diff options
229 files changed, 6887 insertions, 4893 deletions
diff --git a/.github/ISSUE_TEMPLATE/bug_report.md b/.github/ISSUE_TEMPLATE/bug_report.md new file mode 100644 index 000000000..49a1ba697 --- /dev/null +++ b/.github/ISSUE_TEMPLATE/bug_report.md @@ -0,0 +1,31 @@ +--- +name: Bug report +about: Create a bug report to help us improve +title: +labels: + - 'type: bug' +assignees: '' +--- + +**Description** + +A clear description of what the bug is. If possible, explicitly indicate the +expected behavior vs. the observed behavior. + +**Steps to reproduce** + +If available, please include detailed reproduction steps. + +If the bug requires software that is not publicly available, see if it can be +reproduced with software that is publicly available. + +**Environment** + +Please include the following details of your environment: + +* `runsc -v` +* `docker version` or `docker info` (if available) +* `kubectl version` and `kubectl get nodes` (if using Kubernetes) +* `uname -a` +* `git describe` (if built from source) +* `runsc` debug logs (if available) diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml new file mode 100644 index 000000000..772c9a0ac --- /dev/null +++ b/.github/ISSUE_TEMPLATE/config.yml @@ -0,0 +1,11 @@ +blank_issues_enabled: false +contact_links: + - name: gVisor Documentation (FAQ) + url: https://gvisor.dev/docs/user_guide/faq/ + about: Please see our documentation for common questions and answers. + - name: gVisor Documentation (Debugging) + url: https://gvisor.dev/docs/user_guide/debugging/ + about: Please see our documentation for debugging tips. + - name: gVisor User Forum + url: https://groups.google.com/g/gvisor-users + about: Ask and answer general questions here. diff --git a/.github/ISSUE_TEMPLATE/feature_request.md b/.github/ISSUE_TEMPLATE/feature_request.md new file mode 100644 index 000000000..65f60f385 --- /dev/null +++ b/.github/ISSUE_TEMPLATE/feature_request.md @@ -0,0 +1,21 @@ +--- +name: Feature request +about: Suggest an idea or improvement +title: '' +labels: + - 'type: enhancement' +assignees: '' +--- + +**Description** + +A clear description of the feature or enhancement. + +**Is this feature related to a specific bug?** + +Please include a bug references if yes. + +**Do you have a specific solution in mind?** + +Please include any details about a solution that you have in mind, including any +alternatives considered. diff --git a/.github/issue_template.md b/.github/issue_template.md deleted file mode 100644 index 77c401d22..000000000 --- a/.github/issue_template.md +++ /dev/null @@ -1,20 +0,0 @@ -Before filling an issue, please consult our FAQ: -https://gvisor.dev/docs/user_guide/faq/ - -Also check that the issue hasn't been reported before. - -If you have a question, please email gvisor-users@googlegroups.com rather than filing a bug. - -If you believe you've found a security issue, please email gvisor-security@googlegroups.com rather than filing a bug. - -If this is your first time compiling or running gVisor, please make sure that your system meets the minimum requirements: https://github.com/google/gvisor#requirements - -For all other issues, please attach debug logs. To get debug logs, follow the -instructions here: https://gvisor.dev/docs/user_guide/debugging/ - -Other useful information to include is: - -* `runsc -v` -* `docker version` or `docker info` if more relevant -* `uname -a` - `git describe` -* Detailed reproduction steps diff --git a/.github/labeler.yml b/.github/labeler.yml new file mode 100644 index 000000000..b6a17051c --- /dev/null +++ b/.github/labeler.yml @@ -0,0 +1,42 @@ +"arch: arm": + - "**/*_arm64.*" + - "**/*_aarch64.*" +"arch: x86_64": + - "**/*_amd64.*" + - "**/*_x86.*" +"area: bazel": + - "**/BUILD" + - "**/*.bzl" +"area: docs": + - "**/g3doc/**" + - "**/README.md" +"area: filesystem": + - "pkg/sentry/fs/**" + - "pkg/sentry/vfs/**" + - "pkg/sentry/fsimpl/**" +"area: hostinet": + - "pkg/sentry/socket/hostinet/**" +"area: networking": + - "pkg/tcpip/**" + - "pkg/sentry/socket/**" +"area: kernel": + - "pkg/sentry/arch/**" + - "pkg/sentry/kernel/**" + - "pkg/sentry/syscalls/**" +"area: mm": + - "pkg/sentry/mm/**" +"area: tests": + - "**/tests/**" + - "**/*_test.go" + - "**/test/**" +"area: tooling": + - "tools/**" +"dependencies": + - "WORKSPACE" + - "go.mod" + - "go.sum" +"platform: kvm": + - "pkg/sentry/platform/kvm/**" + - "pkg/sentry/platform/ring0/**" +"platform: ptrace": + - "pkg/sentry/platform/ptrace/**" diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md new file mode 100644 index 000000000..264b4e9fa --- /dev/null +++ b/.github/pull_request_template.md @@ -0,0 +1,5 @@ +* [ ] Have you followed the guidelines in [CONTRIBUTING.md](../blob/master/CONTRIBUTING.md)? +* [ ] Have you formatted and linted your code? +* [ ] Have you added relevant tests? +* [ ] Have you added appropriate Fixes & Updates references? +* [ ] If yes, please erase all these lines! diff --git a/.github/workflows/labeler.yml b/.github/workflows/labeler.yml new file mode 100644 index 000000000..b5fd10352 --- /dev/null +++ b/.github/workflows/labeler.yml @@ -0,0 +1,11 @@ +name: "Labeler" +on: +- pull_request + +jobs: + label: + runs-on: ubuntu-latest + steps: + - uses: actions/labeler@v2 + with: + repo-token: "${{ secrets.GITHUB_TOKEN }}" @@ -127,45 +127,6 @@ 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", - digest = "sha256:a137d692a2eb9fc7bf95c5f4a568da090e2c31098e93634421ed88f3a3f1db65", - registry = "gcr.io", - repository = "gvisor-presubmit/iptables-test", -) - -load( - "@io_bazel_rules_docker//go:image.bzl", - _go_image_repos = "repositories", -) - -_go_image_repos() - # Load C++ grpc rules. http_archive( name = "com_github_grpc_grpc", @@ -228,13 +189,6 @@ go_repository( ) go_repository( - name = "com_github_imdario_mergo", - importpath = "github.com/imdario/mergo", - sum = "h1:CGgOkSJeqMRmt0D9XLWExdT4m4F1vd3FV3VPt+0VxkQ=", - version = "v0.3.8", -) - -go_repository( name = "com_github_kr_pretty", importpath = "github.com/kr/pretty", sum = "h1:s5hAObm+yFO5uHYt5dYjxi2rXrsnmRpJx4OYvIWUaQs=", diff --git a/kokoro/benchmark_tests.cfg b/kokoro/benchmark_tests.cfg deleted file mode 100644 index f85cc9681..000000000 --- a/kokoro/benchmark_tests.cfg +++ /dev/null @@ -1,26 +0,0 @@ -build_file : 'repo/scripts/benchmark.sh' - - -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id : 73898 - keyname : 'gvisor-benchmarks-service-account' - }, - } -} - -env_vars { - key : 'PROJECT' - value : 'gvisor-benchmarks' -} - -env_vars { - key : 'ZONE' - value : 'us-central1-b' -} - -env_vars { - key : 'GCLOUD_CREDENTIALS' - value : '73898_gvisor-benchmarks-service-account' -} diff --git a/kokoro/build.cfg b/kokoro/build.cfg deleted file mode 100644 index c9ceda947..000000000 --- a/kokoro/build.cfg +++ /dev/null @@ -1,24 +0,0 @@ -build_file: "repo/scripts/build.sh" - -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id: 73898 - keyname: "kokoro-repo-key" - } - } -} - -env_vars { - key: "KOKORO_REPO_KEY" - value: "73898_kokoro-repo-key" -} - -action { - define_artifacts { - regex: "**/runsc" - regex: "**/runsc.*" - regex: "**/dists/**" - regex: "**/pool/**" - } -} diff --git a/kokoro/build_tests.cfg b/kokoro/build_tests.cfg deleted file mode 100644 index c64b7e679..000000000 --- a/kokoro/build_tests.cfg +++ /dev/null @@ -1 +0,0 @@ -build_file: "repo/scripts/build.sh" diff --git a/kokoro/common.cfg b/kokoro/common.cfg deleted file mode 100644 index 669a2e458..000000000 --- a/kokoro/common.cfg +++ /dev/null @@ -1,29 +0,0 @@ -# Give Kokoro access to Remote Build Executor (RBE) service account key. -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id: 73898 - keyname: "kokoro-rbe-service-account" - } - } -} - -# Configure bazel to access RBE. -bazel_setting { - # Our GCP project name. - project_id: "gvisor-rbe" - - # Use RBE for execution as well as caching. - local_execution: false - - # This must match the values in the job config. - auth_credential: { - keystore_config_id: 73898 - keyname: "kokoro-rbe-service-account" - } - - # Do not change unless you know what you are doing. - bes_backend_address: "buildeventservice.googleapis.com" - foundry_backend_address: "remotebuildexecution.googleapis.com" - upsalite_frontend_address: "https://source.cloud.google.com" -} diff --git a/kokoro/do_tests.cfg b/kokoro/do_tests.cfg deleted file mode 100644 index b45ec0b42..000000000 --- a/kokoro/do_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/do_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/docker_tests.cfg b/kokoro/docker_tests.cfg deleted file mode 100644 index 0a0ef87ed..000000000 --- a/kokoro/docker_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/docker_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc_logs_*.tar.gz" - } -} diff --git a/kokoro/go.cfg b/kokoro/go.cfg deleted file mode 100644 index b9c1fcb12..000000000 --- a/kokoro/go.cfg +++ /dev/null @@ -1,20 +0,0 @@ -build_file: "repo/scripts/go.sh" - -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id: 73898 - keyname: "kokoro-github-access-token" - } - } -} - -env_vars { - key: "KOKORO_GITHUB_ACCESS_TOKEN" - value: "73898_kokoro-github-access-token" -} - -env_vars { - key: "KOKORO_GO_PUSH" - value: "true" -} diff --git a/kokoro/go_tests.cfg b/kokoro/go_tests.cfg deleted file mode 100644 index 5eb51041a..000000000 --- a/kokoro/go_tests.cfg +++ /dev/null @@ -1 +0,0 @@ -build_file: "repo/scripts/go.sh" diff --git a/kokoro/hostnet_tests.cfg b/kokoro/hostnet_tests.cfg deleted file mode 100644 index 520dc55a3..000000000 --- a/kokoro/hostnet_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/hostnet_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc_logs_*.tar.gz" - } -} diff --git a/kokoro/iptables_tests.cfg b/kokoro/iptables_tests.cfg deleted file mode 100644 index a30d82591..000000000 --- a/kokoro/iptables_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/iptables_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc_logs_*.tar.gz" - } -} diff --git a/kokoro/issue_reviver.cfg b/kokoro/issue_reviver.cfg deleted file mode 100644 index 2370d9250..000000000 --- a/kokoro/issue_reviver.cfg +++ /dev/null @@ -1,15 +0,0 @@ -build_file: "repo/scripts/issue_reviver.sh" - -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id: 73898 - keyname: "kokoro-github-access-token" - } - } -} - -env_vars { - key: "KOKORO_GITHUB_ACCESS_TOKEN" - value: "73898_kokoro-github-access-token" -} diff --git a/kokoro/kvm_tests.cfg b/kokoro/kvm_tests.cfg deleted file mode 100644 index 1feb60c8a..000000000 --- a/kokoro/kvm_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/kvm_tests.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 deleted file mode 100644 index ccf657983..000000000 --- a/kokoro/kythe/generate_xrefs.cfg +++ /dev/null @@ -1,29 +0,0 @@ -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" - fail_if_no_artifacts: true - } -} diff --git a/kokoro/kythe/generate_xrefs.sh b/kokoro/kythe/generate_xrefs.sh deleted file mode 100644 index 323b0f77b..000000000 --- a/kokoro/kythe/generate_xrefs.sh +++ /dev/null @@ -1,54 +0,0 @@ -#!/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 - -if command -v use_bazel.sh >/dev/null; then - use_bazel.sh latest -fi -bazel version - -python3 -V - -readonly KYTHE_VERSION='v0.0.41' -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="${KYTHE_DIR}" \ - --define=kythe_corpus=github.com/google/gvisor \ - --cxxopt=-std=c++17 \ - --config=remote \ - --auth_credentials="${KOKORO_BAZEL_AUTH_CREDENTIAL}" \ - //... - -"${KYTHE_DIR}/tools/kzip" merge \ - --output "$KZIP_FILENAME" \ - $(find -L bazel-out/*/extra_actions/ -name '*.kzip') diff --git a/kokoro/make_tests.cfg b/kokoro/make_tests.cfg deleted file mode 100644 index d973130ff..000000000 --- a/kokoro/make_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/make_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/overlay_tests.cfg b/kokoro/overlay_tests.cfg deleted file mode 100644 index 6a2ddbd03..000000000 --- a/kokoro/overlay_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/overlay_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc_logs_*.tar.gz" - } -} diff --git a/kokoro/packetdrill_tests.cfg b/kokoro/packetdrill_tests.cfg deleted file mode 100644 index 258d7deb4..000000000 --- a/kokoro/packetdrill_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/packetdrill_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/packetimpact_tests.cfg b/kokoro/packetimpact_tests.cfg deleted file mode 100644 index db86b52d5..000000000 --- a/kokoro/packetimpact_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/packetimpact_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/release.cfg b/kokoro/release.cfg deleted file mode 100644 index 5cec1790a..000000000 --- a/kokoro/release.cfg +++ /dev/null @@ -1,15 +0,0 @@ -build_file: "repo/scripts/release.sh" - -before_action { - fetch_keystore { - keystore_resource { - keystore_config_id: 73898 - keyname: "kokoro-github-access-token" - } - } -} - -env_vars { - key: "KOKORO_GITHUB_ACCESS_TOKEN" - value: "73898_kokoro-github-access-token" -} diff --git a/kokoro/root_tests.cfg b/kokoro/root_tests.cfg deleted file mode 100644 index 28351695c..000000000 --- a/kokoro/root_tests.cfg +++ /dev/null @@ -1,10 +0,0 @@ -build_file: "repo/scripts/root_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc_logs_*.tar.gz" - } -} diff --git a/kokoro/runtime_tests.cfg b/kokoro/runtime_tests.cfg deleted file mode 100644 index 7d56d5aca..000000000 --- a/kokoro/runtime_tests.cfg +++ /dev/null @@ -1 +0,0 @@ -build_file: "repo/scripts/runtime_tests.sh" diff --git a/kokoro/runtime_tests/go1.12.cfg b/kokoro/runtime_tests/go1.12.cfg deleted file mode 100644 index 04bfe2868..000000000 --- a/kokoro/runtime_tests/go1.12.cfg +++ /dev/null @@ -1,16 +0,0 @@ -build_file: "github/github/scripts/runtime_tests.sh" - -env_vars { - key: "RUNTIME_TEST_NAME" - value: "go1.12" -} - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc" - regex: "**/runsc.*" - } -}
\ No newline at end of file diff --git a/kokoro/runtime_tests/java11.cfg b/kokoro/runtime_tests/java11.cfg deleted file mode 100644 index c82855cd2..000000000 --- a/kokoro/runtime_tests/java11.cfg +++ /dev/null @@ -1,16 +0,0 @@ -build_file: "github/github/scripts/runtime_tests.sh" - -env_vars { - key: "RUNTIME_TEST_NAME" - value: "java11" -} - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc" - regex: "**/runsc.*" - } -}
\ No newline at end of file diff --git a/kokoro/runtime_tests/nodejs12.4.0.cfg b/kokoro/runtime_tests/nodejs12.4.0.cfg deleted file mode 100644 index 5512db5df..000000000 --- a/kokoro/runtime_tests/nodejs12.4.0.cfg +++ /dev/null @@ -1,16 +0,0 @@ -build_file: "github/github/scripts/runtime_tests.sh" - -env_vars { - key: "RUNTIME_TEST_NAME" - value: "nodejs12.4.0" -} - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc" - regex: "**/runsc.*" - } -}
\ No newline at end of file diff --git a/kokoro/runtime_tests/php7.3.6.cfg b/kokoro/runtime_tests/php7.3.6.cfg deleted file mode 100644 index bc9ac92aa..000000000 --- a/kokoro/runtime_tests/php7.3.6.cfg +++ /dev/null @@ -1,16 +0,0 @@ -build_file: "github/github/scripts/runtime_tests.sh" - -env_vars { - key: "RUNTIME_TEST_NAME" - value: "php7.3.6" -} - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc" - regex: "**/runsc.*" - } -}
\ No newline at end of file diff --git a/kokoro/runtime_tests/python3.7.3.cfg b/kokoro/runtime_tests/python3.7.3.cfg deleted file mode 100644 index 12eb13860..000000000 --- a/kokoro/runtime_tests/python3.7.3.cfg +++ /dev/null @@ -1,16 +0,0 @@ -build_file: "github/github/scripts/runtime_tests.sh" - -env_vars { - key: "RUNTIME_TEST_NAME" - value: "python3.7.3" -} - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - regex: "**/runsc" - regex: "**/runsc.*" - } -}
\ No newline at end of file diff --git a/kokoro/runtime_tests/runtime_tests.sh b/kokoro/runtime_tests/runtime_tests.sh deleted file mode 100755 index 73a58f806..000000000 --- a/kokoro/runtime_tests/runtime_tests.sh +++ /dev/null @@ -1,29 +0,0 @@ -#!/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. - -# Run in the root of the repo. -cd "$(dirname "$0")" -cd "$(git rev-parse --show-toplevel)" - -source scripts/common.sh - -if [ ! -v RUNTIME_TEST_NAME ]; then - echo 'Must set $RUNTIME_TEST_NAME' >&2 - exit 1 -fi - -install_runsc_for_test runtimes -test_runsc "//test/runtimes:${RUNTIME_TEST_NAME}_test" diff --git a/kokoro/simple_tests.cfg b/kokoro/simple_tests.cfg deleted file mode 100644 index 32e0a9431..000000000 --- a/kokoro/simple_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/simple_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/swgso_tests.cfg b/kokoro/swgso_tests.cfg deleted file mode 100644 index 101a9c607..000000000 --- a/kokoro/swgso_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/swgso_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/syscall_kvm_tests.cfg b/kokoro/syscall_kvm_tests.cfg deleted file mode 100644 index 3b99e9c13..000000000 --- a/kokoro/syscall_kvm_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/syscall_kvm_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/kokoro/syscall_tests.cfg b/kokoro/syscall_tests.cfg deleted file mode 100644 index ee6e4a3a4..000000000 --- a/kokoro/syscall_tests.cfg +++ /dev/null @@ -1,9 +0,0 @@ -build_file: "repo/scripts/syscall_tests.sh" - -action { - define_artifacts { - regex: "**/sponge_log.xml" - regex: "**/sponge_log.log" - regex: "**/outputs.zip" - } -} diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 322d1ccc4..59b0e138a 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -10,6 +10,7 @@ go_library( name = "linux", srcs = [ "aio.go", + "arch_amd64.go", "audit.go", "bpf.go", "capability.go", diff --git a/test/root/testdata/busybox.go b/pkg/abi/linux/arch_amd64.go index e4dbd2843..0be31e755 100644 --- a/test/root/testdata/busybox.go +++ b/pkg/abi/linux/arch_amd64.go @@ -1,5 +1,5 @@ -// Copyright 2018 The gVisor Authors. -// +// Copyright 2020 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 @@ -12,21 +12,12 @@ // See the License for the specific language governing permissions and // limitations under the License. -package testdata +// +build amd64 + +package linux -// MountOverSymlink is a JSON config for a container that /etc/resolv.conf is a -// symlink to /tmp/resolv.conf. -var MountOverSymlink = ` -{ - "metadata": { - "name": "busybox" - }, - "image": { - "image": "k8s.gcr.io/busybox" - }, - "command": [ - "sleep", - "1000" - ] -} -` +// Start and end addresses of the vsyscall page. +const ( + VSyscallStartAddr uint64 = 0xffffffffff600000 + VSyscallEndAddr uint64 = 0xffffffffff601000 +) diff --git a/pkg/abi/linux/epoll_amd64.go b/pkg/abi/linux/epoll_amd64.go index 34ff18009..7e74b1143 100644 --- a/pkg/abi/linux/epoll_amd64.go +++ b/pkg/abi/linux/epoll_amd64.go @@ -12,11 +12,13 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build amd64 + package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). // -// +marshal +// +marshal slice:EpollEventSlice type EpollEvent struct { Events uint32 // Linux makes struct epoll_event::data a __u64. We represent it as diff --git a/pkg/abi/linux/epoll_arm64.go b/pkg/abi/linux/epoll_arm64.go index f86c35329..a35939cc9 100644 --- a/pkg/abi/linux/epoll_arm64.go +++ b/pkg/abi/linux/epoll_arm64.go @@ -12,11 +12,13 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). // -// +marshal +// +marshal slice:EpollEventSlice type EpollEvent struct { Events uint32 // Linux makes struct epoll_event a __u64, necessitating 4 bytes of padding diff --git a/pkg/abi/linux/mm.go b/pkg/abi/linux/mm.go index cd043dac3..07cc1895e 100644 --- a/pkg/abi/linux/mm.go +++ b/pkg/abi/linux/mm.go @@ -90,14 +90,19 @@ const ( MS_SYNC = 1 << 2 ) +// NumaPolicy is the NUMA memory policy for a memory range. See numa(7). +// +// +marshal +type NumaPolicy int32 + // Policies for get_mempolicy(2)/set_mempolicy(2). const ( - MPOL_DEFAULT = 0 - MPOL_PREFERRED = 1 - MPOL_BIND = 2 - MPOL_INTERLEAVE = 3 - MPOL_LOCAL = 4 - MPOL_MAX = 5 + MPOL_DEFAULT NumaPolicy = 0 + MPOL_PREFERRED NumaPolicy = 1 + MPOL_BIND NumaPolicy = 2 + MPOL_INTERLEAVE NumaPolicy = 3 + MPOL_LOCAL NumaPolicy = 4 + MPOL_MAX NumaPolicy = 5 ) // Flags for get_mempolicy(2). diff --git a/pkg/abi/linux/seccomp.go b/pkg/abi/linux/seccomp.go index 4eeb5cd7a..d0607e256 100644 --- a/pkg/abi/linux/seccomp.go +++ b/pkg/abi/linux/seccomp.go @@ -63,3 +63,10 @@ func (a BPFAction) String() string { func (a BPFAction) Data() uint16 { return uint16(a & SECCOMP_RET_DATA) } + +// SockFprog is sock_fprog taken from <linux/filter.h>. +type SockFprog struct { + Len uint16 + pad [6]byte + Filter *BPFInstruction +} diff --git a/pkg/flipcall/packet_window_allocator.go b/pkg/flipcall/packet_window_allocator.go index ccb918fab..af9cc3d21 100644 --- a/pkg/flipcall/packet_window_allocator.go +++ b/pkg/flipcall/packet_window_allocator.go @@ -134,7 +134,7 @@ func (pwa *PacketWindowAllocator) Allocate(size int) (PacketWindowDescriptor, er start := pwa.nextAlloc pwa.nextAlloc = end return PacketWindowDescriptor{ - FD: pwa.fd, + FD: pwa.FD(), Offset: start, Length: size, }, nil @@ -158,7 +158,7 @@ func (pwa *PacketWindowAllocator) ensureFileSize(min int64) error { } newSize = newNewSize } - if err := syscall.Ftruncate(pwa.fd, newSize); err != nil { + if err := syscall.Ftruncate(pwa.FD(), newSize); err != nil { return fmt.Errorf("ftruncate failed: %v", err) } pwa.fileSize = newSize diff --git a/pkg/seccomp/seccomp_unsafe.go b/pkg/seccomp/seccomp_unsafe.go index be328db12..f7e986589 100644 --- a/pkg/seccomp/seccomp_unsafe.go +++ b/pkg/seccomp/seccomp_unsafe.go @@ -21,13 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" ) -// sockFprog is sock_fprog taken from <linux/filter.h>. -type sockFprog struct { - Len uint16 - pad [6]byte - Filter *linux.BPFInstruction -} - // SetFilter installs the given BPF program. // // This is safe to call from an afterFork context. @@ -39,7 +32,7 @@ func SetFilter(instrs []linux.BPFInstruction) syscall.Errno { return errno } - sockProg := sockFprog{ + sockProg := linux.SockFprog{ Len: uint16(len(instrs)), Filter: (*linux.BPFInstruction)(unsafe.Pointer(&instrs[0])), } diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 0c1db4b13..1cb1adf8c 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -98,9 +98,12 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt if ucSize < 0 { panic("can't get size of UContext64") } - // st.Arch.Width() is for the restorer address. sizeof(siginfo) == 128. - frameSize := int(st.Arch.Width()) + ucSize + 128 - frameBottom := (sp-usermem.Addr(frameSize)) & ^usermem.Addr(15) - 8 + + // frameSize = ucSize + sizeof(siginfo). + // sizeof(siginfo) == 128. + // R30 stores the restorer address. + frameSize := ucSize + 128 + frameBottom := (sp - usermem.Addr(frameSize)) & ^usermem.Addr(15) sp = frameBottom + usermem.Addr(frameSize) st.Bottom = sp @@ -130,12 +133,27 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt c.Regs.Regs[0] = uint64(info.Signo) c.Regs.Regs[1] = uint64(infoAddr) c.Regs.Regs[2] = uint64(ucAddr) - + c.Regs.Regs[30] = uint64(act.Restorer) return nil } // SignalRestore implements Context.SignalRestore. -// Only used on intel. func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { - return 0, SignalStack{}, nil + // Copy out the stack frame. + var uc UContext64 + if _, err := st.Pop(&uc); err != nil { + return 0, SignalStack{}, err + } + var info SignalInfo + if _, err := st.Pop(&info); err != nil { + return 0, SignalStack{}, err + } + + // Restore registers. + c.Regs.Regs = uc.MContext.Regs + c.Regs.Pc = uc.MContext.Pc + c.Regs.Sp = uc.MContext.Sp + c.Regs.Pstate = uc.MContext.Pstate + + return uc.Sigset, uc.Stack, nil } diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 4d42eac83..4bbe90198 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -73,8 +73,7 @@ func checkTaskState(t *kernel.Task) error { type taskDir struct { ramfs.Dir - t *kernel.Task - pidns *kernel.PIDNamespace + t *kernel.Task } var _ fs.InodeOperations = (*taskDir)(nil) diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 12b1c6097..2e9dd2d55 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -27,6 +27,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + const ( // canonMaxBytes is the number of bytes that fit into a single line of // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE @@ -443,3 +445,5 @@ func (l *lineDiscipline) peek(b []byte) int { } return size } + +// LINT.ThenChange(../../fsimpl/devpts/line_discipline.go) diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index f62da49bd..fe07fa929 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -26,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // masterInodeOperations are the fs.InodeOperations for the master end of the // Terminal (ptmx file). // @@ -232,3 +234,5 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { unimpl.EmitUnimplementedEvent(ctx) } } + +// LINT.ThenChange(../../fsimpl/devpts/master.go) diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 1ca79c0b2..ceabb9b1e 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // waitBufMaxBytes is the maximum size of a wait buffer. It is based on // TTYB_DEFAULT_MEM_LIMIT. const waitBufMaxBytes = 131072 @@ -234,3 +236,5 @@ func (q *queue) waitBufAppend(b []byte) { q.waitBuf = append(q.waitBuf, b) q.waitBufLen += uint64(len(b)) } + +// LINT.ThenChange(../../fsimpl/devpts/queue.go) diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 6a2dbc576..9871f6fc6 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // slaveInodeOperations are the fs.InodeOperations for the slave end of the // Terminal (pts file). // @@ -172,3 +174,5 @@ func (sf *slaveFileOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem return 0, syserror.ENOTTY } } + +// LINT.ThenChange(../../fsimpl/devpts/slave.go) diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index 5883f26db..ddcccf4da 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // Terminal is a pseudoterminal. // // +stateify savable @@ -126,3 +128,5 @@ func (tm *Terminal) tty(isMaster bool) *kernel.TTY { } return tm.slaveKTTY } + +// LINT.ThenChange(../../fsimpl/devpts/terminal.go) diff --git a/pkg/sentry/fs/user/BUILD b/pkg/sentry/fs/user/BUILD new file mode 100644 index 000000000..f37f979f1 --- /dev/null +++ b/pkg/sentry/fs/user/BUILD @@ -0,0 +1,34 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "user", + srcs = ["user.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/fs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/usermem", + ], +) + +go_test( + name = "user_test", + size = "small", + srcs = ["user_test.go"], + library = ":user", + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fs", + "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/contexttest", + "//pkg/usermem", + ], +) diff --git a/runsc/boot/user.go b/pkg/sentry/fs/user/user.go index 332e4fce5..fe7f67c00 100644 --- a/runsc/boot/user.go +++ b/pkg/sentry/fs/user/user.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package boot +package user import ( "bufio" @@ -136,10 +136,10 @@ func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth. return homeDir, nil } -// maybeAddExecUserHome returns a new slice with the HOME enviroment variable +// MaybeAddExecUserHome returns a new slice with the HOME enviroment variable // set if the slice does not already contain it, otherwise it returns the // original slice unmodified. -func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { +func MaybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { // Check if the envv already contains HOME. for _, env := range envv { if strings.HasPrefix(env, "HOME=") { @@ -159,7 +159,10 @@ func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth. return append(envv, "HOME="+homeDir), nil } -func maybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { +// MaybeAddExecUserHomeVFS2 returns a new slice with the HOME enviroment +// variable set if the slice does not already contain it, otherwise it returns +// the original slice unmodified. +func MaybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { // Check if the envv already contains HOME. for _, env := range envv { if strings.HasPrefix(env, "HOME=") { diff --git a/runsc/boot/user_test.go b/pkg/sentry/fs/user/user_test.go index fb4e13dfb..7d8e9ac7c 100644 --- a/runsc/boot/user_test.go +++ b/pkg/sentry/fs/user/user_test.go @@ -12,167 +12,111 @@ // See the License for the specific language governing permissions and // limitations under the License. -package boot +package user import ( - "io/ioutil" - "os" - "path/filepath" + "fmt" "strings" - "syscall" "testing" - specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/usermem" ) -func setupTempDir() (string, error) { - tmpDir, err := ioutil.TempDir(os.TempDir(), "exec-user-test") +// createEtcPasswd creates /etc/passwd with the given contents and mode. If +// mode is empty, then no file will be created. If mode is not a regular file +// mode, then contents is ignored. +func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode linux.FileMode) error { + if err := root.CreateDirectory(ctx, root, "etc", fs.FilePermsFromMode(0755)); err != nil { + return err + } + etc, err := root.Walk(ctx, root, "etc") if err != nil { - return "", err + return err } - return tmpDir, nil -} - -func setupPasswd(contents string, perms os.FileMode) func() (string, error) { - return func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - f, err := os.Create(filepath.Join(tmpDir, "etc", "passwd")) - if err != nil { - return "", err - } - defer f.Close() - - _, err = f.WriteString(contents) + defer etc.DecRef() + switch mode.FileType() { + case 0: + // Don't create anything. + return nil + case linux.S_IFREG: + passwd, err := etc.Create(ctx, root, "passwd", fs.FileFlags{Write: true}, fs.FilePermsFromMode(mode)) if err != nil { - return "", err + return err } - - err = f.Chmod(perms) - if err != nil { - return "", err + defer passwd.DecRef() + if _, err := passwd.Writev(ctx, usermem.BytesIOSequence([]byte(contents))); err != nil { + return err } - return tmpDir, nil + return nil + case linux.S_IFDIR: + return etc.CreateDirectory(ctx, root, "passwd", fs.FilePermsFromMode(mode)) + case linux.S_IFIFO: + return etc.CreateFifo(ctx, root, "passwd", fs.FilePermsFromMode(mode)) + default: + return fmt.Errorf("unknown file type %x", mode.FileType()) } } // TestGetExecUserHome tests the getExecUserHome function. func TestGetExecUserHome(t *testing.T) { tests := map[string]struct { - uid auth.KUID - createRoot func() (string, error) - expected string + uid auth.KUID + passwdContents string + passwdMode linux.FileMode + expected string }{ "success": { - uid: 1000, - createRoot: setupPasswd("adin::1000:1111::/home/adin:/bin/sh", 0666), - expected: "/home/adin", + uid: 1000, + passwdContents: "adin::1000:1111::/home/adin:/bin/sh", + passwdMode: linux.S_IFREG | 0666, + expected: "/home/adin", + }, + "no_perms": { + uid: 1000, + passwdContents: "adin::1000:1111::/home/adin:/bin/sh", + passwdMode: linux.S_IFREG, + expected: "/", }, "no_passwd": { - uid: 1000, - createRoot: setupTempDir, - expected: "/", + uid: 1000, + expected: "/", }, - "no_perms": { + "directory": { uid: 1000, - createRoot: setupPasswd("adin::1000:1111::/home/adin:/bin/sh", 0000), + passwdMode: linux.S_IFDIR | 0666, expected: "/", }, - "directory": { - uid: 1000, - createRoot: func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - if err := syscall.Mkdir(filepath.Join(tmpDir, "etc", "passwd"), 0666); err != nil { - return "", err - } - - return tmpDir, nil - }, - expected: "/", - }, // Currently we don't allow named pipes. "named_pipe": { - uid: 1000, - createRoot: func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - if err := syscall.Mkfifo(filepath.Join(tmpDir, "etc", "passwd"), 0666); err != nil { - return "", err - } - - return tmpDir, nil - }, - expected: "/", + uid: 1000, + passwdMode: linux.S_IFIFO | 0666, + expected: "/", }, } for name, tc := range tests { t.Run(name, func(t *testing.T) { - tmpDir, err := tc.createRoot() - if err != nil { - t.Fatalf("failed to create root dir: %v", err) - } - - sandEnd, cleanup, err := startGofer(tmpDir) - if err != nil { - t.Fatalf("failed to create gofer: %v", err) - } - defer cleanup() - ctx := contexttest.Context(t) - conf := &Config{ - RootDir: "unused_root_dir", - Network: NetworkNone, - DisableSeccomp: true, - } + msrc := fs.NewPseudoMountSource(ctx) + rootInode := tmpfs.NewDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0777), msrc) - spec := &specs.Spec{ - Root: &specs.Root{ - Path: tmpDir, - Readonly: true, - }, - // Add /proc mount as tmpfs to avoid needing a kernel. - Mounts: []specs.Mount{ - { - Destination: "/proc", - Type: "tmpfs", - }, - }, - } - - mntr := newContainerMounter(spec, []int{sandEnd}, nil, &podMountHints{}) - mns, err := mntr.createMountNamespace(ctx, conf) + mns, err := fs.NewMountNamespace(ctx, rootInode) if err != nil { - t.Fatalf("failed to create mount namespace: %v", err) + t.Fatalf("NewMountNamespace failed: %v", err) } - ctx = fs.WithRoot(ctx, mns.Root()) - if err := mntr.mountSubmounts(ctx, conf, mns); err != nil { - t.Fatalf("failed to create mount namespace: %v", err) + defer mns.DecRef() + root := mns.Root() + defer root.DecRef() + ctx = fs.WithRoot(ctx, root) + + if err := createEtcPasswd(ctx, root, tc.passwdContents, tc.passwdMode); err != nil { + t.Fatalf("createEtcPasswd failed: %v", err) } got, err := getExecUserHome(ctx, mns, tc.uid) diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD new file mode 100644 index 000000000..585764223 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -0,0 +1,43 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +licenses(["notice"]) + +go_library( + name = "devpts", + srcs = [ + "devpts.go", + "line_discipline.go", + "master.go", + "queue.go", + "slave.go", + "terminal.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/safemem", + "//pkg/sentry/arch", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/unimpl", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) + +go_test( + name = "devpts_test", + size = "small", + srcs = ["devpts_test.go"], + library = ":devpts", + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/contexttest", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go new file mode 100644 index 000000000..07a69b940 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -0,0 +1,207 @@ +// Copyright 2020 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 devpts provides a filesystem implementation that behaves like +// devpts. +package devpts + +import ( + "fmt" + "math" + "sort" + "strconv" + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Name is the filesystem name. +const Name = "devpts" + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +var _ vfs.FilesystemType = (*FilesystemType)(nil) + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + // No data allowed. + if opts.Data != "" { + return nil, nil, syserror.EINVAL + } + + fs, root := fstype.newFilesystem(vfsObj, creds) + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// newFilesystem creates a new devpts filesystem with root directory and ptmx +// master inode. It returns the filesystem and root Dentry. +func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*kernfs.Filesystem, *kernfs.Dentry) { + fs := &kernfs.Filesystem{} + fs.Init(vfsObj, fstype) + + // Construct the root directory. This is always inode id 1. + root := &rootInode{ + slaves: make(map[uint32]*slaveInode), + } + root.InodeAttrs.Init(creds, 1, linux.ModeDirectory|0555) + root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + root.dentry.Init(root) + + // Construct the pts master inode and dentry. Linux always uses inode + // id 2 for ptmx. See fs/devpts/inode.c:mknod_ptmx. + master := &masterInode{ + root: root, + } + master.InodeAttrs.Init(creds, 2, linux.ModeCharacterDevice|0666) + master.dentry.Init(master) + + // Add the master as a child of the root. + links := root.OrderedChildren.Populate(&root.dentry, map[string]*kernfs.Dentry{ + "ptmx": &master.dentry, + }) + root.IncLinks(links) + + return fs, &root.dentry +} + +// rootInode is the root directory inode for the devpts mounts. +type rootInode struct { + kernfs.AlwaysValid + kernfs.InodeAttrs + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeNotSymlink + kernfs.OrderedChildren + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // master is the master pty inode. Immutable. + master *masterInode + + // root is the root directory inode for this filesystem. Immutable. + root *rootInode + + // mu protects the fields below. + mu sync.Mutex + + // slaves maps pty ids to slave inodes. + slaves map[uint32]*slaveInode + + // nextIdx is the next pty index to use. Must be accessed atomically. + // + // TODO(b/29356795): reuse indices when ptys are closed. + nextIdx uint32 +} + +var _ kernfs.Inode = (*rootInode)(nil) + +// allocateTerminal creates a new Terminal and installs a pts node for it. +func (i *rootInode) allocateTerminal(creds *auth.Credentials) (*Terminal, error) { + i.mu.Lock() + defer i.mu.Unlock() + if i.nextIdx == math.MaxUint32 { + return nil, syserror.ENOMEM + } + idx := i.nextIdx + i.nextIdx++ + + // Sanity check that slave with idx does not exist. + if _, ok := i.slaves[idx]; ok { + panic(fmt.Sprintf("pty index collision; index %d already exists", idx)) + } + + // Create the new terminal and slave. + t := newTerminal(idx) + slave := &slaveInode{ + root: i, + t: t, + } + // Linux always uses pty index + 3 as the inode id. See + // fs/devpts/inode.c:devpts_pty_new(). + slave.InodeAttrs.Init(creds, uint64(idx+3), linux.ModeCharacterDevice|0600) + slave.dentry.Init(slave) + i.slaves[idx] = slave + + return t, nil +} + +// masterClose is called when the master end of t is closed. +func (i *rootInode) masterClose(t *Terminal) { + i.mu.Lock() + defer i.mu.Unlock() + + // Sanity check that slave with idx exists. + if _, ok := i.slaves[t.n]; !ok { + panic(fmt.Sprintf("pty with index %d does not exist", t.n)) + } + delete(i.slaves, t.n) +} + +// Open implements kernfs.Inode.Open. +func (i *rootInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + return fd.VFSFileDescription(), nil +} + +// Lookup implements kernfs.Inode.Lookup. +func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + idx, err := strconv.ParseUint(name, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + i.mu.Lock() + defer i.mu.Unlock() + if si, ok := i.slaves[uint32(idx)]; ok { + si.dentry.IncRef() + return si.dentry.VFSDentry(), nil + + } + return nil, syserror.ENOENT +} + +// IterDirents implements kernfs.Inode.IterDirents. +func (i *rootInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + i.mu.Lock() + defer i.mu.Unlock() + ids := make([]int, 0, len(i.slaves)) + for id := range i.slaves { + ids = append(ids, int(id)) + } + sort.Ints(ids) + for _, id := range ids[relOffset:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(id), 10), + Type: linux.DT_CHR, + Ino: i.slaves[uint32(id)].InodeAttrs.Ino(), + NextOff: offset + 1, + } + if err := cb.Handle(dirent); err != nil { + return offset, err + } + offset++ + } + return offset, nil +} diff --git a/pkg/sentry/fsimpl/devpts/devpts_test.go b/pkg/sentry/fsimpl/devpts/devpts_test.go new file mode 100644 index 000000000..b7c149047 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/devpts_test.go @@ -0,0 +1,56 @@ +// 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. + +package devpts + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/usermem" +) + +func TestSimpleMasterToSlave(t *testing.T) { + ld := newLineDiscipline(linux.DefaultSlaveTermios) + ctx := contexttest.Context(t) + inBytes := []byte("hello, tty\n") + src := usermem.BytesIOSequence(inBytes) + outBytes := make([]byte, 32) + dst := usermem.BytesIOSequence(outBytes) + + // Write to the input queue. + nw, err := ld.inputQueueWrite(ctx, src) + if err != nil { + t.Fatalf("error writing to input queue: %v", err) + } + if nw != int64(len(inBytes)) { + t.Fatalf("wrote wrong length: got %d, want %d", nw, len(inBytes)) + } + + // Read from the input queue. + nr, err := ld.inputQueueRead(ctx, dst) + if err != nil { + t.Fatalf("error reading from input queue: %v", err) + } + if nr != int64(len(inBytes)) { + t.Fatalf("read wrong length: got %d, want %d", nr, len(inBytes)) + } + + outStr := string(outBytes[:nr]) + inStr := string(inBytes) + if outStr != inStr { + t.Fatalf("written and read strings do not match: got %q, want %q", outStr, inStr) + } +} diff --git a/pkg/sentry/fsimpl/devpts/line_discipline.go b/pkg/sentry/fsimpl/devpts/line_discipline.go new file mode 100644 index 000000000..e201801d6 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/line_discipline.go @@ -0,0 +1,449 @@ +// 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. + +package devpts + +import ( + "bytes" + "unicode/utf8" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +const ( + // canonMaxBytes is the number of bytes that fit into a single line of + // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE + // in include/linux/tty.h. + canonMaxBytes = 4096 + + // nonCanonMaxBytes is the maximum number of bytes that can be read at + // a time in noncanonical mode. + nonCanonMaxBytes = canonMaxBytes - 1 + + spacesPerTab = 8 +) + +// lineDiscipline dictates how input and output are handled between the +// pseudoterminal (pty) master and slave. It can be configured to alter I/O, +// modify control characters (e.g. Ctrl-C for SIGINT), etc. The following man +// pages are good resources for how to affect the line discipline: +// +// * termios(3) +// * tty_ioctl(4) +// +// This file corresponds most closely to drivers/tty/n_tty.c. +// +// lineDiscipline has a simple structure but supports a multitude of options +// (see the above man pages). It consists of two queues of bytes: one from the +// terminal master to slave (the input queue) and one from slave to master (the +// output queue). When bytes are written to one end of the pty, the line +// discipline reads the bytes, modifies them or takes special action if +// required, and enqueues them to be read by the other end of the pty: +// +// input from terminal +-------------+ input to process (e.g. bash) +// +------------------------>| input queue |---------------------------+ +// | (inputQueueWrite) +-------------+ (inputQueueRead) | +// | | +// | v +// masterFD slaveFD +// ^ | +// | | +// | output to terminal +--------------+ output from process | +// +------------------------| output queue |<--------------------------+ +// (outputQueueRead) +--------------+ (outputQueueWrite) +// +// Lock order: +// termiosMu +// inQueue.mu +// outQueue.mu +// +// +stateify savable +type lineDiscipline struct { + // sizeMu protects size. + sizeMu sync.Mutex `state:"nosave"` + + // size is the terminal size (width and height). + size linux.WindowSize + + // inQueue is the input queue of the terminal. + inQueue queue + + // outQueue is the output queue of the terminal. + outQueue queue + + // termiosMu protects termios. + termiosMu sync.RWMutex `state:"nosave"` + + // termios is the terminal configuration used by the lineDiscipline. + termios linux.KernelTermios + + // column is the location in a row of the cursor. This is important for + // handling certain special characters like backspace. + column int + + // masterWaiter is used to wait on the master end of the TTY. + masterWaiter waiter.Queue `state:"zerovalue"` + + // slaveWaiter is used to wait on the slave end of the TTY. + slaveWaiter waiter.Queue `state:"zerovalue"` +} + +func newLineDiscipline(termios linux.KernelTermios) *lineDiscipline { + ld := lineDiscipline{termios: termios} + ld.inQueue.transformer = &inputQueueTransformer{} + ld.outQueue.transformer = &outputQueueTransformer{} + return &ld +} + +// getTermios gets the linux.Termios for the tty. +func (l *lineDiscipline) getTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + // We must copy a Termios struct, not KernelTermios. + t := l.termios.ToTermios() + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), t, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err +} + +// setTermios sets a linux.Termios for the tty. +func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + l.termiosMu.Lock() + defer l.termiosMu.Unlock() + oldCanonEnabled := l.termios.LEnabled(linux.ICANON) + // We must copy a Termios struct, not KernelTermios. + var t linux.Termios + _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &t, usermem.IOOpts{ + AddressSpaceActive: true, + }) + l.termios.FromTermios(t) + + // If canonical mode is turned off, move bytes from inQueue's wait + // buffer to its read buffer. Anything already in the read buffer is + // now readable. + if oldCanonEnabled && !l.termios.LEnabled(linux.ICANON) { + l.inQueue.mu.Lock() + l.inQueue.pushWaitBufLocked(l) + l.inQueue.readable = true + l.inQueue.mu.Unlock() + l.slaveWaiter.Notify(waiter.EventIn) + } + + return 0, err +} + +func (l *lineDiscipline) windowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + l.sizeMu.Lock() + defer l.sizeMu.Unlock() + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), l.size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err +} + +func (l *lineDiscipline) setWindowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + l.sizeMu.Lock() + defer l.sizeMu.Unlock() + _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &l.size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err +} + +func (l *lineDiscipline) masterReadiness() waiter.EventMask { + // We don't have to lock a termios because the default master termios + // is immutable. + return l.inQueue.writeReadiness(&linux.MasterTermios) | l.outQueue.readReadiness(&linux.MasterTermios) +} + +func (l *lineDiscipline) slaveReadiness() waiter.EventMask { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + return l.outQueue.writeReadiness(&l.termios) | l.inQueue.readReadiness(&l.termios) +} + +func (l *lineDiscipline) inputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + return l.inQueue.readableSize(ctx, io, args) +} + +func (l *lineDiscipline) inputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, pushed, err := l.inQueue.read(ctx, dst, l) + if err != nil { + return 0, err + } + if n > 0 { + l.masterWaiter.Notify(waiter.EventOut) + if pushed { + l.slaveWaiter.Notify(waiter.EventIn) + } + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) inputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, err := l.inQueue.write(ctx, src, l) + if err != nil { + return 0, err + } + if n > 0 { + l.slaveWaiter.Notify(waiter.EventIn) + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) outputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + return l.outQueue.readableSize(ctx, io, args) +} + +func (l *lineDiscipline) outputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, pushed, err := l.outQueue.read(ctx, dst, l) + if err != nil { + return 0, err + } + if n > 0 { + l.slaveWaiter.Notify(waiter.EventOut) + if pushed { + l.masterWaiter.Notify(waiter.EventIn) + } + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) outputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, err := l.outQueue.write(ctx, src, l) + if err != nil { + return 0, err + } + if n > 0 { + l.masterWaiter.Notify(waiter.EventIn) + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +// transformer is a helper interface to make it easier to stateify queue. +type transformer interface { + // transform functions require queue's mutex to be held. + transform(*lineDiscipline, *queue, []byte) int +} + +// outputQueueTransformer implements transformer. It performs line discipline +// transformations on the output queue. +// +// +stateify savable +type outputQueueTransformer struct{} + +// transform does output processing for one end of the pty. See +// drivers/tty/n_tty.c:do_output_char for an analogous kernel function. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (*outputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int { + // transformOutput is effectively always in noncanonical mode, as the + // master termios never has ICANON set. + + if !l.termios.OEnabled(linux.OPOST) { + q.readBuf = append(q.readBuf, buf...) + if len(q.readBuf) > 0 { + q.readable = true + } + return len(buf) + } + + var ret int + for len(buf) > 0 { + size := l.peek(buf) + cBytes := append([]byte{}, buf[:size]...) + ret += size + buf = buf[size:] + // We're guaranteed that cBytes has at least one element. + switch cBytes[0] { + case '\n': + if l.termios.OEnabled(linux.ONLRET) { + l.column = 0 + } + if l.termios.OEnabled(linux.ONLCR) { + q.readBuf = append(q.readBuf, '\r', '\n') + continue + } + case '\r': + if l.termios.OEnabled(linux.ONOCR) && l.column == 0 { + continue + } + if l.termios.OEnabled(linux.OCRNL) { + cBytes[0] = '\n' + if l.termios.OEnabled(linux.ONLRET) { + l.column = 0 + } + break + } + l.column = 0 + case '\t': + spaces := spacesPerTab - l.column%spacesPerTab + if l.termios.OutputFlags&linux.TABDLY == linux.XTABS { + l.column += spaces + q.readBuf = append(q.readBuf, bytes.Repeat([]byte{' '}, spacesPerTab)...) + continue + } + l.column += spaces + case '\b': + if l.column > 0 { + l.column-- + } + default: + l.column++ + } + q.readBuf = append(q.readBuf, cBytes...) + } + if len(q.readBuf) > 0 { + q.readable = true + } + return ret +} + +// inputQueueTransformer implements transformer. It performs line discipline +// transformations on the input queue. +// +// +stateify savable +type inputQueueTransformer struct{} + +// transform does input processing for one end of the pty. Characters read are +// transformed according to flags set in the termios struct. See +// drivers/tty/n_tty.c:n_tty_receive_char_special for an analogous kernel +// function. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (*inputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int { + // If there's a line waiting to be read in canonical mode, don't write + // anything else to the read buffer. + if l.termios.LEnabled(linux.ICANON) && q.readable { + return 0 + } + + maxBytes := nonCanonMaxBytes + if l.termios.LEnabled(linux.ICANON) { + maxBytes = canonMaxBytes + } + + var ret int + for len(buf) > 0 && len(q.readBuf) < canonMaxBytes { + size := l.peek(buf) + cBytes := append([]byte{}, buf[:size]...) + // We're guaranteed that cBytes has at least one element. + switch cBytes[0] { + case '\r': + if l.termios.IEnabled(linux.IGNCR) { + buf = buf[size:] + ret += size + continue + } + if l.termios.IEnabled(linux.ICRNL) { + cBytes[0] = '\n' + } + case '\n': + if l.termios.IEnabled(linux.INLCR) { + cBytes[0] = '\r' + } + } + + // In canonical mode, we discard non-terminating characters + // after the first 4095. + if l.shouldDiscard(q, cBytes) { + buf = buf[size:] + ret += size + continue + } + + // Stop if the buffer would be overfilled. + if len(q.readBuf)+size > maxBytes { + break + } + buf = buf[size:] + ret += size + + // If we get EOF, make the buffer available for reading. + if l.termios.LEnabled(linux.ICANON) && l.termios.IsEOF(cBytes[0]) { + q.readable = true + break + } + + q.readBuf = append(q.readBuf, cBytes...) + + // Anything written to the readBuf will have to be echoed. + if l.termios.LEnabled(linux.ECHO) { + l.outQueue.writeBytes(cBytes, l) + l.masterWaiter.Notify(waiter.EventIn) + } + + // If we finish a line, make it available for reading. + if l.termios.LEnabled(linux.ICANON) && l.termios.IsTerminating(cBytes) { + q.readable = true + break + } + } + + // In noncanonical mode, everything is readable. + if !l.termios.LEnabled(linux.ICANON) && len(q.readBuf) > 0 { + q.readable = true + } + + return ret +} + +// shouldDiscard returns whether c should be discarded. In canonical mode, if +// too many bytes are enqueued, we keep reading input and discarding it until +// we find a terminating character. Signal/echo processing still occurs. +// +// Precondition: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (l *lineDiscipline) shouldDiscard(q *queue, cBytes []byte) bool { + return l.termios.LEnabled(linux.ICANON) && len(q.readBuf)+len(cBytes) >= canonMaxBytes && !l.termios.IsTerminating(cBytes) +} + +// peek returns the size in bytes of the next character to process. As long as +// b isn't empty, peek returns a value of at least 1. +func (l *lineDiscipline) peek(b []byte) int { + size := 1 + // If UTF-8 support is enabled, runes might be multiple bytes. + if l.termios.IEnabled(linux.IUTF8) { + _, size = utf8.DecodeRune(b) + } + return size +} + +// LINT.ThenChange(../../fs/tty/line_discipline.go) diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go new file mode 100644 index 000000000..60340c28e --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -0,0 +1,226 @@ +// Copyright 2020 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 devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/unimpl" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// masterInode is the inode for the master end of the Terminal. +type masterInode struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // root is the devpts root inode. + root *rootInode +} + +var _ kernfs.Inode = (*masterInode)(nil) + +// Open implements kernfs.Inode.Open. +func (mi *masterInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + t, err := mi.root.allocateTerminal(rp.Credentials()) + if err != nil { + return nil, err + } + + mi.IncRef() + fd := &masterFileDescription{ + inode: mi, + t: t, + } + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + mi.DecRef() + return nil, err + } + return &fd.vfsfd, nil +} + +// Stat implements kernfs.Inode.Stat. +func (mi *masterInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := mi.InodeAttrs.Stat(vfsfs, opts) + if err != nil { + return linux.Statx{}, err + } + statx.Blksize = 1024 + statx.RdevMajor = linux.TTYAUX_MAJOR + statx.RdevMinor = linux.PTMX_MINOR + return statx, nil +} + +// SetStat implements kernfs.Inode.SetStat +func (mi *masterInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask&linux.STATX_SIZE != 0 { + return syserror.EINVAL + } + return mi.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) +} + +type masterFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + + inode *masterInode + t *Terminal +} + +var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) + +// Release implements vfs.FileDescriptionImpl.Release. +func (mfd *masterFileDescription) Release() { + mfd.inode.root.masterClose(mfd.t) + mfd.inode.DecRef() +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (mfd *masterFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + mfd.t.ld.masterWaiter.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (mfd *masterFileDescription) EventUnregister(e *waiter.Entry) { + mfd.t.ld.masterWaiter.EventUnregister(e) +} + +// Readiness implements waiter.Waitable.Readiness. +func (mfd *masterFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return mfd.t.ld.masterReadiness() +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (mfd *masterFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return mfd.t.ld.outputQueueRead(ctx, dst) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (mfd *masterFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return mfd.t.ld.inputQueueWrite(ctx, src) +} + +// Ioctl implements vfs.FileDescriptionImpl.Ioctl. +func (mfd *masterFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + switch cmd := args[1].Uint(); cmd { + case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ + // Get the number of bytes in the output queue read buffer. + return 0, mfd.t.ld.outputQueueReadSize(ctx, io, args) + case linux.TCGETS: + // N.B. TCGETS on the master actually returns the configuration + // of the slave end. + return mfd.t.ld.getTermios(ctx, io, args) + case linux.TCSETS: + // N.B. TCSETS on the master actually affects the configuration + // of the slave end. + return mfd.t.ld.setTermios(ctx, io, args) + case linux.TCSETSW: + // TODO(b/29356795): This should drain the output queue first. + return mfd.t.ld.setTermios(ctx, io, args) + case linux.TIOCGPTN: + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(mfd.t.n), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + case linux.TIOCSPTLCK: + // TODO(b/29356795): Implement pty locking. For now just pretend we do. + return 0, nil + case linux.TIOCGWINSZ: + return 0, mfd.t.ld.windowSize(ctx, io, args) + case linux.TIOCSWINSZ: + return 0, mfd.t.ld.setWindowSize(ctx, io, args) + case linux.TIOCSCTTY: + // Make the given terminal the controlling terminal of the + // calling process. + return 0, mfd.t.setControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, mfd.t.releaseControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return mfd.t.foregroundProcessGroup(ctx, io, args, true /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return mfd.t.setForegroundProcessGroup(ctx, io, args, true /* isMaster */) + default: + maybeEmitUnimplementedEvent(ctx, cmd) + return 0, syserror.ENOTTY + } +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (mfd *masterFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + creds := auth.CredentialsFromContext(ctx) + fs := mfd.vfsfd.VirtualDentry().Mount().Filesystem() + return mfd.inode.SetStat(ctx, fs, creds, opts) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (mfd *masterFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := mfd.vfsfd.VirtualDentry().Mount().Filesystem() + return mfd.inode.Stat(fs, opts) +} + +// maybeEmitUnimplementedEvent emits unimplemented event if cmd is valid. +func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { + switch cmd { + case linux.TCGETS, + linux.TCSETS, + linux.TCSETSW, + linux.TCSETSF, + linux.TIOCGWINSZ, + linux.TIOCSWINSZ, + linux.TIOCSETD, + linux.TIOCSBRK, + linux.TIOCCBRK, + linux.TCSBRK, + linux.TCSBRKP, + linux.TIOCSTI, + linux.TIOCCONS, + linux.FIONBIO, + linux.TIOCEXCL, + linux.TIOCNXCL, + linux.TIOCGEXCL, + linux.TIOCGSID, + linux.TIOCGETD, + linux.TIOCVHANGUP, + linux.TIOCGDEV, + linux.TIOCMGET, + linux.TIOCMSET, + linux.TIOCMBIC, + linux.TIOCMBIS, + linux.TIOCGICOUNT, + linux.TCFLSH, + linux.TIOCSSERIAL, + linux.TIOCGPTPEER: + + unimpl.EmitUnimplementedEvent(ctx) + } +} + +// LINT.ThenChange(../../fs/tty/master.go) diff --git a/pkg/sentry/fsimpl/devpts/queue.go b/pkg/sentry/fsimpl/devpts/queue.go new file mode 100644 index 000000000..29a6be858 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/queue.go @@ -0,0 +1,240 @@ +// 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. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// waitBufMaxBytes is the maximum size of a wait buffer. It is based on +// TTYB_DEFAULT_MEM_LIMIT. +const waitBufMaxBytes = 131072 + +// queue represents one of the input or output queues between a pty master and +// slave. Bytes written to a queue are added to the read buffer until it is +// full, at which point they are written to the wait buffer. Bytes are +// processed (i.e. undergo termios transformations) as they are added to the +// read buffer. The read buffer is readable when its length is nonzero and +// readable is true. +// +// +stateify savable +type queue struct { + // mu protects everything in queue. + mu sync.Mutex `state:"nosave"` + + // readBuf is buffer of data ready to be read when readable is true. + // This data has been processed. + readBuf []byte + + // waitBuf contains data that can't fit into readBuf. It is put here + // until it can be loaded into the read buffer. waitBuf contains data + // that hasn't been processed. + waitBuf [][]byte + waitBufLen uint64 + + // readable indicates whether the read buffer can be read from. In + // canonical mode, there can be an unterminated line in the read buffer, + // so readable must be checked. + readable bool + + // transform is the the queue's function for transforming bytes + // entering the queue. For example, transform might convert all '\r's + // entering the queue to '\n's. + transformer +} + +// readReadiness returns whether q is ready to be read from. +func (q *queue) readReadiness(t *linux.KernelTermios) waiter.EventMask { + q.mu.Lock() + defer q.mu.Unlock() + if len(q.readBuf) > 0 && q.readable { + return waiter.EventIn + } + return waiter.EventMask(0) +} + +// writeReadiness returns whether q is ready to be written to. +func (q *queue) writeReadiness(t *linux.KernelTermios) waiter.EventMask { + q.mu.Lock() + defer q.mu.Unlock() + if q.waitBufLen < waitBufMaxBytes { + return waiter.EventOut + } + return waiter.EventMask(0) +} + +// readableSize writes the number of readable bytes to userspace. +func (q *queue) readableSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + q.mu.Lock() + defer q.mu.Unlock() + var size int32 + if q.readable { + size = int32(len(q.readBuf)) + } + + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err + +} + +// read reads from q to userspace. It returns the number of bytes read as well +// as whether the read caused more readable data to become available (whether +// data was pushed from the wait buffer to the read buffer). +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) read(ctx context.Context, dst usermem.IOSequence, l *lineDiscipline) (int64, bool, error) { + q.mu.Lock() + defer q.mu.Unlock() + + if !q.readable { + return 0, false, syserror.ErrWouldBlock + } + + if dst.NumBytes() > canonMaxBytes { + dst = dst.TakeFirst(canonMaxBytes) + } + + n, err := dst.CopyOutFrom(ctx, safemem.ReaderFunc(func(dst safemem.BlockSeq) (uint64, error) { + src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(q.readBuf)) + n, err := safemem.CopySeq(dst, src) + if err != nil { + return 0, err + } + q.readBuf = q.readBuf[n:] + + // If we read everything, this queue is no longer readable. + if len(q.readBuf) == 0 { + q.readable = false + } + + return n, nil + })) + if err != nil { + return 0, false, err + } + + // Move data from the queue's wait buffer to its read buffer. + nPushed := q.pushWaitBufLocked(l) + + return int64(n), nPushed > 0, nil +} + +// write writes to q from userspace. +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) write(ctx context.Context, src usermem.IOSequence, l *lineDiscipline) (int64, error) { + q.mu.Lock() + defer q.mu.Unlock() + + // Copy data into the wait buffer. + n, err := src.CopyInTo(ctx, safemem.WriterFunc(func(src safemem.BlockSeq) (uint64, error) { + copyLen := src.NumBytes() + room := waitBufMaxBytes - q.waitBufLen + // If out of room, return EAGAIN. + if room == 0 && copyLen > 0 { + return 0, syserror.ErrWouldBlock + } + // Cap the size of the wait buffer. + if copyLen > room { + copyLen = room + src = src.TakeFirst64(room) + } + buf := make([]byte, copyLen) + + // Copy the data into the wait buffer. + dst := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)) + n, err := safemem.CopySeq(dst, src) + if err != nil { + return 0, err + } + q.waitBufAppend(buf) + + return n, nil + })) + if err != nil { + return 0, err + } + + // Push data from the wait to the read buffer. + q.pushWaitBufLocked(l) + + return n, nil +} + +// writeBytes writes to q from b. +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) writeBytes(b []byte, l *lineDiscipline) { + q.mu.Lock() + defer q.mu.Unlock() + + // Write to the wait buffer. + q.waitBufAppend(b) + q.pushWaitBufLocked(l) +} + +// pushWaitBufLocked fills the queue's read buffer with data from the wait +// buffer. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be locked. +func (q *queue) pushWaitBufLocked(l *lineDiscipline) int { + if q.waitBufLen == 0 { + return 0 + } + + // Move data from the wait to the read buffer. + var total int + var i int + for i = 0; i < len(q.waitBuf); i++ { + n := q.transform(l, q, q.waitBuf[i]) + total += n + if n != len(q.waitBuf[i]) { + // The read buffer filled up without consuming the + // entire buffer. + q.waitBuf[i] = q.waitBuf[i][n:] + break + } + } + + // Update wait buffer based on consumed data. + q.waitBuf = q.waitBuf[i:] + q.waitBufLen -= uint64(total) + + return total +} + +// Precondition: q.mu must be locked. +func (q *queue) waitBufAppend(b []byte) { + q.waitBuf = append(q.waitBuf, b) + q.waitBufLen += uint64(len(b)) +} + +// LINT.ThenChange(../../fs/tty/queue.go) diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go new file mode 100644 index 000000000..e7e50d51e --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -0,0 +1,186 @@ +// Copyright 2020 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 devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// slaveInode is the inode for the slave end of the Terminal. +type slaveInode struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // root is the devpts root inode. + root *rootInode + + // t is the connected Terminal. + t *Terminal +} + +var _ kernfs.Inode = (*slaveInode)(nil) + +// Open implements kernfs.Inode.Open. +func (si *slaveInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + si.IncRef() + fd := &slaveFileDescription{ + inode: si, + } + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + si.DecRef() + return nil, err + } + return &fd.vfsfd, nil + +} + +// Valid implements kernfs.Inode.Valid. +func (si *slaveInode) Valid(context.Context) bool { + // Return valid if the slave still exists. + si.root.mu.Lock() + defer si.root.mu.Unlock() + _, ok := si.root.slaves[si.t.n] + return ok +} + +// Stat implements kernfs.Inode.Stat. +func (si *slaveInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := si.InodeAttrs.Stat(vfsfs, opts) + if err != nil { + return linux.Statx{}, err + } + statx.Blksize = 1024 + statx.RdevMajor = linux.UNIX98_PTY_SLAVE_MAJOR + statx.RdevMinor = si.t.n + return statx, nil +} + +// SetStat implements kernfs.Inode.SetStat +func (si *slaveInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask&linux.STATX_SIZE != 0 { + return syserror.EINVAL + } + return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) +} + +type slaveFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + + inode *slaveInode +} + +var _ vfs.FileDescriptionImpl = (*slaveFileDescription)(nil) + +// Release implements fs.FileOperations.Release. +func (sfd *slaveFileDescription) Release() { + sfd.inode.DecRef() +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (sfd *slaveFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + sfd.inode.t.ld.slaveWaiter.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (sfd *slaveFileDescription) EventUnregister(e *waiter.Entry) { + sfd.inode.t.ld.slaveWaiter.EventUnregister(e) +} + +// Readiness implements waiter.Waitable.Readiness. +func (sfd *slaveFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return sfd.inode.t.ld.slaveReadiness() +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (sfd *slaveFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return sfd.inode.t.ld.inputQueueRead(ctx, dst) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (sfd *slaveFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return sfd.inode.t.ld.outputQueueWrite(ctx, src) +} + +// Ioctl implements vfs.FileDescripionImpl.Ioctl. +func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + switch cmd := args[1].Uint(); cmd { + case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ + // Get the number of bytes in the input queue read buffer. + return 0, sfd.inode.t.ld.inputQueueReadSize(ctx, io, args) + case linux.TCGETS: + return sfd.inode.t.ld.getTermios(ctx, io, args) + case linux.TCSETS: + return sfd.inode.t.ld.setTermios(ctx, io, args) + case linux.TCSETSW: + // TODO(b/29356795): This should drain the output queue first. + return sfd.inode.t.ld.setTermios(ctx, io, args) + case linux.TIOCGPTN: + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(sfd.inode.t.n), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + case linux.TIOCGWINSZ: + return 0, sfd.inode.t.ld.windowSize(ctx, io, args) + case linux.TIOCSWINSZ: + return 0, sfd.inode.t.ld.setWindowSize(ctx, io, args) + case linux.TIOCSCTTY: + // Make the given terminal the controlling terminal of the + // calling process. + return 0, sfd.inode.t.setControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, sfd.inode.t.releaseControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return sfd.inode.t.foregroundProcessGroup(ctx, io, args, false /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return sfd.inode.t.setForegroundProcessGroup(ctx, io, args, false /* isMaster */) + default: + maybeEmitUnimplementedEvent(ctx, cmd) + return 0, syserror.ENOTTY + } +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (sfd *slaveFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + creds := auth.CredentialsFromContext(ctx) + fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() + return sfd.inode.SetStat(ctx, fs, creds, opts) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() + return sfd.inode.Stat(fs, opts) +} + +// LINT.ThenChange(../../fs/tty/slave.go) diff --git a/pkg/sentry/fsimpl/devpts/terminal.go b/pkg/sentry/fsimpl/devpts/terminal.go new file mode 100644 index 000000000..b44e673d8 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/terminal.go @@ -0,0 +1,124 @@ +// 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. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/usermem" +) + +// LINT.IfChanges + +// Terminal is a pseudoterminal. +// +// +stateify savable +type Terminal struct { + // n is the terminal index. It is immutable. + n uint32 + + // ld is the line discipline of the terminal. It is immutable. + ld *lineDiscipline + + // masterKTTY contains the controlling process of the master end of + // this terminal. This field is immutable. + masterKTTY *kernel.TTY + + // slaveKTTY contains the controlling process of the slave end of this + // terminal. This field is immutable. + slaveKTTY *kernel.TTY +} + +func newTerminal(n uint32) *Terminal { + termios := linux.DefaultSlaveTermios + t := Terminal{ + n: n, + ld: newLineDiscipline(termios), + masterKTTY: &kernel.TTY{Index: n}, + slaveKTTY: &kernel.TTY{Index: n}, + } + return &t +} + +// setControllingTTY makes tm the controlling terminal of the calling thread +// group. +func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setControllingTTY must be called from a task context") + } + + return task.ThreadGroup().SetControllingTTY(tm.tty(isMaster), args[2].Int()) +} + +// releaseControllingTTY removes tm as the controlling terminal of the calling +// thread group. +func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("releaseControllingTTY must be called from a task context") + } + + return task.ThreadGroup().ReleaseControllingTTY(tm.tty(isMaster)) +} + +// foregroundProcessGroup gets the process group ID of tm's foreground process. +func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("foregroundProcessGroup must be called from a task context") + } + + ret, err := task.ThreadGroup().ForegroundProcessGroup(tm.tty(isMaster)) + if err != nil { + return 0, err + } + + // Write it out to *arg. + _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(ret), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err +} + +// foregroundProcessGroup sets tm's foreground process. +func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setForegroundProcessGroup must be called from a task context") + } + + // Read in the process group ID. + var pgid int32 + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgid, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + + ret, err := task.ThreadGroup().SetForegroundProcessGroup(tm.tty(isMaster), kernel.ProcessGroupID(pgid)) + return uintptr(ret), err +} + +func (tm *Terminal) tty(isMaster bool) *kernel.TTY { + if isMaster { + return tm.masterKTTY + } + return tm.slaveKTTY +} + +// LINT.ThenChange(../../fs/tty/terminal.go) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index 64f1b142c..142ee53b0 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -163,16 +163,25 @@ func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind v func (a *Accessor) UserspaceInit(ctx context.Context) error { actx := a.wrapContext(ctx) - // systemd: src/shared/dev-setup.c:dev_setup() + // Initialize symlinks. for _, symlink := range []struct { source string target string }{ - // /proc/kcore is not implemented. + // systemd: src/shared/dev-setup.c:dev_setup() {source: "fd", target: "/proc/self/fd"}, {source: "stdin", target: "/proc/self/fd/0"}, {source: "stdout", target: "/proc/self/fd/1"}, {source: "stderr", target: "/proc/self/fd/2"}, + // /proc/kcore is not implemented. + + // Linux implements /dev/ptmx as a device node, but advises + // container implementations to create /dev/ptmx as a symlink + // to pts/ptmx (Documentation/filesystems/devpts.txt). Systemd + // follows this advice (src/nspawn/nspawn.c:setup_pts()), while + // LXC tries to create a bind mount and falls back to a symlink + // (src/lxc/conf.c:lxc_setup_devpts()). + {source: "ptmx", target: "pts/ptmx"}, } { if err := a.vfsObj.SymlinkAt(actx, a.creds, a.pathOperationAt(symlink.source), symlink.target); err != nil { return fmt.Errorf("failed to create symlink %q => %q: %v", symlink.source, symlink.target, err) diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index a4947c480..ff861d0fe 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -93,8 +93,8 @@ go_test( "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/test/testutil", "//pkg/usermem", - "//runsc/testutil", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", ], diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 29bb73765..64e9a579f 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -32,9 +32,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/pkg/usermem" - - "gvisor.dev/gvisor/runsc/testutil" ) const ( diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index acd061905..b9c4beee4 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -35,7 +35,6 @@ go_library( "fstree.go", "gofer.go", "handle.go", - "handle_unsafe.go", "p9file.go", "pagemath.go", "regular_file.go", @@ -53,6 +52,7 @@ go_library( "//pkg/p9", "//pkg/safemem", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/hostfd", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index d02691232..c67766ab2 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -21,8 +21,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func (d *dentry) isDir() bool { @@ -41,15 +43,46 @@ func (d *dentry) cacheNewChildLocked(child *dentry, name string) { d.children[name] = child } -// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != -// InteropModeShared. -func (d *dentry) cacheNegativeChildLocked(name string) { +// Preconditions: d.dirMu must be locked. d.isDir(). +func (d *dentry) cacheNegativeLookupLocked(name string) { + // Don't cache negative lookups if InteropModeShared is in effect (since + // this makes remote lookup unavoidable), or if d.isSynthetic() (in which + // case the only files in the directory are those for which a dentry exists + // in d.children). Instead, just delete any previously-cached dentry. + if d.fs.opts.interop == InteropModeShared || d.isSynthetic() { + delete(d.children, name) + return + } if d.children == nil { d.children = make(map[string]*dentry) } d.children[name] = nil } +// createSyntheticDirectory creates a synthetic directory with the given name +// in d. +// +// Preconditions: d.dirMu must be locked. d.isDir(). d does not already contain +// a child with the given name. +func (d *dentry) createSyntheticDirectoryLocked(name string, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) { + d2 := &dentry{ + refs: 1, // held by d + fs: d.fs, + mode: uint32(mode) | linux.S_IFDIR, + uid: uint32(kuid), + gid: uint32(kgid), + blockSize: usermem.PageSize, // arbitrary + handle: handle{ + fd: -1, + }, + } + d2.pf.dentry = d2 + d2.vfsd.Init(d2) + + d.cacheNewChildLocked(d2, name) + d.syntheticChildren++ +} + type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl @@ -77,7 +110,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.dirents = ds } - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } @@ -108,10 +141,10 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { // filesystem.renameMu is needed for d.parent, and must be locked before // dentry.dirMu. d.fs.renameMu.RLock() + defer d.fs.renameMu.RUnlock() d.dirMu.Lock() defer d.dirMu.Unlock() if d.dirents != nil { - d.fs.renameMu.RUnlock() return d.dirents, nil } @@ -132,51 +165,81 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { NextOff: 2, }, } - d.fs.renameMu.RUnlock() - off := uint64(0) - const count = 64 * 1024 // for consistency with the vfs1 client - d.handleMu.RLock() - defer d.handleMu.RUnlock() - if !d.handleReadable { - // This should not be possible because a readable handle should have - // been opened when the calling directoryFD was opened. - panic("gofer.dentry.getDirents called without a readable handle") - } - for { - p9ds, err := d.handle.file.readdir(ctx, off, count) - if err != nil { - return nil, err + var realChildren map[string]struct{} + if !d.isSynthetic() { + if d.syntheticChildren != 0 && d.fs.opts.interop == InteropModeShared { + // Record the set of children d actually has so that we don't emit + // duplicate entries for synthetic children. + realChildren = make(map[string]struct{}) } - if len(p9ds) == 0 { - // Cache dirents for future directoryFDs if permitted. - if d.fs.opts.interop != InteropModeShared { - d.dirents = dirents + off := uint64(0) + const count = 64 * 1024 // for consistency with the vfs1 client + d.handleMu.RLock() + if !d.handleReadable { + // This should not be possible because a readable handle should + // have been opened when the calling directoryFD was opened. + d.handleMu.RUnlock() + panic("gofer.dentry.getDirents called without a readable handle") + } + for { + p9ds, err := d.handle.file.readdir(ctx, off, count) + if err != nil { + d.handleMu.RUnlock() + return nil, err + } + if len(p9ds) == 0 { + d.handleMu.RUnlock() + break + } + for _, p9d := range p9ds { + if p9d.Name == "." || p9d.Name == ".." { + continue + } + dirent := vfs.Dirent{ + Name: p9d.Name, + Ino: p9d.QID.Path, + NextOff: int64(len(dirents) + 1), + } + // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or + // DMSOCKET. + switch p9d.Type { + case p9.TypeSymlink: + dirent.Type = linux.DT_LNK + case p9.TypeDir: + dirent.Type = linux.DT_DIR + default: + dirent.Type = linux.DT_REG + } + dirents = append(dirents, dirent) + if realChildren != nil { + realChildren[p9d.Name] = struct{}{} + } } - return dirents, nil + off = p9ds[len(p9ds)-1].Offset } - for _, p9d := range p9ds { - if p9d.Name == "." || p9d.Name == ".." { + } + // Emit entries for synthetic children. + if d.syntheticChildren != 0 { + for _, child := range d.children { + if child == nil || !child.isSynthetic() { continue } - dirent := vfs.Dirent{ - Name: p9d.Name, - Ino: p9d.QID.Path, - NextOff: int64(len(dirents) + 1), - } - // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or - // DMSOCKET. - switch p9d.Type { - case p9.TypeSymlink: - dirent.Type = linux.DT_LNK - case p9.TypeDir: - dirent.Type = linux.DT_DIR - default: - dirent.Type = linux.DT_REG + if _, ok := realChildren[child.name]; ok { + continue } - dirents = append(dirents, dirent) + dirents = append(dirents, vfs.Dirent{ + Name: child.name, + Type: uint8(atomic.LoadUint32(&child.mode) >> 12), + Ino: child.ino, + NextOff: int64(len(dirents) + 1), + }) } - off = p9ds[len(p9ds)-1].Offset } + // Cache dirents for future directoryFDs if permitted. + if d.cachedMetadataAuthoritative() { + d.dirents = dirents + } + return dirents, nil } // Seek implements vfs.FileDescriptionImpl.Seek. diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 43e863c61..98ccb42fd 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -29,14 +29,16 @@ import ( // Sync implements vfs.FilesystemImpl.Sync. func (fs *filesystem) Sync(ctx context.Context) error { - // Snapshot current dentries and special files. + // Snapshot current syncable dentries and special files. fs.syncMu.Lock() - ds := make([]*dentry, 0, len(fs.dentries)) - for d := range fs.dentries { + ds := make([]*dentry, 0, len(fs.syncableDentries)) + for d := range fs.syncableDentries { + d.IncRef() ds = append(ds, d) } sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) for sffd := range fs.specialFileFDs { + sffd.vfsfd.IncRef() sffds = append(sffds, sffd) } fs.syncMu.Unlock() @@ -47,9 +49,6 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync regular files. for _, d := range ds { - if !d.TryIncRef() { - continue - } err := d.syncSharedHandle(ctx) d.DecRef() if err != nil && retErr == nil { @@ -60,9 +59,6 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync special files, which may be writable but do not use dentry shared // handles (so they won't be synced by the above). for _, sffd := range sffds { - if !sffd.vfsfd.TryIncRef() { - continue - } err := sffd.Sync(ctx) sffd.vfsfd.DecRef() if err != nil && retErr == nil { @@ -114,8 +110,8 @@ func putDentrySlice(ds *[]*dentry) { // to *ds. // // Preconditions: fs.renameMu must be locked. d.dirMu must be locked. -// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached -// metadata must be up to date. +// !rp.Done(). If !d.cachedMetadataAuthoritative(), then d's cached metadata +// must be up to date. // // Postconditions: The returned dentry's cached metadata is up to date. func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { @@ -148,7 +144,7 @@ afterSymlink: if err := rp.CheckMount(&d.parent.vfsd); err != nil { return nil, err } - if fs.opts.interop == InteropModeShared && d != d.parent { + if d != d.parent && !d.cachedMetadataAuthoritative() { _, attrMask, attr, err := d.parent.file.getAttr(ctx, dentryAttrMask()) if err != nil { return nil, err @@ -195,7 +191,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil return nil, syserror.ENAMETOOLONG } child, ok := parent.children[name] - if ok && fs.opts.interop != InteropModeShared { + if (ok && fs.opts.interop != InteropModeShared) || parent.isSynthetic() { // Whether child is nil or not, it is cached information that is // assumed to be correct. return child, nil @@ -206,7 +202,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil return fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, ds) } -// Preconditions: As for getChildLocked. +// Preconditions: As for getChildLocked. !parent.isSynthetic(). func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, child *dentry, ds **[]*dentry) (*dentry, error) { qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) if err != nil && err != syserror.ENOENT { @@ -220,24 +216,41 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir child.updateFromP9Attrs(attrMask, &attr) return child, nil } - // The file at this path has changed or no longer exists. Remove - // the stale dentry from the tree, and re-evaluate its caching - // status (i.e. if it has 0 references, drop it). + if file.isNil() && child.isSynthetic() { + // We have a synthetic file, and no remote file has arisen to + // replace it. + return child, nil + } + // The file at this path has changed or no longer exists. Mark the + // dentry invalidated, and re-evaluate its caching status (i.e. if it + // has 0 references, drop it). Wait to update parent.children until we + // know what to replace the existing dentry with (i.e. one of the + // returns below), to avoid a redundant map access. vfsObj.InvalidateDentry(&child.vfsd) + if child.isSynthetic() { + // Normally we don't mark invalidated dentries as deleted since + // they may still exist (but at a different path), and also for + // consistency with Linux. However, synthetic files are guaranteed + // to become unreachable if their dentries are invalidated, so + // treat their invalidation as deletion. + child.setDeleted() + parent.syntheticChildren-- + child.decRefLocked() + parent.dirents = nil + } *ds = appendDentry(*ds, child) } if file.isNil() { // No file exists at this path now. Cache the negative lookup if // allowed. - if fs.opts.interop != InteropModeShared { - parent.cacheNegativeChildLocked(name) - } + parent.cacheNegativeLookupLocked(name) return nil, nil } // Create a new dentry representing the file. child, err = fs.newDentry(ctx, file, qid, attrMask, &attr) if err != nil { file.close(ctx) + delete(parent.children, name) return nil, err } parent.cacheNewChildLocked(child, name) @@ -252,8 +265,9 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // rp.Start().Impl().(*dentry)). It does not check that the returned directory // is searchable by the provider of rp. // -// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == -// InteropModeShared, then d's cached metadata must be up to date. +// Preconditions: fs.renameMu must be locked. !rp.Done(). If +// !d.cachedMetadataAuthoritative(), then d's cached metadata must be up to +// date. func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { for !rp.Final() { d.dirMu.Lock() @@ -275,7 +289,7 @@ func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // Preconditions: fs.renameMu must be locked. func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { d := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !d.cachedMetadataAuthoritative() { // Get updated metadata for rp.Start() as required by fs.stepLocked(). if err := d.updateFromGetattr(ctx); err != nil { return nil, err @@ -297,16 +311,17 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, } // doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. +// createInRemoteDir (if the parent directory is a real remote directory) or +// createInSyntheticDir (if the parent directory is synthetic) to do so. // // Preconditions: !rp.Done(). For the final path component in rp, // !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, createInRemoteDir func(parent *dentry, name string) error, createInSyntheticDir func(parent *dentry, name string) error) error { var ds *[]*dentry fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -340,6 +355,20 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir defer mnt.EndWrite() parent.dirMu.Lock() defer parent.dirMu.Unlock() + if parent.isSynthetic() { + if child := parent.children[name]; child != nil { + return syserror.EEXIST + } + if createInSyntheticDir == nil { + return syserror.EPERM + } + if err := createInSyntheticDir(parent, name); err != nil { + return err + } + parent.touchCMtime() + parent.dirents = nil + return nil + } if fs.opts.interop == InteropModeShared { // The existence of a dentry at name would be inconclusive because the // file it represents may have been deleted from the remote filesystem, @@ -348,21 +377,21 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir // will fail with EEXIST like we would have. If the RPC succeeds, and a // stale dentry exists, the dentry will fail revalidation next time // it's used. - return create(parent, name) + return createInRemoteDir(parent, name) } if child := parent.children[name]; child != nil { return syserror.EEXIST } // No cached dentry exists; however, there might still be an existing file // at name. As above, we attempt the file creation RPC anyway. - if err := create(parent, name); err != nil { + if err := createInRemoteDir(parent, name); err != nil { return err } + if child, ok := parent.children[name]; ok && child == nil { + // Delete the now-stale negative dentry. + delete(parent.children, name) + } parent.touchCMtime() - // Either parent.children[name] doesn't exist (in which case this is a - // no-op) or is nil (in which case this erases the now-stale information - // that the file doesn't exist). - delete(parent.children, name) parent.dirents = nil return nil } @@ -373,7 +402,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -421,8 +450,10 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // only revalidating the dentry if that fails (indicating that the existing // dentry is a mount point). if child != nil { + child.dirMu.Lock() + defer child.dirMu.Unlock() if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { - if fs.opts.interop != InteropModeShared { + if parent.cachedMetadataAuthoritative() { return err } child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, &ds) @@ -437,13 +468,37 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } } flags := uint32(0) + // If a dentry exists, use it for best-effort checks on its deletability. if dir { - if child != nil && !child.isDir() { - vfsObj.AbortDeleteDentry(&child.vfsd) - return syserror.ENOTDIR + if child != nil { + // child must be an empty directory. + if child.syntheticChildren != 0 { + // This is definitely not an empty directory, irrespective of + // fs.opts.interop. + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTEMPTY + } + // If InteropModeShared is in effect and the first call to + // PrepareDeleteDentry above succeeded, then child wasn't + // revalidated (so we can't expect its file type to be correct) and + // individually revalidating its children (to confirm that they + // still exist) would be a waste of time. + if child.cachedMetadataAuthoritative() { + if !child.isDir() { + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTDIR + } + for _, grandchild := range child.children { + if grandchild != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTEMPTY + } + } + } } flags = linux.AT_REMOVEDIR } else { + // child must be a non-directory file. if child != nil && child.isDir() { vfsObj.AbortDeleteDentry(&child.vfsd) return syserror.EISDIR @@ -455,28 +510,36 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b return syserror.ENOTDIR } } - err = parent.file.unlinkAt(ctx, name, flags) - if err != nil { - if child != nil { - vfsObj.AbortDeleteDentry(&child.vfsd) - } - return err - } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime() - if dir { - parent.decLinks() + if parent.isSynthetic() { + if child == nil { + return syserror.ENOENT } - parent.cacheNegativeChildLocked(name) - parent.dirents = nil } else { - delete(parent.children, name) + err = parent.file.unlinkAt(ctx, name, flags) + if err != nil { + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + } + return err + } } if child != nil { - child.setDeleted() vfsObj.CommitDeleteDentry(&child.vfsd) + child.setDeleted() + if child.isSynthetic() { + parent.syntheticChildren-- + child.decRefLocked() + } ds = appendDentry(ds, child) } + parent.cacheNegativeLookupLocked(name) + if parent.cachedMetadataAuthoritative() { + parent.dirents = nil + parent.touchCMtime() + if dir { + parent.decLinks() + } + } return nil } @@ -554,7 +617,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -577,20 +640,32 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } // 9P2000.L supports hard links, but we don't. return syserror.EPERM - }) + }, nil) } // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + creds := rp.Credentials() return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() if _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)); err != nil { - return err + if !opts.ForSyntheticMountpoint || err == syserror.EEXIST { + return err + } + ctx.Infof("Failed to create remote directory %q: %v; falling back to synthetic directory", name, err) + parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) } if fs.opts.interop != InteropModeShared { parent.incLinks() } return nil + }, func(parent *dentry, name string) error { + if !opts.ForSyntheticMountpoint { + // Can't create non-synthetic files in synthetic directories. + return syserror.EPERM + } + parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) + parent.incLinks() + return nil }) } @@ -600,7 +675,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v creds := rp.Credentials() _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) return err - }) + }, nil) } // OpenAt implements vfs.FilesystemImpl.OpenAt. @@ -620,7 +695,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by fs.stepLocked(). if err := start.updateFromGetattr(ctx); err != nil { return nil, err @@ -643,6 +718,10 @@ afterTrailingSymlink: parent.dirMu.Lock() child, err := fs.stepLocked(ctx, rp, parent, &ds) if err == syserror.ENOENT && mayCreate { + if parent.isSynthetic() { + parent.dirMu.Unlock() + return nil, syserror.EPERM + } fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) parent.dirMu.Unlock() return fd, err @@ -702,8 +781,10 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { - return nil, err + if !d.isSynthetic() { + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { + return nil, err + } } fd := &directoryFD{} if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { @@ -733,6 +814,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf } // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. +// !d.isSynthetic(). func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err @@ -811,7 +893,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving child.refs = 1 // Insert the dentry into the tree. d.cacheNewChildLocked(child, name) - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { d.touchCMtime() d.dirents = nil } @@ -888,7 +970,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa defer mnt.EndWrite() oldParent := oldParentVD.Dentry().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !oldParent.cachedMetadataAuthoritative() { if err := oldParent.updateFromGetattr(ctx); err != nil { return err } @@ -933,35 +1015,22 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if newParent.isDeleted() { return syserror.ENOENT } - replaced := newParent.children[newName] - // This is similar to unlinkAt, except: - // - // - If a dentry exists for the file to be replaced, we revalidate it - // unconditionally (instead of only if PrepareRenameDentry fails) for - // simplicity. - // - // - If rp.MustBeDir(), then we need a dentry representing the replaced - // file regardless to confirm that it's a directory. - if replaced != nil || rp.MustBeDir() { - replaced, err = fs.getChildLocked(ctx, rp.VirtualFilesystem(), newParent, newName, &ds) - if err != nil { - return err - } - if replaced != nil { - if replaced.isDir() { - if !renamed.isDir() { - return syserror.EISDIR - } - } else { - if rp.MustBeDir() || renamed.isDir() { - return syserror.ENOTDIR - } - } - } + replaced, err := fs.getChildLocked(ctx, rp.VirtualFilesystem(), newParent, newName, &ds) + if err != nil { + return err } var replacedVFSD *vfs.Dentry if replaced != nil { replacedVFSD = &replaced.vfsd + if replaced.isDir() { + if !renamed.isDir() { + return syserror.EISDIR + } + } else { + if rp.MustBeDir() || renamed.isDir() { + return syserror.ENOTDIR + } + } } if oldParent == newParent && oldName == newName { @@ -972,27 +1041,47 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } - if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { - vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) - return err + + // Update the remote filesystem. + if !renamed.isSynthetic() { + if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } + } else if replaced != nil && !replaced.isSynthetic() { + // We are replacing an existing real file with a synthetic one, so we + // need to unlink the former. + flags := uint32(0) + if replaced.isDir() { + flags = linux.AT_REMOVEDIR + } + if err := newParent.file.unlinkAt(ctx, newName, flags); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } } - if fs.opts.interop != InteropModeShared { - oldParent.cacheNegativeChildLocked(oldName) - oldParent.dirents = nil - newParent.dirents = nil - if renamed.isDir() { - oldParent.decLinks() - newParent.incLinks() + + // Update the dentry tree. + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + if replaced != nil { + replaced.setDeleted() + if replaced.isSynthetic() { + newParent.syntheticChildren-- + replaced.decRefLocked() } - oldParent.touchCMtime() - newParent.touchCMtime() - renamed.touchCtime() - } else { - delete(oldParent.children, oldName) + ds = appendDentry(ds, replaced) } + oldParent.cacheNegativeLookupLocked(oldName) + // We don't use newParent.cacheNewChildLocked() since we don't want to mess + // with reference counts and queue oldParent for checkCachingLocked if the + // parent isn't actually changing. if oldParent != newParent { - appendDentry(ds, oldParent) + ds = appendDentry(ds, oldParent) newParent.IncRef() + if renamed.isSynthetic() { + oldParent.syntheticChildren-- + newParent.syntheticChildren++ + } } renamed.parent = newParent renamed.name = newName @@ -1000,11 +1089,25 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa newParent.children = make(map[string]*dentry) } newParent.children[newName] = renamed - if replaced != nil { - replaced.setDeleted() - appendDentry(ds, replaced) + + // Update metadata. + if renamed.cachedMetadataAuthoritative() { + renamed.touchCtime() + } + if oldParent.cachedMetadataAuthoritative() { + oldParent.dirents = nil + oldParent.touchCMtime() + if renamed.isDir() { + oldParent.decLinks() + } + } + if newParent.cachedMetadataAuthoritative() { + newParent.dirents = nil + newParent.touchCMtime() + if renamed.isDir() { + newParent.incLinks() + } } - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) return nil } @@ -1051,6 +1154,10 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } + // If d is synthetic, invoke statfs on the first ancestor of d that isn't. + for d.isSynthetic() { + d = d.parent + } fsstat, err := d.file.statFS(ctx) if err != nil { return linux.Statfs{}, err @@ -1080,7 +1187,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ creds := rp.Credentials() _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) return err - }) + }, nil) } // UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. @@ -1089,9 +1196,15 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -// -// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + // TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. return nil, syserror.ECONNREFUSED } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 293df2545..8b4e91d17 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -27,8 +27,9 @@ // dentry.handleMu // dentry.dataMu // -// Locking dentry.dirMu in multiple dentries requires holding -// filesystem.renameMu for writing. +// Locking dentry.dirMu in multiple dentries requires that either ancestor +// dentries are locked before descendant dentries, or that filesystem.renameMu +// is locked for writing. package gofer import ( @@ -102,11 +103,12 @@ type filesystem struct { cachedDentries dentryList cachedDentriesLen uint64 - // dentries contains all dentries in this filesystem. specialFileFDs - // contains all open specialFileFDs. These fields are protected by syncMu. - syncMu sync.Mutex - dentries map[*dentry]struct{} - specialFileFDs map[*specialFileFD]struct{} + // syncableDentries contains all dentries in this filesystem for which + // !dentry.file.isNil(). specialFileFDs contains all open specialFileFDs. + // These fields are protected by syncMu. + syncMu sync.Mutex + syncableDentries map[*dentry]struct{} + specialFileFDs map[*specialFileFD]struct{} } type filesystemOptions struct { @@ -187,7 +189,8 @@ const ( // InteropModeShared is appropriate when there are users of the remote // filesystem that may mutate its state other than the client. // - // - The client must verify cached filesystem state before using it. + // - The client must verify ("revalidate") cached filesystem state before + // using it. // // - Client changes to filesystem state must be sent to the remote // filesystem synchronously. @@ -376,14 +379,14 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // Construct the filesystem object. fs := &filesystem{ - mfp: mfp, - opts: fsopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, - client: client, - clock: ktime.RealtimeClockFromContext(ctx), - dentries: make(map[*dentry]struct{}), - specialFileFDs: make(map[*specialFileFD]struct{}), + mfp: mfp, + opts: fsopts, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, + client: client, + clock: ktime.RealtimeClockFromContext(ctx), + syncableDentries: make(map[*dentry]struct{}), + specialFileFDs: make(map[*specialFileFD]struct{}), } fs.vfsfs.Init(vfsObj, &fstype, fs) @@ -409,7 +412,7 @@ func (fs *filesystem) Release() { mf := fs.mfp.MemoryFile() fs.syncMu.Lock() - for d := range fs.dentries { + for d := range fs.syncableDentries { d.handleMu.Lock() d.dataMu.Lock() if d.handleWritable { @@ -444,9 +447,11 @@ type dentry struct { vfsd vfs.Dentry // refs is the reference count. Each dentry holds a reference on its - // parent, even if disowned. refs is accessed using atomic memory - // operations. When refs reaches 0, the dentry may be added to the cache or - // destroyed. If refs==-1 the dentry has already been destroyed. + // parent, even if disowned. An additional reference is held on all + // synthetic dentries until they are unlinked or invalidated. When refs + // reaches 0, the dentry may be added to the cache or destroyed. If refs == + // -1, the dentry has already been destroyed. refs is accessed using atomic + // memory operations. refs int64 // fs is the owning filesystem. fs is immutable. @@ -465,6 +470,12 @@ type dentry struct { // We don't support hard links, so each dentry maps 1:1 to an inode. // file is the unopened p9.File that backs this dentry. file is immutable. + // + // If file.isNil(), this dentry represents a synthetic file, i.e. a file + // that does not exist on the remote filesystem. As of this writing, this + // is only possible for a directory created with + // MkdirOptions.ForSyntheticMountpoint == true. + // TODO(gvisor.dev/issue/1476): Support synthetic sockets (and pipes). file p9file // If deleted is non-zero, the file represented by this dentry has been @@ -484,15 +495,21 @@ type dentry struct { // - Mappings of child filenames to dentries representing those children. // // - Mappings of child filenames that are known not to exist to nil - // dentries (only if InteropModeShared is not in effect). + // dentries (only if InteropModeShared is not in effect and the directory + // is not synthetic). // // children is protected by dirMu. children map[string]*dentry - // If this dentry represents a directory, InteropModeShared is not in - // effect, and dirents is not nil, it is a cache of all entries in the - // directory, in the order they were returned by the server. dirents is - // protected by dirMu. + // If this dentry represents a directory, syntheticChildren is the number + // of child dentries for which dentry.isSynthetic() == true. + // syntheticChildren is protected by dirMu. + syntheticChildren int + + // If this dentry represents a directory, + // dentry.cachedMetadataAuthoritative() == true, and dirents is not nil, it + // is a cache of all entries in the directory, in the order they were + // returned by the server. dirents is protected by dirMu. dirents []vfs.Dirent // Cached metadata; protected by metadataMu and accessed using atomic @@ -589,6 +606,8 @@ func dentryAttrMask() p9.AttrMask { // initially has no references, but is not cached; it is the caller's // responsibility to set the dentry's reference count and/or call // dentry.checkCachingLocked() as appropriate. +// +// Preconditions: !file.isNil(). func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { if !mask.Mode { ctx.Warningf("can't create gofer.dentry without file type") @@ -612,10 +631,10 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma }, } d.pf.dentry = d - if mask.UID { + if mask.UID && attr.UID != auth.NoID { d.uid = uint32(attr.UID) } - if mask.GID { + if mask.GID && attr.GID != auth.NoID { d.gid = uint32(attr.GID) } if mask.Size { @@ -642,11 +661,19 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma d.vfsd.Init(d) fs.syncMu.Lock() - fs.dentries[d] = struct{}{} + fs.syncableDentries[d] = struct{}{} fs.syncMu.Unlock() return d, nil } +func (d *dentry) isSynthetic() bool { + return d.file.isNil() +} + +func (d *dentry) cachedMetadataAuthoritative() bool { + return d.fs.opts.interop != InteropModeShared || d.isSynthetic() +} + // updateFromP9Attrs is called to update d's metadata after an update from the // remote filesystem. func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { @@ -691,6 +718,7 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { d.metadataMu.Unlock() } +// Preconditions: !d.isSynthetic() func (d *dentry) updateFromGetattr(ctx context.Context) error { // Use d.handle.file, which represents a 9P fid that has been opened, in // preference to d.file, which represents a 9P fid that has not. This may @@ -758,7 +786,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin defer mnt.EndWrite() setLocalAtime := false setLocalMtime := false - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { // Timestamp updates will be handled locally. setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 @@ -771,35 +799,37 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } d.metadataMu.Lock() defer d.metadataMu.Unlock() - if stat.Mask != 0 { - if err := d.file.setAttr(ctx, p9.SetAttrMask{ - Permissions: stat.Mask&linux.STATX_MODE != 0, - UID: stat.Mask&linux.STATX_UID != 0, - GID: stat.Mask&linux.STATX_GID != 0, - Size: stat.Mask&linux.STATX_SIZE != 0, - ATime: stat.Mask&linux.STATX_ATIME != 0, - MTime: stat.Mask&linux.STATX_MTIME != 0, - ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, - MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, - }, p9.SetAttr{ - Permissions: p9.FileMode(stat.Mode), - UID: p9.UID(stat.UID), - GID: p9.GID(stat.GID), - Size: stat.Size, - ATimeSeconds: uint64(stat.Atime.Sec), - ATimeNanoSeconds: uint64(stat.Atime.Nsec), - MTimeSeconds: uint64(stat.Mtime.Sec), - MTimeNanoSeconds: uint64(stat.Mtime.Nsec), - }); err != nil { - return err + if !d.isSynthetic() { + if stat.Mask != 0 { + if err := d.file.setAttr(ctx, p9.SetAttrMask{ + Permissions: stat.Mask&linux.STATX_MODE != 0, + UID: stat.Mask&linux.STATX_UID != 0, + GID: stat.Mask&linux.STATX_GID != 0, + Size: stat.Mask&linux.STATX_SIZE != 0, + ATime: stat.Mask&linux.STATX_ATIME != 0, + MTime: stat.Mask&linux.STATX_MTIME != 0, + ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, + MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, + }, p9.SetAttr{ + Permissions: p9.FileMode(stat.Mode), + UID: p9.UID(stat.UID), + GID: p9.GID(stat.GID), + Size: stat.Size, + ATimeSeconds: uint64(stat.Atime.Sec), + ATimeNanoSeconds: uint64(stat.Atime.Nsec), + MTimeSeconds: uint64(stat.Mtime.Sec), + MTimeNanoSeconds: uint64(stat.Mtime.Nsec), + }); err != nil { + return err + } + } + if d.fs.opts.interop == InteropModeShared { + // There's no point to updating d's metadata in this case since + // it'll be overwritten by revalidation before the next time it's + // used anyway. (InteropModeShared inhibits client caching of + // regular file data, so there's no cache to truncate either.) + return nil } - } - if d.fs.opts.interop == InteropModeShared { - // There's no point to updating d's metadata in this case since it'll - // be overwritten by revalidation before the next time it's used - // anyway. (InteropModeShared inhibits client caching of regular file - // data, so there's no cache to truncate either.) - return nil } now := d.fs.clock.Now().Nanoseconds() if stat.Mask&linux.STATX_MODE != 0 { @@ -897,6 +927,15 @@ func (d *dentry) DecRef() { } } +// decRefLocked decrements d's reference count without calling +// d.checkCachingLocked, even if d's reference count reaches 0; callers are +// responsible for ensuring that d.checkCachingLocked will be called later. +func (d *dentry) decRefLocked() { + if refs := atomic.AddInt64(&d.refs, -1); refs < 0 { + panic("gofer.dentry.decRefLocked() called without holding a reference") + } +} + // checkCachingLocked should be called after d's reference count becomes 0 or it // becomes disowned. // @@ -1013,11 +1052,11 @@ func (d *dentry) destroyLocked() { if !d.file.isNil() { d.file.close(ctx) d.file = p9file{} + // Remove d from the set of syncable dentries. + d.fs.syncMu.Lock() + delete(d.fs.syncableDentries, d) + d.fs.syncMu.Unlock() } - // Remove d from the set of all dentries. - d.fs.syncMu.Lock() - delete(d.fs.dentries, d) - d.fs.syncMu.Unlock() // Drop the reference held by d on its parent without recursively locking // d.fs.renameMu. if d.parent != nil { @@ -1040,6 +1079,9 @@ func (d *dentry) setDeleted() { // We only support xattrs prefixed with "user." (see b/148380782). Currently, // there is no need to expose any other xattrs through a gofer. func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) { + if d.file.isNil() { + return nil, nil + } xattrMap, err := d.file.listXattr(ctx, size) if err != nil { return nil, err @@ -1054,6 +1096,9 @@ func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size ui } func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) { + if d.file.isNil() { + return "", syserror.ENODATA + } if err := d.checkPermissions(creds, vfs.MayRead); err != nil { return "", err } @@ -1064,6 +1109,9 @@ func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vf } func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.SetxattrOptions) error { + if d.file.isNil() { + return syserror.EPERM + } if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { return err } @@ -1074,6 +1122,9 @@ func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vf } func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name string) error { + if d.file.isNil() { + return syserror.EPERM + } if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { return err } @@ -1083,7 +1134,7 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name return d.file.removeXattr(ctx, name) } -// Preconditions: d.isRegularFile() || d.isDirectory(). +// Preconditions: !d.file.isNil(). d.isRegularFile() || d.isDirectory(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with // O_TRUNC). @@ -1213,7 +1264,7 @@ func (fd *fileDescription) dentry() *dentry { func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { d := fd.dentry() const validMask = uint32(linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME) - if d.fs.opts.interop == InteropModeShared && opts.Mask&(validMask) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { + if !d.cachedMetadataAuthoritative() && opts.Mask&validMask != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if // available? if err := d.updateFromGetattr(ctx); err != nil { diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index 4041fb252..adff39490 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -24,7 +24,7 @@ import ( func TestDestroyIdempotent(t *testing.T) { fs := filesystem{ - dentries: make(map[*dentry]struct{}), + syncableDentries: make(map[*dentry]struct{}), opts: filesystemOptions{ // Test relies on no dentry being held in the cache. maxCachedDentries: 0, diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index cfe66f797..724a3f1f7 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/hostfd" ) // handle represents a remote "open file descriptor", consisting of an opened @@ -77,7 +78,7 @@ func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offs } if h.fd >= 0 { ctx.UninterruptibleSleepStart(false) - n, err := hostPreadv(h.fd, dsts, int64(offset)) + n, err := hostfd.Preadv2(h.fd, dsts, int64(offset), 0 /* flags */) ctx.UninterruptibleSleepFinish(false) return n, err } @@ -103,7 +104,7 @@ func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, o } if h.fd >= 0 { ctx.UninterruptibleSleepStart(false) - n, err := hostPwritev(h.fd, srcs, int64(offset)) + n, err := hostfd.Pwritev2(h.fd, srcs, int64(offset), 0 /* flags */) ctx.UninterruptibleSleepFinish(false) return n, err } diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go deleted file mode 100644 index 19560ab26..000000000 --- a/pkg/sentry/fsimpl/gofer/handle_unsafe.go +++ /dev/null @@ -1,66 +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 gofer - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/safemem" -) - -// Preconditions: !dsts.IsEmpty(). -func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if dsts.NumBlocks() == 1 { - // Use pread() instead of preadv() to avoid iovec allocation and - // copying. - dst := dsts.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(dsts) - n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} - -// Preconditions: !srcs.IsEmpty(). -func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if srcs.NumBlocks() == 1 { - // Use pwrite() instead of pwritev() to avoid iovec allocation and - // copying. - src := srcs.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(srcs) - n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 82e1fb74b..44dd9f672 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -15,12 +15,11 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", - "//pkg/fd", "//pkg/log", "//pkg/refs", - "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/hostfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index fe14476f1..ae94cfa6e 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -25,11 +25,10 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -492,19 +491,9 @@ func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, off if flags != 0 { return 0, syserror.EOPNOTSUPP } - - var reader safemem.Reader - if offset == -1 { - reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} - } else { - reader = safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv(hostFD, srcs, offset) - return int64(n), err - }, - } - } + reader := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) n, err := dst.CopyOutFrom(ctx, reader) + hostfd.PutReadWriterAt(reader) return int64(n), err } @@ -542,19 +531,9 @@ func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offs if flags != 0 { return 0, syserror.EOPNOTSUPP } - - var writer safemem.Writer - if offset == -1 { - writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} - } else { - writer = safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev(hostFD, srcs, offset) - return int64(n), err - }, - } - } + writer := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) n, err := src.CopyInTo(ctx, writer) + hostfd.PutReadWriterAt(writer) return int64(n), err } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 01c23d192..1d46dba25 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -246,8 +246,8 @@ func (fs *Filesystem) Sync(ctx context.Context) error { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() - defer fs.mu.RUnlock() defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() _, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { @@ -391,7 +391,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // 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 + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK ats := vfs.AccessTypesForOpenFlags(&opts) // Do not create new file. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 9f526359e..a946645f6 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -216,6 +216,11 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo atomic.StoreUint32(&a.nlink, nlink) } +// Ino returns the inode id. +func (a *InodeAttrs) Ino() uint64 { + return atomic.LoadUint64(&a.ino) +} + // Mode implements Inode.Mode. func (a *InodeAttrs) Mode() linux.FileMode { return linux.FileMode(atomic.LoadUint32(&a.mode)) @@ -359,8 +364,8 @@ func (o *OrderedChildren) Destroy() { // 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. +// Precondition: 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 { diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 2c6f8bdfc..f3173e197 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -111,17 +111,18 @@ func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { } defer m.DecUsers(ctx) - // Space for buffer with AT_NULL (0) terminator at the end. auxv := m.Auxv() + // Space for buffer with AT_NULL (0) terminator at the end. buf.Grow((len(auxv) + 1) * 16) for _, e := range auxv { - var tmp [8]byte - usermem.ByteOrder.PutUint64(tmp[:], e.Key) - buf.Write(tmp[:]) - - usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) + var tmp [16]byte + usermem.ByteOrder.PutUint64(tmp[:8], e.Key) + usermem.ByteOrder.PutUint64(tmp[8:], uint64(e.Value)) buf.Write(tmp[:]) } + var atNull [16]byte + buf.Write(atNull[:]) + return nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 3d5dc463c..f08668ca2 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -39,7 +39,7 @@ func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), }), "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), + "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{k: k}), "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), }), "net": newSysNetDir(root, inoGen, k), diff --git a/pkg/sentry/hostfd/BUILD b/pkg/sentry/hostfd/BUILD new file mode 100644 index 000000000..364a78306 --- /dev/null +++ b/pkg/sentry/hostfd/BUILD @@ -0,0 +1,17 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "hostfd", + srcs = [ + "hostfd.go", + "hostfd_unsafe.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/safemem", + "//pkg/sync", + "@org_golang_x_sys//unix:go_default_library", + ], +) diff --git a/pkg/sentry/hostfd/hostfd.go b/pkg/sentry/hostfd/hostfd.go new file mode 100644 index 000000000..70dd9cafb --- /dev/null +++ b/pkg/sentry/hostfd/hostfd.go @@ -0,0 +1,84 @@ +// Copyright 2020 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 hostfd provides efficient I/O with host file descriptors. +package hostfd + +import ( + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sync" +) + +// ReadWriterAt implements safemem.Reader and safemem.Writer by reading from +// and writing to a host file descriptor respectively. ReadWriterAts should be +// obtained by calling GetReadWriterAt. +// +// Clients should usually prefer to use Preadv2 and Pwritev2 directly. +type ReadWriterAt struct { + fd int32 + offset int64 + flags uint32 +} + +var rwpool = sync.Pool{ + New: func() interface{} { + return &ReadWriterAt{} + }, +} + +// GetReadWriterAt returns a ReadWriterAt that reads from / writes to the given +// host file descriptor, starting at the given offset and using the given +// preadv2(2)/pwritev2(2) flags. If offset is -1, the host file descriptor's +// offset is used instead. Users are responsible for ensuring that fd remains +// valid for the lifetime of the returned ReadWriterAt, and must call +// PutReadWriterAt when it is no longer needed. +func GetReadWriterAt(fd int32, offset int64, flags uint32) *ReadWriterAt { + rw := rwpool.Get().(*ReadWriterAt) + *rw = ReadWriterAt{ + fd: fd, + offset: offset, + flags: flags, + } + return rw +} + +// PutReadWriterAt releases a ReadWriterAt returned by a previous call to +// GetReadWriterAt that is no longer in use. +func PutReadWriterAt(rw *ReadWriterAt) { + rwpool.Put(rw) +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +func (rw *ReadWriterAt) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + n, err := Preadv2(rw.fd, dsts, rw.offset, rw.flags) + if rw.offset >= 0 { + rw.offset += int64(n) + } + return n, err +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +func (rw *ReadWriterAt) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + n, err := Pwritev2(rw.fd, srcs, rw.offset, rw.flags) + if rw.offset >= 0 { + rw.offset += int64(n) + } + return n, err +} diff --git a/pkg/sentry/hostfd/hostfd_unsafe.go b/pkg/sentry/hostfd/hostfd_unsafe.go new file mode 100644 index 000000000..5e9e60fc4 --- /dev/null +++ b/pkg/sentry/hostfd/hostfd_unsafe.go @@ -0,0 +1,107 @@ +// Copyright 2020 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 hostfd + +import ( + "io" + "syscall" + "unsafe" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/safemem" +) + +// Preadv2 reads up to dsts.NumBytes() bytes from host file descriptor fd into +// dsts. offset and flags are interpreted as for preadv2(2). +// +// Preconditions: !dsts.IsEmpty(). +func Preadv2(fd int32, dsts safemem.BlockSeq, offset int64, flags uint32) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + var ( + n uintptr + e syscall.Errno + ) + // Avoid preadv2(2) if possible, since it's relatively new and thus least + // likely to be supported by the host kernel. + if flags == 0 { + if dsts.NumBlocks() == 1 { + // Use read() or pread() to avoid iovec allocation and copying. + dst := dsts.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_READ, uintptr(fd), dst.Addr(), uintptr(dst.Len())) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } else { + iovs := safemem.IovecsFromBlockSeq(dsts) + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_READV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } + } else { + iovs := safemem.IovecsFromBlockSeq(dsts) + n, _, e = syscall.Syscall6(unix.SYS_PREADV2, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, uintptr(flags)) + } + if e != 0 { + return 0, e + } + if n == 0 { + return 0, io.EOF + } + return uint64(n), nil +} + +// Pwritev2 writes up to srcs.NumBytes() from srcs into host file descriptor +// fd. offset and flags are interpreted as for pwritev2(2). +// +// Preconditions: !srcs.IsEmpty(). +func Pwritev2(fd int32, srcs safemem.BlockSeq, offset int64, flags uint32) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + var ( + n uintptr + e syscall.Errno + ) + // Avoid pwritev2(2) if possible, since it's relatively new and thus least + // likely to be supported by the host kernel. + if flags == 0 { + if srcs.NumBlocks() == 1 { + // Use write() or pwrite() to avoid iovec allocation and copying. + src := srcs.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_WRITE, uintptr(fd), src.Addr(), uintptr(src.Len())) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } else { + iovs := safemem.IovecsFromBlockSeq(srcs) + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_WRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } + } else { + iovs := safemem.IovecsFromBlockSeq(srcs) + n, _, e = syscall.Syscall6(unix.SYS_PWRITEV2, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, uintptr(flags)) + } + if e != 0 { + return 0, e + } + return uint64(n), nil +} diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD index dedf0fa15..75eedd5a2 100644 --- a/pkg/sentry/kernel/epoll/BUILD +++ b/pkg/sentry/kernel/epoll/BUILD @@ -24,6 +24,7 @@ go_library( ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/context", "//pkg/refs", "//pkg/sentry/fs", diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 592650923..3d78cd48f 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -20,6 +20,7 @@ import ( "fmt" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -30,19 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// Event describes the event mask that was observed and the user data to be -// returned when one of the events occurs. It has this format to match the linux -// format to avoid extra copying/allocation when writing events to userspace. -type Event struct { - // Events is the event mask containing the set of events that have been - // observed on an entry. - Events uint32 - - // Data is an opaque 64-bit value provided by the caller when adding the - // entry, and returned to the caller when the entry reports an event. - Data [2]int32 -} - // EntryFlags is a bitmask that holds an entry's flags. type EntryFlags int @@ -227,9 +215,9 @@ func (e *EventPoll) Readiness(mask waiter.EventMask) waiter.EventMask { } // ReadEvents returns up to max available events. -func (e *EventPoll) ReadEvents(max int) []Event { +func (e *EventPoll) ReadEvents(max int) []linux.EpollEvent { var local pollEntryList - var ret []Event + var ret []linux.EpollEvent e.listsMu.Lock() @@ -251,7 +239,7 @@ func (e *EventPoll) ReadEvents(max int) []Event { } // Add event to the array that will be returned to caller. - ret = append(ret, Event{ + ret = append(ret, linux.EpollEvent{ Events: uint32(ready), Data: entry.userData, }) diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index e5d133d6c..f48247c94 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -484,7 +484,7 @@ type Task struct { // bit. // // numaPolicy and numaNodeMask are protected by mu. - numaPolicy int32 + numaPolicy linux.NumaPolicy numaNodeMask uint64 // netns is the task's network namespace. netns is never nil. diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 2ba8d7e63..d654dd997 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -96,6 +96,7 @@ func (t *Task) run(threadID uintptr) { t.tg.liveGoroutines.Done() t.tg.pidns.owner.liveGoroutines.Done() t.tg.pidns.owner.runningGoroutines.Done() + t.p.Release() // Keep argument alive because stack trace for dead variables may not be correct. runtime.KeepAlive(threadID) diff --git a/pkg/sentry/kernel/task_sched.go b/pkg/sentry/kernel/task_sched.go index 8b148db35..09366b60c 100644 --- a/pkg/sentry/kernel/task_sched.go +++ b/pkg/sentry/kernel/task_sched.go @@ -653,14 +653,14 @@ func (t *Task) SetNiceness(n int) { } // NumaPolicy returns t's current numa policy. -func (t *Task) NumaPolicy() (policy int32, nodeMask uint64) { +func (t *Task) NumaPolicy() (policy linux.NumaPolicy, nodeMask uint64) { t.mu.Lock() defer t.mu.Unlock() return t.numaPolicy, t.numaNodeMask } // SetNumaPolicy sets t's numa policy. -func (t *Task) SetNumaPolicy(policy int32, nodeMask uint64) { +func (t *Task) SetNumaPolicy(policy linux.NumaPolicy, nodeMask uint64) { t.mu.Lock() defer t.mu.Unlock() t.numaPolicy = policy diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index f07de2089..7d25e98f7 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -263,6 +263,19 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) if t.haveSavedSignalMask { mask = t.savedSignalMask } + + // Set up the restorer. + // x86-64 should always uses SA_RESTORER, but this flag is optional on other platforms. + // Please see the linux code as reference: + // linux/arch/x86/kernel/signal.c:__setup_rt_frame() + // If SA_RESTORER is not configured, we can use the sigreturn trampolines + // the vdso provides instead. + // Please see the linux code as reference: + // linux/arch/arm64/kernel/signal.c:setup_return() + if act.Flags&linux.SA_RESTORER == 0 { + act.Restorer = t.MemoryManager().VDSOSigReturn() + } + if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index d6675b8f0..88449fe95 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -311,6 +311,15 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V m.SetAuxv(auxv) m.SetExecutable(file) + symbolValue, err := getSymbolValueFromVDSO("rt_sigreturn") + if err != nil { + return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to find rt_sigreturn in vdso: %v", err), syserr.FromError(err).ToLinux()) + } + + // Found rt_sigretrun. + addr := uint64(vdsoAddr) + symbolValue - vdsoPrelink + m.SetVDSOSigReturn(addr) + ac.SetIP(uintptr(loaded.entry)) ac.SetStack(uintptr(stack.Bottom)) diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index 161b28c2c..00977fc08 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -15,9 +15,11 @@ package loader import ( + "bytes" "debug/elf" "fmt" "io" + "strings" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/context" @@ -38,6 +40,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +const vdsoPrelink = 0xffffffffff700000 + type fileContext struct { context.Context } @@ -221,6 +225,27 @@ type VDSO struct { phdrs []elf.ProgHeader `state:".([]elfProgHeader)"` } +// getSymbolValueFromVDSO returns the specific symbol value in vdso.so. +func getSymbolValueFromVDSO(symbol string) (uint64, error) { + f, err := elf.NewFile(bytes.NewReader(vdsoBin)) + if err != nil { + return 0, err + } + syms, err := f.Symbols() + if err != nil { + return 0, err + } + + for _, sym := range syms { + if elf.ST_BIND(sym.Info) != elf.STB_LOCAL && sym.Section != elf.SHN_UNDEF { + if strings.Contains(sym.Name, symbol) { + return sym.Value, nil + } + } + } + return 0, fmt.Errorf("no %v in vdso.so", symbol) +} + // PrepareVDSO validates the system VDSO and returns a VDSO, containing the // param page for updating by the kernel. func PrepareVDSO(ctx context.Context, mfp pgalloc.MemoryFileProvider) (*VDSO, error) { diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 34d3bde7a..6db7c3d40 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -35,6 +35,7 @@ package mm import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -286,7 +287,7 @@ type vma struct { mlockMode memmap.MLockMode // numaPolicy is the NUMA policy for this vma set by mbind(). - numaPolicy int32 + numaPolicy linux.NumaPolicy // numaNodemask is the NUMA nodemask for this vma set by mbind(). numaNodemask uint64 diff --git a/pkg/sentry/mm/procfs.go b/pkg/sentry/mm/procfs.go index 1ab92f046..6efe5102b 100644 --- a/pkg/sentry/mm/procfs.go +++ b/pkg/sentry/mm/procfs.go @@ -148,7 +148,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI // Do not include the guard page: fs/proc/task_mmu.c:show_map_vma() => // stack_guard_page_start(). - fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", + lineLen, _ := fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", vseg.Start(), vseg.End(), vma.realPerms, private, vma.off, devMajor, devMinor, ino) // Figure out our filename or hint. @@ -165,7 +165,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI } if s != "" { // Per linux, we pad until the 74th character. - if pad := 73 - b.Len(); pad > 0 { + if pad := 73 - lineLen; pad > 0 { b.WriteString(strings.Repeat(" ", pad)) } b.WriteString(s) diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index c5dfa5972..3f496aa9f 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -974,7 +974,7 @@ func (mm *MemoryManager) MLockAll(ctx context.Context, opts MLockAllOpts) error } // NumaPolicy implements the semantics of Linux's get_mempolicy(MPOL_F_ADDR). -func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { +func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (linux.NumaPolicy, uint64, error) { mm.mappingMu.RLock() defer mm.mappingMu.RUnlock() vseg := mm.vmas.FindSegment(addr) @@ -986,7 +986,7 @@ func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { } // SetNumaPolicy implements the semantics of Linux's mbind(). -func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy int32, nodemask uint64) error { +func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy linux.NumaPolicy, nodemask uint64) error { if !addr.IsPageAligned() { return syserror.EINVAL } diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index c769ac7b4..6507121ea 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -85,3 +85,6 @@ func (c *context) Switch(as platform.AddressSpace, ac arch.Context, _ int32) (*a func (c *context) Interrupt() { c.interrupt.NotifyInterrupt() } + +// Release implements platform.Context.Release(). +func (c *context) Release() {} diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index a9b4af43e..ae813e24e 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -191,6 +191,11 @@ func (*constructor) OpenDevice() (*os.File, error) { return OpenDevice() } +// Flags implements platform.Constructor.Flags(). +func (*constructor) Requirements() platform.Requirements { + return platform.Requirements{} +} + func init() { platform.Register("kvm", &constructor{}) } diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 2ca696382..171513f3f 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -148,6 +148,9 @@ type Context interface { // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. Interrupt() + + // Release() releases any resources associated with this context. + Release() } var ( @@ -353,10 +356,28 @@ func (fr FileRange) String() string { return fmt.Sprintf("[%#x, %#x)", fr.Start, fr.End) } +// Requirements is used to specify platform specific requirements. +type Requirements struct { + // RequiresCurrentPIDNS indicates that the sandbox has to be started in the + // current pid namespace. + RequiresCurrentPIDNS bool + // RequiresCapSysPtrace indicates that the sandbox has to be started with + // the CAP_SYS_PTRACE capability. + RequiresCapSysPtrace bool +} + // Constructor represents a platform type. type Constructor interface { + // New returns a new platform instance. + // + // Arguments: + // + // * deviceFile - the device file (e.g. /dev/kvm for the KVM platform). New(deviceFile *os.File) (Platform, error) OpenDevice() (*os.File, error) + + // Requirements returns platform specific requirements. + Requirements() Requirements } // platforms contains all available platform types. diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 03adb624b..08d055e05 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -177,6 +177,9 @@ func (c *context) Interrupt() { c.interrupt.NotifyInterrupt() } +// Release implements platform.Context.Release(). +func (c *context) Release() {} + // PTrace represents a collection of ptrace subprocesses. type PTrace struct { platform.MMapMinAddr @@ -248,6 +251,16 @@ func (*constructor) OpenDevice() (*os.File, error) { return nil, nil } +// Flags implements platform.Constructor.Flags(). +func (*constructor) Requirements() platform.Requirements { + // TODO(b/75837838): Also set a new PID namespace so that we limit + // access to other host processes. + return platform.Requirements{ + RequiresCapSysPtrace: true, + RequiresCurrentPIDNS: true, + } +} + func init() { platform.Register("ptrace", &constructor{}) } diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index a644609ef..773ddb1ed 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -332,7 +332,7 @@ func (t *thread) unexpectedStubExit() { msg, err := t.getEventMessage() status := syscall.WaitStatus(msg) if status.Signaled() && status.Signal() == syscall.SIGKILL { - // SIGKILL can be only sent by an user or OOM-killer. In both + // SIGKILL can be only sent by a user or OOM-killer. In both // these cases, we don't need to panic. There is no reasons to // think that something wrong in gVisor. log.Warningf("The ptrace stub process %v has been killed by SIGKILL.", t.tgid) diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ff1cfd8f6..55c0f04f3 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,12 +121,13 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.TCPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + if !ok { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(pkt.Data.First()) + tcpHeader = header.TCP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3359418c1..04d03d494 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,12 +120,13 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.UDPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(pkt.Data.First()) + udpHeader = header.UDP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index 87dcad18b..d9fb808c0 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -17,6 +17,7 @@ package syscalls import ( "time" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -118,7 +119,7 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { } // WaitEpoll implements the epoll_wait(2) linux syscall. -func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]epoll.Event, error) { +func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]linux.EpollEvent, error) { // Get epoll from the file descriptor. epollfile := t.GetFile(fd) if epollfile == nil { diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 3ab93fbde..51bf205cf 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" "gvisor.dev/gvisor/pkg/sentry/syscalls" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -72,7 +71,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc var data [2]int32 if op != linux.EPOLL_CTL_DEL { var e linux.EpollEvent - if _, err := t.CopyIn(eventAddr, &e); err != nil { + if _, err := e.CopyIn(t, eventAddr); err != nil { return 0, nil, err } @@ -105,28 +104,6 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } } -// copyOutEvents copies epoll events from the kernel to user memory. -func copyOutEvents(t *kernel.Task, addr usermem.Addr, e []epoll.Event) error { - const itemLen = 12 - buffLen := len(e) * itemLen - if _, ok := addr.AddLength(uint64(buffLen)); !ok { - return syserror.EFAULT - } - - b := t.CopyScratchBuffer(buffLen) - for i := range e { - usermem.ByteOrder.PutUint32(b[i*itemLen:], e[i].Events) - usermem.ByteOrder.PutUint32(b[i*itemLen+4:], uint32(e[i].Data[0])) - usermem.ByteOrder.PutUint32(b[i*itemLen+8:], uint32(e[i].Data[1])) - } - - if _, err := t.CopyOutBytes(addr, b); err != nil { - return err - } - - return nil -} - // EpollWait implements the epoll_wait(2) linux syscall. func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { epfd := args[0].Int() @@ -140,7 +117,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } if len(r) != 0 { - if err := copyOutEvents(t, eventsAddr, r); err != nil { + if _, err := linux.CopyEpollEventSliceOut(t, eventsAddr, r); err != nil { return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/sys_mempolicy.go b/pkg/sentry/syscalls/linux/sys_mempolicy.go index ac934dc6f..9b4a5c3f1 100644 --- a/pkg/sentry/syscalls/linux/sys_mempolicy.go +++ b/pkg/sentry/syscalls/linux/sys_mempolicy.go @@ -162,10 +162,10 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - policy = 0 // maxNodes == 1 + policy = linux.MPOL_DEFAULT // maxNodes == 1 } if mode != 0 { - if _, err := t.CopyOut(mode, policy); err != nil { + if _, err := policy.CopyOut(t, mode); err != nil { return 0, nil, err } } @@ -199,10 +199,10 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if policy&^linux.MPOL_MODE_FLAGS != linux.MPOL_INTERLEAVE { return 0, nil, syserror.EINVAL } - policy = 0 // maxNodes == 1 + policy = linux.MPOL_DEFAULT // maxNodes == 1 } if mode != 0 { - if _, err := t.CopyOut(mode, policy); err != nil { + if _, err := policy.CopyOut(t, mode); err != nil { return 0, nil, err } } @@ -216,7 +216,7 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. // SetMempolicy implements the syscall set_mempolicy(2). func SetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - modeWithFlags := args[0].Int() + modeWithFlags := linux.NumaPolicy(args[0].Int()) nodemask := args[1].Pointer() maxnode := args[2].Uint() @@ -233,7 +233,7 @@ func SetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. func Mbind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() length := args[1].Uint64() - mode := args[2].Int() + mode := linux.NumaPolicy(args[2].Int()) nodemask := args[3].Pointer() maxnode := args[4].Uint() flags := args[5].Uint() @@ -258,9 +258,9 @@ func Mbind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } -func copyInMempolicyNodemask(t *kernel.Task, modeWithFlags int32, nodemask usermem.Addr, maxnode uint32) (int32, uint64, error) { - flags := modeWithFlags & linux.MPOL_MODE_FLAGS - mode := modeWithFlags &^ linux.MPOL_MODE_FLAGS +func copyInMempolicyNodemask(t *kernel.Task, modeWithFlags linux.NumaPolicy, nodemask usermem.Addr, maxnode uint32) (linux.NumaPolicy, uint64, error) { + flags := linux.NumaPolicy(modeWithFlags & linux.MPOL_MODE_FLAGS) + mode := linux.NumaPolicy(modeWithFlags &^ linux.MPOL_MODE_FLAGS) if flags == linux.MPOL_MODE_FLAGS { // Can't specify both mode flags simultaneously. return 0, 0, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 6ff2d84d2..f6fb0f219 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -6,7 +6,6 @@ go_library( name = "vfs2", srcs = [ "epoll.go", - "epoll_unsafe.go", "execve.go", "fd.go", "filesystem.go", diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index 5a938cee2..34c90ae3e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +var sizeofEpollEvent = (*linux.EpollEvent)(nil).SizeBytes() + // EpollCreate1 implements Linux syscall epoll_create1(2). func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { flags := args[0].Int() @@ -124,7 +126,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys maxEvents := int(args[2].Int()) timeout := int(args[3].Int()) - const _EP_MAX_EVENTS = math.MaxInt32 / sizeofEpollEvent // Linux: fs/eventpoll.c:EP_MAX_EVENTS + var _EP_MAX_EVENTS = math.MaxInt32 / sizeofEpollEvent // Linux: fs/eventpoll.c:EP_MAX_EVENTS if maxEvents <= 0 || maxEvents > _EP_MAX_EVENTS { return 0, nil, syserror.EINVAL } @@ -157,7 +159,8 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys maxEvents -= n if n != 0 { // Copy what we read out. - copiedEvents, err := copyOutEvents(t, eventsAddr, events[:n]) + copiedBytes, err := linux.CopyEpollEventSliceOut(t, eventsAddr, events[:n]) + copiedEvents := copiedBytes / sizeofEpollEvent // rounded down eventsAddr += usermem.Addr(copiedEvents * sizeofEpollEvent) total += copiedEvents if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go deleted file mode 100644 index 825f325bf..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go +++ /dev/null @@ -1,44 +0,0 @@ -// Copyright 2020 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 vfs2 - -import ( - "reflect" - "runtime" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/gohacks" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/usermem" -) - -const sizeofEpollEvent = int(unsafe.Sizeof(linux.EpollEvent{})) - -func copyOutEvents(t *kernel.Task, addr usermem.Addr, events []linux.EpollEvent) (int, error) { - if len(events) == 0 { - return 0, nil - } - // Cast events to a byte slice for copying. - var eventBytes []byte - eventBytesHdr := (*reflect.SliceHeader)(unsafe.Pointer(&eventBytes)) - eventBytesHdr.Data = uintptr(gohacks.Noescape(unsafe.Pointer(&events[0]))) - eventBytesHdr.Len = len(events) * sizeofEpollEvent - eventBytesHdr.Cap = len(events) * sizeofEpollEvent - copiedBytes, err := t.CopyOutBytes(addr, eventBytes) - runtime.KeepAlive(events) - copiedEvents := copiedBytes / sizeofEpollEvent // rounded down - return copiedEvents, err -} diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index 62e98817d..c7c7bf7ce 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -130,7 +130,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { if cb.t.Arch().Width() != 8 { panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width())) } - size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name) + size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name) size = (size + 7) &^ 7 // round up to multiple of sizeof(long) if size > cb.remaining { return syserror.EINVAL @@ -143,11 +143,11 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { // Zero out all remaining bytes in buf, including the NUL terminator // after dirent.Name and the zero padding byte between the name and // dirent type. - bufTail := buf[18+len(dirent.Name):] + bufTail := buf[18+len(dirent.Name) : size-1] for i := range bufTail { bufTail[i] = 0 } - bufTail[2] = dirent.Type + buf[size-1] = dirent.Type } n, err := cb.t.CopyOutBytes(cb.addr, buf) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 21eb98444..74920f785 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -58,8 +58,8 @@ func Override(table map[uintptr]kernel.Syscall) { table[51] = syscalls.PartiallySupported("getsockname", GetSockName, "In process of porting socket syscalls to VFS2.", nil) table[52] = syscalls.PartiallySupported("getpeername", GetPeerName, "In process of porting socket syscalls to VFS2.", nil) table[53] = syscalls.PartiallySupported("socketpair", SocketPair, "In process of porting socket syscalls to VFS2.", nil) - table[54] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) - table[55] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[54] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[55] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) table[59] = syscalls.Supported("execve", Execve) table[72] = syscalls.Supported("fcntl", Fcntl) delete(table, 73) // flock diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 15cc091e2..418d69b96 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -111,10 +111,10 @@ type FileDescriptionOptions struct { } // Init must be called before first use of fd. If it succeeds, it takes -// references on mnt and d. statusFlags is the initial file description status -// flags, which is usually the full set of flags passed to open(2). -func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) error { - writable := MayWriteFileWithOpenFlags(statusFlags) +// references on mnt and d. flags is the initial file description flags, which +// is usually the full set of flags passed to open(2). +func (fd *FileDescription) Init(impl FileDescriptionImpl, flags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) error { + writable := MayWriteFileWithOpenFlags(flags) if writable { if err := mnt.CheckBeginWrite(); err != nil { return err @@ -122,7 +122,10 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn } fd.refs = 1 - fd.statusFlags = statusFlags + + // Remove "file creation flags" to mirror the behavior from file.f_flags in + // fs/open.c:do_dentry_open + fd.statusFlags = flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC) fd.vd = VirtualDentry{ mount: mnt, dentry: d, @@ -130,7 +133,7 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn mnt.IncRef() d.IncRef() fd.opts = *opts - fd.readable = MayReadFileWithOpenFlags(statusFlags) + fd.readable = MayReadFileWithOpenFlags(flags) fd.writable = writable fd.impl = impl return nil diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 74577bc2f..20e5bb072 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -443,8 +443,7 @@ type FilesystemImpl interface { // Errors: // // - If extended attributes are not supported by the filesystem, - // ListxattrAt returns nil. (See FileDescription.Listxattr for an - // explanation.) + // ListxattrAt returns ENOTSUP. // // - If the size of the list (including a NUL terminating byte after every // entry) would exceed size, ERANGE may be returned. Note that diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 534528ce6..022bac127 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -33,6 +33,25 @@ type GetDentryOptions struct { type MkdirOptions struct { // Mode is the file mode bits for the created directory. Mode linux.FileMode + + // If ForSyntheticMountpoint is true, FilesystemImpl.MkdirAt() may create + // the given directory in memory only (as opposed to persistent storage). + // The created directory should be able to support the creation of + // subdirectories with ForSyntheticMountpoint == true. It does not need to + // support the creation of subdirectories with ForSyntheticMountpoint == + // false, or files of other types. + // + // FilesystemImpls are permitted to ignore the ForSyntheticMountpoint + // option. + // + // The ForSyntheticMountpoint option exists because, unlike mount(2), the + // OCI Runtime Specification permits the specification of mount points that + // do not exist, under the expectation that container runtimes will create + // them. (More accurately, the OCI Runtime Specification completely fails + // to document this feature, but it's implemented by runc.) + // ForSyntheticMountpoint allows such mount points to be created even when + // the underlying persistent filesystem is immutable. + ForSyntheticMountpoint bool } // MknodOptions contains options to VirtualFilesystem.MknodAt() and diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 8ec5d5d5c..f01217c91 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,7 +77,8 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. +// TrimFront removes the first "count" bytes of the vectorised view. It panics +// if count > vv.Size(). func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -86,7 +87,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } } @@ -104,7 +105,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } if copied == 0 { return 0, io.EOF @@ -126,7 +127,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } return copied } @@ -162,22 +163,37 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// First returns the first view of the vectorised view. -func (vv *VectorisedView) First() View { +// PullUp returns the first "count" bytes of the vectorised view. If those +// bytes aren't already contiguous inside the vectorised view, PullUp will +// reallocate as needed to make them contiguous. PullUp fails and returns false +// when count > vv.Size(). +func (vv *VectorisedView) PullUp(count int) (View, bool) { if len(vv.views) == 0 { - return nil + return nil, count == 0 + } + if count <= len(vv.views[0]) { + return vv.views[0][:count], true + } + if count > vv.size { + return nil, false } - return vv.views[0] -} -// RemoveFirst removes the first view of the vectorised view. -func (vv *VectorisedView) RemoveFirst() { - if len(vv.views) == 0 { - return + newFirst := NewView(count) + i := 0 + for offset := 0; offset < count; i++ { + copy(newFirst[offset:], vv.views[i]) + if count-offset < len(vv.views[i]) { + vv.views[i].TrimFront(count - offset) + break + } + offset += len(vv.views[i]) + vv.views[i] = nil } - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] + // We're guaranteed that i > 0, since count is too large for the first + // view. + vv.views[i-1] = newFirst + vv.views = vv.views[i-1:] + return newFirst, true } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -225,3 +241,10 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } + +// removeFirst panics when len(vv.views) < 1. +func (vv *VectorisedView) removeFirst() { + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] +} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index 106e1994c..c56795c7b 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,6 +16,7 @@ package buffer import ( + "bytes" "reflect" "testing" ) @@ -370,3 +371,115 @@ func TestVVRead(t *testing.T) { }) } } + +var pullUpTestCases = []struct { + comment string + in VectorisedView + count int + want []byte + result VectorisedView + ok bool +}{ + { + comment: "simple case", + in: vv(2, "12"), + count: 1, + want: []byte("1"), + result: vv(2, "12"), + ok: true, + }, + { + comment: "entire View", + in: vv(2, "1", "2"), + count: 1, + want: []byte("1"), + result: vv(2, "1", "2"), + ok: true, + }, + { + comment: "spanning across two Views", + in: vv(3, "1", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, + { + comment: "spanning across all Views", + in: vv(5, "1", "23", "45"), + count: 5, + want: []byte("12345"), + result: vv(5, "12345"), + ok: true, + }, + { + comment: "count = 0", + in: vv(1, "1"), + count: 0, + want: []byte{}, + result: vv(1, "1"), + ok: true, + }, + { + comment: "count = size", + in: vv(1, "1"), + count: 1, + want: []byte("1"), + result: vv(1, "1"), + ok: true, + }, + { + comment: "count too large", + in: vv(3, "1", "23"), + count: 4, + want: nil, + result: vv(3, "1", "23"), + ok: false, + }, + { + comment: "empty vv", + in: vv(0, ""), + count: 1, + want: nil, + result: vv(0, ""), + ok: false, + }, + { + comment: "empty vv, count = 0", + in: vv(0, ""), + count: 0, + want: nil, + result: vv(0, ""), + ok: true, + }, + { + comment: "empty views", + in: vv(3, "", "1", "", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, +} + +func TestPullUp(t *testing.T) { + for _, c := range pullUpTestCases { + got, ok := c.in.PullUp(c.count) + + // Is the return value right? + if ok != c.ok { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", + c.comment, c.count, c.in, ok, c.ok) + } + if bytes.Compare(got, View(c.want)) != 0 { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", + c.comment, c.count, c.in, got, c.want) + } + + // Is the underlying structure right? + if !reflect.DeepEqual(c.in, c.result) { + t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", + c.comment, c.count, c.in, c.result) + } + } +} diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 1e2255bfa..073c84ef9 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // Reject the packet if it's shorter than an ethernet header. - if vv.Size() < header.EthernetMinimumSize { + // There should be an ethernet header at the beginning of vv. + hdr, ok := vv.PullUp(header.EthernetMinimumSize) + if !ok { + // Reject the packet if it's shorter than an ethernet header. return tcpip.ErrBadAddress } - - // There should be an ethernet header at the beginning of vv. - linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + linkHeader := header.Ethernet(hdr) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 14b527bc2..9cc08d0e2 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -18,3 +18,10 @@ go_library( "@org_golang_x_sys//unix:go_default_library", ], ) + +go_test( + name = "rawfile_test", + size = "small", + srcs = ["rawfile_test.go"], + library = ":rawfile", +) diff --git a/pkg/tcpip/link/rawfile/rawfile_test.go b/pkg/tcpip/link/rawfile/rawfile_test.go new file mode 100644 index 000000000..8f14ba761 --- /dev/null +++ b/pkg/tcpip/link/rawfile/rawfile_test.go @@ -0,0 +1,46 @@ +// Copyright 2020 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 linux + +package rawfile + +import ( + "syscall" + "testing" +) + +func TestNonBlockingWrite3ZeroLength(t *testing.T) { + fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) + if err != nil { + t.Fatalf("failed to open /dev/null: %v", err) + } + defer syscall.Close(fd) + + if err := NonBlockingWrite3(fd, []byte{}, []byte{0}, nil); err != nil { + t.Fatalf("failed to write: %v", err) + } +} + +func TestNonBlockingWrite3Nil(t *testing.T) { + fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) + if err != nil { + t.Fatalf("failed to open /dev/null: %v", err) + } + defer syscall.Close(fd) + + if err := NonBlockingWrite3(fd, nil, []byte{0}, nil); err != nil { + t.Fatalf("failed to write: %v", err) + } +} diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 44e25d475..92efd0bf8 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -76,9 +76,13 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { // We have two buffers. Build the iovec that represents them and issue // a writev syscall. + var base *byte + if len(b1) > 0 { + base = &b1[0] + } iovec := [3]syscall.Iovec{ { - Base: &b1[0], + Base: base, Len: uint64(len(b1)), }, { diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 27ea3f531..33f640b85 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.First()) + rcvd := []byte(c.packets[0].vv.ToView()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index be2537a82..0799c8f4d 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,11 +171,7 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - first := pkt.Header.View() - if len(first) == 0 { - first = pkt.Data.First() - } - logPacket(prefix, protocol, first, gso) + logPacket(prefix, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -238,7 +234,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -247,28 +243,49 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie size := uint16(0) var fragmentOffset uint16 var moreFragments bool + + // Create a clone of pkt, including any headers if present. Avoid allocating + // backing memory for the clone. + views := [8]buffer.View{} + vv := buffer.NewVectorisedView(0, views[:0]) + vv.AppendView(pkt.Header.View()) + vv.Append(pkt.Data) + switch protocol { case header.IPv4ProtocolNumber: - ipv4 := header.IPv4(b) + hdr, ok := vv.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + ipv4 := header.IPv4(hdr) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - b = b[ipv4.HeaderLength():] + vv.TrimFront(int(ipv4.HeaderLength())) id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - ipv6 := header.IPv6(b) + hdr, ok := vv.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + ipv6 := header.IPv6(hdr) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - b = b[header.IPv6MinimumSize:] + vv.TrimFront(header.IPv6MinimumSize) case header.ARPProtocolNumber: - arp := header.ARP(b) + hdr, ok := vv.PullUp(header.ARPSize) + if !ok { + return + } + vv.TrimFront(header.ARPSize) + arp := header.ARP(hdr) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -284,7 +301,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -297,7 +314,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - icmp := header.ICMPv4(b) + hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) + if !ok { + break + } + icmp := header.ICMPv4(hdr) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -330,7 +351,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.ICMPv6ProtocolNumber: transName = "icmp" - icmp := header.ICMPv6(b) + hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) + if !ok { + break + } + icmp := header.ICMPv6(hdr) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -361,7 +386,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.UDPProtocolNumber: transName = "udp" - udp := header.UDP(b) + hdr, ok := vv.PullUp(header.UDPMinimumSize) + if !ok { + break + } + udp := header.UDP(hdr) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -371,7 +400,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.TCPProtocolNumber: transName = "tcp" - tcp := header.TCP(b) + hdr, ok := vv.PullUp(header.TCPMinimumSize) + if !ok { + break + } + tcp := header.TCP(hdr) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 7acbfa0a8..cf73a939e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -93,7 +93,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v := pkt.Data.First() + v, ok := pkt.Data.PullUp(header.ARPSize) + if !ok { + return + } h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index c4bf1ba5c..4cbefe5ab 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,7 +25,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + hdr := header.IPv4(h) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -34,12 +38,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } - hlen := int(h.HeaderLength()) - if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { + hlen := int(hdr.HeaderLength()) + if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -48,15 +52,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := h.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := hdr.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv4MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 104aafbed..17202cc7a 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -328,7 +328,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return tcpip.ErrInvalidOptionValue + } + ip := header.IPv4(h) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -378,7 +382,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b68983d10..bdf3a0d25 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,7 +28,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv6(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + hdr := header.IPv6(h) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -36,17 +40,21 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := h.TransportProtocol() + p := hdr.TransportProtocol() if p == header.IPv6FragmentHeader { - f := header.IPv6Fragment(pkt.Data.First()) - if !f.IsValid() || f.FragmentOffset() != 0 { + f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) + if !ok { + return + } + fragHdr := header.IPv6Fragment(f) + if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -55,19 +63,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = f.TransportProtocol() + p = fragHdr.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv6MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) + if !ok { received.Invalid.Increment() return } @@ -76,11 +84,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // - // Only the first view in vv is accounted for by h. To account for the - // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.RemoveFirst() + payload.TrimFront(len(h)) if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -101,34 +107,40 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - if len(v) < header.ICMPv6PacketTooBigMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := h.MTU() + mtu := header.ICMPv6(hdr).MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - if len(v) < header.ICMPv6DstUnreachableMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch h.Code() { + switch header.ICMPv6(hdr).Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - ns := header.NDPNeighborSolicit(h.NDPPayload()) + // The remainder of payload must be only the neighbor solicitation, so + // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, + // NDP messages cannot be fragmented. Also note that in the common case NDP + // datagrams are very small and ToView() will not incur allocations. + ns := header.NDPNeighborSolicit(payload.ToView()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -286,12 +298,16 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - na := header.NDPNeighborAdvert(h.NDPPayload()) + // The remainder of payload must be only the neighbor advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + na := header.NDPNeighborAdvert(payload.ToView()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -363,14 +379,15 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, h) + copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -384,7 +401,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -406,8 +423,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - p := h.NDPPayload() - if len(p) < header.NDPRAMinimumSize || !isNDPValid() { + // Is the NDP payload of sufficient size to hold a Router + // Advertisement? + if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -425,7 +443,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - ra := header.NDPRouterAdvert(p) + // The remainder of payload must be only the router advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + ra := header.NDPRouterAdvert(payload.ToView()) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index bd099a7f8..d412ff688 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,7 +166,8 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize}, + size: header.ICMPv6NeighborSolicitMinimumSize, + }, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 331b0817b..486725131 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,7 +171,11 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index e9c652042..c7c663498 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,7 +70,10 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -473,7 +476,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -517,7 +520,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -564,7 +567,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -619,7 +622,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6c0a4b24d..6b91159d4 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,6 +212,11 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. +// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -226,7 +231,9 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -271,14 +278,21 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pk.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data.First(). + // pkt.Data. if pkt.NetworkHeader == nil { - pkt.NetworkHeader = pkt.Data.First() + var ok bool + pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // Precondition has been violated. + panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) + } } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 7b4543caf..8be61f4b1 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,9 +96,12 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView := newPkt.Data.First() + headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return RuleDrop, 0 + } netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] + newPkt.NetworkHeader = headerView hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -117,10 +120,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.UDPMinimumSize { + if pkt.Data.Size() < header.UDPMinimumSize { + return RuleDrop, 0 + } + hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) + if !ok { return RuleDrop, 0 } - udpHeader = header.UDP(newPkt.Data.First()) + udpHeader = header.UDP(hdr) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -128,10 +135,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.TCPMinimumSize { + if pkt.Data.Size() < header.TCPMinimumSize { return RuleDrop, 0 } - tcpHeader = header.TCP(newPkt.TransportHeader) + hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) + if !ok { + return RuleDrop, 0 + } + tcpHeader = header.TCP(hdr) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 016dbe15e..0c2b1f36a 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1203,12 +1203,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - if len(pkt.Data.First()) < netProto.MinimumPacketSize() { + netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - - src, dst := netProto.ParseAddresses(pkt.Data.First()) + src, dst := netProto.ParseAddresses(netHeader) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1289,22 +1289,8 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - - firstData := pkt.Data.First() - pkt.Data.RemoveFirst() - - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { - pkt.Header = buffer.NewPrependableFromView(firstData) - } else { - firstDataLen := len(firstData) - - // pkt.Header should have enough capacity to hold n.linkEP's headers. - pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) - - // TODO(b/151227689): avoid copying the packet when forwarding - if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) - } + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { + pkt.Header = buffer.NewPrependable(linkHeaderLen) } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1332,12 +1318,13 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - if len(pkt.Data.First()) < transProto.MinimumPacketSize() { + transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1375,11 +1362,12 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(pkt.Data.First()) < 8 { + transHeader, ok := pkt.Data.PullUp(8) + if !ok { return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index dc125f25e..7d36f8e84 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,7 +37,13 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. + // down the stack, each layer adds to Header. Note that forwarded + // packets don't populate Headers on their way out -- their headers and + // payload are never parsed out and remain in Data. + // + // TODO(gvisor.dev/issue/170): Forwarded packets don't currently + // populate Header, but should. This will be doable once early parsing + // (https://github.com/google/gvisor/pull/1995) is supported. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index c7634ceb1..d45d2cc1f 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,16 +95,18 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb := pkt.Data.First() - if len(nb) < fakeNetHeaderLen { + nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { return } - pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 3084e6593..a611e44ab 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,10 +642,11 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Pkt.Header.View()[0]; dst != 3 { + hdrs := p.Pkt.Data.ToView() + if dst := hdrs[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Pkt.Header.View()[1]; src != 1 { + if src := hdrs[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index feef8dca0..b1d820372 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.Data.First()) - if h.Type() != header.ICMPv4EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.Data.First()) - if h.Type() != header.ICMPv6EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) + if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 61426623c..f2aa69069 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -105,8 +105,8 @@ go_test( "//pkg/tcpip/seqnum", "//pkg/tcpip/stack", "//pkg/tcpip/transport/tcp/testing/context", + "//pkg/test/testutil", "//pkg/waiter", - "//runsc/testutil", ], ) diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 45f2aa78b..07d3e64c8 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2158,8 +2158,6 @@ func (e *endpoint) shutdownLocked(flags tcpip.ShutdownFlags) *tcpip.Error { // // By not removing this endpoint from the demuxer mapping, we // ensure that any other bind to the same port fails, as on Linux. - // TODO(gvisor.dev/issue/2468): We need to enable applications to - // start listening on this endpoint again similar to Linux. e.rcvListMu.Lock() e.rcvClosed = true e.rcvListMu.Unlock() @@ -2188,26 +2186,31 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { e.LockUser() defer e.UnlockUser() - // Allow the backlog to be adjusted if the endpoint is not shutting down. - // When the endpoint shuts down, it sets workerCleanup to true, and from - // that point onward, acceptedChan is the responsibility of the cleanup() - // method (and should not be touched anywhere else, including here). - if e.EndpointState() == StateListen && !e.workerCleanup { - // Adjust the size of the channel iff we can fix existing - // pending connections into the new one. + if e.EndpointState() == StateListen && !e.closed { e.acceptMu.Lock() defer e.acceptMu.Unlock() - if len(e.acceptedChan) > backlog { - return tcpip.ErrInvalidEndpointState - } - if cap(e.acceptedChan) == backlog { - return nil - } - origChan := e.acceptedChan - e.acceptedChan = make(chan *endpoint, backlog) - close(origChan) - for ep := range origChan { - e.acceptedChan <- ep + if e.acceptedChan == nil { + // listen is called after shutdown. + e.acceptedChan = make(chan *endpoint, backlog) + e.shutdownFlags = 0 + e.rcvListMu.Lock() + e.rcvClosed = false + e.rcvListMu.Unlock() + } else { + // Adjust the size of the channel iff we can fix + // existing pending connections into the new one. + if len(e.acceptedChan) > backlog { + return tcpip.ErrInvalidEndpointState + } + if cap(e.acceptedChan) == backlog { + return nil + } + origChan := e.acceptedChan + e.acceptedChan = make(chan *endpoint, backlog) + close(origChan) + for ep := range origChan { + e.acceptedChan <- ep + } } // Notify any blocked goroutines that they can attempt to diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index c3c692555..8b7562396 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -247,6 +247,11 @@ func (e *endpoint) Resume(s *stack.Stack) { if err := e.Listen(backlog); err != nil { panic("endpoint listening failed: " + err.String()) } + e.LockUser() + if e.shutdownFlags != 0 { + e.shutdownLocked(e.shutdownFlags) + } + e.UnlockUser() listenLoading.Done() tcpip.AsyncLoading.Done() }() diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 40461fd31..7712ce652 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,7 +144,11 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h := header.TCP(s.data.First()) + h, ok := s.data.PullUp(header.TCPMinimumSize) + if !ok { + return false + } + hdr := header.TCP(h) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -156,12 +160,16 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(h.DataOffset()) - if offset < header.TCPMinimumSize || offset > len(h) { + offset := int(hdr.DataOffset()) + if offset < header.TCPMinimumSize { + return false + } + hdrWithOpts, ok := s.data.PullUp(offset) + if !ok { return false } - s.options = []byte(h[header.TCPMinimumSize:offset]) + s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -173,18 +181,19 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - s.csum = h.Checksum() + hdr = header.TCP(hdrWithOpts) + s.csum = hdr.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = h.CalculateChecksum(xsum) + xsum = hdr.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(h.SequenceNumber()) - s.ackNumber = seqnum.Value(h.AckNumber()) - s.flags = h.Flags() - s.window = seqnum.Size(h.WindowSize()) + s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) + s.ackNumber = seqnum.Value(hdr.AckNumber()) + s.flags = hdr.Flags() + s.window = seqnum.Size(hdr.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go index 359a75e73..5fe23113b 100644 --- a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go +++ b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go @@ -31,7 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) func TestFastRecovery(t *testing.T) { diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index ab1014c7f..286c66cf5 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3548,7 +3548,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3575,7 +3575,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index edb54f0be..756ab913a 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: hdr.SourcePort(), + Port: header.UDP(hdr).SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 6e31a9bac..52af6de22 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,8 +68,13 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + h, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { + // Malformed packet. + r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() + return true + } + if int(header.UDP(h).Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true diff --git a/runsc/criutil/BUILD b/pkg/test/criutil/BUILD index 8a571a000..a7b082cee 100644 --- a/runsc/criutil/BUILD +++ b/pkg/test/criutil/BUILD @@ -7,5 +7,8 @@ go_library( testonly = 1, srcs = ["criutil.go"], visibility = ["//:sandbox"], - deps = ["//runsc/testutil"], + deps = [ + "//pkg/test/dockerutil", + "//pkg/test/testutil", + ], ) diff --git a/runsc/criutil/criutil.go b/pkg/test/criutil/criutil.go index 773f5a1c4..bebebb48e 100644 --- a/runsc/criutil/criutil.go +++ b/pkg/test/criutil/criutil.go @@ -25,40 +25,45 @@ import ( "strings" "time" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) -const endpointPrefix = "unix://" - // Crictl contains information required to run the crictl utility. type Crictl struct { - executable string - timeout time.Duration - imageEndpoint string - runtimeEndpoint string + logger testutil.Logger + endpoint string + cleanup []func() +} + +// resolvePath attempts to find binary paths. It may set the path to invalid, +// which will cause the execution to fail with a sensible error. +func resolvePath(executable string) string { + guess, err := exec.LookPath(executable) + if err != nil { + guess = fmt.Sprintf("/usr/local/bin/%s", executable) + } + return guess } // NewCrictl returns a Crictl configured with a timeout and an endpoint over // which it will talk to containerd. -func NewCrictl(timeout time.Duration, endpoint string) *Crictl { - // Bazel doesn't pass PATH through, assume the location of crictl - // unless specified by environment variable. - executable := os.Getenv("CRICTL_PATH") - if executable == "" { - executable = "/usr/local/bin/crictl" - } +func NewCrictl(logger testutil.Logger, endpoint string) *Crictl { + // Attempt to find the executable, but don't bother propagating the + // error at this point. The first command executed will return with a + // binary not found error. return &Crictl{ - executable: executable, - timeout: timeout, - imageEndpoint: endpointPrefix + endpoint, - runtimeEndpoint: endpointPrefix + endpoint, + logger: logger, + endpoint: endpoint, } } -// Pull pulls an container image. It corresponds to `crictl pull`. -func (cc *Crictl) Pull(imageName string) error { - _, err := cc.run("pull", imageName) - return err +// CleanUp executes cleanup functions. +func (cc *Crictl) CleanUp() { + for _, c := range cc.cleanup { + c() + } + cc.cleanup = nil } // RunPod creates a sandbox. It corresponds to `crictl runp`. @@ -157,27 +162,66 @@ func (cc *Crictl) RmPod(podID string) error { return err } +// Import imports the given container from the local Docker instance. +func (cc *Crictl) Import(image string) error { + // Note that we provide a 10 minute timeout after connect because we may + // be pushing a lot of bytes in order to import the image. The connect + // timeout stays the same and is inherited from the Crictl instance. + cmd := testutil.Command(cc.logger, + resolvePath("ctr"), + fmt.Sprintf("--connect-timeout=%s", 30*time.Second), + fmt.Sprintf("--address=%s", cc.endpoint), + "-n", "k8s.io", "images", "import", "-") + cmd.Stderr = os.Stderr // Pass through errors. + + // Create a pipe and start the program. + w, err := cmd.StdinPipe() + if err != nil { + return err + } + if err := cmd.Start(); err != nil { + return err + } + + // Save the image on the other end. + if err := dockerutil.Save(cc.logger, image, w); err != nil { + cmd.Wait() + return err + } + + // Close our pipe reference & see if it was loaded. + if err := w.Close(); err != nil { + return w.Close() + } + + return cmd.Wait() +} + // StartContainer pulls the given image ands starts the container in the // sandbox with the given podID. +// +// Note that the image will always be imported from the local docker daemon. func (cc *Crictl) StartContainer(podID, image, sbSpec, contSpec string) (string, error) { + if err := cc.Import(image); err != nil { + return "", err + } + // Write the specs to files that can be read by crictl. - sbSpecFile, err := testutil.WriteTmpFile("sbSpec", sbSpec) + sbSpecFile, cleanup, err := testutil.WriteTmpFile("sbSpec", sbSpec) if err != nil { return "", fmt.Errorf("failed to write sandbox spec: %v", err) } - contSpecFile, err := testutil.WriteTmpFile("contSpec", contSpec) + cc.cleanup = append(cc.cleanup, cleanup) + contSpecFile, cleanup, err := testutil.WriteTmpFile("contSpec", contSpec) if err != nil { return "", fmt.Errorf("failed to write container spec: %v", err) } + cc.cleanup = append(cc.cleanup, cleanup) return cc.startContainer(podID, image, sbSpecFile, contSpecFile) } func (cc *Crictl) startContainer(podID, image, sbSpecFile, contSpecFile string) (string, error) { - if err := cc.Pull(image); err != nil { - return "", fmt.Errorf("failed to pull %s: %v", image, err) - } - contID, err := cc.Create(podID, contSpecFile, sbSpecFile) if err != nil { return "", fmt.Errorf("failed to create container in pod %q: %v", podID, err) @@ -203,18 +247,24 @@ func (cc *Crictl) StopContainer(contID string) error { return nil } -// StartPodAndContainer pulls an image, then starts a sandbox and container in -// that sandbox. It returns the pod ID and container ID. +// StartPodAndContainer starts a sandbox and container in that sandbox. It +// returns the pod ID and container ID. func (cc *Crictl) StartPodAndContainer(image, sbSpec, contSpec string) (string, string, error) { + if err := cc.Import(image); err != nil { + return "", "", err + } + // Write the specs to files that can be read by crictl. - sbSpecFile, err := testutil.WriteTmpFile("sbSpec", sbSpec) + sbSpecFile, cleanup, err := testutil.WriteTmpFile("sbSpec", sbSpec) if err != nil { return "", "", fmt.Errorf("failed to write sandbox spec: %v", err) } - contSpecFile, err := testutil.WriteTmpFile("contSpec", contSpec) + cc.cleanup = append(cc.cleanup, cleanup) + contSpecFile, cleanup, err := testutil.WriteTmpFile("contSpec", contSpec) if err != nil { return "", "", fmt.Errorf("failed to write container spec: %v", err) } + cc.cleanup = append(cc.cleanup, cleanup) podID, err := cc.RunPod(sbSpecFile) if err != nil { @@ -243,35 +293,14 @@ func (cc *Crictl) StopPodAndContainer(podID, contID string) error { return nil } -// run runs crictl with the given args and returns an error if it takes longer -// than cc.Timeout to run. +// run runs crictl with the given args. func (cc *Crictl) run(args ...string) (string, error) { defaultArgs := []string{ - "--image-endpoint", cc.imageEndpoint, - "--runtime-endpoint", cc.runtimeEndpoint, - } - cmd := exec.Command(cc.executable, append(defaultArgs, args...)...) - - // Run the command with a timeout. - done := make(chan string) - errCh := make(chan error) - go func() { - output, err := cmd.CombinedOutput() - if err != nil { - errCh <- fmt.Errorf("error: \"%v\", output: %s", err, string(output)) - return - } - done <- string(output) - }() - select { - case output := <-done: - return output, nil - case err := <-errCh: - return "", err - case <-time.After(cc.timeout): - if err := testutil.KillCommand(cmd); err != nil { - return "", fmt.Errorf("timed out, then couldn't kill process %+v: %v", cmd, err) - } - return "", fmt.Errorf("timed out: %+v", cmd) + resolvePath("crictl"), + "--image-endpoint", fmt.Sprintf("unix://%s", cc.endpoint), + "--runtime-endpoint", fmt.Sprintf("unix://%s", cc.endpoint), } + fullArgs := append(defaultArgs, args...) + out, err := testutil.Command(cc.logger, fullArgs...).CombinedOutput() + return string(out), err } diff --git a/runsc/dockerutil/BUILD b/pkg/test/dockerutil/BUILD index 8621af901..7c8758e35 100644 --- a/runsc/dockerutil/BUILD +++ b/pkg/test/dockerutil/BUILD @@ -8,7 +8,7 @@ go_library( srcs = ["dockerutil.go"], visibility = ["//:sandbox"], deps = [ - "//runsc/testutil", + "//pkg/test/testutil", "@com_github_kr_pty//:go_default_library", ], ) diff --git a/pkg/test/dockerutil/dockerutil.go b/pkg/test/dockerutil/dockerutil.go new file mode 100644 index 000000000..baa8fc2f2 --- /dev/null +++ b/pkg/test/dockerutil/dockerutil.go @@ -0,0 +1,581 @@ +// 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. + +// Package dockerutil is a collection of utility functions. +package dockerutil + +import ( + "encoding/json" + "flag" + "fmt" + "io" + "io/ioutil" + "log" + "net" + "os" + "os/exec" + "path" + "regexp" + "strconv" + "strings" + "syscall" + "time" + + "github.com/kr/pty" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +var ( + // runtime is the runtime to use for tests. This will be applied to all + // containers. Note that the default here ("runsc") corresponds to the + // default used by the installations. This is important, because the + // default installer for vm_tests (in tools/installers:head, invoked + // via tools/vm:defs.bzl) will install with this name. So without + // changing anything, tests should have a runsc runtime available to + // them. Otherwise installers should update the existing runtime + // instead of installing a new one. + runtime = flag.String("runtime", "runsc", "specify which runtime to use") + + // config is the default Docker daemon configuration path. + config = flag.String("config_path", "/etc/docker/daemon.json", "configuration file for reading paths") +) + +// EnsureSupportedDockerVersion checks if correct docker is installed. +// +// This logs directly to stderr, as it is typically called from a Main wrapper. +func EnsureSupportedDockerVersion() { + cmd := exec.Command("docker", "version") + out, err := cmd.CombinedOutput() + if err != nil { + log.Fatalf("error running %q: %v", "docker version", err) + } + re := regexp.MustCompile(`Version:\s+(\d+)\.(\d+)\.\d.*`) + matches := re.FindStringSubmatch(string(out)) + if len(matches) != 3 { + log.Fatalf("Invalid docker output: %s", out) + } + major, _ := strconv.Atoi(matches[1]) + minor, _ := strconv.Atoi(matches[2]) + if major < 17 || (major == 17 && minor < 9) { + log.Fatalf("Docker version 17.09.0 or greater is required, found: %02d.%02d", major, minor) + } +} + +// RuntimePath returns the binary path for the current runtime. +func RuntimePath() (string, error) { + // Read the configuration data; the file must exist. + configBytes, err := ioutil.ReadFile(*config) + if err != nil { + return "", err + } + + // Unmarshal the configuration. + c := make(map[string]interface{}) + if err := json.Unmarshal(configBytes, &c); err != nil { + return "", err + } + + // Decode the expected configuration. + r, ok := c["runtimes"] + if !ok { + return "", fmt.Errorf("no runtimes declared: %v", c) + } + rs, ok := r.(map[string]interface{}) + if !ok { + // The runtimes are not a map. + return "", fmt.Errorf("unexpected format: %v", c) + } + r, ok = rs[*runtime] + if !ok { + // The expected runtime is not declared. + return "", fmt.Errorf("runtime %q not found: %v", *runtime, c) + } + rs, ok = r.(map[string]interface{}) + if !ok { + // The runtime is not a map. + return "", fmt.Errorf("unexpected format: %v", c) + } + p, ok := rs["path"].(string) + if !ok { + // The runtime does not declare a path. + return "", fmt.Errorf("unexpected format: %v", c) + } + return p, nil +} + +// Save exports a container image to the given Writer. +// +// Note that the writer should be actively consuming the output, otherwise it +// is not guaranteed that the Save will make any progress and the call may +// stall indefinitely. +// +// This is called by criutil in order to import imports. +func Save(logger testutil.Logger, image string, w io.Writer) error { + cmd := testutil.Command(logger, "docker", "save", testutil.ImageByName(image)) + cmd.Stdout = w // Send directly to the writer. + return cmd.Run() +} + +// MountMode describes if the mount should be ro or rw. +type MountMode int + +const ( + // ReadOnly is what the name says. + ReadOnly MountMode = iota + // ReadWrite is what the name says. + ReadWrite +) + +// String returns the mount mode argument for this MountMode. +func (m MountMode) String() string { + switch m { + case ReadOnly: + return "ro" + case ReadWrite: + return "rw" + } + panic(fmt.Sprintf("invalid mode: %d", m)) +} + +// Docker contains the name and the runtime of a docker container. +type Docker struct { + logger testutil.Logger + Runtime string + Name string + copyErr error + mounts []string + cleanups []func() +} + +// MakeDocker sets up the struct for a Docker container. +// +// Names of containers will be unique. +func MakeDocker(logger testutil.Logger) *Docker { + return &Docker{ + logger: logger, + Name: testutil.RandomID(logger.Name()), + Runtime: *runtime, + } +} + +// Mount mounts the given source and makes it available in the container. +func (d *Docker) Mount(target, source string, mode MountMode) { + d.mounts = append(d.mounts, fmt.Sprintf("-v=%s:%s:%v", source, target, mode)) +} + +// CopyFiles copies in and mounts the given files. They are always ReadOnly. +func (d *Docker) CopyFiles(target string, sources ...string) { + dir, err := ioutil.TempDir("", d.Name) + if err != nil { + d.copyErr = fmt.Errorf("ioutil.TempDir failed: %v", err) + return + } + d.cleanups = append(d.cleanups, func() { os.RemoveAll(dir) }) + if err := os.Chmod(dir, 0755); err != nil { + d.copyErr = fmt.Errorf("os.Chmod(%q, 0755) failed: %v", dir, err) + return + } + for _, name := range sources { + src, err := testutil.FindFile(name) + if err != nil { + d.copyErr = fmt.Errorf("testutil.FindFile(%q) failed: %v", name, err) + return + } + dst := path.Join(dir, path.Base(name)) + if err := testutil.Copy(src, dst); err != nil { + d.copyErr = fmt.Errorf("testutil.Copy(%q, %q) failed: %v", src, dst, err) + return + } + d.logger.Logf("copy: %s -> %s", src, dst) + } + d.Mount(target, dir, ReadOnly) +} + +// Link links the given target. +func (d *Docker) Link(target string, source *Docker) { + d.mounts = append(d.mounts, fmt.Sprintf("--link=%s:%s", source.Name, target)) +} + +// RunOpts are options for running a container. +type RunOpts struct { + // Image is the image relative to images/. This will be mangled + // appropriately, to ensure that only first-party images are used. + Image string + + // Memory is the memory limit in kB. + Memory int + + // Ports are the ports to be allocated. + Ports []int + + // WorkDir sets the working directory. + WorkDir string + + // ReadOnly sets the read-only flag. + ReadOnly bool + + // Env are additional environment variables. + Env []string + + // User is the user to use. + User string + + // Privileged enables privileged mode. + Privileged bool + + // CapAdd are the extra set of capabilities to add. + CapAdd []string + + // CapDrop are the extra set of capabilities to drop. + CapDrop []string + + // Pty indicates that a pty will be allocated. If this is non-nil, then + // this will run after start-up with the *exec.Command and Pty file + // passed in to the function. + Pty func(*exec.Cmd, *os.File) + + // Foreground indicates that the container should be run in the + // foreground. If this is true, then the output will be available as a + // return value from the Run function. + Foreground bool + + // Extra are extra arguments that may be passed. + Extra []string +} + +// args returns common arguments. +// +// Note that this does not define the complete behavior. +func (d *Docker) argsFor(r *RunOpts, command string, p []string) (rv []string) { + isExec := command == "exec" + isRun := command == "run" + + if isRun || isExec { + rv = append(rv, "-i") + } + if r.Pty != nil { + rv = append(rv, "-t") + } + if r.User != "" { + rv = append(rv, fmt.Sprintf("--user=%s", r.User)) + } + if r.Privileged { + rv = append(rv, "--privileged") + } + for _, c := range r.CapAdd { + rv = append(rv, fmt.Sprintf("--cap-add=%s", c)) + } + for _, c := range r.CapDrop { + rv = append(rv, fmt.Sprintf("--cap-drop=%s", c)) + } + for _, e := range r.Env { + rv = append(rv, fmt.Sprintf("--env=%s", e)) + } + if r.WorkDir != "" { + rv = append(rv, fmt.Sprintf("--workdir=%s", r.WorkDir)) + } + if !isExec { + if r.Memory != 0 { + rv = append(rv, fmt.Sprintf("--memory=%dk", r.Memory)) + } + for _, p := range r.Ports { + rv = append(rv, fmt.Sprintf("--publish=%d", p)) + } + if r.ReadOnly { + rv = append(rv, fmt.Sprintf("--read-only")) + } + if len(p) > 0 { + rv = append(rv, "--entrypoint=") + } + } + + // Always attach the test environment & Extra. + rv = append(rv, fmt.Sprintf("--env=RUNSC_TEST_NAME=%s", d.Name)) + rv = append(rv, r.Extra...) + + // Attach necessary bits. + if isExec { + rv = append(rv, d.Name) + } else { + rv = append(rv, d.mounts...) + rv = append(rv, fmt.Sprintf("--runtime=%s", d.Runtime)) + rv = append(rv, fmt.Sprintf("--name=%s", d.Name)) + rv = append(rv, testutil.ImageByName(r.Image)) + } + + // Attach other arguments. + rv = append(rv, p...) + return rv +} + +// run runs a complete command. +func (d *Docker) run(r RunOpts, command string, p ...string) (string, error) { + if d.copyErr != nil { + return "", d.copyErr + } + basicArgs := []string{"docker"} + if command == "spawn" { + command = "run" + basicArgs = append(basicArgs, command) + basicArgs = append(basicArgs, "-d") + } else { + basicArgs = append(basicArgs, command) + } + customArgs := d.argsFor(&r, command, p) + cmd := testutil.Command(d.logger, append(basicArgs, customArgs...)...) + if r.Pty != nil { + // If allocating a terminal, then we just ignore the output + // from the command. + ptmx, err := pty.Start(cmd.Cmd) + if err != nil { + return "", err + } + defer cmd.Wait() // Best effort. + r.Pty(cmd.Cmd, ptmx) + } else { + // Can't support PTY or streaming. + out, err := cmd.CombinedOutput() + return string(out), err + } + return "", nil +} + +// Create calls 'docker create' with the arguments provided. +func (d *Docker) Create(r RunOpts, args ...string) error { + _, err := d.run(r, "create", args...) + return err +} + +// Start calls 'docker start'. +func (d *Docker) Start() error { + return testutil.Command(d.logger, "docker", "start", d.Name).Run() +} + +// Stop calls 'docker stop'. +func (d *Docker) Stop() error { + return testutil.Command(d.logger, "docker", "stop", d.Name).Run() +} + +// Run calls 'docker run' with the arguments provided. +func (d *Docker) Run(r RunOpts, args ...string) (string, error) { + return d.run(r, "run", args...) +} + +// Spawn starts the container and detaches. +func (d *Docker) Spawn(r RunOpts, args ...string) error { + _, err := d.run(r, "spawn", args...) + return err +} + +// Logs calls 'docker logs'. +func (d *Docker) Logs() (string, error) { + // Don't capture the output; since it will swamp the logs. + out, err := exec.Command("docker", "logs", d.Name).CombinedOutput() + return string(out), err +} + +// Exec calls 'docker exec' with the arguments provided. +func (d *Docker) Exec(r RunOpts, args ...string) (string, error) { + return d.run(r, "exec", args...) +} + +// Pause calls 'docker pause'. +func (d *Docker) Pause() error { + return testutil.Command(d.logger, "docker", "pause", d.Name).Run() +} + +// Unpause calls 'docker pause'. +func (d *Docker) Unpause() error { + return testutil.Command(d.logger, "docker", "unpause", d.Name).Run() +} + +// Checkpoint calls 'docker checkpoint'. +func (d *Docker) Checkpoint(name string) error { + return testutil.Command(d.logger, "docker", "checkpoint", "create", d.Name, name).Run() +} + +// Restore calls 'docker start --checkname [name]'. +func (d *Docker) Restore(name string) error { + return testutil.Command(d.logger, "docker", "start", fmt.Sprintf("--checkpoint=%s", name), d.Name).Run() +} + +// Kill calls 'docker kill'. +func (d *Docker) Kill() error { + // Skip logging this command, it will likely be an error. + out, err := exec.Command("docker", "kill", d.Name).CombinedOutput() + if err != nil && !strings.Contains(string(out), "is not running") { + return err + } + return nil +} + +// Remove calls 'docker rm'. +func (d *Docker) Remove() error { + return testutil.Command(d.logger, "docker", "rm", d.Name).Run() +} + +// CleanUp kills and deletes the container (best effort). +func (d *Docker) CleanUp() { + // Kill the container. + if err := d.Kill(); err != nil { + // Just log; can't do anything here. + d.logger.Logf("error killing container %q: %v", d.Name, err) + } + // Remove the image. + if err := d.Remove(); err != nil { + d.logger.Logf("error removing container %q: %v", d.Name, err) + } + // Forget all mounts. + d.mounts = nil + // Execute all cleanups. + for _, c := range d.cleanups { + c() + } + d.cleanups = nil +} + +// FindPort returns the host port that is mapped to 'sandboxPort'. This calls +// docker to allocate a free port in the host and prevent conflicts. +func (d *Docker) FindPort(sandboxPort int) (int, error) { + format := fmt.Sprintf(`{{ (index (index .NetworkSettings.Ports "%d/tcp") 0).HostPort }}`, sandboxPort) + out, err := testutil.Command(d.logger, "docker", "inspect", "-f", format, d.Name).CombinedOutput() + if err != nil { + return -1, fmt.Errorf("error retrieving port: %v", err) + } + port, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + return -1, fmt.Errorf("error parsing port %q: %v", out, err) + } + return port, nil +} + +// FindIP returns the IP address of the container. +func (d *Docker) FindIP() (net.IP, error) { + const format = `{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}` + out, err := testutil.Command(d.logger, "docker", "inspect", "-f", format, d.Name).CombinedOutput() + if err != nil { + return net.IP{}, fmt.Errorf("error retrieving IP: %v", err) + } + ip := net.ParseIP(strings.TrimSpace(string(out))) + if ip == nil { + return net.IP{}, fmt.Errorf("invalid IP: %q", string(out)) + } + return ip, nil +} + +// SandboxPid returns the PID to the sandbox process. +func (d *Docker) SandboxPid() (int, error) { + out, err := testutil.Command(d.logger, "docker", "inspect", "-f={{.State.Pid}}", d.Name).CombinedOutput() + if err != nil { + return -1, fmt.Errorf("error retrieving pid: %v", err) + } + pid, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + return -1, fmt.Errorf("error parsing pid %q: %v", out, err) + } + return pid, nil +} + +// ID returns the container ID. +func (d *Docker) ID() (string, error) { + out, err := testutil.Command(d.logger, "docker", "inspect", "-f={{.Id}}", d.Name).CombinedOutput() + if err != nil { + return "", fmt.Errorf("error retrieving ID: %v", err) + } + return strings.TrimSpace(string(out)), nil +} + +// Wait waits for container to exit, up to the given timeout. Returns error if +// wait fails or timeout is hit. Returns the application return code otherwise. +// Note that the application may have failed even if err == nil, always check +// the exit code. +func (d *Docker) Wait(timeout time.Duration) (syscall.WaitStatus, error) { + timeoutChan := time.After(timeout) + waitChan := make(chan (syscall.WaitStatus)) + errChan := make(chan (error)) + + go func() { + out, err := testutil.Command(d.logger, "docker", "wait", d.Name).CombinedOutput() + if err != nil { + errChan <- fmt.Errorf("error waiting for container %q: %v", d.Name, err) + } + exit, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + errChan <- fmt.Errorf("error parsing exit code %q: %v", out, err) + } + waitChan <- syscall.WaitStatus(uint32(exit)) + }() + + select { + case ws := <-waitChan: + return ws, nil + case err := <-errChan: + return syscall.WaitStatus(1), err + case <-timeoutChan: + return syscall.WaitStatus(1), fmt.Errorf("timeout waiting for container %q", d.Name) + } +} + +// WaitForOutput calls 'docker logs' to retrieve containers output and searches +// for the given pattern. +func (d *Docker) WaitForOutput(pattern string, timeout time.Duration) (string, error) { + matches, err := d.WaitForOutputSubmatch(pattern, timeout) + if err != nil { + return "", err + } + if len(matches) == 0 { + return "", nil + } + return matches[0], nil +} + +// WaitForOutputSubmatch calls 'docker logs' to retrieve containers output and +// searches for the given pattern. It returns any regexp submatches as well. +func (d *Docker) WaitForOutputSubmatch(pattern string, timeout time.Duration) ([]string, error) { + re := regexp.MustCompile(pattern) + var ( + lastOut string + stopped bool + ) + for exp := time.Now().Add(timeout); time.Now().Before(exp); { + out, err := d.Logs() + if err != nil { + return nil, err + } + if out != lastOut { + if lastOut == "" { + d.logger.Logf("output (start): %s", out) + } else if strings.HasPrefix(out, lastOut) { + d.logger.Logf("output (contn): %s", out[len(lastOut):]) + } else { + d.logger.Logf("output (trunc): %s", out) + } + lastOut = out // Save for future. + if matches := re.FindStringSubmatch(lastOut); matches != nil { + return matches, nil // Success! + } + } else if stopped { + // The sandbox stopped and we looked at the + // logs at least once since determining that. + return nil, fmt.Errorf("no longer running: %v", err) + } else if pid, err := d.SandboxPid(); pid == 0 || err != nil { + // The sandbox may have stopped, but it's + // possible that it has emitted the terminal + // line between the last call to Logs and here. + stopped = true + } + time.Sleep(100 * time.Millisecond) + } + return nil, fmt.Errorf("timeout waiting for output %q: %s", re.String(), lastOut) +} diff --git a/runsc/testutil/BUILD b/pkg/test/testutil/BUILD index 945405303..03b1b4677 100644 --- a/runsc/testutil/BUILD +++ b/pkg/test/testutil/BUILD @@ -11,7 +11,6 @@ go_library( ], visibility = ["//:sandbox"], deps = [ - "//pkg/log", "//pkg/sync", "//runsc/boot", "//runsc/specutils", diff --git a/runsc/testutil/testutil.go b/pkg/test/testutil/testutil.go index 5e09f8f16..d75ceca3d 100644 --- a/runsc/testutil/testutil.go +++ b/pkg/test/testutil/testutil.go @@ -25,6 +25,7 @@ import ( "fmt" "io" "io/ioutil" + "log" "math" "math/rand" "net/http" @@ -42,7 +43,6 @@ import ( "github.com/cenkalti/backoff" specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/specutils" @@ -52,15 +52,59 @@ var ( checkpoint = flag.Bool("checkpoint", true, "control checkpoint/restore support") ) -func init() { - rand.Seed(time.Now().UnixNano()) -} - // IsCheckpointSupported returns the relevant command line flag. func IsCheckpointSupported() bool { return *checkpoint } +// nameToActual is used by ImageByName (for now). +var nameToActual = map[string]string{ + "basic/alpine": "alpine", + "basic/busybox": "busybox:1.31.1", + "basic/httpd": "httpd", + "basic/mysql": "mysql", + "basic/nginx": "nginx", + "basic/python": "gcr.io/gvisor-presubmit/python-hello", + "basic/resolv": "k8s.gcr.io/busybox", + "basic/ruby": "ruby", + "basic/tomcat": "tomcat:8.0", + "basic/ubuntu": "ubuntu:trusty", + "iptables": "gcr.io/gvisor-presubmit/iptables-test", + "packetdrill": "gcr.io/gvisor-presubmit/packetdrill", + "packetimpact": "gcr.io/gvisor-presubmit/packetimpact", + "runtimes/go1.12": "gcr.io/gvisor-presubmit/go1.12", + "runtimes/java11": "gcr.io/gvisor-presubmit/java11", + "runtimes/nodejs12.4.0": "gcr.io/gvisor-presubmit/nodejs12.4.0", + "runtimes/php7.3.6": "gcr.io/gvisor-presubmit/php7.3.6", + "runtimes/python3.7.3": "gcr.io/gvisor-presubmit/python3.7.3", +} + +// ImageByName mangles the image name used locally. +// +// For now, this is implemented as a static lookup table. In a subsequent +// change, this will be used to reference a locally-generated image. +func ImageByName(name string) string { + actual, ok := nameToActual[name] + if !ok { + panic(fmt.Sprintf("unknown image: %v", name)) + } + // A terrible hack, for now execute a manual pull. + if out, err := exec.Command("docker", "pull", actual).CombinedOutput(); err != nil { + panic(fmt.Sprintf("error pulling image %q -> %q: %v, out: %s", name, actual, err, string(out))) + } + return actual +} + +// ConfigureExePath configures the executable for runsc in the test environment. +func ConfigureExePath() error { + path, err := FindFile("runsc/runsc") + if err != nil { + return err + } + specutils.ExePath = path + return nil +} + // TmpDir returns the absolute path to a writable directory that can be used as // scratch by the test. func TmpDir() string { @@ -71,20 +115,58 @@ func TmpDir() string { return dir } -// ConfigureExePath configures the executable for runsc in the test environment. -func ConfigureExePath() error { - path, err := FindFile("runsc/runsc") +// Logger is a simple logging wrapper. +// +// This is designed to be implemented by *testing.T. +type Logger interface { + Name() string + Logf(fmt string, args ...interface{}) +} + +// DefaultLogger logs using the log package. +type DefaultLogger string + +// Name implements Logger.Name. +func (d DefaultLogger) Name() string { + return string(d) +} + +// Logf implements Logger.Logf. +func (d DefaultLogger) Logf(fmt string, args ...interface{}) { + log.Printf(fmt, args...) +} + +// Cmd is a simple wrapper. +type Cmd struct { + logger Logger + *exec.Cmd +} + +// CombinedOutput returns the output and logs. +func (c *Cmd) CombinedOutput() ([]byte, error) { + out, err := c.Cmd.CombinedOutput() + if len(out) > 0 { + c.logger.Logf("output: %s", string(out)) + } if err != nil { - return err + c.logger.Logf("error: %v", err) + } + return out, err +} + +// Command is a simple wrapper around exec.Command, that logs. +func Command(logger Logger, args ...string) *Cmd { + logger.Logf("command: %s", strings.Join(args, " ")) + return &Cmd{ + logger: logger, + Cmd: exec.Command(args[0], args[1:]...), } - specutils.ExePath = path - return nil } // TestConfig returns the default configuration to use in tests. Note that // 'RootDir' must be set by caller if required. func TestConfig(t *testing.T) *boot.Config { - logDir := "" + logDir := os.TempDir() if dir, ok := os.LookupEnv("TEST_UNDECLARED_OUTPUTS_DIR"); ok { logDir = dir + "/" } @@ -143,37 +225,45 @@ func NewSpecWithArgs(args ...string) *specs.Spec { } // SetupRootDir creates a root directory for containers. -func SetupRootDir() (string, error) { +func SetupRootDir() (string, func(), error) { rootDir, err := ioutil.TempDir(TmpDir(), "containers") if err != nil { - return "", fmt.Errorf("error creating root dir: %v", err) + return "", nil, fmt.Errorf("error creating root dir: %v", err) } - return rootDir, nil + return rootDir, func() { os.RemoveAll(rootDir) }, nil } // SetupContainer creates a bundle and root dir for the container, generates a // test config, and writes the spec to config.json in the bundle dir. -func SetupContainer(spec *specs.Spec, conf *boot.Config) (rootDir, bundleDir string, err error) { - rootDir, err = SetupRootDir() +func SetupContainer(spec *specs.Spec, conf *boot.Config) (rootDir, bundleDir string, cleanup func(), err error) { + rootDir, rootCleanup, err := SetupRootDir() if err != nil { - return "", "", err + return "", "", nil, err } conf.RootDir = rootDir - bundleDir, err = SetupBundleDir(spec) - return rootDir, bundleDir, err + bundleDir, bundleCleanup, err := SetupBundleDir(spec) + if err != nil { + rootCleanup() + return "", "", nil, err + } + return rootDir, bundleDir, func() { + bundleCleanup() + rootCleanup() + }, err } // SetupBundleDir creates a bundle dir and writes the spec to config.json. -func SetupBundleDir(spec *specs.Spec) (bundleDir string, err error) { - bundleDir, err = ioutil.TempDir(TmpDir(), "bundle") +func SetupBundleDir(spec *specs.Spec) (string, func(), error) { + bundleDir, err := ioutil.TempDir(TmpDir(), "bundle") if err != nil { - return "", fmt.Errorf("error creating bundle dir: %v", err) + return "", nil, fmt.Errorf("error creating bundle dir: %v", err) } - - if err = writeSpec(bundleDir, spec); err != nil { - return "", fmt.Errorf("error writing spec: %v", err) + cleanup := func() { os.RemoveAll(bundleDir) } + if err := writeSpec(bundleDir, spec); err != nil { + cleanup() + return "", nil, fmt.Errorf("error writing spec: %v", err) } - return bundleDir, nil + return bundleDir, cleanup, nil } // writeSpec writes the spec to disk in the given directory. @@ -185,22 +275,25 @@ func writeSpec(dir string, spec *specs.Spec) error { return ioutil.WriteFile(filepath.Join(dir, "config.json"), b, 0755) } -// UniqueContainerID generates a unique container id for each test. -// -// The container id is used to create an abstract unix domain socket, which must -// be unique. While the container forbids creating two containers with the same -// name, sometimes between test runs the socket does not get cleaned up quickly -// enough, causing container creation to fail. -func UniqueContainerID() string { +// RandomID returns 20 random bytes following the given prefix. +func RandomID(prefix string) string { // Read 20 random bytes. b := make([]byte, 20) // "[Read] always returns len(p) and a nil error." --godoc if _, err := rand.Read(b); err != nil { panic("rand.Read failed: " + err.Error()) } - // base32 encode the random bytes, so that the name is a valid - // container id and can be used as a socket name in the filesystem. - return fmt.Sprintf("test-container-%s", base32.StdEncoding.EncodeToString(b)) + return fmt.Sprintf("%s-%s", prefix, base32.StdEncoding.EncodeToString(b)) +} + +// RandomContainerID generates a random container id for each test. +// +// The container id is used to create an abstract unix domain socket, which +// must be unique. While the container forbids creating two containers with the +// same name, sometimes between test runs the socket does not get cleaned up +// quickly enough, causing container creation to fail. +func RandomContainerID() string { + return RandomID("test-container-") } // Copy copies file from src to dst. @@ -211,12 +304,39 @@ func Copy(src, dst string) error { } defer in.Close() - out, err := os.Create(dst) + st, err := in.Stat() + if err != nil { + return err + } + + out, err := os.OpenFile(dst, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, st.Mode().Perm()) if err != nil { return err } defer out.Close() + // Mirror the local user's permissions across all users. This is + // because as we inject things into the container, the UID/GID will + // change. Also, the build system may generate artifacts with different + // modes. At the top-level (volume mapping) we have a big read-only + // knob that can be applied to prevent modifications. + // + // Note that this must be done via a separate Chmod call, otherwise the + // current process's umask will get in the way. + var mode os.FileMode + if st.Mode()&0100 != 0 { + mode |= 0111 + } + if st.Mode()&0200 != 0 { + mode |= 0222 + } + if st.Mode()&0400 != 0 { + mode |= 0444 + } + if err := os.Chmod(dst, mode); err != nil { + return err + } + _, err = io.Copy(out, in) return err } @@ -239,7 +359,7 @@ func WaitForHTTP(port int, timeout time.Duration) error { url := fmt.Sprintf("http://localhost:%d/", port) resp, err := c.Get(url) if err != nil { - log.Infof("Waiting %s: %v", url, err) + log.Printf("Waiting %s: %v", url, err) return err } resp.Body.Close() @@ -349,6 +469,8 @@ func WaitUntilRead(r io.Reader, want string, split bufio.SplitFunc, timeout time // KillCommand kills the process running cmd unless it hasn't been started. It // returns an error if it cannot kill the process unless the reason is that the // process has already exited. +// +// KillCommand will also reap the process. func KillCommand(cmd *exec.Cmd) error { if cmd.Process == nil { return nil @@ -358,26 +480,21 @@ func KillCommand(cmd *exec.Cmd) error { return fmt.Errorf("failed to kill process %v: %v", cmd, err) } } - return nil + return cmd.Wait() } // WriteTmpFile writes text to a temporary file, closes the file, and returns -// the name of the file. -func WriteTmpFile(pattern, text string) (string, error) { +// the name of the file. A cleanup function is also returned. +func WriteTmpFile(pattern, text string) (string, func(), error) { file, err := ioutil.TempFile(TmpDir(), pattern) if err != nil { - return "", err + return "", nil, err } defer file.Close() if _, err := file.Write([]byte(text)); err != nil { - return "", err + return "", nil, err } - return file.Name(), nil -} - -// RandomName create a name with a 6 digit random number appended to it. -func RandomName(prefix string) string { - return fmt.Sprintf("%s-%06d", prefix, rand.Int31n(1000000)) + return file.Name(), func() { os.RemoveAll(file.Name()) }, nil } // IsStatic returns true iff the given file is a static binary. diff --git a/runsc/testutil/testutil_runfiles.go b/pkg/test/testutil/testutil_runfiles.go index ece9ea9a1..ece9ea9a1 100644 --- a/runsc/testutil/testutil_runfiles.go +++ b/pkg/test/testutil/testutil_runfiles.go diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 5451f1eba..69dcc74f2 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -20,10 +20,10 @@ go_library( "loader_arm64.go", "network.go", "strace.go", - "user.go", "vfs.go", ], visibility = [ + "//pkg/test:__subpackages__", "//runsc:__subpackages__", "//test:__subpackages__", ], @@ -52,6 +52,7 @@ go_library( "//pkg/sentry/fs/sys", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/fs/tty", + "//pkg/sentry/fs/user", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/gofer", "//pkg/sentry/fsimpl/host", @@ -97,7 +98,6 @@ go_library( "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/urpc", - "//pkg/usermem", "//runsc/boot/filter", "//runsc/boot/platforms", "//runsc/boot/pprof", @@ -115,7 +115,6 @@ go_test( "compat_test.go", "fs_test.go", "loader_test.go", - "user_test.go", ], library = ":boot", deps = [ @@ -125,7 +124,6 @@ go_test( "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", "//pkg/sync", "//pkg/unet", "//runsc/fsgofer", diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index cf1f47bc7..096b0e9f0 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -550,11 +551,11 @@ func (l *Loader) run() error { // Add the HOME enviroment variable if it is not already set. var envv []string if kernel.VFS2Enabled { - envv, err = maybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, + envv, err = user.MaybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) } else { - envv, err = maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, + envv, err = user.MaybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) } if err != nil { @@ -860,7 +861,7 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { root := args.MountNamespace.Root() defer root.DecRef() ctx := fs.WithRoot(l.k.SupervisorContext(), root) - envv, err := maybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) + envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) if err != nil { return 0, err } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 82083c57d..bce3a3593 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -251,6 +251,12 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, // All writes go to upper, be paranoid and make lower readonly. opts.ReadOnly = useOverlay + if err := c.k.VFS().MkdirAt(ctx, creds, target, &vfs.MkdirOptions{ + ForSyntheticMountpoint: true, + }); err != nil && err != syserror.EEXIST { + // Log a warning, but attempt the mount anyway. + log.Warningf("Failed to create mount point at %q: %v", submount.Destination, err) + } if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) } diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index d0bb4613a..af3538ef0 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -44,13 +44,13 @@ go_library( "//pkg/sentry/control", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/platform", "//pkg/state", "//pkg/state/statefile", "//pkg/sync", "//pkg/unet", "//pkg/urpc", "//runsc/boot", - "//runsc/boot/platforms", "//runsc/console", "//runsc/container", "//runsc/flag", @@ -82,11 +82,11 @@ go_test( "//pkg/log", "//pkg/sentry/control", "//pkg/sentry/kernel/auth", + "//pkg/test/testutil", "//pkg/urpc", "//runsc/boot", "//runsc/container", "//runsc/specutils", - "//runsc/testutil", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/cmd/boot.go b/runsc/cmd/boot.go index 0938944a6..4c2ac6ff0 100644 --- a/runsc/cmd/boot.go +++ b/runsc/cmd/boot.go @@ -25,8 +25,8 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/runsc/boot" - "gvisor.dev/gvisor/runsc/boot/platforms" "gvisor.dev/gvisor/runsc/flag" "gvisor.dev/gvisor/runsc/specutils" ) @@ -183,7 +183,12 @@ func (b *Boot) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if caps == nil { caps = &specs.LinuxCapabilities{} } - if conf.Platform == platforms.Ptrace { + + gPlatform, err := platform.Lookup(conf.Platform) + if err != nil { + Fatalf("loading platform: %v", err) + } + if gPlatform.Requirements().RequiresCapSysPtrace { // Ptrace platform requires extra capabilities. const c = "CAP_SYS_PTRACE" caps.Bounding = append(caps.Bounding, c) diff --git a/runsc/cmd/capability_test.go b/runsc/cmd/capability_test.go index 9360d7442..a84067112 100644 --- a/runsc/cmd/capability_test.go +++ b/runsc/cmd/capability_test.go @@ -23,10 +23,10 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "github.com/syndtr/gocapability/capability" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/container" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) func init() { @@ -90,16 +90,15 @@ func TestCapabilities(t *testing.T) { // Use --network=host to make sandbox use spec's capabilities. conf.Network = boot.NetworkHost - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := container.Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/container/BUILD b/runsc/container/BUILD index 0aaeea3a8..331b8e866 100644 --- a/runsc/container/BUILD +++ b/runsc/container/BUILD @@ -33,13 +33,15 @@ go_test( size = "large", srcs = [ "console_test.go", + "container_norace_test.go", + "container_race_test.go", "container_test.go", "multi_container_test.go", "shared_volume_test.go", ], data = [ "//runsc", - "//runsc/container/test_app", + "//test/cmd/test_app", ], library = ":container", shard_count = 5, @@ -54,12 +56,12 @@ go_test( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sync", + "//pkg/test/testutil", "//pkg/unet", "//pkg/urpc", "//runsc/boot", "//runsc/boot/platforms", "//runsc/specutils", - "//runsc/testutil", "@com_github_cenkalti_backoff//:go_default_library", "@com_github_kr_pty//:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index af245b6d8..294dca5e7 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -29,9 +29,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" - "gvisor.dev/gvisor/runsc/testutil" ) // socketPath creates a path inside bundleDir and ensures that the returned @@ -58,25 +58,26 @@ func socketPath(bundleDir string) (string, error) { } // createConsoleSocket creates a socket at the given path that will receive a -// console fd from the sandbox. If no error occurs, it returns the server -// socket and a cleanup function. -func createConsoleSocket(path string) (*unet.ServerSocket, func() error, error) { +// console fd from the sandbox. If an error occurs, t.Fatalf will be called. +// The function returning should be deferred as cleanup. +func createConsoleSocket(t *testing.T, path string) (*unet.ServerSocket, func()) { + t.Helper() srv, err := unet.BindAndListen(path, false) if err != nil { - return nil, nil, fmt.Errorf("error binding and listening to socket %q: %v", path, err) + t.Fatalf("error binding and listening to socket %q: %v", path, err) } - cleanup := func() error { + cleanup := func() { + // Log errors; nothing can be done. if err := srv.Close(); err != nil { - return fmt.Errorf("error closing socket %q: %v", path, err) + t.Logf("error closing socket %q: %v", path, err) } if err := os.Remove(path); err != nil { - return fmt.Errorf("error removing socket %q: %v", path, err) + t.Logf("error removing socket %q: %v", path, err) } - return nil } - return srv, cleanup, nil + return srv, cleanup } // receiveConsolePTY accepts a connection on the server socket and reads fds. @@ -118,45 +119,42 @@ func receiveConsolePTY(srv *unet.ServerSocket) (*os.File, error) { // Test that an pty FD is sent over the console socket if one is provided. func TestConsoleSocket(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - spec := testutil.NewSpecWithArgs("true") - 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) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("true") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - sock, err := socketPath(bundleDir) - if err != nil { - t.Fatalf("error getting socket path: %v", err) - } - srv, cleanup, err := createConsoleSocket(sock) - if err != nil { - t.Fatalf("error creating socket at %q: %v", sock, err) - } - defer cleanup() - - // Create the container and pass the socket name. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - ConsoleSocket: sock, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() + sock, err := socketPath(bundleDir) + if err != nil { + t.Fatalf("error getting socket path: %v", err) + } + srv, cleanup := createConsoleSocket(t, sock) + defer cleanup() + + // Create the container and pass the socket name. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + ConsoleSocket: sock, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() - // Make sure we get a console PTY. - ptyMaster, err := receiveConsolePTY(srv) - if err != nil { - t.Fatalf("error receiving console FD: %v", err) - } - ptyMaster.Close() + // Make sure we get a console PTY. + ptyMaster, err := receiveConsolePTY(srv) + if err != nil { + t.Fatalf("error receiving console FD: %v", err) + } + ptyMaster.Close() + }) } } @@ -165,16 +163,15 @@ func TestJobControlSignalExec(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "10000") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -292,26 +289,22 @@ func TestJobControlSignalRootContainer(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/bash", "--noprofile", "--norc") spec.Process.Terminal = true - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() sock, err := socketPath(bundleDir) if err != nil { t.Fatalf("error getting socket path: %v", err) } - srv, cleanup, err := createConsoleSocket(sock) - if err != nil { - t.Fatalf("error creating socket at %q: %v", sock, err) - } + srv, cleanup := createConsoleSocket(t, sock) defer cleanup() // Create the container and pass the socket name. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, ConsoleSocket: sock, @@ -368,7 +361,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { {PID: 1, Cmd: "bash", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(c, expectedPL); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Execute sleep via the terminal. @@ -377,7 +370,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { // Wait for sleep to start. 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) + t.Fatalf("error waiting for processes: %v", err) } // Reset the pty buffer, so there is less output for us to scan later. diff --git a/runsc/container/container.go b/runsc/container/container.go index 7233659b1..117ea7d7b 100644 --- a/runsc/container/container.go +++ b/runsc/container/container.go @@ -274,7 +274,7 @@ func New(conf *boot.Config, args Args) (*Container, error) { } if err := os.MkdirAll(conf.RootDir, 0711); err != nil { - return nil, fmt.Errorf("creating container root directory: %v", err) + return nil, fmt.Errorf("creating container root directory %q: %v", conf.RootDir, err) } c := &Container{ diff --git a/test/root/testdata/httpd.go b/runsc/container/container_norace_test.go index 45d5e33d4..838c1e20a 100644 --- a/test/root/testdata/httpd.go +++ b/runsc/container/container_norace_test.go @@ -12,21 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -package testdata +// +build !race -// Httpd is a JSON config for an httpd container. -const Httpd = ` -{ - "metadata": { - "name": "httpd" - }, - "image":{ - "image": "httpd" - }, - "mounts": [ - ], - "linux": { - }, - "log_path": "httpd.log" -} -` +package container + +// Allow both kvm and ptrace for non-race builds. +var platformOptions = []configOption{ptrace, kvm} diff --git a/test/root/testdata/sandbox.go b/runsc/container/container_race_test.go index 0db210370..9fb4c4fc0 100644 --- a/test/root/testdata/sandbox.go +++ b/runsc/container/container_race_test.go @@ -12,19 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -package testdata +// +build race -// Sandbox is a default JSON config for a sandbox. -const Sandbox = ` -{ - "metadata": { - "name": "default-sandbox", - "namespace": "default", - "attempt": 1, - "uid": "hdishd83djaidwnduwk28bcsb" - }, - "linux": { - }, - "log_directory": "/tmp" -} -` +package container + +// Only enabled ptrace with race builds. +var platformOptions = []configOption{ptrace} diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 5db6d64aa..3ff89f38c 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -39,10 +39,10 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) // waitForProcessList waits for the given process list to show up in the container. @@ -215,16 +215,15 @@ func readOutputNum(file string, position int) (int, error) { // run starts the sandbox and waits for it to exit, checking that the // application succeeded. func run(spec *specs.Spec, conf *boot.Config) error { - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { return fmt.Errorf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create, start and wait for the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, Attached: true, @@ -243,35 +242,41 @@ type configOption int const ( overlay configOption = iota + ptrace kvm nonExclusiveFS ) -var noOverlay = []configOption{kvm, nonExclusiveFS} -var all = append(noOverlay, overlay) +var ( + noOverlay = append(platformOptions, nonExclusiveFS) + all = append(noOverlay, overlay) +) // configs generates different configurations to run tests. -func configs(t *testing.T, opts ...configOption) []*boot.Config { +func configs(t *testing.T, opts ...configOption) map[string]*boot.Config { // Always load the default config. - cs := []*boot.Config{testutil.TestConfig(t)} - + cs := make(map[string]*boot.Config) for _, o := range opts { - c := testutil.TestConfig(t) switch o { case overlay: + c := testutil.TestConfig(t) c.Overlay = true + cs["overlay"] = c + case ptrace: + c := testutil.TestConfig(t) + c.Platform = platforms.Ptrace + cs["ptrace"] = c case kvm: - // TODO(b/112165693): KVM tests are flaky. Disable until fixed. - continue - + c := testutil.TestConfig(t) c.Platform = platforms.KVM + cs["kvm"] = c case nonExclusiveFS: + c := testutil.TestConfig(t) c.FileAccess = boot.FileAccessShared + cs["non-exclusive"] = c default: panic(fmt.Sprintf("unknown config option %v", o)) - } - cs = append(cs, c) } return cs } @@ -285,133 +290,133 @@ func TestLifecycle(t *testing.T) { childReaper.Start() defer childReaper.Stop() - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - // The container will just sleep for a long time. We will kill it before - // it finishes sleeping. - spec := testutil.NewSpecWithArgs("sleep", "100") + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + // The container will just sleep for a long time. We will kill it before + // it finishes sleeping. + spec := testutil.NewSpecWithArgs("sleep", "100") - 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) - - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - } - // Create 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() - - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Created; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // List should return the container id. - ids, err := List(rootDir) - if err != nil { - t.Fatalf("error listing containers: %v", err) - } - if got, want := ids, []string{args.ID}; !reflect.DeepEqual(got, want) { - t.Errorf("container list got %v, want %v", got, want) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + } + // Create the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() - // Start the container. - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) + if err != nil { + t.Fatalf("error loading container: %v", err) + } + if got, want := c.Status, Created; got != want { + t.Errorf("container status got %v, want %v", got, want) + } - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Running; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + // List should return the container id. + ids, err := List(rootDir) + if err != nil { + t.Fatalf("error listing containers: %v", err) + } + if got, want := ids, []string{args.ID}; !reflect.DeepEqual(got, want) { + t.Errorf("container list got %v, want %v", got, want) + } - // Verify that "sleep 100" is running. - if err := waitForProcessList(c, expectedPL); err != nil { - t.Error(err) - } + // Start the container. + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Wait on the container. - var wg sync.WaitGroup - wg.Add(1) - ch := make(chan struct{}) - go func() { - ch <- struct{}{} - ws, err := c.Wait() + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) if err != nil { - t.Fatalf("error waiting on container: %v", err) + t.Fatalf("error loading container: %v", err) } - if got, want := ws.Signal(), syscall.SIGTERM; got != want { - t.Fatalf("got signal %v, want %v", got, want) + if got, want := c.Status, Running; got != want { + t.Errorf("container status got %v, want %v", got, want) } - wg.Done() - }() - // Wait a bit to ensure that we've started waiting on the - // container before we signal. - <-ch - time.Sleep(100 * time.Millisecond) - // Send the container a SIGTERM which will cause it to stop. - if err := c.SignalContainer(syscall.SIGTERM, false); err != nil { - t.Fatalf("error sending signal %v to container: %v", syscall.SIGTERM, err) - } - // Wait for it to die. - wg.Wait() + // Verify that "sleep 100" is running. + if err := waitForProcessList(c, expectedPL); err != nil { + t.Error(err) + } - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Stopped; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + // Wait on the container. + ch := make(chan error) + go func() { + ws, err := c.Wait() + if err != nil { + ch <- err + } + if got, want := ws.Signal(), syscall.SIGTERM; got != want { + ch <- fmt.Errorf("got signal %v, want %v", got, want) + } + ch <- nil + }() - // Destroy the container. - if err := c.Destroy(); err != nil { - t.Fatalf("error destroying container: %v", err) - } + // Wait a bit to ensure that we've started waiting on + // the container before we signal. + time.Sleep(time.Second) - // List should not return the container id. - ids, err = List(rootDir) - if err != nil { - t.Fatalf("error listing containers: %v", err) - } - if len(ids) != 0 { - t.Errorf("expected container list to be empty, but got %v", ids) - } + // Send the container a SIGTERM which will cause it to stop. + if err := c.SignalContainer(syscall.SIGTERM, false); err != nil { + t.Fatalf("error sending signal %v to container: %v", syscall.SIGTERM, err) + } - // Loading the container by id should fail. - if _, err = Load(rootDir, args.ID); err == nil { - t.Errorf("expected loading destroyed container to fail, but it did not") - } + // Wait for it to die. + if err := <-ch; err != nil { + t.Fatalf("error waiting for container: %v", err) + } + + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) + if err != nil { + t.Fatalf("error loading container: %v", err) + } + if got, want := c.Status, Stopped; got != want { + t.Errorf("container status got %v, want %v", got, want) + } + + // Destroy the container. + if err := c.Destroy(); err != nil { + t.Fatalf("error destroying container: %v", err) + } + + // List should not return the container id. + ids, err = List(rootDir) + if err != nil { + t.Fatalf("error listing containers: %v", err) + } + if len(ids) != 0 { + t.Errorf("expected container list to be empty, but got %v", ids) + } + + // Loading the container by id should fail. + if _, err = Load(rootDir, args.ID); err == nil { + t.Errorf("expected loading destroyed container to fail, but it did not") + } + }) } } @@ -420,12 +425,14 @@ func TestExePath(t *testing.T) { // Create two directories that will be prepended to PATH. firstPath, err := ioutil.TempDir(testutil.TmpDir(), "first") if err != nil { - t.Fatal(err) + t.Fatalf("error creating temporary directory: %v", err) } + defer os.RemoveAll(firstPath) secondPath, err := ioutil.TempDir(testutil.TmpDir(), "second") if err != nil { - t.Fatal(err) + t.Fatalf("error creating temporary directory: %v", err) } + defer os.RemoveAll(secondPath) // Create two minimal executables in the second path, two of which // will be masked by files in first path. @@ -433,11 +440,11 @@ func TestExePath(t *testing.T) { path := filepath.Join(secondPath, p) f, err := os.OpenFile(path, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0777) if err != nil { - t.Fatal(err) + t.Fatalf("error opening path: %v", err) } defer f.Close() if _, err := io.WriteString(f, "#!/bin/true\n"); err != nil { - t.Fatal(err) + t.Fatalf("error writing contents: %v", err) } } @@ -446,7 +453,7 @@ func TestExePath(t *testing.T) { nonExecutable := filepath.Join(firstPath, "masked1") f2, err := os.OpenFile(nonExecutable, os.O_CREATE|os.O_EXCL, 0666) if err != nil { - t.Fatal(err) + t.Fatalf("error opening file: %v", err) } f2.Close() @@ -454,68 +461,69 @@ func TestExePath(t *testing.T) { // executable in the second. nonRegular := filepath.Join(firstPath, "masked2") if err := os.Mkdir(nonRegular, 0777); err != nil { - t.Fatal(err) - } - - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - for _, test := range []struct { - path string - success bool - }{ - {path: "true", success: true}, - {path: "bin/true", success: true}, - {path: "/bin/true", success: true}, - {path: "thisfiledoesntexit", success: false}, - {path: "bin/thisfiledoesntexit", success: false}, - {path: "/bin/thisfiledoesntexit", success: false}, - - {path: "unmasked", success: true}, - {path: filepath.Join(firstPath, "unmasked"), success: false}, - {path: filepath.Join(secondPath, "unmasked"), success: true}, - - {path: "masked1", success: true}, - {path: filepath.Join(firstPath, "masked1"), success: false}, - {path: filepath.Join(secondPath, "masked1"), success: true}, - - {path: "masked2", success: true}, - {path: filepath.Join(firstPath, "masked2"), success: false}, - {path: filepath.Join(secondPath, "masked2"), success: true}, - } { - spec := testutil.NewSpecWithArgs(test.path) - spec.Process.Env = []string{ - fmt.Sprintf("PATH=%s:%s:%s", firstPath, secondPath, os.Getenv("PATH")), - } - - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("exec: %s, error setting up container: %v", test.path, err) - } - - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - Attached: true, - } - ws, err := Run(conf, args) + t.Fatalf("error making directory: %v", err) + } + + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + for _, test := range []struct { + path string + success bool + }{ + {path: "true", success: true}, + {path: "bin/true", success: true}, + {path: "/bin/true", success: true}, + {path: "thisfiledoesntexit", success: false}, + {path: "bin/thisfiledoesntexit", success: false}, + {path: "/bin/thisfiledoesntexit", success: false}, + + {path: "unmasked", success: true}, + {path: filepath.Join(firstPath, "unmasked"), success: false}, + {path: filepath.Join(secondPath, "unmasked"), success: true}, + + {path: "masked1", success: true}, + {path: filepath.Join(firstPath, "masked1"), success: false}, + {path: filepath.Join(secondPath, "masked1"), success: true}, + + {path: "masked2", success: true}, + {path: filepath.Join(firstPath, "masked2"), success: false}, + {path: filepath.Join(secondPath, "masked2"), success: true}, + } { + t.Run(fmt.Sprintf("path=%s,success=%t", test.path, test.success), func(t *testing.T) { + spec := testutil.NewSpecWithArgs(test.path) + spec.Process.Env = []string{ + fmt.Sprintf("PATH=%s:%s:%s", firstPath, secondPath, os.Getenv("PATH")), + } - os.RemoveAll(rootDir) - os.RemoveAll(bundleDir) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("exec: error setting up container: %v", err) + } + defer cleanup() - if test.success { - if err != nil { - t.Errorf("exec: %s, error running container: %v", test.path, err) - } - if ws.ExitStatus() != 0 { - t.Errorf("exec: %s, got exit status %v want %v", test.path, ws.ExitStatus(), 0) - } - } else { - if err == nil { - t.Errorf("exec: %s, got: no error, want: error", test.path) - } + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + Attached: true, + } + ws, err := Run(conf, args) + + if test.success { + if err != nil { + t.Errorf("exec: error running container: %v", err) + } + if ws.ExitStatus() != 0 { + t.Errorf("exec: got exit status %v want %v", ws.ExitStatus(), 0) + } + } else { + if err == nil { + t.Errorf("exec: got: no error, want: error") + } + } + }) } - } + }) } } @@ -534,15 +542,14 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { succSpec := testutil.NewSpecWithArgs("true") conf := testutil.TestConfig(t) conf.VFS2 = vfs2 - rootDir, bundleDir, err := testutil.SetupContainer(succSpec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(succSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: succSpec, BundleDir: bundleDir, Attached: true, @@ -559,15 +566,14 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { wantStatus := 123 errSpec := testutil.NewSpecWithArgs("bash", "-c", fmt.Sprintf("exit %d", wantStatus)) - rootDir2, bundleDir2, err := testutil.SetupContainer(errSpec, conf) + _, bundleDir2, cleanup2, err := testutil.SetupContainer(errSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir2) - defer os.RemoveAll(bundleDir2) + defer cleanup2() args2 := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: errSpec, BundleDir: bundleDir2, Attached: true, @@ -583,166 +589,163 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { // TestExec verifies that a container can exec a new program. func TestExec(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + const uid = 343 + spec := testutil.NewSpecWithArgs("sleep", "100") - const uid = 343 - spec := testutil.NewSpecWithArgs("sleep", "100") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - 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.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + { + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{2}, + }, + } - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{2}, - }, - } + // Verify that "sleep 100" is running. + if err := waitForProcessList(cont, expectedPL[:1]); err != nil { + t.Error(err) + } - // Verify that "sleep 100" is running. - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { - t.Error(err) - } + execArgs := &control.ExecArgs{ + Filename: "/bin/sleep", + Argv: []string{"/bin/sleep", "5"}, + WorkingDirectory: "/", + KUID: uid, + } - execArgs := &control.ExecArgs{ - Filename: "/bin/sleep", - Argv: []string{"/bin/sleep", "5"}, - WorkingDirectory: "/", - KUID: uid, - } + // Verify that "sleep 100" and "sleep 5" are running + // after exec. First, start running exec (whick + // blocks). + ch := make(chan error) + go func() { + exitStatus, err := cont.executeSync(execArgs) + if err != nil { + ch <- err + } else if exitStatus != 0 { + ch <- fmt.Errorf("failed with exit status: %v", exitStatus) + } else { + ch <- nil + } + }() - // Verify that "sleep 100" and "sleep 5" are running after exec. - // First, start running exec (whick blocks). - status := make(chan error, 1) - go func() { - exitStatus, err := cont.executeSync(execArgs) - if err != nil { - log.Debugf("error executing: %v", err) - status <- err - } else if exitStatus != 0 { - log.Debugf("bad status: %d", exitStatus) - status <- fmt.Errorf("failed with exit status: %v", exitStatus) - } else { - status <- nil + if err := waitForProcessList(cont, expectedPL); err != nil { + t.Fatalf("error waiting for processes: %v", err) } - }() - if err := waitForProcessList(cont, expectedPL); err != nil { - t.Fatal(err) - } - - // Ensure that exec finished without error. - select { - case <-time.After(10 * time.Second): - t.Fatalf("container timed out waiting for exec to finish.") - case st := <-status: - if st != nil { - t.Errorf("container failed to exec %v: %v", args, err) + // Ensure that exec finished without error. + select { + case <-time.After(10 * time.Second): + t.Fatalf("container timed out waiting for exec to finish.") + case err := <-ch: + if err != nil { + t.Errorf("container failed to exec %v: %v", args, err) + } } - } + }) } } // TestKillPid verifies that we can signal individual exec'd processes. func TestKillPid(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - app, err := testutil.FindFile("runsc/container/test_app/test_app") - if err != nil { - t.Fatal("error finding test_app:", err) - } + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + app, err := testutil.FindFile("test/cmd/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } - const nProcs = 4 - spec := testutil.NewSpecWithArgs(app, "task-tree", "--depth", strconv.Itoa(nProcs-1), "--width=1", "--pause=true") - 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) + const nProcs = 4 + spec := testutil.NewSpecWithArgs(app, "task-tree", "--depth", strconv.Itoa(nProcs-1), "--width=1", "--pause=true") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Verify that all processes are running. - if err := waitForProcessCount(cont, nProcs); err != nil { - t.Fatalf("timed out waiting for processes to start: %v", err) - } + // Verify that all processes are running. + if err := waitForProcessCount(cont, nProcs); err != nil { + t.Fatalf("timed out waiting for processes to start: %v", err) + } - // Kill the child process with the largest PID. - procs, err := cont.Processes() - if err != nil { - t.Fatalf("failed to get process list: %v", err) - } - var pid int32 - for _, p := range procs { - if pid < int32(p.PID) { - pid = int32(p.PID) + // Kill the child process with the largest PID. + procs, err := cont.Processes() + if err != nil { + t.Fatalf("failed to get process list: %v", err) + } + var pid int32 + for _, p := range procs { + if pid < int32(p.PID) { + pid = int32(p.PID) + } + } + if err := cont.SignalProcess(syscall.SIGKILL, pid); err != nil { + t.Fatalf("failed to signal process %d: %v", pid, err) } - } - if err := cont.SignalProcess(syscall.SIGKILL, pid); err != nil { - t.Fatalf("failed to signal process %d: %v", pid, err) - } - // Verify that one process is gone. - if err := waitForProcessCount(cont, nProcs-1); err != nil { - t.Fatal(err) - } + // Verify that one process is gone. + if err := waitForProcessCount(cont, nProcs-1); err != nil { + t.Fatalf("error waiting for processes: %v", err) + } - procs, err = cont.Processes() - if err != nil { - t.Fatalf("failed to get process list: %v", err) - } - for _, p := range procs { - if pid == int32(p.PID) { - t.Fatalf("pid %d is still alive, which should be killed", pid) + procs, err = cont.Processes() + if err != nil { + t.Fatalf("failed to get process list: %v", err) } - } + for _, p := range procs { + if pid == int32(p.PID) { + t.Fatalf("pid %d is still alive, which should be killed", pid) + } + } + }) } } @@ -753,160 +756,160 @@ func TestKillPid(t *testing.T) { // be the next consecutive number after the last number from the checkpointed container. func TestCheckpointRestore(t *testing.T) { // Skip overlay because test requires writing to host file. - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "checkpoint-test") - if err != nil { - t.Fatalf("ioutil.TempDir failed: %v", err) - } - if err := os.Chmod(dir, 0777); err != nil { - t.Fatalf("error chmoding file: %q, %v", dir, err) - } + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "checkpoint-test") + if err != nil { + t.Fatalf("ioutil.TempDir failed: %v", err) + } + defer os.RemoveAll(dir) + if err := os.Chmod(dir, 0777); err != nil { + t.Fatalf("error chmoding file: %q, %v", dir, err) + } - outputPath := filepath.Join(dir, "output") - outputFile, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile.Close() + outputPath := filepath.Join(dir, "output") + outputFile, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile.Close() - script := fmt.Sprintf("for ((i=0; ;i++)); do echo $i >> %q; sleep 1; done", outputPath) - spec := testutil.NewSpecWithArgs("bash", "-c", script) - 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) + script := fmt.Sprintf("for ((i=0; ;i++)); do echo $i >> %q; sleep 1; done", outputPath) + spec := testutil.NewSpecWithArgs("bash", "-c", script) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Set the image path, which is where the checkpoint image will be saved. - imagePath := filepath.Join(dir, "test-image-file") + // Set the image path, which is where the checkpoint image will be saved. + imagePath := filepath.Join(dir, "test-image-file") - // Create the image file and open for writing. - file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) - if err != nil { - t.Fatalf("error opening new file at imagePath: %v", err) - } - defer file.Close() + // Create the image file and open for writing. + file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) + if err != nil { + t.Fatalf("error opening new file at imagePath: %v", err) + } + defer file.Close() - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Checkpoint running container; save state into new file. - if err := cont.Checkpoint(file); err != nil { - t.Fatalf("error checkpointing container to empty file: %v", err) - } - defer os.RemoveAll(imagePath) + // Checkpoint running container; save state into new file. + if err := cont.Checkpoint(file); err != nil { + t.Fatalf("error checkpointing container to empty file: %v", err) + } + defer os.RemoveAll(imagePath) - lastNum, err := readOutputNum(outputPath, -1) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + lastNum, err := readOutputNum(outputPath, -1) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile2, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile2.Close() + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile2, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile2.Close() - // Restore into a new container. - args2 := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont2, err := New(conf, args2) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont2.Destroy() + // Restore into a new container. + args2 := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont2, err := New(conf, args2) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont2.Destroy() - if err := cont2.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := cont2.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile2); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile2); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - firstNum, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + firstNum, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum and that the container picks - // up from where it left off. - if lastNum+1 != firstNum { - t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum) - } - cont2.Destroy() + // Check that lastNum is one less than firstNum and that the container picks + // up from where it left off. + if lastNum+1 != firstNum { + t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum) + } + cont2.Destroy() - // Restore into another container! - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile3, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile3.Close() + // Restore into another container! + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile3, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile3.Close() - // Restore into a new container. - args3 := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont3, err := New(conf, args3) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont3.Destroy() + // Restore into a new container. + args3 := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont3, err := New(conf, args3) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont3.Destroy() - if err := cont3.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := cont3.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile3); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile3); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - firstNum2, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + firstNum2, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum and that the container picks - // up from where it left off. - if lastNum+1 != firstNum2 { - t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum2) - } - cont3.Destroy() + // Check that lastNum is one less than firstNum and that the container picks + // up from where it left off. + if lastNum+1 != firstNum2 { + t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum2) + } + cont3.Destroy() + }) } } @@ -914,135 +917,134 @@ func TestCheckpointRestore(t *testing.T) { // with filesystem Unix Domain Socket use. func TestUnixDomainSockets(t *testing.T) { // Skip overlay because test requires writing to host file. - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - // UDS path is limited to 108 chars for compatibility with older systems. - // Use '/tmp' (instead of testutil.TmpDir) to ensure the size limit is - // not exceeded. Assumes '/tmp' exists in the system. - dir, err := ioutil.TempDir("/tmp", "uds-test") - if err != nil { - t.Fatalf("ioutil.TempDir failed: %v", err) - } - defer os.RemoveAll(dir) + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + // UDS path is limited to 108 chars for compatibility with older systems. + // Use '/tmp' (instead of testutil.TmpDir) to ensure the size limit is + // not exceeded. Assumes '/tmp' exists in the system. + dir, err := ioutil.TempDir("/tmp", "uds-test") + if err != nil { + t.Fatalf("ioutil.TempDir failed: %v", err) + } + defer os.RemoveAll(dir) - outputPath := filepath.Join(dir, "uds_output") - outputFile, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile.Close() + outputPath := filepath.Join(dir, "uds_output") + outputFile, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile.Close() - app, err := testutil.FindFile("runsc/container/test_app/test_app") - if err != nil { - t.Fatal("error finding test_app:", err) - } + app, err := testutil.FindFile("test/cmd/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } - socketPath := filepath.Join(dir, "uds_socket") - defer os.Remove(socketPath) + socketPath := filepath.Join(dir, "uds_socket") + defer os.Remove(socketPath) - spec := testutil.NewSpecWithArgs(app, "uds", "--file", outputPath, "--socket", socketPath) - spec.Process.User = specs.User{ - UID: uint32(os.Getuid()), - GID: uint32(os.Getgid()), - } - spec.Mounts = []specs.Mount{{ - Type: "bind", - Destination: dir, - Source: dir, - }} + spec := testutil.NewSpecWithArgs(app, "uds", "--file", outputPath, "--socket", socketPath) + spec.Process.User = specs.User{ + UID: uint32(os.Getuid()), + GID: uint32(os.Getgid()), + } + spec.Mounts = []specs.Mount{{ + Type: "bind", + Destination: dir, + Source: dir, + }} - 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) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Set the image path, the location where the checkpoint image will be saved. - imagePath := filepath.Join(dir, "test-image-file") + // Set the image path, the location where the checkpoint image will be saved. + imagePath := filepath.Join(dir, "test-image-file") - // Create the image file and open for writing. - file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) - if err != nil { - t.Fatalf("error opening new file at imagePath: %v", err) - } - defer file.Close() - defer os.RemoveAll(imagePath) + // Create the image file and open for writing. + file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) + if err != nil { + t.Fatalf("error opening new file at imagePath: %v", err) + } + defer file.Close() + defer os.RemoveAll(imagePath) - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Checkpoint running container; save state into new file. - if err := cont.Checkpoint(file); err != nil { - t.Fatalf("error checkpointing container to empty file: %v", err) - } + // Checkpoint running container; save state into new file. + if err := cont.Checkpoint(file); err != nil { + t.Fatalf("error checkpointing container to empty file: %v", err) + } - // Read last number outputted before checkpoint. - lastNum, err := readOutputNum(outputPath, -1) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + // Read last number outputted before checkpoint. + lastNum, err := readOutputNum(outputPath, -1) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile2, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile2.Close() + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile2, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile2.Close() - // Restore into a new container. - argsRestore := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - contRestore, err := New(conf, argsRestore) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer contRestore.Destroy() + // Restore into a new container. + argsRestore := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + contRestore, err := New(conf, argsRestore) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer contRestore.Destroy() - if err := contRestore.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := contRestore.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile2); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile2); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Read first number outputted after restore. - firstNum, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + // Read first number outputted after restore. + firstNum, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum. - if lastNum+1 != firstNum { - t.Errorf("error numbers not consecutive, previous: %d, next: %d", lastNum, firstNum) - } - contRestore.Destroy() + // Check that lastNum is one less than firstNum. + if lastNum+1 != firstNum { + t.Errorf("error numbers not consecutive, previous: %d, next: %d", lastNum, firstNum) + } + contRestore.Destroy() + }) } } @@ -1052,10 +1054,8 @@ func TestUnixDomainSockets(t *testing.T) { // recreated. Then it resumes the container, verify that the file gets created // again. func TestPauseResume(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Run(fmt.Sprintf("conf: %+v", conf), func(t *testing.T) { - t.Logf("Running test with conf: %+v", conf) - + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "lock") if err != nil { t.Fatalf("error creating temp dir: %v", err) @@ -1066,16 +1066,15 @@ func TestPauseResume(t *testing.T) { script := fmt.Sprintf("while [[ true ]]; do touch %q; sleep 0.1; done", running) spec := testutil.NewSpecWithArgs("/bin/bash", "-c", script) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1134,16 +1133,15 @@ func TestPauseResume(t *testing.T) { func TestPauseResumeStatus(t *testing.T) { spec := testutil.NewSpecWithArgs("sleep", "20") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1199,359 +1197,356 @@ func TestCapabilities(t *testing.T) { uid := auth.KUID(os.Getuid() + 1) gid := auth.KGID(os.Getgid() + 1) - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("sleep", "100") - 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) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("sleep", "100") + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "exe", - Threads: []kernel.ThreadID{2}, - }, - } - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { - t.Fatalf("Failed to wait for sleep to start, err: %v", err) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + { + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "exe", + Threads: []kernel.ThreadID{2}, + }, + } + if err := waitForProcessList(cont, expectedPL[:1]); err != nil { + t.Fatalf("Failed to wait for sleep to start, err: %v", err) + } - // Create an executable that can't be run with the specified UID:GID. - // This shouldn't be callable within the container until we add the - // CAP_DAC_OVERRIDE capability to skip the access check. - exePath := filepath.Join(rootDir, "exe") - if err := ioutil.WriteFile(exePath, []byte("#!/bin/sh\necho hello"), 0770); err != nil { - t.Fatalf("couldn't create executable: %v", err) - } - defer os.Remove(exePath) - - // Need to traverse the intermediate directory. - os.Chmod(rootDir, 0755) - - execArgs := &control.ExecArgs{ - Filename: exePath, - Argv: []string{exePath}, - WorkingDirectory: "/", - KUID: uid, - KGID: gid, - Capabilities: &auth.TaskCapabilities{}, - } + // Create an executable that can't be run with the specified UID:GID. + // This shouldn't be callable within the container until we add the + // CAP_DAC_OVERRIDE capability to skip the access check. + exePath := filepath.Join(rootDir, "exe") + if err := ioutil.WriteFile(exePath, []byte("#!/bin/sh\necho hello"), 0770); err != nil { + t.Fatalf("couldn't create executable: %v", err) + } + defer os.Remove(exePath) + + // Need to traverse the intermediate directory. + os.Chmod(rootDir, 0755) + + execArgs := &control.ExecArgs{ + Filename: exePath, + Argv: []string{exePath}, + WorkingDirectory: "/", + KUID: uid, + KGID: gid, + Capabilities: &auth.TaskCapabilities{}, + } - // "exe" should fail because we don't have the necessary permissions. - if _, err := cont.executeSync(execArgs); err == nil { - t.Fatalf("container executed without error, but an error was expected") - } + // "exe" should fail because we don't have the necessary permissions. + if _, err := cont.executeSync(execArgs); err == nil { + t.Fatalf("container executed without error, but an error was expected") + } - // Now we run with the capability enabled and should succeed. - execArgs.Capabilities = &auth.TaskCapabilities{ - EffectiveCaps: auth.CapabilitySetOf(linux.CAP_DAC_OVERRIDE), - } - // "exe" should not fail this time. - if _, err := cont.executeSync(execArgs); err != nil { - t.Fatalf("container failed to exec %v: %v", args, err) - } + // Now we run with the capability enabled and should succeed. + execArgs.Capabilities = &auth.TaskCapabilities{ + EffectiveCaps: auth.CapabilitySetOf(linux.CAP_DAC_OVERRIDE), + } + // "exe" should not fail this time. + if _, err := cont.executeSync(execArgs); err != nil { + t.Fatalf("container failed to exec %v: %v", args, err) + } + }) } } // TestRunNonRoot checks that sandbox can be configured when running as // non-privileged user. func TestRunNonRoot(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("/bin/true") - - // Set a random user/group with no access to "blocked" dir. - spec.Process.User.UID = 343 - spec.Process.User.GID = 2401 - spec.Process.Capabilities = nil + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("/bin/true") + + // Set a random user/group with no access to "blocked" dir. + spec.Process.User.UID = 343 + spec.Process.User.GID = 2401 + spec.Process.Capabilities = nil + + // User running inside container can't list '$TMP/blocked' and would fail to + // mount it. + dir, err := ioutil.TempDir(testutil.TmpDir(), "blocked") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + if err := os.Chmod(dir, 0700); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", dir, err) + } + dir = path.Join(dir, "test") + if err := os.Mkdir(dir, 0755); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", dir, err) + } - // User running inside container can't list '$TMP/blocked' and would fail to - // mount it. - dir, err := ioutil.TempDir(testutil.TmpDir(), "blocked") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - if err := os.Chmod(dir, 0700); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", dir, err) - } - dir = path.Join(dir, "test") - if err := os.Mkdir(dir, 0755); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", dir, err) - } + src, err := ioutil.TempDir(testutil.TmpDir(), "src") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } - src, err := ioutil.TempDir(testutil.TmpDir(), "src") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: dir, + Source: src, + Type: "bind", + }) - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: dir, - Source: src, - Type: "bind", + if err := run(spec, conf); err != nil { + t.Fatalf("error running sandbox: %v", err) + } }) - - if err := run(spec, conf); err != nil { - t.Fatalf("error running sandbox: %v", err) - } } } // TestMountNewDir checks that runsc will create destination directory if it // doesn't exit. func TestMountNewDir(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + root, err := ioutil.TempDir(testutil.TmpDir(), "root") + if err != nil { + t.Fatal("ioutil.TempDir() failed:", err) + } - root, err := ioutil.TempDir(testutil.TmpDir(), "root") - if err != nil { - t.Fatal("ioutil.TempDir() failed:", err) - } + srcDir := path.Join(root, "src", "dir", "anotherdir") + if err := os.MkdirAll(srcDir, 0755); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", srcDir, err) + } - srcDir := path.Join(root, "src", "dir", "anotherdir") - if err := os.MkdirAll(srcDir, 0755); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", srcDir, err) - } + mountDir := path.Join(root, "dir", "anotherdir") - mountDir := path.Join(root, "dir", "anotherdir") + spec := testutil.NewSpecWithArgs("/bin/ls", mountDir) + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: mountDir, + Source: srcDir, + Type: "bind", + }) - spec := testutil.NewSpecWithArgs("/bin/ls", mountDir) - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: mountDir, - Source: srcDir, - Type: "bind", + if err := run(spec, conf); err != nil { + t.Fatalf("error running sandbox: %v", err) + } }) - - if err := run(spec, conf); err != nil { - t.Fatalf("error running sandbox: %v", err) - } } } func TestReadonlyRoot(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("/bin/touch", "/foo") - spec.Root.Readonly = true - 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) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("/bin/touch", "/foo") + spec.Root.Readonly = true + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for 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) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + 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) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { - t.Fatalf("container failed, waitStatus: %v", ws) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { + t.Fatalf("container failed, waitStatus: %v", ws) + } + }) } } func TestUIDMap(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - testDir, err := ioutil.TempDir(testutil.TmpDir(), "test-mount") - if err != nil { - t.Fatal(err) - } - defer os.RemoveAll(testDir) - testFile := path.Join(testDir, "testfile") - - spec := testutil.NewSpecWithArgs("touch", "/tmp/testfile") - uid := os.Getuid() - gid := os.Getgid() - spec.Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - {Type: specs.UserNamespace}, - {Type: specs.PIDNamespace}, - {Type: specs.MountNamespace}, - }, - UIDMappings: []specs.LinuxIDMapping{ - { - ContainerID: 0, - HostID: uint32(uid), - Size: 1, + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + testDir, err := ioutil.TempDir(testutil.TmpDir(), "test-mount") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + defer os.RemoveAll(testDir) + testFile := path.Join(testDir, "testfile") + + spec := testutil.NewSpecWithArgs("touch", "/tmp/testfile") + uid := os.Getuid() + gid := os.Getgid() + spec.Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + {Type: specs.UserNamespace}, + {Type: specs.PIDNamespace}, + {Type: specs.MountNamespace}, }, - }, - GIDMappings: []specs.LinuxIDMapping{ - { - ContainerID: 0, - HostID: uint32(gid), - Size: 1, + UIDMappings: []specs.LinuxIDMapping{ + { + ContainerID: 0, + HostID: uint32(uid), + Size: 1, + }, }, - }, - } + GIDMappings: []specs.LinuxIDMapping{ + { + ContainerID: 0, + HostID: uint32(gid), + Size: 1, + }, + }, + } - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: "/tmp", - Source: testDir, - Type: "bind", - }) + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: "/tmp", + Source: testDir, + Type: "bind", + }) - 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) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for 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) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + 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) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || ws.ExitStatus() != 0 { - t.Fatalf("container failed, waitStatus: %v", ws) - } - st := syscall.Stat_t{} - if err := syscall.Stat(testFile, &st); err != nil { - t.Fatalf("error stat /testfile: %v", err) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || ws.ExitStatus() != 0 { + t.Fatalf("container failed, waitStatus: %v", ws) + } + st := syscall.Stat_t{} + if err := syscall.Stat(testFile, &st); err != nil { + t.Fatalf("error stat /testfile: %v", err) + } - if st.Uid != uint32(uid) || st.Gid != uint32(gid) { - t.Fatalf("UID: %d (%d) GID: %d (%d)", st.Uid, uid, st.Gid, gid) - } + if st.Uid != uint32(uid) || st.Gid != uint32(gid) { + t.Fatalf("UID: %d (%d) GID: %d (%d)", st.Uid, uid, st.Gid, gid) + } + }) } } func TestReadonlyMount(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "ro-mount") - spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: dir, - Source: dir, - Type: "bind", - Options: []string{"ro"}, - }) - spec.Root.Readonly = false - - 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) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "ro-mount") + spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: dir, + Source: dir, + Type: "bind", + Options: []string{"ro"}, + }) + spec.Root.Readonly = false + + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for 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) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + 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) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { - t.Fatalf("container failed, waitStatus: %v", ws) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { + t.Fatalf("container failed, waitStatus: %v", ws) + } + }) } } // TestAbbreviatedIDs checks that runsc supports using abbreviated container // IDs in place of full IDs. func TestAbbreviatedIDs(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir cids := []string{ - "foo-" + testutil.UniqueContainerID(), - "bar-" + testutil.UniqueContainerID(), - "baz-" + testutil.UniqueContainerID(), + "foo-" + testutil.RandomContainerID(), + "bar-" + testutil.RandomContainerID(), + "baz-" + testutil.RandomContainerID(), } for _, cid := range cids { spec := testutil.NewSpecWithArgs("sleep", "100") - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ @@ -1596,16 +1591,15 @@ func TestAbbreviatedIDs(t *testing.T) { func TestGoferExits(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "10000") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1634,7 +1628,7 @@ func TestGoferExits(t *testing.T) { } func TestRootNotMount(t *testing.T) { - appSym, err := testutil.FindFile("runsc/container/test_app/test_app") + appSym, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1671,7 +1665,7 @@ func TestRootNotMount(t *testing.T) { } func TestUserLog(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1679,12 +1673,11 @@ func TestUserLog(t *testing.T) { // sched_rr_get_interval = 148 - not implemented in gvisor. spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall=148") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() dir, err := ioutil.TempDir(testutil.TmpDir(), "user_log_test") if err != nil { @@ -1694,7 +1687,7 @@ func TestUserLog(t *testing.T) { // Create, start and wait for the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, UserLog: userLog, @@ -1718,72 +1711,70 @@ func TestUserLog(t *testing.T) { } func TestWaitOnExitedSandbox(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - // Run a shell that sleeps for 1 second and then exits with a - // non-zero code. - const wantExit = 17 - cmd := fmt.Sprintf("sleep 1; exit %d", wantExit) - spec := testutil.NewSpecWithArgs("/bin/sh", "-c", 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) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + // Run a shell that sleeps for 1 second and then exits with a + // non-zero code. + const wantExit = 17 + cmd := fmt.Sprintf("sleep 1; exit %d", wantExit) + spec := testutil.NewSpecWithArgs("/bin/sh", "-c", cmd) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // 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) - } + // Create and Start the container. + args := Args{ + ID: testutil.RandomContainerID(), + 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 on the sandbox. This will make an RPC to the sandbox - // and get the actual exit status of the application. - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if got := ws.ExitStatus(); got != wantExit { - t.Errorf("got exit status %d, want %d", got, wantExit) - } + // Wait on the sandbox. This will make an RPC to the sandbox + // and get the actual exit status of the application. + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if got := ws.ExitStatus(); got != wantExit { + t.Errorf("got exit status %d, want %d", got, wantExit) + } - // Now the sandbox has exited, but the zombie sandbox process - // still exists. Calling Wait() now will return the sandbox - // exit status. - ws, err = c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if got := ws.ExitStatus(); got != wantExit { - t.Errorf("got exit status %d, want %d", got, wantExit) - } + // Now the sandbox has exited, but the zombie sandbox process + // still exists. Calling Wait() now will return the sandbox + // exit status. + ws, err = c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if got := ws.ExitStatus(); got != wantExit { + t.Errorf("got exit status %d, want %d", got, wantExit) + } + }) } } func TestDestroyNotStarted(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "100") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create the container and check that it can be destroyed. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1801,16 +1792,15 @@ func TestDestroyStarting(t *testing.T) { for i := 0; i < 10; i++ { spec := testutil.NewSpecWithArgs("/bin/sleep", "100") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create the container and check that it can be destroyed. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1845,23 +1835,23 @@ func TestDestroyStarting(t *testing.T) { } func TestCreateWorkingDir(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "cwd-create") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - dir := path.Join(tmpDir, "new/working/dir") + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "cwd-create") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + dir := path.Join(tmpDir, "new/working/dir") - // touch will fail if the directory doesn't exist. - spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) - spec.Process.Cwd = dir - spec.Root.Readonly = true + // touch will fail if the directory doesn't exist. + spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) + spec.Process.Cwd = dir + spec.Root.Readonly = true - if err := run(spec, conf); err != nil { - t.Fatalf("Error running container: %v", err) - } + if err := run(spec, conf); err != nil { + t.Fatalf("Error running container: %v", err) + } + }) } } @@ -1919,15 +1909,14 @@ func TestMountPropagation(t *testing.T) { } conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1969,81 +1958,81 @@ func TestMountPropagation(t *testing.T) { } func TestMountSymlink(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "mount-symlink") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "mount-symlink") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + defer os.RemoveAll(dir) - source := path.Join(dir, "source") - target := path.Join(dir, "target") - for _, path := range []string{source, target} { - if err := os.MkdirAll(path, 0777); err != nil { - t.Fatalf("os.MkdirAll(): %v", err) + source := path.Join(dir, "source") + target := path.Join(dir, "target") + for _, path := range []string{source, target} { + if err := os.MkdirAll(path, 0777); err != nil { + t.Fatalf("os.MkdirAll(): %v", err) + } } - } - f, err := os.Create(path.Join(source, "file")) - if err != nil { - t.Fatalf("os.Create(): %v", err) - } - f.Close() + f, err := os.Create(path.Join(source, "file")) + if err != nil { + t.Fatalf("os.Create(): %v", err) + } + f.Close() - link := path.Join(dir, "link") - if err := os.Symlink(target, link); err != nil { - t.Fatalf("os.Symlink(%q, %q): %v", target, link, err) - } + link := path.Join(dir, "link") + if err := os.Symlink(target, link); err != nil { + t.Fatalf("os.Symlink(%q, %q): %v", target, link, err) + } - spec := testutil.NewSpecWithArgs("/bin/sleep", "1000") + spec := testutil.NewSpecWithArgs("/bin/sleep", "1000") - // Mount to a symlink to ensure the mount code will follow it and mount - // at the symlink target. - spec.Mounts = append(spec.Mounts, specs.Mount{ - Type: "bind", - Destination: link, - Source: source, - }) + // Mount to a symlink to ensure the mount code will follow it and mount + // at the symlink target. + spec.Mounts = append(spec.Mounts, specs.Mount{ + Type: "bind", + Destination: link, + Source: source, + }) - 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) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("creating container: %v", err) - } - defer cont.Destroy() + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("creating container: %v", err) + } + defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("starting container: %v", err) - } + if err := cont.Start(conf); err != nil { + t.Fatalf("starting container: %v", err) + } - // Check that symlink was resolved and mount was created where the symlink - // is pointing to. - file := path.Join(target, "file") - execArgs := &control.ExecArgs{ - Filename: "/usr/bin/test", - Argv: []string{"test", "-f", file}, - } - if ws, err := cont.executeSync(execArgs); err != nil || ws != 0 { - t.Fatalf("exec: test -f %q, ws: %v, err: %v", file, ws, err) - } + // Check that symlink was resolved and mount was created where the symlink + // is pointing to. + file := path.Join(target, "file") + execArgs := &control.ExecArgs{ + Filename: "/usr/bin/test", + Argv: []string{"test", "-f", file}, + } + if ws, err := cont.executeSync(execArgs); err != nil || ws != 0 { + t.Fatalf("exec: test -f %q, ws: %v, err: %v", file, ws, err) + } + }) } } // Check that --net-raw disables the CAP_NET_RAW capability. func TestNetRaw(t *testing.T) { capNetRaw := strconv.FormatUint(bits.MaskOf64(int(linux.CAP_NET_RAW)), 10) - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -2106,7 +2095,7 @@ func TestTTYField(t *testing.T) { stop := testutil.StartReaper() defer stop() - testApp, err := testutil.FindFile("runsc/container/test_app/test_app") + testApp, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -2140,16 +2129,15 @@ func TestTTYField(t *testing.T) { } spec := testutil.NewSpecWithArgs(cmd...) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index dc2fb42ce..e3704b453 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -30,15 +30,15 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { var specs []*specs.Spec var ids []string - rootID := testutil.UniqueContainerID() + rootID := testutil.RandomContainerID() for i, cmd := range cmds { spec := testutil.NewSpecWithArgs(cmd...) @@ -52,7 +52,7 @@ func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { specutils.ContainerdContainerTypeAnnotation: specutils.ContainerdContainerTypeContainer, specutils.ContainerdSandboxIDAnnotation: rootID, } - ids = append(ids, testutil.UniqueContainerID()) + ids = append(ids, testutil.RandomContainerID()) } specs = append(specs, spec) } @@ -64,23 +64,29 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C panic("conf.RootDir not set. Call testutil.SetupRootDir() to set.") } - var containers []*Container - var bundles []string - cleanup := func() { + var ( + containers []*Container + cleanups []func() + ) + cleanups = append(cleanups, func() { for _, c := range containers { c.Destroy() } - for _, b := range bundles { - os.RemoveAll(b) + }) + cleanupAll := func() { + for _, c := range cleanups { + c() } } + localClean := specutils.MakeCleanup(cleanupAll) + defer localClean.Clean() + for i, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error setting up container: %v", err) } - bundles = append(bundles, bundleDir) + cleanups = append(cleanups, cleanup) args := Args{ ID: ids[i], @@ -89,17 +95,17 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C } cont, err := New(conf, args) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error creating container: %v", err) } containers = append(containers, cont) if err := cont.Start(conf); err != nil { - cleanup() return nil, nil, fmt.Errorf("error starting container: %v", err) } } - return containers, cleanup, nil + + localClean.Release() + return containers, cleanupAll, nil } type execDesc struct { @@ -135,159 +141,159 @@ func createSharedMount(mount specs.Mount, name string, pod ...*specs.Spec) { // TestMultiContainerSanity checks that it is possible to run 2 dead-simple // containers in the same sandbox. func TestMultiContainerSanity(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir - // Setup the containers. - sleep := []string{"sleep", "100"} - specs, ids := createSpecs(sleep, sleep) - containers, cleanup, err := startContainers(conf, specs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + // Setup the containers. + sleep := []string{"sleep", "100"} + specs, ids := createSpecs(sleep, sleep) + containers, cleanup, err := startContainers(conf, specs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {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", Threads: []kernel.ThreadID{2}}, - } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {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", Threads: []kernel.ThreadID{2}}, + } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + }) } } // TestMultiPIDNS checks that it is possible to run 2 dead-simple // containers in the same sandbox with different pidns. func TestMultiPIDNS(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - testSpecs, ids := createSpecs(sleep, sleep) - testSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + testSpecs, ids := createSpecs(sleep, sleep) + testSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + }, }, - }, - } + } - containers, cleanup, err := startContainers(conf, testSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, testSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {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", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {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", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + }) } } // TestMultiPIDNSPath checks the pidns path. func TestMultiPIDNSPath(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - testSpecs, ids := createSpecs(sleep, sleep, sleep) - testSpecs[0].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/1/ns/pid", + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + testSpecs, ids := createSpecs(sleep, sleep, sleep) + testSpecs[0].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/1/ns/pid", + }, }, - }, - } - testSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/1/ns/pid", + } + testSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/1/ns/pid", + }, }, - }, - } - testSpecs[2].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/2/ns/pid", + } + testSpecs[2].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/2/ns/pid", + }, }, - }, - } + } - containers, cleanup, err := startContainers(conf, testSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, testSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {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) - } - if err := waitForProcessList(containers[2], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {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) + } + if err := waitForProcessList(containers[2], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - expectedPL = []*control.Process{ - {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) - } + expectedPL = []*control.Process{ + {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) + } + }) } } func TestMultiContainerWait(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -361,11 +367,11 @@ func TestMultiContainerWait(t *testing.T) { // TestExecWait ensures what we can wait containers and individual processes in the // sandbox that have already exited. func TestExecWait(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -457,11 +463,11 @@ func TestMultiContainerMount(t *testing.T) { }) // Setup the containers. - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -484,174 +490,174 @@ func TestMultiContainerMount(t *testing.T) { // TestMultiContainerSignal checks that it is possible to signal individual // containers without killing the entire sandbox. func TestMultiContainerSignal(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir - // Setup the containers. - sleep := []string{"sleep", "100"} - specs, ids := createSpecs(sleep, sleep) - containers, cleanup, err := startContainers(conf, specs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + // Setup the containers. + sleep := []string{"sleep", "100"} + specs, ids := createSpecs(sleep, sleep) + containers, cleanup, err := startContainers(conf, specs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that container 1 process is running. - expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, - } + // Check via ps that container 1 process is running. + expectedPL := []*control.Process{ + {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) - } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - // Kill process 2. - if err := containers[1].SignalContainer(syscall.SIGKILL, false); err != nil { - t.Errorf("failed to kill process 2: %v", err) - } + // Kill process 2. + if err := containers[1].SignalContainer(syscall.SIGKILL, false); err != nil { + t.Errorf("failed to kill process 2: %v", err) + } - // Make sure process 1 is still running. - expectedPL = []*control.Process{ - {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) - } + // Make sure process 1 is still running. + expectedPL = []*control.Process{ + {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) + } - // goferPid is reset when container is destroyed. - goferPid := containers[1].GoferPid + // goferPid is reset when container is destroyed. + goferPid := containers[1].GoferPid - // Destroy container and ensure container's gofer process has exited. - if err := containers[1].Destroy(); err != nil { - t.Errorf("failed to destroy container: %v", err) - } - _, _, err = specutils.RetryEintr(func() (uintptr, uintptr, error) { - cpid, err := syscall.Wait4(goferPid, nil, 0, nil) - return uintptr(cpid), 0, err - }) - if err != syscall.ECHILD { - t.Errorf("error waiting for gofer to exit: %v", err) - } - // Make sure process 1 is still running. - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Destroy container and ensure container's gofer process has exited. + if err := containers[1].Destroy(); err != nil { + t.Errorf("failed to destroy container: %v", err) + } + _, _, err = specutils.RetryEintr(func() (uintptr, uintptr, error) { + cpid, err := syscall.Wait4(goferPid, nil, 0, nil) + return uintptr(cpid), 0, err + }) + if err != syscall.ECHILD { + t.Errorf("error waiting for gofer to exit: %v", err) + } + // Make sure process 1 is still running. + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - // Now that process 2 is gone, ensure we get an error trying to - // signal it again. - if err := containers[1].SignalContainer(syscall.SIGKILL, false); err == nil { - t.Errorf("container %q shouldn't exist, but we were able to signal it", containers[1].ID) - } + // Now that process 2 is gone, ensure we get an error trying to + // signal it again. + if err := containers[1].SignalContainer(syscall.SIGKILL, false); err == nil { + t.Errorf("container %q shouldn't exist, but we were able to signal it", containers[1].ID) + } - // Kill process 1. - if err := containers[0].SignalContainer(syscall.SIGKILL, false); err != nil { - t.Errorf("failed to kill process 1: %v", err) - } + // Kill process 1. + if err := containers[0].SignalContainer(syscall.SIGKILL, false); err != nil { + t.Errorf("failed to kill process 1: %v", err) + } - // Ensure that container's gofer and sandbox process are no more. - err = blockUntilWaitable(containers[0].GoferPid) - if err != nil && err != syscall.ECHILD { - t.Errorf("error waiting for gofer to exit: %v", err) - } + // Ensure that container's gofer and sandbox process are no more. + err = blockUntilWaitable(containers[0].GoferPid) + if err != nil && err != syscall.ECHILD { + t.Errorf("error waiting for gofer to exit: %v", err) + } - err = blockUntilWaitable(containers[0].Sandbox.Pid) - if err != nil && err != syscall.ECHILD { - t.Errorf("error waiting for sandbox to exit: %v", err) - } + err = blockUntilWaitable(containers[0].Sandbox.Pid) + if err != nil && err != syscall.ECHILD { + t.Errorf("error waiting for sandbox to exit: %v", err) + } - // The sentry should be gone, so signaling should yield an error. - if err := containers[0].SignalContainer(syscall.SIGKILL, false); err == nil { - t.Errorf("sandbox %q shouldn't exist, but we were able to signal it", containers[0].Sandbox.ID) - } + // The sentry should be gone, so signaling should yield an error. + if err := containers[0].SignalContainer(syscall.SIGKILL, false); err == nil { + t.Errorf("sandbox %q shouldn't exist, but we were able to signal it", containers[0].Sandbox.ID) + } - if err := containers[0].Destroy(); err != nil { - t.Errorf("failed to destroy container: %v", err) - } + if err := containers[0].Destroy(); err != nil { + t.Errorf("failed to destroy container: %v", err) + } + }) } } // TestMultiContainerDestroy checks that container are properly cleaned-up when // they are destroyed. func TestMultiContainerDestroy(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // First container will remain intact while the second container is killed. - podSpecs, ids := createSpecs( - []string{"sleep", "100"}, - []string{app, "fork-bomb"}) - - // Run the fork bomb in a PID namespace to prevent processes to be - // re-parented to PID=1 in the root container. - podSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{{Type: "pid"}}, - } - containers, cleanup, err := startContainers(conf, podSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // First container will remain intact while the second container is killed. + podSpecs, ids := createSpecs( + []string{"sleep", "100"}, + []string{app, "fork-bomb"}) + + // Run the fork bomb in a PID namespace to prevent processes to be + // re-parented to PID=1 in the root container. + podSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{{Type: "pid"}}, + } + containers, cleanup, err := startContainers(conf, podSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Exec more processes to ensure signal all works for exec'd processes too. - args := &control.ExecArgs{ - Filename: app, - Argv: []string{app, "fork-bomb"}, - } - if _, err := containers[1].Execute(args); err != nil { - t.Fatalf("error exec'ing: %v", err) - } + // Exec more processes to ensure signal all works for exec'd processes too. + args := &control.ExecArgs{ + Filename: app, + Argv: []string{app, "fork-bomb"}, + } + if _, err := containers[1].Execute(args); err != nil { + t.Fatalf("error exec'ing: %v", err) + } - // Let it brew... - time.Sleep(500 * time.Millisecond) + // Let it brew... + time.Sleep(500 * time.Millisecond) - if err := containers[1].Destroy(); err != nil { - t.Fatalf("error destroying container: %v", err) - } + if err := containers[1].Destroy(); err != nil { + t.Fatalf("error destroying container: %v", err) + } - // Check that destroy killed all processes belonging to the container and - // waited for them to exit before returning. - pss, err := containers[0].Sandbox.Processes("") - if err != nil { - t.Fatalf("error getting process data from sandbox: %v", err) - } - 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 destroy killed all processes belonging to the container and + // waited for them to exit before returning. + pss, err := containers[0].Sandbox.Processes("") + if err != nil { + t.Fatalf("error getting process data from sandbox: %v", err) + } + 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. - if err := containers[1].Destroy(); err != nil { - t.Errorf("error destroying container: %v", err) - } + // Check that cont.Destroy is safe to call multiple times. + if err := containers[1].Destroy(); err != nil { + t.Errorf("error destroying container: %v", err) + } + }) } } func TestMultiContainerProcesses(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -706,11 +712,11 @@ func TestMultiContainerProcesses(t *testing.T) { // TestMultiContainerKillAll checks that all process that belong to a container // are killed when SIGKILL is sent to *all* processes in that container. func TestMultiContainerKillAll(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -721,7 +727,7 @@ func TestMultiContainerKillAll(t *testing.T) { {killContainer: true}, {killContainer: false}, } { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -739,11 +745,11 @@ func TestMultiContainerKillAll(t *testing.T) { // Wait until all processes are created. rootProcCount := int(math.Pow(2, 3) - 1) if err := waitForProcessCount(containers[0], rootProcCount); err != nil { - t.Fatal(err) + t.Fatalf("error waitting for processes: %v", err) } procCount := int(math.Pow(2, 5) - 1) if err := waitForProcessCount(containers[1], procCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Exec more processes to ensure signal works for exec'd processes too. @@ -757,7 +763,7 @@ func TestMultiContainerKillAll(t *testing.T) { // Wait for these new processes to start. procCount += int(math.Pow(2, 3) - 1) if err := waitForProcessCount(containers[1], procCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } if tc.killContainer { @@ -790,11 +796,11 @@ func TestMultiContainerKillAll(t *testing.T) { // Check that all processes are gone. if err := waitForProcessCount(containers[1], 0); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Check that root container was not affected. if err := waitForProcessCount(containers[0], rootProcCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } } } @@ -805,17 +811,16 @@ func TestMultiContainerDestroyNotStarted(t *testing.T) { []string{"/bin/sleep", "100"}) conf := testutil.TestConfig(t) - rootDir, rootBundleDir, err := testutil.SetupContainer(specs[0], conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(specs[0], conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(rootBundleDir) + defer cleanup() rootArgs := Args{ ID: ids[0], Spec: specs[0], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } root, err := New(conf, rootArgs) if err != nil { @@ -827,11 +832,11 @@ func TestMultiContainerDestroyNotStarted(t *testing.T) { } // Create and destroy sub-container. - bundleDir, err := testutil.SetupBundleDir(specs[1]) + bundleDir, cleanupSub, err := testutil.SetupBundleDir(specs[1]) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanupSub() args := Args{ ID: ids[1], @@ -859,17 +864,16 @@ func TestMultiContainerDestroyStarting(t *testing.T) { specs, ids := createSpecs(cmds...) conf := testutil.TestConfig(t) - rootDir, rootBundleDir, err := testutil.SetupContainer(specs[0], conf) + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(specs[0], conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(rootBundleDir) + defer cleanup() rootArgs := Args{ ID: ids[0], Spec: specs[0], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } root, err := New(conf, rootArgs) if err != nil { @@ -886,16 +890,16 @@ func TestMultiContainerDestroyStarting(t *testing.T) { continue // skip root container } - bundleDir, err := testutil.SetupBundleDir(specs[i]) + bundleDir, cleanup, err := testutil.SetupBundleDir(specs[i]) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() rootArgs := Args{ ID: ids[i], Spec: specs[i], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } cont, err := New(conf, rootArgs) if err != nil { @@ -937,11 +941,11 @@ func TestMultiContainerDifferentFilesystems(t *testing.T) { script := fmt.Sprintf("if [ -f %q ]; then exit 1; else touch %q; fi", filename, filename) cmd := []string{"sh", "-c", script} - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -977,7 +981,7 @@ func TestMultiContainerDifferentFilesystems(t *testing.T) { // TestMultiContainerContainerDestroyStress tests that IO operations continue // to work after containers have been stopped and gofers killed. func TestMultiContainerContainerDestroyStress(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1007,12 +1011,11 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { childrenIDs := allIDs[1:] conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(rootSpec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(rootSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Start root container. rootArgs := Args{ @@ -1038,11 +1041,11 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { var children []*Container for j, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ ID: ids[j], @@ -1080,306 +1083,306 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { // Test that pod shared mounts are properly mounted in 2 containers and that // changes from one container is reflected in the other. func TestMultiContainerSharedMount(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: nil, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: nil, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", - }, - { - c: containers[1], - cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", - }, - { - c: containers[1], - cmd: []string{"/bin/mkdir", file1}, - desc: "create directory in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", file0}, - desc: "dir appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", file1}, - desc: "dir appears in container1", - }, - { - c: containers[0], - cmd: []string{"/bin/rmdir", file0}, - desc: "create directory in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-d", file0}, - desc: "dir removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-d", file1}, - desc: "dir removed from container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, + desc: "directory is mounted in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, + desc: "directory is mounted in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + desc: "create file in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file appears in container1", + }, + { + c: containers[1], + cmd: []string{"/bin/rm", file1}, + desc: "file removed from container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-f", file0}, + desc: "file removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-f", file1}, + desc: "file removed from container1", + }, + { + c: containers[1], + cmd: []string{"/bin/mkdir", file1}, + desc: "create directory in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", file0}, + desc: "dir appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", file1}, + desc: "dir appears in container1", + }, + { + c: containers[0], + cmd: []string{"/bin/rmdir", file0}, + desc: "create directory in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-d", file0}, + desc: "dir removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-d", file1}, + desc: "dir removed from container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that pod mounts are mounted as readonly when requested. func TestMultiContainerSharedMountReadonly(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: []string{"ro"}, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: []string{"ro"}, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - want: 1, - desc: "fails to write to container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/touch", file1}, - want: 1, - desc: "fails to write to container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, + desc: "directory is mounted in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, + desc: "directory is mounted in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + want: 1, + desc: "fails to write to container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/touch", file1}, + want: 1, + desc: "fails to write to container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that shared pod mounts continue to work after container is restarted. func TestMultiContainerSharedMountRestart(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: nil, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: nil, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + desc: "create file in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file appears in container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } - containers[1].Destroy() + containers[1].Destroy() - bundleDir, err := testutil.SetupBundleDir(podSpec[1]) - if err != nil { - t.Fatalf("error restarting container: %v", err) - } - defer os.RemoveAll(bundleDir) + bundleDir, cleanup, err := testutil.SetupBundleDir(podSpec[1]) + if err != nil { + t.Fatalf("error restarting container: %v", err) + } + defer cleanup() - args := Args{ - ID: ids[1], - Spec: podSpec[1], - BundleDir: bundleDir, - } - containers[1], err = New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - if err := containers[1].Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + args := Args{ + ID: ids[1], + Spec: podSpec[1], + BundleDir: bundleDir, + } + containers[1], err = New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + if err := containers[1].Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - execs = []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file is still in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file is still in container1", - }, - { - c: containers[1], - cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execs = []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file is still in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file is still in container1", + }, + { + c: containers[1], + cmd: []string{"/bin/rm", file1}, + desc: "file removed from container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-f", file0}, + desc: "file removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-f", file1}, + desc: "file removed from container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that unsupported pod mounts options are ignored when matching master and // slave mounts. func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1428,7 +1431,7 @@ func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) { // Test that one container can send an FD to another container, even though // they have distinct MountNamespaces. func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1457,11 +1460,11 @@ func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { Type: "tmpfs", } - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1494,11 +1497,11 @@ func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { // Test that container is destroyed when Gofer is killed. func TestMultiContainerGoferKilled(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1581,11 +1584,11 @@ func TestMultiContainerLoadSandbox(t *testing.T) { sleep := []string{"sleep", "100"} specs, ids := createSpecs(sleep, sleep, sleep) - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1614,7 +1617,7 @@ func TestMultiContainerLoadSandbox(t *testing.T) { } // Create a valid but empty container directory. - randomCID := testutil.UniqueContainerID() + randomCID := testutil.RandomContainerID() dir = filepath.Join(conf.RootDir, randomCID) if err := os.MkdirAll(dir, 0755); err != nil { t.Fatalf("os.MkdirAll(%q)=%v", dir, err) @@ -1681,11 +1684,11 @@ func TestMultiContainerRunNonRoot(t *testing.T) { Type: "bind", }) - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir diff --git a/runsc/container/shared_volume_test.go b/runsc/container/shared_volume_test.go index f80852414..bac177a88 100644 --- a/runsc/container/shared_volume_test.go +++ b/runsc/container/shared_volume_test.go @@ -24,8 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" - "gvisor.dev/gvisor/runsc/testutil" ) // TestSharedVolume checks that modifications to a volume mount are propagated @@ -33,7 +33,6 @@ import ( func TestSharedVolume(t *testing.T) { conf := testutil.TestConfig(t) conf.FileAccess = boot.FileAccessShared - t.Logf("Running test with conf: %+v", conf) // Main process just sleeps. We will use "exec" to probe the state of // the filesystem. @@ -44,16 +43,15 @@ func TestSharedVolume(t *testing.T) { t.Fatalf("TempDir failed: %v", err) } - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -192,7 +190,6 @@ func checkFile(c *Container, filename string, want []byte) error { func TestSharedVolumeFile(t *testing.T) { conf := testutil.TestConfig(t) conf.FileAccess = boot.FileAccessShared - t.Logf("Running test with conf: %+v", conf) // Main process just sleeps. We will use "exec" to probe the state of // the filesystem. @@ -203,16 +200,15 @@ func TestSharedVolumeFile(t *testing.T) { t.Fatalf("TempDir failed: %v", err) } - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go deleted file mode 100644 index f009486bc..000000000 --- a/runsc/dockerutil/dockerutil.go +++ /dev/null @@ -1,486 +0,0 @@ -// 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. - -// Package dockerutil is a collection of utility functions, primarily for -// testing. -package dockerutil - -import ( - "encoding/json" - "flag" - "fmt" - "io/ioutil" - "log" - "os" - "os/exec" - "path" - "regexp" - "strconv" - "strings" - "syscall" - "time" - - "github.com/kr/pty" - "gvisor.dev/gvisor/runsc/testutil" -) - -var ( - // runtime is the runtime to use for tests. This will be applied to all - // containers. Note that the default here ("runsc") corresponds to the - // default used by the installations. This is important, because the - // default installer for vm_tests (in tools/installers:head, invoked - // via tools/vm:defs.bzl) will install with this name. So without - // changing anything, tests should have a runsc runtime available to - // them. Otherwise installers should update the existing runtime - // instead of installing a new one. - runtime = flag.String("runtime", "runsc", "specify which runtime to use") - - // config is the default Docker daemon configuration path. - config = flag.String("config_path", "/etc/docker/daemon.json", "configuration file for reading paths") -) - -// EnsureSupportedDockerVersion checks if correct docker is installed. -func EnsureSupportedDockerVersion() { - cmd := exec.Command("docker", "version") - out, err := cmd.CombinedOutput() - if err != nil { - log.Fatalf("Error running %q: %v", "docker version", err) - } - re := regexp.MustCompile(`Version:\s+(\d+)\.(\d+)\.\d.*`) - matches := re.FindStringSubmatch(string(out)) - if len(matches) != 3 { - log.Fatalf("Invalid docker output: %s", out) - } - major, _ := strconv.Atoi(matches[1]) - minor, _ := strconv.Atoi(matches[2]) - if major < 17 || (major == 17 && minor < 9) { - log.Fatalf("Docker version 17.09.0 or greater is required, found: %02d.%02d", major, minor) - } -} - -// RuntimePath returns the binary path for the current runtime. -func RuntimePath() (string, error) { - // Read the configuration data; the file must exist. - configBytes, err := ioutil.ReadFile(*config) - if err != nil { - return "", err - } - - // Unmarshal the configuration. - c := make(map[string]interface{}) - if err := json.Unmarshal(configBytes, &c); err != nil { - return "", err - } - - // Decode the expected configuration. - r, ok := c["runtimes"] - if !ok { - return "", fmt.Errorf("no runtimes declared: %v", c) - } - rs, ok := r.(map[string]interface{}) - if !ok { - // The runtimes are not a map. - return "", fmt.Errorf("unexpected format: %v", c) - } - r, ok = rs[*runtime] - if !ok { - // The expected runtime is not declared. - return "", fmt.Errorf("runtime %q not found: %v", *runtime, c) - } - rs, ok = r.(map[string]interface{}) - if !ok { - // The runtime is not a map. - return "", fmt.Errorf("unexpected format: %v", c) - } - p, ok := rs["path"].(string) - if !ok { - // The runtime does not declare a path. - return "", fmt.Errorf("unexpected format: %v", c) - } - return p, nil -} - -// MountMode describes if the mount should be ro or rw. -type MountMode int - -const ( - // ReadOnly is what the name says. - ReadOnly MountMode = iota - // ReadWrite is what the name says. - ReadWrite -) - -// String returns the mount mode argument for this MountMode. -func (m MountMode) String() string { - switch m { - case ReadOnly: - return "ro" - case ReadWrite: - return "rw" - } - panic(fmt.Sprintf("invalid mode: %d", m)) -} - -// MountArg formats the volume argument to mount in the container. -func MountArg(source, target string, mode MountMode) string { - return fmt.Sprintf("-v=%s:%s:%v", source, target, mode) -} - -// LinkArg formats the link argument. -func LinkArg(source *Docker, target string) string { - return fmt.Sprintf("--link=%s:%s", source.Name, target) -} - -// PrepareFiles creates temp directory to copy files there. The sandbox doesn't -// have access to files in the test dir. -func PrepareFiles(names ...string) (string, error) { - dir, err := ioutil.TempDir("", "image-test") - if err != nil { - return "", fmt.Errorf("ioutil.TempDir failed: %v", err) - } - if err := os.Chmod(dir, 0777); err != nil { - return "", fmt.Errorf("os.Chmod(%q, 0777) failed: %v", dir, err) - } - for _, name := range names { - src, err := testutil.FindFile(name) - if err != nil { - return "", fmt.Errorf("testutil.Preparefiles(%q) failed: %v", name, err) - } - dst := path.Join(dir, path.Base(name)) - if err := testutil.Copy(src, dst); err != nil { - return "", fmt.Errorf("testutil.Copy(%q, %q) failed: %v", src, dst, err) - } - } - return dir, nil -} - -// do executes docker command. -func do(args ...string) (string, error) { - log.Printf("Running: docker %s\n", args) - cmd := exec.Command("docker", args...) - out, err := cmd.CombinedOutput() - if err != nil { - return "", fmt.Errorf("error executing docker %s: %v\nout: %s", args, err, out) - } - return string(out), nil -} - -// doWithPty executes docker command with stdio attached to a pty. -func doWithPty(args ...string) (*exec.Cmd, *os.File, error) { - log.Printf("Running with pty: docker %s\n", args) - cmd := exec.Command("docker", args...) - ptmx, err := pty.Start(cmd) - if err != nil { - return nil, nil, fmt.Errorf("error executing docker %s with a pty: %v", args, err) - } - return cmd, ptmx, nil -} - -// Pull pulls a docker image. This is used in tests to isolate the -// time to pull the image off the network from the time to actually -// start the container, to avoid timeouts over slow networks. -func Pull(image string) error { - _, err := do("pull", image) - return err -} - -// Docker contains the name and the runtime of a docker container. -type Docker struct { - Runtime string - Name string -} - -// MakeDocker sets up the struct for a Docker container. -// Names of containers will be unique. -func MakeDocker(namePrefix string) Docker { - return Docker{ - Name: testutil.RandomName(namePrefix), - Runtime: *runtime, - } -} - -// logDockerID logs a container id, which is needed to find container runsc logs. -func (d *Docker) logDockerID() { - id, err := d.ID() - if err != nil { - log.Printf("%v\n", err) - } - log.Printf("Name: %s ID: %v\n", d.Name, id) -} - -// Create calls 'docker create' with the arguments provided. -func (d *Docker) Create(args ...string) error { - a := []string{"create", "--runtime", d.Runtime, "--name", d.Name} - a = append(a, args...) - _, err := do(a...) - if err == nil { - d.logDockerID() - } - return err -} - -// Start calls 'docker start'. -func (d *Docker) Start() error { - if _, err := do("start", d.Name); err != nil { - return fmt.Errorf("error starting container %q: %v", d.Name, err) - } - return nil -} - -// Stop calls 'docker stop'. -func (d *Docker) Stop() error { - if _, err := do("stop", d.Name); err != nil { - return fmt.Errorf("error stopping container %q: %v", d.Name, err) - } - return nil -} - -// Run calls 'docker run' with the arguments provided. The container starts -// running in the background and the call returns immediately. -func (d *Docker) Run(args ...string) error { - a := d.runArgs("-d") - a = append(a, args...) - _, err := do(a...) - if err == nil { - d.logDockerID() - } - return err -} - -// RunWithPty is like Run but with an attached pty. -func (d *Docker) RunWithPty(args ...string) (*exec.Cmd, *os.File, error) { - a := d.runArgs("-it") - a = append(a, args...) - return doWithPty(a...) -} - -// RunFg calls 'docker run' with the arguments provided in the foreground. It -// blocks until the container exits and returns the output. -func (d *Docker) RunFg(args ...string) (string, error) { - a := d.runArgs(args...) - out, err := do(a...) - if err == nil { - d.logDockerID() - } - return string(out), err -} - -func (d *Docker) runArgs(args ...string) []string { - // Environment variable RUNSC_TEST_NAME is picked up by the runtime and added - // to the log name, so one can easily identify the corresponding logs for - // this test. - rv := []string{"run", "--runtime", d.Runtime, "--name", d.Name, "-e", "RUNSC_TEST_NAME=" + d.Name} - return append(rv, args...) -} - -// Logs calls 'docker logs'. -func (d *Docker) Logs() (string, error) { - return do("logs", d.Name) -} - -// Exec calls 'docker exec' with the arguments provided. -func (d *Docker) Exec(args ...string) (string, error) { - return d.ExecWithFlags(nil, args...) -} - -// ExecWithFlags calls 'docker exec <flags> name <args>'. -func (d *Docker) ExecWithFlags(flags []string, args ...string) (string, error) { - a := []string{"exec"} - a = append(a, flags...) - a = append(a, d.Name) - a = append(a, args...) - return do(a...) -} - -// ExecAsUser calls 'docker exec' as the given user with the arguments -// provided. -func (d *Docker) ExecAsUser(user string, args ...string) (string, error) { - a := []string{"exec", "--user", user, d.Name} - a = append(a, args...) - return do(a...) -} - -// ExecWithTerminal calls 'docker exec -it' with the arguments provided and -// attaches a pty to stdio. -func (d *Docker) ExecWithTerminal(args ...string) (*exec.Cmd, *os.File, error) { - a := []string{"exec", "-it", d.Name} - a = append(a, args...) - return doWithPty(a...) -} - -// Pause calls 'docker pause'. -func (d *Docker) Pause() error { - if _, err := do("pause", d.Name); err != nil { - return fmt.Errorf("error pausing container %q: %v", d.Name, err) - } - return nil -} - -// Unpause calls 'docker pause'. -func (d *Docker) Unpause() error { - if _, err := do("unpause", d.Name); err != nil { - return fmt.Errorf("error unpausing container %q: %v", d.Name, err) - } - return nil -} - -// Checkpoint calls 'docker checkpoint'. -func (d *Docker) Checkpoint(name string) error { - if _, err := do("checkpoint", "create", d.Name, name); err != nil { - return fmt.Errorf("error pausing container %q: %v", d.Name, err) - } - return nil -} - -// Restore calls 'docker start --checkname [name]'. -func (d *Docker) Restore(name string) error { - if _, err := do("start", "--checkpoint", name, d.Name); err != nil { - return fmt.Errorf("error starting container %q: %v", d.Name, err) - } - return nil -} - -// Remove calls 'docker rm'. -func (d *Docker) Remove() error { - if _, err := do("rm", d.Name); err != nil { - return fmt.Errorf("error deleting container %q: %v", d.Name, err) - } - return nil -} - -// CleanUp kills and deletes the container (best effort). -func (d *Docker) CleanUp() { - d.logDockerID() - if _, err := do("kill", d.Name); err != nil { - if strings.Contains(err.Error(), "is not running") { - // Nothing to kill. Don't log the error in this case. - } else { - log.Printf("error killing container %q: %v", d.Name, err) - } - } - if err := d.Remove(); err != nil { - log.Print(err) - } -} - -// FindPort returns the host port that is mapped to 'sandboxPort'. This calls -// docker to allocate a free port in the host and prevent conflicts. -func (d *Docker) FindPort(sandboxPort int) (int, error) { - format := fmt.Sprintf(`{{ (index (index .NetworkSettings.Ports "%d/tcp") 0).HostPort }}`, sandboxPort) - out, err := do("inspect", "-f", format, d.Name) - if err != nil { - return -1, fmt.Errorf("error retrieving port: %v", err) - } - port, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - return -1, fmt.Errorf("error parsing port %q: %v", out, err) - } - 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) - if err != nil { - return -1, fmt.Errorf("error retrieving pid: %v", err) - } - pid, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - return -1, fmt.Errorf("error parsing pid %q: %v", out, err) - } - return pid, nil -} - -// ID returns the container ID. -func (d *Docker) ID() (string, error) { - out, err := do("inspect", "-f={{.Id}}", d.Name) - if err != nil { - return "", fmt.Errorf("error retrieving ID: %v", err) - } - return strings.TrimSpace(string(out)), nil -} - -// Wait waits for container to exit, up to the given timeout. Returns error if -// wait fails or timeout is hit. Returns the application return code otherwise. -// Note that the application may have failed even if err == nil, always check -// the exit code. -func (d *Docker) Wait(timeout time.Duration) (syscall.WaitStatus, error) { - timeoutChan := time.After(timeout) - waitChan := make(chan (syscall.WaitStatus)) - errChan := make(chan (error)) - - go func() { - out, err := do("wait", d.Name) - if err != nil { - errChan <- fmt.Errorf("error waiting for container %q: %v", d.Name, err) - } - exit, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - errChan <- fmt.Errorf("error parsing exit code %q: %v", out, err) - } - waitChan <- syscall.WaitStatus(uint32(exit)) - }() - - select { - case ws := <-waitChan: - return ws, nil - case err := <-errChan: - return syscall.WaitStatus(1), err - case <-timeoutChan: - return syscall.WaitStatus(1), fmt.Errorf("timeout waiting for container %q", d.Name) - } -} - -// WaitForOutput calls 'docker logs' to retrieve containers output and searches -// for the given pattern. -func (d *Docker) WaitForOutput(pattern string, timeout time.Duration) (string, error) { - matches, err := d.WaitForOutputSubmatch(pattern, timeout) - if err != nil { - return "", err - } - if len(matches) == 0 { - return "", nil - } - return matches[0], nil -} - -// WaitForOutputSubmatch calls 'docker logs' to retrieve containers output and -// searches for the given pattern. It returns any regexp submatches as well. -func (d *Docker) WaitForOutputSubmatch(pattern string, timeout time.Duration) ([]string, error) { - re := regexp.MustCompile(pattern) - var out string - for exp := time.Now().Add(timeout); time.Now().Before(exp); { - var err error - out, err = d.Logs() - if err != nil { - return nil, err - } - if matches := re.FindStringSubmatch(out); matches != nil { - // Success! - return matches, nil - } - time.Sleep(100 * time.Millisecond) - } - return nil, fmt.Errorf("timeout waiting for output %q: %s", re.String(), out) -} diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index e82bcef6f..e4ec16e2f 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -446,9 +446,13 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF nextFD++ } - // If the platform needs a device FD we must pass it in. - if deviceFile, err := deviceFileForPlatform(conf.Platform); err != nil { + gPlatform, err := platform.Lookup(conf.Platform) + if err != nil { return err + } + + if deviceFile, err := gPlatform.OpenDevice(); err != nil { + return fmt.Errorf("opening device file for platform %q: %v", gPlatform, err) } else if deviceFile != nil { defer deviceFile.Close() cmd.ExtraFiles = append(cmd.ExtraFiles, deviceFile) @@ -539,7 +543,7 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF {Type: specs.UTSNamespace}, } - if conf.Platform == platforms.Ptrace { + if gPlatform.Requirements().RequiresCurrentPIDNS { // TODO(b/75837838): Also set a new PID namespace so that we limit // access to other host processes. log.Infof("Sandbox will be started in the current PID namespace") diff --git a/scripts/benchmarks.sh b/scripts/benchmarks.sh deleted file mode 100755 index 6b9065b07..000000000 --- a/scripts/benchmarks.sh +++ /dev/null @@ -1,53 +0,0 @@ -#!/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. -#!/usr/bin/env bash - -if [ "$#" -lt "1" ]; then - echo "usage: $0 <--mock |--env=<filename>> ..." - echo "example: $0 --mock --runs=8" - exit 1 -fi - -source $(dirname $0)/common.sh - -readonly TIMESTAMP=`date "+%Y%m%d-%H%M%S"` -readonly OUTDIR="$(mktemp --tmpdir -d run-${TIMESTAMP}-XXX)" -readonly DEFAULT_RUNTIMES="--runtime=runc --runtime=runsc --runtime=runsc-kvm" -readonly ALL_RUNTIMES="--runtime=runc --runtime=runsc --runtime=runsc-kvm" - -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} 'fio.(read|write)' --metric=bandwidth --size=5g --ioengine=sync --blocksize=1m > "${OUTDIR}/fio.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} fio.rand --metric=bandwidth --size=5g --ioengine=sync --blocksize=4k --time=30 > "${OUTDIR}/tmp_fio.csv" -cat "${OUTDIR}/tmp_fio.csv" | grep "\(runc\|runsc\)" >> "${OUTDIR}/fio.csv" && rm "${OUTDIR}/tmp_fio.csv" - -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} 'fio.(read|write)' --metric=bandwidth --tmpfs=True --size=5g --ioengine=sync --blocksize=1m > "${OUTDIR}/fio-tmpfs.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} fio.rand --metric=bandwidth --tmpfs=True --size=5g --ioengine=sync --blocksize=4k --time=30 > "${OUTDIR}/tmp_fio.csv" -cat "${OUTDIR}/tmp_fio.csv" | grep "\(runc\|runsc\)" >> "${OUTDIR}/fio-tmpfs.csv" && rm "${OUTDIR}/tmp_fio.csv" - -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} startup --count=50 > "${OUTDIR}/startup.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} density > "${OUTDIR}/density.csv" - -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} sysbench.cpu --threads=1 --max_prime=50000 --options='--max-time=5' > "${OUTDIR}/sysbench-cpu.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} sysbench.memory --threads=1 --options='--memory-block-size=1M --memory-total-size=500G' > "${OUTDIR}/sysbench-memory.csv" -run //benchmarks:perf -- run "$@" ${ALL_RUNTIMES} syscall > "${OUTDIR}/syscall.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} 'network.(upload|download)' --runs=20 > "${OUTDIR}/iperf.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} ml.tensorflow > "${OUTDIR}/tensorflow.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} media.ffmpeg > "${OUTDIR}/ffmpeg.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} http.httpd --path=latin100k.txt --connections=1 --connections=5 --connections=10 --connections=25 > "${OUTDIR}/httpd100k.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} http.httpd --path=latin10240k.txt --connections=1 --connections=5 --connections=10 --connections=25 > "${OUTDIR}/httpd10240k.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} redis > "${OUTDIR}/redis.csv" -run //benchmarks:perf -- run "$@" ${DEFAULT_RUNTIMES} 'http.(ruby|node)' > "${OUTDIR}/applications.csv" - -echo "${OUTPUT}" && exit 0 diff --git a/scripts/iptables_tests.sh b/scripts/iptables_tests.sh index 0f46909ac..c8da1f32d 100755 --- a/scripts/iptables_tests.sh +++ b/scripts/iptables_tests.sh @@ -17,14 +17,5 @@ source $(dirname $0)/common.sh install_runsc_for_test iptables --net-raw - -# Build the docker image for the test. -run //test/iptables/runner:runner-image --norun - -test //test/iptables:iptables_test \ - "--test_arg=--runtime=runc" \ - "--test_arg=--image=bazel/test/iptables/runner:runner-image" - -test //test/iptables:iptables_test \ - "--test_arg=--runtime=${RUNTIME}" \ - "--test_arg=--image=bazel/test/iptables/runner:runner-image" +test //test/iptables:iptables_test --test_arg=--runtime=runc +test //test/iptables:iptables_test --test_arg=--runtime=${RUNTIME} diff --git a/runsc/container/test_app/BUILD b/test/cmd/test_app/BUILD index 0defbd9fc..98ba5a3d9 100644 --- a/runsc/container/test_app/BUILD +++ b/test/cmd/test_app/BUILD @@ -12,9 +12,9 @@ go_binary( pure = True, visibility = ["//runsc/container:__pkg__"], deps = [ + "//pkg/test/testutil", "//pkg/unet", "//runsc/flag", - "//runsc/testutil", "@com_github_google_subcommands//:go_default_library", "@com_github_kr_pty//:go_default_library", ], diff --git a/runsc/container/test_app/fds.go b/test/cmd/test_app/fds.go index 2a146a2c3..a7658eefd 100644 --- a/runsc/container/test_app/fds.go +++ b/test/cmd/test_app/fds.go @@ -22,9 +22,9 @@ import ( "time" "github.com/google/subcommands" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/flag" - "gvisor.dev/gvisor/runsc/testutil" ) const fileContents = "foobarbaz" diff --git a/runsc/container/test_app/test_app.go b/test/cmd/test_app/test_app.go index 5f1c4b7d6..3ba4f38f8 100644 --- a/runsc/container/test_app/test_app.go +++ b/test/cmd/test_app/test_app.go @@ -32,8 +32,8 @@ import ( "github.com/google/subcommands" "github.com/kr/pty" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/flag" - "gvisor.dev/gvisor/runsc/testutil" ) func main() { diff --git a/test/e2e/BUILD b/test/e2e/BUILD index 76e04f878..44cce0e3b 100644 --- a/test/e2e/BUILD +++ b/test/e2e/BUILD @@ -20,9 +20,9 @@ go_test( deps = [ "//pkg/abi/linux", "//pkg/bits", - "//runsc/dockerutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", "//runsc/specutils", - "//runsc/testutil", ], ) diff --git a/test/e2e/exec_test.go b/test/e2e/exec_test.go index 594c8e752..6a63b1232 100644 --- a/test/e2e/exec_test.go +++ b/test/e2e/exec_test.go @@ -23,6 +23,8 @@ package integration import ( "fmt" + "os" + "os/exec" "strconv" "strings" "syscall" @@ -31,23 +33,23 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bits" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/runsc/specutils" ) // Test that exec uses the exact same capability set as the container. func TestExecCapabilities(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-capabilities-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Check that capability. matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) if err != nil { t.Fatalf("WaitForOutputSubmatch() timeout: %v", err) @@ -59,7 +61,7 @@ func TestExecCapabilities(t *testing.T) { t.Log("Root capabilities:", want) // Now check that exec'd process capabilities match the root. - got, err := d.Exec("grep", "CapEff:", "/proc/self/status") + got, err := d.Exec(dockerutil.RunOpts{}, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -72,16 +74,16 @@ func TestExecCapabilities(t *testing.T) { // Test that 'exec --privileged' adds all capabilities, except for CAP_NET_RAW // which is removed from the container when --net-raw=false. func TestExecPrivileged(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-privileged-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with all capabilities dropped. - if err := d.Run("--cap-drop=all", "alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + CapDrop: []string{"all"}, + }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Check that all capabilities where dropped from container. matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) @@ -100,9 +102,11 @@ func TestExecPrivileged(t *testing.T) { t.Fatalf("Container should have no capabilities: %x", containerCaps) } - // Check that 'exec --privileged' adds all capabilities, except - // for CAP_NET_RAW. - got, err := d.ExecWithFlags([]string{"--privileged"}, "grep", "CapEff:", "/proc/self/status") + // Check that 'exec --privileged' adds all capabilities, except for + // CAP_NET_RAW. + got, err := d.Exec(dockerutil.RunOpts{ + Privileged: true, + }, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -114,97 +118,99 @@ func TestExecPrivileged(t *testing.T) { } func TestExecJobControl(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-job-control-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec 'sh' with an attached pty. - cmd, ptmx, err := d.ExecWithTerminal("sh") - if err != nil { + if _, err := d.Exec(dockerutil.RunOpts{ + Pty: func(cmd *exec.Cmd, ptmx *os.File) { + // Call "sleep 100 | cat" in the shell. We pipe to cat + // so that there will be two processes in the + // foreground process group. + if _, err := ptmx.Write([]byte("sleep 100 | cat\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Give shell a few seconds to start executing the sleep. + time.Sleep(2 * time.Second) + + // Send a ^C to the pty, which should kill sleep and + // cat, but not the shell. \x03 is ASCII "end of + // text", which is the same as ^C. + if _, err := ptmx.Write([]byte{'\x03'}); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // The shell should still be alive at this point. Sleep + // should have exited with code 2+128=130. We'll exit + // with 10 plus that number, so that we can be sure + // that the shell did not get signalled. + if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Exec process should exit with code 10+130=140. + ps, err := cmd.Process.Wait() + if err != nil { + t.Fatalf("error waiting for exec process: %v", err) + } + ws := ps.Sys().(syscall.WaitStatus) + if !ws.Exited() { + t.Errorf("ws.Exited got false, want true") + } + if got, want := ws.ExitStatus(), 140; got != want { + t.Errorf("ws.ExitedStatus got %d, want %d", got, want) + } + }, + }, "sh"); err != nil { t.Fatalf("docker exec failed: %v", err) } - defer ptmx.Close() - - // Call "sleep 100 | cat" in the shell. We pipe to cat so that there - // will be two processes in the foreground process group. - if _, err := ptmx.Write([]byte("sleep 100 | cat\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Give shell a few seconds to start executing the sleep. - time.Sleep(2 * time.Second) - - // Send a ^C to the pty, which should kill sleep and cat, but not the - // shell. \x03 is ASCII "end of text", which is the same as ^C. - if _, err := ptmx.Write([]byte{'\x03'}); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // The shell should still be alive at this point. Sleep should have - // exited with code 2+128=130. We'll exit with 10 plus that number, so - // that we can be sure that the shell did not get signalled. - if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Exec process should exit with code 10+130=140. - ps, err := cmd.Process.Wait() - if err != nil { - t.Fatalf("error waiting for exec process: %v", err) - } - ws := ps.Sys().(syscall.WaitStatus) - if !ws.Exited() { - t.Errorf("ws.Exited got false, want true") - } - if got, want := ws.ExitStatus(), 140; got != want { - t.Errorf("ws.ExitedStatus got %d, want %d", got, want) - } } // Test that failure to exec returns proper error message. func TestExecError(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-error-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() - _, err := d.Exec("no_can_find") + // Attempt to exec a binary that doesn't exist. + out, err := d.Exec(dockerutil.RunOpts{}, "no_can_find") if err == nil { t.Fatalf("docker exec didn't fail") } - if want := `error finding executable "no_can_find" in PATH`; !strings.Contains(err.Error(), want) { - t.Fatalf("docker exec wrong error, got: %s, want: .*%s.*", err.Error(), want) + if want := `error finding executable "no_can_find" in PATH`; !strings.Contains(out, want) { + t.Fatalf("docker exec wrong error, got: %s, want: .*%s.*", out, want) } } // Test that exec inherits environment from run. func TestExecEnv(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-env-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with env FOO=BAR. - if err := d.Run("-e", "FOO=BAR", "alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Env: []string{"FOO=BAR"}, + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec "echo $FOO". - got, err := d.Exec("/bin/sh", "-c", "echo $FOO") + got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $FOO") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -216,17 +222,19 @@ func TestExecEnv(t *testing.T) { // TestRunEnvHasHome tests that run always has HOME environment set. func TestRunEnvHasHome(t *testing.T) { // Base alpine image does not have any environment variables set. - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("run-env-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Exec "echo $HOME". The 'bin' user's home dir is '/bin'. - got, err := d.RunFg("--user", "bin", "alpine", "/bin/sh", "-c", "echo $HOME") + got, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + User: "bin", + }, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + + // Check that the directory matches. if got, want := strings.TrimSpace(got), "/bin"; got != want { t.Errorf("bad output from 'docker run'. Got %q; Want %q.", got, want) } @@ -235,18 +243,17 @@ func TestRunEnvHasHome(t *testing.T) { // Test that exec always has HOME environment set, even when not set in run. func TestExecEnvHasHome(t *testing.T) { // Base alpine image does not have any environment variables set. - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-env-home-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec "echo $HOME", and expect to see "/root". - got, err := d.Exec("/bin/sh", "-c", "echo $HOME") + got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -258,12 +265,14 @@ func TestExecEnvHasHome(t *testing.T) { newUID := 1234 newHome := "/foo/bar" cmd := fmt.Sprintf("mkdir -p -m 777 %q && adduser foo -D -u %d -h %q", newHome, newUID, newHome) - if _, err := d.Exec("/bin/sh", "-c", cmd); err != nil { + if _, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", cmd); err != nil { t.Fatalf("docker exec failed: %v", err) } // Execute the same as the new user and expect newHome. - got, err = d.ExecAsUser(strconv.Itoa(newUID), "/bin/sh", "-c", "echo $HOME") + got, err = d.Exec(dockerutil.RunOpts{ + User: strconv.Itoa(newUID), + }, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker exec failed: %v", err) } diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index cc4fbbaed..404e37689 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -27,14 +27,15 @@ import ( "net" "net/http" "os" + "os/exec" "strconv" "strings" "syscall" "testing" "time" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) // httpRequestSucceeds sends a request to a given url and checks that the status is OK. @@ -53,65 +54,66 @@ func httpRequestSucceeds(client http.Client, server string, port int) error { // TestLifeCycle tests a basic Create/Start/Stop docker container life cycle. func TestLifeCycle(t *testing.T) { - if err := dockerutil.Pull("nginx"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("lifecycle-test") - if err := d.Create("-p", "80", "nginx"); err != nil { - t.Fatal("docker create failed:", err) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Create(dockerutil.RunOpts{ + Image: "basic/nginx", + Ports: []int{80}, + }); err != nil { + t.Fatalf("docker create failed: %v", err) } if err := d.Start(); err != nil { - d.CleanUp() - t.Fatal("docker start failed:", err) + t.Fatalf("docker start failed: %v", err) } - // Test that container is working + // Test that container is working. port, err := d.FindPort(80) if err != nil { - t.Fatal("docker.FindPort(80) failed: ", err) + t.Fatalf("docker.FindPort(80) failed: %v", err) } if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } client := http.Client{Timeout: time.Duration(2 * time.Second)} if err := httpRequestSucceeds(client, "localhost", port); err != nil { - t.Error("http request failed:", err) + t.Errorf("http request failed: %v", err) } if err := d.Stop(); err != nil { - d.CleanUp() - t.Fatal("docker stop failed:", err) + t.Fatalf("docker stop failed: %v", err) } if err := d.Remove(); err != nil { - t.Fatal("docker rm failed:", err) + t.Fatalf("docker rm failed: %v", err) } } func TestPauseResume(t *testing.T) { - const img = "gcr.io/gvisor-presubmit/python-hello" if !testutil.IsCheckpointSupported() { - t.Log("Checkpoint is not supported, skipping test.") - return + t.Skip("Checkpoint is not supported.") } - if err := dockerutil.Pull(img); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("pause-resume-test") - if err := d.Run("-p", "8080", img); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Ports: []int{8080}, // See Dockerfile. + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatal("docker.FindPort(8080) failed:", err) + t.Fatalf("docker.FindPort(8080) failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check that container is working. @@ -121,7 +123,7 @@ func TestPauseResume(t *testing.T) { } if err := d.Pause(); err != nil { - t.Fatal("docker pause failed:", err) + t.Fatalf("docker pause failed: %v", err) } // Check if container is paused. @@ -137,12 +139,12 @@ func TestPauseResume(t *testing.T) { } if err := d.Unpause(); err != nil { - t.Fatal("docker unpause failed:", err) + t.Fatalf("docker unpause failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. @@ -152,43 +154,43 @@ func TestPauseResume(t *testing.T) { } func TestCheckpointRestore(t *testing.T) { - const img = "gcr.io/gvisor-presubmit/python-hello" if !testutil.IsCheckpointSupported() { - t.Log("Pause/resume is not supported, skipping test.") - return + t.Skip("Pause/resume is not supported.") } - if err := dockerutil.Pull(img); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("save-restore-test") - if err := d.Run("-p", "8080", img); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Ports: []int{8080}, // See Dockerfile. + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Create a snapshot. if err := d.Checkpoint("test"); err != nil { - t.Fatal("docker checkpoint failed:", err) + t.Fatalf("docker checkpoint failed: %v", err) } - if _, err := d.Wait(30 * time.Second); err != nil { - t.Fatal(err) + t.Fatalf("wait failed: %v", err) } // TODO(b/143498576): Remove Poll after github.com/moby/moby/issues/38963 is fixed. if err := testutil.Poll(func() error { return d.Restore("test") }, 15*time.Second); err != nil { - t.Fatal("docker restore failed:", err) + t.Fatalf("docker restore failed: %v", err) } // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatal("docker.FindPort(8080) failed:", err) + t.Fatalf("docker.FindPort(8080) failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. @@ -200,26 +202,28 @@ func TestCheckpointRestore(t *testing.T) { // Create client and server that talk to each other using the local IP. func TestConnectToSelf(t *testing.T) { - d := dockerutil.MakeDocker("connect-to-self-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Creates server that replies "server" and exists. Sleeps at the end because // 'docker exec' gets killed if the init process exists before it can finish. - if err := d.Run("ubuntu:trusty", "/bin/sh", "-c", "echo server | nc -l -p 8080 && sleep 1"); err != nil { - t.Fatal("docker run failed:", err) + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/ubuntu", + }, "/bin/sh", "-c", "echo server | nc -l -p 8080 && sleep 1"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Finds IP address for host. - ip, err := d.Exec("/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") + ip, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") if err != nil { - t.Fatal("docker exec failed:", err) + t.Fatalf("docker exec failed: %v", err) } ip = strings.TrimRight(ip, "\n") // Runs client that sends "client" to the server and exits. - reply, err := d.Exec("/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) + reply, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) if err != nil { - t.Fatal("docker exec failed:", err) + t.Fatalf("docker exec failed: %v", err) } // Ensure both client and server got the message from each other. @@ -227,21 +231,22 @@ func TestConnectToSelf(t *testing.T) { t.Errorf("Error on server, want: %q, got: %q", want, reply) } if _, err := d.WaitForOutput("^client\n$", 1*time.Second); err != nil { - t.Fatal("docker.WaitForOutput(client) timeout:", err) + t.Fatalf("docker.WaitForOutput(client) timeout: %v", err) } } func TestMemLimit(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") - cmd := "cat /proc/meminfo | grep MemTotal: | awk '{print $2}'" - out, err := d.RunFg("--memory=500MB", "alpine", "sh", "-c", cmd) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + allocMemory := 500 * 1024 + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Memory: allocMemory, // In kB. + }, "sh", "-c", "cat /proc/meminfo | grep MemTotal: | awk '{print $2}'") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Remove warning message that swap isn't present. if strings.HasPrefix(out, "WARNING") { @@ -252,27 +257,30 @@ func TestMemLimit(t *testing.T) { out = lines[1] } + // Ensure the memory matches what we want. got, err := strconv.ParseUint(strings.TrimSpace(out), 10, 64) if err != nil { t.Fatalf("failed to parse %q: %v", out, err) } - if want := uint64(500 * 1024); got != want { + if want := uint64(allocMemory); got != want { t.Errorf("MemTotal got: %d, want: %d", got, want) } } func TestNumCPU(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") - cmd := "cat /proc/cpuinfo | grep 'processor.*:' | wc -l" - out, err := d.RunFg("--cpuset-cpus=0", "alpine", "sh", "-c", cmd) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Read how many cores are in the container. + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: []string{"--cpuset-cpus=0"}, + }, "sh", "-c", "cat /proc/cpuinfo | grep 'processor.*:' | wc -l") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Ensure it matches what we want. got, err := strconv.Atoi(strings.TrimSpace(out)) if err != nil { t.Fatalf("failed to parse %q: %v", out, err) @@ -284,39 +292,39 @@ func TestNumCPU(t *testing.T) { // TestJobControl tests that job control characters are handled properly. func TestJobControl(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("job-control-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with an attached PTY. - _, ptmx, err := d.RunWithPty("alpine", "sh") - if err != nil { + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Pty: func(_ *exec.Cmd, ptmx *os.File) { + // Call "sleep 100" in the shell. + if _, err := ptmx.Write([]byte("sleep 100\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Give shell a few seconds to start executing the sleep. + time.Sleep(2 * time.Second) + + // Send a ^C to the pty, which should kill sleep, but + // not the shell. \x03 is ASCII "end of text", which + // is the same as ^C. + if _, err := ptmx.Write([]byte{'\x03'}); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // The shell should still be alive at this point. Sleep + // should have exited with code 2+128=130. We'll exit + // with 10 plus that number, so that we can be sure + // that the shell did not get signalled. + if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + }, + }, "sh"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer ptmx.Close() - defer d.CleanUp() - - // Call "sleep 100" in the shell. - if _, err := ptmx.Write([]byte("sleep 100\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Give shell a few seconds to start executing the sleep. - time.Sleep(2 * time.Second) - - // Send a ^C to the pty, which should kill sleep, but not the shell. - // \x03 is ASCII "end of text", which is the same as ^C. - if _, err := ptmx.Write([]byte{'\x03'}); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // The shell should still be alive at this point. Sleep should have - // exited with code 2+128=130. We'll exit with 10 plus that number, so - // that we can be sure that the shell did not get signalled. - if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } // Wait for the container to exit. got, err := d.Wait(5 * time.Second) @@ -332,14 +340,25 @@ func TestJobControl(t *testing.T) { // TestTmpFile checks that files inside '/tmp' are not overridden. In addition, // it checks that working dir is created if it doesn't exit. func TestTmpFile(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Should work without ReadOnly + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + WorkDir: "/tmp/foo/bar", + }, "touch", "/tmp/foo/bar/file"); err != nil { + t.Fatalf("docker run failed: %v", err) } - d := dockerutil.MakeDocker("tmp-file-test") - if err := d.Run("-w=/tmp/foo/bar", "--read-only", "alpine", "touch", "/tmp/foo/bar/file"); err != nil { - t.Fatal("docker run failed:", err) + + // Expect failure. + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + WorkDir: "/tmp/foo/bar", + ReadOnly: true, + }, "touch", "/tmp/foo/bar/file"); err == nil { + t.Fatalf("docker run expected failure, but succeeded") } - defer d.CleanUp() } func TestMain(m *testing.M) { diff --git a/test/e2e/regression_test.go b/test/e2e/regression_test.go index 2488be383..327a2174c 100644 --- a/test/e2e/regression_test.go +++ b/test/e2e/regression_test.go @@ -18,7 +18,7 @@ import ( "strings" "testing" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" ) // Test that UDS can be created using overlay when parent directory is in lower @@ -27,19 +27,19 @@ import ( // Prerequisite: the directory where the socket file is created must not have // been open for write before bind(2) is called. func TestBindOverlay(t *testing.T) { - if err := dockerutil.Pull("ubuntu:trusty"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("bind-overlay-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - cmd := "nc -l -U /var/run/sock & p=$! && sleep 1 && echo foobar-asdf | nc -U /var/run/sock && wait $p" - got, err := d.RunFg("ubuntu:trusty", "bash", "-c", cmd) + // Run the container. + got, err := d.Run(dockerutil.RunOpts{ + Image: "basic/ubuntu", + }, "bash", "-c", "nc -l -U /var/run/sock & p=$! && sleep 1 && echo foobar-asdf | nc -U /var/run/sock && wait $p") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } + // Check the output contains what we want. if want := "foobar-asdf"; !strings.Contains(got, want) { t.Fatalf("docker run output is missing %q: %s", want, got) } - defer d.CleanUp() } diff --git a/test/image/BUILD b/test/image/BUILD index 7392ac54e..e749e47d4 100644 --- a/test/image/BUILD +++ b/test/image/BUILD @@ -22,8 +22,8 @@ go_test( ], visibility = ["//:sandbox"], deps = [ - "//runsc/dockerutil", - "//runsc/testutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", ], ) diff --git a/test/image/image_test.go b/test/image/image_test.go index 0a1e19d6f..2e3543109 100644 --- a/test/image/image_test.go +++ b/test/image/image_test.go @@ -28,24 +28,29 @@ import ( "log" "net/http" "os" - "path/filepath" "strings" "testing" "time" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) func TestHelloWorld(t *testing.T) { - d := dockerutil.MakeDocker("hello-test") - if err := d.Run("hello-world"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Run the basic container. + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "echo", "Hello world!") + if err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() - if _, err := d.WaitForOutput("Hello from Docker!", 5*time.Second); err != nil { - t.Fatalf("docker didn't say hello: %v", err) + // Check the output. + if !strings.Contains(out, "Hello world!") { + t.Fatalf("docker didn't say hello: got %s", out) } } @@ -102,27 +107,22 @@ func testHTTPServer(t *testing.T, port int) { } func TestHttpd(t *testing.T) { - if err := dockerutil.Pull("httpd"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("http-test") - - dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - mountArg := dockerutil.MountArg(dir, "/usr/local/apache2/htdocs", dockerutil.ReadOnly) - if err := d.Run("-p", "80", mountArg, "httpd"); err != nil { + d.CopyFiles("/usr/local/apache2/htdocs", "test/image/latin10k.txt") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/httpd", + Ports: []int{80}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 80 is mapped to. port, err := d.FindPort(80) if err != nil { - t.Fatalf("docker.FindPort(80) failed: %v", err) + t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. @@ -134,27 +134,22 @@ func TestHttpd(t *testing.T) { } func TestNginx(t *testing.T) { - if err := dockerutil.Pull("nginx"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("net-test") - - dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - mountArg := dockerutil.MountArg(dir, "/usr/share/nginx/html", dockerutil.ReadOnly) - if err := d.Run("-p", "80", mountArg, "nginx"); err != nil { + d.CopyFiles("/usr/share/nginx/html", "test/image/latin10k.txt") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/nginx", + Ports: []int{80}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 80 is mapped to. port, err := d.FindPort(80) if err != nil { - t.Fatalf("docker.FindPort(80) failed: %v", err) + t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. @@ -166,99 +161,58 @@ func TestNginx(t *testing.T) { } func TestMysql(t *testing.T) { - if err := dockerutil.Pull("mysql"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("mysql-test") + server := dockerutil.MakeDocker(t) + defer server.CleanUp() // Start the container. - if err := d.Run("-e", "MYSQL_ROOT_PASSWORD=foobar123", "mysql"); err != nil { + if err := server.Spawn(dockerutil.RunOpts{ + Image: "basic/mysql", + Env: []string{"MYSQL_ROOT_PASSWORD=foobar123"}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Wait until it's up and running. - if _, err := d.WaitForOutput("port: 3306 MySQL Community Server", 3*time.Minute); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) + if _, err := server.WaitForOutput("port: 3306 MySQL Community Server", 3*time.Minute); err != nil { + t.Fatalf("WaitForOutput() timeout: %v", err) } - client := dockerutil.MakeDocker("mysql-client-test") - dir, err := dockerutil.PrepareFiles("test/image/mysql.sql") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + // Generate the client and copy in the SQL payload. + client := dockerutil.MakeDocker(t) + defer client.CleanUp() - // Tell mysql client to connect to the server and execute the file in verbose - // mode to verify the output. - args := []string{ - dockerutil.LinkArg(&d, "mysql"), - dockerutil.MountArg(dir, "/sql", dockerutil.ReadWrite), - "mysql", - "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql", - } - if err := client.Run(args...); err != nil { + // Tell mysql client to connect to the server and execute the file in + // verbose mode to verify the output. + client.CopyFiles("/sql", "test/image/mysql.sql") + client.Link("mysql", server) + if _, err := client.Run(dockerutil.RunOpts{ + Image: "basic/mysql", + }, "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer client.CleanUp() // Ensure file executed to the end and shutdown mysql. - if _, err := client.WaitForOutput("--------------\nshutdown\n--------------", 15*time.Second); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) - } - if _, err := d.WaitForOutput("mysqld: Shutdown complete", 30*time.Second); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) + if _, err := server.WaitForOutput("mysqld: Shutdown complete", 30*time.Second); err != nil { + t.Fatalf("WaitForOutput() timeout: %v", err) } } -func TestPythonHello(t *testing.T) { - // TODO(b/136503277): Once we have more complete python runtime tests, - // we can drop this one. - const img = "gcr.io/gvisor-presubmit/python-hello" - if err := dockerutil.Pull(img); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("python-hello-test") - if err := d.Run("-p", "8080", img); err != nil { - t.Fatalf("docker run failed: %v", err) - } +func TestTomcat(t *testing.T) { + d := dockerutil.MakeDocker(t) defer d.CleanUp() - // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) - if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) - } - - // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatalf("WaitForHTTP() timeout: %v", err) - } - - // Ensure that content is being served. - url := fmt.Sprintf("http://localhost:%d", port) - resp, err := http.Get(url) - if err != nil { - t.Errorf("Error reaching http server: %v", err) - } - if want := http.StatusOK; resp.StatusCode != want { - t.Errorf("Wrong response code, got: %d, want: %d", resp.StatusCode, want) - } -} - -func TestTomcat(t *testing.T) { - if err := dockerutil.Pull("tomcat:8.0"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("tomcat-test") - if err := d.Run("-p", "8080", "tomcat:8.0"); err != nil { + // Start the server. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/tomcat", + Ports: []int{8080}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) + t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running. @@ -278,28 +232,22 @@ func TestTomcat(t *testing.T) { } func TestRuby(t *testing.T) { - if err := dockerutil.Pull("ruby"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("ruby-test") - - dir, err := dockerutil.PrepareFiles("test/image/ruby.rb", "test/image/ruby.sh") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } - if err := os.Chmod(filepath.Join(dir, "ruby.sh"), 0333); err != nil { - t.Fatalf("os.Chmod(%q, 0333) failed: %v", dir, err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - if err := d.Run("-p", "8080", dockerutil.MountArg(dir, "/src", dockerutil.ReadOnly), "ruby", "/src/ruby.sh"); err != nil { + // Execute the ruby workload. + d.CopyFiles("/src", "test/image/ruby.rb", "test/image/ruby.sh") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/ruby", + Ports: []int{8080}, + }, "/src/ruby.sh"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) + t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running, 'gem install' can take some time. @@ -326,18 +274,17 @@ func TestRuby(t *testing.T) { } func TestStdio(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("stdio-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() wantStdout := "hello stdout" wantStderr := "bonjour stderr" cmd := fmt.Sprintf("echo %q; echo %q 1>&2;", wantStdout, wantStderr) - if err := d.Run("alpine", "/bin/sh", "-c", cmd); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "/bin/sh", "-c", cmd); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() for _, want := range []string{wantStdout, wantStderr} { if _, err := d.WaitForOutput(want, 5*time.Second); err != nil { diff --git a/test/image/ruby.sh b/test/image/ruby.sh index ebe8d5b0e..ebe8d5b0e 100644..100755 --- a/test/image/ruby.sh +++ b/test/image/ruby.sh diff --git a/test/iptables/BUILD b/test/iptables/BUILD index 6bb3b82b5..3e29ca90d 100644 --- a/test/iptables/BUILD +++ b/test/iptables/BUILD @@ -14,7 +14,7 @@ go_library( ], visibility = ["//test/iptables:__subpackages__"], deps = [ - "//runsc/testutil", + "//pkg/test/testutil", ], ) @@ -23,14 +23,14 @@ go_test( srcs = [ "iptables_test.go", ], + data = ["//test/iptables/runner"], library = ":iptables", tags = [ "local", "manual", ], deps = [ - "//pkg/log", - "//runsc/dockerutil", - "//runsc/testutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", ], ) diff --git a/test/iptables/README.md b/test/iptables/README.md index cc8a2fcac..b9f44bd40 100644 --- a/test/iptables/README.md +++ b/test/iptables/README.md @@ -38,7 +38,7 @@ Build the testing Docker container. Re-run this when you modify the test code in this directory: ```bash -$ bazel run //test/iptables/runner:runner-image -- --norun +$ make load-iptables ``` Run an individual test via: diff --git a/test/iptables/iptables.go b/test/iptables/iptables.go index 2e565d988..16cb4f4da 100644 --- a/test/iptables/iptables.go +++ b/test/iptables/iptables.go @@ -18,12 +18,19 @@ package iptables import ( "fmt" "net" + "time" ) // IPExchangePort is the port the container listens on to receive the IP // address of the local process. const IPExchangePort = 2349 +// TerminalStatement is the last statement in the test runner. +const TerminalStatement = "Finished!" + +// TestTimeout is the timeout used for all tests. +const TestTimeout = 10 * time.Minute + // 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. diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 493d69052..334d8e676 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -15,28 +15,14 @@ package iptables import ( - "flag" "fmt" "net" - "os" - "path" "testing" - "time" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) -const timeout = 18 * time.Second - -var image = flag.String("image", "bazel/test/iptables/runner:runner-image", "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. @@ -46,77 +32,45 @@ type result struct { // // Container output is logged to $TEST_UNDECLARED_OUTPUTS_DIR if it exists, or // to stderr. -func singleTest(test TestCase) error { +func singleTest(t *testing.T, test TestCase) { if _, ok := Tests[test.Name()]; !ok { - return fmt.Errorf("no test found with name %q. Has it been registered?", test.Name()) + t.Fatalf("no test found with name %q. Has it been registered?", test.Name()) } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + // 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} - }() + d.CopyFiles("/runner", "test/iptables/runner/runner") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "iptables", + CapAdd: []string{"NET_ADMIN"}, + }, "/runner/runner", "-name", test.Name()); err != nil { + t.Fatalf("docker run failed: %v", err) + } // Get the container IP. - ip, err := getIP(cont) + ip, err := d.FindIP() if err != nil { - return fmt.Errorf("failed to get container IP: %v", err) + t.Fatalf("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) + t.Fatalf("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()) - } + if err := test.LocalAction(ip); err != nil { + t.Fatalf("LocalAction failed: %v", err) } - 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) + // Wait for the final statement. This structure has the side effect + // that all container logs will appear within the individual test + // context. + if _, err := d.WaitForOutput(TerminalStatement, TestTimeout); err != nil { + t.Fatalf("test failed: %v", err) } - log.Infof("Container has IP of %s", ipStr) - return ip, nil } func sendIP(ip net.IP) error { @@ -132,7 +86,7 @@ func sendIP(ip net.IP) error { conn = c return err } - if err := testutil.Poll(cb, timeout); err != nil { + if err := testutil.Poll(cb, TestTimeout); err != nil { return fmt.Errorf("timed out waiting to send IP, most recent error: %v", err) } if _, err := conn.Write([]byte{0}); err != nil { @@ -141,281 +95,184 @@ func sendIP(ip net.IP) error { 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) - } + singleTest(t, FilterInputDropUDP{}) } func TestFilterInputDropUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropUDPPort{}) } func TestFilterInputDropDifferentUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropDifferentUDPPort{}) } func TestFilterInputDropAll(t *testing.T) { - if err := singleTest(FilterInputDropAll{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropAll{}) } func TestFilterInputDropOnlyUDP(t *testing.T) { - if err := singleTest(FilterInputDropOnlyUDP{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropOnlyUDP{}) } func TestNATRedirectUDPPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectUDPPort{}) } func TestNATRedirectTCPPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectTCPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectTCPPort{}) } func TestNATDropUDP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATDropUDP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATDropUDP{}) } func TestNATAcceptAll(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATAcceptAll{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATAcceptAll{}) } func TestFilterInputDropTCPDestPort(t *testing.T) { - if err := singleTest(FilterInputDropTCPDestPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropTCPDestPort{}) } func TestFilterInputDropTCPSrcPort(t *testing.T) { - if err := singleTest(FilterInputDropTCPSrcPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropTCPSrcPort{}) } func TestFilterInputCreateUserChain(t *testing.T) { - if err := singleTest(FilterInputCreateUserChain{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputCreateUserChain{}) } func TestFilterInputDefaultPolicyAccept(t *testing.T) { - if err := singleTest(FilterInputDefaultPolicyAccept{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDefaultPolicyAccept{}) } func TestFilterInputDefaultPolicyDrop(t *testing.T) { - if err := singleTest(FilterInputDefaultPolicyDrop{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDefaultPolicyDrop{}) } func TestFilterInputReturnUnderflow(t *testing.T) { - if err := singleTest(FilterInputReturnUnderflow{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputReturnUnderflow{}) } func TestFilterOutputDropTCPDestPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(FilterOutputDropTCPDestPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPDestPort{}) } func TestFilterOutputDropTCPSrcPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(FilterOutputDropTCPSrcPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPSrcPort{}) } func TestFilterOutputAcceptTCPOwner(t *testing.T) { - if err := singleTest(FilterOutputAcceptTCPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputAcceptTCPOwner{}) } func TestFilterOutputDropTCPOwner(t *testing.T) { - if err := singleTest(FilterOutputDropTCPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPOwner{}) } func TestFilterOutputAcceptUDPOwner(t *testing.T) { - if err := singleTest(FilterOutputAcceptUDPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputAcceptUDPOwner{}) } func TestFilterOutputDropUDPOwner(t *testing.T) { - if err := singleTest(FilterOutputDropUDPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropUDPOwner{}) } func TestFilterOutputOwnerFail(t *testing.T) { - if err := singleTest(FilterOutputOwnerFail{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputOwnerFail{}) } func TestJumpSerialize(t *testing.T) { - if err := singleTest(FilterInputSerializeJump{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputSerializeJump{}) } func TestJumpBasic(t *testing.T) { - if err := singleTest(FilterInputJumpBasic{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpBasic{}) } func TestJumpReturn(t *testing.T) { - if err := singleTest(FilterInputJumpReturn{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpReturn{}) } func TestJumpReturnDrop(t *testing.T) { - if err := singleTest(FilterInputJumpReturnDrop{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpReturnDrop{}) } func TestJumpBuiltin(t *testing.T) { - if err := singleTest(FilterInputJumpBuiltin{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpBuiltin{}) } func TestJumpTwice(t *testing.T) { - if err := singleTest(FilterInputJumpTwice{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpTwice{}) } func TestInputDestination(t *testing.T) { - if err := singleTest(FilterInputDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDestination{}) } func TestInputInvertDestination(t *testing.T) { - if err := singleTest(FilterInputInvertDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputInvertDestination{}) } func TestOutputDestination(t *testing.T) { - if err := singleTest(FilterOutputDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDestination{}) } func TestOutputInvertDestination(t *testing.T) { - if err := singleTest(FilterOutputInvertDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputInvertDestination{}) } func TestNATOutRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutRedirectIP{}) } func TestNATOutDontRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutDontRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutDontRedirectIP{}) } func TestNATOutRedirectInvert(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutRedirectInvert{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutRedirectInvert{}) } func TestNATPreRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreRedirectIP{}) } func TestNATPreDontRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreDontRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreDontRedirectIP{}) } func TestNATPreRedirectInvert(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreRedirectInvert{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreRedirectInvert{}) } func TestNATRedirectRequiresProtocol(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectRequiresProtocol{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectRequiresProtocol{}) } diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 134391e8d..2a00677be 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -20,7 +20,7 @@ import ( "os/exec" "time" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) const iptablesBinary = "iptables" diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD index b9199387a..24504a1b9 100644 --- a/test/iptables/runner/BUILD +++ b/test/iptables/runner/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "container_image", "go_binary", "go_image") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) @@ -6,18 +6,7 @@ go_binary( name = "runner", testonly = 1, srcs = ["main.go"], - deps = ["//test/iptables"], -) - -container_image( - name = "iptables-base", - base = "@iptables-test//image", -) - -go_image( - name = "runner-image", - testonly = 1, - srcs = ["main.go"], - base = ":iptables-base", + pure = True, + visibility = ["//test/iptables:__subpackages__"], deps = ["//test/iptables"], ) diff --git a/test/iptables/runner/main.go b/test/iptables/runner/main.go index 3c794114e..6f77c0684 100644 --- a/test/iptables/runner/main.go +++ b/test/iptables/runner/main.go @@ -46,6 +46,9 @@ func main() { if err := test.ContainerAction(ip); err != nil { log.Fatalf("Failed running test %q: %v", *name, err) } + + // Emit the final line. + log.Printf("%s", iptables.TerminalStatement) } // getIP listens for a connection from the local process and returns the source diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh index c8268170f..922547d65 100755 --- a/test/packetdrill/packetdrill_test.sh +++ b/test/packetdrill/packetdrill_test.sh @@ -85,23 +85,26 @@ if [[ ! -x "${INIT_SCRIPT-}" ]]; then exit 2 fi +function new_net_prefix() { + # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. + echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" +} + # Variables specific to the control network and interface start with CTRL_. # Variables specific to the test network and interface start with TEST_. # Variables specific to the DUT start with DUT_. # Variables specific to the test runner start with TEST_RUNNER_. declare -r PACKETDRILL="/packetdrill/gtests/net/packetdrill/packetdrill" # Use random numbers so that test networks don't collide. -declare -r CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" -declare -r TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" +declare CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" +declare CTRL_NET_PREFIX=$(new_net_prefix) +declare TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" +declare TEST_NET_PREFIX=$(new_net_prefix) declare -r tolerance_usecs=100000 # On both DUT and test runner, testing packets are on the eth2 interface. declare -r TEST_DEVICE="eth2" # Number of bits in the *_NET_PREFIX variables. declare -r NET_MASK="24" -function new_net_prefix() { - # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. - echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" -} # Last bits of the DUT's IP address. declare -r DUT_NET_SUFFIX=".10" # Control port. @@ -137,23 +140,21 @@ function finish { trap finish EXIT # Subnet for control packets between test runner and DUT. -declare CTRL_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do sleep 0.1 - declare CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" done # Subnet for the packets that are part of the test. -declare TEST_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do sleep 0.1 - declare TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" done -docker pull "${IMAGE_TAG}" - # Create the DUT container and connect to network. DUT=$(docker create ${RUNTIME_ARG} --privileged --rm \ --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG}) diff --git a/test/packetimpact/testbench/BUILD b/test/packetimpact/testbench/BUILD index b6a254882..3ceceb9d7 100644 --- a/test/packetimpact/testbench/BUILD +++ b/test/packetimpact/testbench/BUILD @@ -23,7 +23,6 @@ go_library( "//test/packetimpact/proto:posix_server_go_proto", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", - "@com_github_imdario_mergo//:go_default_library", "@com_github_mohae_deepcopy//:go_default_library", "@org_golang_google_grpc//:go_default_library", "@org_golang_google_grpc//keepalive:go_default_library", @@ -37,5 +36,8 @@ go_test( size = "small", srcs = ["layers_test.go"], library = ":testbench", - deps = ["//pkg/tcpip"], + deps = [ + "//pkg/tcpip", + "@com_github_mohae_deepcopy//:go_default_library", + ], ) diff --git a/test/packetimpact/testbench/connections.go b/test/packetimpact/testbench/connections.go index f84fd8ba7..952a717e0 100644 --- a/test/packetimpact/testbench/connections.go +++ b/test/packetimpact/testbench/connections.go @@ -72,7 +72,8 @@ type layerState interface { // incoming creates an expected Layer for comparing against a received Layer. // Because the expectation can depend on values in the received Layer, it is // an input to incoming. For example, the ACK number needs to be checked in a - // TCP packet but only if the ACK flag is set in the received packet. + // TCP packet but only if the ACK flag is set in the received packet. The + // calles takes ownership of the returned Layer. incoming(received Layer) Layer // sent updates the layerState based on the Layer that was sent. The input is @@ -124,6 +125,7 @@ func (s *etherState) outgoing() Layer { return &s.out } +// incoming implements layerState.incoming. func (s *etherState) incoming(Layer) Layer { return deepcopy.Copy(&s.in).(Layer) } @@ -168,6 +170,7 @@ func (s *ipv4State) outgoing() Layer { return &s.out } +// incoming implements layerState.incoming. func (s *ipv4State) incoming(Layer) Layer { return deepcopy.Copy(&s.in).(Layer) } @@ -234,6 +237,7 @@ func (s *tcpState) outgoing() Layer { return &newOutgoing } +// incoming implements layerState.incoming. func (s *tcpState) incoming(received Layer) Layer { tcpReceived, ok := received.(*TCP) if !ok { @@ -328,6 +332,7 @@ func (s *udpState) outgoing() Layer { return &s.out } +// incoming implements layerState.incoming. func (s *udpState) incoming(Layer) Layer { return deepcopy.Copy(&s.in).(Layer) } @@ -363,16 +368,33 @@ type Connection struct { // reverse is never a match. override overrides the default matchers for each // Layer. func (conn *Connection) match(override, received Layers) bool { - if len(received) < len(conn.layerStates) { + var layersToMatch int + if len(override) < len(conn.layerStates) { + layersToMatch = len(conn.layerStates) + } else { + layersToMatch = len(override) + } + if len(received) < layersToMatch { return false } - for i, s := range conn.layerStates { - toMatch := s.incoming(received[i]) - if toMatch == nil { - return false - } - if i < len(override) { - toMatch.merge(override[i]) + for i := 0; i < layersToMatch; i++ { + var toMatch Layer + if i < len(conn.layerStates) { + s := conn.layerStates[i] + toMatch = s.incoming(received[i]) + if toMatch == nil { + return false + } + if i < len(override) { + if err := toMatch.merge(override[i]); err != nil { + conn.t.Fatalf("failed to merge: %s", err) + } + } + } else { + toMatch = override[i] + if toMatch == nil { + conn.t.Fatalf("expect the overriding layers to be non-nil") + } } if !toMatch.match(received[i]) { return false diff --git a/test/packetimpact/testbench/dut.go b/test/packetimpact/testbench/dut.go index 9335909c0..3f340c6bc 100644 --- a/test/packetimpact/testbench/dut.go +++ b/test/packetimpact/testbench/dut.go @@ -132,7 +132,7 @@ func (dut *DUT) CreateBoundSocket(typ, proto int32, addr net.IP) (int32, uint16) copy(sa.Addr[:], addr.To16()) dut.Bind(fd, &sa) } else { - dut.t.Fatal("unknown ip addr type for remoteIP") + dut.t.Fatalf("unknown ip addr type for remoteIP") } sa := dut.GetSockName(fd) var port int diff --git a/test/packetimpact/testbench/layers.go b/test/packetimpact/testbench/layers.go index 5ce324f0d..01e99567d 100644 --- a/test/packetimpact/testbench/layers.go +++ b/test/packetimpact/testbench/layers.go @@ -22,7 +22,6 @@ import ( "github.com/google/go-cmp/cmp" "github.com/google/go-cmp/cmp/cmpopts" - "github.com/imdario/mergo" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -111,13 +110,31 @@ func equalLayer(x, y Layer) bool { return cmp.Equal(x, y, opt, cmpopts.IgnoreTypes(LayerBase{})) } -// mergeLayer merges other in layer. Any non-nil value in other overrides the -// corresponding value in layer. If other is nil, no action is performed. -func mergeLayer(layer, other Layer) error { - if other == nil { +// mergeLayer merges y into x. Any fields for which y has a non-nil value, that +// value overwrite the corresponding fields in x. +func mergeLayer(x, y Layer) error { + if y == nil { return nil } - return mergo.Merge(layer, other, mergo.WithOverride) + if reflect.TypeOf(x) != reflect.TypeOf(y) { + return fmt.Errorf("can't merge %T into %T", y, x) + } + vx := reflect.ValueOf(x).Elem() + vy := reflect.ValueOf(y).Elem() + t := vy.Type() + for i := 0; i < vy.NumField(); i++ { + t := t.Field(i) + if t.Anonymous { + // Ignore the LayerBase in the Layer struct. + continue + } + v := vy.Field(i) + if v.IsNil() { + continue + } + vx.Field(i).Set(v) + } + return nil } func stringLayer(l Layer) string { @@ -243,8 +260,7 @@ func (l *Ether) length() int { return header.EthernetMinimumSize } -// merge overrides the values in l with the values from other but only in fields -// where the value is not nil. +// merge implements Layer.merge. func (l *Ether) merge(other Layer) error { return mergeLayer(l, other) } @@ -399,8 +415,7 @@ func (l *IPv4) length() int { return int(*l.IHL) } -// merge overrides the values in l with the values from other but only in fields -// where the value is not nil. +// merge implements Layer.merge. func (l *IPv4) merge(other Layer) error { return mergeLayer(l, other) } @@ -544,8 +559,7 @@ func (l *TCP) length() int { return int(*l.DataOffset) } -// merge overrides the values in l with the values from other but only in fields -// where the value is not nil. +// merge implements Layer.merge. func (l *TCP) merge(other Layer) error { return mergeLayer(l, other) } @@ -622,8 +636,7 @@ func (l *UDP) length() int { return int(*l.Length) } -// merge overrides the values in l with the values from other but only in fields -// where the value is not nil. +// merge implements Layer.merge. func (l *UDP) merge(other Layer) error { return mergeLayer(l, other) } @@ -659,8 +672,7 @@ func (l *Payload) length() int { return len(l.Bytes) } -// merge overrides the values in l with the values from other but only in fields -// where the value is not nil. +// merge implements Layer.merge. func (l *Payload) merge(other Layer) error { return mergeLayer(l, other) } diff --git a/test/packetimpact/testbench/layers_test.go b/test/packetimpact/testbench/layers_test.go index b32efda93..f07ec5eb2 100644 --- a/test/packetimpact/testbench/layers_test.go +++ b/test/packetimpact/testbench/layers_test.go @@ -17,6 +17,7 @@ package testbench import ( "testing" + "github.com/mohae/deepcopy" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -52,6 +53,114 @@ func TestLayerMatch(t *testing.T) { } } +func TestLayerMergeMismatch(t *testing.T) { + tcp := &TCP{} + otherTCP := &TCP{} + ipv4 := &IPv4{} + ether := &Ether{} + for _, tt := range []struct { + a, b Layer + success bool + }{ + {tcp, tcp, true}, + {tcp, otherTCP, true}, + {tcp, ipv4, false}, + {tcp, ether, false}, + {tcp, nil, true}, + + {otherTCP, otherTCP, true}, + {otherTCP, ipv4, false}, + {otherTCP, ether, false}, + {otherTCP, nil, true}, + + {ipv4, ipv4, true}, + {ipv4, ether, false}, + {ipv4, nil, true}, + + {ether, ether, true}, + {ether, nil, true}, + } { + if err := tt.a.merge(tt.b); (err == nil) != tt.success { + t.Errorf("%s.merge(%s) got %s, wanted the opposite", tt.a, tt.b, err) + } + if tt.b != nil { + if err := tt.b.merge(tt.a); (err == nil) != tt.success { + t.Errorf("%s.merge(%s) got %s, wanted the opposite", tt.b, tt.a, err) + } + } + } +} + +func TestLayerMerge(t *testing.T) { + zero := Uint32(0) + one := Uint32(1) + two := Uint32(2) + empty := []byte{} + foo := []byte("foo") + bar := []byte("bar") + for _, tt := range []struct { + a, b Layer + want Layer + }{ + {&TCP{AckNum: nil}, &TCP{AckNum: nil}, &TCP{AckNum: nil}}, + {&TCP{AckNum: nil}, &TCP{AckNum: zero}, &TCP{AckNum: zero}}, + {&TCP{AckNum: nil}, &TCP{AckNum: one}, &TCP{AckNum: one}}, + {&TCP{AckNum: nil}, &TCP{AckNum: two}, &TCP{AckNum: two}}, + {&TCP{AckNum: nil}, nil, &TCP{AckNum: nil}}, + + {&TCP{AckNum: zero}, &TCP{AckNum: nil}, &TCP{AckNum: zero}}, + {&TCP{AckNum: zero}, &TCP{AckNum: zero}, &TCP{AckNum: zero}}, + {&TCP{AckNum: zero}, &TCP{AckNum: one}, &TCP{AckNum: one}}, + {&TCP{AckNum: zero}, &TCP{AckNum: two}, &TCP{AckNum: two}}, + {&TCP{AckNum: zero}, nil, &TCP{AckNum: zero}}, + + {&TCP{AckNum: one}, &TCP{AckNum: nil}, &TCP{AckNum: one}}, + {&TCP{AckNum: one}, &TCP{AckNum: zero}, &TCP{AckNum: zero}}, + {&TCP{AckNum: one}, &TCP{AckNum: one}, &TCP{AckNum: one}}, + {&TCP{AckNum: one}, &TCP{AckNum: two}, &TCP{AckNum: two}}, + {&TCP{AckNum: one}, nil, &TCP{AckNum: one}}, + + {&TCP{AckNum: two}, &TCP{AckNum: nil}, &TCP{AckNum: two}}, + {&TCP{AckNum: two}, &TCP{AckNum: zero}, &TCP{AckNum: zero}}, + {&TCP{AckNum: two}, &TCP{AckNum: one}, &TCP{AckNum: one}}, + {&TCP{AckNum: two}, &TCP{AckNum: two}, &TCP{AckNum: two}}, + {&TCP{AckNum: two}, nil, &TCP{AckNum: two}}, + + {&Payload{Bytes: nil}, &Payload{Bytes: nil}, &Payload{Bytes: nil}}, + {&Payload{Bytes: nil}, &Payload{Bytes: empty}, &Payload{Bytes: empty}}, + {&Payload{Bytes: nil}, &Payload{Bytes: foo}, &Payload{Bytes: foo}}, + {&Payload{Bytes: nil}, &Payload{Bytes: bar}, &Payload{Bytes: bar}}, + {&Payload{Bytes: nil}, nil, &Payload{Bytes: nil}}, + + {&Payload{Bytes: empty}, &Payload{Bytes: nil}, &Payload{Bytes: empty}}, + {&Payload{Bytes: empty}, &Payload{Bytes: empty}, &Payload{Bytes: empty}}, + {&Payload{Bytes: empty}, &Payload{Bytes: foo}, &Payload{Bytes: foo}}, + {&Payload{Bytes: empty}, &Payload{Bytes: bar}, &Payload{Bytes: bar}}, + {&Payload{Bytes: empty}, nil, &Payload{Bytes: empty}}, + + {&Payload{Bytes: foo}, &Payload{Bytes: nil}, &Payload{Bytes: foo}}, + {&Payload{Bytes: foo}, &Payload{Bytes: empty}, &Payload{Bytes: empty}}, + {&Payload{Bytes: foo}, &Payload{Bytes: foo}, &Payload{Bytes: foo}}, + {&Payload{Bytes: foo}, &Payload{Bytes: bar}, &Payload{Bytes: bar}}, + {&Payload{Bytes: foo}, nil, &Payload{Bytes: foo}}, + + {&Payload{Bytes: bar}, &Payload{Bytes: nil}, &Payload{Bytes: bar}}, + {&Payload{Bytes: bar}, &Payload{Bytes: empty}, &Payload{Bytes: empty}}, + {&Payload{Bytes: bar}, &Payload{Bytes: foo}, &Payload{Bytes: foo}}, + {&Payload{Bytes: bar}, &Payload{Bytes: bar}, &Payload{Bytes: bar}}, + {&Payload{Bytes: bar}, nil, &Payload{Bytes: bar}}, + } { + a := deepcopy.Copy(tt.a).(Layer) + if err := a.merge(tt.b); err != nil { + t.Errorf("%s.merge(%s) = %s, wanted nil", tt.a, tt.b, err) + continue + } + if a.String() != tt.want.String() { + t.Errorf("%s.merge(%s) merge result got %s, want %s", tt.a, tt.b, a, tt.want) + } + } +} + func TestLayerStringFormat(t *testing.T) { for _, tt := range []struct { name string @@ -154,3 +263,53 @@ func TestLayerStringFormat(t *testing.T) { }) } } + +func TestConnectionMatch(t *testing.T) { + conn := Connection{ + layerStates: []layerState{ðerState{}}, + } + protoNum0 := tcpip.NetworkProtocolNumber(0) + protoNum1 := tcpip.NetworkProtocolNumber(1) + for _, tt := range []struct { + description string + override, received Layers + wantMatch bool + }{ + { + description: "shorter override", + override: []Layer{&Ether{}}, + received: []Layer{&Ether{}, &Payload{Bytes: []byte("hello")}}, + wantMatch: true, + }, + { + description: "longer override", + override: []Layer{&Ether{}, &Payload{Bytes: []byte("hello")}}, + received: []Layer{&Ether{}}, + wantMatch: false, + }, + { + description: "ether layer mismatch", + override: []Layer{&Ether{Type: &protoNum0}}, + received: []Layer{&Ether{Type: &protoNum1}}, + wantMatch: false, + }, + { + description: "both nil", + override: nil, + received: nil, + wantMatch: false, + }, + { + description: "nil override", + override: nil, + received: []Layer{&Ether{}}, + wantMatch: true, + }, + } { + t.Run(tt.description, func(t *testing.T) { + if gotMatch := conn.match(tt.override, tt.received); gotMatch != tt.wantMatch { + t.Fatalf("conn.match(%s, %s) = %t, want %t", tt.override, tt.received, gotMatch, tt.wantMatch) + } + }) + } +} diff --git a/test/packetimpact/tests/defs.bzl b/test/packetimpact/tests/defs.bzl index 8c0d058b2..27c5de375 100644 --- a/test/packetimpact/tests/defs.bzl +++ b/test/packetimpact/tests/defs.bzl @@ -59,7 +59,11 @@ _packetimpact_test = rule( PACKETIMPACT_TAGS = ["local", "manual"] -def packetimpact_linux_test(name, testbench_binary, **kwargs): +def packetimpact_linux_test( + name, + testbench_binary, + expect_failure = False, + **kwargs): """Add a packetimpact test on linux. Args: @@ -67,28 +71,37 @@ def packetimpact_linux_test(name, testbench_binary, **kwargs): testbench_binary: the testbench binary **kwargs: all the other args, forwarded to _packetimpact_test """ + expect_failure_flag = ["--expect_failure"] if expect_failure else [] _packetimpact_test( name = name + "_linux_test", testbench_binary = testbench_binary, - flags = ["--dut_platform", "linux"], + flags = ["--dut_platform", "linux"] + expect_failure_flag, tags = PACKETIMPACT_TAGS + ["packetimpact"], **kwargs ) -def packetimpact_netstack_test(name, testbench_binary, **kwargs): +def packetimpact_netstack_test( + name, + testbench_binary, + expect_failure = False, + **kwargs): """Add a packetimpact test on netstack. Args: name: name of the test testbench_binary: the testbench binary + expect_failure: the test must fail **kwargs: all the other args, forwarded to _packetimpact_test """ + expect_failure_flag = [] + if expect_failure: + expect_failure_flag = ["--expect_failure"] _packetimpact_test( name = name + "_netstack_test", testbench_binary = testbench_binary, # This is the default runtime unless # "--test_arg=--runtime=OTHER_RUNTIME" is used to override the value. - flags = ["--dut_platform", "netstack", "--runtime=runsc-d"], + flags = ["--dut_platform", "netstack", "--runtime=runsc-d"] + expect_failure_flag, tags = PACKETIMPACT_TAGS + ["packetimpact"], **kwargs ) @@ -112,7 +125,13 @@ def packetimpact_go_test(name, size = "small", pure = True, linux = True, netsta tags = PACKETIMPACT_TAGS, **kwargs ) - if linux: - packetimpact_linux_test(name = name, testbench_binary = testbench_binary) - if netstack: - packetimpact_netstack_test(name = name, testbench_binary = testbench_binary) + packetimpact_linux_test( + name = name, + expect_failure = not linux, + testbench_binary = testbench_binary, + ) + packetimpact_netstack_test( + name = name, + expect_failure = not netstack, + testbench_binary = testbench_binary, + ) diff --git a/test/packetimpact/tests/test_runner.sh b/test/packetimpact/tests/test_runner.sh index e99fc7d09..e938de782 100755 --- a/test/packetimpact/tests/test_runner.sh +++ b/test/packetimpact/tests/test_runner.sh @@ -29,7 +29,7 @@ function failure() { } trap 'failure ${LINENO} "$BASH_COMMAND"' ERR -declare -r LONGOPTS="dut_platform:,posix_server_binary:,testbench_binary:,runtime:,tshark,extra_test_arg:" +declare -r LONGOPTS="dut_platform:,posix_server_binary:,testbench_binary:,runtime:,tshark,extra_test_arg:,expect_failure" # Don't use declare below so that the error from getopt will end the script. PARSED=$(getopt --options "" --longoptions=$LONGOPTS --name "$0" -- "$@") @@ -68,6 +68,10 @@ while true; do EXTRA_TEST_ARGS+="$2" shift 2 ;; + --expect_failure) + declare -r EXPECT_FAILURE="1" + shift 1 + ;; --) shift break @@ -103,21 +107,24 @@ if [[ ! -f "${TESTBENCH_BINARY-}" ]]; then exit 2 fi +function new_net_prefix() { + # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. + echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" +} + # Variables specific to the control network and interface start with CTRL_. # Variables specific to the test network and interface start with TEST_. # Variables specific to the DUT start with DUT_. # Variables specific to the test bench start with TESTBENCH_. # Use random numbers so that test networks don't collide. -declare -r CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" -declare -r TEST_NET="test_net-${RANDOM}${RANDOM}" +declare CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" +declare CTRL_NET_PREFIX=$(new_net_prefix) +declare TEST_NET="test_net-${RANDOM}${RANDOM}" +declare TEST_NET_PREFIX=$(new_net_prefix) # On both DUT and test bench, testing packets are on the eth2 interface. declare -r TEST_DEVICE="eth2" # Number of bits in the *_NET_PREFIX variables. declare -r NET_MASK="24" -function new_net_prefix() { - # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. - echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" -} # Last bits of the DUT's IP address. declare -r DUT_NET_SUFFIX=".10" # Control port. @@ -126,6 +133,7 @@ declare -r CTRL_PORT="40000" declare -r TESTBENCH_NET_SUFFIX=".20" declare -r TIMEOUT="60" declare -r IMAGE_TAG="gcr.io/gvisor-presubmit/packetimpact" + # Make sure that docker is installed. docker --version @@ -165,19 +173,19 @@ function finish { trap finish EXIT # Subnet for control packets between test bench and DUT. -declare CTRL_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do sleep 0.1 - declare CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" done # Subnet for the packets that are part of the test. -declare TEST_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do sleep 0.1 - declare TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET="test_net-${RANDOM}${RANDOM}" done docker pull "${IMAGE_TAG}" @@ -254,7 +262,10 @@ sleep 3 # Start a packetimpact test on the test bench. The packetimpact test sends and # receives packets and also sends POSIX socket commands to the posix_server to # be executed on the DUT. -docker exec -t "${TESTBENCH}" \ +docker exec \ + -e XML_OUTPUT_FILE="/test.xml" \ + -e TEST_TARGET \ + -t "${TESTBENCH}" \ /bin/bash -c "${DOCKER_TESTBENCH_BINARY} \ ${EXTRA_TEST_ARGS[@]-} \ --posix_server_ip=${CTRL_NET_PREFIX}${DUT_NET_SUFFIX} \ @@ -263,6 +274,15 @@ docker exec -t "${TESTBENCH}" \ --local_ipv4=${TEST_NET_PREFIX}${TESTBENCH_NET_SUFFIX} \ --remote_mac=${REMOTE_MAC} \ --local_mac=${LOCAL_MAC} \ - --device=${TEST_DEVICE}" - + --device=${TEST_DEVICE}" && true +declare -r TEST_RESULT="${?}" +if [[ -z "${EXPECT_FAILURE-}" && "${TEST_RESULT}" != 0 ]]; then + echo 'FAIL: This test was expected to pass.' + exit ${TEST_RESULT} +fi +if [[ ! -z "${EXPECT_FAILURE-}" && "${TEST_RESULT}" == 0 ]]; then + echo 'FAIL: This test was expected to fail but passed. Enable the test and' \ + 'mark the corresponding bug as fixed.' + exit 1 +fi echo PASS: No errors. diff --git a/test/root/BUILD b/test/root/BUILD index 05166673a..17e51e66e 100644 --- a/test/root/BUILD +++ b/test/root/BUILD @@ -33,14 +33,12 @@ go_test( ], visibility = ["//:sandbox"], deps = [ - "//runsc/boot", + "//pkg/test/criutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", "//runsc/cgroup", "//runsc/container", - "//runsc/criutil", - "//runsc/dockerutil", "//runsc/specutils", - "//runsc/testutil", - "//test/root/testdata", "@com_github_cenkalti_backoff//:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", "@com_github_syndtr_gocapability//capability:go_default_library", diff --git a/test/root/cgroup_test.go b/test/root/cgroup_test.go index 679342def..8876d0d61 100644 --- a/test/root/cgroup_test.go +++ b/test/root/cgroup_test.go @@ -26,9 +26,9 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/cgroup" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" ) func verifyPid(pid int, path string) error { @@ -56,54 +56,70 @@ func verifyPid(pid int, path string) error { return fmt.Errorf("got: %v, want: %d", gots, pid) } -// TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestMemCGroup(t *testing.T) { - allocMemSize := 128 << 20 - if err := dockerutil.Pull("python"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("memusage-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start a new container and allocate the specified about of memory. - args := []string{ - "--memory=256MB", - "python", - "python", - "-c", - fmt.Sprintf("import time; s = 'a' * %d; time.sleep(100)", allocMemSize), - } - if err := d.Run(args...); err != nil { - t.Fatal("docker create failed:", err) + allocMemSize := 128 << 20 + allocMemLimit := 2 * allocMemSize + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Memory: allocMemLimit / 1024, // Must be in Kb. + }, "python", "-c", fmt.Sprintf("import time; s = 'a' * %d; time.sleep(100)", allocMemSize)); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Extract the ID to lookup the cgroup. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) } t.Logf("cgroup ID: %s", gid) - path := filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.usage_in_bytes") - memUsage := 0 - // Wait when the container will allocate memory. + memUsage := 0 start := time.Now() - for time.Now().Sub(start) < 30*time.Second { + for time.Since(start) < 30*time.Second { + // Sleep for a brief period of time after spawning the + // container (so that Docker can create the cgroup etc. + // or after looping below (so the application can start). + time.Sleep(100 * time.Millisecond) + + // Read the cgroup memory limit. + path := filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.limit_in_bytes") outRaw, err := ioutil.ReadFile(path) if err != nil { - t.Fatalf("failed to read %q: %v", path, err) + // It's possible that the container does not exist yet. + continue } out := strings.TrimSpace(string(outRaw)) + memLimit, err := strconv.Atoi(out) + if err != nil { + t.Fatalf("Atoi(%v): %v", out, err) + } + if memLimit != allocMemLimit { + // The group may not have had the correct limit set yet. + continue + } + + // Read the cgroup memory usage. + path = filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.max_usage_in_bytes") + outRaw, err = ioutil.ReadFile(path) + if err != nil { + t.Fatalf("error reading usage: %v", err) + } + out = strings.TrimSpace(string(outRaw)) memUsage, err = strconv.Atoi(out) if err != nil { t.Fatalf("Atoi(%v): %v", out, err) } + t.Logf("read usage: %v, wanted: %v", memUsage, allocMemSize) - if memUsage > allocMemSize { + // Are we done? + if memUsage >= allocMemSize { return } - - time.Sleep(100 * time.Millisecond) } t.Fatalf("%vMB is less than %vMB", memUsage>>20, allocMemSize>>20) @@ -111,10 +127,8 @@ func TestMemCGroup(t *testing.T) { // TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestCgroup(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // This is not a comprehensive list of attributes. // @@ -179,10 +193,11 @@ func TestCgroup(t *testing.T) { want: "5", }, { - arg: "--blkio-weight=750", - ctrl: "blkio", - file: "blkio.weight", - want: "750", + arg: "--blkio-weight=750", + ctrl: "blkio", + file: "blkio.weight", + want: "750", + skipIfNotFound: true, // blkio groups may not be available. }, } @@ -191,12 +206,15 @@ func TestCgroup(t *testing.T) { args = append(args, attr.arg) } - args = append(args, "alpine", "sleep", "10000") - if err := d.Run(args...); err != nil { - t.Fatal("docker create failed:", err) + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: args, // Cgroup arguments. + }, "sleep", "10000"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Lookup the relevant cgroup ID. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) @@ -245,17 +263,21 @@ func TestCgroup(t *testing.T) { } } +// TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestCgroupParent(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - parent := testutil.RandomName("runsc") - if err := d.Run("--cgroup-parent", parent, "alpine", "sleep", "10000"); err != nil { - t.Fatal("docker create failed:", err) + // Construct a known cgroup name. + parent := testutil.RandomID("runsc-") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: []string{fmt.Sprintf("--cgroup-parent=%s", parent)}, + }, "sleep", "10000"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + + // Extract the ID to look up the cgroup. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) diff --git a/test/root/chroot_test.go b/test/root/chroot_test.go index be0f63d18..a306132a4 100644 --- a/test/root/chroot_test.go +++ b/test/root/chroot_test.go @@ -24,17 +24,20 @@ import ( "strings" "testing" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" ) // TestChroot verifies that the sandbox is chroot'd and that mounts are cleaned // up after the sandbox is destroyed. func TestChroot(t *testing.T) { - d := dockerutil.MakeDocker("chroot-test") - if err := d.Run("alpine", "sleep", "10000"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() pid, err := d.SandboxPid() if err != nil { @@ -76,11 +79,14 @@ func TestChroot(t *testing.T) { } func TestChrootGofer(t *testing.T) { - d := dockerutil.MakeDocker("chroot-test") - if err := d.Run("alpine", "sleep", "10000"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // It's tricky to find gofers. Get sandbox PID first, then find parent. From // parent get all immediate children, remove the sandbox, and everything else diff --git a/test/root/crictl_test.go b/test/root/crictl_test.go index 3f90c4c6a..85007dcce 100644 --- a/test/root/crictl_test.go +++ b/test/root/crictl_test.go @@ -16,6 +16,7 @@ package root import ( "bytes" + "encoding/json" "fmt" "io" "io/ioutil" @@ -29,16 +30,58 @@ import ( "testing" "time" - "gvisor.dev/gvisor/runsc/criutil" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/criutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" - "gvisor.dev/gvisor/test/root/testdata" ) // Tests for crictl have to be run as root (rather than in a user namespace) // because crictl creates named network namespaces in /var/run/netns/. +// SimpleSpec returns a JSON config for a simple container that runs the +// specified command in the specified image. +func SimpleSpec(name, image string, cmd []string, extra map[string]interface{}) string { + s := map[string]interface{}{ + "metadata": map[string]string{ + "name": name, + }, + "image": map[string]string{ + "image": testutil.ImageByName(image), + }, + "log_path": fmt.Sprintf("%s.log", name), + } + if len(cmd) > 0 { // Omit if empty. + s["command"] = cmd + } + for k, v := range extra { + s[k] = v // Extra settings. + } + v, err := json.Marshal(s) + if err != nil { + // This shouldn't happen. + panic(err) + } + return string(v) +} + +// Sandbox is a default JSON config for a sandbox. +var Sandbox = `{ + "metadata": { + "name": "default-sandbox", + "namespace": "default", + "attempt": 1, + "uid": "hdishd83djaidwnduwk28bcsb" + }, + "linux": { + }, + "log_directory": "/tmp" +} +` + +// Httpd is a JSON config for an httpd container. +var Httpd = SimpleSpec("httpd", "basic/httpd", nil, nil) + // TestCrictlSanity refers to b/112433158. func TestCrictlSanity(t *testing.T) { // Setup containerd and crictl. @@ -47,9 +90,9 @@ func TestCrictlSanity(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("httpd", testdata.Sandbox, testdata.Httpd) + podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox, Httpd) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } // Look for the httpd page. @@ -59,10 +102,38 @@ func TestCrictlSanity(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } +// HttpdMountPaths is a JSON config for an httpd container with additional +// mounts. +var HttpdMountPaths = SimpleSpec("httpd", "basic/httpd", nil, map[string]interface{}{ + "mounts": []map[string]interface{}{ + map[string]interface{}{ + "container_path": "/var/run/secrets/kubernetes.io/serviceaccount", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/volumes/kubernetes.io~secret/default-token-2rpfx", + "readonly": true, + }, + map[string]interface{}{ + "container_path": "/etc/hosts", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/etc-hosts", + "readonly": false, + }, + map[string]interface{}{ + "container_path": "/dev/termination-log", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/containers/httpd/d1709580", + "readonly": false, + }, + map[string]interface{}{ + "container_path": "/usr/local/apache2/htdocs/test", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064", + "readonly": true, + }, + }, + "linux": map[string]interface{}{}, +}) + // TestMountPaths refers to b/117635704. func TestMountPaths(t *testing.T) { // Setup containerd and crictl. @@ -71,9 +142,9 @@ func TestMountPaths(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("httpd", testdata.Sandbox, testdata.HttpdMountPaths) + podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox, HttpdMountPaths) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } // Look for the directory available at /test. @@ -83,7 +154,7 @@ func TestMountPaths(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } @@ -95,14 +166,16 @@ func TestMountOverSymlinks(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("k8s.gcr.io/busybox", testdata.Sandbox, testdata.MountOverSymlink) + + spec := SimpleSpec("busybox", "basic/resolv", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer("basic/resolv", Sandbox, spec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } out, err := crictl.Exec(contID, "readlink", "/etc/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("readlink failed: %v, out: %s", err, out) } if want := "/tmp/resolv.conf"; !strings.Contains(string(out), want) { t.Fatalf("/etc/resolv.conf is not pointing to %q: %q", want, string(out)) @@ -110,11 +183,11 @@ func TestMountOverSymlinks(t *testing.T) { etc, err := crictl.Exec(contID, "cat", "/etc/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("cat failed: %v, out: %s", err, etc) } tmp, err := crictl.Exec(contID, "cat", "/tmp/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("cat failed: %v, out: %s", err, out) } if tmp != etc { t.Fatalf("file content doesn't match:\n\t/etc/resolv.conf: %s\n\t/tmp/resolv.conf: %s", string(etc), string(tmp)) @@ -122,7 +195,7 @@ func TestMountOverSymlinks(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } @@ -135,16 +208,16 @@ func TestHomeDir(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - contSpec := testdata.SimpleSpec("root", "k8s.gcr.io/busybox", []string{"sleep", "1000"}) - podID, contID, err := crictl.StartPodAndContainer("k8s.gcr.io/busybox", testdata.Sandbox, contSpec) + contSpec := SimpleSpec("root", "basic/busybox", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer("basic/busybox", Sandbox, contSpec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } t.Run("root container", func(t *testing.T) { out, err := crictl.Exec(contID, "sh", "-c", "echo $HOME") if err != nil { - t.Fatal(err) + t.Fatalf("exec failed: %v, out: %s", err, out) } if got, want := strings.TrimSpace(string(out)), "/root"; got != want { t.Fatalf("Home directory invalid. Got %q, Want : %q", got, want) @@ -153,32 +226,47 @@ func TestHomeDir(t *testing.T) { t.Run("sub-container", func(t *testing.T) { // Create a sub container in the same pod. - subContSpec := testdata.SimpleSpec("subcontainer", "k8s.gcr.io/busybox", []string{"sleep", "1000"}) - subContID, err := crictl.StartContainer(podID, "k8s.gcr.io/busybox", testdata.Sandbox, subContSpec) + subContSpec := SimpleSpec("subcontainer", "basic/busybox", []string{"sleep", "1000"}, nil) + subContID, err := crictl.StartContainer(podID, "basic/busybox", Sandbox, subContSpec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } out, err := crictl.Exec(subContID, "sh", "-c", "echo $HOME") if err != nil { - t.Fatal(err) + t.Fatalf("exec failed: %v, out: %s", err, out) } if got, want := strings.TrimSpace(string(out)), "/root"; got != want { t.Fatalf("Home directory invalid. Got %q, Want: %q", got, want) } if err := crictl.StopContainer(subContID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } }) // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } +// containerdConfigTemplate is a .toml config for containerd. It contains a +// formatting verb so the runtime field can be set via fmt.Sprintf. +const containerdConfigTemplate = ` +disabled_plugins = ["restart"] +[plugins.linux] + runtime = "%s" + runtime_root = "/tmp/test-containerd/runsc" + shim = "/usr/local/bin/gvisor-containerd-shim" + shim_debug = true + +[plugins.cri.containerd.runtimes.runsc] + runtime_type = "io.containerd.runtime.v1.linux" + runtime_engine = "%s" +` + // setup sets up before a test. Specifically it: // * Creates directories and a socket for containerd to utilize. // * Runs containerd and waits for it to reach a "ready" state for testing. @@ -213,50 +301,52 @@ func setup(t *testing.T) (*criutil.Crictl, func(), error) { if err != nil { t.Fatalf("error discovering runtime path: %v", err) } - config, err := testutil.WriteTmpFile("containerd-config", testdata.ContainerdConfig(runtime)) + config, configCleanup, err := testutil.WriteTmpFile("containerd-config", fmt.Sprintf(containerdConfigTemplate, runtime, runtime)) if err != nil { t.Fatalf("failed to write containerd config") } - cleanups = append(cleanups, func() { os.RemoveAll(config) }) + cleanups = append(cleanups, configCleanup) // Start containerd. - containerd := exec.Command(getContainerd(), + cmd := exec.Command(getContainerd(), "--config", config, "--log-level", "debug", "--root", containerdRoot, "--state", containerdState, "--address", sockAddr) + startupR, startupW := io.Pipe() + defer startupR.Close() + defer startupW.Close() + stderr := &bytes.Buffer{} + stdout := &bytes.Buffer{} + cmd.Stderr = io.MultiWriter(startupW, stderr) + cmd.Stdout = io.MultiWriter(startupW, stdout) cleanups = append(cleanups, func() { - if err := testutil.KillCommand(containerd); err != nil { - log.Printf("error killing containerd: %v", err) - } + t.Logf("containerd stdout: %s", stdout.String()) + t.Logf("containerd stderr: %s", stderr.String()) }) - containerdStderr, err := containerd.StderrPipe() - if err != nil { - t.Fatalf("failed to get containerd stderr: %v", err) - } - containerdStdout, err := containerd.StdoutPipe() - if err != nil { - t.Fatalf("failed to get containerd stdout: %v", err) - } - if err := containerd.Start(); err != nil { + + // Start the process. + if err := cmd.Start(); err != nil { t.Fatalf("failed running containerd: %v", err) } - // Wait for containerd to boot. Then put all containerd output into a - // buffer to be logged at the end of the test. - testutil.WaitUntilRead(containerdStderr, "Start streaming server", nil, 10*time.Second) - stdoutBuf := &bytes.Buffer{} - stderrBuf := &bytes.Buffer{} - go func() { io.Copy(stdoutBuf, containerdStdout) }() - go func() { io.Copy(stderrBuf, containerdStderr) }() + // Wait for containerd to boot. + if err := testutil.WaitUntilRead(startupR, "Start streaming server", nil, 10*time.Second); err != nil { + t.Fatalf("failed to start containerd: %v", err) + } + + // Kill must be the last cleanup (as it will be executed first). + cc := criutil.NewCrictl(t, sockAddr) cleanups = append(cleanups, func() { - t.Logf("containerd stdout: %s", string(stdoutBuf.Bytes())) - t.Logf("containerd stderr: %s", string(stderrBuf.Bytes())) + cc.CleanUp() // Remove tmp files, etc. + if err := testutil.KillCommand(cmd); err != nil { + log.Printf("error killing containerd: %v", err) + } }) cleanup.Release() - return criutil.NewCrictl(20*time.Second, sockAddr), cleanupFunc, nil + return cc, cleanupFunc, nil } // httpGet GETs the contents of a file served from a pod on port 80. diff --git a/test/root/main_test.go b/test/root/main_test.go index d74dec85f..9fb17e0dd 100644 --- a/test/root/main_test.go +++ b/test/root/main_test.go @@ -21,7 +21,7 @@ import ( "testing" "github.com/syndtr/gocapability/capability" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/runsc/specutils" ) diff --git a/test/root/oom_score_adj_test.go b/test/root/oom_score_adj_test.go index 22488b05d..9a3cecd97 100644 --- a/test/root/oom_score_adj_test.go +++ b/test/root/oom_score_adj_test.go @@ -20,10 +20,9 @@ import ( "testing" specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/runsc/boot" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/container" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) var ( @@ -40,15 +39,6 @@ var ( // TestOOMScoreAdjSingle tests that oom_score_adj is set properly in a // single container sandbox. func TestOOMScoreAdjSingle(t *testing.T) { - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - - conf := testutil.TestConfig(t) - conf.RootDir = rootDir - ppid, err := specutils.GetParentPid(os.Getpid()) if err != nil { t.Fatalf("getting parent pid: %v", err) @@ -89,11 +79,11 @@ func TestOOMScoreAdjSingle(t *testing.T) { for _, testCase := range testCases { t.Run(testCase.Name, func(t *testing.T) { - id := testutil.UniqueContainerID() + id := testutil.RandomContainerID() s := testutil.NewSpecWithArgs("sleep", "1000") s.Process.OOMScoreAdj = testCase.OOMScoreAdj - containers, cleanup, err := startContainers(conf, []*specs.Spec{s}, []string{id}) + containers, cleanup, err := startContainers(t, []*specs.Spec{s}, []string{id}) if err != nil { t.Fatalf("error starting containers: %v", err) } @@ -131,15 +121,6 @@ func TestOOMScoreAdjSingle(t *testing.T) { // TestOOMScoreAdjMulti tests that oom_score_adj is set properly in a // multi-container sandbox. func TestOOMScoreAdjMulti(t *testing.T) { - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - - conf := testutil.TestConfig(t) - conf.RootDir = rootDir - ppid, err := specutils.GetParentPid(os.Getpid()) if err != nil { t.Fatalf("getting parent pid: %v", err) @@ -257,7 +238,7 @@ func TestOOMScoreAdjMulti(t *testing.T) { } } - containers, cleanup, err := startContainers(conf, specs, ids) + containers, cleanup, err := startContainers(t, specs, ids) if err != nil { t.Fatalf("error starting containers: %v", err) } @@ -321,7 +302,7 @@ func TestOOMScoreAdjMulti(t *testing.T) { func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { var specs []*specs.Spec var ids []string - rootID := testutil.UniqueContainerID() + rootID := testutil.RandomContainerID() for i, cmd := range cmds { spec := testutil.NewSpecWithArgs(cmd...) @@ -335,35 +316,48 @@ func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { specutils.ContainerdContainerTypeAnnotation: specutils.ContainerdContainerTypeContainer, specutils.ContainerdSandboxIDAnnotation: rootID, } - ids = append(ids, testutil.UniqueContainerID()) + ids = append(ids, testutil.RandomContainerID()) } specs = append(specs, spec) } return specs, ids } -func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*container.Container, func(), error) { - if len(conf.RootDir) == 0 { - panic("conf.RootDir not set. Call testutil.SetupRootDir() to set.") - } - - var containers []*container.Container - var bundles []string - cleanup := func() { +func startContainers(t *testing.T, specs []*specs.Spec, ids []string) ([]*container.Container, func(), error) { + var ( + containers []*container.Container + cleanups []func() + ) + cleanups = append(cleanups, func() { for _, c := range containers { c.Destroy() } - for _, b := range bundles { - os.RemoveAll(b) + }) + cleanupAll := func() { + for _, c := range cleanups { + c() } } + localClean := specutils.MakeCleanup(cleanupAll) + defer localClean.Clean() + + // All containers must share the same root. + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + cleanups = append(cleanups, cleanup) + + // Point this to from the configuration. + conf := testutil.TestConfig(t) + conf.RootDir = rootDir + for i, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { - cleanup() - return nil, nil, fmt.Errorf("error setting up container: %v", err) + return nil, nil, fmt.Errorf("error setting up bundle: %v", err) } - bundles = append(bundles, bundleDir) + cleanups = append(cleanups, cleanup) args := container.Args{ ID: ids[i], @@ -372,15 +366,15 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*c } cont, err := container.New(conf, args) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error creating container: %v", err) } containers = append(containers, cont) if err := cont.Start(conf); err != nil { - cleanup() return nil, nil, fmt.Errorf("error starting container: %v", err) } } - return containers, cleanup, nil + + localClean.Release() + return containers, cleanupAll, nil } diff --git a/test/root/runsc_test.go b/test/root/runsc_test.go index 90373e2db..25204bebb 100644 --- a/test/root/runsc_test.go +++ b/test/root/runsc_test.go @@ -28,8 +28,8 @@ import ( "github.com/cenkalti/backoff" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) // TestDoKill checks that when "runsc do..." is killed, the sandbox process is diff --git a/test/root/testdata/BUILD b/test/root/testdata/BUILD deleted file mode 100644 index 6859541ad..000000000 --- a/test/root/testdata/BUILD +++ /dev/null @@ -1,18 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "testdata", - srcs = [ - "busybox.go", - "containerd_config.go", - "httpd.go", - "httpd_mount_paths.go", - "sandbox.go", - "simple.go", - ], - visibility = [ - "//:sandbox", - ], -) diff --git a/test/root/testdata/containerd_config.go b/test/root/testdata/containerd_config.go deleted file mode 100644 index e12f1ec88..000000000 --- a/test/root/testdata/containerd_config.go +++ /dev/null @@ -1,39 +0,0 @@ -// 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. - -// Package testdata contains data required for root tests. -package testdata - -import "fmt" - -// containerdConfigTemplate is a .toml config for containerd. It contains a -// formatting verb so the runtime field can be set via fmt.Sprintf. -const containerdConfigTemplate = ` -disabled_plugins = ["restart"] -[plugins.linux] - runtime = "%s" - runtime_root = "/tmp/test-containerd/runsc" - shim = "/usr/local/bin/gvisor-containerd-shim" - shim_debug = true - -[plugins.cri.containerd.runtimes.runsc] - runtime_type = "io.containerd.runtime.v1.linux" - runtime_engine = "%s" -` - -// ContainerdConfig returns a containerd config file with the specified -// runtime. -func ContainerdConfig(runtime string) string { - return fmt.Sprintf(containerdConfigTemplate, runtime, runtime) -} diff --git a/test/root/testdata/httpd_mount_paths.go b/test/root/testdata/httpd_mount_paths.go deleted file mode 100644 index ac3f4446a..000000000 --- a/test/root/testdata/httpd_mount_paths.go +++ /dev/null @@ -1,53 +0,0 @@ -// 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. - -package testdata - -// HttpdMountPaths is a JSON config for an httpd container with additional -// mounts. -const HttpdMountPaths = ` -{ - "metadata": { - "name": "httpd" - }, - "image":{ - "image": "httpd" - }, - "mounts": [ - { - "container_path": "/var/run/secrets/kubernetes.io/serviceaccount", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/volumes/kubernetes.io~secret/default-token-2rpfx", - "readonly": true - }, - { - "container_path": "/etc/hosts", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/etc-hosts", - "readonly": false - }, - { - "container_path": "/dev/termination-log", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/containers/httpd/d1709580", - "readonly": false - }, - { - "container_path": "/usr/local/apache2/htdocs/test", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064", - "readonly": true - } - ], - "linux": { - }, - "log_path": "httpd.log" -} -` diff --git a/test/root/testdata/simple.go b/test/root/testdata/simple.go deleted file mode 100644 index 1cca53f0c..000000000 --- a/test/root/testdata/simple.go +++ /dev/null @@ -1,41 +0,0 @@ -// 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. - -package testdata - -import ( - "encoding/json" - "fmt" -) - -// SimpleSpec returns a JSON config for a simple container that runs the -// specified command in the specified image. -func SimpleSpec(name, image string, cmd []string) string { - cmds, err := json.Marshal(cmd) - if err != nil { - // This shouldn't happen. - panic(err) - } - return fmt.Sprintf(` -{ - "metadata": { - "name": %q - }, - "image": { - "image": %q - }, - "command": %s - } -`, name, image, cmds) -} diff --git a/test/runner/BUILD b/test/runner/BUILD index 9959ef9b0..6833c9986 100644 --- a/test/runner/BUILD +++ b/test/runner/BUILD @@ -12,8 +12,8 @@ go_binary( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/test/testutil", "//runsc/specutils", - "//runsc/testutil", "//test/runner/gtest", "//test/uds", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/test/runner/runner.go b/test/runner/runner.go index 0d3742f71..14c9cbc47 100644 --- a/test/runner/runner.go +++ b/test/runner/runner.go @@ -32,8 +32,8 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" "gvisor.dev/gvisor/test/runner/gtest" "gvisor.dev/gvisor/test/uds" ) @@ -115,20 +115,20 @@ func runTestCaseNative(testBin string, tc gtest.TestCase, t *testing.T) { // // Returns an error if the sandboxed application exits non-zero. func runRunsc(tc gtest.TestCase, spec *specs.Spec) error { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { return fmt.Errorf("SetupBundleDir failed: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { return fmt.Errorf("SetupRootDir failed: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() name := tc.FullName() - id := testutil.UniqueContainerID() + id := testutil.RandomContainerID() log.Infof("Running test %q in container %q", name, id) specutils.LogSpec(spec) diff --git a/test/runtimes/BUILD b/test/runtimes/BUILD index 2c472bf8d..4cd627222 100644 --- a/test/runtimes/BUILD +++ b/test/runtimes/BUILD @@ -1,20 +1,7 @@ -# These packages are used to run language runtime tests inside gVisor sandboxes. - -load("//tools:defs.bzl", "go_binary", "go_test") -load("//test/runtimes:build_defs.bzl", "runtime_test") +load("//test/runtimes:defs.bzl", "runtime_test") package(licenses = ["notice"]) -go_binary( - name = "runner", - testonly = 1, - srcs = ["runner.go"], - deps = [ - "//runsc/dockerutil", - "//runsc/testutil", - ], -) - runtime_test( name = "go1.12", blacklist_file = "blacklist_go1.12.csv", @@ -44,10 +31,3 @@ runtime_test( blacklist_file = "blacklist_python3.7.3.csv", lang = "python", ) - -go_test( - name = "blacklist_test", - size = "small", - srcs = ["blacklist_test.go"], - library = ":runner", -) diff --git a/test/runtimes/README.md b/test/runtimes/README.md deleted file mode 100644 index 42d722553..000000000 --- a/test/runtimes/README.md +++ /dev/null @@ -1,56 +0,0 @@ -# Runtimes Tests Dockerfiles - -The Dockerfiles defined under this path are configured to host the execution of -the runtimes language tests. Each Dockerfile can support the language indicated -by its directory. - -The following runtimes are currently supported: - -- Go 1.12 -- Java 11 -- Node.js 12 -- PHP 7.3 -- Python 3.7 - -### Building and pushing the images: - -The canonical source of images is the -[gvisor-presubmit container registry](https://gcr.io/gvisor-presubmit/). You can -build new images with the following command: - -```bash -$ cd images -$ docker build -f Dockerfile_$LANG [-t $NAME] . -``` - -To push them to our container registry, set the tag in the command above to -`gcr.io/gvisor-presubmit/$LANG`, then push them. (Note that you will need -appropriate permissions to the `gvisor-presubmit` GCP project.) - -```bash -gcloud docker -- push gcr.io/gvisor-presubmit/$LANG -``` - -#### Running in Docker locally: - -1) [Install and configure Docker](https://docs.docker.com/install/) - -2) Pull the image you want to run: - -```bash -$ docker pull gcr.io/gvisor-presubmit/$LANG -``` - -3) Run docker with the image. - -```bash -$ docker run [--runtime=runsc] --rm -it $NAME [FLAG] -``` - -Running the command with no flags will cause all the available tests to execute. - -Flags can be added for additional functionality: - -- --list: Print a list of all available tests -- --test <name>: Run a single test from the list of available tests -- --v: Print the language version diff --git a/test/runtimes/build_defs.bzl b/test/runtimes/build_defs.bzl deleted file mode 100644 index 92e275a76..000000000 --- a/test/runtimes/build_defs.bzl +++ /dev/null @@ -1,75 +0,0 @@ -"""Defines a rule for runtime test targets.""" - -load("//tools:defs.bzl", "go_test", "loopback") - -def runtime_test( - name, - lang, - image_repo = "gcr.io/gvisor-presubmit", - image_name = None, - blacklist_file = None, - shard_count = 50, - size = "enormous"): - """Generates sh_test and blacklist test targets for a given runtime. - - Args: - name: The name of the runtime being tested. Typically, the lang + version. - This is used in the names of the generated test targets. - lang: The language being tested. - image_repo: The docker repository containing the proctor image to run. - i.e., the prefix to the fully qualified docker image id. - image_name: The name of the image in the image_repo. - Defaults to the test name. - blacklist_file: A test blacklist to pass to the runtime test's runner. - shard_count: See Bazel common test attributes. - size: See Bazel common test attributes. - """ - if image_name == None: - image_name = name - args = [ - "--lang", - lang, - "--image", - "/".join([image_repo, image_name]), - ] - data = [ - ":runner", - loopback, - ] - if blacklist_file: - args += ["--blacklist_file", "test/runtimes/" + blacklist_file] - data += [blacklist_file] - - # Add a test that the blacklist parses correctly. - blacklist_test(name, blacklist_file) - - sh_test( - name = name + "_test", - srcs = ["runner.sh"], - args = args, - data = data, - size = size, - shard_count = shard_count, - tags = [ - # Requires docker and runsc to be configured before the test runs. - "local", - # Don't include test target in wildcard target patterns. - "manual", - ], - ) - -def blacklist_test(name, blacklist_file): - """Test that a blacklist parses correctly.""" - go_test( - name = name + "_blacklist_test", - library = ":runner", - srcs = ["blacklist_test.go"], - args = ["--blacklist_file", "test/runtimes/" + blacklist_file], - data = [blacklist_file], - ) - -def sh_test(**kwargs): - """Wraps the standard sh_test.""" - native.sh_test( - **kwargs - ) diff --git a/test/runtimes/defs.bzl b/test/runtimes/defs.bzl new file mode 100644 index 000000000..f836dd952 --- /dev/null +++ b/test/runtimes/defs.bzl @@ -0,0 +1,79 @@ +"""Defines a rule for runtime test targets.""" + +load("//tools:defs.bzl", "go_test") + +def _runtime_test_impl(ctx): + # Construct arguments. + args = [ + "--lang", + ctx.attr.lang, + "--image", + ctx.attr.image, + ] + if ctx.attr.blacklist_file: + args += [ + "--blacklist_file", + ctx.files.blacklist_file[0].short_path, + ] + + # Build a runner. + runner = ctx.actions.declare_file("%s-executer" % ctx.label.name) + runner_content = "\n".join([ + "#!/bin/bash", + "%s %s\n" % (ctx.files._runner[0].short_path, " ".join(args)), + ]) + ctx.actions.write(runner, runner_content, is_executable = True) + + # Return the runner. + return [DefaultInfo( + executable = runner, + runfiles = ctx.runfiles( + files = ctx.files._runner + ctx.files.blacklist_file + ctx.files._proctor, + collect_default = True, + collect_data = True, + ), + )] + +_runtime_test = rule( + implementation = _runtime_test_impl, + attrs = { + "image": attr.string( + mandatory = False, + ), + "lang": attr.string( + mandatory = True, + ), + "blacklist_file": attr.label( + mandatory = False, + allow_single_file = True, + ), + "_runner": attr.label( + default = "//test/runtimes/runner:runner", + ), + "_proctor": attr.label( + default = "//test/runtimes/proctor:proctor", + ), + }, + test = True, +) + +def runtime_test(name, **kwargs): + _runtime_test( + name = name, + image = name, # Resolved as images/runtimes/%s. + tags = [ + "local", + "manual", + ], + **kwargs + ) + +def blacklist_test(name, blacklist_file): + """Test that a blacklist parses correctly.""" + go_test( + name = name + "_blacklist_test", + library = ":runner", + srcs = ["blacklist_test.go"], + args = ["--blacklist_file", "test/runtimes/" + blacklist_file], + data = [blacklist_file], + ) diff --git a/test/runtimes/images/proctor/BUILD b/test/runtimes/proctor/BUILD index 85e004c45..50a26d182 100644 --- a/test/runtimes/images/proctor/BUILD +++ b/test/runtimes/proctor/BUILD @@ -12,7 +12,8 @@ go_binary( "proctor.go", "python.go", ], - visibility = ["//test/runtimes/images:__subpackages__"], + pure = True, + visibility = ["//test/runtimes:__pkg__"], ) go_test( @@ -21,6 +22,6 @@ go_test( srcs = ["proctor_test.go"], library = ":proctor", deps = [ - "//runsc/testutil", + "//pkg/test/testutil", ], ) diff --git a/test/runtimes/images/proctor/go.go b/test/runtimes/proctor/go.go index 3e2d5d8db..3e2d5d8db 100644 --- a/test/runtimes/images/proctor/go.go +++ b/test/runtimes/proctor/go.go diff --git a/test/runtimes/images/proctor/java.go b/test/runtimes/proctor/java.go index 8b362029d..8b362029d 100644 --- a/test/runtimes/images/proctor/java.go +++ b/test/runtimes/proctor/java.go diff --git a/test/runtimes/images/proctor/nodejs.go b/test/runtimes/proctor/nodejs.go index bd57db444..bd57db444 100644 --- a/test/runtimes/images/proctor/nodejs.go +++ b/test/runtimes/proctor/nodejs.go diff --git a/test/runtimes/images/proctor/php.go b/test/runtimes/proctor/php.go index 9115040e1..9115040e1 100644 --- a/test/runtimes/images/proctor/php.go +++ b/test/runtimes/proctor/php.go diff --git a/test/runtimes/images/proctor/proctor.go b/test/runtimes/proctor/proctor.go index b54abe434..b54abe434 100644 --- a/test/runtimes/images/proctor/proctor.go +++ b/test/runtimes/proctor/proctor.go diff --git a/test/runtimes/images/proctor/proctor_test.go b/test/runtimes/proctor/proctor_test.go index 6bb61d142..6ef2de085 100644 --- a/test/runtimes/images/proctor/proctor_test.go +++ b/test/runtimes/proctor/proctor_test.go @@ -23,24 +23,24 @@ import ( "strings" "testing" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) func touch(t *testing.T, name string) { t.Helper() f, err := os.Create(name) if err != nil { - t.Fatal(err) + t.Fatalf("error creating file %q: %v", name, err) } if err := f.Close(); err != nil { - t.Fatal(err) + t.Fatalf("error closing file %q: %v", name, err) } } func TestSearchEmptyDir(t *testing.T) { td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") if err != nil { - t.Fatal(err) + t.Fatalf("error creating searchtest: %v", err) } defer os.RemoveAll(td) @@ -60,7 +60,7 @@ func TestSearchEmptyDir(t *testing.T) { func TestSearch(t *testing.T) { td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") if err != nil { - t.Fatal(err) + t.Fatalf("error creating searchtest: %v", err) } defer os.RemoveAll(td) @@ -101,14 +101,14 @@ func TestSearch(t *testing.T) { if strings.HasSuffix(item, "/") { // This item is a directory, create it. if err := os.MkdirAll(filepath.Join(td, item), 0755); err != nil { - t.Fatal(err) + t.Fatalf("error making directory: %v", err) } } else { // This item is a file, create the directory and touch file. // Create directory in which file should be created fullDirPath := filepath.Join(td, filepath.Dir(item)) if err := os.MkdirAll(fullDirPath, 0755); err != nil { - t.Fatal(err) + t.Fatalf("error making directory: %v", err) } // Create file with full path to file. touch(t, filepath.Join(td, item)) diff --git a/test/runtimes/images/proctor/python.go b/test/runtimes/proctor/python.go index b9e0fbe6f..b9e0fbe6f 100644 --- a/test/runtimes/images/proctor/python.go +++ b/test/runtimes/proctor/python.go diff --git a/test/runtimes/runner.sh b/test/runtimes/runner.sh deleted file mode 100755 index a8d9a3460..000000000 --- a/test/runtimes/runner.sh +++ /dev/null @@ -1,35 +0,0 @@ -#!/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. - -set -euf -x -o pipefail - -echo -- "$@" - -# Create outputs dir if it does not exist. -if [[ -n "${TEST_UNDECLARED_OUTPUTS_DIR}" ]]; then - mkdir -p "${TEST_UNDECLARED_OUTPUTS_DIR}" - chmod a+rwx "${TEST_UNDECLARED_OUTPUTS_DIR}" -fi - -# Update the timestamp on the shard status file. Bazel looks for this. -touch "${TEST_SHARD_STATUS_FILE}" - -# Get location of runner binary. -readonly runner=$(find "${TEST_SRCDIR}" -name runner) - -# Pass the arguments of this script directly to the runner. -exec "${runner}" "$@" - diff --git a/test/runtimes/runner/BUILD b/test/runtimes/runner/BUILD new file mode 100644 index 000000000..63924b9c5 --- /dev/null +++ b/test/runtimes/runner/BUILD @@ -0,0 +1,21 @@ +load("//tools:defs.bzl", "go_binary", "go_test") + +package(licenses = ["notice"]) + +go_binary( + name = "runner", + testonly = 1, + srcs = ["main.go"], + visibility = ["//test/runtimes:__pkg__"], + deps = [ + "//pkg/test/dockerutil", + "//pkg/test/testutil", + ], +) + +go_test( + name = "blacklist_test", + size = "small", + srcs = ["blacklist_test.go"], + library = ":runner", +) diff --git a/test/runtimes/blacklist_test.go b/test/runtimes/runner/blacklist_test.go index 0ff69ab18..0ff69ab18 100644 --- a/test/runtimes/blacklist_test.go +++ b/test/runtimes/runner/blacklist_test.go diff --git a/test/runtimes/runner.go b/test/runtimes/runner/main.go index 3c98f4570..57540e00e 100644 --- a/test/runtimes/runner.go +++ b/test/runtimes/runner/main.go @@ -26,8 +26,8 @@ import ( "testing" "time" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) var ( @@ -45,7 +45,6 @@ func main() { fmt.Fprintf(os.Stderr, "lang and image flags must not be empty\n") os.Exit(1) } - os.Exit(runTests()) } @@ -60,8 +59,8 @@ func runTests() int { return 1 } - // Create a single docker container that will be used for all tests. - d := dockerutil.MakeDocker("gvisor-" + *lang) + // Construct the shared docker instance. + d := dockerutil.MakeDocker(testutil.DefaultLogger(*lang)) defer d.CleanUp() // Get a slice of tests to run. This will also start a single Docker @@ -77,21 +76,18 @@ func runTests() int { return m.Run() } -// getTests returns a slice of tests to run, subject to the shard size and -// index. -func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.InternalTest, error) { - // Pull the image. - if err := dockerutil.Pull(*image); err != nil { - return nil, fmt.Errorf("docker pull %q failed: %v", *image, err) - } - - // Run proctor with --pause flag to keep container alive forever. - if err := d.Run(*image, "--pause"); err != nil { +// getTests executes all tests as table tests. +func getTests(d *dockerutil.Docker, blacklist map[string]struct{}) ([]testing.InternalTest, error) { + // Start the container. + d.CopyFiles("/proctor", "test/runtimes/proctor/proctor") + if err := d.Spawn(dockerutil.RunOpts{ + Image: fmt.Sprintf("runtimes/%s", *image), + }, "/proctor/proctor", "--pause"); err != nil { return nil, fmt.Errorf("docker run failed: %v", err) } // Get a list of all tests in the image. - list, err := d.Exec("/proctor", "--runtime", *lang, "--list") + list, err := d.Exec(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--list") if err != nil { return nil, fmt.Errorf("docker exec failed: %v", err) } @@ -126,7 +122,7 @@ func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.Int go func() { fmt.Printf("RUNNING %s...\n", tc) - output, err = d.Exec("/proctor", "--runtime", *lang, "--test", tc) + output, err = d.Exec(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--test", tc) close(done) }() @@ -143,6 +139,7 @@ func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.Int }, }) } + return itests, nil } @@ -153,11 +150,7 @@ func getBlacklist() (map[string]struct{}, error) { if *blacklistFile == "" { return blacklist, nil } - file, err := testutil.FindFile(*blacklistFile) - if err != nil { - return nil, err - } - f, err := os.Open(file) + f, err := os.Open(*blacklistFile) if err != nil { return nil, err } diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index d3000dbc6..9400ffaeb 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -319,6 +319,49 @@ TEST_P(SocketInetLoopbackTest, TCPListenUnbound) { tcpSimpleConnectTest(listener, connector, false); } +TEST_P(SocketInetLoopbackTest, TCPListenShutdownListen) { + const auto& param = GetParam(); + + const TestAddress& listener = param.listener; + const TestAddress& connector = param.connector; + + constexpr int kBacklog = 5; + + // Create the listening socket. + 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(), kBacklog), SyscallSucceeds()); + ASSERT_THAT(shutdown(listen_fd.get(), SHUT_RD), SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), kBacklog), 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)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + + for (int i = 0; i < kBacklog; i++) { + auto client = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + ASSERT_THAT(connect(client.get(), reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + } + for (int i = 0; i < kBacklog; i++) { + ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr), SyscallSucceeds()); + } +} + TEST_P(SocketInetLoopbackTest, TCPListenShutdown) { auto const& param = GetParam(); diff --git a/tools/bazeldefs/defs.bzl b/tools/bazeldefs/defs.bzl index 2207b9b34..3c22aec24 100644 --- a/tools/bazeldefs/defs.bzl +++ b/tools/bazeldefs/defs.bzl @@ -5,18 +5,14 @@ load("@io_bazel_rules_go//go:def.bzl", "GoLibrary", _go_binary = "go_binary", _g load("@io_bazel_rules_go//proto:def.bzl", _go_grpc_library = "go_grpc_library", _go_proto_library = "go_proto_library") load("@rules_cc//cc:defs.bzl", _cc_binary = "cc_binary", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test") load("@rules_pkg//:pkg.bzl", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar") -load("@io_bazel_rules_docker//go:image.bzl", _go_image = "go_image") -load("@io_bazel_rules_docker//container:container.bzl", _container_image = "container_image") load("@pydeps//:requirements.bzl", _py_requirement = "requirement") load("@com_github_grpc_grpc//bazel:cc_grpc_library.bzl", _cc_grpc_library = "cc_grpc_library") -container_image = _container_image cc_library = _cc_library cc_flags_supplier = _cc_flags_supplier cc_proto_library = _cc_proto_library cc_test = _cc_test cc_toolchain = "@bazel_tools//tools/cpp:current_cc_toolchain" -go_image = _go_image go_embed_data = _go_embed_data gtest = "@com_google_googletest//:gtest" grpcpp = "@com_github_grpc_grpc//:grpc++" diff --git a/tools/defs.bzl b/tools/defs.bzl index 6a224d7d5..cdaf281f3 100644 --- a/tools/defs.bzl +++ b/tools/defs.bzl @@ -7,7 +7,7 @@ change for Google-internal and bazel-compatible rules. load("//tools/go_stateify:defs.bzl", "go_stateify") load("//tools/go_marshal:defs.bzl", "go_marshal", "marshal_deps", "marshal_test_deps") -load("//tools/bazeldefs:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_grpc_library = "cc_grpc_library", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _gbenchmark = "gbenchmark", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_grpc_and_proto_libraries = "go_grpc_and_proto_libraries", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _grpcpp = "grpcpp", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") +load("//tools/bazeldefs:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_grpc_library = "cc_grpc_library", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _default_installer = "default_installer", _default_net_util = "default_net_util", _gbenchmark = "gbenchmark", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_grpc_and_proto_libraries = "go_grpc_and_proto_libraries", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _grpcpp = "grpcpp", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") load("//tools/bazeldefs:platforms.bzl", _default_platform = "default_platform", _platforms = "platforms") load("//tools/bazeldefs:tags.bzl", "go_suffixes") load("//tools/nogo:defs.bzl", "nogo_test") @@ -19,12 +19,10 @@ cc_grpc_library = _cc_grpc_library cc_library = _cc_library cc_test = _cc_test cc_toolchain = _cc_toolchain -container_image = _container_image default_installer = _default_installer default_net_util = _default_net_util gbenchmark = _gbenchmark go_embed_data = _go_embed_data -go_image = _go_image go_test = _go_test gtest = _gtest grpcpp = _grpcpp @@ -92,7 +90,7 @@ def go_imports(name, src, out): cmd = ("$(location @org_golang_x_tools//cmd/goimports:goimports) $(SRCS) > $@"), ) -def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = False, marshal_debug = False, nogo = True, **kwargs): +def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = False, marshal_debug = False, nogo = False, **kwargs): """Wraps the standard go_library and does stateification and marshalling. The recommended way is to use this rule with mostly identical configuration as the native diff --git a/tools/nogo/config.go b/tools/nogo/config.go index 0c4b7dd40..6958fca69 100644 --- a/tools/nogo/config.go +++ b/tools/nogo/config.go @@ -103,6 +103,9 @@ var analyzerConfig = map[*analysis.Analyzer]matcher{ "pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go", // Special case. "pkg/sentry/platform/safecopy/safecopy_unsafe.go", // Special case. "pkg/sentry/vfs/mount_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/stub_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/switchto_google_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/sysmsg_thread_unsafe.go", // Special case. ), ), unusedresult.Analyzer: alwaysMatches(), diff --git a/vdso/syscalls.h b/vdso/syscalls.h index b6d15a7d3..0c6a922a0 100644 --- a/vdso/syscalls.h +++ b/vdso/syscalls.h @@ -26,6 +26,9 @@ #include <stddef.h> #include <sys/types.h> +#define __stringify_1(x...) #x +#define __stringify(x...) __stringify_1(x) + namespace vdso { #if __x86_64__ @@ -51,20 +54,13 @@ static inline int sys_getcpu(unsigned* cpu, unsigned* node, return num; } -#elif __aarch64__ - -static inline int sys_rt_sigreturn(void) { - int num = __NR_rt_sigreturn; - - asm volatile( - "mov x8, %0\n" - "svc #0 \n" - : "+r"(num) - : - :); - return num; +static inline void sys_rt_sigreturn(void) { + asm volatile("movl $" __stringify(__NR_rt_sigreturn)", %eax \n" + "syscall \n"); } +#elif __aarch64__ + static inline int sys_clock_gettime(clockid_t _clkid, struct timespec* _ts) { register struct timespec* ts asm("x1") = _ts; register clockid_t clkid asm("x0") = _clkid; @@ -91,6 +87,11 @@ static inline int sys_clock_getres(clockid_t _clkid, struct timespec* _ts) { return ret; } +static inline void sys_rt_sigreturn(void) { + asm volatile("mov x8, #" __stringify(__NR_rt_sigreturn)" \n" + "svc #0 \n"); +} + #else #error "unsupported architecture" #endif diff --git a/vdso/vdso.cc b/vdso/vdso.cc index c2585d592..3b6653b5d 100644 --- a/vdso/vdso.cc +++ b/vdso/vdso.cc @@ -69,6 +69,12 @@ int __common_gettimeofday(struct timeval* tv, struct timezone* tz) { } } // namespace +// __kernel_rt_sigreturn() implements rt_sigreturn() +extern "C" void __kernel_rt_sigreturn(unsigned long unused) { + // No optimizations yet, just make the real system call. + sys_rt_sigreturn(); +} + #if __x86_64__ // __vdso_clock_gettime() implements clock_gettime() @@ -143,12 +149,6 @@ extern "C" int __kernel_clock_getres(clockid_t clock, struct timespec* res) { return ret; } -// __kernel_rt_sigreturn() implements gettimeofday() -extern "C" int __kernel_rt_sigreturn(unsigned long unused) { - // No optimizations yet, just make the real system call. - return sys_rt_sigreturn(); -} - #else #error "unsupported architecture" #endif diff --git a/vdso/vdso_amd64.lds b/vdso/vdso_amd64.lds index e2615ae9e..d114290da 100644 --- a/vdso/vdso_amd64.lds +++ b/vdso/vdso_amd64.lds @@ -95,6 +95,7 @@ VERSION { __vdso_getcpu; time; __vdso_time; + __kernel_rt_sigreturn; local: *; }; |