to invalidate prefix3. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, 0)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix3, true, false, 0, 0)) waitForEvent(subnet3, false, defaultTimeout) // Should not have any routes. @@ -1364,10 +1469,10 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { // invalidate the prefix. const testInfiniteLifetimeSeconds = 2 const testInfiniteLifetime = testInfiniteLifetimeSeconds * time.Second - saved := header.NDPPrefixInformationInfiniteLifetime - header.NDPPrefixInformationInfiniteLifetime = testInfiniteLifetime + saved := header.NDPInfiniteLifetime + header.NDPInfiniteLifetime = testInfiniteLifetime defer func() { - header.NDPPrefixInformationInfiniteLifetime = saved + header.NDPInfiniteLifetime = saved }() prefix := tcpip.AddressWithPrefix{ @@ -1415,7 +1520,7 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { // Receive an RA with prefix in an NDP Prefix Information option (PI) // with infinite valid lifetime which should not get invalidated. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, testInfiniteLifetimeSeconds)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0)) waitForEvent(true, defaultTimeout) select { case <-ndpDisp.prefixC: @@ -1425,16 +1530,16 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { // Receive an RA with finite lifetime. // The prefix should get invalidated after 1s. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, testInfiniteLifetimeSeconds-1)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0)) waitForEvent(false, testInfiniteLifetime) // Receive an RA with finite lifetime. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, testInfiniteLifetimeSeconds-1)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds-1, 0)) waitForEvent(true, defaultTimeout) // Receive an RA with prefix with an infinite lifetime. // The prefix should not be invalidated. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, testInfiniteLifetimeSeconds)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds, 0)) select { case <-ndpDisp.prefixC: t.Fatal("unexpectedly invalidated a prefix with infinite lifetime") @@ -1443,7 +1548,7 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { // Receive an RA with a prefix with a lifetime value greater than the // set infinite lifetime value. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, testInfiniteLifetimeSeconds+1)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, testInfiniteLifetimeSeconds+1, 0)) select { case <-ndpDisp.prefixC: t.Fatal("unexpectedly invalidated a prefix with infinite lifetime") @@ -1452,13 +1557,15 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { // Receive an RA with 0 lifetime. // The prefix should get invalidated. - e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, 0)) + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, false, 0, 0)) waitForEvent(false, defaultTimeout) } // TestPrefixDiscoveryMaxRouters tests that only // stack.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { + t.Parallel() + ndpDisp := ndpDispatcher{ prefixC: make(chan ndpPrefixEvent, stack.MaxDiscoveredOnLinkPrefixes+3), rememberPrefix: true, @@ -1537,3 +1644,606 @@ func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { t.Fatalf("got GetRouteTable = %v, want = %v", got, expectedRt) } } + +// Checks to see if list contains an IPv6 address, item. +func contains(list []tcpip.ProtocolAddress, item tcpip.AddressWithPrefix) bool { + protocolAddress := tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: item, + } + + for _, i := range list { + if i == protocolAddress { + return true + } + } + + return false +} + +// TestNoAutoGenAddr tests that SLAAC is not performed when configured not to. +func TestNoAutoGenAddr(t *testing.T) { + t.Parallel() + + prefix, _, _ := prefixSubnetAddr(0, "") + + // Being configured to auto-generate addresses means handle and + // autogen are set to true and forwarding is set to false. + // This tests all possible combinations of the configurations, + // except for the configuration where handle = true, autogen = + // true and forwarding = false (the required configuration to do + // SLAAC) - that will done in other tests. + for i := 0; i < 7; i++ { + handle := i&1 != 0 + autogen := i&2 != 0 + forwarding := i&4 == 0 + + t.Run(fmt.Sprintf("HandleRAs(%t), AutoGenAddr(%t), Forwarding(%t)", handle, autogen, forwarding), func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: handle, + AutoGenGlobalAddresses: autogen, + }, + NDPDisp: &ndpDisp, + }) + s.SetForwarding(forwarding) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Rx an RA with prefix with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, false, true, 10, 0)) + + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address when configured not to") + case <-time.After(defaultTimeout): + } + }) + } +} + +// TestAutoGenAddr tests that an address is properly generated and invalidated +// when configured to do so. +func TestAutoGenAddr(t *testing.T) { + const newMinVL = 2 + newMinVLDuration := newMinVL * time.Second + saved := stack.MinPrefixInformationValidLifetimeForUpdate + defer func() { + stack.MinPrefixInformationValidLifetimeForUpdate = saved + }() + stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + + prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) + prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + waitForEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType, timeout time.Duration) { + t.Helper() + + select { + case r := <-ndpDisp.autoGenAddrC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.eventType != eventType { + t.Fatalf("got r.eventType = %v, want = %v", r.eventType, eventType) + } + case <-time.After(timeout): + t.Fatal("timeout waiting for addr auto gen event") + } + } + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with zero valid lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 0, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address with 0 lifetime") + case <-time.After(defaultTimeout): + } + + // Receive an RA with prefix1 in an NDP Prefix Information option (PI) + // with non-zero lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 100, 0)) + waitForEvent(addr1, newAddr, defaultTimeout) + if !contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Receive an RA with prefix2 in an NDP Prefix Information option (PI) + // with preferred lifetime > valid lifetime + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, 5, 6)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address with preferred lifetime > valid lifetime") + case <-time.After(defaultTimeout): + } + + // Receive an RA with prefix2 in a PI. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix2, true, true, 100, 0)) + waitForEvent(addr2, newAddr, defaultTimeout) + if !contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr2) { + t.Fatalf("Should have %s in the list of addresses", addr2) + } + + // Refresh valid lifetime for addr of prefix1. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix1, true, true, newMinVL, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly auto-generated an address when we already have an address for a prefix") + case <-time.After(defaultTimeout): + } + + // Wait for addr of prefix1 to be invalidated. + waitForEvent(addr1, invalidatedAddr, newMinVLDuration+defaultTimeout) + if contains(s.NICInfo()[1].ProtocolAddresses, addr1) { + t.Fatalf("Should not have %s in the list of addresses", addr1) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr2) { + t.Fatalf("Should have %s in the list of addresses", addr2) + } +} + +// TestAutoGenAddrValidLifetimeUpdates tests that the valid lifetime of an +// auto-generated address only gets updated when required to, as specified in +// RFC 4862 section 5.5.3.e. +func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { + const infiniteVL = 4294967295 + const newMinVL = 5 + saved := stack.MinPrefixInformationValidLifetimeForUpdate + defer func() { + stack.MinPrefixInformationValidLifetimeForUpdate = saved + }() + stack.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + tests := []struct { + name string + ovl uint32 + nvl uint32 + evl uint32 + }{ + // Should update the VL to the minimum VL for updating if the + // new VL is less than newMinVL but was originally greater than + // it. + { + "LargeVLToVLLessThanMinVLForUpdate", + 9999, + 1, + newMinVL, + }, + { + "LargeVLTo0", + 9999, + 0, + newMinVL, + }, + { + "InfiniteVLToVLLessThanMinVLForUpdate", + infiniteVL, + 1, + newMinVL, + }, + { + "InfiniteVLTo0", + infiniteVL, + 0, + newMinVL, + }, + + // Should not update VL if original VL was less than newMinVL + // and the new VL is also less than newMinVL. + { + "ShouldNotUpdateWhenBothOldAndNewAreLessThanMinVLForUpdate", + newMinVL - 1, + newMinVL - 3, + newMinVL - 1, + }, + + // Should take the new VL if the new VL is greater than the + // remaining time or is greater than newMinVL. + { + "MorethanMinVLToLesserButStillMoreThanMinVLForUpdate", + newMinVL + 5, + newMinVL + 3, + newMinVL + 3, + }, + { + "SmallVLToGreaterVLButStillLessThanMinVLForUpdate", + newMinVL - 3, + newMinVL - 1, + newMinVL - 1, + }, + { + "SmallVLToGreaterVLThatIsMoreThaMinVLForUpdate", + newMinVL - 3, + newMinVL + 1, + newMinVL + 1, + }, + } + + const delta = 500 * time.Millisecond + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Receive an RA with prefix with initial VL, test.ovl. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.ovl, 0)) + select { + case r := <-ndpDisp.autoGenAddrC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.eventType != newAddr { + t.Fatalf("got r.eventType = %v, want = %v", r.eventType, newAddr) + } + case <-time.After(defaultTimeout): + t.Fatal("timeout waiting for addr auto gen event") + } + + // Receive an new RA with prefix with new VL, test.nvl. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, test.nvl, 0)) + + // + // Validate that the VL for the address got set to + // test.evl. + // + + // Make sure we do not get any invalidation events + // until atleast 500ms (delta) before test.evl. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatalf("unexpectedly received an auto gen addr event") + case <-time.After(time.Duration(test.evl)*time.Second - delta): + } + + // Wait for another second (2x delta), but now we expect + // the invalidation event. + select { + case r := <-ndpDisp.autoGenAddrC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.eventType != invalidatedAddr { + t.Fatalf("got r.eventType = %v, want = %v", r.eventType, newAddr) + } + case <-time.After(2 * delta): + t.Fatal("timeout waiting for addr auto gen event") + } + }) + } +} + +// TestAutoGenAddrRemoval tests that when auto-generated addresses are removed +// by the user, its resources will be cleaned up and an invalidation event will +// be sent to the integrator. +func TestAutoGenAddrRemoval(t *testing.T) { + t.Parallel() + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Receive an RA with prefix with its valid lifetime = lifetime. + const lifetime = 5 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetime, 0)) + select { + case r := <-ndpDisp.autoGenAddrC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.eventType != newAddr { + t.Fatalf("got r.eventType = %v, want = %v", r.eventType, newAddr) + } + case <-time.After(defaultTimeout): + t.Fatal("timeout waiting for addr auto gen event") + } + + // Remove the address. + if err := s.RemoveAddress(1, addr.Address); err != nil { + t.Fatalf("RemoveAddress(_, %s) = %s", addr.Address, err) + } + + // Should get the invalidation event immediately. + select { + case r := <-ndpDisp.autoGenAddrC: + if r.nicID != 1 { + t.Fatalf("got r.nicID = %d, want = 1", r.nicID) + } + if r.addr != addr { + t.Fatalf("got r.addr = %s, want = %s", r.addr, addr) + } + if r.eventType != invalidatedAddr { + t.Fatalf("got r.eventType = %v, want = %v", r.eventType, newAddr) + } + case <-time.After(defaultTimeout): + t.Fatal("timeout waiting for addr auto gen event") + } + + // Wait for the original valid lifetime to make sure the original timer + // got stopped/cleaned up. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatalf("unexpectedly received an auto gen addr event") + case <-time.After(lifetime*time.Second + defaultTimeout): + } +} + +// TestAutoGenAddrStaticConflict tests that if SLAAC generates an address that +// is already assigned to the NIC, the static address remains. +func TestAutoGenAddrStaticConflict(t *testing.T) { + t.Parallel() + + prefix, _, addr := prefixSubnetAddr(0, linkAddr1) + + ndpDisp := ndpDispatcher{ + autoGenAddrC: make(chan ndpAutoGenAddrEvent, 10), + } + e := channel.New(10, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + }) + + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Add the address as a static address before SLAAC tries to add it. + if err := s.AddProtocolAddress(1, tcpip.ProtocolAddress{Protocol: header.IPv6ProtocolNumber, AddressWithPrefix: addr}); err != nil { + t.Fatalf("AddAddress(_, %d, %s) = %s", header.IPv6ProtocolNumber, addr.Address, err) + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Receive a PI where the generated address will be the same as the one + // that we already have assigned statically. + const lifetime = 5 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, lifetime, 0)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly received an auto gen addr event for an address we already have statically") + case <-time.After(defaultTimeout): + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } + + // Should not get an invalidation event after the PI's invalidation + // time. + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly received an auto gen addr event") + case <-time.After(lifetime*time.Second + defaultTimeout): + } + if !contains(s.NICInfo()[1].ProtocolAddresses, addr) { + t.Fatalf("Should have %s in the list of addresses", addr1) + } +} + +// TestNDPRecursiveDNSServerDispatch tests that we properly dispatch an event +// to the integrator when an RA is received with the NDP Recursive DNS Server +// option with at least one valid address. +func TestNDPRecursiveDNSServerDispatch(t *testing.T) { + t.Parallel() + + tests := []struct { + name string + opt header.NDPRecursiveDNSServer + expected *ndpRDNSS + }{ + { + "Unspecified", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + }), + nil, + }, + { + "Multicast", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, + }), + nil, + }, + { + "OptionTooSmall", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, + }), + nil, + }, + { + "0Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + }), + nil, + }, + { + "Valid1Address", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + }, + 2 * time.Second, + }, + }, + { + "Valid2Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02", + }, + time.Second, + }, + }, + { + "Valid3Addresses", + header.NDPRecursiveDNSServer([]byte{ + 0, 0, + 0, 0, 0, 0, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 1, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 2, + 1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 3, + }), + &ndpRDNSS{ + []tcpip.Address{ + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x01", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x02", + "\x01\x02\x03\x04\x05\x06\x07\x08\x00\x00\x00\x00\x00\x00\x00\x03", + }, + 0, + }, + }, + } + + for _, test := range tests { + test := test + + t.Run(test.name, func(t *testing.T) { + t.Parallel() + + ndpDisp := ndpDispatcher{ + // We do not expect more than a single RDNSS + // event at any time for this test. + rdnssC: make(chan ndpRDNSSEvent, 1), + } + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, header.NDPOptionsSerializer{test.opt})) + + if test.expected != nil { + select { + case e := <-ndpDisp.rdnssC: + if e.nicID != 1 { + t.Errorf("got rdnss nicID = %d, want = 1", e.nicID) + } + if diff := cmp.Diff(e.rdnss.addrs, test.expected.addrs); diff != "" { + t.Errorf("rdnss addrs mismatch (-want +got):\n%s", diff) + } + if e.rdnss.lifetime != test.expected.lifetime { + t.Errorf("got rdnss lifetime = %s, want = %s", e.rdnss.lifetime, test.expected.lifetime) + } + default: + t.Fatal("expected an RDNSS option event") + } + } + + // Should have no more RDNSS options. + select { + case e := <-ndpDisp.rdnssC: + t.Fatalf("unexpectedly got a new RDNSS option event: %+v", e) + default: + } + }) + } +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 3f8d7312c..e8401c673 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -115,10 +115,11 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback }, }, ndp: ndpState{ - configs: stack.ndpConfigs, - dad: make(map[tcpip.Address]dadState), - defaultRouters: make(map[tcpip.Address]defaultRouterState), - onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), + configs: stack.ndpConfigs, + dad: make(map[tcpip.Address]dadState), + defaultRouters: make(map[tcpip.Address]defaultRouterState), + onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), + autoGenAddresses: make(map[tcpip.Address]autoGenAddressState), }, } nic.ndp.nic = nic @@ -244,6 +245,20 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN return nil } +// hasPermanentAddrLocked returns true if n has a permanent (including currently +// tentative) address, addr. +func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { + ref, ok := n.endpoints[NetworkEndpointID{addr}] + + if !ok { + return false + } + + kind := ref.getKind() + + return kind == permanent || kind == permanentTentative +} + func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, n.promiscuous) } @@ -335,7 +350,7 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t Address: address, PrefixLen: netProto.DefaultPrefixLen(), }, - }, peb, temporary) + }, peb, temporary, static) n.mu.Unlock() return ref @@ -384,10 +399,10 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p } } - return n.addAddressLocked(protocolAddress, peb, permanent) + return n.addAddressLocked(protocolAddress, peb, permanent, static) } -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind) (*referencedNetworkEndpoint, *tcpip.Error) { +func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType) (*referencedNetworkEndpoint, *tcpip.Error) { // TODO(b/141022673): Validate IP address before adding them. // Sanity check. @@ -417,11 +432,12 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar } ref := &referencedNetworkEndpoint{ - refs: 1, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, + refs: 1, + ep: ep, + nic: n, + protocol: protocolAddress.Protocol, + kind: kind, + configType: configType, } // Set up cache if link address resolution exists for this protocol. @@ -624,9 +640,18 @@ func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { isIPv6Unicast := r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr) - // If we are removing a tentative IPv6 unicast address, stop DAD. - if isIPv6Unicast && kind == permanentTentative { - n.ndp.stopDuplicateAddressDetection(addr) + if isIPv6Unicast { + // If we are removing a tentative IPv6 unicast address, stop + // DAD. + if kind == permanentTentative { + n.ndp.stopDuplicateAddressDetection(addr) + } + + // If we are removing an address generated via SLAAC, cleanup + // its SLAAC resources and notify the integrator. + if r.configType == slaac { + n.ndp.cleanupAutoGenAddrResourcesAndNotify(addr) + } } r.setKind(permanentExpired) @@ -989,7 +1014,7 @@ const ( // removing the permanent address from the NIC. permanent - // An expired permanent endoint is a permanent endoint that had its address + // An expired permanent endpoint is a permanent endpoint that had its address // removed from the NIC, and it is waiting to be removed once no more routes // hold a reference to it. This is achieved by decreasing its reference count // by 1. If its address is re-added before the endpoint is removed, its type @@ -1035,6 +1060,19 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep } } +type networkEndpointConfigType int32 + +const ( + // A statically configured endpoint is an address that was added by + // some user-specified action (adding an explicit address, joining a + // multicast group). + static networkEndpointConfigType = iota + + // A slaac configured endpoint is an IPv6 endpoint that was + // added by SLAAC as per RFC 4862 section 5.5.3. + slaac +) + type referencedNetworkEndpoint struct { ep NetworkEndpoint nic *NIC @@ -1050,6 +1088,10 @@ type referencedNetworkEndpoint struct { // networkEndpointKind must only be accessed using {get,set}Kind(). kind networkEndpointKind + + // configType is the method that was used to configure this endpoint. + // This must never change after the endpoint is added to a NIC. + configType networkEndpointConfigType } func (r *referencedNetworkEndpoint) getKind() networkEndpointKind { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 5746043cc..f62fd729f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -576,6 +576,11 @@ type KeepaliveIntervalOption time.Duration // closed. type KeepaliveCountOption int +// TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user +// specified timeout for a given TCP connection. +// See: RFC5482 for details. +type TCPUserTimeoutOption time.Duration + // CongestionControlOption is used by SetSockOpt/GetSockOpt to set/get // the current congestion control algorithm. type CongestionControlOption string @@ -924,6 +929,14 @@ type TCPStats struct { // ESTABLISHED state or the CLOSE-WAIT state. EstablishedResets *StatCounter + // EstablishedClosed is the number of times established TCP connections + // made a transition to CLOSED state. + EstablishedClosed *StatCounter + + // EstablishedTimedout is the number of times an established connection + // was reset because of keep-alive time out. + EstablishedTimedout *StatCounter + // ListenOverflowSynDrop is the number of times the listen queue overflowed // and a SYN was dropped. ListenOverflowSynDrop *StatCounter diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index dd1728f9c..3b353d56c 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -28,6 +28,7 @@ go_library( "forwarder.go", "protocol.go", "rcv.go", + "rcv_state.go", "reno.go", "sack.go", "sack_scoreboard.go", @@ -52,6 +53,7 @@ go_library( "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", "//pkg/tcpip/iptables", + "//pkg/tcpip/ports", "//pkg/tcpip/seqnum", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index f543a6105..5422ae80c 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -242,6 +242,13 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() + // Now inherit any socket options that should be inherited from the + // listening endpoint. + // In case of Forwarder listenEP will be nil and hence this check. + if l.listenEP != nil { + l.listenEP.propagateInheritableOptions(n) + } + // Register new endpoint so that packets are routed to it. if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { n.Close() @@ -298,8 +305,6 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, err } ep.mu.Lock() - ep.stack.Stats().TCP.CurrentEstablished.Increment() - ep.state = StateEstablished ep.isConnectNotified = true ep.mu.Unlock() @@ -352,6 +357,14 @@ func (e *endpoint) deliverAccepted(n *endpoint) { } } +// propagateInheritableOptions propagates any options set on the listening +// endpoint to the newly created endpoint. +func (e *endpoint) propagateInheritableOptions(n *endpoint) { + e.mu.Lock() + n.userTimeout = e.userTimeout + e.mu.Unlock() +} + // handleSynSegment is called in its own goroutine once the listening endpoint // receives a SYN segment. It is responsible for completing the handshake and // queueing the new endpoint for acceptance. @@ -546,6 +559,8 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n.tsOffset = 0 // Switch state to connected. + // We do not use transitionToStateEstablishedLocked here as there is + // no handshake state available when doing a SYN cookie based accept. n.stack.Stats().TCP.CurrentEstablished.Increment() n.state = StateEstablished n.isConnectNotified = true diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 16f8aea12..cdd69f360 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -218,6 +218,14 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // acceptable if the ack field acknowledges the SYN. if s.flagIsSet(header.TCPFlagRst) { if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == h.iss+1 { + // RFC 793, page 67, states that "If the RST bit is set [and] If the ACK + // was acceptable then signal the user "error: connection reset", drop + // the segment, enter CLOSED state, delete TCB, and return." + h.ep.mu.Lock() + h.ep.workerCleanup = true + h.ep.mu.Unlock() + // Although the RFC above calls out ECONNRESET, Linux actually returns + // ECONNREFUSED here so we do as well. return tcpip.ErrConnectionRefused } return nil @@ -252,6 +260,11 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // and the handshake is completed. if s.flagIsSet(header.TCPFlagAck) { h.state = handshakeCompleted + + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd>>h.effectiveRcvWndScale()) return nil } @@ -352,6 +365,10 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { h.ep.updateRecentTimestamp(s.parsedOptions.TSVal, h.ackNum, s.sequenceNumber) } h.state = handshakeCompleted + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + return nil } @@ -853,7 +870,7 @@ func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { } e.state = StateError e.HardError = err - if err != tcpip.ErrConnectionReset { + if err != tcpip.ErrConnectionReset && err != tcpip.ErrTimeout { // The exact sequence number to be used for the RST is the same as the // one used by Linux. We need to handle the case of window being shrunk // which can cause sndNxt to be outside the acceptable window on the @@ -880,6 +897,30 @@ func (e *endpoint) completeWorkerLocked() { } } +// transitionToStateEstablisedLocked transitions a given endpoint +// to an established state using the handshake parameters provided. +// It also initializes sender/receiver if required. +func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { + if e.snd == nil { + // Transfer handshake state to TCP connection. We disable + // receive window scaling if the peer doesn't support it + // (indicated by a negative send window scale). + e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) + } + if e.rcv == nil { + rcvBufSize := seqnum.Size(e.receiveBufferSize()) + e.rcvListMu.Lock() + e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) + // Bootstrap the auto tuning algorithm. Starting at zero will + // result in a really large receive window after the first auto + // tuning adjustment. + e.rcvAutoParams.prevCopied = int(h.rcvWnd) + e.rcvListMu.Unlock() + } + h.ep.stack.Stats().TCP.CurrentEstablished.Increment() + e.state = StateEstablished +} + // transitionToStateCloseLocked ensures that the endpoint is // cleaned up from the transport demuxer, "before" moving to // StateClose. This will ensure that no packet will be @@ -891,6 +932,7 @@ func (e *endpoint) transitionToStateCloseLocked() { } e.cleanupLocked() e.state = StateClose + e.stack.Stats().TCP.EstablishedClosed.Increment() } // tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed @@ -1053,14 +1095,27 @@ func (e *endpoint) handleSegments() *tcpip.Error { // keepalive packets periodically when the connection is idle. If we don't hear // from the other side after a number of tries, we terminate the connection. func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { + e.mu.RLock() + userTimeout := e.userTimeout + e.mu.RUnlock() + e.keepalive.Lock() if !e.keepalive.enabled || !e.keepalive.timer.checkExpiration() { e.keepalive.Unlock() return nil } + // If a userTimeout is set then abort the connection if it is + // exceeded. + if userTimeout != 0 && time.Since(e.rcv.lastRcvdAckTime) >= userTimeout && e.keepalive.unacked > 0 { + e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() + return tcpip.ErrTimeout + } + if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } @@ -1077,7 +1132,6 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { // whether it is enabled for this endpoint. func (e *endpoint) resetKeepaliveTimer(receivedData bool) { e.keepalive.Lock() - defer e.keepalive.Unlock() if receivedData { e.keepalive.unacked = 0 } @@ -1085,6 +1139,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { // data to send. if !e.keepalive.enabled || e.snd == nil || e.snd.sndUna != e.snd.sndNxt { e.keepalive.timer.disable() + e.keepalive.Unlock() return } if e.keepalive.unacked > 0 { @@ -1092,6 +1147,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { } else { e.keepalive.timer.enable(e.keepalive.idle) } + e.keepalive.Unlock() } // disableKeepaliveTimer stops the keepalive timer. @@ -1146,8 +1202,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() - e.stack.Stats().TCP.EstablishedResets.Increment() - e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateError e.HardError = err @@ -1156,25 +1210,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return err } - - // Transfer handshake state to TCP connection. We disable - // receive window scaling if the peer doesn't support it - // (indicated by a negative send window scale). - e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) - - rcvBufSize := seqnum.Size(e.receiveBufferSize()) - e.rcvListMu.Lock() - e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) - // boot strap the auto tuning algorithm. Starting at zero will - // result in a large step function on the first proper causing - // the window to just go to a really large value after the first - // RTT itself. - e.rcvAutoParams.prevCopied = initialRcvWnd - e.rcvListMu.Unlock() - e.stack.Stats().TCP.CurrentEstablished.Increment() - e.mu.Lock() - e.state = StateEstablished - e.mu.Unlock() } e.keepalive.timer.init(&e.keepalive.waker) @@ -1225,6 +1260,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { w: &e.snd.resendWaker, f: func() *tcpip.Error { if !e.snd.retransmitTimerExpired() { + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } return nil @@ -1375,7 +1411,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Mark endpoint as closed. e.mu.Lock() if e.state != StateError { - e.stack.Stats().TCP.EstablishedResets.Increment() e.stack.Stats().TCP.CurrentEstablished.Decrement() e.transitionToStateCloseLocked() } @@ -1392,6 +1427,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { if s == nil { break } + e.tryDeliverSegmentFromClosedEndpoint(s) } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 9d4a87e30..dd8b47cbe 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -30,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tmutex" @@ -340,9 +341,11 @@ type endpoint struct { // TCP should never broadcast but Linux nevertheless supports enabling/ // disabling SO_BROADCAST, albeit as a NOOP. broadcast bool + // Values used to reserve a port or register a transport endpoint // (which ever happens first). boundBindToDevice tcpip.NICID + boundPortFlags ports.Flags // effectiveNetProtos contains the network protocols actually in use. In // most cases it will only contain "netProto", but in cases like IPv6 @@ -472,6 +475,12 @@ type endpoint struct { // without hearing a response, the connection is closed. keepalive keepalive + // userTimeout if non-zero specifies a user specified timeout for + // a connection w/ pending data to send. A connection that has pending + // unacked data will be forcibily aborted if the timeout is reached + // without any data being acked. + userTimeout time.Duration + // pendingAccepted is a synchronization primitive used to track number // of connections that are queued up to be delivered to the accepted // channel. We use this to ensure that all goroutines blocked on writing @@ -737,9 +746,10 @@ func (e *endpoint) Close() { e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.isPortReserved = false e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } // Mark endpoint as closed. @@ -800,10 +810,11 @@ func (e *endpoint) cleanupLocked() { } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.isPortReserved = false } e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} e.route.Release() e.stack.CompleteTransportEndpointCleanup(e) @@ -1329,6 +1340,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.notifyProtocolGoroutine(notifyKeepaliveChanged) return nil + case tcpip.TCPUserTimeoutOption: + e.mu.Lock() + e.userTimeout = time.Duration(v) + e.mu.Unlock() + return nil + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -1587,6 +1604,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.keepalive.Unlock() return nil + case *tcpip.TCPUserTimeoutOption: + e.mu.Lock() + *o = tcpip.TCPUserTimeoutOption(e.userTimeout) + e.mu.Unlock() + return nil + case *tcpip.OutOfBandInlineOption: // We don't currently support disabling this option. *o = 1 @@ -1775,7 +1798,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } // reusePort is false below because connect cannot reuse a port even if // reusePort was set. - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.ID.LocalAddress, p, false /* reusePort */, e.bindToDevice) { + if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.ID.LocalAddress, p, ports.Flags{LoadBalanced: false}, e.bindToDevice) { return false, nil } @@ -1802,7 +1825,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.isPortReserved = false } @@ -2034,28 +2057,33 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) + flags := ports.Flags{ + LoadBalanced: e.reusePort, + } + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, flags, e.bindToDevice) if err != nil { return err } e.boundBindToDevice = e.bindToDevice + e.boundPortFlags = flags e.isPortReserved = true e.effectiveNetProtos = netProtos e.ID.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func(bindToDevice tcpip.NICID) { + defer func(portFlags ports.Flags, bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, portFlags, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil e.ID.LocalPort = 0 e.ID.LocalAddress = "" e.boundNICID = 0 e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } - }(e.boundBindToDevice) + }(e.boundPortFlags, e.boundBindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 89b965c23..bc718064c 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -162,13 +162,26 @@ func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Transpo func replyWithReset(s *segment) { // Get the seqnum from the packet if the ack flag is set. seq := seqnum.Value(0) + ack := seqnum.Value(0) + flags := byte(header.TCPFlagRst) + // As per RFC 793 page 35 (Reset Generation) + // 1. If the connection does not exist (CLOSED) then a reset is sent + // in response to any incoming segment except another reset. In + // particular, SYNs addressed to a non-existent connection are rejected + // by this means. + + // If the incoming segment has an ACK field, the reset takes its + // sequence number from the ACK field of the segment, otherwise the + // reset has sequence number zero and the ACK field is set to the sum + // of the sequence number and segment length of the incoming segment. + // The connection remains in the CLOSED state. if s.flagIsSet(header.TCPFlagAck) { seq = s.ackNumber + } else { + flags |= header.TCPFlagAck + ack = s.sequenceNumber.Add(s.logicalLen()) } - - ack := s.sequenceNumber.Add(s.logicalLen()) - - sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, header.TCPFlagRst|header.TCPFlagAck, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) + sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, flags, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) } // SetOption implements TransportProtocol.SetOption. diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 5ee499c36..0a5534959 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -50,16 +50,20 @@ type receiver struct { pendingRcvdSegments segmentHeap pendingBufUsed seqnum.Size pendingBufSize seqnum.Size + + // Time when the last ack was received. + lastRcvdAckTime time.Time `state:".(unixTime)"` } func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8, pendingBufSize seqnum.Size) *receiver { return &receiver{ - ep: ep, - rcvNxt: irs + 1, - rcvAcc: irs.Add(rcvWnd + 1), - rcvWnd: rcvWnd, - rcvWndScale: rcvWndScale, - pendingBufSize: pendingBufSize, + ep: ep, + rcvNxt: irs + 1, + rcvAcc: irs.Add(rcvWnd + 1), + rcvWnd: rcvWnd, + rcvWndScale: rcvWndScale, + pendingBufSize: pendingBufSize, + lastRcvdAckTime: time.Now(), } } @@ -360,6 +364,9 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { return true, nil } + // Store the time of the last ack. + r.lastRcvdAckTime = time.Now() + // Defer segment processing if it can't be consumed now. if !r.consumeSegment(s, segSeq, segLen) { if segLen > 0 || s.flagIsSet(header.TCPFlagFin) { diff --git a/pkg/tcpip/transport/tcp/rcv_state.go b/pkg/tcpip/transport/tcp/rcv_state.go new file mode 100644 index 000000000..2bf21a2e7 --- /dev/null +++ b/pkg/tcpip/transport/tcp/rcv_state.go @@ -0,0 +1,29 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tcp + +import ( + "time" +) + +// saveLastRcvdAckTime is invoked by stateify. +func (r *receiver) saveLastRcvdAckTime() unixTime { + return unixTime{r.lastRcvdAckTime.Unix(), r.lastRcvdAckTime.UnixNano()} +} + +// loadLastRcvdAckTime is invoked by stateify. +func (r *receiver) loadLastRcvdAckTime(unix unixTime) { + r.lastRcvdAckTime = time.Unix(unix.second, unix.nano) +} diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index d3f7c9125..8a947dc66 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -28,8 +28,11 @@ import ( ) const ( - // minRTO is the minimum allowed value for the retransmit timeout. - minRTO = 200 * time.Millisecond + // MinRTO is the minimum allowed value for the retransmit timeout. + MinRTO = 200 * time.Millisecond + + // MaxRTO is the maximum allowed value for the retransmit timeout. + MaxRTO = 120 * time.Second // InitialCwnd is the initial congestion window. InitialCwnd = 10 @@ -134,6 +137,10 @@ type sender struct { // rttMeasureTime is the time when the rttMeasureSeqNum was sent. rttMeasureTime time.Time `state:".(unixTime)"` + // firstRetransmittedSegXmitTime is the original transmit time of + // the first segment that was retransmitted due to RTO expiration. + firstRetransmittedSegXmitTime time.Time `state:".(unixTime)"` + closed bool writeNext *segment writeList segmentList @@ -392,8 +399,8 @@ func (s *sender) updateRTO(rtt time.Duration) { s.rto = s.rtt.srtt + 4*s.rtt.rttvar s.rtt.Unlock() - if s.rto < minRTO { - s.rto = minRTO + if s.rto < MinRTO { + s.rto = MinRTO } } @@ -438,8 +445,30 @@ func (s *sender) retransmitTimerExpired() bool { s.ep.stack.Stats().TCP.Timeouts.Increment() s.ep.stats.SendErrors.Timeouts.Increment() - // Give up if we've waited more than a minute since the last resend. - if s.rto >= 60*time.Second { + // Give up if we've waited more than a minute since the last resend or + // if a user time out is set and we have exceeded the user specified + // timeout since the first retransmission. + s.ep.mu.RLock() + uto := s.ep.userTimeout + s.ep.mu.RUnlock() + + if s.firstRetransmittedSegXmitTime.IsZero() { + // We store the original xmitTime of the segment that we are + // about to retransmit as the retransmission time. This is + // required as by the time the retransmitTimer has expired the + // segment has already been sent and unacked for the RTO at the + // time the segment was sent. + s.firstRetransmittedSegXmitTime = s.writeList.Front().xmitTime + } + + elapsed := time.Since(s.firstRetransmittedSegXmitTime) + remaining := MaxRTO + if uto != 0 { + // Cap to the user specified timeout if one is specified. + remaining = uto - elapsed + } + + if remaining <= 0 || s.rto >= MaxRTO { return false } @@ -447,6 +476,11 @@ func (s *sender) retransmitTimerExpired() bool { // below. s.rto *= 2 + // Cap RTO to remaining time. + if s.rto > remaining { + s.rto = remaining + } + // See: https://tools.ietf.org/html/rfc6582#section-3.2 Step 4. // // Retransmit timeouts: @@ -674,7 +708,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.state = StateFinWait1 } - s.ep.stack.Stats().TCP.CurrentEstablished.Decrement() s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. @@ -1169,6 +1202,8 @@ func (s *sender) handleRcvdSegment(seg *segment) { // RFC 6298 Rule 5.3 if s.sndUna == s.sndNxt { s.outstanding = 0 + // Reset firstRetransmittedSegXmitTime to the zero value. + s.firstRetransmittedSegXmitTime = time.Time{} s.resendTimer.disable() } } diff --git a/pkg/tcpip/transport/tcp/snd_state.go b/pkg/tcpip/transport/tcp/snd_state.go index 12eff8afc..8b20c3455 100644 --- a/pkg/tcpip/transport/tcp/snd_state.go +++ b/pkg/tcpip/transport/tcp/snd_state.go @@ -48,3 +48,13 @@ func (s *sender) loadRttMeasureTime(unix unixTime) { func (s *sender) afterLoad() { s.resendTimer.init(&s.resendWaker) } + +// saveFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) saveFirstRetransmittedSegXmitTime() unixTime { + return unixTime{s.firstRetransmittedSegXmitTime.Unix(), s.firstRetransmittedSegXmitTime.UnixNano()} +} + +// loadFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) loadFirstRetransmittedSegXmitTime(unix unixTime) { + s.firstRetransmittedSegXmitTime = time.Unix(unix.second, unix.nano) +} diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index d1f0d6ce7..e8fe4dab5 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -75,6 +75,20 @@ func TestGiveUpConnect(t *testing.T) { if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != tcpip.ErrAborted { t.Fatalf("got ep.GetSockOpt(tcpip.ErrorOption{}) = %v, want = %v", err, tcpip.ErrAborted) } + + // Call Connect again to retreive the handshake failure status + // and stats updates. + if err := ep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrAborted { + t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrAborted) + } + + if got := c.Stack().Stats().TCP.FailedConnectionAttempts.Value(); got != 1 { + t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestConnectIncrementActiveConnection(t *testing.T) { @@ -309,8 +323,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) } func TestTCPResetsReceivedIncrement(t *testing.T) { @@ -446,18 +460,17 @@ func TestConnectResetAfterClose(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), ), ) break } } -// TestClosingWithEnqueuedSegments tests handling of -// still enqueued segments when the endpoint transitions -// to StateClose. The in-flight segments would be re-enqueued -// to a any listening endpoint. +// TestClosingWithEnqueuedSegments tests handling of still enqueued segments +// when the endpoint transitions to StateClose. The in-flight segments would be +// re-enqueued to a any listening endpoint. func TestClosingWithEnqueuedSegments(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -541,21 +554,29 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { ep.(interface{ ResumeWork() }).ResumeWork() // Wait for the protocolMainLoop to resume and update state. - time.Sleep(1 * time.Millisecond) + time.Sleep(10 * time.Millisecond) // Expect the endpoint to be closed. if got, want := tcp.EndpointState(ep.State()), tcp.StateClose; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + // Check if the endpoint was moved to CLOSED and netstack a reset in // response to the ACK packet that we sent after last-ACK. checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(793), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), ), ) } @@ -892,7 +913,7 @@ func TestSendRstOnListenerRxAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.TCPFlags(header.TCPFlagRst), checker.SeqNum(200))) } @@ -920,7 +941,7 @@ func TestSendRstOnListenerRxAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.TCPFlags(header.TCPFlagRst), checker.SeqNum(200))) } @@ -1119,6 +1140,71 @@ func TestConnectBindToDevice(t *testing.T) { } } +func TestRstOnSynSent(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Create an endpoint, don't handshake because we want to interfere with the + // handshake process. + c.Create(-1) + + // Start connection attempt. + waitEntry, ch := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + addr := tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort} + if err := c.EP.Connect(addr); err != tcpip.ErrConnectStarted { + t.Fatalf("got Connect(%+v) = %v, want %s", addr, err, tcpip.ErrConnectStarted) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + + // Ensure that we've reached SynSent state + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + // Send a packet with a proper ACK and a RST flag to cause the socket + // to Error and close out + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagRst | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + // Wait for receive to be notified. + select { + case <-ch: + case <-time.After(3 * time.Second): + t.Fatal("timed out waiting for packet to arrive") + } + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionRefused { + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrConnectionRefused) + } + + // Due to the RST the endpoint should be in an error state. + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -2694,6 +2780,13 @@ loop: if tcp.EndpointState(c.EP.State()) != tcp.StateError { t.Fatalf("got EP state is not StateError") } + + if got := c.Stack().Stats().TCP.EstablishedResets.Value(); got != 1 { + t.Errorf("got stats.TCP.EstablishedResets.Value() = %v, want = 1", got) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestSendOnResetConnection(t *testing.T) { @@ -4262,8 +4355,9 @@ func TestKeepalive(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) - c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) @@ -4353,19 +4447,43 @@ func TestKeepalive(t *testing.T) { ) } + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + // The connection should be terminated after 5 unacked keepalives. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(next)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), ), ) + if got := c.Stack().Stats().TCP.EstablishedTimedout.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout.Value() = %v, want = 1", got) + } + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { @@ -5992,6 +6110,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) } + want := c.Stack().Stats().TCP.EstablishedClosed.Value() + 1 + wq := &waiter.Queue{} ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) if err != nil { @@ -6118,8 +6238,15 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(uint32(ackHeaders.SeqNum)), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) + + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = %v", got, want) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestTCPCloseWithData(t *testing.T) { @@ -6290,7 +6417,147 @@ func TestTCPCloseWithData(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(uint32(ackHeaders.SeqNum)), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) +} + +func TestTCPUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + userTimeout := 50 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Send some data and wait before ACKing it. + view := buffer.NewView(3) + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %v", err) + } + + next := uint32(c.IRS) + 1 + checker.IPv4(t, c.GetPacket(), + checker.PayloadLen(len(view)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(next), + checker.AckNum(790), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + + // Wait for a little over the minimum retransmit timeout of 200ms for + // the retransmitTimer to fire and close the connection. + time.Sleep(tcp.MinRTO + 10*time.Millisecond) + + // No packet should be received as the connection should be silently + // closed due to timeout. + c.CheckNoPacket("unexpected packet received after userTimeout has expired") + + next += uint32(len(view)) + + // The connection should be terminated after userTimeout has expired. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(next)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } +} + +func TestKeepaliveWithUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + + const keepAliveInterval = 10 * time.Millisecond + c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) + c.EP.SetSockOpt(tcpip.KeepaliveCountOption(10)) + c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + + // Set userTimeout to be the duration for 3 keepalive probes. + userTimeout := 30 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Check that the connection is still alive. + if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) + } + + // Now receive 2 keepalives, but don't ACK them. The connection should + // be reset when the 3rd one should be sent due to userTimeout being + // 30ms and each keepalive probe should be sent 10ms apart as set above after + // the connection has been idle for 10ms. + for i := 0; i < 2; i++ { + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)), + checker.AckNum(uint32(790)), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + } + + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + + // The connection should be terminated after 30ms. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(c.IRS + 1), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 6cb66c1af..b0a376eba 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -231,6 +231,7 @@ func (c *Context) CheckNoPacket(errMsg string) { // addresses. It will fail with an error if no packet is received for // 2 seconds. func (c *Context) GetPacket() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { @@ -259,6 +260,7 @@ func (c *Context) GetPacket() []byte { // and destination address. If no packet is available it will return // nil immediately. func (c *Context) GetPacketNonBlocking() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { @@ -483,6 +485,7 @@ func (c *Context) CreateV6Endpoint(v6only bool) { // GetV6Packet reads a single packet from the link layer endpoint of the context // and asserts that it is an IPv6 Packet with the expected src/dest addresses. func (c *Context) GetV6Packet() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv6.ProtocolNumber { diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index 8d4c3808f..97e4d5825 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/iptables", + "//pkg/tcpip/ports", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", "//pkg/waiter", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 4b161e404..1ac4705af 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -107,6 +108,7 @@ type endpoint struct { // Values used to reserve a port or register a transport endpoint. // (which ever happens first). boundBindToDevice tcpip.NICID + boundPortFlags ports.Flags // sendTOS represents IPv4 TOS or IPv6 TrafficClass, // applied while sending packets. Defaults to 0 as on Linux. @@ -180,8 +182,9 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) e.boundBindToDevice = 0 + e.boundPortFlags = ports.Flags{} } for _, mem := range e.multicastMemberships { @@ -895,7 +898,8 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.ID.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundBindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice) + e.boundPortFlags = ports.Flags{} } e.state = StateInitial } @@ -1042,16 +1046,23 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, tcpip.NICID, *tcpip.Error) { if e.ID.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) + flags := ports.Flags{ + LoadBalanced: e.reusePort, + // FIXME(b/129164367): Support SO_REUSEADDR. + MostRecent: false, + } + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, flags, e.bindToDevice) if err != nil { return id, e.bindToDevice, err } + e.boundPortFlags = flags id.LocalPort = port } err := e.stack.RegisterTransportEndpoint(nicID, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.boundPortFlags, e.bindToDevice) + e.boundPortFlags = ports.Flags{} } return id, e.bindToDevice, err } diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 847d2f91c..6226b63f8 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "compat.go", "compat_amd64.go", + "compat_arm64.go", "config.go", "controller.go", "debug.go", @@ -110,7 +111,6 @@ go_test( "//pkg/control/server", "//pkg/log", "//pkg/p9", - "//pkg/sentry/arch:registers_go_proto", "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", diff --git a/runsc/boot/compat.go b/runsc/boot/compat.go index 07e35ab10..352e710d2 100644 --- a/runsc/boot/compat.go +++ b/runsc/boot/compat.go @@ -21,10 +21,8 @@ import ( "syscall" "github.com/golang/protobuf/proto" - "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/eventchannel" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/arch" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" ucspb "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto" "gvisor.dev/gvisor/pkg/sentry/strace" @@ -53,9 +51,9 @@ type compatEmitter struct { } func newCompatEmitter(logFD int) (*compatEmitter, error) { - nameMap, ok := strace.Lookup(abi.Linux, arch.AMD64) + nameMap, ok := getSyscallNameMap() if !ok { - return nil, fmt.Errorf("amd64 Linux syscall table not found") + return nil, fmt.Errorf("Linux syscall table not found") } c := &compatEmitter{ @@ -86,16 +84,16 @@ func (c *compatEmitter) Emit(msg proto.Message) (bool, error) { } func (c *compatEmitter) emitUnimplementedSyscall(us *spb.UnimplementedSyscall) { - regs := us.Registers.GetArch().(*rpb.Registers_Amd64).Amd64 + regs := us.Registers c.mu.Lock() defer c.mu.Unlock() - sysnr := regs.OrigRax + sysnr := syscallNum(regs) tr := c.trackers[sysnr] if tr == nil { switch sysnr { - case syscall.SYS_PRCTL, syscall.SYS_ARCH_PRCTL: + case syscall.SYS_PRCTL: // args: cmd, ... tr = newArgsTracker(0) @@ -112,10 +110,14 @@ func (c *compatEmitter) emitUnimplementedSyscall(us *spb.UnimplementedSyscall) { tr = newArgsTracker(2) default: - tr = &onceTracker{} + tr = newArchArgsTracker(sysnr) + if tr == nil { + tr = &onceTracker{} + } } c.trackers[sysnr] = tr } + if tr.shouldReport(regs) { c.sink.Infof("Unsupported syscall: %s, regs: %+v", c.nameMap.Name(uintptr(sysnr)), regs) tr.onReported(regs) @@ -139,10 +141,10 @@ func (c *compatEmitter) Close() error { // the syscall and arguments. type syscallTracker interface { // shouldReport returns true is the syscall should be reported. - shouldReport(regs *rpb.AMD64Registers) bool + shouldReport(regs *rpb.Registers) bool // onReported marks the syscall as reported. - onReported(regs *rpb.AMD64Registers) + onReported(regs *rpb.Registers) } // onceTracker reports only a single time, used for most syscalls. @@ -150,10 +152,45 @@ type onceTracker struct { reported bool } -func (o *onceTracker) shouldReport(_ *rpb.AMD64Registers) bool { +func (o *onceTracker) shouldReport(_ *rpb.Registers) bool { return !o.reported } -func (o *onceTracker) onReported(_ *rpb.AMD64Registers) { +func (o *onceTracker) onReported(_ *rpb.Registers) { o.reported = true } + +// argsTracker reports only once for each different combination of arguments. +// It's used for generic syscalls like ioctl to report once per 'cmd'. +type argsTracker struct { + // argsIdx is the syscall arguments to use as unique ID. + argsIdx []int + reported map[string]struct{} + count int +} + +func newArgsTracker(argIdx ...int) *argsTracker { + return &argsTracker{argsIdx: argIdx, reported: make(map[string]struct{})} +} + +// key returns the command based on the syscall argument index. +func (a *argsTracker) key(regs *rpb.Registers) string { + var rv string + for _, idx := range a.argsIdx { + rv += fmt.Sprintf("%d|", argVal(idx, regs)) + } + return rv +} + +func (a *argsTracker) shouldReport(regs *rpb.Registers) bool { + if a.count >= reportLimit { + return false + } + _, ok := a.reported[a.key(regs)] + return !ok +} + +func (a *argsTracker) onReported(regs *rpb.Registers) { + a.count++ + a.reported[a.key(regs)] = struct{}{} +} diff --git a/runsc/boot/compat_amd64.go b/runsc/boot/compat_amd64.go index 43cd0db94..42b0ca8b0 100644 --- a/runsc/boot/compat_amd64.go +++ b/runsc/boot/compat_amd64.go @@ -16,62 +16,81 @@ package boot import ( "fmt" + "syscall" + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/sentry/arch" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + "gvisor.dev/gvisor/pkg/sentry/strace" ) // reportLimit is the max number of events that should be reported per tracker. const reportLimit = 100 -// argsTracker reports only once for each different combination of arguments. -// It's used for generic syscalls like ioctl to report once per 'cmd'. -type argsTracker struct { - // argsIdx is the syscall arguments to use as unique ID. - argsIdx []int - reported map[string]struct{} - count int +// newRegs create a empty Registers instance. +func newRegs() *rpb.Registers { + return &rpb.Registers{ + Arch: &rpb.Registers_Amd64{ + Amd64: &rpb.AMD64Registers{}, + }, + } } -func newArgsTracker(argIdx ...int) *argsTracker { - return &argsTracker{argsIdx: argIdx, reported: make(map[string]struct{})} -} +func argVal(argIdx int, regs *rpb.Registers) uint32 { + amd64Regs := regs.GetArch().(*rpb.Registers_Amd64).Amd64 -// cmd returns the command based on the syscall argument index. -func (a *argsTracker) key(regs *rpb.AMD64Registers) string { - var rv string - for _, idx := range a.argsIdx { - rv += fmt.Sprintf("%d|", argVal(idx, regs)) + switch argIdx { + case 0: + return uint32(amd64Regs.Rdi) + case 1: + return uint32(amd64Regs.Rsi) + case 2: + return uint32(amd64Regs.Rdx) + case 3: + return uint32(amd64Regs.R10) + case 4: + return uint32(amd64Regs.R8) + case 5: + return uint32(amd64Regs.R9) } - return rv + panic(fmt.Sprintf("invalid syscall argument index %d", argIdx)) } -func argVal(argIdx int, regs *rpb.AMD64Registers) uint32 { +func setArgVal(argIdx int, argVal uint64, regs *rpb.Registers) { + amd64Regs := regs.GetArch().(*rpb.Registers_Amd64).Amd64 + switch argIdx { case 0: - return uint32(regs.Rdi) + amd64Regs.Rdi = argVal case 1: - return uint32(regs.Rsi) + amd64Regs.Rsi = argVal case 2: - return uint32(regs.Rdx) + amd64Regs.Rdx = argVal case 3: - return uint32(regs.R10) + amd64Regs.R10 = argVal case 4: - return uint32(regs.R8) + amd64Regs.R8 = argVal case 5: - return uint32(regs.R9) + amd64Regs.R9 = argVal + default: + panic(fmt.Sprintf("invalid syscall argument index %d", argIdx)) } - panic(fmt.Sprintf("invalid syscall argument index %d", argIdx)) } -func (a *argsTracker) shouldReport(regs *rpb.AMD64Registers) bool { - if a.count >= reportLimit { - return false - } - _, ok := a.reported[a.key(regs)] - return !ok +func getSyscallNameMap() (strace.SyscallMap, bool) { + return strace.Lookup(abi.Linux, arch.AMD64) +} + +func syscallNum(regs *rpb.Registers) uint64 { + amd64Regs := regs.GetArch().(*rpb.Registers_Amd64).Amd64 + return amd64Regs.OrigRax } -func (a *argsTracker) onReported(regs *rpb.AMD64Registers) { - a.count++ - a.reported[a.key(regs)] = struct{}{} +func newArchArgsTracker(sysnr uint64) syscallTracker { + switch sysnr { + case syscall.SYS_ARCH_PRCTL: + // args: cmd, ... + return newArgsTracker(0) + } + return nil } diff --git a/runsc/boot/compat_arm64.go b/runsc/boot/compat_arm64.go new file mode 100644 index 000000000..f784cd237 --- /dev/null +++ b/runsc/boot/compat_arm64.go @@ -0,0 +1,91 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package boot + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/sentry/arch" + rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + "gvisor.dev/gvisor/pkg/sentry/strace" +) + +// reportLimit is the max number of events that should be reported per tracker. +const reportLimit = 100 + +// newRegs create a empty Registers instance. +func newRegs() *rpb.Registers { + return &rpb.Registers{ + Arch: &rpb.Registers_Arm64{ + Arm64: &rpb.ARM64Registers{}, + }, + } +} + +func argVal(argIdx int, regs *rpb.Registers) uint32 { + arm64Regs := regs.GetArch().(*rpb.Registers_Arm64).Arm64 + + switch argIdx { + case 0: + return uint32(arm64Regs.R0) + case 1: + return uint32(arm64Regs.R1) + case 2: + return uint32(arm64Regs.R2) + case 3: + return uint32(arm64Regs.R3) + case 4: + return uint32(arm64Regs.R4) + case 5: + return uint32(arm64Regs.R5) + } + panic(fmt.Sprintf("invalid syscall argument index %d", argIdx)) +} + +func setArgVal(argIdx int, argVal uint64, regs *rpb.Registers) { + arm64Regs := regs.GetArch().(*rpb.Registers_Arm64).Arm64 + + switch argIdx { + case 0: + arm64Regs.R0 = argVal + case 1: + arm64Regs.R1 = argVal + case 2: + arm64Regs.R2 = argVal + case 3: + arm64Regs.R3 = argVal + case 4: + arm64Regs.R4 = argVal + case 5: + arm64Regs.R5 = argVal + default: + panic(fmt.Sprintf("invalid syscall argument index %d", argIdx)) + } +} + +func getSyscallNameMap() (strace.SyscallMap, bool) { + return strace.Lookup(abi.Linux, arch.ARM64) +} + +func syscallNum(regs *rpb.Registers) uint64 { + arm64Regs := regs.GetArch().(*rpb.Registers_Arm64).Arm64 + return arm64Regs.R8 +} + +func newArchArgsTracker(sysnr uint64) syscallTracker { + // currently, no arch specific syscalls need to be handled here. + return nil +} diff --git a/runsc/boot/compat_test.go b/runsc/boot/compat_test.go index 388298d8d..839c5303b 100644 --- a/runsc/boot/compat_test.go +++ b/runsc/boot/compat_test.go @@ -16,8 +16,6 @@ package boot import ( "testing" - - rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" ) func TestOnceTracker(t *testing.T) { @@ -35,31 +33,34 @@ func TestOnceTracker(t *testing.T) { func TestArgsTracker(t *testing.T) { for _, tc := range []struct { - name string - idx []int - rdi1 uint64 - rdi2 uint64 - rsi1 uint64 - rsi2 uint64 - want bool + name string + idx []int + arg1_1 uint64 + arg1_2 uint64 + arg2_1 uint64 + arg2_2 uint64 + want bool }{ - {name: "same rdi", idx: []int{0}, rdi1: 123, rdi2: 123, want: false}, - {name: "same rsi", idx: []int{1}, rsi1: 123, rsi2: 123, want: false}, - {name: "diff rdi", idx: []int{0}, rdi1: 123, rdi2: 321, want: true}, - {name: "diff rsi", idx: []int{1}, rsi1: 123, rsi2: 321, want: true}, - {name: "cmd is uint32", idx: []int{0}, rsi1: 0xdead00000123, rsi2: 0xbeef00000123, want: false}, - {name: "same 2 args", idx: []int{0, 1}, rsi1: 123, rdi1: 321, rsi2: 123, rdi2: 321, want: false}, - {name: "diff 2 args", idx: []int{0, 1}, rsi1: 123, rdi1: 321, rsi2: 789, rdi2: 987, want: true}, + {name: "same arg1", idx: []int{0}, arg1_1: 123, arg1_2: 123, want: false}, + {name: "same arg2", idx: []int{1}, arg2_1: 123, arg2_2: 123, want: false}, + {name: "diff arg1", idx: []int{0}, arg1_1: 123, arg1_2: 321, want: true}, + {name: "diff arg2", idx: []int{1}, arg2_1: 123, arg2_2: 321, want: true}, + {name: "cmd is uint32", idx: []int{0}, arg2_1: 0xdead00000123, arg2_2: 0xbeef00000123, want: false}, + {name: "same 2 args", idx: []int{0, 1}, arg2_1: 123, arg1_1: 321, arg2_2: 123, arg1_2: 321, want: false}, + {name: "diff 2 args", idx: []int{0, 1}, arg2_1: 123, arg1_1: 321, arg2_2: 789, arg1_2: 987, want: true}, } { t.Run(tc.name, func(t *testing.T) { c := newArgsTracker(tc.idx...) - regs := &rpb.AMD64Registers{Rdi: tc.rdi1, Rsi: tc.rsi1} + regs := newRegs() + setArgVal(0, tc.arg1_1, regs) + setArgVal(1, tc.arg2_1, regs) if !c.shouldReport(regs) { t.Error("first call to shouldReport, got: false, want: true") } c.onReported(regs) - regs.Rdi, regs.Rsi = tc.rdi2, tc.rsi2 + setArgVal(0, tc.arg1_2, regs) + setArgVal(1, tc.arg2_2, regs) if got := c.shouldReport(regs); tc.want != got { t.Errorf("second call to shouldReport, got: %t, want: %t", got, tc.want) } @@ -70,7 +71,9 @@ func TestArgsTracker(t *testing.T) { func TestArgsTrackerLimit(t *testing.T) { c := newArgsTracker(0, 1) for i := 0; i < reportLimit; i++ { - regs := &rpb.AMD64Registers{Rdi: 123, Rsi: uint64(i)} + regs := newRegs() + setArgVal(0, 123, regs) + setArgVal(1, uint64(i), regs) if !c.shouldReport(regs) { t.Error("shouldReport before limit was reached, got: false, want: true") } @@ -78,7 +81,9 @@ func TestArgsTrackerLimit(t *testing.T) { } // Should hit the count limit now. - regs := &rpb.AMD64Registers{Rdi: 123, Rsi: 123456} + regs := newRegs() + setArgVal(0, 123, regs) + setArgVal(1, 123456, regs) if c.shouldReport(regs) { t.Error("shouldReport after limit was reached, got: true, want: false") } diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index f62be4c59..9c9e94864 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -152,7 +152,9 @@ func newController(fd int, l *Loader) (*controller, error) { srv.Register(&debug{}) srv.Register(&control.Logging{}) if l.conf.ProfileEnable { - srv.Register(&control.Profile{}) + srv.Register(&control.Profile{ + Kernel: l.k, + }) } return &controller{ diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index bc9ffaf81..421ccd255 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -16,7 +16,6 @@ package boot import ( "fmt" - "path" "path/filepath" "sort" "strconv" @@ -52,7 +51,7 @@ const ( rootDevice = "9pfs-/" // MountPrefix is the annotation prefix for mount hints. - MountPrefix = "gvisor.dev/spec/mount" + MountPrefix = "dev.gvisor.spec.mount." // Filesystems that runsc supports. bind = "bind" @@ -490,14 +489,15 @@ type podMountHints struct { func newPodMountHints(spec *specs.Spec) (*podMountHints, error) { mnts := make(map[string]*mountHint) for k, v := range spec.Annotations { - // Look for 'gvisor.dev/spec/mount' annotations and parse them. + // Look for 'dev.gvisor.spec.mount' annotations and parse them. if strings.HasPrefix(k, MountPrefix) { - parts := strings.Split(k, "/") - if len(parts) != 5 { + // Remove the prefix and split the rest. + parts := strings.Split(k[len(MountPrefix):], ".") + if len(parts) != 2 { return nil, fmt.Errorf("invalid mount annotation: %s=%s", k, v) } - name := parts[3] - if len(name) == 0 || path.Clean(name) != name { + name := parts[0] + if len(name) == 0 { return nil, fmt.Errorf("invalid mount name: %s", name) } mnt := mnts[name] @@ -505,7 +505,7 @@ func newPodMountHints(spec *specs.Spec) (*podMountHints, error) { mnt = &mountHint{name: name} mnts[name] = mnt } - if err := mnt.setField(parts[4], v); err != nil { + if err := mnt.setField(parts[1], v); err != nil { return nil, err } } @@ -575,6 +575,11 @@ func newContainerMounter(spec *specs.Spec, goferFDs []int, k *kernel.Kernel, hin func (c *containerMounter) processHints(conf *Config) error { ctx := c.k.SupervisorContext() for _, hint := range c.hints.mounts { + // TODO(b/142076984): Only support tmpfs for now. Bind mounts require a + // common gofer to mount all shared volumes. + if hint.mount.Type != tmpfs { + continue + } log.Infof("Mounting master of shared mount %q from %q type %q", hint.name, hint.mount.Source, hint.mount.Type) inode, err := c.mountSharedMaster(ctx, conf, hint) if err != nil { @@ -851,7 +856,7 @@ func (c *containerMounter) mountSubmount(ctx context.Context, conf *Config, mns return fmt.Errorf("mount %q error: %v", m.Destination, err) } - log.Infof("Mounted %q to %q type %s", m.Source, m.Destination, m.Type) + log.Infof("Mounted %q to %q type: %s, internal-options: %q", m.Source, m.Destination, m.Type, opts) return nil } diff --git a/runsc/boot/fs_test.go b/runsc/boot/fs_test.go index 0396a4cfb..912037075 100644 --- a/runsc/boot/fs_test.go +++ b/runsc/boot/fs_test.go @@ -15,7 +15,6 @@ package boot import ( - "path" "reflect" "strings" "testing" @@ -26,19 +25,19 @@ import ( func TestPodMountHintsHappy(t *testing.T) { spec := &specs.Spec{ Annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "pod", - path.Join(MountPrefix, "mount2", "source"): "bar", - path.Join(MountPrefix, "mount2", "type"): "bind", - path.Join(MountPrefix, "mount2", "share"): "container", - path.Join(MountPrefix, "mount2", "options"): "rw,private", + MountPrefix + "mount2.source": "bar", + MountPrefix + "mount2.type": "bind", + MountPrefix + "mount2.share": "container", + MountPrefix + "mount2.options": "rw,private", }, } podHints, err := newPodMountHints(spec) if err != nil { - t.Errorf("newPodMountHints failed: %v", err) + t.Fatalf("newPodMountHints failed: %v", err) } // Check that fields were set correctly. @@ -86,95 +85,95 @@ func TestPodMountHintsErrors(t *testing.T) { { name: "too short", annotations: map[string]string{ - path.Join(MountPrefix, "mount1"): "foo", + MountPrefix + "mount1": "foo", }, error: "invalid mount annotation", }, { name: "no name", annotations: map[string]string{ - MountPrefix + "//source": "foo", + MountPrefix + ".source": "foo", }, error: "invalid mount name", }, { name: "missing source", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "pod", }, error: "source field", }, { name: "missing type", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.share": "pod", }, error: "type field", }, { name: "missing share", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "tmpfs", }, error: "share field", }, { name: "invalid field name", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "invalid"): "foo", + MountPrefix + "mount1.invalid": "foo", }, error: "invalid mount annotation", }, { name: "invalid source", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": "", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "pod", }, error: "source cannot be empty", }, { name: "invalid type", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "invalid-type", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "invalid-type", + MountPrefix + "mount1.share": "pod", }, error: "invalid type", }, { name: "invalid share", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "invalid-share", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "invalid-share", }, error: "invalid share", }, { name: "invalid options", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "pod", - path.Join(MountPrefix, "mount1", "options"): "invalid-option", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "pod", + MountPrefix + "mount1.options": "invalid-option", }, error: "unknown mount option", }, { name: "duplicate source", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): "foo", - path.Join(MountPrefix, "mount1", "type"): "tmpfs", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": "foo", + MountPrefix + "mount1.type": "tmpfs", + MountPrefix + "mount1.share": "pod", - path.Join(MountPrefix, "mount2", "source"): "foo", - path.Join(MountPrefix, "mount2", "type"): "bind", - path.Join(MountPrefix, "mount2", "share"): "container", + MountPrefix + "mount2.source": "foo", + MountPrefix + "mount2.type": "bind", + MountPrefix + "mount2.share": "container", }, error: "have the same mount source", }, @@ -202,36 +201,36 @@ func TestGetMountAccessType(t *testing.T) { { name: "container=exclusive", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): source, - path.Join(MountPrefix, "mount1", "type"): "bind", - path.Join(MountPrefix, "mount1", "share"): "container", + MountPrefix + "mount1.source": source, + MountPrefix + "mount1.type": "bind", + MountPrefix + "mount1.share": "container", }, want: FileAccessExclusive, }, { name: "pod=shared", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): source, - path.Join(MountPrefix, "mount1", "type"): "bind", - path.Join(MountPrefix, "mount1", "share"): "pod", + MountPrefix + "mount1.source": source, + MountPrefix + "mount1.type": "bind", + MountPrefix + "mount1.share": "pod", }, want: FileAccessShared, }, { name: "shared=shared", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): source, - path.Join(MountPrefix, "mount1", "type"): "bind", - path.Join(MountPrefix, "mount1", "share"): "shared", + MountPrefix + "mount1.source": source, + MountPrefix + "mount1.type": "bind", + MountPrefix + "mount1.share": "shared", }, want: FileAccessShared, }, { name: "default=shared", annotations: map[string]string{ - path.Join(MountPrefix, "mount1", "source"): source + "mismatch", - path.Join(MountPrefix, "mount1", "type"): "bind", - path.Join(MountPrefix, "mount1", "share"): "container", + MountPrefix + "mount1.source": source + "mismatch", + MountPrefix + "mount1.type": "bind", + MountPrefix + "mount1.share": "container", }, want: FileAccessShared, }, diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index df6052c88..bc1d0c1bb 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -93,10 +93,6 @@ type Loader struct { // spec is the base configuration for the root container. spec *specs.Spec - // startSignalForwarding enables forwarding of signals to the sandboxed - // container. It should be called after the init process is loaded. - startSignalForwarding func() func() - // stopSignalForwarding disables forwarding of signals to the sandboxed // container. It should be called when a sandbox is destroyed. stopSignalForwarding func() @@ -336,29 +332,6 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("ignore child stop signals failed: %v", err) } - // Handle signals by forwarding them to the root container process - // (except for panic signal, which should cause a panic). - l.startSignalForwarding = sighandling.PrepareHandler(func(sig linux.Signal) { - // Panic signal should cause a panic. - if args.Conf.PanicSignal != -1 && sig == linux.Signal(args.Conf.PanicSignal) { - panic("Signal-induced panic") - } - - // Otherwise forward to root container. - deliveryMode := DeliverToProcess - if args.Console { - // Since we are running with a console, we should - // forward the signal to the foreground process group - // so that job control signals like ^C can be handled - // properly. - deliveryMode = DeliverToForegroundProcessGroup - } - log.Infof("Received external signal %d, mode: %v", sig, deliveryMode) - if err := l.signal(args.ID, 0, int32(sig), deliveryMode); err != nil { - log.Warningf("error sending signal %v to container %q: %v", sig, args.ID, err) - } - }) - // Create the control server using the provided FD. // // This must be done *after* we have initialized the kernel since the @@ -566,8 +539,27 @@ func (l *Loader) run() error { ep.tty.InitForegroundProcessGroup(ep.tg.ProcessGroup()) } - // Start signal forwarding only after an init process is created. - l.stopSignalForwarding = l.startSignalForwarding() + // Handle signals by forwarding them to the root container process + // (except for panic signal, which should cause a panic). + l.stopSignalForwarding = sighandling.StartSignalForwarding(func(sig linux.Signal) { + // Panic signal should cause a panic. + if l.conf.PanicSignal != -1 && sig == linux.Signal(l.conf.PanicSignal) { + panic("Signal-induced panic") + } + + // Otherwise forward to root container. + deliveryMode := DeliverToProcess + if l.console { + // Since we are running with a console, we should forward the signal to + // the foreground process group so that job control signals like ^C can + // be handled properly. + deliveryMode = DeliverToForegroundProcessGroup + } + log.Infof("Received external signal %d, mode: %v", sig, deliveryMode) + if err := l.signal(l.sandboxID, 0, int32(sig), deliveryMode); err != nil { + log.Warningf("error sending signal %v to container %q: %v", sig, l.sandboxID, err) + } + }) log.Infof("Process should have started...") l.watchdog.Start() diff --git a/runsc/boot/network.go b/runsc/boot/network.go index f98c5fd36..dd4926bb9 100644 --- a/runsc/boot/network.go +++ b/runsc/boot/network.go @@ -80,7 +80,8 @@ type CreateLinksAndRoutesArgs struct { LoopbackLinks []LoopbackLink FDBasedLinks []FDBasedLink - DefaultGateway DefaultRoute + Defaultv4Gateway DefaultRoute + Defaultv6Gateway DefaultRoute } // Empty returns true if route hasn't been set. @@ -122,10 +123,10 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct nicID++ nicids[link.Name] = nicID - ep := loopback.New() + linkEP := loopback.New() log.Infof("Enabling loopback interface %q with id %d on addresses %+v", link.Name, nicID, link.Addresses) - if err := n.createNICWithAddrs(nicID, link.Name, ep, link.Addresses, true /* loopback */); err != nil { + if err := n.createNICWithAddrs(nicID, link.Name, linkEP, link.Addresses, true /* loopback */); err != nil { return err } @@ -157,7 +158,7 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct } mac := tcpip.LinkAddress(link.LinkAddress) - ep, err := fdbased.New(&fdbased.Options{ + linkEP, err := fdbased.New(&fdbased.Options{ FDs: FDs, MTU: uint32(link.MTU), EthernetHeader: true, @@ -172,7 +173,7 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct } log.Infof("Enabling interface %q with id %d on addresses %+v (%v) w/ %d channels", link.Name, nicID, link.Addresses, mac, link.NumChannels) - if err := n.createNICWithAddrs(nicID, link.Name, ep, link.Addresses, false /* loopback */); err != nil { + if err := n.createNICWithAddrs(nicID, link.Name, linkEP, link.Addresses, false /* loopback */); err != nil { return err } @@ -186,12 +187,24 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct } } - if !args.DefaultGateway.Route.Empty() { - nicID, ok := nicids[args.DefaultGateway.Name] + if !args.Defaultv4Gateway.Route.Empty() { + nicID, ok := nicids[args.Defaultv4Gateway.Name] if !ok { - return fmt.Errorf("invalid interface name %q for default route", args.DefaultGateway.Name) + return fmt.Errorf("invalid interface name %q for default route", args.Defaultv4Gateway.Name) } - route, err := args.DefaultGateway.Route.toTcpipRoute(nicID) + route, err := args.Defaultv4Gateway.Route.toTcpipRoute(nicID) + if err != nil { + return err + } + routes = append(routes, route) + } + + if !args.Defaultv6Gateway.Route.Empty() { + nicID, ok := nicids[args.Defaultv6Gateway.Name] + if !ok { + return fmt.Errorf("invalid interface name %q for default route", args.Defaultv6Gateway.Name) + } + route, err := args.Defaultv6Gateway.Route.toTcpipRoute(nicID) if err != nil { return err } @@ -208,11 +221,11 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct func (n *Network) createNICWithAddrs(id tcpip.NICID, name string, ep stack.LinkEndpoint, addrs []net.IP, loopback bool) error { if loopback { if err := n.Stack.CreateNamedLoopbackNIC(id, name, sniffer.New(ep)); err != nil { - return fmt.Errorf("CreateNamedLoopbackNIC(%v, %v) failed: %v", id, name, err) + return fmt.Errorf("CreateNamedLoopbackNIC(%v, %v, %v) failed: %v", id, name, ep, err) } } else { if err := n.Stack.CreateNamedNIC(id, name, sniffer.New(ep)); err != nil { - return fmt.Errorf("CreateNamedNIC(%v, %v) failed: %v", id, name, err) + return fmt.Errorf("CreateNamedNIC(%v, %v, %v) failed: %v", id, name, ep, err) } } diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go index 7313e473f..f37415810 100644 --- a/runsc/cmd/debug.go +++ b/runsc/cmd/debug.go @@ -32,16 +32,17 @@ import ( // Debug implements subcommands.Command for the "debug" command. type Debug struct { - pid int - stacks bool - signal int - profileHeap string - profileCPU string - profileDelay int - trace string - strace string - logLevel string - logPackets string + pid int + stacks bool + signal int + profileHeap string + profileCPU string + trace string + strace string + logLevel string + logPackets string + duration time.Duration + ps bool } // Name implements subcommands.Command. @@ -65,12 +66,13 @@ func (d *Debug) SetFlags(f *flag.FlagSet) { f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log") f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.") f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.") - f.IntVar(&d.profileDelay, "profile-delay", 5, "amount of time to wait before stoping CPU profile") + f.DurationVar(&d.duration, "duration", time.Second, "amount of time to wait for CPU and trace profiles") f.StringVar(&d.trace, "trace", "", "writes an execution trace to the given file.") f.IntVar(&d.signal, "signal", -1, "sends signal to the sandbox") f.StringVar(&d.strace, "strace", "", `A comma separated list of syscalls to trace. "all" enables all traces, "off" disables all`) f.StringVar(&d.logLevel, "log-level", "", "The log level to set: warning (0), info (1), or debug (2).") f.StringVar(&d.logPackets, "log-packets", "", "A boolean value to enable or disable packet logging: true or false.") + f.BoolVar(&d.ps, "ps", false, "lists processes") } // Execute implements subcommands.Command.Execute. @@ -163,7 +165,7 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if err := c.Sandbox.StartCPUProfile(f); err != nil { return Errorf(err.Error()) } - log.Infof("CPU profile started for %d sec, writing to %q", d.profileDelay, d.profileCPU) + log.Infof("CPU profile started for %v, writing to %q", d.duration, d.profileCPU) } if d.trace != "" { delay = true @@ -181,8 +183,7 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if err := c.Sandbox.StartTrace(f); err != nil { return Errorf(err.Error()) } - log.Infof("Tracing started for %d sec, writing to %q", d.profileDelay, d.trace) - + log.Infof("Tracing started for %v, writing to %q", d.duration, d.trace) } if d.strace != "" || len(d.logLevel) != 0 || len(d.logPackets) != 0 { @@ -241,9 +242,20 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) } log.Infof("Logging options changed") } + if d.ps { + pList, err := c.Processes() + if err != nil { + Fatalf("getting processes for container: %v", err) + } + o, err := control.ProcessListToJSON(pList) + if err != nil { + Fatalf("generating JSON: %v", err) + } + log.Infof(o) + } if delay { - time.Sleep(time.Duration(d.profileDelay) * time.Second) + time.Sleep(d.duration) } return subcommands.ExitSuccess diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index 7d67c3a75..5ed131a7f 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -28,6 +28,7 @@ import ( "github.com/kr/pty" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/sentry/control" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" "gvisor.dev/gvisor/runsc/testutil" @@ -219,9 +220,9 @@ func TestJobControlSignalExec(t *testing.T) { // Make sure all the processes are running. expectedPL := []*control.Process{ // Root container process. - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, // Bash from exec process. - {PID: 2, Cmd: "bash"}, + {PID: 2, Cmd: "bash", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Error(err) @@ -231,7 +232,7 @@ func TestJobControlSignalExec(t *testing.T) { ptyMaster.Write([]byte("sleep 100\n")) // Wait for it to start. Sleep's PPID is bash's PID. - expectedPL = append(expectedPL, &control.Process{PID: 3, PPID: 2, Cmd: "sleep"}) + expectedPL = append(expectedPL, &control.Process{PID: 3, PPID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{3}}) if err := waitForProcessList(c, expectedPL); err != nil { t.Error(err) } @@ -361,7 +362,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { // Wait for bash to start. expectedPL := []*control.Process{ - {PID: 1, Cmd: "bash"}, + {PID: 1, Cmd: "bash", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Fatal(err) @@ -371,7 +372,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { ptyMaster.Write([]byte("sleep 100\n")) // Wait for sleep to start. - expectedPL = append(expectedPL, &control.Process{PID: 2, PPID: 1, Cmd: "sleep"}) + expectedPL = append(expectedPL, &control.Process{PID: 2, PPID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{2}}) if err := waitForProcessList(c, expectedPL); err != nil { t.Fatal(err) } diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 07eacaac0..2ced028f6 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -37,6 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/control" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" @@ -52,13 +53,14 @@ func waitForProcessList(cont *Container, want []*control.Process) error { err = fmt.Errorf("error getting process data from container: %v", err) return &backoff.PermanentError{Err: err} } - if !procListsEqual(got, want) { - return fmt.Errorf("container got process list: %s, want: %s", procListToString(got), procListToString(want)) + if r, err := procListsEqual(got, want); !r { + return fmt.Errorf("container got process list: %s, want: %s: error: %v", + procListToString(got), procListToString(want), err) } return nil } // Gives plenty of time as tests can run slow under --race. - return testutil.Poll(cb, 30*time.Second) + return testutil.Poll(cb, 10*time.Second) } func waitForProcessCount(cont *Container, want int) error { @@ -91,22 +93,34 @@ func blockUntilWaitable(pid int) error { // procListsEqual is used to check whether 2 Process lists are equal for all // implemented fields. -func procListsEqual(got, want []*control.Process) bool { +func procListsEqual(got, want []*control.Process) (bool, error) { if len(got) != len(want) { - return false + return false, nil } for i := range got { pd1 := got[i] pd2 := want[i] - // Zero out unimplemented and timing dependant fields. + // Zero out timing dependant fields. pd1.Time = "" pd1.STime = "" pd1.C = 0 - if *pd1 != *pd2 { - return false + // Ignore TTY field too, since it's not relevant in the cases + // where we use this method. Tests that care about the TTY + // field should check for it themselves. + pd1.TTY = "" + pd1Json, err := control.ProcessListToJSON([]*control.Process{pd1}) + if err != nil { + return false, err + } + pd2Json, err := control.ProcessListToJSON([]*control.Process{pd2}) + if err != nil { + return false, err + } + if pd1Json != pd2Json { + return false, nil } } - return true + return true, nil } // getAndCheckProcLists is similar to waitForProcessList, but does not wait and retry the @@ -116,7 +130,11 @@ func getAndCheckProcLists(cont *Container, want []*control.Process) error { if err != nil { return fmt.Errorf("error getting process data from container: %v", err) } - if procListsEqual(got, want) { + equal, err := procListsEqual(got, want) + if err != nil { + return err + } + if equal { return nil } return fmt.Errorf("container got process list: %s, want: %s", procListToString(got), procListToString(want)) @@ -288,11 +306,12 @@ func TestLifecycle(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, } // Create the container. @@ -590,18 +609,20 @@ func TestExec(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{2}, }, } @@ -1062,18 +1083,20 @@ func TestPauseResume(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "bash", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "bash", + Threads: []kernel.ThreadID{2}, }, } @@ -1126,11 +1149,12 @@ func TestPauseResume(t *testing.T) { expectedPL2 := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, } @@ -1241,18 +1265,20 @@ func TestCapabilities(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "exe", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "exe", + Threads: []kernel.ThreadID{2}, }, } if err := waitForProcessList(cont, expectedPL[:1]); err != nil { @@ -2112,6 +2138,95 @@ func TestOverlayfsStaleRead(t *testing.T) { } } +// TestTTYField checks TTY field returned by container.Processes(). +func TestTTYField(t *testing.T) { + stop := testutil.StartReaper() + defer stop() + + testApp, err := testutil.FindFile("runsc/container/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } + + testCases := []struct { + name string + useTTY bool + wantTTYField string + }{ + { + name: "no tty", + useTTY: false, + wantTTYField: "?", + }, + { + name: "tty used", + useTTY: true, + wantTTYField: "pts/0", + }, + } + + for _, test := range testCases { + t.Run(test.name, func(t *testing.T) { + conf := testutil.TestConfig() + + // We will run /bin/sleep, possibly with an open TTY. + cmd := []string{"/bin/sleep", "10000"} + if test.useTTY { + // Run inside the "pty-runner". + cmd = append([]string{testApp, "pty-runner"}, cmd...) + } + + spec := testutil.NewSpecWithArgs(cmd...) + rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer os.RemoveAll(rootDir) + defer os.RemoveAll(bundleDir) + + // Create and start the container. + args := Args{ + ID: testutil.UniqueContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } + + // Wait for sleep to be running, and check the TTY + // field. + var gotTTYField string + cb := func() error { + ps, err := c.Processes() + if err != nil { + err = fmt.Errorf("error getting process data from container: %v", err) + return &backoff.PermanentError{Err: err} + } + for _, p := range ps { + if strings.Contains(p.Cmd, "sleep") { + gotTTYField = p.TTY + return nil + } + } + return fmt.Errorf("sleep not running") + } + if err := testutil.Poll(cb, 30*time.Second); err != nil { + t.Fatalf("error waiting for sleep process: %v", err) + } + + if gotTTYField != test.wantTTYField { + t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField) + } + }) + } +} + // executeSync synchronously executes a new process. func (cont *Container) executeSync(args *control.ExecArgs) (syscall.WaitStatus, error) { pid, err := cont.Execute(args) diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index a5a62378c..4ad09ceab 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -123,11 +123,11 @@ func execMany(execs []execDesc) error { func createSharedMount(mount specs.Mount, name string, pod ...*specs.Spec) { for _, spec := range pod { - spec.Annotations[path.Join(boot.MountPrefix, name, "source")] = mount.Source - spec.Annotations[path.Join(boot.MountPrefix, name, "type")] = mount.Type - spec.Annotations[path.Join(boot.MountPrefix, name, "share")] = "pod" + spec.Annotations[boot.MountPrefix+name+".source"] = mount.Source + spec.Annotations[boot.MountPrefix+name+".type"] = mount.Type + spec.Annotations[boot.MountPrefix+name+".share"] = "pod" if len(mount.Options) > 0 { - spec.Annotations[path.Join(boot.MountPrefix, name, "options")] = strings.Join(mount.Options, ",") + spec.Annotations[boot.MountPrefix+name+".options"] = strings.Join(mount.Options, ",") } } } @@ -156,13 +156,13 @@ func TestMultiContainerSanity(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -202,13 +202,13 @@ func TestMultiPIDNS(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -264,7 +264,7 @@ func TestMultiPIDNSPath(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -274,7 +274,7 @@ func TestMultiPIDNSPath(t *testing.T) { } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -306,7 +306,7 @@ func TestMultiContainerWait(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -351,7 +351,7 @@ func TestMultiContainerWait(t *testing.T) { // After Wait returns, ensure that the root container is running and // the child has finished. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for %q to start: %v", strings.Join(containers[0].Spec.Process.Args, " "), err) @@ -383,7 +383,7 @@ func TestExecWait(t *testing.T) { // Check via ps that process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Fatalf("failed to wait for sleep to start: %v", err) @@ -418,7 +418,7 @@ func TestExecWait(t *testing.T) { // Wait for the exec'd process to exit. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Fatalf("failed to wait for second container to stop: %v", err) @@ -505,7 +505,7 @@ func TestMultiContainerSignal(t *testing.T) { // Check via ps that container 1 process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { @@ -519,7 +519,7 @@ func TestMultiContainerSignal(t *testing.T) { // Make sure process 1 is still running. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -633,9 +633,10 @@ func TestMultiContainerDestroy(t *testing.T) { if err != nil { t.Fatalf("error getting process data from sandbox: %v", err) } - expectedPL := []*control.Process{{PID: 1, Cmd: "sleep"}} - if !procListsEqual(pss, expectedPL) { - t.Errorf("container got process list: %s, want: %s", procListToString(pss), procListToString(expectedPL)) + expectedPL := []*control.Process{{PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}} + if r, err := procListsEqual(pss, expectedPL); !r { + t.Errorf("container got process list: %s, want: %s: error: %v", + procListToString(pss), procListToString(expectedPL), err) } // Check that cont.Destroy is safe to call multiple times. @@ -669,7 +670,7 @@ func TestMultiContainerProcesses(t *testing.T) { // Check root's container process list doesn't include other containers. expectedPL0 := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL0); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -677,8 +678,8 @@ func TestMultiContainerProcesses(t *testing.T) { // Same for the other container. expectedPL1 := []*control.Process{ - {PID: 2, Cmd: "sh"}, - {PID: 3, PPID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sh", Threads: []kernel.ThreadID{2}}, + {PID: 3, PPID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, } if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -692,7 +693,7 @@ func TestMultiContainerProcesses(t *testing.T) { if _, err := containers[1].Execute(args); err != nil { t.Fatalf("error exec'ing: %v", err) } - expectedPL1 = append(expectedPL1, &control.Process{PID: 4, Cmd: "sleep"}) + expectedPL1 = append(expectedPL1, &control.Process{PID: 4, Cmd: "sleep", Threads: []kernel.ThreadID{4}}) if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) } @@ -1513,7 +1514,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Ensure container is running c := containers[2] expectedPL := []*control.Process{ - {PID: 3, Cmd: "sleep"}, + {PID: 3, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -1541,7 +1542,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { continue // container[2] has been killed. } pl := []*control.Process{ - {PID: kernel.ThreadID(i + 1), Cmd: "sleep"}, + {PID: kernel.ThreadID(i + 1), Cmd: "sleep", Threads: []kernel.ThreadID{kernel.ThreadID(i + 1)}}, } if err := waitForProcessList(c, pl); err != nil { t.Errorf("Container %q was affected by another container: %v", c.ID, err) @@ -1561,7 +1562,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Wait until sandbox stops. waitForProcessList will loop until sandbox exits // and RPC errors out. impossiblePL := []*control.Process{ - {PID: 100, Cmd: "non-existent-process"}, + {PID: 100, Cmd: "non-existent-process", Threads: []kernel.ThreadID{100}}, } if err := waitForProcessList(c, impossiblePL); err == nil { t.Fatalf("Sandbox was not killed after gofer death") diff --git a/runsc/container/test_app/BUILD b/runsc/container/test_app/BUILD index 9bf9e6e9d..bfd338bb6 100644 --- a/runsc/container/test_app/BUILD +++ b/runsc/container/test_app/BUILD @@ -15,5 +15,6 @@ go_binary( "//pkg/unet", "//runsc/testutil", "@com_github_google_subcommands//:go_default_library", + "@com_github_kr_pty//:go_default_library", ], ) diff --git a/runsc/container/test_app/test_app.go b/runsc/container/test_app/test_app.go index 913d781c6..a1c8a741a 100644 --- a/runsc/container/test_app/test_app.go +++ b/runsc/container/test_app/test_app.go @@ -19,6 +19,7 @@ package main import ( "context" "fmt" + "io" "io/ioutil" "log" "net" @@ -31,6 +32,7 @@ import ( "flag" "github.com/google/subcommands" + "github.com/kr/pty" "gvisor.dev/gvisor/runsc/testutil" ) @@ -41,6 +43,7 @@ func main() { subcommands.Register(new(fdReceiver), "") subcommands.Register(new(fdSender), "") subcommands.Register(new(forkBomb), "") + subcommands.Register(new(ptyRunner), "") subcommands.Register(new(reaper), "") subcommands.Register(new(syscall), "") subcommands.Register(new(taskTree), "") @@ -352,3 +355,40 @@ func (c *capability) Execute(ctx context.Context, f *flag.FlagSet, args ...inter return subcommands.ExitSuccess } + +type ptyRunner struct{} + +// Name implements subcommands.Command. +func (*ptyRunner) Name() string { + return "pty-runner" +} + +// Synopsis implements subcommands.Command. +func (*ptyRunner) Synopsis() string { + return "runs the given command with an open pty terminal" +} + +// Usage implements subcommands.Command. +func (*ptyRunner) Usage() string { + return "pty-runner [command]" +} + +// SetFlags implements subcommands.Command.SetFlags. +func (*ptyRunner) SetFlags(f *flag.FlagSet) {} + +// Execute implements subcommands.Command. +func (*ptyRunner) Execute(_ context.Context, fs *flag.FlagSet, _ ...interface{}) subcommands.ExitStatus { + c := exec.Command(fs.Args()[0], fs.Args()[1:]...) + f, err := pty.Start(c) + if err != nil { + fmt.Printf("pty.Start failed: %v", err) + return subcommands.ExitFailure + } + defer f.Close() + + // Copy stdout from the command to keep this process alive until the + // subprocess exits. + io.Copy(os.Stdout, f) + + return subcommands.ExitSuccess +} diff --git a/runsc/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go index 57f6ae8de..9b6346ca2 100644 --- a/runsc/dockerutil/dockerutil.go +++ b/runsc/dockerutil/dockerutil.go @@ -380,6 +380,16 @@ func (d *Docker) FindPort(sandboxPort int) (int, error) { return port, nil } +// FindIP returns the IP address of the container as a string. +func (d *Docker) FindIP() (string, error) { + const format = `{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}` + out, err := do("inspect", "-f", format, d.Name) + if err != nil { + return "", fmt.Errorf("error retrieving IP: %v", err) + } + return strings.TrimSpace(out), nil +} + // SandboxPid returns the PID to the sandbox process. func (d *Docker) SandboxPid() (int, error) { out, err := do("inspect", "-f={{.State.Pid}}", d.Name) diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index c9add64ec..b59e1a70e 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -199,6 +199,7 @@ func (a *attachPoint) makeQID(stat syscall.Stat_t) p9.QID { // The reason that the file is not opened initially as read-write is for better // performance with 'overlay2' storage driver. overlay2 eagerly copies the // entire file up when it's opened in write mode, and would perform badly when +// multiple files are only being opened for read (esp. startup). type localFile struct { p9.DefaultWalkGetAttr diff --git a/runsc/sandbox/BUILD b/runsc/sandbox/BUILD index 27459e6d1..8001949d5 100644 --- a/runsc/sandbox/BUILD +++ b/runsc/sandbox/BUILD @@ -19,6 +19,7 @@ go_library( "//pkg/log", "//pkg/sentry/control", "//pkg/sentry/platform", + "//pkg/tcpip/header", "//pkg/tcpip/stack", "//pkg/urpc", "//runsc/boot", diff --git a/runsc/sandbox/network.go b/runsc/sandbox/network.go index d42de0176..be8b72b3e 100644 --- a/runsc/sandbox/network.go +++ b/runsc/sandbox/network.go @@ -28,6 +28,7 @@ import ( "github.com/vishvananda/netlink" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/urpc" "gvisor.dev/gvisor/runsc/boot" @@ -183,36 +184,39 @@ func createInterfacesAndRoutesFromNS(conn *urpc.Client, nsPath string, hardwareG continue } - // Keep only IPv4 addresses. - var ip4addrs []*net.IPNet + var ipAddrs []*net.IPNet for _, ifaddr := range allAddrs { ipNet, ok := ifaddr.(*net.IPNet) if !ok { return fmt.Errorf("address is not IPNet: %+v", ifaddr) } - if ipNet.IP.To4() == nil { - log.Warningf("IPv6 is not supported, skipping: %v", ipNet) - continue - } - ip4addrs = append(ip4addrs, ipNet) + ipAddrs = append(ipAddrs, ipNet) } - if len(ip4addrs) == 0 { - log.Warningf("No IPv4 address found for interface %q, skipping", iface.Name) + if len(ipAddrs) == 0 { + log.Warningf("No usable IP addresses found for interface %q, skipping", iface.Name) continue } // Scrape the routes before removing the address, since that // will remove the routes as well. - routes, def, err := routesForIface(iface) + routes, defv4, defv6, err := routesForIface(iface) if err != nil { return fmt.Errorf("getting routes for interface %q: %v", iface.Name, err) } - if def != nil { - if !args.DefaultGateway.Route.Empty() { - return fmt.Errorf("more than one default route found, interface: %v, route: %v, default route: %+v", iface.Name, def, args.DefaultGateway) + if defv4 != nil { + if !args.Defaultv4Gateway.Route.Empty() { + return fmt.Errorf("more than one default route found, interface: %v, route: %v, default route: %+v", iface.Name, defv4, args.Defaultv4Gateway) } - args.DefaultGateway.Route = *def - args.DefaultGateway.Name = iface.Name + args.Defaultv4Gateway.Route = *defv4 + args.Defaultv4Gateway.Name = iface.Name + } + + if defv6 != nil { + if !args.Defaultv6Gateway.Route.Empty() { + return fmt.Errorf("more than one default route found, interface: %v, route: %v, default route: %+v", iface.Name, defv6, args.Defaultv6Gateway) + } + args.Defaultv6Gateway.Route = *defv6 + args.Defaultv6Gateway.Name = iface.Name } link := boot.FDBasedLink{ @@ -247,6 +251,7 @@ func createInterfacesAndRoutesFromNS(conn *urpc.Client, nsPath string, hardwareG } args.FilePayload.Files = append(args.FilePayload.Files, socketEntry.deviceFile) } + if link.GSOMaxSize == 0 && softwareGSO { // Hardware GSO is disabled. Let's enable software GSO. link.GSOMaxSize = stack.SoftwareGSOMaxSize @@ -255,7 +260,7 @@ func createInterfacesAndRoutesFromNS(conn *urpc.Client, nsPath string, hardwareG // Collect the addresses for the interface, enable forwarding, // and remove them from the host. - for _, addr := range ip4addrs { + for _, addr := range ipAddrs { link.Addresses = append(link.Addresses, addr.IP) // Steal IP address from NIC. @@ -351,46 +356,56 @@ func loopbackLinks(iface net.Interface, addrs []net.Addr) ([]boot.LoopbackLink, } // routesForIface iterates over all routes for the given interface and converts -// them to boot.Routes. -func routesForIface(iface net.Interface) ([]boot.Route, *boot.Route, error) { +// them to boot.Routes. It also returns the a default v4/v6 route if found. +func routesForIface(iface net.Interface) ([]boot.Route, *boot.Route, *boot.Route, error) { link, err := netlink.LinkByIndex(iface.Index) if err != nil { - return nil, nil, err + return nil, nil, nil, err } rs, err := netlink.RouteList(link, netlink.FAMILY_ALL) if err != nil { - return nil, nil, fmt.Errorf("getting routes from %q: %v", iface.Name, err) + return nil, nil, nil, fmt.Errorf("getting routes from %q: %v", iface.Name, err) } - var def *boot.Route + var defv4, defv6 *boot.Route var routes []boot.Route for _, r := range rs { // Is it a default route? if r.Dst == nil { if r.Gw == nil { - return nil, nil, fmt.Errorf("default route with no gateway %q: %+v", iface.Name, r) - } - if r.Gw.To4() == nil { - log.Warningf("IPv6 is not supported, skipping default route: %v", r) - continue - } - if def != nil { - return nil, nil, fmt.Errorf("more than one default route found %q, def: %+v, route: %+v", iface.Name, def, r) + return nil, nil, nil, fmt.Errorf("default route with no gateway %q: %+v", iface.Name, r) } // Create a catch all route to the gateway. - def = &boot.Route{ - Destination: net.IPNet{ - IP: net.IPv4zero, - Mask: net.IPMask(net.IPv4zero), - }, - Gateway: r.Gw, + switch len(r.Gw) { + case header.IPv4AddressSize: + if defv4 != nil { + return nil, nil, nil, fmt.Errorf("more than one default route found %q, def: %+v, route: %+v", iface.Name, defv4, r) + } + defv4 = &boot.Route{ + Destination: net.IPNet{ + IP: net.IPv4zero, + Mask: net.IPMask(net.IPv4zero), + }, + Gateway: r.Gw, + } + case header.IPv6AddressSize: + if defv6 != nil { + return nil, nil, nil, fmt.Errorf("more than one default route found %q, def: %+v, route: %+v", iface.Name, defv6, r) + } + + defv6 = &boot.Route{ + Destination: net.IPNet{ + IP: net.IPv6zero, + Mask: net.IPMask(net.IPv6zero), + }, + Gateway: r.Gw, + } + default: + return nil, nil, nil, fmt.Errorf("unexpected address size for gateway: %+v for route: %+v", r.Gw, r) } continue } - if r.Dst.IP.To4() == nil { - log.Warningf("IPv6 is not supported, skipping route: %v", r) - continue - } + dst := *r.Dst dst.IP = dst.IP.Mask(dst.Mask) routes = append(routes, boot.Route{ @@ -398,7 +413,7 @@ func routesForIface(iface net.Interface) ([]boot.Route, *boot.Route, error) { Gateway: r.Gw, }) } - return routes, def, nil + return routes, defv4, defv6, nil } // removeAddress removes IP address from network device. It's equivalent to: diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index ee9327fc8..805233184 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -1004,16 +1004,22 @@ func (s *Sandbox) ChangeLogging(args control.LoggingArgs) error { // DestroyContainer destroys the given container. If it is the root container, // then the entire sandbox is destroyed. func (s *Sandbox) DestroyContainer(cid string) error { + if err := s.destroyContainer(cid); err != nil { + // If the sandbox isn't running, the container has already been destroyed, + // ignore the error in this case. + if s.IsRunning() { + return err + } + } + return nil +} + +func (s *Sandbox) destroyContainer(cid string) error { if s.IsRootContainer(cid) { log.Debugf("Destroying root container %q by destroying sandbox", cid) return s.destroy() } - if !s.IsRunning() { - // Sandbox isn't running anymore, container is already destroyed. - return nil - } - log.Debugf("Destroying container %q in sandbox %q", cid, s.ID) conn, err := s.sandboxConnect() if err != nil { diff --git a/scripts/dev.sh b/scripts/dev.sh index c67003018..6238b4d0b 100755 --- a/scripts/dev.sh +++ b/scripts/dev.sh @@ -54,9 +54,10 @@ declare OUTPUT="$(build //runsc)" if [[ ${REFRESH} -eq 0 ]]; then install_runsc "${RUNTIME}" --net-raw install_runsc "${RUNTIME}-d" --net-raw --debug --strace --log-packets + install_runsc "${RUNTIME}-p" --net-raw --profile echo - echo "Runtimes ${RUNTIME} and ${RUNTIME}-d (debug enabled) setup." + echo "Runtimes ${RUNTIME}, ${RUNTIME}-d (debug enabled), and ${RUNTIME}-p installed." echo "Use --runtime="${RUNTIME}" with your Docker command." echo " docker run --rm --runtime="${RUNTIME}" hello-world" echo diff --git a/scripts/go.sh b/scripts/go.sh index 0dbfb7747..626ed8fa4 100755 --- a/scripts/go.sh +++ b/scripts/go.sh @@ -25,6 +25,8 @@ tools/go_branch.sh # Checkout the new branch. git checkout go && git clean -f +go version + # Build everything. go build ./... diff --git a/scripts/iptables_tests.sh b/scripts/iptables_tests.sh new file mode 100755 index 000000000..c47cbd675 --- /dev/null +++ b/scripts/iptables_tests.sh @@ -0,0 +1,27 @@ +#!/bin/bash + +# Copyright 2018 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +source $(dirname $0)/common.sh + +install_runsc_for_test iptables + +# Build the docker image for the test. +run //test/iptables/runner --norun + +# TODO(gvisor.dev/issue/170): Also test this on runsc once iptables are better +# supported +test //test/iptables:iptables_test "--test_arg=--runtime=runc" \ + "--test_arg=--image=bazel/test/iptables/runner:runner" diff --git a/scripts/simple_tests.sh b/scripts/simple_tests.sh index ef25afc2e..3a15050c2 100755 --- a/scripts/simple_tests.sh +++ b/scripts/simple_tests.sh @@ -17,4 +17,4 @@ source $(dirname $0)/common.sh # Run all simple tests (locally). -test //pkg/... //runsc/... //tools/... //benchmarks/... +test //pkg/... //runsc/... //tools/... //benchmarks/... //benchmarks/runner:runner_test diff --git a/test/iptables/BUILD b/test/iptables/BUILD new file mode 100644 index 000000000..fa833c3b2 --- /dev/null +++ b/test/iptables/BUILD @@ -0,0 +1,31 @@ +load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "iptables", + srcs = [ + "filter_input.go", + "iptables.go", + "iptables_util.go", + ], + importpath = "gvisor.dev/gvisor/test/iptables", + visibility = ["//test/iptables:__subpackages__"], +) + +go_test( + name = "iptables_test", + srcs = [ + "iptables_test.go", + ], + embed = [":iptables"], + tags = [ + "local", + "manual", + ], + deps = [ + "//pkg/log", + "//runsc/dockerutil", + "//runsc/testutil", + ], +) diff --git a/test/iptables/README.md b/test/iptables/README.md new file mode 100644 index 000000000..b37cb2a96 --- /dev/null +++ b/test/iptables/README.md @@ -0,0 +1,44 @@ +# iptables Tests + +iptables tests are run via `scripts/iptables\_test.sh`. + +## Test Structure + +Each test implements `TestCase`, providing (1) a function to run inside the +container and (2) a function to run locally. Those processes are given each +others' IP addresses. The test succeeds when both functions succeed. + +The function inside the container (`ContainerAction`) typically sets some +iptables rules and then tries to send or receive packets. The local function +(`LocalAction`) will typically just send or receive packets. + +### Adding Tests + +1) Add your test to the `iptables` package. + +2) Register the test in an `init` function via `RegisterTestCase` (see +`filter_input.go` as an example). + +3) Add it to `iptables_test.go` (see the other tests in that file). + +Your test is now runnable with bazel! + +## Run individual tests + +Build the testing Docker container: + +```bash +$ bazel run //test/iptables/runner -- --norun +``` + +Run an individual test via: + +```bash +$ bazel test //test/iptables:iptables_test --test_filter=<TESTNAME> +``` + +To run an individual test with `runc`: + +```bash +$ bazel test //test/iptables:iptables_test --test_filter=<TESTNAME> --test_arg=--runtime=runc +``` diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go new file mode 100644 index 000000000..923f44e68 --- /dev/null +++ b/test/iptables/filter_input.go @@ -0,0 +1,124 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + "net" + "time" +) + +const ( + dropPort = 2401 + acceptPort = 2402 + sendloopDuration = 2 * time.Second + network = "udp4" +) + +func init() { + RegisterTestCase(FilterInputDropUDP{}) + RegisterTestCase(FilterInputDropUDPPort{}) + RegisterTestCase(FilterInputDropDifferentUDPPort{}) +} + +// FilterInputDropUDP tests that we can drop UDP traffic. +type FilterInputDropUDP struct{} + +// Name implements TestCase.Name. +func (FilterInputDropUDP) Name() string { + return "FilterInputDropUDP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDropUDP) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-p", "udp", "-j", "DROP"); err != nil { + return err + } + + // Listen for UDP packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets on port %d should have been dropped, but got a packet", dropPort) + } else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() { + return fmt.Errorf("error reading: %v", err) + } + + // At this point we know that reading timed out and never received a + // packet. + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDropUDP) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// FilterInputDropUDPPort tests that we can drop UDP traffic by port. +type FilterInputDropUDPPort struct{} + +// Name implements TestCase.Name. +func (FilterInputDropUDPPort) Name() string { + return "FilterInputDropUDPPort" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDropUDPPort) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { + return err + } + + // Listen for UDP packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets on port %d should have been dropped, but got a packet", dropPort) + } else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() { + return fmt.Errorf("error reading: %v", err) + } + + // At this point we know that reading timed out and never received a + // packet. + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDropUDPPort) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// FilterInputDropDifferentUDPPort tests that dropping traffic for a single UDP port +// doesn't drop packets on other ports. +type FilterInputDropDifferentUDPPort struct{} + +// Name implements TestCase.Name. +func (FilterInputDropDifferentUDPPort) Name() string { + return "FilterInputDropDifferentUDPPort" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDropDifferentUDPPort) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { + return err + } + + // Listen for UDP packets on another port. + if err := listenUDP(acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("packets on port %d should be allowed, but encountered an error: %v", acceptPort, err) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDropDifferentUDPPort) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables.go b/test/iptables/iptables.go new file mode 100644 index 000000000..2e565d988 --- /dev/null +++ b/test/iptables/iptables.go @@ -0,0 +1,53 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package iptables contains a set of iptables tests implemented as TestCases +package iptables + +import ( + "fmt" + "net" +) + +// IPExchangePort is the port the container listens on to receive the IP +// address of the local process. +const IPExchangePort = 2349 + +// A TestCase contains one action to run in the container and one to run +// locally. The actions run concurrently and each must succeed for the test +// pass. +type TestCase interface { + // Name returns the name of the test. + Name() string + + // ContainerAction runs inside the container. It receives the IP of the + // local process. + ContainerAction(ip net.IP) error + + // LocalAction runs locally. It receives the IP of the container. + LocalAction(ip net.IP) error +} + +// Tests maps test names to TestCase. +// +// New TestCases are added by calling RegisterTestCase in an init function. +var Tests = map[string]TestCase{} + +// RegisterTestCase registers tc so it can be run. +func RegisterTestCase(tc TestCase) { + if _, ok := Tests[tc.Name()]; ok { + panic(fmt.Sprintf("TestCase %s already registered.", tc.Name())) + } + Tests[tc.Name()] = tc +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go new file mode 100644 index 000000000..bfbf1bb87 --- /dev/null +++ b/test/iptables/iptables_test.go @@ -0,0 +1,179 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + "net" + "os" + "path" + "testing" + "time" + + "flag" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/runsc/testutil" +) + +const timeout time.Duration = 10 * time.Second + +var image = flag.String("image", "bazel/test/iptables/runner:runner", "image to run tests in") + +type result struct { + output string + err error +} + +// singleTest runs a TestCase. Each test follows a pattern: +// - Create a container. +// - Get the container's IP. +// - Send the container our IP. +// - Start a new goroutine running the local action of the test. +// - Wait for both the container and local actions to finish. +// +// Container output is logged to $TEST_UNDECLARED_OUTPUTS_DIR if it exists, or +// to stderr. +func singleTest(test TestCase) error { + if _, ok := Tests[test.Name()]; !ok { + return fmt.Errorf("no test found with name %q. Has it been registered?", test.Name()) + } + + // Create and start the container. + cont := dockerutil.MakeDocker("gvisor-iptables") + defer cont.CleanUp() + resultChan := make(chan *result) + go func() { + output, err := cont.RunFg("--cap-add=NET_ADMIN", *image, "-name", test.Name()) + logContainer(output, err) + resultChan <- &result{output, err} + }() + + // Get the container IP. + ip, err := getIP(cont) + if err != nil { + return fmt.Errorf("failed to get container IP: %v", err) + } + + // Give the container our IP. + if err := sendIP(ip); err != nil { + return fmt.Errorf("failed to send IP to container: %v", err) + } + + // Run our side of the test. + errChan := make(chan error) + go func() { + errChan <- test.LocalAction(ip) + }() + + // Wait for both the container and local tests to finish. + var res *result + to := time.After(timeout) + for localDone := false; res == nil || !localDone; { + select { + case res = <-resultChan: + log.Infof("Container finished.") + case err, localDone = <-errChan: + log.Infof("Local finished.") + if err != nil { + return fmt.Errorf("local test failed: %v", err) + } + case <-to: + return fmt.Errorf("timed out after %f seconds", timeout.Seconds()) + } + } + + return res.err +} + +func getIP(cont dockerutil.Docker) (net.IP, error) { + // The container might not have started yet, so retry a few times. + var ipStr string + to := time.After(timeout) + for ipStr == "" { + ipStr, _ = cont.FindIP() + select { + case <-to: + return net.IP{}, fmt.Errorf("timed out getting IP after %f seconds", timeout.Seconds()) + default: + time.Sleep(250 * time.Millisecond) + } + } + ip := net.ParseIP(ipStr) + if ip == nil { + return net.IP{}, fmt.Errorf("invalid IP: %q", ipStr) + } + log.Infof("Container has IP of %s", ipStr) + return ip, nil +} + +func sendIP(ip net.IP) error { + contAddr := net.TCPAddr{ + IP: ip, + Port: IPExchangePort, + } + var conn *net.TCPConn + // The container may not be listening when we first connect, so retry + // upon error. + cb := func() error { + c, err := net.DialTCP("tcp4", nil, &contAddr) + conn = c + return err + } + if err := testutil.Poll(cb, timeout); err != nil { + return fmt.Errorf("timed out waiting to send IP, most recent error: %v", err) + } + if _, err := conn.Write([]byte{0}); err != nil { + return fmt.Errorf("error writing to container: %v", err) + } + return nil +} + +func logContainer(output string, err error) { + msg := fmt.Sprintf("Container error: %v\nContainer output:\n%v", err, output) + if artifactsDir := os.Getenv("TEST_UNDECLARED_OUTPUTS_DIR"); artifactsDir != "" { + fpath := path.Join(artifactsDir, "container.log") + if file, err := os.OpenFile(fpath, os.O_WRONLY|os.O_CREATE, 0644); err != nil { + log.Warningf("Failed to open log file %q: %v", fpath, err) + } else { + defer file.Close() + if _, err := file.Write([]byte(msg)); err == nil { + return + } + log.Warningf("Failed to write to log file %s: %v", fpath, err) + } + } + + // We couldn't write to the output directory -- just log to stderr. + log.Infof(msg) +} + +func TestFilterInputDropUDP(t *testing.T) { + if err := singleTest(FilterInputDropUDP{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropUDPPort{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropDifferentUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { + t.Fatal(err) + } +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go new file mode 100644 index 000000000..3a4d11f1a --- /dev/null +++ b/test/iptables/iptables_util.go @@ -0,0 +1,82 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + "net" + "os/exec" + "time" +) + +const iptablesBinary = "iptables" + +// filterTable calls `iptables -t filter` with the given args. +func filterTable(args ...string) error { + args = append([]string{"-t", "filter"}, args...) + cmd := exec.Command(iptablesBinary, args...) + if out, err := cmd.CombinedOutput(); err != nil { + return fmt.Errorf("error running iptables with args %v\nerror: %v\noutput: %s", args, err, string(out)) + } + return nil +} + +// listenUDP listens on a UDP port and returns the value of net.Conn.Read() for +// the first read on that port. +func listenUDP(port int, timeout time.Duration) error { + localAddr := net.UDPAddr{ + Port: port, + } + conn, err := net.ListenUDP(network, &localAddr) + if err != nil { + return err + } + defer conn.Close() + conn.SetDeadline(time.Now().Add(timeout)) + _, err = conn.Read([]byte{0}) + return err +} + +// sendUDPLoop sends 1 byte UDP packets repeatedly to the IP and port specified +// over a duration. +func sendUDPLoop(ip net.IP, port int, duration time.Duration) error { + // Send packets for a few seconds. + remote := net.UDPAddr{ + IP: ip, + Port: port, + } + conn, err := net.DialUDP(network, nil, &remote) + if err != nil { + return err + } + defer conn.Close() + + to := time.After(duration) + for timedOut := false; !timedOut; { + // This may return an error (connection refused) if the remote + // hasn't started listening yet or they're dropping our + // packets. So we ignore Write errors and depend on the remote + // to report a failure if it doesn't get a packet it needs. + conn.Write([]byte{0}) + select { + case <-to: + timedOut = true + default: + time.Sleep(200 * time.Millisecond) + } + } + + return nil +} diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD new file mode 100644 index 000000000..1c59e26b9 --- /dev/null +++ b/test/iptables/runner/BUILD @@ -0,0 +1,16 @@ +load("@io_bazel_rules_docker//container:container.bzl", "container_image") +load("@io_bazel_rules_docker//go:image.bzl", "go_image") + +package(licenses = ["notice"]) + +container_image( + name = "iptables-base", + base = "@iptables-test//image", +) + +go_image( + name = "runner", + srcs = ["main.go"], + base = ":iptables-base", + deps = ["//test/iptables"], +) diff --git a/test/iptables/runner/Dockerfile b/test/iptables/runner/Dockerfile new file mode 100644 index 000000000..b77db44a1 --- /dev/null +++ b/test/iptables/runner/Dockerfile @@ -0,0 +1,4 @@ +# This Dockerfile builds the image hosted at +# gcr.io/gvisor-presubmit/iptables-test. +FROM ubuntu +RUN apt update && apt install -y iptables diff --git a/test/iptables/runner/main.go b/test/iptables/runner/main.go new file mode 100644 index 000000000..3c794114e --- /dev/null +++ b/test/iptables/runner/main.go @@ -0,0 +1,70 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package main runs iptables tests from within a docker container. +package main + +import ( + "flag" + "fmt" + "log" + "net" + + "gvisor.dev/gvisor/test/iptables" +) + +var name = flag.String("name", "", "name of the test to run") + +func main() { + flag.Parse() + + // Find out which test we're running. + test, ok := iptables.Tests[*name] + if !ok { + log.Fatalf("No test found named %q", *name) + } + log.Printf("Running test %q", *name) + + // Get the IP of the local process. + ip, err := getIP() + if err != nil { + log.Fatal(err) + } + + // Run the test. + if err := test.ContainerAction(ip); err != nil { + log.Fatalf("Failed running test %q: %v", *name, err) + } +} + +// getIP listens for a connection from the local process and returns the source +// IP of that connection. +func getIP() (net.IP, error) { + localAddr := net.TCPAddr{ + Port: iptables.IPExchangePort, + } + listener, err := net.ListenTCP("tcp4", &localAddr) + if err != nil { + return net.IP{}, fmt.Errorf("failed listening for IP: %v", err) + } + defer listener.Close() + conn, err := listener.AcceptTCP() + if err != nil { + return net.IP{}, fmt.Errorf("failed accepting IP: %v", err) + } + defer conn.Close() + log.Printf("Connected to %v", conn.RemoteAddr()) + + return conn.RemoteAddr().(*net.TCPAddr).IP, nil +} diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 722d14b53..a3a85917d 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -376,6 +376,8 @@ syscall_test( syscall_test(test = "//test/syscalls/linux:rlimits_test") +syscall_test(test = "//test/syscalls/linux:rseq_test") + syscall_test(test = "//test/syscalls/linux:rtsignal_test") syscall_test(test = "//test/syscalls/linux:sched_test") @@ -669,6 +671,7 @@ syscall_test(test = "//test/syscalls/linux:udp_bind_test") syscall_test( size = "medium", + add_hostinet = True, shard_count = 10, test = "//test/syscalls/linux:udp_socket_test", ) @@ -714,6 +717,11 @@ syscall_test(test = "//test/syscalls/linux:proc_net_tcp_test") syscall_test(test = "//test/syscalls/linux:proc_net_udp_test") +syscall_test( + add_overlay = True, + test = "//test/syscalls/linux:xattr_test", +) + go_binary( name = "syscall_test_runner", testonly = 1, diff --git a/test/syscalls/build_defs.bzl b/test/syscalls/build_defs.bzl index dcf5b73ed..aaf77c65b 100644 --- a/test/syscalls/build_defs.bzl +++ b/test/syscalls/build_defs.bzl @@ -9,6 +9,7 @@ def syscall_test( use_tmpfs = False, add_overlay = False, add_uds_tree = False, + add_hostinet = False, tags = None): _syscall_test( test = test, @@ -65,6 +66,18 @@ def syscall_test( file_access = "shared", ) + if add_hostinet: + _syscall_test( + test = test, + shard_count = shard_count, + size = size, + platform = "ptrace", + use_tmpfs = use_tmpfs, + network = "host", + add_uds_tree = add_uds_tree, + tags = tags, + ) + def _syscall_test( test, shard_count, @@ -72,6 +85,7 @@ def _syscall_test( platform, use_tmpfs, tags, + network = "none", file_access = "exclusive", overlay = False, add_uds_tree = False): @@ -85,6 +99,8 @@ def _syscall_test( name += "_shared" if overlay: name += "_overlay" + if network != "none": + name += "_" + network + "net" if tags == None: tags = [] @@ -107,6 +123,7 @@ def _syscall_test( # Arguments are passed directly to syscall_test_runner binary. "--test-name=" + test_name, "--platform=" + platform, + "--network=" + network, "--use-tmpfs=" + str(use_tmpfs), "--file-access=" + file_access, "--overlay=" + str(overlay), diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 9cca78a93..e6568128e 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -6,6 +6,17 @@ package( licenses = ["notice"], ) +exports_files( + [ + "socket.cc", + "socket_ip_loopback_blocking.cc", + "socket_ipv4_udp_unbound_loopback.cc", + "tcp_socket.cc", + "udp_socket.cc", + ], + visibility = ["//:sandbox"], +) + cc_binary( name = "sigaltstack_check", testonly = 1, @@ -743,6 +754,7 @@ cc_binary( "//test/util:eventfd_util", "//test/util:multiprocess_util", "//test/util:posix_error", + "//test/util:save_util", "//test/util:temp_path", "//test/util:test_util", "//test/util:timer_util", @@ -1842,6 +1854,22 @@ cc_binary( ) cc_binary( + name = "rseq_test", + testonly = 1, + srcs = ["rseq.cc"], + data = ["//test/syscalls/linux/rseq"], + linkstatic = 1, + deps = [ + "//test/syscalls/linux/rseq:lib", + "//test/util:logging", + "//test/util:multiprocess_util", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( name = "rtsignal_test", testonly = 1, srcs = ["rtsignal.cc"], @@ -3245,8 +3273,6 @@ cc_binary( testonly = 1, srcs = ["tcp_socket.cc"], linkstatic = 1, - # FIXME(b/135470853) - tags = ["flaky"], deps = [ ":socket_test_util", "//test/util:file_descriptor", @@ -3697,3 +3723,24 @@ cc_binary( "@com_google_googletest//:gtest", ], ) + +cc_binary( + name = "xattr_test", + testonly = 1, + srcs = [ + "file_base.h", + "xattr.cc", + ], + linkstatic = 1, + deps = [ + "//test/util:capability_util", + "//test/util:file_descriptor", + "//test/util:fs_util", + "//test/util:posix_error", + "//test/util:temp_path", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_absl//absl/strings", + "@com_google_googletest//:gtest", + ], +) diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 8a45be12a..4f3aa81d6 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -14,6 +14,7 @@ #include <fcntl.h> #include <signal.h> +#include <sys/types.h> #include <syscall.h> #include <unistd.h> @@ -32,6 +33,7 @@ #include "test/util/eventfd_util.h" #include "test/util/multiprocess_util.h" #include "test/util/posix_error.h" +#include "test/util/save_util.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" #include "test/util/timer_util.h" @@ -910,8 +912,166 @@ TEST(FcntlTest, GetOwn) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), 0); + MaybeSave(); +} + +TEST(FcntlTest, GetOwnEx) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &owner), + SyscallSucceedsWithValue(0)); +} + +TEST(FcntlTest, SetOwnExInvalidType) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = __pid_type(-1); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(FcntlTest, SetOwnExInvalidTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_TID; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExInvalidPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExInvalidPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PGRP; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_TID; + EXPECT_THAT(owner.pid = syscall(__NR_gettid), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnExPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnExPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PGRP; + EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + // NOTE(igudger): I don't understand why, but this is flaky on Linux. + // GetOwnExPgrp (below) does not have this issue. + SKIP_IF(!IsRunningOnGvisor()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), -owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, GetOwnExTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_TID; + EXPECT_THAT(set_owner.pid = syscall(__NR_gettid), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); +} + +TEST(FcntlTest, GetOwnExPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PID; + EXPECT_THAT(set_owner.pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); +} + +TEST(FcntlTest, GetOwnExPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PGRP; + EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); } } // namespace diff --git a/test/syscalls/linux/ioctl.cc b/test/syscalls/linux/ioctl.cc index c4f8bff08..b0a07a064 100644 --- a/test/syscalls/linux/ioctl.cc +++ b/test/syscalls/linux/ioctl.cc @@ -215,7 +215,8 @@ TEST_F(IoctlTest, FIOASYNCSelfTarget2) { auto mask_cleanup = ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_UNBLOCK, SIGIO)); - pid_t pid = getpid(); + pid_t pid = -1; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); EXPECT_THAT(ioctl(pair->second_fd(), FIOSETOWN, &pid), SyscallSucceeds()); int set = 1; diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index 072230d85..9cb4566db 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -26,25 +26,6 @@ namespace gvisor { namespace testing { -// Possible values of the "st" field in a /proc/net/{tcp,udp} entry. Source: -// Linux kernel, include/net/tcp_states.h. -enum { - TCP_ESTABLISHED = 1, - TCP_SYN_SENT, - TCP_SYN_RECV, - TCP_FIN_WAIT1, - TCP_FIN_WAIT2, - TCP_TIME_WAIT, - TCP_CLOSE, - TCP_CLOSE_WAIT, - TCP_LAST_ACK, - TCP_LISTEN, - TCP_CLOSING, - TCP_NEW_SYN_RECV, - - TCP_MAX_STATES -}; - // Extracts the IP address from an inet sockaddr in network byte order. uint32_t IPFromInetSockaddr(const struct sockaddr* addr); diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 512de5ee0..8cf08991b 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -37,6 +37,7 @@ #include <map> #include <memory> #include <ostream> +#include <regex> #include <string> #include <unordered_set> #include <utility> @@ -51,6 +52,7 @@ #include "absl/strings/str_split.h" #include "absl/strings/string_view.h" #include "absl/synchronization/mutex.h" +#include "absl/synchronization/notification.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/capability_util.h" @@ -1988,6 +1990,44 @@ TEST(Proc, GetdentsEnoent) { SyscallFailsWithErrno(ENOENT)); } +void CheckSyscwFromIOFile(const std::string& path, const std::string& regex) { + std::string output; + ASSERT_NO_ERRNO(GetContents(path, &output)); + ASSERT_THAT(output, ContainsRegex(absl::StrCat("syscw:\\s+", regex, "\n"))); +} + +// Checks that there is variable accounting of IO between threads/tasks. +TEST(Proc, PidTidIOAccounting) { + absl::Notification notification; + + // Run a thread with a bunch of writes. Check that io account records exactly + // the number of write calls. File open/close is there to prevent buffering. + ScopedThread writer([¬ification] { + const int num_writes = 100; + for (int i = 0; i < num_writes; i++) { + auto path = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + ASSERT_NO_ERRNO(SetContents(path.path(), "a")); + } + notification.Notify(); + const std::string& writer_dir = + absl::StrCat("/proc/", getpid(), "/task/", gettid(), "/io"); + + CheckSyscwFromIOFile(writer_dir, std::to_string(num_writes)); + }); + + // Run a thread and do no writes. Check that no writes are recorded. + ScopedThread noop([¬ification] { + notification.WaitForNotification(); + const std::string& noop_dir = + absl::StrCat("/proc/", getpid(), "/task/", gettid(), "/io"); + + CheckSyscwFromIOFile(noop_dir, "0"); + }); + + writer.Join(); + noop.Join(); +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/proc_net_tcp.cc b/test/syscalls/linux/proc_net_tcp.cc index 2659f6a98..5b6e3e3cd 100644 --- a/test/syscalls/linux/proc_net_tcp.cc +++ b/test/syscalls/linux/proc_net_tcp.cc @@ -12,6 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include <netinet/tcp.h> #include <sys/socket.h> #include <sys/stat.h> #include <sys/types.h> diff --git a/test/syscalls/linux/proc_net_udp.cc b/test/syscalls/linux/proc_net_udp.cc index f06f1a24b..786b4b4af 100644 --- a/test/syscalls/linux/proc_net_udp.cc +++ b/test/syscalls/linux/proc_net_udp.cc @@ -12,6 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include <netinet/tcp.h> #include <sys/socket.h> #include <sys/stat.h> #include <sys/types.h> diff --git a/test/syscalls/linux/rseq.cc b/test/syscalls/linux/rseq.cc new file mode 100644 index 000000000..106c045e3 --- /dev/null +++ b/test/syscalls/linux/rseq.cc @@ -0,0 +1,198 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include <errno.h> +#include <signal.h> +#include <sys/syscall.h> +#include <sys/types.h> +#include <sys/wait.h> +#include <unistd.h> + +#include "gtest/gtest.h" +#include "test/syscalls/linux/rseq/test.h" +#include "test/syscalls/linux/rseq/uapi.h" +#include "test/util/logging.h" +#include "test/util/multiprocess_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +// Syscall test for rseq (restartable sequences). +// +// We must be very careful about how these tests are written. Each thread may +// only have one struct rseq registration, which may be done automatically at +// thread start (as of 2019-11-13, glibc does *not* support rseq and thus does +// not do so). +// +// Testing of rseq is thus done primarily in a child process with no +// registration. This means exec'ing a nostdlib binary, as rseq registration can +// only be cleared by execve (or knowing the old rseq address), and glibc (based +// on the current unmerged patches) register rseq before calling main()). + +int RSeq(struct rseq* rseq, uint32_t rseq_len, int flags, uint32_t sig) { + return syscall(kRseqSyscall, rseq, rseq_len, flags, sig); +} + +// Returns true if this kernel supports the rseq syscall. +PosixErrorOr<bool> RSeqSupported() { + // We have to be careful here, there are three possible cases: + // + // 1. rseq is not supported -> ENOSYS + // 2. rseq is supported and not registered -> success, but we should + // unregister. + // 3. rseq is supported and registered -> EINVAL (most likely). + + // The only validation done on new registrations is that rseq is aligned and + // writable. + rseq rseq = {}; + int ret = RSeq(&rseq, sizeof(rseq), 0, 0); + if (ret == 0) { + // Successfully registered, rseq is supported. Unregister. + ret = RSeq(&rseq, sizeof(rseq), kRseqFlagUnregister, 0); + if (ret != 0) { + return PosixError(errno); + } + return true; + } + + switch (errno) { + case ENOSYS: + // Not supported. + return false; + case EINVAL: + // Supported, but already registered. EINVAL returned because we provided + // a different address. + return true; + default: + // Unknown error. + return PosixError(errno); + } +} + +constexpr char kRseqBinary[] = "test/syscalls/linux/rseq/rseq"; + +void RunChildTest(std::string test_case, int want_status) { + std::string path = RunfilePath(kRseqBinary); + + pid_t child_pid = -1; + int execve_errno = 0; + auto cleanup = ASSERT_NO_ERRNO_AND_VALUE( + ForkAndExec(path, {path, test_case}, {}, &child_pid, &execve_errno)); + + ASSERT_GT(child_pid, 0); + ASSERT_EQ(execve_errno, 0); + + int status = 0; + ASSERT_THAT(RetryEINTR(waitpid)(child_pid, &status, 0), SyscallSucceeds()); + ASSERT_EQ(status, want_status); +} + +// Test that rseq must be aligned. +TEST(RseqTest, Unaligned) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestUnaligned, 0); +} + +// Sanity test that registration works. +TEST(RseqTest, Register) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestRegister, 0); +} + +// Registration can't be done twice. +TEST(RseqTest, DoubleRegister) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestDoubleRegister, 0); +} + +// Registration can be done again after unregister. +TEST(RseqTest, RegisterUnregister) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestRegisterUnregister, 0); +} + +// The pointer to rseq must match on register/unregister. +TEST(RseqTest, UnregisterDifferentPtr) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestUnregisterDifferentPtr, 0); +} + +// The signature must match on register/unregister. +TEST(RseqTest, UnregisterDifferentSignature) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestUnregisterDifferentSignature, 0); +} + +// The CPU ID is initialized. +TEST(RseqTest, CPU) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestCPU, 0); +} + +// Critical section is eventually aborted. +TEST(RseqTest, Abort) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestAbort, 0); +} + +// Abort may be before the critical section. +TEST(RseqTest, AbortBefore) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestAbortBefore, 0); +} + +// Signature must match. +TEST(RseqTest, AbortSignature) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestAbortSignature, SIGSEGV); +} + +// Abort must not be in the critical section. +TEST(RseqTest, AbortPreCommit) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestAbortPreCommit, SIGSEGV); +} + +// rseq.rseq_cs is cleared on abort. +TEST(RseqTest, AbortClearsCS) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestAbortClearsCS, 0); +} + +// rseq.rseq_cs is cleared on abort outside of critical section. +TEST(RseqTest, InvalidAbortClearsCS) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(RSeqSupported())); + + RunChildTest(kRseqTestInvalidAbortClearsCS, 0); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/rseq/BUILD b/test/syscalls/linux/rseq/BUILD new file mode 100644 index 000000000..5cfe4e56f --- /dev/null +++ b/test/syscalls/linux/rseq/BUILD @@ -0,0 +1,59 @@ +# This package contains a standalone rseq test binary. This binary must not +# depend on libc, which might use rseq itself. + +load("@bazel_tools//tools/cpp:cc_flags_supplier.bzl", "cc_flags_supplier") +load("@rules_cc//cc:defs.bzl", "cc_library") + +package(licenses = ["notice"]) + +genrule( + name = "rseq_binary", + srcs = [ + "critical.h", + "critical.S", + "rseq.cc", + "syscalls.h", + "start.S", + "test.h", + "types.h", + "uapi.h", + ], + outs = ["rseq"], + cmd = " ".join([ + "$(CC)", + "$(CC_FLAGS) ", + "-I.", + "-Wall", + "-Werror", + "-O2", + "-std=c++17", + "-static", + "-nostdlib", + "-ffreestanding", + "-o", + "$(location rseq)", + "$(location critical.S)", + "$(location rseq.cc)", + "$(location start.S)", + ]), + toolchains = [ + ":no_pie_cc_flags", + "@bazel_tools//tools/cpp:current_cc_toolchain", + ], + visibility = ["//:sandbox"], +) + +cc_flags_supplier( + name = "no_pie_cc_flags", + features = ["-pie"], +) + +cc_library( + name = "lib", + testonly = 1, + hdrs = [ + "test.h", + "uapi.h", + ], + visibility = ["//:sandbox"], +) diff --git a/test/syscalls/linux/rseq/critical.S b/test/syscalls/linux/rseq/critical.S new file mode 100644 index 000000000..8c0687e6d --- /dev/null +++ b/test/syscalls/linux/rseq/critical.S @@ -0,0 +1,66 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Restartable sequences critical sections. + +// Loops continuously until aborted. +// +// void rseq_loop(struct rseq* r, struct rseq_cs* cs) + + .text + .globl rseq_loop + .type rseq_loop, @function + +rseq_loop: + jmp begin + + // Abort block before the critical section. + // Abort signature is 4 nops for simplicity. + .byte 0x90, 0x90, 0x90, 0x90 + .globl rseq_loop_early_abort +rseq_loop_early_abort: + ret + +begin: + // r->rseq_cs = cs + movq %rsi, 8(%rdi) + + // N.B. rseq_cs will be cleared by any preempt, even outside the critical + // section. Thus it must be set in or immediately before the critical section + // to ensure it is not cleared before the section begins. + .globl rseq_loop_start +rseq_loop_start: + jmp rseq_loop_start + + // "Pre-commit": extra instructions inside the critical section. These are + // used as the abort point in TestAbortPreCommit, which is not valid. + .globl rseq_loop_pre_commit +rseq_loop_pre_commit: + // Extra abort signature + nop for TestAbortPostCommit. + .byte 0x90, 0x90, 0x90, 0x90 + nop + + // "Post-commit": never reached in this case. + .globl rseq_loop_post_commit +rseq_loop_post_commit: + + // Abort signature is 4 nops for simplicity. + .byte 0x90, 0x90, 0x90, 0x90 + + .globl rseq_loop_abort +rseq_loop_abort: + ret + + .size rseq_loop,.-rseq_loop + .section .note.GNU-stack,"",@progbits diff --git a/test/syscalls/linux/rseq/critical.h b/test/syscalls/linux/rseq/critical.h new file mode 100644 index 000000000..ac987a25e --- /dev/null +++ b/test/syscalls/linux/rseq/critical.h @@ -0,0 +1,39 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_RSEQ_CRITICAL_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_RSEQ_CRITICAL_H_ + +#include "test/syscalls/linux/rseq/types.h" +#include "test/syscalls/linux/rseq/uapi.h" + +constexpr uint32_t kRseqSignature = 0x90909090; + +extern "C" { + +extern void rseq_loop(struct rseq* r, struct rseq_cs* cs); +extern void* rseq_loop_early_abort; +extern void* rseq_loop_start; +extern void* rseq_loop_pre_commit; +extern void* rseq_loop_post_commit; +extern void* rseq_loop_abort; + +extern int rseq_getpid(struct rseq* r, struct rseq_cs* cs); +extern void* rseq_getpid_start; +extern void* rseq_getpid_post_commit; +extern void* rseq_getpid_abort; + +} // extern "C" + +#endif // GVISOR_TEST_SYSCALLS_LINUX_RSEQ_CRITICAL_H_ diff --git a/test/syscalls/linux/rseq/rseq.cc b/test/syscalls/linux/rseq/rseq.cc new file mode 100644 index 000000000..f036db26d --- /dev/null +++ b/test/syscalls/linux/rseq/rseq.cc @@ -0,0 +1,366 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "test/syscalls/linux/rseq/critical.h" +#include "test/syscalls/linux/rseq/syscalls.h" +#include "test/syscalls/linux/rseq/test.h" +#include "test/syscalls/linux/rseq/types.h" +#include "test/syscalls/linux/rseq/uapi.h" + +namespace gvisor { +namespace testing { + +extern "C" int main(int argc, char** argv, char** envp); + +// Standalone initialization before calling main(). +extern "C" void __init(uintptr_t* sp) { + int argc = sp[0]; + char** argv = reinterpret_cast<char**>(&sp[1]); + char** envp = &argv[argc + 1]; + + // Call main() and exit. + sys_exit_group(main(argc, argv, envp)); + + // sys_exit_group does not return +} + +int strcmp(const char* s1, const char* s2) { + const unsigned char* p1 = reinterpret_cast<const unsigned char*>(s1); + const unsigned char* p2 = reinterpret_cast<const unsigned char*>(s2); + + while (*p1 == *p2) { + if (!*p1) { + return 0; + } + ++p1; + ++p2; + } + return static_cast<int>(*p1) - static_cast<int>(*p2); +} + +int sys_rseq(struct rseq* rseq, uint32_t rseq_len, int flags, uint32_t sig) { + return raw_syscall(kRseqSyscall, rseq, rseq_len, flags, sig); +} + +// Test that rseq must be aligned. +int TestUnaligned() { + constexpr uintptr_t kRequiredAlignment = alignof(rseq); + + char buf[2 * kRequiredAlignment] = {}; + uintptr_t ptr = reinterpret_cast<uintptr_t>(&buf[0]); + if ((ptr & (kRequiredAlignment - 1)) == 0) { + // buf is already aligned. Misalign it. + ptr++; + } + + int ret = sys_rseq(reinterpret_cast<rseq*>(ptr), sizeof(rseq), 0, 0); + if (sys_errno(ret) != EINVAL) { + return 1; + } + return 0; +} + +// Sanity test that registration works. +int TestRegister() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + return 0; +}; + +// Registration can't be done twice. +int TestDoubleRegister() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != EBUSY) { + return 1; + } + + return 0; +}; + +// Registration can be done again after unregister. +int TestRegisterUnregister() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + + if (int ret = sys_rseq(&r, sizeof(r), kRseqFlagUnregister, 0); + sys_errno(ret) != 0) { + return 1; + } + + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + + return 0; +}; + +// The pointer to rseq must match on register/unregister. +int TestUnregisterDifferentPtr() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + + struct rseq r2 = {}; + if (int ret = sys_rseq(&r2, sizeof(r2), kRseqFlagUnregister, 0); + sys_errno(ret) != EINVAL) { + return 1; + } + + return 0; +}; + +// The signature must match on register/unregister. +int TestUnregisterDifferentSignature() { + constexpr int kSignature = 0; + + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kSignature); sys_errno(ret) != 0) { + return 1; + } + + if (int ret = sys_rseq(&r, sizeof(r), kRseqFlagUnregister, kSignature + 1); + sys_errno(ret) != EPERM) { + return 1; + } + + return 0; +}; + +// The CPU ID is initialized. +int TestCPU() { + struct rseq r = {}; + r.cpu_id = kRseqCPUIDUninitialized; + + if (int ret = sys_rseq(&r, sizeof(r), 0, 0); sys_errno(ret) != 0) { + return 1; + } + + if (__atomic_load_n(&r.cpu_id, __ATOMIC_RELAXED) < 0) { + return 1; + } + if (__atomic_load_n(&r.cpu_id_start, __ATOMIC_RELAXED) < 0) { + return 1; + } + + return 0; +}; + +// Critical section is eventually aborted. +int TestAbort() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_abort); + + // Loops until abort. If this returns then abort occurred. + rseq_loop(&r, &cs); + + return 0; +}; + +// Abort may be before the critical section. +int TestAbortBefore() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_early_abort); + + // Loops until abort. If this returns then abort occurred. + rseq_loop(&r, &cs); + + return 0; +}; + +// Signature must match. +int TestAbortSignature() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature + 1); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_abort); + + // Loops until abort. This should SIGSEGV on abort. + rseq_loop(&r, &cs); + + return 1; +}; + +// Abort must not be in the critical section. +int TestAbortPreCommit() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature + 1); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_pre_commit); + + // Loops until abort. This should SIGSEGV on abort. + rseq_loop(&r, &cs); + + return 1; +}; + +// rseq.rseq_cs is cleared on abort. +int TestAbortClearsCS() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_abort); + + // Loops until abort. If this returns then abort occurred. + rseq_loop(&r, &cs); + + if (__atomic_load_n(&r.rseq_cs, __ATOMIC_RELAXED)) { + return 1; + } + + return 0; +}; + +// rseq.rseq_cs is cleared on abort outside of critical section. +int TestInvalidAbortClearsCS() { + struct rseq r = {}; + if (int ret = sys_rseq(&r, sizeof(r), 0, kRseqSignature); + sys_errno(ret) != 0) { + return 1; + } + + struct rseq_cs cs = {}; + cs.version = 0; + cs.flags = 0; + cs.start_ip = reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.post_commit_offset = reinterpret_cast<uint64_t>(&rseq_loop_post_commit) - + reinterpret_cast<uint64_t>(&rseq_loop_start); + cs.abort_ip = reinterpret_cast<uint64_t>(&rseq_loop_abort); + + __atomic_store_n(&r.rseq_cs, &cs, __ATOMIC_RELAXED); + + // When the next abort condition occurs, the kernel will clear cs once it + // determines we aren't in the critical section. + while (1) { + if (!__atomic_load_n(&r.rseq_cs, __ATOMIC_RELAXED)) { + break; + } + } + + return 0; +}; + +// Exit codes: +// 0 - Pass +// 1 - Fail +// 2 - Missing argument +// 3 - Unknown test case +extern "C" int main(int argc, char** argv, char** envp) { + if (argc != 2) { + // Usage: rseq <test case> + return 2; + } + + if (strcmp(argv[1], kRseqTestUnaligned) == 0) { + return TestUnaligned(); + } + if (strcmp(argv[1], kRseqTestRegister) == 0) { + return TestRegister(); + } + if (strcmp(argv[1], kRseqTestDoubleRegister) == 0) { + return TestDoubleRegister(); + } + if (strcmp(argv[1], kRseqTestRegisterUnregister) == 0) { + return TestRegisterUnregister(); + } + if (strcmp(argv[1], kRseqTestUnregisterDifferentPtr) == 0) { + return TestUnregisterDifferentPtr(); + } + if (strcmp(argv[1], kRseqTestUnregisterDifferentSignature) == 0) { + return TestUnregisterDifferentSignature(); + } + if (strcmp(argv[1], kRseqTestCPU) == 0) { + return TestCPU(); + } + if (strcmp(argv[1], kRseqTestAbort) == 0) { + return TestAbort(); + } + if (strcmp(argv[1], kRseqTestAbortBefore) == 0) { + return TestAbortBefore(); + } + if (strcmp(argv[1], kRseqTestAbortSignature) == 0) { + return TestAbortSignature(); + } + if (strcmp(argv[1], kRseqTestAbortPreCommit) == 0) { + return TestAbortPreCommit(); + } + if (strcmp(argv[1], kRseqTestAbortClearsCS) == 0) { + return TestAbortClearsCS(); + } + if (strcmp(argv[1], kRseqTestInvalidAbortClearsCS) == 0) { + return TestInvalidAbortClearsCS(); + } + + return 3; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/rseq/start.S b/test/syscalls/linux/rseq/start.S new file mode 100644 index 000000000..b9611b276 --- /dev/null +++ b/test/syscalls/linux/rseq/start.S @@ -0,0 +1,45 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + + + .text + .align 4 + .type _start,@function + .globl _start + +_start: + movq %rsp,%rdi + call __init + hlt + + .size _start,.-_start + .section .note.GNU-stack,"",@progbits + + .text + .globl raw_syscall + .type raw_syscall, @function + +raw_syscall: + mov %rdi,%rax // syscall # + mov %rsi,%rdi // arg0 + mov %rdx,%rsi // arg1 + mov %rcx,%rdx // arg2 + mov %r8,%r10 // arg3 (goes in r10 instead of rcx for system calls) + mov %r9,%r8 // arg4 + mov 0x8(%rsp),%r9 // arg5 + syscall + ret + + .size raw_syscall,.-raw_syscall + .section .note.GNU-stack,"",@progbits diff --git a/test/syscalls/linux/rseq/syscalls.h b/test/syscalls/linux/rseq/syscalls.h new file mode 100644 index 000000000..e5299c188 --- /dev/null +++ b/test/syscalls/linux/rseq/syscalls.h @@ -0,0 +1,66 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_RSEQ_SYSCALLS_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_RSEQ_SYSCALLS_H_ + +#include "test/syscalls/linux/rseq/types.h" + +#ifdef __x86_64__ +// Syscall numbers. +constexpr int kGetpid = 39; +constexpr int kExitGroup = 231; +#else +#error "Unknown architecture" +#endif + +namespace gvisor { +namespace testing { + +// Standalone system call interfaces. +// Note that these are all "raw" system call interfaces which encode +// errors by setting the return value to a small negative number. +// Use sys_errno() to check system call return values for errors. + +// Maximum Linux error number. +constexpr int kMaxErrno = 4095; + +// Errno values. +#define EPERM 1 +#define EFAULT 14 +#define EBUSY 16 +#define EINVAL 22 + +// Get the error number from a raw system call return value. +// Returns a positive error number or 0 if there was no error. +static inline int sys_errno(uintptr_t rval) { + if (rval >= static_cast<uintptr_t>(-kMaxErrno)) { + return -static_cast<int>(rval); + } + return 0; +} + +extern "C" uintptr_t raw_syscall(int number, ...); + +static inline void sys_exit_group(int status) { + raw_syscall(kExitGroup, status); +} +static inline int sys_getpid() { + return static_cast<int>(raw_syscall(kGetpid)); +} + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_LINUX_RSEQ_SYSCALLS_H_ diff --git a/test/syscalls/linux/rseq/test.h b/test/syscalls/linux/rseq/test.h new file mode 100644 index 000000000..3b7bb74b1 --- /dev/null +++ b/test/syscalls/linux/rseq/test.h @@ -0,0 +1,43 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TEST_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TEST_H_ + +namespace gvisor { +namespace testing { + +// Test cases supported by rseq binary. + +inline constexpr char kRseqTestUnaligned[] = "unaligned"; +inline constexpr char kRseqTestRegister[] = "register"; +inline constexpr char kRseqTestDoubleRegister[] = "double-register"; +inline constexpr char kRseqTestRegisterUnregister[] = "register-unregister"; +inline constexpr char kRseqTestUnregisterDifferentPtr[] = + "unregister-different-ptr"; +inline constexpr char kRseqTestUnregisterDifferentSignature[] = + "unregister-different-signature"; +inline constexpr char kRseqTestCPU[] = "cpu"; +inline constexpr char kRseqTestAbort[] = "abort"; +inline constexpr char kRseqTestAbortBefore[] = "abort-before"; +inline constexpr char kRseqTestAbortSignature[] = "abort-signature"; +inline constexpr char kRseqTestAbortPreCommit[] = "abort-precommit"; +inline constexpr char kRseqTestAbortClearsCS[] = "abort-clears-cs"; +inline constexpr char kRseqTestInvalidAbortClearsCS[] = + "invalid-abort-clears-cs"; + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TEST_H_ diff --git a/test/syscalls/linux/rseq/types.h b/test/syscalls/linux/rseq/types.h new file mode 100644 index 000000000..b6afe9817 --- /dev/null +++ b/test/syscalls/linux/rseq/types.h @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TYPES_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TYPES_H_ + +using size_t = __SIZE_TYPE__; +using uintptr_t = __UINTPTR_TYPE__; + +using uint8_t = __UINT8_TYPE__; +using uint16_t = __UINT16_TYPE__; +using uint32_t = __UINT32_TYPE__; +using uint64_t = __UINT64_TYPE__; + +using int8_t = __INT8_TYPE__; +using int16_t = __INT16_TYPE__; +using int32_t = __INT32_TYPE__; +using int64_t = __INT64_TYPE__; + +#endif // GVISOR_TEST_SYSCALLS_LINUX_RSEQ_TYPES_H_ diff --git a/test/syscalls/linux/rseq/uapi.h b/test/syscalls/linux/rseq/uapi.h new file mode 100644 index 000000000..e3ff0579a --- /dev/null +++ b/test/syscalls/linux/rseq/uapi.h @@ -0,0 +1,54 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_RSEQ_UAPI_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_RSEQ_UAPI_H_ + +// User-kernel ABI for restartable sequences. + +// Standard types. +// +// N.B. This header will be included in targets that do have the standard +// library, so we can't shadow the standard type names. +using __u32 = __UINT32_TYPE__; +using __u64 = __UINT64_TYPE__; + +#ifdef __x86_64__ +// Syscall numbers. +constexpr int kRseqSyscall = 334; +#else +#error "Unknown architecture" +#endif // __x86_64__ + +struct rseq_cs { + __u32 version; + __u32 flags; + __u64 start_ip; + __u64 post_commit_offset; + __u64 abort_ip; +} __attribute__((aligned(4 * sizeof(__u64)))); + +// N.B. alignment is enforced by the kernel. +struct rseq { + __u32 cpu_id_start; + __u32 cpu_id; + struct rseq_cs* rseq_cs; + __u32 flags; +} __attribute__((aligned(4 * sizeof(__u64)))); + +constexpr int kRseqFlagUnregister = 1 << 0; + +constexpr int kRseqCPUIDUninitialized = -1; + +#endif // GVISOR_TEST_SYSCALLS_LINUX_RSEQ_UAPI_H_ diff --git a/test/syscalls/linux/socket_bind_to_device_sequence.cc b/test/syscalls/linux/socket_bind_to_device_sequence.cc index e4641c62e..34b1058a9 100644 --- a/test/syscalls/linux/socket_bind_to_device_sequence.cc +++ b/test/syscalls/linux/socket_bind_to_device_sequence.cc @@ -97,12 +97,22 @@ class BindToDeviceSequenceTest : public ::testing::TestWithParam<SocketKind> { sockets_to_close_.erase(socket_id); } - // Bind a socket with the reuse option and bind_to_device options. Checks + // SetDevice changes the bind_to_device option. It does not bind or re-bind. + void SetDevice(int socket_id, int device_id) { + auto socket_fd = sockets_to_close_[socket_id]->get(); + string device_name; + ASSERT_NO_FATAL_FAILURE(GetDevice(device_id, &device_name)); + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, + device_name.c_str(), device_name.size() + 1), + SyscallSucceedsWithValue(0)); + } + + // Bind a socket with the reuse options and bind_to_device options. Checks // that all steps succeed and that the bind command's error matches want. // Sets the socket_id to uniquely identify the socket bound if it is not // nullptr. - void BindSocket(bool reuse, int device_id = 0, int want = 0, - int *socket_id = nullptr) { + void BindSocket(bool reuse_port, bool reuse_addr, int device_id = 0, + int want = 0, int *socket_id = nullptr) { next_socket_id_++; sockets_to_close_[next_socket_id_] = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket_fd = sockets_to_close_[next_socket_id_]->get(); @@ -110,13 +120,20 @@ class BindToDeviceSequenceTest : public ::testing::TestWithParam<SocketKind> { *socket_id = next_socket_id_; } - // If reuse is indicated, do that. - if (reuse) { + // If reuse_port is indicated, do that. + if (reuse_port) { EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, sizeof(kSockOptOn)), SyscallSucceedsWithValue(0)); } + // If reuse_addr is indicated, do that. + if (reuse_addr) { + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + } + // If the device is non-zero, bind to that device. if (device_id != 0) { string device_name; @@ -182,129 +199,308 @@ class BindToDeviceSequenceTest : public ::testing::TestWithParam<SocketKind> { }; TEST_P(BindToDeviceSequenceTest, BindTwiceWithDeviceFails) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 3)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 3, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ false, /* bind_to_device */ 3)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 3, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, BindToDevice) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 1)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 2)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ false, /* bind_to_device */ 1)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ false, /* bind_to_device */ 2)); } TEST_P(BindToDeviceSequenceTest, BindToDeviceAndThenWithoutDevice) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, BindWithoutDevice) { - ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ false)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, BindWithDevice) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 456, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 789, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, BindWithReuse) { - ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true, /* bind_to_device */ 0)); + BindSocket(/* reusePort */ true, /* reuse_addr */ false)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, + /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 0)); } TEST_P(BindToDeviceSequenceTest, BindingWithReuseAndDevice) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 456)); - ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 789)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 999, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse_port */ true, /* reuse_addr */ false)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 789)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 999, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, MixingReuseAndNotReuseByBindingToDevice) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 456, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 789, 0)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 999, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 999, 0)); } TEST_P(BindToDeviceSequenceTest, CannotBindTo0AfterMixingReuseAndNotReuse) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 456)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); } TEST_P(BindToDeviceSequenceTest, BindAndRelease) { - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 123)); int to_release; - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, 0, &to_release)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 345, EADDRINUSE)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 789)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, 0, &to_release)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 345, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 789)); // Release the bind to device 0 and try again. ASSERT_NO_FATAL_FAILURE(ReleaseSocket(to_release)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 345)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 345)); } TEST_P(BindToDeviceSequenceTest, BindTwiceWithReuseOnce) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithReuseAddr) { + // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. + SKIP_IF(IsRunningOnGvisor()); + ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ false, /* bind_to_device */ 123)); - ASSERT_NO_FATAL_FAILURE( - BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + BindSocket(/* reusePort */ false, /* reuse_addr */ true)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ true, /* bind_to_device */ 0)); +} + +TEST_P(BindToDeviceSequenceTest, + CannotBindTo0AfterMixingReuseAddrAndNotReuseAddr) { + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindReuseAddrReusePortThenReusePort) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindReuseAddrReusePortThenReuseAddr) { + // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. + SKIP_IF(IsRunningOnGvisor()); + + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindDoubleReuseAddrReusePortThenReusePort) { + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ true, /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindDoubleReuseAddrReusePortThenReuseAddr) { + // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. + SKIP_IF(IsRunningOnGvisor()); + + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ true, /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindReusePortThenReuseAddrReusePort) { + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ true, /* reuse_addr */ false, /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ true, + /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindReuseAddrThenReuseAddr) { + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ true, /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0, EADDRINUSE)); +} + +// This behavior seems like a bug? +TEST_P(BindToDeviceSequenceTest, + BindReuseAddrThenReuseAddrReusePortThenReuseAddr) { + // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. + SKIP_IF(IsRunningOnGvisor()); + + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ true, /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ true, + /* bind_to_device */ 0)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, + /* reuse_addr */ false, + /* bind_to_device */ 0)); +} + +// Repro test for gvisor.dev/issue/1217. Not replicated in ports_test.go as this +// test is different from the others and wouldn't fit well there. +TEST_P(BindToDeviceSequenceTest, BindAndReleaseDifferentDevice) { + int to_release; + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 3, 0, &to_release)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, + /* reuse_addr */ false, + /* bind_to_device */ 3, EADDRINUSE)); + // Change the device. Since the socket was already bound, this should have no + // effect. + SetDevice(to_release, 2); + // Release the bind to device 3 and try again. + ASSERT_NO_FATAL_FAILURE(ReleaseSocket(to_release)); + ASSERT_NO_FATAL_FAILURE(BindSocket( + /* reuse_port */ false, /* reuse_addr */ false, /* bind_to_device */ 3)); } INSTANTIATE_TEST_SUITE_P(BindToDeviceTest, BindToDeviceSequenceTest, diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 96a1731cf..761c3a9fe 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -206,7 +206,7 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { } // TODO(b/138400178): Fix cooperative S/R failure when ds.reset() is invoked // before function end. - // ds.reset() + // ds.reset(); } TEST_P(SocketInetLoopbackTest, TCPbacklog) { @@ -603,6 +603,60 @@ TEST_P(SocketInetLoopbackTest, TCPTimeWaitTest_NoRandomSave) { SyscallSucceeds()); } +TEST_P(SocketInetLoopbackTest, AcceptedInheritsTCPUserTimeout) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the userTimeout on the listening socket. + constexpr int kUserTimeout = 10; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kUserTimeout, sizeof(kUserTimeout)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + // Verify that the accepted socket inherited the user timeout set on + // listening socket. + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(accepted.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kUserTimeout); +} + INSTANTIATE_TEST_SUITE_P( All, SocketInetLoopbackTest, ::testing::Values( @@ -635,7 +689,9 @@ INSTANTIATE_TEST_SUITE_P( using SocketInetReusePortTest = ::testing::TestWithParam<TestParam>; -TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread) { +// TODO(gvisor.dev/issue/940): Remove _NoRandomSave when portHint/stack.Seed is +// saved/restored. +TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread_NoRandomSave) { auto const& param = GetParam(); TestAddress const& listener = param.listener; @@ -643,6 +699,7 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread) { sockaddr_storage listen_addr = listener.addr; sockaddr_storage conn_addr = connector.addr; constexpr int kThreadCount = 3; + constexpr int kConnectAttempts = 4096; // Create the listening socket. FileDescriptor listener_fds[kThreadCount]; @@ -657,7 +714,7 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread) { ASSERT_THAT( bind(fd, reinterpret_cast<sockaddr*>(&listen_addr), listener.addr_len), SyscallSucceeds()); - ASSERT_THAT(listen(fd, 40), SyscallSucceeds()); + ASSERT_THAT(listen(fd, kConnectAttempts / 3), SyscallSucceeds()); // On the first bind we need to determine which port was bound. if (i != 0) { @@ -676,7 +733,6 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread) { ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); } - constexpr int kConnectAttempts = 10000; std::atomic<int> connects_received = ATOMIC_VAR_INIT(0); std::unique_ptr<ScopedThread> listen_thread[kThreadCount]; int accept_counts[kThreadCount] = {}; diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index c74273436..57ce8e169 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -812,5 +812,68 @@ TEST_P(TCPSocketPairTest, TestTCPCloseWithData) { ASSERT_THAT(close(sockets->release_first_fd()), SyscallSucceeds()); } +TEST_P(TCPSocketPairTest, TCPUserTimeoutDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kZero = 0; + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kZero, sizeof(kZero)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutBelowZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kNeg = -10; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kNeg, sizeof(kNeg)), + SyscallFailsWithErrno(EINVAL)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutAboveZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kAbove = 10; + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kAbove, sizeof(kAbove)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kAbove); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix.cc b/test/syscalls/linux/socket_unix.cc index 8a28202a8..4cf1f76f1 100644 --- a/test/syscalls/linux/socket_unix.cc +++ b/test/syscalls/linux/socket_unix.cc @@ -65,6 +65,21 @@ TEST_P(UnixSocketPairTest, BindToBadName) { SyscallFailsWithErrno(ENOENT)); } +TEST_P(UnixSocketPairTest, BindToBadFamily) { + auto pair = + ASSERT_NO_ERRNO_AND_VALUE(UnixDomainSocketPair(SOCK_SEQPACKET).Create()); + + constexpr char kBadName[] = "/some/path/that/does/not/exist"; + sockaddr_un sockaddr; + sockaddr.sun_family = AF_INET; + memcpy(sockaddr.sun_path, kBadName, sizeof(kBadName)); + + EXPECT_THAT( + bind(pair->first_fd(), reinterpret_cast<struct sockaddr*>(&sockaddr), + sizeof(sockaddr)), + SyscallFailsWithErrno(EINVAL)); +} + TEST_P(UnixSocketPairTest, RecvmmsgTimeoutAfterRecv) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); char sent_data[10]; diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 99863b0ed..6b99c021d 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -967,6 +967,78 @@ TEST_P(SimpleTcpSocketTest, BlockingConnectRefused) { EXPECT_THAT(close(s.release()), SyscallSucceeds()); } +// Test that connecting to a non-listening port and thus receiving a RST is +// handled appropriately by the socket - the port that the socket was bound to +// is released and the expected error is returned. +TEST_P(SimpleTcpSocketTest, CleanupOnConnectionRefused) { + // Create a socket that is known to not be listening. As is it bound but not + // listening, when another socket connects to the port, it will refuse.. + FileDescriptor bound_s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage bound_addr = + ASSERT_NO_ERRNO_AND_VALUE(InetLoopbackAddr(GetParam())); + socklen_t bound_addrlen = sizeof(bound_addr); + + ASSERT_THAT( + bind(bound_s.get(), reinterpret_cast<struct sockaddr*>(&bound_addr), + bound_addrlen), + SyscallSucceeds()); + + // Get the addresses the socket is bound to because the port is chosen by the + // stack. + ASSERT_THAT(getsockname(bound_s.get(), + reinterpret_cast<struct sockaddr*>(&bound_addr), + &bound_addrlen), + SyscallSucceeds()); + + // Create, initialize, and bind the socket that is used to test connecting to + // the non-listening port. + FileDescriptor client_s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // Initialize client address to the loopback one. + sockaddr_storage client_addr = + ASSERT_NO_ERRNO_AND_VALUE(InetLoopbackAddr(GetParam())); + socklen_t client_addrlen = sizeof(client_addr); + + ASSERT_THAT( + bind(client_s.get(), reinterpret_cast<struct sockaddr*>(&client_addr), + client_addrlen), + SyscallSucceeds()); + + ASSERT_THAT(getsockname(client_s.get(), + reinterpret_cast<struct sockaddr*>(&client_addr), + &client_addrlen), + SyscallSucceeds()); + + // Now the test: connect to the bound but not listening socket with the + // client socket. The bound socket should return a RST and cause the client + // socket to return an error and clean itself up immediately. + // The error being ECONNREFUSED diverges with RFC 793, page 37, but does what + // Linux does. + ASSERT_THAT(connect(client_s.get(), + reinterpret_cast<const struct sockaddr*>(&bound_addr), + bound_addrlen), + SyscallFailsWithErrno(ECONNREFUSED)); + + FileDescriptor new_s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + // Test binding to the address from the client socket. This should be okay + // if it was dropped correctly. + ASSERT_THAT( + bind(new_s.get(), reinterpret_cast<struct sockaddr*>(&client_addr), + client_addrlen), + SyscallSucceeds()); + + // Attempt #2, with the new socket and reused addr our connect should fail in + // the same way as before, not with an EADDRINUSE. + ASSERT_THAT(connect(client_s.get(), + reinterpret_cast<const struct sockaddr*>(&bound_addr), + bound_addrlen), + SyscallFailsWithErrno(ECONNREFUSED)); +} + // Test that we get an ECONNREFUSED with a nonblocking socket. TEST_P(SimpleTcpSocketTest, NonBlockingConnectRefused) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( @@ -1175,6 +1247,31 @@ TEST_P(SimpleTcpSocketTest, SetMaxSegFailsForInvalidMSSValues) { } } +TEST_P(SimpleTcpSocketTest, SetTCPUserTimeout) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + { + constexpr int kTCPUserTimeout = -1; + EXPECT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kTCPUserTimeout, sizeof(kTCPUserTimeout)), + SyscallFailsWithErrno(EINVAL)); + } + + // kTCPUserTimeout is in milliseconds. + constexpr int kTCPUserTimeout = 100; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kTCPUserTimeout, sizeof(kTCPUserTimeout)), + SyscallSucceedsWithValue(0)); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPUserTimeout); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 63b92d6a7..dc35c2f50 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -656,6 +656,9 @@ TEST_P(UdpSocketTest, SendToAddressOtherThanConnected) { } TEST_P(UdpSocketTest, ZerolengthWriteAllowed) { + // TODO(gvisor.dev/issue/1202): Hostinet does not support zero length writes. + SKIP_IF(IsRunningWithHostinet()); + // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); @@ -673,6 +676,9 @@ TEST_P(UdpSocketTest, ZerolengthWriteAllowed) { } TEST_P(UdpSocketTest, ZerolengthWriteAllowedNonBlockRead) { + // TODO(gvisor.dev/issue/1202): Hostinet does not support zero length writes. + SKIP_IF(IsRunningWithHostinet()); + // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); @@ -878,6 +884,10 @@ TEST_P(UdpSocketTest, ReadShutdownSameSocketResetsShutdownState) { } TEST_P(UdpSocketTest, ReadShutdown) { + // TODO(gvisor.dev/issue/1202): Calling recv() after shutdown without + // MSG_DONTWAIT blocks indefinitely. + SKIP_IF(IsRunningWithHostinet()); + char received[512]; EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); @@ -900,6 +910,10 @@ TEST_P(UdpSocketTest, ReadShutdown) { } TEST_P(UdpSocketTest, ReadShutdownDifferentThread) { + // TODO(gvisor.dev/issue/1202): Calling recv() after shutdown without + // MSG_DONTWAIT blocks indefinitely. + SKIP_IF(IsRunningWithHostinet()); + char received[512]; EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); @@ -1189,6 +1203,10 @@ TEST_P(UdpSocketTest, FIONREADZeroLengthWriteShutdown) { } TEST_P(UdpSocketTest, SoTimestampOffByDefault) { + // TODO(gvisor.dev/issue/1202): SO_TIMESTAMP socket option not supported by + // hostinet. + SKIP_IF(IsRunningWithHostinet()); + int v = -1; socklen_t optlen = sizeof(v); ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_TIMESTAMP, &v, &optlen), @@ -1198,6 +1216,10 @@ TEST_P(UdpSocketTest, SoTimestampOffByDefault) { } TEST_P(UdpSocketTest, SoTimestamp) { + // TODO(gvisor.dev/issue/1202): ioctl() and SO_TIMESTAMP socket option are not + // supported by hostinet. + SKIP_IF(IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1241,6 +1263,9 @@ TEST_P(UdpSocketTest, WriteShutdownNotConnected) { } TEST_P(UdpSocketTest, TimestampIoctl) { + // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. + SKIP_IF(IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1259,7 +1284,10 @@ TEST_P(UdpSocketTest, TimestampIoctl) { ASSERT_TRUE(tv.tv_sec != 0 || tv.tv_usec != 0); } -TEST_P(UdpSocketTest, TimetstampIoctlNothingRead) { +TEST_P(UdpSocketTest, TimestampIoctlNothingRead) { + // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. + SKIP_IF(IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1270,6 +1298,10 @@ TEST_P(UdpSocketTest, TimetstampIoctlNothingRead) { // Test that the timestamp accessed via SIOCGSTAMP is still accessible after // SO_TIMESTAMP is enabled and used to retrieve a timestamp. TEST_P(UdpSocketTest, TimestampIoctlPersistence) { + // TODO(gvisor.dev/issue/1202): ioctl() and SO_TIMESTAMP socket option are not + // supported by hostinet. + SKIP_IF(IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1304,7 +1336,6 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { msg.msg_controllen = sizeof(cmsgbuf); ASSERT_THAT(RetryEINTR(recvmsg)(s_, &msg, 0), SyscallSucceedsWithValue(0)); struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg); - cmsg = CMSG_FIRSTHDR(&msg); ASSERT_NE(cmsg, nullptr); // The ioctl should return the exact same values as before. @@ -1314,5 +1345,154 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { ASSERT_EQ(tv.tv_usec, tv2.tv_usec); } +// Test that a socket with IP_TOS or IPV6_TCLASS set will set the TOS byte on +// outgoing packets, and that a receiving socket with IP_RECVTOS or +// IPV6_RECVTCLASS will create the corresponding control message. +TEST_P(UdpSocketTest, SetAndReceiveTOS) { + // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. + SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + + // Allow socket to receive control message. + int recv_level = SOL_IP; + int recv_type = IP_RECVTOS; + if (GetParam() != AddressFamily::kIpv4) { + recv_level = SOL_IPV6; + recv_type = IPV6_RECVTCLASS; + } + ASSERT_THAT( + setsockopt(s_, recv_level, recv_type, &kSockOptOn, sizeof(kSockOptOn)), + SyscallSucceeds()); + + // Set socket TOS. + int sent_level = recv_level; + int sent_type = IP_TOS; + if (sent_level == SOL_IPV6) { + sent_type = IPV6_TCLASS; + } + int sent_tos = IPTOS_LOWDELAY; // Choose some TOS value. + ASSERT_THAT( + setsockopt(t_, sent_level, sent_type, &sent_tos, sizeof(sent_tos)), + SyscallSucceeds()); + + // Prepare message to send. + constexpr size_t kDataLength = 1024; + struct msghdr sent_msg = {}; + struct iovec sent_iov = {}; + char sent_data[kDataLength]; + sent_iov.iov_base = &sent_data[0]; + sent_iov.iov_len = kDataLength; + sent_msg.msg_iov = &sent_iov; + sent_msg.msg_iovlen = 1; + + ASSERT_THAT(RetryEINTR(sendmsg)(t_, &sent_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + // Receive message. + struct msghdr received_msg = {}; + struct iovec received_iov = {}; + char received_data[kDataLength]; + received_iov.iov_base = &received_data[0]; + received_iov.iov_len = kDataLength; + received_msg.msg_iov = &received_iov; + received_msg.msg_iovlen = 1; + size_t cmsg_data_len = sizeof(int8_t); + if (sent_type == IPV6_TCLASS) { + cmsg_data_len = sizeof(int); + } + std::vector<char> received_cmsgbuf(CMSG_SPACE(cmsg_data_len)); + received_msg.msg_control = &received_cmsgbuf[0]; + received_msg.msg_controllen = received_cmsgbuf.size(); + ASSERT_THAT(RetryEINTR(recvmsg)(s_, &received_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + struct cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); + ASSERT_NE(cmsg, nullptr); + EXPECT_EQ(cmsg->cmsg_len, CMSG_LEN(cmsg_data_len)); + EXPECT_EQ(cmsg->cmsg_level, sent_level); + EXPECT_EQ(cmsg->cmsg_type, sent_type); + int8_t received_tos = 0; + memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); + EXPECT_EQ(received_tos, sent_tos); +} + +// Test that sendmsg with IP_TOS and IPV6_TCLASS control messages will set the +// TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or +// IPV6_RECVTCLASS will create the corresponding control message. +TEST_P(UdpSocketTest, SendAndReceiveTOS) { + // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. + SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + + // Allow socket to receive control message. + int recv_level = SOL_IP; + int recv_type = IP_RECVTOS; + if (GetParam() != AddressFamily::kIpv4) { + recv_level = SOL_IPV6; + recv_type = IPV6_RECVTCLASS; + } + int recv_opt = kSockOptOn; + ASSERT_THAT( + setsockopt(s_, recv_level, recv_type, &recv_opt, sizeof(recv_opt)), + SyscallSucceeds()); + + // Prepare message to send. + constexpr size_t kDataLength = 1024; + int sent_level = recv_level; + int sent_type = IP_TOS; + int sent_tos = IPTOS_LOWDELAY; // Choose some TOS value. + + struct msghdr sent_msg = {}; + struct iovec sent_iov = {}; + char sent_data[kDataLength]; + sent_iov.iov_base = &sent_data[0]; + sent_iov.iov_len = kDataLength; + sent_msg.msg_iov = &sent_iov; + sent_msg.msg_iovlen = 1; + size_t cmsg_data_len = sizeof(int8_t); + if (sent_level == SOL_IPV6) { + sent_type = IPV6_TCLASS; + cmsg_data_len = sizeof(int); + } + std::vector<char> sent_cmsgbuf(CMSG_SPACE(cmsg_data_len)); + sent_msg.msg_control = &sent_cmsgbuf[0]; + sent_msg.msg_controllen = CMSG_LEN(cmsg_data_len); + + // Manually add control message. + struct cmsghdr* sent_cmsg = CMSG_FIRSTHDR(&sent_msg); + sent_cmsg->cmsg_len = CMSG_LEN(cmsg_data_len); + sent_cmsg->cmsg_level = sent_level; + sent_cmsg->cmsg_type = sent_type; + *(int8_t*)CMSG_DATA(sent_cmsg) = sent_tos; + + ASSERT_THAT(RetryEINTR(sendmsg)(t_, &sent_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + // Receive message. + struct msghdr received_msg = {}; + struct iovec received_iov = {}; + char received_data[kDataLength]; + received_iov.iov_base = &received_data[0]; + received_iov.iov_len = kDataLength; + received_msg.msg_iov = &received_iov; + received_msg.msg_iovlen = 1; + std::vector<char> received_cmsgbuf(CMSG_SPACE(cmsg_data_len)); + received_msg.msg_control = &received_cmsgbuf[0]; + received_msg.msg_controllen = CMSG_LEN(cmsg_data_len); + ASSERT_THAT(RetryEINTR(recvmsg)(s_, &received_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + struct cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); + ASSERT_NE(cmsg, nullptr); + EXPECT_EQ(cmsg->cmsg_len, CMSG_LEN(cmsg_data_len)); + EXPECT_EQ(cmsg->cmsg_level, sent_level); + EXPECT_EQ(cmsg->cmsg_type, sent_type); + int8_t received_tos = 0; + memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); + EXPECT_EQ(received_tos, sent_tos); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc new file mode 100644 index 000000000..75740238c --- /dev/null +++ b/test/syscalls/linux/xattr.cc @@ -0,0 +1,488 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include <errno.h> +#include <fcntl.h> +#include <limits.h> +#include <sys/types.h> +#include <sys/xattr.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/file_base.h" +#include "test/util/capability_util.h" +#include "test/util/posix_error.h" +#include "test/util/temp_path.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class XattrTest : public FileTest {}; + +TEST_F(XattrTest, XattrNullName) { + const char* path = test_file_name_.c_str(); + + EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(EFAULT)); + EXPECT_THAT(getxattr(path, nullptr, nullptr, 0), + SyscallFailsWithErrno(EFAULT)); +} + +TEST_F(XattrTest, XattrEmptyName) { + const char* path = test_file_name_.c_str(); + + EXPECT_THAT(setxattr(path, "", nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(ERANGE)); + EXPECT_THAT(getxattr(path, "", nullptr, 0), SyscallFailsWithErrno(ERANGE)); +} + +TEST_F(XattrTest, XattrLargeName) { + const char* path = test_file_name_.c_str(); + std::string name = "user."; + name += std::string(XATTR_NAME_MAX - name.length(), 'a'); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), + SyscallSucceedsWithValue(0)); + } + + name += "a"; + EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(ERANGE)); + EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), + SyscallFailsWithErrno(ERANGE)); +} + +TEST_F(XattrTest, XattrInvalidPrefix) { + const char* path = test_file_name_.c_str(); + std::string name(XATTR_NAME_MAX, 'a'); + EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(EOPNOTSUPP)); + EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), + SyscallFailsWithErrno(EOPNOTSUPP)); +} + +TEST_F(XattrTest, XattrReadOnly) { + // Drop capabilities that allow us to override file and directory permissions. + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + size_t size = sizeof(val); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), + SyscallSucceeds()); + } + + ASSERT_NO_ERRNO(testing::Chmod(test_file_name_, S_IRUSR)); + + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), + SyscallFailsWithErrno(EACCES)); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + char buf = '-'; + EXPECT_THAT(getxattr(path, name, &buf, size), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, val); + } +} + +TEST_F(XattrTest, XattrWriteOnly) { + // Drop capabilities that allow us to override file and directory permissions. + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); + + ASSERT_NO_ERRNO(testing::Chmod(test_file_name_, S_IWUSR)); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + size_t size = sizeof(val); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), + SyscallSucceeds()); + } + + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(EACCES)); +} + +TEST_F(XattrTest, XattrTrustedWithNonadmin) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + SKIP_IF(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + const char* path = test_file_name_.c_str(); + const char name[] = "trusted.abc"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); +} + +TEST_F(XattrTest, XattrOnDirectory) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(dir.path().c_str(), name, NULL, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(getxattr(dir.path().c_str(), name, NULL, 0), + SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, XattrOnSymlink) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(link.path().c_str(), name, NULL, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(getxattr(link.path().c_str(), name, NULL, 0), + SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, XattrOnInvalidFileTypes) { + char name[] = "user.abc"; + + char char_device[] = "/dev/zero"; + EXPECT_THAT(setxattr(char_device, name, NULL, 0, /*flags=*/0), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(getxattr(char_device, name, NULL, 0), + SyscallFailsWithErrno(ENODATA)); + + // Use tmpfs, where creation of named pipes is supported. + const std::string fifo = NewTempAbsPathInDir("/dev/shm"); + const char* path = fifo.c_str(); + EXPECT_THAT(mknod(path, S_IFIFO | S_IRUSR | S_IWUSR, 0), SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name, NULL, 0, /*flags=*/0), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(getxattr(path, name, NULL, 0), SyscallFailsWithErrno(ENODATA)); +} + +TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + std::vector<char> val = {'a', 'a'}; + size_t size = 1; + EXPECT_THAT(setxattr(path, name, val.data(), size, /*flags=*/0), + SyscallSucceeds()); + + std::vector<char> buf = {'-', '-'}; + std::vector<char> expected_buf = {'a', '-'}; + EXPECT_THAT(getxattr(path, name, buf.data(), buf.size()), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, expected_buf); +} + +TEST_F(XattrTest, SetxattrZeroSize) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + EXPECT_THAT(setxattr(path, name, &val, 0, /*flags=*/0), SyscallSucceeds()); + + char buf = '-'; + EXPECT_THAT(getxattr(path, name, &buf, XATTR_SIZE_MAX), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(buf, '-'); +} + +TEST_F(XattrTest, SetxattrSizeTooLarge) { + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + + // Note that each particular fs implementation may stipulate a lower size + // limit, in which case we actually may fail (e.g. error with ENOSPC) for + // some sizes under XATTR_SIZE_MAX. + size_t size = XATTR_SIZE_MAX + 1; + std::vector<char> val(size); + EXPECT_THAT(setxattr(path, name, val.data(), size, /*flags=*/0), + SyscallFailsWithErrno(E2BIG)); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(getxattr(path, name, nullptr, 0), + SyscallFailsWithErrno(ENODATA)); + } +} + +TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(path, name, nullptr, 1, /*flags=*/0), + SyscallFailsWithErrno(EFAULT)); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(getxattr(path, name, nullptr, 0), + SyscallFailsWithErrno(ENODATA)); + } +} + +TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + std::vector<char> val(XATTR_SIZE_MAX + 1); + std::fill(val.begin(), val.end(), 'a'); + size_t size = 1; + EXPECT_THAT(setxattr(path, name, val.data(), size, /*flags=*/0), + SyscallSucceeds()); + + std::vector<char> buf = {'-', '-'}; + std::vector<char> expected_buf = {'a', '-'}; + EXPECT_THAT(getxattr(path, name, buf.data(), size), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, expected_buf); +} + +TEST_F(XattrTest, SetxattrReplaceWithSmaller) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + std::vector<char> val = {'a', 'a'}; + EXPECT_THAT(setxattr(path, name, val.data(), 2, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name, val.data(), 1, /*flags=*/0), + SyscallSucceeds()); + + std::vector<char> buf = {'-', '-'}; + std::vector<char> expected_buf = {'a', '-'}; + EXPECT_THAT(getxattr(path, name, buf.data(), 2), SyscallSucceedsWithValue(1)); + EXPECT_EQ(buf, expected_buf); +} + +TEST_F(XattrTest, SetxattrReplaceWithLarger) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + std::vector<char> val = {'a', 'a'}; + EXPECT_THAT(setxattr(path, name, val.data(), 1, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name, val.data(), 2, /*flags=*/0), + SyscallSucceeds()); + + std::vector<char> buf = {'-', '-'}; + EXPECT_THAT(getxattr(path, name, buf.data(), 2), SyscallSucceedsWithValue(2)); + EXPECT_EQ(buf, val); +} + +TEST_F(XattrTest, SetxattrCreateFlag) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_CREATE), + SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_CREATE), + SyscallFailsWithErrno(EEXIST)); + + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, SetxattrReplaceFlag) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_REPLACE), + SyscallFailsWithErrno(ENODATA)); + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_REPLACE), + SyscallSucceeds()); + + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, SetxattrInvalidFlags) { + const char* path = test_file_name_.c_str(); + int invalid_flags = 0xff; + EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, invalid_flags), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_F(XattrTest, Getxattr) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + int val = 1234; + size_t size = sizeof(val); + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + + int buf = 0; + EXPECT_THAT(getxattr(path, name, &buf, size), SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, val); +} + +TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + std::vector<char> val = {'a', 'a'}; + size_t size = val.size(); + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + + char buf = '-'; + EXPECT_THAT(getxattr(path, name, &buf, 1), SyscallFailsWithErrno(ERANGE)); + EXPECT_EQ(buf, '-'); +} + +TEST_F(XattrTest, GetxattrSizeLargerThanValue) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + EXPECT_THAT(setxattr(path, name, &val, 1, /*flags=*/0), SyscallSucceeds()); + + std::vector<char> buf(XATTR_SIZE_MAX); + std::fill(buf.begin(), buf.end(), '-'); + std::vector<char> expected_buf = buf; + expected_buf[0] = 'a'; + EXPECT_THAT(getxattr(path, name, buf.data(), buf.size()), + SyscallSucceedsWithValue(1)); + EXPECT_EQ(buf, expected_buf); +} + +TEST_F(XattrTest, GetxattrZeroSize) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + EXPECT_THAT(setxattr(path, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + + char buf = '-'; + EXPECT_THAT(getxattr(path, name, &buf, 0), + SyscallSucceedsWithValue(sizeof(val))); + EXPECT_EQ(buf, '-'); +} + +TEST_F(XattrTest, GetxattrSizeTooLarge) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + EXPECT_THAT(setxattr(path, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + + std::vector<char> buf(XATTR_SIZE_MAX + 1); + std::fill(buf.begin(), buf.end(), '-'); + std::vector<char> expected_buf = buf; + expected_buf[0] = 'a'; + EXPECT_THAT(getxattr(path, name, buf.data(), buf.size()), + SyscallSucceedsWithValue(sizeof(val))); + EXPECT_EQ(buf, expected_buf); +} + +TEST_F(XattrTest, GetxattrNullValue) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + size_t size = sizeof(val); + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + + EXPECT_THAT(getxattr(path, name, nullptr, size), + SyscallFailsWithErrno(EFAULT)); +} + +TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { + // TODO(b/127675828): Support setxattr and getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + char name[] = "user.abc"; + char val = 'a'; + size_t size = sizeof(val); + // Set value with zero size. + EXPECT_THAT(setxattr(path, name, &val, 0, /*flags=*/0), SyscallSucceeds()); + // Get value with nonzero size. + EXPECT_THAT(getxattr(path, name, nullptr, size), SyscallSucceedsWithValue(0)); + + // Set value with nonzero size. + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + // Get value with zero size. + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallSucceedsWithValue(size)); +} + +TEST_F(XattrTest, GetxattrNonexistentName) { + // TODO(b/127675828): Support getxattr. + SKIP_IF(IsRunningOnGvisor()); + + const char* path = test_file_name_.c_str(); + std::string name = "user.nonexistent"; + EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), + SyscallFailsWithErrno(ENODATA)); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/syscall_test_runner.go b/test/syscalls/syscall_test_runner.go index accf46347..b9fd885ff 100644 --- a/test/syscalls/syscall_test_runner.go +++ b/test/syscalls/syscall_test_runner.go @@ -46,6 +46,7 @@ var ( debug = flag.Bool("debug", false, "enable debug logs") strace = flag.Bool("strace", false, "enable strace logs") platform = flag.String("platform", "ptrace", "platform to run on") + network = flag.String("network", "none", "network stack to run on (sandbox, host, none)") useTmpfs = flag.Bool("use-tmpfs", false, "mounts tmpfs for /tmp") fileAccess = flag.String("file-access", "exclusive", "mounts root in exclusive or shared mode") overlay = flag.Bool("overlay", false, "wrap filesystem mounts with writable tmpfs overlay") @@ -137,7 +138,7 @@ func runRunsc(tc gtest.TestCase, spec *specs.Spec) error { args := []string{ "-root", rootDir, - "-network=none", + "-network", *network, "-log-format=text", "-TESTONLY-unsafe-nonroot=true", "-net-raw=true", @@ -335,10 +336,11 @@ func runTestCaseRunsc(testBin string, tc gtest.TestCase, t *testing.T) { }) } - // Set environment variable that indicates we are - // running in gVisor and with the given platform. + // Set environment variables that indicate we are + // running in gVisor with the given platform and network. platformVar := "TEST_ON_GVISOR" - env := append(os.Environ(), platformVar+"="+*platform) + networkVar := "GVISOR_NETWORK" + env := append(os.Environ(), platformVar+"="+*platform, networkVar+"="+*network) // Remove env variables that cause the gunit binary to write output // files, since they will stomp on eachother, and on the output files diff --git a/test/util/fs_util.cc b/test/util/fs_util.cc index 88b1e7911..042cec94a 100644 --- a/test/util/fs_util.cc +++ b/test/util/fs_util.cc @@ -105,6 +105,15 @@ PosixErrorOr<struct stat> Stat(absl::string_view path) { return stat_buf; } +PosixErrorOr<struct stat> Lstat(absl::string_view path) { + struct stat stat_buf; + int res = lstat(std::string(path).c_str(), &stat_buf); + if (res < 0) { + return PosixError(errno, absl::StrCat("lstat ", path)); + } + return stat_buf; +} + PosixErrorOr<struct stat> Fstat(int fd) { struct stat stat_buf; int res = fstat(fd, &stat_buf); @@ -127,7 +136,7 @@ PosixErrorOr<bool> Exists(absl::string_view path) { } PosixErrorOr<bool> IsDirectory(absl::string_view path) { - ASSIGN_OR_RETURN_ERRNO(struct stat stat_buf, Stat(path)); + ASSIGN_OR_RETURN_ERRNO(struct stat stat_buf, Lstat(path)); if (S_ISDIR(stat_buf.st_mode)) { return true; } diff --git a/test/util/test_util.cc b/test/util/test_util.cc index 9cb050735..848504c88 100644 --- a/test/util/test_util.cc +++ b/test/util/test_util.cc @@ -41,6 +41,7 @@ namespace gvisor { namespace testing { #define TEST_ON_GVISOR "TEST_ON_GVISOR" +#define GVISOR_NETWORK "GVISOR_NETWORK" bool IsRunningOnGvisor() { return GvisorPlatform() != Platform::kNative; } @@ -60,6 +61,11 @@ Platform GvisorPlatform() { abort(); } +bool IsRunningWithHostinet() { + char* env = getenv(GVISOR_NETWORK); + return env && strcmp(env, "host") == 0; +} + // Inline cpuid instruction. Preserve %ebx/%rbx register. In PIC compilations // %ebx contains the address of the global offset table. %rbx is occasionally // used to address stack variables in presence of dynamic allocas. diff --git a/test/util/test_util.h b/test/util/test_util.h index ee6c2bf4d..b3235c7e3 100644 --- a/test/util/test_util.h +++ b/test/util/test_util.h @@ -220,6 +220,7 @@ enum class Platform { }; bool IsRunningOnGvisor(); Platform GvisorPlatform(); +bool IsRunningWithHostinet(); #ifdef __linux__ void SetupGvisorDeathTest(); |