diff options
Diffstat (limited to 'test')
85 files changed, 4153 insertions, 808 deletions
diff --git a/test/benchmarks/README.md b/test/benchmarks/README.md new file mode 100644 index 000000000..9ff602cf1 --- /dev/null +++ b/test/benchmarks/README.md @@ -0,0 +1,118 @@ +# Benchmark tools + +This package and subpackages are for running macro benchmarks on `runsc`. They +are meant to replace the previous //benchmarks benchmark-tools written in +python. + +Benchmarks are meant to look like regular golang benchmarks using the testing.B +library. + +## Setup + +To run benchmarks you will need: + +* Docker installed (17.09.0 or greater). + +The easiest way to run benchmarks is to use the script at +//scripts/benchmark.sh. + +If not using the script, you will need: + +* `runsc` configured with docker + +Note: benchmarks call the runtime by name. If docker can run it with +`--runtime=` flag, these tools should work. + +## Running benchmarks + +The easiest way to run is with the script at //scripts/benchmarks.sh. The script +will run all benchmarks under //test/benchmarks if a target is not provided. + +```bash +./script/benchmarks.sh //path/to/target +``` + +If you want to run benchmarks manually: + +* Run `make load-all-images` from `//` +* Run with: + +```bash +bazel test --test_arg=--runtime=RUNTIME -c opt --test_output=streamed --test_timeout=600 --test_arg=-test.bench=. --nocache_test_results //path/to/target +``` + +## Writing benchmarks + +Benchmarks consist of docker images as Dockerfiles and golang testing.B +benchmarks. + +### Dockerfiles: + +* Are stored at //images. +* New Dockerfiles go in an appropriately named directory at + `//images/benchmarks/my-cool-dockerfile`. +* Dockerfiles for benchmarks should: + * Use explicitly versioned packages. + * Not use ENV and CMD statements...it is easy to add these in the API. +* Note: A common pattern for getting access to a tmpfs mount is to copy files + there after container start. See: //test/benchmarks/build/bazel_test.go. You + can also make your own with `RunOpts.Mounts`. + +### testing.B packages + +In general, benchmarks should look like this: + +```golang + +var h harness.Harness + +func BenchmarkMyCoolOne(b *testing.B) { + machine, err := h.GetMachine() + // check err + + ctx := context.Background() + container := machine.GetContainer(ctx, b) + defer container.CleanUp(ctx) + + b.ResetTimer() + + //Respect b.N. + for i := 0; i < b.N; i++ { + out, err := container.Run(ctx, dockerutil.RunOpts{ + Image: "benchmarks/my-cool-image", + Env: []string{"MY_VAR=awesome"}, + other options...see dockerutil + }, "sh", "-c", "echo MY_VAR" ...) + //check err + b.StopTimer() + + // Do parsing and reporting outside of the timer. + number := parseMyMetric(out) + b.ReportMetric(number, "my-cool-custom-metric") + + b.StartTimer() + } +} + +func TestMain(m *testing.M) { + h.Init() + os.Exit(m.Run()) +} +``` + +Some notes on the above: + +* The harness is initiated in the TestMain method and made global to test + module. The harness will handle any presetup that needs to happen with + flags, remote virtual machines (eventually), and other services. +* Respect `b.N` in that users of the benchmark may want to "run for an hour" + or something of the sort. +* Use the `b.ReportMetric` method to report custom metrics. +* Set the timer if time is useful for reporting. There isn't a way to turn off + default metrics in testing.B (B/op, allocs/op, ns/op). +* Take a look at dockerutil at //pkg/test/dockerutil to see all methods + available from containers. The API is based on the "official" + [docker API for golang](https://pkg.go.dev/mod/github.com/docker/docker). +* `harness.GetMachine` marks how many machines this tests needs. If you have a + client and server and to mark them as multiple machines, call it + `GetMachine` twice. diff --git a/test/benchmarks/fs/BUILD b/test/benchmarks/fs/BUILD new file mode 100644 index 000000000..2874cdbb3 --- /dev/null +++ b/test/benchmarks/fs/BUILD @@ -0,0 +1,23 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "fs", + testonly = 1, + srcs = ["fs.go"], + deps = ["//test/benchmarks/harness"], +) + +go_test( + name = "fs_test", + size = "large", + srcs = ["bazel_test.go"], + library = ":fs", + tags = [ + # Requires docker and runsc to be configured before test runs. + "local", + "manual", + ], + deps = ["//pkg/test/dockerutil"], +) diff --git a/test/benchmarks/fs/bazel_test.go b/test/benchmarks/fs/bazel_test.go new file mode 100644 index 000000000..fdcac1a7a --- /dev/null +++ b/test/benchmarks/fs/bazel_test.go @@ -0,0 +1,95 @@ +// 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 fs + +import ( + "context" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/test/dockerutil" +) + +// Note: CleanCache versions of this test require running with root permissions. +func BenchmarkABSL(b *testing.B) { + // Get a machine from the Harness on which to run. + machine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get machine: %v", err) + } + defer machine.CleanUp() + + // Dimensions here are clean/dirty cache (do or don't drop caches) + // and if the mount on which we are compiling is a tmpfs/bind mount. + benchmarks := []struct { + name string + clearCache bool // clearCache drops caches before running. + tmpfs bool // tmpfs will run compilation on a tmpfs. + }{ + {name: "CleanCache", clearCache: true, tmpfs: false}, + {name: "DirtyCache", clearCache: false, tmpfs: false}, + {name: "CleanCacheTmpfs", clearCache: true, tmpfs: true}, + {name: "DirtyCacheTmpfs", clearCache: false, tmpfs: true}, + } + for _, bm := range benchmarks { + b.Run(bm.name, func(b *testing.B) { + // Grab a container. + ctx := context.Background() + container := machine.GetContainer(ctx, b) + defer container.CleanUp(ctx) + + workdir := "/abseil-cpp" + + // Start a container. + if err := container.Spawn(ctx, dockerutil.RunOpts{ + Image: "benchmarks/absl", + }, "sleep", "1000"); err != nil { + b.Fatalf("run failed with: %v", err) + } + + // If we are running on a tmpfs, copy to /tmp which is a tmpfs. + if bm.tmpfs { + if _, err := container.Exec(ctx, dockerutil.ExecOpts{}, + "cp", "-r", "/abseil-cpp", "/tmp/."); err != nil { + b.Fatal("failed to copy directory: %v", err) + } + workdir = "/tmp" + workdir + } + + // Drop Caches. + if bm.clearCache { + if out, err := machine.RunCommand("/bin/sh -c sync; echo 3 > /proc/sys/vm/drop_caches"); err != nil { + b.Fatalf("failed to drop caches: %v %s", err, out) + } + } + + b.ResetTimer() + for i := 0; i < b.N; i++ { + got, err := container.Exec(ctx, dockerutil.ExecOpts{ + WorkDir: workdir, + }, "bazel", "build", "-c", "opt", "absl/base/...") + if err != nil { + b.Fatalf("build failed with: %v", err) + } + b.StopTimer() + + want := "Build completed successfully" + if !strings.Contains(got, want) { + b.Fatalf("string %s not in: %s", want, got) + } + b.StartTimer() + } + }) + } +} diff --git a/test/benchmarks/fs/fs.go b/test/benchmarks/fs/fs.go new file mode 100644 index 000000000..e5ca28c3b --- /dev/null +++ b/test/benchmarks/fs/fs.go @@ -0,0 +1,31 @@ +// 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 fs holds benchmarks around filesystem performance. +package fs + +import ( + "os" + "testing" + + "gvisor.dev/gvisor/test/benchmarks/harness" +) + +var h harness.Harness + +// TestMain is the main method for package fs. +func TestMain(m *testing.M) { + h.Init() + os.Exit(m.Run()) +} diff --git a/test/benchmarks/harness/BUILD b/test/benchmarks/harness/BUILD new file mode 100644 index 000000000..c2e316709 --- /dev/null +++ b/test/benchmarks/harness/BUILD @@ -0,0 +1,18 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "harness", + testonly = 1, + srcs = [ + "harness.go", + "machine.go", + "util.go", + ], + visibility = ["//:sandbox"], + deps = [ + "//pkg/test/dockerutil", + "//pkg/test/testutil", + ], +) diff --git a/test/benchmarks/harness/harness.go b/test/benchmarks/harness/harness.go new file mode 100644 index 000000000..68bd7b4cf --- /dev/null +++ b/test/benchmarks/harness/harness.go @@ -0,0 +1,38 @@ +// 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 harness holds utility code for running benchmarks on Docker. +package harness + +import ( + "flag" + + "gvisor.dev/gvisor/pkg/test/dockerutil" +) + +// Harness is a handle for managing state in benchmark runs. +type Harness struct { +} + +// Init performs any harness initilialization before runs. +func (h *Harness) Init() error { + flag.Parse() + dockerutil.EnsureSupportedDockerVersion() + return nil +} + +// GetMachine returns this run's implementation of machine. +func (h *Harness) GetMachine() (Machine, error) { + return &localMachine{}, nil +} diff --git a/test/benchmarks/harness/machine.go b/test/benchmarks/harness/machine.go new file mode 100644 index 000000000..93c0db9ce --- /dev/null +++ b/test/benchmarks/harness/machine.go @@ -0,0 +1,71 @@ +// 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 harness + +import ( + "context" + "net" + "os/exec" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +// Machine describes a real machine for use in benchmarks. +type Machine interface { + // GetContainer gets a container from the machine, + GetContainer(ctx context.Context, log testutil.Logger) *dockerutil.Container + + // RunCommand runs cmd on this machine. + RunCommand(cmd string, args ...string) (string, error) + + // Returns IP Address for the machine. + IPAddress() (net.IP, error) + + // CleanUp cleans up this machine. + CleanUp() +} + +// localMachine describes this machine. +type localMachine struct { +} + +// GetContainer implements Machine.GetContainer for localMachine. +func (l *localMachine) GetContainer(ctx context.Context, logger testutil.Logger) *dockerutil.Container { + return dockerutil.MakeContainer(ctx, logger) +} + +// RunCommand implements Machine.RunCommand for localMachine. +func (l *localMachine) RunCommand(cmd string, args ...string) (string, error) { + c := exec.Command(cmd, args...) + out, err := c.CombinedOutput() + return string(out), err +} + +// IPAddress implements Machine.IPAddress. +func (l *localMachine) IPAddress() (net.IP, error) { + conn, err := net.Dial("udp", "8.8.8.8:80") + if err != nil { + return nil, err + } + defer conn.Close() + + addr := conn.LocalAddr().(*net.UDPAddr) + return addr.IP, nil +} + +// CleanUp implements Machine.CleanUp and does nothing for localMachine. +func (*localMachine) CleanUp() { +} diff --git a/test/benchmarks/harness/util.go b/test/benchmarks/harness/util.go new file mode 100644 index 000000000..cc7de6426 --- /dev/null +++ b/test/benchmarks/harness/util.go @@ -0,0 +1,38 @@ +// 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 harness + +import ( + "context" + "fmt" + "net" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +// WaitUntilServing grabs a container from `machine` and waits for a server at +// IP:port. +func WaitUntilServing(ctx context.Context, machine Machine, server net.IP, port int) error { + var logger testutil.DefaultLogger = "netcat" + netcat := machine.GetContainer(ctx, logger) + defer netcat.CleanUp(ctx) + + cmd := fmt.Sprintf("while ! nc -zv %s %d; do true; done", server.String(), port) + _, err := netcat.Run(ctx, dockerutil.RunOpts{ + Image: "packetdrill", + }, "sh", "-c", cmd) + return err +} diff --git a/test/benchmarks/network/BUILD b/test/benchmarks/network/BUILD new file mode 100644 index 000000000..16d267bc8 --- /dev/null +++ b/test/benchmarks/network/BUILD @@ -0,0 +1,29 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "network", + testonly = 1, + srcs = ["network.go"], + deps = ["//test/benchmarks/harness"], +) + +go_test( + name = "network_test", + size = "large", + srcs = [ + "httpd_test.go", + "iperf_test.go", + ], + library = ":network", + tags = [ + # Requires docker and runsc to be configured before test runs. + "manual", + "local", + ], + deps = [ + "//pkg/test/dockerutil", + "//test/benchmarks/harness", + ], +) diff --git a/test/benchmarks/network/httpd_test.go b/test/benchmarks/network/httpd_test.go new file mode 100644 index 000000000..f9afdf15f --- /dev/null +++ b/test/benchmarks/network/httpd_test.go @@ -0,0 +1,276 @@ +// 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 network + +import ( + "context" + "fmt" + "regexp" + "strconv" + "testing" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/test/benchmarks/harness" +) + +// see Dockerfile '//images/benchmarks/httpd'. +var docs = map[string]string{ + "notfound": "notfound", + "1Kb": "latin1k.txt", + "10Kb": "latin10k.txt", + "100Kb": "latin100k.txt", + "1000Kb": "latin1000k.txt", + "1Mb": "latin1024k.txt", + "10Mb": "latin10240k.txt", +} + +// BenchmarkHttpdConcurrency iterates the concurrency argument and tests +// how well the runtime under test handles requests in parallel. +func BenchmarkHttpdConcurrency(b *testing.B) { + // Grab a machine for the client and server. + clientMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get client: %v", err) + } + defer clientMachine.CleanUp() + + serverMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get server: %v", err) + } + defer serverMachine.CleanUp() + + // The test iterates over client concurrency, so set other parameters. + requests := 1000 + concurrency := []int{1, 5, 10, 25} + doc := docs["10Kb"] + + for _, c := range concurrency { + b.Run(fmt.Sprintf("%dConcurrency", c), func(b *testing.B) { + runHttpd(b, clientMachine, serverMachine, doc, requests, c) + }) + } +} + +// BenchmarkHttpdDocSize iterates over different sized payloads, testing how +// well the runtime handles different payload sizes. +func BenchmarkHttpdDocSize(b *testing.B) { + clientMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get machine: %v", err) + } + defer clientMachine.CleanUp() + + serverMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get machine: %v", err) + } + defer serverMachine.CleanUp() + + requests := 1000 + concurrency := 1 + + for name, filename := range docs { + b.Run(name, func(b *testing.B) { + runHttpd(b, clientMachine, serverMachine, filename, requests, concurrency) + }) + } +} + +// runHttpd runs a single test run. +func runHttpd(b *testing.B, clientMachine, serverMachine harness.Machine, doc string, requests, concurrency int) { + b.Helper() + + // Grab a container from the server. + ctx := context.Background() + server := serverMachine.GetContainer(ctx, b) + defer server.CleanUp(ctx) + + // Copy the docs to /tmp and serve from there. + cmd := "mkdir -p /tmp/html; cp -r /local /tmp/html/.; apache2 -X" + port := 80 + + // Start the server. + server.Spawn(ctx, dockerutil.RunOpts{ + Image: "benchmarks/httpd", + Ports: []int{port}, + Env: []string{ + // Standard environmental variables for httpd. + "APACHE_RUN_DIR=/tmp", + "APACHE_RUN_USER=nobody", + "APACHE_RUN_GROUP=nogroup", + "APACHE_LOG_DIR=/tmp", + "APACHE_PID_FILE=/tmp/apache.pid", + }, + }, "sh", "-c", cmd) + + ip, err := serverMachine.IPAddress() + if err != nil { + b.Fatalf("failed to find server ip: %v", err) + } + + servingPort, err := server.FindPort(ctx, port) + if err != nil { + b.Fatalf("failed to find server port %d: %v", port, err) + } + + // Check the server is serving. + harness.WaitUntilServing(ctx, clientMachine, ip, servingPort) + + // Grab a client. + client := clientMachine.GetContainer(ctx, b) + defer client.CleanUp(ctx) + + path := fmt.Sprintf("http://%s:%d/%s", ip, servingPort, doc) + // See apachebench (ab) for flags. + cmd = fmt.Sprintf("ab -n %d -c %d %s", requests, concurrency, path) + + b.ResetTimer() + for i := 0; i < b.N; i++ { + out, err := client.Run(ctx, dockerutil.RunOpts{ + Image: "benchmarks/ab", + }, "sh", "-c", cmd) + if err != nil { + b.Fatalf("run failed with: %v", err) + } + + b.StopTimer() + + // Parse and report custom metrics. + transferRate, err := parseTransferRate(out) + if err != nil { + b.Logf("failed to parse transferrate: %v", err) + } + b.ReportMetric(transferRate*1024, "transfer_rate") // Convert from Kb/s to b/s. + + latency, err := parseLatency(out) + if err != nil { + b.Logf("failed to parse latency: %v", err) + } + b.ReportMetric(latency/1000, "mean_latency") // Convert from ms to s. + + reqPerSecond, err := parseRequestsPerSecond(out) + if err != nil { + b.Logf("failed to parse requests per second: %v", err) + } + b.ReportMetric(reqPerSecond, "requests_per_second") + + b.StartTimer() + } +} + +var transferRateRE = regexp.MustCompile(`Transfer rate:\s+(\d+\.?\d+?)\s+\[Kbytes/sec\]\s+received`) + +// parseTransferRate parses transfer rate from apachebench output. +func parseTransferRate(data string) (float64, error) { + match := transferRateRE.FindStringSubmatch(data) + if len(match) < 2 { + return 0, fmt.Errorf("failed get bandwidth: %s", data) + } + return strconv.ParseFloat(match[1], 64) +} + +var latencyRE = regexp.MustCompile(`Total:\s+\d+\s+(\d+)\s+(\d+\.?\d+?)\s+\d+\s+\d+\s`) + +// parseLatency parses latency from apachebench output. +func parseLatency(data string) (float64, error) { + match := latencyRE.FindStringSubmatch(data) + if len(match) < 2 { + return 0, fmt.Errorf("failed get bandwidth: %s", data) + } + return strconv.ParseFloat(match[1], 64) +} + +var requestsPerSecondRE = regexp.MustCompile(`Requests per second:\s+(\d+\.?\d+?)\s+`) + +// parseRequestsPerSecond parses requests per second from apachebench output. +func parseRequestsPerSecond(data string) (float64, error) { + match := requestsPerSecondRE.FindStringSubmatch(data) + if len(match) < 2 { + return 0, fmt.Errorf("failed get bandwidth: %s", data) + } + return strconv.ParseFloat(match[1], 64) +} + +// Sample output from apachebench. +const sampleData = `This is ApacheBench, Version 2.3 <$Revision: 1826891 $> +Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/ +Licensed to The Apache Software Foundation, http://www.apache.org/ + +Benchmarking 10.10.10.10 (be patient).....done + + +Server Software: Apache/2.4.38 +Server Hostname: 10.10.10.10 +Server Port: 80 + +Document Path: /latin10k.txt +Document Length: 210 bytes + +Concurrency Level: 1 +Time taken for tests: 0.180 seconds +Complete requests: 100 +Failed requests: 0 +Non-2xx responses: 100 +Total transferred: 38800 bytes +HTML transferred: 21000 bytes +Requests per second: 556.44 [#/sec] (mean) +Time per request: 1.797 [ms] (mean) +Time per request: 1.797 [ms] (mean, across all concurrent requests) +Transfer rate: 210.84 [Kbytes/sec] received + +Connection Times (ms) + min mean[+/-sd] median max +Connect: 0 0 0.2 0 2 +Processing: 1 2 1.0 1 8 +Waiting: 1 1 1.0 1 7 +Total: 1 2 1.2 1 10 + +Percentage of the requests served within a certain time (ms) + 50% 1 + 66% 2 + 75% 2 + 80% 2 + 90% 2 + 95% 3 + 98% 7 + 99% 10 + 100% 10 (longest request)` + +// TestParsers checks the parsers work. +func TestParsers(t *testing.T) { + want := 210.84 + got, err := parseTransferRate(sampleData) + if err != nil { + t.Fatalf("failed to parse transfer rate with error: %v", err) + } else if got != want { + t.Fatalf("parseTransferRate got: %f, want: %f", got, want) + } + + want = 2.0 + got, err = parseLatency(sampleData) + if err != nil { + t.Fatalf("failed to parse transfer rate with error: %v", err) + } else if got != want { + t.Fatalf("parseLatency got: %f, want: %f", got, want) + } + + want = 556.44 + got, err = parseRequestsPerSecond(sampleData) + if err != nil { + t.Fatalf("failed to parse transfer rate with error: %v", err) + } else if got != want { + t.Fatalf("parseRequestsPerSecond got: %f, want: %f", got, want) + } +} diff --git a/test/benchmarks/network/iperf_test.go b/test/benchmarks/network/iperf_test.go new file mode 100644 index 000000000..664e0797e --- /dev/null +++ b/test/benchmarks/network/iperf_test.go @@ -0,0 +1,144 @@ +// 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 network + +import ( + "context" + "fmt" + "regexp" + "strconv" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/test/benchmarks/harness" +) + +func BenchmarkIperf(b *testing.B) { + + // Get two machines + clientMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get machine: %v", err) + } + defer clientMachine.CleanUp() + + serverMachine, err := h.GetMachine() + if err != nil { + b.Fatalf("failed to get machine: %v", err) + } + defer serverMachine.CleanUp() + + for _, bm := range []struct { + name string + clientRuntime string + serverRuntime string + }{ + // We are either measuring the server or the client. The other should be + // runc. e.g. Upload sees how fast the runtime under test uploads to a native + // server. + {name: "Upload", clientRuntime: dockerutil.Runtime(), serverRuntime: "runc"}, + {name: "Download", clientRuntime: "runc", serverRuntime: dockerutil.Runtime()}, + } { + b.Run(bm.name, func(b *testing.B) { + + // Get a container from the server and set its runtime. + ctx := context.Background() + server := serverMachine.GetContainer(ctx, b) + defer server.CleanUp(ctx) + server.Runtime = bm.serverRuntime + + // Get a container from the client and set its runtime. + client := clientMachine.GetContainer(ctx, b) + defer client.CleanUp(ctx) + client.Runtime = bm.clientRuntime + + // iperf serves on port 5001 by default. + port := 5001 + + // Start the server. + if err := server.Spawn(ctx, dockerutil.RunOpts{ + Image: "benchmarks/iperf", + Ports: []int{port}, + }, "iperf", "-s"); err != nil { + b.Fatalf("failed to start server with: %v", err) + } + + ip, err := serverMachine.IPAddress() + if err != nil { + b.Fatalf("failed to find server ip: %v", err) + } + + servingPort, err := server.FindPort(ctx, port) + if err != nil { + b.Fatalf("failed to find port %d: %v", port, err) + } + + // Make sure the server is up and serving before we run. + if err := harness.WaitUntilServing(ctx, clientMachine, ip, servingPort); err != nil { + b.Fatalf("failed to wait for server: %v", err) + } + + // iperf report in Kb realtime + cmd := fmt.Sprintf("iperf -f K --realtime -c %s -p %d", ip.String(), servingPort) + + // Run the client. + b.ResetTimer() + + for i := 0; i < b.N; i++ { + out, err := client.Run(ctx, dockerutil.RunOpts{ + Image: "benchmarks/iperf", + }, strings.Split(cmd, " ")...) + if err != nil { + b.Fatalf("failed to run client: %v", err) + } + b.StopTimer() + + // Parse bandwidth and report it. + bW, err := bandwidth(out) + if err != nil { + b.Fatalf("failed to parse bandwitdth from %s: %v", out, err) + } + b.ReportMetric(bW*1024, "bandwidth") // Convert from Kb/s to b/s. + b.StartTimer() + } + }) + } +} + +// bandwidth parses the Bandwidth number from an iperf report. A sample is below. +func bandwidth(data string) (float64, error) { + re := regexp.MustCompile(`\[\s*\d+\][^\n]+\s+(\d+\.?\d*)\s+KBytes/sec`) + match := re.FindStringSubmatch(data) + if len(match) < 1 { + return 0, fmt.Errorf("failed get bandwidth: %s", data) + } + return strconv.ParseFloat(match[1], 64) +} + +func TestParser(t *testing.T) { + sampleData := ` +------------------------------------------------------------ +Client connecting to 10.138.15.215, TCP port 32779 +TCP window size: 45.0 KByte (default) +------------------------------------------------------------ +[ 3] local 10.138.15.216 port 32866 connected with 10.138.15.215 port 32779 +[ ID] Interval Transfer Bandwidth +[ 3] 0.0-10.0 sec 459520 KBytes 45900 KBytes/sec +` + bandwidth, err := bandwidth(sampleData) + if err != nil || bandwidth != 45900 { + t.Fatalf("failed with: %v and %f", err, bandwidth) + } +} diff --git a/test/benchmarks/network/network.go b/test/benchmarks/network/network.go new file mode 100644 index 000000000..ce17ddb94 --- /dev/null +++ b/test/benchmarks/network/network.go @@ -0,0 +1,31 @@ +// 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 network holds benchmarks around raw network performance. +package network + +import ( + "os" + "testing" + + "gvisor.dev/gvisor/test/benchmarks/harness" +) + +var h harness.Harness + +// TestMain is the main method for package network. +func TestMain(m *testing.M) { + h.Init() + os.Exit(m.Run()) +} diff --git a/test/e2e/BUILD b/test/e2e/BUILD index 44cce0e3b..29a84f184 100644 --- a/test/e2e/BUILD +++ b/test/e2e/BUILD @@ -23,6 +23,7 @@ go_test( "//pkg/test/dockerutil", "//pkg/test/testutil", "//runsc/specutils", + "@com_github_docker_docker//api/types/mount:go_default_library", ], ) diff --git a/test/e2e/exec_test.go b/test/e2e/exec_test.go index 6a63b1232..b47df447c 100644 --- a/test/e2e/exec_test.go +++ b/test/e2e/exec_test.go @@ -22,12 +22,10 @@ package integration import ( + "context" "fmt" - "os" - "os/exec" "strconv" "strings" - "syscall" "testing" "time" @@ -39,18 +37,19 @@ import ( // Test that exec uses the exact same capability set as the container. func TestExecCapabilities(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { t.Fatalf("docker run failed: %v", err) } // Check that capability. - matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) + matches, err := d.WaitForOutputSubmatch(ctx, "CapEff:\t([0-9a-f]+)\n", 5*time.Second) if err != nil { t.Fatalf("WaitForOutputSubmatch() timeout: %v", err) } @@ -61,7 +60,7 @@ func TestExecCapabilities(t *testing.T) { t.Log("Root capabilities:", want) // Now check that exec'd process capabilities match the root. - got, err := d.Exec(dockerutil.RunOpts{}, "grep", "CapEff:", "/proc/self/status") + got, err := d.Exec(ctx, dockerutil.ExecOpts{}, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -74,11 +73,12 @@ 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) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container with all capabilities dropped. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", CapDrop: []string{"all"}, }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { @@ -86,7 +86,7 @@ func TestExecPrivileged(t *testing.T) { } // Check that all capabilities where dropped from container. - matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) + matches, err := d.WaitForOutputSubmatch(ctx, "CapEff:\t([0-9a-f]+)\n", 5*time.Second) if err != nil { t.Fatalf("WaitForOutputSubmatch() timeout: %v", err) } @@ -104,7 +104,7 @@ func TestExecPrivileged(t *testing.T) { // Check that 'exec --privileged' adds all capabilities, except for // CAP_NET_RAW. - got, err := d.Exec(dockerutil.RunOpts{ + got, err := d.Exec(ctx, dockerutil.ExecOpts{ Privileged: true, }, "grep", "CapEff:", "/proc/self/status") if err != nil { @@ -118,76 +118,59 @@ func TestExecPrivileged(t *testing.T) { } func TestExecJobControl(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - // Exec 'sh' with an attached pty. - 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 { + p, err := d.ExecProcess(ctx, dockerutil.ExecOpts{UseTTY: true}, "/bin/sh") + if err != nil { t.Fatalf("docker exec failed: %v", err) } + + if _, err = p.Write(time.Second, []byte("sleep 100 | cat\n")); err != nil { + t.Fatalf("error exit: %v", err) + } + time.Sleep(time.Second) + + if _, err = p.Write(time.Second, []byte{0x03}); err != nil { + t.Fatalf("error exit: %v", err) + } + + if _, err = p.Write(time.Second, []byte("exit $(expr $? + 10)\n")); err != nil { + t.Fatalf("error exit: %v", err) + } + + want := 140 + got, err := p.WaitExitStatus(ctx) + if err != nil { + t.Fatalf("wait for exit failed with: %v", err) + } else if got != want { + t.Fatalf("wait for exit returned: %d want: %d", got, want) + } } // Test that failure to exec returns proper error message. func TestExecError(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } // Attempt to exec a binary that doesn't exist. - out, err := d.Exec(dockerutil.RunOpts{}, "no_can_find") + out, err := d.Exec(ctx, dockerutil.ExecOpts{}, "no_can_find") if err == nil { t.Fatalf("docker exec didn't fail") } @@ -198,11 +181,12 @@ func TestExecError(t *testing.T) { // Test that exec inherits environment from run. func TestExecEnv(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container with env FOO=BAR. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", Env: []string{"FOO=BAR"}, }, "sleep", "1000"); err != nil { @@ -210,7 +194,7 @@ func TestExecEnv(t *testing.T) { } // Exec "echo $FOO". - got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $FOO") + got, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/bin/sh", "-c", "echo $FOO") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -222,11 +206,12 @@ 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. - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Exec "echo $HOME". The 'bin' user's home dir is '/bin'. - got, err := d.Run(dockerutil.RunOpts{ + got, err := d.Run(ctx, dockerutil.RunOpts{ Image: "basic/alpine", User: "bin", }, "/bin/sh", "-c", "echo $HOME") @@ -243,17 +228,18 @@ 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. - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } // Exec "echo $HOME", and expect to see "/root". - got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $HOME") + got, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -265,12 +251,12 @@ 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(dockerutil.RunOpts{}, "/bin/sh", "-c", cmd); err != nil { + if _, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/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.Exec(dockerutil.RunOpts{ + got, err = d.Exec(ctx, dockerutil.ExecOpts{ User: strconv.Itoa(newUID), }, "/bin/sh", "-c", "echo $HOME") if err != nil { diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index 60e739c6a..ef42b689a 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -22,24 +22,27 @@ package integration import ( + "context" "flag" "fmt" "io/ioutil" "net" "net/http" "os" - "os/exec" "path/filepath" "strconv" "strings" - "syscall" "testing" "time" + "github.com/docker/docker/api/types/mount" "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/pkg/test/testutil" ) +// defaultWait is the default wait time used for tests. +const defaultWait = time.Minute + // httpRequestSucceeds sends a request to a given url and checks that the status is OK. func httpRequestSucceeds(client http.Client, server string, port int) error { url := fmt.Sprintf("http://%s:%d", server, port) @@ -56,37 +59,38 @@ 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) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Create(dockerutil.RunOpts{ + if err := d.Create(ctx, dockerutil.RunOpts{ Image: "basic/nginx", Ports: []int{80}, }); err != nil { t.Fatalf("docker create failed: %v", err) } - if err := d.Start(); err != nil { + if err := d.Start(ctx); err != nil { t.Fatalf("docker start failed: %v", err) } // Test that container is working. - port, err := d.FindPort(80) + port, err := d.FindPort(ctx, 80) if err != nil { t.Fatalf("docker.FindPort(80) failed: %v", err) } - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } - client := http.Client{Timeout: time.Duration(2 * time.Second)} + client := http.Client{Timeout: defaultWait} if err := httpRequestSucceeds(client, "localhost", port); err != nil { t.Errorf("http request failed: %v", err) } - if err := d.Stop(); err != nil { + if err := d.Stop(ctx); err != nil { t.Fatalf("docker stop failed: %v", err) } - if err := d.Remove(); err != nil { + if err := d.Remove(ctx); err != nil { t.Fatalf("docker rm failed: %v", err) } } @@ -96,11 +100,12 @@ func TestPauseResume(t *testing.T) { t.Skip("Checkpoint is not supported.") } - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/python", Ports: []int{8080}, // See Dockerfile. }); err != nil { @@ -108,27 +113,28 @@ func TestPauseResume(t *testing.T) { } // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) + port, err := d.FindPort(ctx, 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 { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check that container is working. - client := http.Client{Timeout: time.Duration(2 * time.Second)} + client := http.Client{Timeout: defaultWait} if err := httpRequestSucceeds(client, "localhost", port); err != nil { t.Error("http request failed:", err) } - if err := d.Pause(); err != nil { + if err := d.Pause(ctx); err != nil { t.Fatalf("docker pause failed: %v", err) } // Check if container is paused. + client = http.Client{Timeout: 10 * time.Millisecond} // Don't wait a minute. switch _, err := client.Get(fmt.Sprintf("http://localhost:%d", port)); v := err.(type) { case nil: t.Errorf("http req expected to fail but it succeeded") @@ -140,16 +146,17 @@ func TestPauseResume(t *testing.T) { t.Errorf("http req got unexpected error %v", v) } - if err := d.Unpause(); err != nil { + if err := d.Unpause(ctx); err != nil { t.Fatalf("docker unpause failed: %v", err) } // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. + client = http.Client{Timeout: defaultWait} if err := httpRequestSucceeds(client, "localhost", port); err != nil { t.Error("http request failed:", err) } @@ -160,11 +167,12 @@ func TestCheckpointRestore(t *testing.T) { t.Skip("Pause/resume is not supported.") } - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/python", Ports: []int{8080}, // See Dockerfile. }); err != nil { @@ -172,31 +180,31 @@ func TestCheckpointRestore(t *testing.T) { } // Create a snapshot. - if err := d.Checkpoint("test"); err != nil { + if err := d.Checkpoint(ctx, "test"); err != nil { t.Fatalf("docker checkpoint failed: %v", err) } - if _, err := d.Wait(30 * time.Second); err != nil { + if err := d.WaitTimeout(ctx, defaultWait); err != nil { 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 { + if err := testutil.Poll(func() error { return d.Restore(ctx, "test") }, defaultWait); err != nil { t.Fatalf("docker restore failed: %v", err) } // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) + port, err := d.FindPort(ctx, 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 { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. - client := http.Client{Timeout: time.Duration(2 * time.Second)} + client := http.Client{Timeout: defaultWait} if err := httpRequestSucceeds(client, "localhost", port); err != nil { t.Error("http request failed:", err) } @@ -204,26 +212,27 @@ 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(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // 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.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, 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) } // Finds IP address for host. - ip, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") + ip, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") if err != nil { t.Fatalf("docker exec failed: %v", err) } ip = strings.TrimRight(ip, "\n") // Runs client that sends "client" to the server and exits. - reply, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) + reply, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -232,21 +241,22 @@ func TestConnectToSelf(t *testing.T) { if want := "server\n"; reply != want { t.Errorf("Error on server, want: %q, got: %q", want, reply) } - if _, err := d.WaitForOutput("^client\n$", 1*time.Second); err != nil { + if _, err := d.WaitForOutput(ctx, "^client\n$", defaultWait); err != nil { t.Fatalf("docker.WaitForOutput(client) timeout: %v", err) } } func TestMemLimit(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // N.B. Because the size of the memory file may grow in large chunks, // there is a minimum threshold of 1GB for the MemTotal figure. - allocMemory := 1024 * 1024 - out, err := d.Run(dockerutil.RunOpts{ + allocMemory := 1024 * 1024 // In kb. + out, err := d.Run(ctx, dockerutil.RunOpts{ Image: "basic/alpine", - Memory: allocMemory, // In kB. + Memory: allocMemory * 1024, // In bytes. }, "sh", "-c", "cat /proc/meminfo | grep MemTotal: | awk '{print $2}'") if err != nil { t.Fatalf("docker run failed: %v", err) @@ -272,13 +282,14 @@ func TestMemLimit(t *testing.T) { } func TestNumCPU(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Read how many cores are in the container. - out, err := d.Run(dockerutil.RunOpts{ - Image: "basic/alpine", - Extra: []string{"--cpuset-cpus=0"}, + out, err := d.Run(ctx, dockerutil.RunOpts{ + Image: "basic/alpine", + CpusetCpus: "0", }, "sh", "-c", "cat /proc/cpuinfo | grep 'processor.*:' | wc -l") if err != nil { t.Fatalf("docker run failed: %v", err) @@ -296,48 +307,34 @@ func TestNumCPU(t *testing.T) { // TestJobControl tests that job control characters are handled properly. func TestJobControl(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container with an attached PTY. - if _, err := d.Run(dockerutil.RunOpts{ + p, err := d.SpawnProcess(ctx, 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) + }, "sh", "-c", "sleep 100 | cat") + if err != nil { + t.Fatalf("docker run failed: %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) - } + if _, err := p.Write(time.Second, []byte{0x03}); err != nil { + t.Fatalf("error exit: %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) + if err := d.WaitTimeout(ctx, 3*time.Second); err != nil { + t.Fatalf("WaitTimeout failed: %v", err) } - // Wait for the container to exit. - got, err := d.Wait(5 * time.Second) + want := 130 + got, err := p.WaitExitStatus(ctx) if err != nil { - t.Fatalf("error getting exit code: %v", err) - } - // Container should exit with code 10+130=140. - if want := syscall.WaitStatus(140); got != want { - t.Errorf("container exited with code %d want %d", got, want) + t.Fatalf("wait for exit failed with: %v", err) + } else if got != want { + t.Fatalf("got: %d want: %d", got, want) } } @@ -356,15 +353,16 @@ func TestWorkingDirCreation(t *testing.T) { name += "-readonly" } t.Run(name, func(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) opts := dockerutil.RunOpts{ Image: "basic/alpine", WorkDir: tc.workingDir, ReadOnly: readonly, } - got, err := d.Run(opts, "sh", "-c", "echo ${PWD}") + got, err := d.Run(ctx, opts, "sh", "-c", "echo ${PWD}") if err != nil { t.Fatalf("docker run failed: %v", err) } @@ -378,11 +376,12 @@ func TestWorkingDirCreation(t *testing.T) { // TestTmpFile checks that files inside '/tmp' are not overridden. func TestTmpFile(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) - opts := dockerutil.RunOpts{Image: "tmpfile"} - got, err := d.Run(opts, "cat", "/tmp/foo/file.txt") + opts := dockerutil.RunOpts{Image: "basic/tmpfile"} + got, err := d.Run(ctx, opts, "cat", "/tmp/foo/file.txt") if err != nil { t.Fatalf("docker run failed: %v", err) } @@ -393,6 +392,7 @@ func TestTmpFile(t *testing.T) { // TestTmpMount checks that mounts inside '/tmp' are not overridden. func TestTmpMount(t *testing.T) { + ctx := context.Background() dir, err := ioutil.TempDir(testutil.TmpDir(), "tmp-mount") if err != nil { t.Fatalf("TempDir(): %v", err) @@ -401,19 +401,20 @@ func TestTmpMount(t *testing.T) { if err := ioutil.WriteFile(filepath.Join(dir, "file.txt"), []byte("123"), 0666); err != nil { t.Fatalf("WriteFile(): %v", err) } - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) opts := dockerutil.RunOpts{ Image: "basic/alpine", - Mounts: []dockerutil.Mount{ + Mounts: []mount.Mount{ { + Type: mount.TypeBind, Source: dir, Target: "/tmp/foo", }, }, } - got, err := d.Run(opts, "cat", "/tmp/foo/file.txt") + got, err := d.Run(ctx, opts, "cat", "/tmp/foo/file.txt") if err != nil { t.Fatalf("docker run failed: %v", err) } @@ -426,11 +427,12 @@ func TestTmpMount(t *testing.T) { // runsc to hide the incoherence of FDs opened before and after overlayfs // copy-up on the host. func TestHostOverlayfsCopyUp(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) - if _, err := d.Run(dockerutil.RunOpts{ - Image: "hostoverlaytest", + if _, err := d.Run(ctx, dockerutil.RunOpts{ + Image: "basic/hostoverlaytest", WorkDir: "/root", }, "./test"); err != nil { t.Fatalf("docker run failed: %v", err) diff --git a/test/e2e/regression_test.go b/test/e2e/regression_test.go index 327a2174c..70bbe5121 100644 --- a/test/e2e/regression_test.go +++ b/test/e2e/regression_test.go @@ -15,6 +15,7 @@ package integration import ( + "context" "strings" "testing" @@ -27,11 +28,12 @@ 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) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Run the container. - got, err := d.Run(dockerutil.RunOpts{ + got, err := d.Run(ctx, 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 { diff --git a/test/image/image_test.go b/test/image/image_test.go index 3e4321480..ac6186688 100644 --- a/test/image/image_test.go +++ b/test/image/image_test.go @@ -22,6 +22,7 @@ package image import ( + "context" "flag" "fmt" "io/ioutil" @@ -36,12 +37,20 @@ import ( "gvisor.dev/gvisor/pkg/test/testutil" ) +// defaultWait defines how long to wait for progress. +// +// See BUILD: This is at least a "large" test, so allow up to 1 minute for any +// given "wait" step. Note that all tests are run in parallel, which may cause +// individual slow-downs (but a huge speed-up in aggregate). +const defaultWait = time.Minute + func TestHelloWorld(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Run the basic container. - out, err := d.Run(dockerutil.RunOpts{ + out, err := d.Run(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "echo", "Hello world!") if err != nil { @@ -107,8 +116,9 @@ func testHTTPServer(t *testing.T, port int) { } func TestHttpd(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. opts := dockerutil.RunOpts{ @@ -116,18 +126,18 @@ func TestHttpd(t *testing.T) { Ports: []int{80}, } d.CopyFiles(&opts, "/usr/local/apache2/htdocs", "test/image/latin10k.txt") - if err := d.Spawn(opts); err != nil { + if err := d.Spawn(ctx, opts); err != nil { t.Fatalf("docker run failed: %v", err) } // Find where port 80 is mapped to. - port, err := d.FindPort(80) + port, err := d.FindPort(ctx, 80) if err != nil { t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Errorf("WaitForHTTP() timeout: %v", err) } @@ -135,8 +145,9 @@ func TestHttpd(t *testing.T) { } func TestNginx(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the container. opts := dockerutil.RunOpts{ @@ -144,18 +155,18 @@ func TestNginx(t *testing.T) { Ports: []int{80}, } d.CopyFiles(&opts, "/usr/share/nginx/html", "test/image/latin10k.txt") - if err := d.Spawn(opts); err != nil { + if err := d.Spawn(ctx, opts); err != nil { t.Fatalf("docker run failed: %v", err) } // Find where port 80 is mapped to. - port, err := d.FindPort(80) + port, err := d.FindPort(ctx, 80) if err != nil { t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Errorf("WaitForHTTP() timeout: %v", err) } @@ -163,11 +174,12 @@ func TestNginx(t *testing.T) { } func TestMysql(t *testing.T) { - server := dockerutil.MakeDocker(t) - defer server.CleanUp() + ctx := context.Background() + server := dockerutil.MakeContainer(ctx, t) + defer server.CleanUp(ctx) // Start the container. - if err := server.Spawn(dockerutil.RunOpts{ + if err := server.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/mysql", Env: []string{"MYSQL_ROOT_PASSWORD=foobar123"}, }); err != nil { @@ -175,42 +187,38 @@ func TestMysql(t *testing.T) { } // Wait until it's up and running. - if _, err := server.WaitForOutput("port: 3306 MySQL Community Server", 3*time.Minute); err != nil { + if _, err := server.WaitForOutput(ctx, "port: 3306 MySQL Community Server", defaultWait); err != nil { t.Fatalf("WaitForOutput() timeout: %v", err) } // Generate the client and copy in the SQL payload. - client := dockerutil.MakeDocker(t) - defer client.CleanUp() + client := dockerutil.MakeContainer(ctx, t) + defer client.CleanUp(ctx) // Tell mysql client to connect to the server and execute the file in // verbose mode to verify the output. opts := dockerutil.RunOpts{ Image: "basic/mysql", - Links: []dockerutil.Link{ - { - Source: server, - Target: "mysql", - }, - }, + Links: []string{server.MakeLink("mysql")}, } client.CopyFiles(&opts, "/sql", "test/image/mysql.sql") - if _, err := client.Run(opts, "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql"); err != nil { + if _, err := client.Run(ctx, opts, "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql"); err != nil { t.Fatalf("docker run failed: %v", err) } // Ensure file executed to the end and shutdown mysql. - if _, err := server.WaitForOutput("mysqld: Shutdown complete", 30*time.Second); err != nil { + if _, err := server.WaitForOutput(ctx, "mysqld: Shutdown complete", defaultWait); err != nil { t.Fatalf("WaitForOutput() timeout: %v", err) } } func TestTomcat(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start the server. - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/tomcat", Ports: []int{8080}, }); err != nil { @@ -218,13 +226,13 @@ func TestTomcat(t *testing.T) { } // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) + port, err := d.FindPort(ctx, 8080) if err != nil { t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { + if err := testutil.WaitForHTTP(port, defaultWait); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } @@ -240,8 +248,9 @@ func TestTomcat(t *testing.T) { } func TestRuby(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Execute the ruby workload. opts := dockerutil.RunOpts{ @@ -249,18 +258,18 @@ func TestRuby(t *testing.T) { Ports: []int{8080}, } d.CopyFiles(&opts, "/src", "test/image/ruby.rb", "test/image/ruby.sh") - if err := d.Spawn(opts, "/src/ruby.sh"); err != nil { + if err := d.Spawn(ctx, opts, "/src/ruby.sh"); err != nil { t.Fatalf("docker run failed: %v", err) } // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) + port, err := d.FindPort(ctx, 8080) if err != nil { t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running, 'gem install' can take some time. - if err := testutil.WaitForHTTP(port, 1*time.Minute); err != nil { + if err := testutil.WaitForHTTP(port, time.Minute); err != nil { t.Fatalf("WaitForHTTP() timeout: %v", err) } @@ -283,20 +292,21 @@ func TestRuby(t *testing.T) { } func TestStdio(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) wantStdout := "hello stdout" wantStderr := "bonjour stderr" cmd := fmt.Sprintf("echo %q; echo %q 1>&2;", wantStdout, wantStderr) - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "/bin/sh", "-c", cmd); err != nil { t.Fatalf("docker run failed: %v", err) } for _, want := range []string{wantStdout, wantStderr} { - if _, err := d.WaitForOutput(want, 5*time.Second); err != nil { + if _, err := d.WaitForOutput(ctx, want, defaultWait); err != nil { t.Fatalf("docker didn't get output %q : %v", want, err) } } diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 872021358..068f228bd 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -618,7 +618,7 @@ func (FilterInputDestination) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterInputDestination) ContainerAction(ip net.IP) error { - addrs, err := localAddrs() + addrs, err := localAddrs(false) if err != nil { return err } diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 340f9426e..f303030aa 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -15,8 +15,10 @@ package iptables import ( + "context" "fmt" "net" + "reflect" "testing" "gvisor.dev/gvisor/pkg/test/dockerutil" @@ -37,8 +39,9 @@ func singleTest(t *testing.T, test TestCase) { t.Fatalf("no test found with name %q. Has it been registered?", test.Name()) } - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Create and start the container. opts := dockerutil.RunOpts{ @@ -46,12 +49,12 @@ func singleTest(t *testing.T, test TestCase) { CapAdd: []string{"NET_ADMIN"}, } d.CopyFiles(&opts, "/runner", "test/iptables/runner/runner") - if err := d.Spawn(opts, "/runner/runner", "-name", test.Name()); err != nil { + if err := d.Spawn(ctx, opts, "/runner/runner", "-name", test.Name()); err != nil { t.Fatalf("docker run failed: %v", err) } // Get the container IP. - ip, err := d.FindIP() + ip, err := d.FindIP(ctx) if err != nil { t.Fatalf("failed to get container IP: %v", err) } @@ -69,7 +72,7 @@ func singleTest(t *testing.T, test TestCase) { // 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 { + if _, err := d.WaitForOutput(ctx, TerminalStatement, TestTimeout); err != nil { t.Fatalf("test failed: %v", err) } } @@ -260,6 +263,13 @@ func TestNATPreRedirectTCPPort(t *testing.T) { singleTest(t, NATPreRedirectTCPPort{}) } +func TestNATPreRedirectTCPOutgoing(t *testing.T) { + singleTest(t, NATPreRedirectTCPOutgoing{}) +} + +func TestNATOutRedirectTCPIncoming(t *testing.T) { + singleTest(t, NATOutRedirectTCPIncoming{}) +} func TestNATOutRedirectUDPPort(t *testing.T) { singleTest(t, NATOutRedirectUDPPort{}) } @@ -315,3 +325,28 @@ func TestInputSource(t *testing.T) { func TestInputInvertSource(t *testing.T) { singleTest(t, FilterInputInvertSource{}) } + +func TestFilterAddrs(t *testing.T) { + tcs := []struct { + ipv6 bool + addrs []string + want []string + }{ + { + ipv6: false, + addrs: []string{"192.168.0.1", "192.168.0.2/24", "::1", "::2/128"}, + want: []string{"192.168.0.1", "192.168.0.2"}, + }, + { + ipv6: true, + addrs: []string{"192.168.0.1", "192.168.0.2/24", "::1", "::2/128"}, + want: []string{"::1", "::2"}, + }, + } + + for _, tc := range tcs { + if got := filterAddrs(tc.addrs, tc.ipv6); !reflect.DeepEqual(got, tc.want) { + t.Errorf("%v with IPv6 %t: got %v, but wanted %v", tc.addrs, tc.ipv6, got, tc.want) + } + } +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 7146edbb9..d4bc55b24 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -18,6 +18,7 @@ import ( "fmt" "net" "os/exec" + "strings" "time" "gvisor.dev/gvisor/pkg/test/testutil" @@ -157,8 +158,10 @@ func connectTCP(ip net.IP, port int, timeout time.Duration) error { return nil } -// localAddrs returns a list of local network interface addresses. -func localAddrs() ([]string, error) { +// localAddrs returns a list of local network interface addresses. When ipv6 is +// true, only IPv6 addresses are returned. Otherwise only IPv4 addresses are +// returned. +func localAddrs(ipv6 bool) ([]string, error) { addrs, err := net.InterfaceAddrs() if err != nil { return nil, err @@ -167,7 +170,19 @@ func localAddrs() ([]string, error) { for _, addr := range addrs { addrStrs = append(addrStrs, addr.String()) } - return addrStrs, nil + return filterAddrs(addrStrs, ipv6), nil +} + +func filterAddrs(addrs []string, ipv6 bool) []string { + addrStrs := make([]string, 0, len(addrs)) + for _, addr := range addrs { + // Add only IPv4 or only IPv6 addresses. + parts := strings.Split(addr, "/") + if isIPv6 := net.ParseIP(parts[0]).To4() == nil; isIPv6 == ipv6 { + addrStrs = append(addrStrs, parts[0]) + } + } + return addrStrs } // getInterfaceName returns the name of the interface other than loopback. diff --git a/test/iptables/nat.go b/test/iptables/nat.go index 5e54a3963..149dec2bb 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -28,6 +28,8 @@ const ( func init() { RegisterTestCase(NATPreRedirectUDPPort{}) RegisterTestCase(NATPreRedirectTCPPort{}) + RegisterTestCase(NATPreRedirectTCPOutgoing{}) + RegisterTestCase(NATOutRedirectTCPIncoming{}) RegisterTestCase(NATOutRedirectUDPPort{}) RegisterTestCase(NATOutRedirectTCPPort{}) RegisterTestCase(NATDropUDP{}) @@ -91,6 +93,56 @@ func (NATPreRedirectTCPPort) LocalAction(ip net.IP) error { return connectTCP(ip, dropPort, sendloopDuration) } +// NATPreRedirectTCPOutgoing verifies that outgoing TCP connections aren't +// affected by PREROUTING connection tracking. +type NATPreRedirectTCPOutgoing struct{} + +// Name implements TestCase.Name. +func (NATPreRedirectTCPOutgoing) Name() string { + return "NATPreRedirectTCPOutgoing" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreRedirectTCPOutgoing) ContainerAction(ip net.IP) error { + // Redirect all incoming TCP traffic to a closed port. + if err := natTable("-A", "PREROUTING", "-p", "tcp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + + // Establish a connection to the host process. + return connectTCP(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreRedirectTCPOutgoing) LocalAction(ip net.IP) error { + return listenTCP(acceptPort, sendloopDuration) +} + +// NATOutRedirectTCPIncoming verifies that incoming TCP connections aren't +// affected by OUTPUT connection tracking. +type NATOutRedirectTCPIncoming struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectTCPIncoming) Name() string { + return "NATOutRedirectTCPIncoming" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectTCPIncoming) ContainerAction(ip net.IP) error { + // Redirect all outgoing TCP traffic to a closed port. + if err := natTable("-A", "OUTPUT", "-p", "tcp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + + // Establish a connection to the host process. + return listenTCP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectTCPIncoming) LocalAction(ip net.IP) error { + return connectTCP(ip, acceptPort, sendloopDuration) +} + // NATOutRedirectUDPPort tests that packets are redirected to different port. type NATOutRedirectUDPPort struct{} @@ -241,7 +293,7 @@ func (NATPreRedirectIP) Name() string { // ContainerAction implements TestCase.ContainerAction. func (NATPreRedirectIP) ContainerAction(ip net.IP) error { - addrs, err := localAddrs() + addrs, err := localAddrs(false) if err != nil { return err } diff --git a/test/packetdrill/defs.bzl b/test/packetdrill/defs.bzl index f499c177b..fc28ce9ba 100644 --- a/test/packetdrill/defs.bzl +++ b/test/packetdrill/defs.bzl @@ -26,7 +26,7 @@ def _packetdrill_test_impl(ctx): transitive_files = depset() if hasattr(ctx.attr._test_runner, "data_runfiles"): - transitive_files = depset(ctx.attr._test_runner.data_runfiles.files) + transitive_files = ctx.attr._test_runner.data_runfiles.files runfiles = ctx.runfiles( files = [test_runner] + ctx.files._init_script + ctx.files.scripts, transitive_files = transitive_files, @@ -60,11 +60,15 @@ _packetdrill_test = rule( implementation = _packetdrill_test_impl, ) -_PACKETDRILL_TAGS = ["local", "manual"] +PACKETDRILL_TAGS = [ + "local", + "manual", + "packetdrill", +] def packetdrill_linux_test(name, **kwargs): if "tags" not in kwargs: - kwargs["tags"] = _PACKETDRILL_TAGS + kwargs["tags"] = PACKETDRILL_TAGS _packetdrill_test( name = name, flags = ["--dut_platform", "linux"], @@ -73,7 +77,7 @@ def packetdrill_linux_test(name, **kwargs): def packetdrill_netstack_test(name, **kwargs): if "tags" not in kwargs: - kwargs["tags"] = _PACKETDRILL_TAGS + kwargs["tags"] = PACKETDRILL_TAGS _packetdrill_test( name = name, # This is the default runtime unless diff --git a/test/packetimpact/dut/posix_server.cc b/test/packetimpact/dut/posix_server.cc index a1a5c3612..29d4cc6fe 100644 --- a/test/packetimpact/dut/posix_server.cc +++ b/test/packetimpact/dut/posix_server.cc @@ -53,7 +53,10 @@ response_in6->set_flowinfo(ntohl(addr_in6->sin6_flowinfo)); response_in6->mutable_addr()->assign( reinterpret_cast<const char *>(&addr_in6->sin6_addr.s6_addr), 16); - response_in6->set_scope_id(ntohl(addr_in6->sin6_scope_id)); + // sin6_scope_id is stored in host byte order. + // + // https://www.gnu.org/software/libc/manual/html_node/Internet-Address-Formats.html + response_in6->set_scope_id(addr_in6->sin6_scope_id); return ::grpc::Status::OK; } } @@ -89,7 +92,10 @@ addr_in6->sin6_flowinfo = htonl(proto_in6.flowinfo()); proto_in6.addr().copy( reinterpret_cast<char *>(&addr_in6->sin6_addr.s6_addr), 16); - addr_in6->sin6_scope_id = htonl(proto_in6.scope_id()); + // sin6_scope_id is stored in host byte order. + // + // https://www.gnu.org/software/libc/manual/html_node/Internet-Address-Formats.html + addr_in6->sin6_scope_id = proto_in6.scope_id(); *addr_len = sizeof(*addr_in6); break; } diff --git a/test/packetimpact/netdevs/BUILD b/test/packetimpact/netdevs/BUILD index 422bb9b0c..8d1193fed 100644 --- a/test/packetimpact/netdevs/BUILD +++ b/test/packetimpact/netdevs/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package( licenses = ["notice"], @@ -13,3 +13,11 @@ go_library( "//pkg/tcpip/header", ], ) + +go_test( + name = "netdevs_test", + size = "small", + srcs = ["netdevs_test.go"], + library = ":netdevs", + deps = ["@com_github_google_go_cmp//cmp:go_default_library"], +) diff --git a/test/packetimpact/netdevs/netdevs.go b/test/packetimpact/netdevs/netdevs.go index d2c9cfeaf..eecfe0730 100644 --- a/test/packetimpact/netdevs/netdevs.go +++ b/test/packetimpact/netdevs/netdevs.go @@ -19,6 +19,7 @@ import ( "fmt" "net" "regexp" + "strconv" "strings" "gvisor.dev/gvisor/pkg/tcpip" @@ -27,6 +28,7 @@ import ( // A DeviceInfo represents a network device. type DeviceInfo struct { + ID uint32 MAC net.HardwareAddr IPv4Addr net.IP IPv4Net *net.IPNet @@ -35,7 +37,7 @@ type DeviceInfo struct { } var ( - deviceLine = regexp.MustCompile(`^\s*\d+: (\w+)`) + deviceLine = regexp.MustCompile(`^\s*(\d+): (\w+)`) linkLine = regexp.MustCompile(`^\s*link/\w+ ([0-9a-fA-F:]+)`) inetLine = regexp.MustCompile(`^\s*inet ([0-9./]+)`) inet6Line = regexp.MustCompile(`^\s*inet6 ([0-9a-fA-Z:/]+)`) @@ -43,6 +45,11 @@ var ( // ParseDevices parses the output from `ip addr show` into a map from device // name to information about the device. +// +// Note: if multiple IPv6 addresses are assigned to a device, the last address +// displayed by `ip addr show` will be used. This is fine for packetimpact +// because we will always only have at most one IPv6 address assigned to each +// device. func ParseDevices(cmdOutput string) (map[string]DeviceInfo, error) { var currentDevice string var currentInfo DeviceInfo @@ -52,8 +59,12 @@ func ParseDevices(cmdOutput string) (map[string]DeviceInfo, error) { if currentDevice != "" { deviceInfos[currentDevice] = currentInfo } - currentInfo = DeviceInfo{} - currentDevice = m[1] + id, err := strconv.ParseUint(m[1], 10, 32) + if err != nil { + return nil, fmt.Errorf("parsing device ID %s: %w", m[1], err) + } + currentInfo = DeviceInfo{ID: uint32(id)} + currentDevice = m[2] } else if m := linkLine.FindStringSubmatch(line); m != nil { mac, err := net.ParseMAC(m[1]) if err != nil { diff --git a/test/packetimpact/netdevs/netdevs_test.go b/test/packetimpact/netdevs/netdevs_test.go new file mode 100644 index 000000000..24ad12198 --- /dev/null +++ b/test/packetimpact/netdevs/netdevs_test.go @@ -0,0 +1,227 @@ +// 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 netdevs + +import ( + "fmt" + "net" + "testing" + + "github.com/google/go-cmp/cmp" +) + +func mustParseMAC(s string) net.HardwareAddr { + mac, err := net.ParseMAC(s) + if err != nil { + panic(fmt.Sprintf("failed to parse test MAC %q: %s", s, err)) + } + return mac +} + +func TestParseDevices(t *testing.T) { + for _, v := range []struct { + desc string + cmdOutput string + want map[string]DeviceInfo + }{ + { + desc: "v4 and v6", + cmdOutput: ` +1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000 + link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 + inet 127.0.0.1/8 scope host lo + valid_lft forever preferred_lft forever + inet6 ::1/128 scope host + valid_lft forever preferred_lft forever +2613: eth0@if2614: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:c0:a8:09:02 brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 192.168.9.2/24 brd 192.168.9.255 scope global eth0 + valid_lft forever preferred_lft forever + inet6 fe80::42:c0ff:fea8:902/64 scope link tentative + valid_lft forever preferred_lft forever +2615: eth2@if2616: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:df:f5:e1:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 223.245.225.10/24 brd 223.245.225.255 scope global eth2 + valid_lft forever preferred_lft forever + inet6 fe80::42:dfff:fef5:e10a/64 scope link tentative + valid_lft forever preferred_lft forever +2617: eth1@if2618: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:da:33:13:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 218.51.19.10/24 brd 218.51.19.255 scope global eth1 + valid_lft forever preferred_lft forever + inet6 fe80::42:daff:fe33:130a/64 scope link tentative + valid_lft forever preferred_lft forever`, + want: map[string]DeviceInfo{ + "lo": DeviceInfo{ + ID: 1, + MAC: mustParseMAC("00:00:00:00:00:00"), + IPv4Addr: net.IPv4(127, 0, 0, 1), + IPv4Net: &net.IPNet{ + IP: net.IPv4(127, 0, 0, 0), + Mask: net.CIDRMask(8, 32), + }, + IPv6Addr: net.ParseIP("::1"), + IPv6Net: &net.IPNet{ + IP: net.ParseIP("::1"), + Mask: net.CIDRMask(128, 128), + }, + }, + "eth0": DeviceInfo{ + ID: 2613, + MAC: mustParseMAC("02:42:c0:a8:09:02"), + IPv4Addr: net.IPv4(192, 168, 9, 2), + IPv4Net: &net.IPNet{ + IP: net.IPv4(192, 168, 9, 0), + Mask: net.CIDRMask(24, 32), + }, + IPv6Addr: net.ParseIP("fe80::42:c0ff:fea8:902"), + IPv6Net: &net.IPNet{ + IP: net.ParseIP("fe80::"), + Mask: net.CIDRMask(64, 128), + }, + }, + "eth1": DeviceInfo{ + ID: 2617, + MAC: mustParseMAC("02:42:da:33:13:0a"), + IPv4Addr: net.IPv4(218, 51, 19, 10), + IPv4Net: &net.IPNet{ + IP: net.IPv4(218, 51, 19, 0), + Mask: net.CIDRMask(24, 32), + }, + IPv6Addr: net.ParseIP("fe80::42:daff:fe33:130a"), + IPv6Net: &net.IPNet{ + IP: net.ParseIP("fe80::"), + Mask: net.CIDRMask(64, 128), + }, + }, + "eth2": DeviceInfo{ + ID: 2615, + MAC: mustParseMAC("02:42:df:f5:e1:0a"), + IPv4Addr: net.IPv4(223, 245, 225, 10), + IPv4Net: &net.IPNet{ + IP: net.IPv4(223, 245, 225, 0), + Mask: net.CIDRMask(24, 32), + }, + IPv6Addr: net.ParseIP("fe80::42:dfff:fef5:e10a"), + IPv6Net: &net.IPNet{ + IP: net.ParseIP("fe80::"), + Mask: net.CIDRMask(64, 128), + }, + }, + }, + }, + { + desc: "v4 only", + cmdOutput: ` +2613: eth0@if2614: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:c0:a8:09:02 brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 192.168.9.2/24 brd 192.168.9.255 scope global eth0 + valid_lft forever preferred_lft forever`, + want: map[string]DeviceInfo{ + "eth0": DeviceInfo{ + ID: 2613, + MAC: mustParseMAC("02:42:c0:a8:09:02"), + IPv4Addr: net.IPv4(192, 168, 9, 2), + IPv4Net: &net.IPNet{ + IP: net.IPv4(192, 168, 9, 0), + Mask: net.CIDRMask(24, 32), + }, + }, + }, + }, + { + desc: "v6 only", + cmdOutput: ` +2615: eth2@if2616: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:df:f5:e1:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet6 fe80::42:dfff:fef5:e10a/64 scope link tentative + valid_lft forever preferred_lft forever`, + want: map[string]DeviceInfo{ + "eth2": DeviceInfo{ + ID: 2615, + MAC: mustParseMAC("02:42:df:f5:e1:0a"), + IPv6Addr: net.ParseIP("fe80::42:dfff:fef5:e10a"), + IPv6Net: &net.IPNet{ + IP: net.ParseIP("fe80::"), + Mask: net.CIDRMask(64, 128), + }, + }, + }, + }, + } { + t.Run(v.desc, func(t *testing.T) { + got, err := ParseDevices(v.cmdOutput) + if err != nil { + t.Errorf("ParseDevices(\n%s\n) got unexpected error: %s", v.cmdOutput, err) + } + if diff := cmp.Diff(v.want, got); diff != "" { + t.Errorf("ParseDevices(\n%s\n) got output diff (-want, +got):\n%s", v.cmdOutput, diff) + } + }) + } +} + +func TestParseDevicesErrors(t *testing.T) { + for _, v := range []struct { + desc string + cmdOutput string + }{ + { + desc: "invalid MAC addr", + cmdOutput: ` +2617: eth1@if2618: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:da:33:13:0a:ffffffff brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 218.51.19.10/24 brd 218.51.19.255 scope global eth1 + valid_lft forever preferred_lft forever + inet6 fe80::42:daff:fe33:130a/64 scope link tentative + valid_lft forever preferred_lft forever`, + }, + { + desc: "invalid v4 addr", + cmdOutput: ` +2617: eth1@if2618: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:da:33:13:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 1234.4321.424242.0/24 brd 218.51.19.255 scope global eth1 + valid_lft forever preferred_lft forever + inet6 fe80::42:daff:fe33:130a/64 scope link tentative + valid_lft forever preferred_lft forever`, + }, + { + desc: "invalid v6 addr", + cmdOutput: ` +2617: eth1@if2618: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:da:33:13:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 218.51.19.10/24 brd 218.51.19.255 scope global eth1 + valid_lft forever preferred_lft forever + inet6 fe80:ffffffff::42:daff:fe33:130a/64 scope link tentative + valid_lft forever preferred_lft forever`, + }, + { + desc: "invalid CIDR missing prefixlen", + cmdOutput: ` +2617: eth1@if2618: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default + link/ether 02:42:da:33:13:0a brd ff:ff:ff:ff:ff:ff link-netnsid 0 + inet 218.51.19.10 brd 218.51.19.255 scope global eth1 + valid_lft forever preferred_lft forever + inet6 fe80::42:daff:fe33:130a scope link tentative + valid_lft forever preferred_lft forever`, + }, + } { + t.Run(v.desc, func(t *testing.T) { + if _, err := ParseDevices(v.cmdOutput); err == nil { + t.Errorf("ParseDevices(\n%s\n) succeeded unexpectedly, want error", v.cmdOutput) + } + }) + } +} diff --git a/test/packetimpact/runner/BUILD b/test/packetimpact/runner/BUILD index 0b68a760a..bad4f0183 100644 --- a/test/packetimpact/runner/BUILD +++ b/test/packetimpact/runner/BUILD @@ -16,5 +16,6 @@ go_test( deps = [ "//pkg/test/dockerutil", "//test/packetimpact/netdevs", + "@com_github_docker_docker//api/types/mount:go_default_library", ], ) diff --git a/test/packetimpact/runner/defs.bzl b/test/packetimpact/runner/defs.bzl index 77cdfea12..79b3c9162 100644 --- a/test/packetimpact/runner/defs.bzl +++ b/test/packetimpact/runner/defs.bzl @@ -55,7 +55,11 @@ _packetimpact_test = rule( implementation = _packetimpact_test_impl, ) -PACKETIMPACT_TAGS = ["local", "manual"] +PACKETIMPACT_TAGS = [ + "local", + "manual", + "packetimpact", +] def packetimpact_linux_test( name, @@ -75,7 +79,7 @@ def packetimpact_linux_test( name = name + "_linux_test", testbench_binary = testbench_binary, flags = ["--dut_platform", "linux"] + expect_failure_flag, - tags = PACKETIMPACT_TAGS + ["packetimpact"], + tags = PACKETIMPACT_TAGS, **kwargs ) @@ -101,7 +105,7 @@ def packetimpact_netstack_test( # This is the default runtime unless # "--test_arg=--runtime=OTHER_RUNTIME" is used to override the value. flags = ["--dut_platform", "netstack", "--runtime=runsc-d"] + expect_failure_flag, - tags = PACKETIMPACT_TAGS + ["packetimpact"], + tags = PACKETIMPACT_TAGS, **kwargs ) @@ -121,7 +125,10 @@ def packetimpact_go_test(name, size = "small", pure = True, expect_linux_failure name = testbench_binary, size = size, pure = pure, - tags = PACKETIMPACT_TAGS, + tags = [ + "local", + "manual", + ], **kwargs ) packetimpact_linux_test( diff --git a/test/packetimpact/runner/packetimpact_test.go b/test/packetimpact/runner/packetimpact_test.go index c0a2620de..1a0221893 100644 --- a/test/packetimpact/runner/packetimpact_test.go +++ b/test/packetimpact/runner/packetimpact_test.go @@ -16,6 +16,7 @@ package packetimpact_test import ( + "context" "flag" "fmt" "io/ioutil" @@ -29,6 +30,7 @@ import ( "testing" "time" + "github.com/docker/docker/api/types/mount" "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/test/packetimpact/netdevs" ) @@ -94,15 +96,16 @@ func TestOne(t *testing.T) { } } dockerutil.EnsureSupportedDockerVersion() + ctx := context.Background() // Create the networks needed for the test. One control network is needed for // the gRPC control packets and one test network on which to transmit the test // packets. - ctrlNet := dockerutil.NewDockerNetwork(logger("ctrlNet")) - testNet := dockerutil.NewDockerNetwork(logger("testNet")) - for _, dn := range []*dockerutil.DockerNetwork{ctrlNet, testNet} { + ctrlNet := dockerutil.NewNetwork(ctx, logger("ctrlNet")) + testNet := dockerutil.NewNetwork(ctx, logger("testNet")) + for _, dn := range []*dockerutil.Network{ctrlNet, testNet} { for { - if err := createDockerNetwork(dn); err != nil { + if err := createDockerNetwork(ctx, dn); err != nil { t.Log("creating docker network:", err) const wait = 100 * time.Millisecond t.Logf("sleeping %s and will try creating docker network again", wait) @@ -113,11 +116,19 @@ func TestOne(t *testing.T) { } break } - defer func(dn *dockerutil.DockerNetwork) { - if err := dn.Cleanup(); err != nil { + defer func(dn *dockerutil.Network) { + if err := dn.Cleanup(ctx); err != nil { t.Errorf("unable to cleanup container %s: %s", dn.Name, err) } }(dn) + // Sanity check. + inspect, err := dn.Inspect(ctx) + if err != nil { + t.Fatalf("failed to inspect network %s: %v", dn.Name, err) + } else if inspect.Name != dn.Name { + t.Fatalf("name mismatch for network want: %s got: %s", dn.Name, inspect.Name) + } + } tmpDir, err := ioutil.TempDir("", "container-output") @@ -128,42 +139,51 @@ func TestOne(t *testing.T) { const testOutputDir = "/tmp/testoutput" - runOpts := dockerutil.RunOpts{ - Image: "packetimpact", - CapAdd: []string{"NET_ADMIN"}, - Extra: []string{"--sysctl", "net.ipv6.conf.all.disable_ipv6=0", "--rm", "-v", tmpDir + ":" + testOutputDir}, - Foreground: true, - } - // Create the Docker container for the DUT. - dut := dockerutil.MakeDocker(logger("dut")) + dut := dockerutil.MakeContainer(ctx, logger("dut")) if *dutPlatform == "linux" { dut.Runtime = "" } + runOpts := dockerutil.RunOpts{ + Image: "packetimpact", + CapAdd: []string{"NET_ADMIN"}, + Mounts: []mount.Mount{mount.Mount{ + Type: mount.TypeBind, + Source: tmpDir, + Target: testOutputDir, + ReadOnly: false, + }}, + } + const containerPosixServerBinary = "/packetimpact/posix_server" dut.CopyFiles(&runOpts, "/packetimpact", "/test/packetimpact/dut/posix_server") - if err := dut.Create(runOpts, containerPosixServerBinary, "--ip=0.0.0.0", "--port="+ctrlPort); err != nil { - t.Fatalf("unable to create container %s: %s", dut.Name, err) + conf, hostconf, _ := dut.ConfigsFrom(runOpts, containerPosixServerBinary, "--ip=0.0.0.0", "--port="+ctrlPort) + hostconf.AutoRemove = true + hostconf.Sysctls = map[string]string{"net.ipv6.conf.all.disable_ipv6": "0"} + + if err := dut.CreateFrom(ctx, conf, hostconf, nil); err != nil { + t.Fatalf("unable to create container %s: %v", dut.Name, err) } - defer dut.CleanUp() + + defer dut.CleanUp(ctx) // Add ctrlNet as eth1 and testNet as eth2. const testNetDev = "eth2" - if err := addNetworks(dut, dutAddr, []*dockerutil.DockerNetwork{ctrlNet, testNet}); err != nil { + if err := addNetworks(ctx, dut, dutAddr, []*dockerutil.Network{ctrlNet, testNet}); err != nil { t.Fatal(err) } - if err := dut.Start(); err != nil { + if err := dut.Start(ctx); err != nil { t.Fatalf("unable to start container %s: %s", dut.Name, err) } - if _, err := dut.WaitForOutput("Server listening.*\n", 60*time.Second); err != nil { + if _, err := dut.WaitForOutput(ctx, "Server listening.*\n", 60*time.Second); err != nil { t.Fatalf("%s on container %s never listened: %s", containerPosixServerBinary, dut.Name, err) } - dutTestDevice, dutDeviceInfo, err := deviceByIP(dut, addressInSubnet(dutAddr, *testNet.Subnet)) + dutTestDevice, dutDeviceInfo, err := deviceByIP(ctx, dut, addressInSubnet(dutAddr, *testNet.Subnet)) if err != nil { t.Fatal(err) } @@ -173,11 +193,11 @@ func TestOne(t *testing.T) { // Netstack as DUT doesn't assign IPv6 addresses automatically so do it if // needed. if remoteIPv6 == nil { - if _, err := dut.Exec(dockerutil.RunOpts{}, "ip", "addr", "add", netdevs.MACToIP(remoteMAC).String(), "scope", "link", "dev", dutTestDevice); err != nil { + if _, err := dut.Exec(ctx, dockerutil.ExecOpts{}, "ip", "addr", "add", netdevs.MACToIP(remoteMAC).String(), "scope", "link", "dev", dutTestDevice); err != nil { t.Fatalf("unable to ip addr add on container %s: %s", dut.Name, err) } // Now try again, to make sure that it worked. - _, dutDeviceInfo, err = deviceByIP(dut, addressInSubnet(dutAddr, *testNet.Subnet)) + _, dutDeviceInfo, err = deviceByIP(ctx, dut, addressInSubnet(dutAddr, *testNet.Subnet)) if err != nil { t.Fatal(err) } @@ -188,16 +208,20 @@ func TestOne(t *testing.T) { } // Create the Docker container for the testbench. - testbench := dockerutil.MakeDocker(logger("testbench")) + testbench := dockerutil.MakeContainer(ctx, logger("testbench")) testbench.Runtime = "" // The testbench always runs on Linux. tbb := path.Base(*testbenchBinary) containerTestbenchBinary := "/packetimpact/" + tbb runOpts = dockerutil.RunOpts{ - Image: "packetimpact", - CapAdd: []string{"NET_ADMIN"}, - Extra: []string{"--sysctl", "net.ipv6.conf.all.disable_ipv6=0", "--rm", "-v", tmpDir + ":" + testOutputDir}, - Foreground: true, + Image: "packetimpact", + CapAdd: []string{"NET_ADMIN"}, + Mounts: []mount.Mount{mount.Mount{ + Type: mount.TypeBind, + Source: tmpDir, + Target: testOutputDir, + ReadOnly: false, + }}, } testbench.CopyFiles(&runOpts, "/packetimpact", "/test/packetimpact/tests/"+tbb) @@ -227,39 +251,42 @@ func TestOne(t *testing.T) { } }() - if err := testbench.Create(runOpts, snifferArgs...); err != nil { + conf, hostconf, _ = testbench.ConfigsFrom(runOpts, snifferArgs...) + hostconf.AutoRemove = true + hostconf.Sysctls = map[string]string{"net.ipv6.conf.all.disable_ipv6": "0"} + + if err := testbench.CreateFrom(ctx, conf, hostconf, nil); err != nil { t.Fatalf("unable to create container %s: %s", testbench.Name, err) } - defer testbench.CleanUp() + defer testbench.CleanUp(ctx) // Add ctrlNet as eth1 and testNet as eth2. - if err := addNetworks(testbench, testbenchAddr, []*dockerutil.DockerNetwork{ctrlNet, testNet}); err != nil { + if err := addNetworks(ctx, testbench, testbenchAddr, []*dockerutil.Network{ctrlNet, testNet}); err != nil { t.Fatal(err) } - if err := testbench.Start(); err != nil { + if err := testbench.Start(ctx); err != nil { t.Fatalf("unable to start container %s: %s", testbench.Name, err) } // Kill so that it will flush output. defer func() { - // Wait 1 second before killing tcpdump to give it time to flush - // any packets. On linux tests killing it immediately can - // sometimes result in partial pcaps. time.Sleep(1 * time.Second) - testbench.Exec(dockerutil.RunOpts{}, "killall", snifferArgs[0]) + testbench.Exec(ctx, dockerutil.ExecOpts{}, "killall", snifferArgs[0]) }() - if _, err := testbench.WaitForOutput(snifferRegex, 60*time.Second); err != nil { + if _, err := testbench.WaitForOutput(ctx, snifferRegex, 60*time.Second); err != nil { t.Fatalf("sniffer on %s never listened: %s", dut.Name, err) } // Because the Linux kernel receives the SYN-ACK but didn't send the SYN it - // will issue a RST. To prevent this IPtables can be used to filter out all + // will issue an RST. To prevent this IPtables can be used to filter out all // incoming packets. The raw socket that packetimpact tests use will still see // everything. - if _, err := testbench.Exec(dockerutil.RunOpts{}, "iptables", "-A", "INPUT", "-i", testNetDev, "-j", "DROP"); err != nil { - t.Fatalf("unable to Exec iptables on container %s: %s", testbench.Name, err) + for _, bin := range []string{"iptables", "ip6tables"} { + if logs, err := testbench.Exec(ctx, dockerutil.ExecOpts{}, bin, "-A", "INPUT", "-i", testNetDev, "-p", "tcp", "-j", "DROP"); err != nil { + t.Fatalf("unable to Exec %s on container %s: %s, logs from testbench:\n%s", bin, testbench.Name, err, logs) + } } // FIXME(b/156449515): Some piece of the system has a race. The old @@ -279,23 +306,41 @@ func TestOne(t *testing.T) { "--local_ipv4", addressInSubnet(testbenchAddr, *testNet.Subnet).String(), "--remote_ipv6", remoteIPv6.String(), "--remote_mac", remoteMAC.String(), + "--remote_interface_id", fmt.Sprintf("%d", dutDeviceInfo.ID), "--device", testNetDev, "--dut_type", *dutPlatform, ) - _, err = testbench.Exec(dockerutil.RunOpts{}, testArgs...) - if !*expectFailure && err != nil { - t.Fatal("test failed:", err) - } - if *expectFailure && err == nil { - t.Fatal("test failure expected but the test succeeded, enable the test and mark the corresponding bug as fixed") + testbenchLogs, err := testbench.Exec(ctx, dockerutil.ExecOpts{}, testArgs...) + if (err != nil) != *expectFailure { + var dutLogs string + if logs, err := dut.Logs(ctx); err != nil { + dutLogs = fmt.Sprintf("failed to fetch DUT logs: %s", err) + } else { + dutLogs = logs + } + + t.Errorf(`test error: %v, expect failure: %t + +====== Begin of DUT Logs ====== + +%s + +====== End of DUT Logs ====== + +====== Begin of Testbench Logs ====== + +%s + +====== End of Testbench Logs ======`, + err, *expectFailure, dutLogs, testbenchLogs) } } -func addNetworks(d *dockerutil.Docker, addr net.IP, networks []*dockerutil.DockerNetwork) error { +func addNetworks(ctx context.Context, d *dockerutil.Container, addr net.IP, networks []*dockerutil.Network) error { for _, dn := range networks { ip := addressInSubnet(addr, *dn.Subnet) // Connect to the network with the specified IP address. - if err := dn.Connect(d, "--ip", ip.String()); err != nil { + if err := dn.Connect(ctx, d, ip.String(), ""); err != nil { return fmt.Errorf("unable to connect container %s to network %s: %w", d.Name, dn.Name, err) } } @@ -313,9 +358,9 @@ func addressInSubnet(addr net.IP, subnet net.IPNet) net.IP { return net.IP(octets) } -// makeDockerNetwork makes a randomly-named network that will start with the +// createDockerNetwork makes a randomly-named network that will start with the // namePrefix. The network will be a random /24 subnet. -func createDockerNetwork(n *dockerutil.DockerNetwork) error { +func createDockerNetwork(ctx context.Context, n *dockerutil.Network) error { randSource := rand.NewSource(time.Now().UnixNano()) r1 := rand.New(randSource) // Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. @@ -324,12 +369,12 @@ func createDockerNetwork(n *dockerutil.DockerNetwork) error { IP: ip, Mask: ip.DefaultMask(), } - return n.Create() + return n.Create(ctx) } // deviceByIP finds a deviceInfo and device name from an IP address. -func deviceByIP(d *dockerutil.Docker, ip net.IP) (string, netdevs.DeviceInfo, error) { - out, err := d.Exec(dockerutil.RunOpts{}, "ip", "addr", "show") +func deviceByIP(ctx context.Context, d *dockerutil.Container, ip net.IP) (string, netdevs.DeviceInfo, error) { + out, err := d.Exec(ctx, dockerutil.ExecOpts{}, "ip", "addr", "show") if err != nil { return "", netdevs.DeviceInfo{}, fmt.Errorf("listing devices on %s container: %w", d.Name, err) } diff --git a/test/packetimpact/testbench/BUILD b/test/packetimpact/testbench/BUILD index d19ec07d4..5a0ee1367 100644 --- a/test/packetimpact/testbench/BUILD +++ b/test/packetimpact/testbench/BUILD @@ -23,8 +23,8 @@ go_library( "//pkg/usermem", "//test/packetimpact/netdevs", "//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_google_go_cmp//cmp:go_default_library", + "@com_github_google_go_cmp//cmp/cmpopts: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", diff --git a/test/packetimpact/testbench/connections.go b/test/packetimpact/testbench/connections.go index 8b4a4d905..87ce58c24 100644 --- a/test/packetimpact/testbench/connections.go +++ b/test/packetimpact/testbench/connections.go @@ -41,16 +41,19 @@ func portFromSockaddr(sa unix.Sockaddr) (uint16, error) { return 0, fmt.Errorf("sockaddr type %T does not contain port", sa) } -// pickPort makes a new socket and returns the socket FD and port. The domain should be AF_INET or AF_INET6. The caller must close the FD when done with +// pickPort makes a new socket and returns the socket FD and port. The domain +// should be AF_INET or AF_INET6. The caller must close the FD when done with // the port if there is no error. -func pickPort(domain, typ int) (int, uint16, error) { - fd, err := unix.Socket(domain, typ, 0) +func pickPort(domain, typ int) (fd int, port uint16, err error) { + fd, err = unix.Socket(domain, typ, 0) if err != nil { - return -1, 0, err + return -1, 0, fmt.Errorf("creating socket: %w", err) } defer func() { if err != nil { - err = multierr.Append(err, unix.Close(fd)) + if cerr := unix.Close(fd); cerr != nil { + err = multierr.Append(err, fmt.Errorf("failed to close socket %d: %w", fd, cerr)) + } } }() var sa unix.Sockaddr @@ -60,22 +63,22 @@ func pickPort(domain, typ int) (int, uint16, error) { copy(sa4.Addr[:], net.ParseIP(LocalIPv4).To4()) sa = &sa4 case unix.AF_INET6: - var sa6 unix.SockaddrInet6 + sa6 := unix.SockaddrInet6{ZoneId: uint32(LocalInterfaceID)} copy(sa6.Addr[:], net.ParseIP(LocalIPv6).To16()) sa = &sa6 default: return -1, 0, fmt.Errorf("invalid domain %d, it should be one of unix.AF_INET or unix.AF_INET6", domain) } if err = unix.Bind(fd, sa); err != nil { - return -1, 0, err + return -1, 0, fmt.Errorf("binding to %+v: %w", sa, err) } sa, err = unix.Getsockname(fd) if err != nil { - return -1, 0, err + return -1, 0, fmt.Errorf("Getsocketname(%d): %w", fd, err) } - port, err := portFromSockaddr(sa) + port, err = portFromSockaddr(sa) if err != nil { - return -1, 0, err + return -1, 0, fmt.Errorf("extracting port from socket address %+v: %w", sa, err) } return fd, port, nil } @@ -378,7 +381,7 @@ var _ layerState = (*udpState)(nil) func newUDPState(domain int, out, in UDP) (*udpState, error) { portPickerFD, localPort, err := pickPort(domain, unix.SOCK_DGRAM) if err != nil { - return nil, err + return nil, fmt.Errorf("picking port: %w", err) } s := udpState{ out: UDP{SrcPort: &localPort}, @@ -916,14 +919,14 @@ func (conn *UDPIPv4) SendIP(ip IPv4, udp UDP, additionalLayers ...Layer) { func (conn *UDPIPv4) Expect(udp UDP, timeout time.Duration) (*UDP, error) { conn.t.Helper() layer, err := (*Connection)(conn).Expect(&udp, timeout) - if layer == nil { + if err != nil { return nil, err } gotUDP, ok := layer.(*UDP) if !ok { conn.t.Fatalf("expected %s to be UDP", layer) } - return gotUDP, err + return gotUDP, nil } // ExpectData is a convenient method that expects a Layer and the Layer after @@ -948,3 +951,169 @@ func (conn *UDPIPv4) Close() { func (conn *UDPIPv4) Drain() { conn.sniffer.Drain() } + +// UDPIPv6 maintains the state for all the layers in a UDP/IPv6 connection. +type UDPIPv6 Connection + +// NewUDPIPv6 creates a new UDPIPv6 connection with reasonable defaults. +func NewUDPIPv6(t *testing.T, outgoingUDP, incomingUDP UDP) UDPIPv6 { + etherState, err := newEtherState(Ether{}, Ether{}) + if err != nil { + t.Fatalf("can't make etherState: %s", err) + } + ipv6State, err := newIPv6State(IPv6{}, IPv6{}) + if err != nil { + t.Fatalf("can't make IPv6State: %s", err) + } + udpState, err := newUDPState(unix.AF_INET6, outgoingUDP, incomingUDP) + if err != nil { + t.Fatalf("can't make udpState: %s", err) + } + injector, err := NewInjector(t) + if err != nil { + t.Fatalf("can't make injector: %s", err) + } + sniffer, err := NewSniffer(t) + if err != nil { + t.Fatalf("can't make sniffer: %s", err) + } + return UDPIPv6{ + layerStates: []layerState{etherState, ipv6State, udpState}, + injector: injector, + sniffer: sniffer, + t: t, + } +} + +func (conn *UDPIPv6) udpState() *udpState { + state, ok := conn.layerStates[2].(*udpState) + if !ok { + conn.t.Fatalf("got transport-layer state type=%T, expected udpState", conn.layerStates[2]) + } + return state +} + +func (conn *UDPIPv6) ipv6State() *ipv6State { + state, ok := conn.layerStates[1].(*ipv6State) + if !ok { + conn.t.Fatalf("got network-layer state type=%T, expected ipv6State", conn.layerStates[1]) + } + return state +} + +// LocalAddr gets the local socket address of this connection. +func (conn *UDPIPv6) LocalAddr() *unix.SockaddrInet6 { + sa := &unix.SockaddrInet6{ + Port: int(*conn.udpState().out.SrcPort), + // Local address is in perspective to the remote host, so it's scoped to the + // ID of the remote interface. + ZoneId: uint32(RemoteInterfaceID), + } + copy(sa.Addr[:], *conn.ipv6State().out.SrcAddr) + return sa +} + +// Send sends a packet with reasonable defaults, potentially overriding the UDP +// layer and adding additionLayers. +func (conn *UDPIPv6) Send(udp UDP, additionalLayers ...Layer) { + (*Connection)(conn).send(Layers{&udp}, additionalLayers...) +} + +// SendIPv6 sends a packet with reasonable defaults, potentially overriding the +// UDP and IPv6 headers and adding additionLayers. +func (conn *UDPIPv6) SendIPv6(ip IPv6, udp UDP, additionalLayers ...Layer) { + (*Connection)(conn).send(Layers{&ip, &udp}, additionalLayers...) +} + +// Expect expects a frame with the UDP layer matching the provided UDP within +// the timeout specified. If it doesn't arrive in time, an error is returned. +func (conn *UDPIPv6) Expect(udp UDP, timeout time.Duration) (*UDP, error) { + conn.t.Helper() + layer, err := (*Connection)(conn).Expect(&udp, timeout) + if err != nil { + return nil, err + } + gotUDP, ok := layer.(*UDP) + if !ok { + conn.t.Fatalf("expected %s to be UDP", layer) + } + return gotUDP, nil +} + +// ExpectData is a convenient method that expects a Layer and the Layer after +// it. If it doens't arrive in time, it returns nil. +func (conn *UDPIPv6) ExpectData(udp UDP, payload Payload, timeout time.Duration) (Layers, error) { + conn.t.Helper() + expected := make([]Layer, len(conn.layerStates)) + expected[len(expected)-1] = &udp + if payload.length() != 0 { + expected = append(expected, &payload) + } + return (*Connection)(conn).ExpectFrame(expected, timeout) +} + +// Close frees associated resources held by the UDPIPv6 connection. +func (conn *UDPIPv6) Close() { + (*Connection)(conn).Close() +} + +// Drain drains the sniffer's receive buffer by receiving packets until there's +// nothing else to receive. +func (conn *UDPIPv6) Drain() { + conn.sniffer.Drain() +} + +// TCPIPv6 maintains the state for all the layers in a TCP/IPv6 connection. +type TCPIPv6 Connection + +// NewTCPIPv6 creates a new TCPIPv6 connection with reasonable defaults. +func NewTCPIPv6(t *testing.T, outgoingTCP, incomingTCP TCP) TCPIPv6 { + etherState, err := newEtherState(Ether{}, Ether{}) + if err != nil { + t.Fatalf("can't make etherState: %s", err) + } + ipv6State, err := newIPv6State(IPv6{}, IPv6{}) + if err != nil { + t.Fatalf("can't make ipv6State: %s", err) + } + tcpState, err := newTCPState(unix.AF_INET6, outgoingTCP, incomingTCP) + if err != nil { + t.Fatalf("can't make tcpState: %s", err) + } + injector, err := NewInjector(t) + if err != nil { + t.Fatalf("can't make injector: %s", err) + } + sniffer, err := NewSniffer(t) + if err != nil { + t.Fatalf("can't make sniffer: %s", err) + } + + return TCPIPv6{ + layerStates: []layerState{etherState, ipv6State, tcpState}, + injector: injector, + sniffer: sniffer, + t: t, + } +} + +func (conn *TCPIPv6) SrcPort() uint16 { + state := conn.layerStates[2].(*tcpState) + return *state.out.SrcPort +} + +// ExpectData is a convenient method that expects a Layer and the Layer after +// it. If it doens't arrive in time, it returns nil. +func (conn *TCPIPv6) ExpectData(tcp *TCP, payload *Payload, timeout time.Duration) (Layers, error) { + expected := make([]Layer, len(conn.layerStates)) + expected[len(expected)-1] = tcp + if payload != nil { + expected = append(expected, payload) + } + return (*Connection)(conn).ExpectFrame(expected, timeout) +} + +// Close frees associated resources held by the TCPIPv6 connection. +func (conn *TCPIPv6) Close() { + (*Connection)(conn).Close() +} diff --git a/test/packetimpact/testbench/dut.go b/test/packetimpact/testbench/dut.go index 2a2afecb5..51be13759 100644 --- a/test/packetimpact/testbench/dut.go +++ b/test/packetimpact/testbench/dut.go @@ -87,7 +87,7 @@ func (dut *DUT) sockaddrToProto(sa unix.Sockaddr) *pb.Sockaddr { }, } } - dut.t.Fatalf("can't parse Sockaddr: %+v", sa) + dut.t.Fatalf("can't parse Sockaddr struct: %+v", sa) return nil } @@ -106,8 +106,9 @@ func (dut *DUT) protoToSockaddr(sa *pb.Sockaddr) unix.Sockaddr { ZoneId: s.In6.GetScopeId(), } copy(ret.Addr[:], s.In6.GetAddr()) + return &ret } - dut.t.Fatalf("can't parse Sockaddr: %+v", sa) + dut.t.Fatalf("can't parse Sockaddr proto: %+v", sa) return nil } @@ -126,6 +127,7 @@ func (dut *DUT) CreateBoundSocket(typ, proto int32, addr net.IP) (int32, uint16) fd = dut.Socket(unix.AF_INET6, typ, proto) sa := unix.SockaddrInet6{} copy(sa.Addr[:], addr.To16()) + sa.ZoneId = uint32(RemoteInterfaceID) dut.Bind(fd, &sa) } else { dut.t.Fatalf("unknown ip addr type for remoteIP") diff --git a/test/packetimpact/testbench/layers.go b/test/packetimpact/testbench/layers.go index a8121b0da..24aa46cce 100644 --- a/test/packetimpact/testbench/layers.go +++ b/test/packetimpact/testbench/layers.go @@ -15,6 +15,7 @@ package testbench import ( + "encoding/binary" "encoding/hex" "fmt" "reflect" @@ -470,21 +471,11 @@ func (l *IPv6) ToBytes() ([]byte, error) { if l.NextHeader != nil { fields.NextHeader = *l.NextHeader } else { - switch n := l.next().(type) { - case *TCP: - fields.NextHeader = uint8(header.TCPProtocolNumber) - case *UDP: - fields.NextHeader = uint8(header.UDPProtocolNumber) - case *ICMPv6: - fields.NextHeader = uint8(header.ICMPv6ProtocolNumber) - case *IPv6HopByHopOptionsExtHdr: - fields.NextHeader = uint8(header.IPv6HopByHopOptionsExtHdrIdentifier) - case *IPv6DestinationOptionsExtHdr: - fields.NextHeader = uint8(header.IPv6DestinationOptionsExtHdrIdentifier) - default: - // TODO(b/150301488): Support more protocols as needed. - return nil, fmt.Errorf("ToBytes can't deduce the IPv6 header's next protocol: %#v", n) + nh, err := nextHeaderByLayer(l.next()) + if err != nil { + return nil, err } + fields.NextHeader = nh } if l.HopLimit != nil { fields.HopLimit = *l.HopLimit @@ -514,6 +505,8 @@ func nextIPv6PayloadParser(nextHeader uint8) layerParser { return parseIPv6HopByHopOptionsExtHdr case header.IPv6DestinationOptionsExtHdrIdentifier: return parseIPv6DestinationOptionsExtHdr + case header.IPv6FragmentExtHdrIdentifier: + return parseIPv6FragmentExtHdr } return parsePayload } @@ -566,14 +559,56 @@ type IPv6DestinationOptionsExtHdr struct { Options []byte } +// IPv6FragmentExtHdr can construct and match an IPv6 Fragment Extension Header. +type IPv6FragmentExtHdr struct { + LayerBase + NextHeader *header.IPv6ExtensionHeaderIdentifier + FragmentOffset *uint16 + MoreFragments *bool + Identification *uint32 +} + +// nextHeaderByLayer finds the correct next header protocol value for layer l. +func nextHeaderByLayer(l Layer) (uint8, error) { + if l == nil { + return uint8(header.IPv6NoNextHeaderIdentifier), nil + } + switch l.(type) { + case *TCP: + return uint8(header.TCPProtocolNumber), nil + case *UDP: + return uint8(header.UDPProtocolNumber), nil + case *ICMPv6: + return uint8(header.ICMPv6ProtocolNumber), nil + case *Payload: + return uint8(header.IPv6NoNextHeaderIdentifier), nil + case *IPv6HopByHopOptionsExtHdr: + return uint8(header.IPv6HopByHopOptionsExtHdrIdentifier), nil + case *IPv6DestinationOptionsExtHdr: + return uint8(header.IPv6DestinationOptionsExtHdrIdentifier), nil + case *IPv6FragmentExtHdr: + return uint8(header.IPv6FragmentExtHdrIdentifier), nil + default: + // TODO(b/161005083): Support more protocols as needed. + return 0, fmt.Errorf("failed to deduce the IPv6 header's next protocol: %T", l) + } +} + // ipv6OptionsExtHdrToBytes serializes an options extension header into bytes. -func ipv6OptionsExtHdrToBytes(nextHeader *header.IPv6ExtensionHeaderIdentifier, options []byte) []byte { +func ipv6OptionsExtHdrToBytes(nextHeader *header.IPv6ExtensionHeaderIdentifier, nextLayer Layer, options []byte) ([]byte, error) { length := len(options) + 2 + if length%8 != 0 { + return nil, fmt.Errorf("IPv6 extension headers must be a multiple of 8 octets long, but the length given: %d, options: %s", length, hex.Dump(options)) + } bytes := make([]byte, length) - if nextHeader == nil { - bytes[0] = byte(header.IPv6NoNextHeaderIdentifier) - } else { + if nextHeader != nil { bytes[0] = byte(*nextHeader) + } else { + nh, err := nextHeaderByLayer(nextLayer) + if err != nil { + return nil, err + } + bytes[0] = nh } // ExtHdrLen field is the length of the extension header // in 8-octet unit, ignoring the first 8 octets. @@ -581,7 +616,7 @@ func ipv6OptionsExtHdrToBytes(nextHeader *header.IPv6ExtensionHeaderIdentifier, // https://tools.ietf.org/html/rfc2460#section-4.6 bytes[1] = uint8((length - 8) / 8) copy(bytes[2:], options) - return bytes + return bytes, nil } // IPv6ExtHdrIdent is a helper routine that allocates a new @@ -591,14 +626,45 @@ func IPv6ExtHdrIdent(id header.IPv6ExtensionHeaderIdentifier) *header.IPv6Extens return &id } -// ToBytes implements Layer.ToBytes +// ToBytes implements Layer.ToBytes. func (l *IPv6HopByHopOptionsExtHdr) ToBytes() ([]byte, error) { - return ipv6OptionsExtHdrToBytes(l.NextHeader, l.Options), nil + return ipv6OptionsExtHdrToBytes(l.NextHeader, l.next(), l.Options) } -// ToBytes implements Layer.ToBytes +// ToBytes implements Layer.ToBytes. func (l *IPv6DestinationOptionsExtHdr) ToBytes() ([]byte, error) { - return ipv6OptionsExtHdrToBytes(l.NextHeader, l.Options), nil + return ipv6OptionsExtHdrToBytes(l.NextHeader, l.next(), l.Options) +} + +// ToBytes implements Layer.ToBytes. +func (l *IPv6FragmentExtHdr) ToBytes() ([]byte, error) { + var offset, mflag uint16 + var ident uint32 + bytes := make([]byte, header.IPv6FragmentExtHdrLength) + if l.NextHeader != nil { + bytes[0] = byte(*l.NextHeader) + } else { + nh, err := nextHeaderByLayer(l.next()) + if err != nil { + return nil, err + } + bytes[0] = nh + } + bytes[1] = 0 // reserved + if l.MoreFragments != nil && *l.MoreFragments { + mflag = 1 + } + if l.FragmentOffset != nil { + offset = *l.FragmentOffset + } + if l.Identification != nil { + ident = *l.Identification + } + offsetAndMflag := offset<<3 | mflag + binary.BigEndian.PutUint16(bytes[2:], offsetAndMflag) + binary.BigEndian.PutUint32(bytes[4:], ident) + + return bytes, nil } // parseIPv6ExtHdr parses an IPv6 extension header and returns the NextHeader @@ -631,6 +697,26 @@ func parseIPv6DestinationOptionsExtHdr(b []byte) (Layer, layerParser) { return &IPv6DestinationOptionsExtHdr{NextHeader: &nextHeader, Options: options}, nextParser } +// Bool is a helper routine that allocates a new +// bool value to store v and returns a pointer to it. +func Bool(v bool) *bool { + return &v +} + +// parseIPv6FragmentExtHdr parses the bytes assuming that they start +// with an IPv6 Fragment Extension Header. +func parseIPv6FragmentExtHdr(b []byte) (Layer, layerParser) { + nextHeader := b[0] + var extHdr header.IPv6FragmentExtHdr + copy(extHdr[:], b[2:]) + return &IPv6FragmentExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6ExtensionHeaderIdentifier(nextHeader)), + FragmentOffset: Uint16(extHdr.FragmentOffset()), + MoreFragments: Bool(extHdr.More()), + Identification: Uint32(extHdr.ID()), + }, nextIPv6PayloadParser(nextHeader) +} + func (l *IPv6HopByHopOptionsExtHdr) length() int { return len(l.Options) + 2 } @@ -667,13 +753,31 @@ func (l *IPv6DestinationOptionsExtHdr) String() string { return stringLayer(l) } +func (*IPv6FragmentExtHdr) length() int { + return header.IPv6FragmentExtHdrLength +} + +func (l *IPv6FragmentExtHdr) match(other Layer) bool { + return equalLayer(l, other) +} + +// merge overrides the values in l with the values from other but only in fields +// where the value is not nil. +func (l *IPv6FragmentExtHdr) merge(other Layer) error { + return mergeLayer(l, other) +} + +func (l *IPv6FragmentExtHdr) String() string { + return stringLayer(l) +} + // ICMPv6 can construct and match an ICMPv6 encapsulation. type ICMPv6 struct { LayerBase - Type *header.ICMPv6Type - Code *byte - Checksum *uint16 - NDPPayload []byte + Type *header.ICMPv6Type + Code *byte + Checksum *uint16 + Payload []byte } func (l *ICMPv6) String() string { @@ -684,7 +788,7 @@ func (l *ICMPv6) String() string { // ToBytes implements Layer.ToBytes. func (l *ICMPv6) ToBytes() ([]byte, error) { - b := make([]byte, header.ICMPv6HeaderSize+len(l.NDPPayload)) + b := make([]byte, header.ICMPv6HeaderSize+len(l.Payload)) h := header.ICMPv6(b) if l.Type != nil { h.SetType(*l.Type) @@ -692,7 +796,7 @@ func (l *ICMPv6) ToBytes() ([]byte, error) { if l.Code != nil { h.SetCode(*l.Code) } - copy(h.NDPPayload(), l.NDPPayload) + copy(h.NDPPayload(), l.Payload) if l.Checksum != nil { h.SetChecksum(*l.Checksum) } else { @@ -701,7 +805,11 @@ func (l *ICMPv6) ToBytes() ([]byte, error) { // We need to search forward to find the IPv6 header. for prev := l.Prev(); prev != nil; prev = prev.Prev() { if ipv6, ok := prev.(*IPv6); ok { - h.SetChecksum(header.ICMPv6Checksum(h, *ipv6.SrcAddr, *ipv6.DstAddr, buffer.VectorisedView{})) + payload, err := payload(l) + if err != nil { + return nil, err + } + h.SetChecksum(header.ICMPv6Checksum(h, *ipv6.SrcAddr, *ipv6.DstAddr, payload)) break } } @@ -725,10 +833,10 @@ func Byte(v byte) *byte { func parseICMPv6(b []byte) (Layer, layerParser) { h := header.ICMPv6(b) icmpv6 := ICMPv6{ - Type: ICMPv6Type(h.Type()), - Code: Byte(h.Code()), - Checksum: Uint16(h.Checksum()), - NDPPayload: h.NDPPayload(), + Type: ICMPv6Type(h.Type()), + Code: Byte(h.Code()), + Checksum: Uint16(h.Checksum()), + Payload: h.NDPPayload(), } return &icmpv6, nil } @@ -738,7 +846,7 @@ func (l *ICMPv6) match(other Layer) bool { } func (l *ICMPv6) length() int { - return header.ICMPv6HeaderSize + len(l.NDPPayload) + return header.ICMPv6HeaderSize + len(l.Payload) } // merge overrides the values in l with the values from other but only in fields @@ -904,12 +1012,14 @@ func payload(l Layer) (buffer.VectorisedView, error) { func layerChecksum(l Layer, protoNumber tcpip.TransportProtocolNumber) (uint16, error) { totalLength := uint16(totalLength(l)) var xsum uint16 - switch s := l.Prev().(type) { + switch p := l.Prev().(type) { case *IPv4: - xsum = header.PseudoHeaderChecksum(protoNumber, *s.SrcAddr, *s.DstAddr, totalLength) + xsum = header.PseudoHeaderChecksum(protoNumber, *p.SrcAddr, *p.DstAddr, totalLength) + case *IPv6: + xsum = header.PseudoHeaderChecksum(protoNumber, *p.SrcAddr, *p.DstAddr, totalLength) default: - // TODO(b/150301488): Support more protocols, like IPv6. - return 0, fmt.Errorf("can't get src and dst addr from previous layer: %#v", s) + // TODO(b/161246171): Support more protocols. + return 0, fmt.Errorf("checksum for protocol %d is not supported when previous layer is %T", protoNumber, p) } payloadBytes, err := payload(l) if err != nil { diff --git a/test/packetimpact/testbench/layers_test.go b/test/packetimpact/testbench/layers_test.go index 382a983a1..a2a763034 100644 --- a/test/packetimpact/testbench/layers_test.go +++ b/test/packetimpact/testbench/layers_test.go @@ -593,10 +593,107 @@ func TestIPv6ExtHdrOptions(t *testing.T) { Options: []byte{0x05, 0x02, 0x00, 0x00, 0x01, 0x00}, }, &ICMPv6{ - Type: ICMPv6Type(header.ICMPv6ParamProblem), - Code: Byte(0), - Checksum: Uint16(0x5f98), - NDPPayload: []byte{0x00, 0x00, 0x00, 0x06}, + Type: ICMPv6Type(header.ICMPv6ParamProblem), + Code: Byte(0), + Checksum: Uint16(0x5f98), + Payload: []byte{0x00, 0x00, 0x00, 0x06}, + }, + }, + }, + { + description: "IPv6/HopByHop/Fragment", + wantBytes: []byte{ + // IPv6 Header + 0x60, 0x00, 0x00, 0x00, 0x00, 0x10, 0x00, 0x40, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x01, 0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xde, 0xad, 0xbe, 0xef, + // HopByHop Options + 0x2c, 0x00, 0x05, 0x02, 0x00, 0x00, 0x01, 0x00, + // Fragment ExtHdr + 0x3b, 0x00, 0x03, 0x20, 0x00, 0x00, 0x00, 0x2a, + }, + wantLayers: []Layer{ + &IPv6{ + SrcAddr: Address(tcpip.Address(net.ParseIP("::1"))), + DstAddr: Address(tcpip.Address(net.ParseIP("fe80::dead:beef"))), + }, + &IPv6HopByHopOptionsExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6FragmentExtHdrIdentifier), + Options: []byte{0x05, 0x02, 0x00, 0x00, 0x01, 0x00}, + }, + &IPv6FragmentExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6NoNextHeaderIdentifier), + FragmentOffset: Uint16(100), + MoreFragments: Bool(false), + Identification: Uint32(42), + }, + &Payload{ + Bytes: nil, + }, + }, + }, + { + description: "IPv6/DestOpt/Fragment/Payload", + wantBytes: []byte{ + // IPv6 Header + 0x60, 0x00, 0x00, 0x00, 0x00, 0x1b, 0x3c, 0x40, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x01, 0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xde, 0xad, 0xbe, 0xef, + // Destination Options + 0x2c, 0x00, 0x05, 0x02, 0x00, 0x00, 0x01, 0x00, + // Fragment ExtHdr + 0x3b, 0x00, 0x03, 0x21, 0x00, 0x00, 0x00, 0x2a, + // Sample Data + 0x53, 0x61, 0x6d, 0x70, 0x6c, 0x65, 0x20, 0x44, 0x61, 0x74, 0x61, + }, + wantLayers: []Layer{ + &IPv6{ + SrcAddr: Address(tcpip.Address(net.ParseIP("::1"))), + DstAddr: Address(tcpip.Address(net.ParseIP("fe80::dead:beef"))), + }, + &IPv6DestinationOptionsExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6FragmentExtHdrIdentifier), + Options: []byte{0x05, 0x02, 0x00, 0x00, 0x01, 0x00}, + }, + &IPv6FragmentExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6NoNextHeaderIdentifier), + FragmentOffset: Uint16(100), + MoreFragments: Bool(true), + Identification: Uint32(42), + }, + &Payload{ + Bytes: []byte("Sample Data"), + }, + }, + }, + { + description: "IPv6/Fragment/Payload", + wantBytes: []byte{ + // IPv6 Header + 0x60, 0x00, 0x00, 0x00, 0x00, 0x13, 0x2c, 0x40, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x01, 0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xde, 0xad, 0xbe, 0xef, + // Fragment ExtHdr + 0x3b, 0x00, 0x03, 0x21, 0x00, 0x00, 0x00, 0x2a, + // Sample Data + 0x53, 0x61, 0x6d, 0x70, 0x6c, 0x65, 0x20, 0x44, 0x61, 0x74, 0x61, + }, + wantLayers: []Layer{ + &IPv6{ + SrcAddr: Address(tcpip.Address(net.ParseIP("::1"))), + DstAddr: Address(tcpip.Address(net.ParseIP("fe80::dead:beef"))), + }, + &IPv6FragmentExtHdr{ + NextHeader: IPv6ExtHdrIdent(header.IPv6NoNextHeaderIdentifier), + FragmentOffset: Uint16(100), + MoreFragments: Bool(true), + Identification: Uint32(42), + }, + &Payload{ + Bytes: []byte("Sample Data"), }, }, }, @@ -606,6 +703,19 @@ func TestIPv6ExtHdrOptions(t *testing.T) { if !layers.match(tt.wantLayers) { t.Fatalf("match failed with diff: %s", layers.diff(tt.wantLayers)) } + // Make sure we can generate correct next header values and checksums + for _, layer := range layers { + switch layer := layer.(type) { + case *IPv6HopByHopOptionsExtHdr: + layer.NextHeader = nil + case *IPv6DestinationOptionsExtHdr: + layer.NextHeader = nil + case *IPv6FragmentExtHdr: + layer.NextHeader = nil + case *ICMPv6: + layer.Checksum = nil + } + } gotBytes, err := layers.ToBytes() if err != nil { t.Fatalf("ToBytes() failed on %s: %s", &layers, err) diff --git a/test/packetimpact/testbench/testbench.go b/test/packetimpact/testbench/testbench.go index d64f32a5b..242464e3a 100644 --- a/test/packetimpact/testbench/testbench.go +++ b/test/packetimpact/testbench/testbench.go @@ -31,23 +31,37 @@ var ( DUTType = "" // Device is the local device on the test network. Device = "" + // LocalIPv4 is the local IPv4 address on the test network. LocalIPv4 = "" + // RemoteIPv4 is the DUT's IPv4 address on the test network. + RemoteIPv4 = "" + // IPv4PrefixLength is the network prefix length of the IPv4 test network. + IPv4PrefixLength = 0 + // LocalIPv6 is the local IPv6 address on the test network. LocalIPv6 = "" + // RemoteIPv6 is the DUT's IPv6 address on the test network. + RemoteIPv6 = "" + + // LocalInterfaceID is the ID of the local interface on the test network. + LocalInterfaceID uint32 + // RemoteInterfaceID is the ID of the remote interface on the test network. + // + // Not using uint32 because package flag does not support uint32. + RemoteInterfaceID uint64 + // LocalMAC is the local MAC address on the test network. LocalMAC = "" + // RemoteMAC is the DUT's MAC address on the test network. + RemoteMAC = "" + // POSIXServerIP is the POSIX server's IP address on the control network. POSIXServerIP = "" // POSIXServerPort is the UDP port the POSIX server is bound to on the // control network. POSIXServerPort = 40000 - // RemoteIPv4 is the DUT's IPv4 address on the test network. - RemoteIPv4 = "" - // RemoteIPv6 is the DUT's IPv6 address on the test network. - RemoteIPv6 = "" - // RemoteMAC is the DUT's MAC address on the test network. - RemoteMAC = "" + // RPCKeepalive is the gRPC keepalive. RPCKeepalive = 10 * time.Second // RPCTimeout is the gRPC timeout. @@ -68,6 +82,7 @@ func RegisterFlags(fs *flag.FlagSet) { fs.StringVar(&RemoteMAC, "remote_mac", RemoteMAC, "remote mac address for test packets") fs.StringVar(&Device, "device", Device, "local device for test packets") fs.StringVar(&DUTType, "dut_type", DUTType, "type of device under test") + fs.Uint64Var(&RemoteInterfaceID, "remote_interface_id", RemoteInterfaceID, "remote interface ID for test packets") } // genPseudoFlags populates flag-like global config based on real flags. @@ -90,6 +105,13 @@ func genPseudoFlags() error { LocalMAC = deviceInfo.MAC.String() LocalIPv6 = deviceInfo.IPv6Addr.String() + LocalInterfaceID = deviceInfo.ID + + if deviceInfo.IPv4Net != nil { + IPv4PrefixLength, _ = deviceInfo.IPv4Net.Mask.Size() + } else { + IPv4PrefixLength, _ = net.ParseIP(LocalIPv4).DefaultMask().Size() + } return nil } diff --git a/test/packetimpact/tests/BUILD b/test/packetimpact/tests/BUILD index 85749c559..27905dcff 100644 --- a/test/packetimpact/tests/BUILD +++ b/test/packetimpact/tests/BUILD @@ -18,8 +18,6 @@ packetimpact_go_test( packetimpact_go_test( name = "ipv4_id_uniqueness", srcs = ["ipv4_id_uniqueness_test.go"], - # TODO(b/157506701) Fix netstack then remove the line below. - expect_netstack_failure = True, deps = [ "//pkg/abi/linux", "//pkg/tcpip/header", @@ -29,8 +27,19 @@ packetimpact_go_test( ) packetimpact_go_test( - name = "udp_recv_multicast", - srcs = ["udp_recv_multicast_test.go"], + name = "udp_discard_mcast_source_addr", + srcs = ["udp_discard_mcast_source_addr_test.go"], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/header", + "//test/packetimpact/testbench", + "@org_golang_x_sys//unix:go_default_library", + ], +) + +packetimpact_go_test( + name = "udp_recv_mcast_bcast", + srcs = ["udp_recv_mcast_bcast_test.go"], # TODO(b/152813495): Fix netstack then remove the line below. expect_netstack_failure = True, deps = [ @@ -211,6 +220,16 @@ packetimpact_go_test( ) packetimpact_go_test( + name = "tcp_network_unreachable", + srcs = ["tcp_network_unreachable_test.go"], + deps = [ + "//pkg/tcpip/header", + "//test/packetimpact/testbench", + "@org_golang_x_sys//unix:go_default_library", + ], +) + +packetimpact_go_test( name = "tcp_cork_mss", srcs = ["tcp_cork_mss_test.go"], deps = [ @@ -257,6 +276,20 @@ packetimpact_go_test( ) packetimpact_go_test( + name = "ipv6_fragment_reassembly", + srcs = ["ipv6_fragment_reassembly_test.go"], + # TODO(b/160919104): Fix netstack then remove the line below. + expect_netstack_failure = True, + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//test/packetimpact/testbench", + "@org_golang_x_sys//unix:go_default_library", + ], +) + +packetimpact_go_test( name = "udp_send_recv_dgram", srcs = ["udp_send_recv_dgram_test.go"], deps = [ diff --git a/test/packetimpact/tests/icmpv6_param_problem_test.go b/test/packetimpact/tests/icmpv6_param_problem_test.go index 4d1d9a7f5..8dfd26ee8 100644 --- a/test/packetimpact/tests/icmpv6_param_problem_test.go +++ b/test/packetimpact/tests/icmpv6_param_problem_test.go @@ -41,8 +41,8 @@ func TestICMPv6ParamProblemTest(t *testing.T) { NextHeader: testbench.Uint8(254), } icmpv6 := testbench.ICMPv6{ - Type: testbench.ICMPv6Type(header.ICMPv6EchoRequest), - NDPPayload: []byte("hello world"), + Type: testbench.ICMPv6Type(header.ICMPv6EchoRequest), + Payload: []byte("hello world"), } toSend := (*testbench.Connection)(&conn).CreateFrame(testbench.Layers{&ipv6}, &icmpv6) @@ -62,8 +62,8 @@ func TestICMPv6ParamProblemTest(t *testing.T) { binary.BigEndian.PutUint32(b, header.IPv6NextHeaderOffset) expectedPayload = append(b, expectedPayload...) expectedICMPv6 := testbench.ICMPv6{ - Type: testbench.ICMPv6Type(header.ICMPv6ParamProblem), - NDPPayload: expectedPayload, + Type: testbench.ICMPv6Type(header.ICMPv6ParamProblem), + Payload: expectedPayload, } paramProblem := testbench.Layers{ diff --git a/test/packetimpact/tests/ipv6_fragment_reassembly_test.go b/test/packetimpact/tests/ipv6_fragment_reassembly_test.go new file mode 100644 index 000000000..7b462c8e2 --- /dev/null +++ b/test/packetimpact/tests/ipv6_fragment_reassembly_test.go @@ -0,0 +1,168 @@ +// 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 ipv6_fragment_reassembly_test + +import ( + "bytes" + "encoding/binary" + "encoding/hex" + "flag" + "net" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/test/packetimpact/testbench" +) + +const ( + // The payload length for the first fragment we send. This number + // is a multiple of 8 near 750 (half of 1500). + firstPayloadLength = 752 + // The ID field for our outgoing fragments. + fragmentID = 1 + // A node must be able to accept a fragmented packet that, + // after reassembly, is as large as 1500 octets. + reassemblyCap = 1500 +) + +func init() { + testbench.RegisterFlags(flag.CommandLine) +} + +func TestIPv6FragmentReassembly(t *testing.T) { + dut := testbench.NewDUT(t) + defer dut.TearDown() + conn := testbench.NewIPv6Conn(t, testbench.IPv6{}, testbench.IPv6{}) + defer conn.Close() + + firstPayloadToSend := make([]byte, firstPayloadLength) + for i := range firstPayloadToSend { + firstPayloadToSend[i] = 'A' + } + + secondPayloadLength := reassemblyCap - firstPayloadLength - header.ICMPv6EchoMinimumSize + secondPayloadToSend := firstPayloadToSend[:secondPayloadLength] + + icmpv6EchoPayload := make([]byte, 4) + binary.BigEndian.PutUint16(icmpv6EchoPayload[0:], 0) + binary.BigEndian.PutUint16(icmpv6EchoPayload[2:], 0) + icmpv6EchoPayload = append(icmpv6EchoPayload, firstPayloadToSend...) + + lIP := tcpip.Address(net.ParseIP(testbench.LocalIPv6).To16()) + rIP := tcpip.Address(net.ParseIP(testbench.RemoteIPv6).To16()) + icmpv6 := testbench.ICMPv6{ + Type: testbench.ICMPv6Type(header.ICMPv6EchoRequest), + Code: testbench.Byte(0), + Payload: icmpv6EchoPayload, + } + icmpv6Bytes, err := icmpv6.ToBytes() + if err != nil { + t.Fatalf("failed to serialize ICMPv6: %s", err) + } + cksum := header.ICMPv6Checksum( + header.ICMPv6(icmpv6Bytes), + lIP, + rIP, + buffer.NewVectorisedView(len(secondPayloadToSend), []buffer.View{secondPayloadToSend}), + ) + + conn.Send(testbench.IPv6{}, + &testbench.IPv6FragmentExtHdr{ + FragmentOffset: testbench.Uint16(0), + MoreFragments: testbench.Bool(true), + Identification: testbench.Uint32(fragmentID), + }, + &testbench.ICMPv6{ + Type: testbench.ICMPv6Type(header.ICMPv6EchoRequest), + Code: testbench.Byte(0), + Payload: icmpv6EchoPayload, + Checksum: &cksum, + }) + + icmpv6ProtoNum := header.IPv6ExtensionHeaderIdentifier(header.ICMPv6ProtocolNumber) + + conn.Send(testbench.IPv6{}, + &testbench.IPv6FragmentExtHdr{ + NextHeader: &icmpv6ProtoNum, + FragmentOffset: testbench.Uint16((firstPayloadLength + header.ICMPv6EchoMinimumSize) / 8), + MoreFragments: testbench.Bool(false), + Identification: testbench.Uint32(fragmentID), + }, + &testbench.Payload{ + Bytes: secondPayloadToSend, + }) + + gotEchoReplyFirstPart, err := conn.ExpectFrame(testbench.Layers{ + &testbench.Ether{}, + &testbench.IPv6{}, + &testbench.IPv6FragmentExtHdr{ + FragmentOffset: testbench.Uint16(0), + MoreFragments: testbench.Bool(true), + }, + &testbench.ICMPv6{ + Type: testbench.ICMPv6Type(header.ICMPv6EchoReply), + Code: testbench.Byte(0), + }, + }, time.Second) + if err != nil { + t.Fatalf("expected a fragmented ICMPv6 Echo Reply, but got none: %s", err) + } + + id := *gotEchoReplyFirstPart[2].(*testbench.IPv6FragmentExtHdr).Identification + gotFirstPayload, err := gotEchoReplyFirstPart[len(gotEchoReplyFirstPart)-1].ToBytes() + if err != nil { + t.Fatalf("failed to serialize ICMPv6: %s", err) + } + icmpPayload := gotFirstPayload[header.ICMPv6EchoMinimumSize:] + receivedLen := len(icmpPayload) + wantSecondPayloadLen := reassemblyCap - header.ICMPv6EchoMinimumSize - receivedLen + wantFirstPayload := make([]byte, receivedLen) + for i := range wantFirstPayload { + wantFirstPayload[i] = 'A' + } + wantSecondPayload := wantFirstPayload[:wantSecondPayloadLen] + if !bytes.Equal(icmpPayload, wantFirstPayload) { + t.Fatalf("received unexpected payload, got: %s, want: %s", + hex.Dump(icmpPayload), + hex.Dump(wantFirstPayload)) + } + + gotEchoReplySecondPart, err := conn.ExpectFrame(testbench.Layers{ + &testbench.Ether{}, + &testbench.IPv6{}, + &testbench.IPv6FragmentExtHdr{ + NextHeader: &icmpv6ProtoNum, + FragmentOffset: testbench.Uint16(uint16((receivedLen + header.ICMPv6EchoMinimumSize) / 8)), + MoreFragments: testbench.Bool(false), + Identification: &id, + }, + &testbench.ICMPv6{}, + }, time.Second) + if err != nil { + t.Fatalf("expected the rest of ICMPv6 Echo Reply, but got none: %s", err) + } + secondPayload, err := gotEchoReplySecondPart[len(gotEchoReplySecondPart)-1].ToBytes() + if err != nil { + t.Fatalf("failed to serialize ICMPv6 Echo Reply: %s", err) + } + if !bytes.Equal(secondPayload, wantSecondPayload) { + t.Fatalf("received unexpected payload, got: %s, want: %s", + hex.Dump(secondPayload), + hex.Dump(wantSecondPayload)) + } +} diff --git a/test/packetimpact/tests/ipv6_unknown_options_action_test.go b/test/packetimpact/tests/ipv6_unknown_options_action_test.go index d301d8829..100b30ad7 100644 --- a/test/packetimpact/tests/ipv6_unknown_options_action_test.go +++ b/test/packetimpact/tests/ipv6_unknown_options_action_test.go @@ -171,9 +171,9 @@ func TestIPv6UnknownOptionAction(t *testing.T) { &tb.Ether{}, &tb.IPv6{}, &tb.ICMPv6{ - Type: tb.ICMPv6Type(header.ICMPv6ParamProblem), - Code: tb.Byte(2), - NDPPayload: icmpv6Payload, + Type: tb.ICMPv6Type(header.ICMPv6ParamProblem), + Code: tb.Byte(2), + Payload: icmpv6Payload, }, }, time.Second) if tt.wantICMPv6 && err != nil { diff --git a/test/packetimpact/tests/tcp_network_unreachable_test.go b/test/packetimpact/tests/tcp_network_unreachable_test.go new file mode 100644 index 000000000..868a08da8 --- /dev/null +++ b/test/packetimpact/tests/tcp_network_unreachable_test.go @@ -0,0 +1,139 @@ +// 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 tcp_synsent_reset_test + +import ( + "context" + "flag" + "net" + "syscall" + "testing" + "time" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/test/packetimpact/testbench" +) + +func init() { + testbench.RegisterFlags(flag.CommandLine) +} + +// TestTCPSynSentUnreachable verifies that TCP connections fail immediately when +// an ICMP destination unreachable message is sent in response to the inital +// SYN. +func TestTCPSynSentUnreachable(t *testing.T) { + // Create the DUT and connection. + dut := testbench.NewDUT(t) + defer dut.TearDown() + clientFD, clientPort := dut.CreateBoundSocket(unix.SOCK_STREAM|unix.SOCK_NONBLOCK, unix.IPPROTO_TCP, net.ParseIP(testbench.RemoteIPv4)) + port := uint16(9001) + conn := testbench.NewTCPIPv4(t, testbench.TCP{SrcPort: &port, DstPort: &clientPort}, testbench.TCP{SrcPort: &clientPort, DstPort: &port}) + defer conn.Close() + + // Bring the DUT to SYN-SENT state with a non-blocking connect. + ctx, cancel := context.WithTimeout(context.Background(), testbench.RPCTimeout) + defer cancel() + sa := unix.SockaddrInet4{Port: int(port)} + copy(sa.Addr[:], net.IP(net.ParseIP(testbench.LocalIPv4)).To4()) + if _, err := dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EINPROGRESS) { + t.Errorf("expected connect to fail with EINPROGRESS, but got %v", err) + } + + // Get the SYN. + tcpLayers, err := conn.ExpectData(&testbench.TCP{Flags: testbench.Uint8(header.TCPFlagSyn)}, nil, time.Second) + if err != nil { + t.Fatalf("expected SYN: %s", err) + } + + // Send a host unreachable message. + rawConn := (*testbench.Connection)(&conn) + layers := rawConn.CreateFrame(nil) + layers = layers[:len(layers)-1] + const ipLayer = 1 + const tcpLayer = ipLayer + 1 + ip, ok := tcpLayers[ipLayer].(*testbench.IPv4) + if !ok { + t.Fatalf("expected %s to be IPv4", tcpLayers[ipLayer]) + } + tcp, ok := tcpLayers[tcpLayer].(*testbench.TCP) + if !ok { + t.Fatalf("expected %s to be TCP", tcpLayers[tcpLayer]) + } + var icmpv4 testbench.ICMPv4 = testbench.ICMPv4{Type: testbench.ICMPv4Type(header.ICMPv4DstUnreachable), Code: testbench.Uint8(header.ICMPv4HostUnreachable)} + layers = append(layers, &icmpv4, ip, tcp) + rawConn.SendFrameStateless(layers) + + if _, err = dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EHOSTUNREACH) { + t.Errorf("expected connect to fail with EHOSTUNREACH, but got %v", err) + } +} + +// TestTCPSynSentUnreachable6 verifies that TCP connections fail immediately when +// an ICMP destination unreachable message is sent in response to the inital +// SYN. +func TestTCPSynSentUnreachable6(t *testing.T) { + // Create the DUT and connection. + dut := testbench.NewDUT(t) + defer dut.TearDown() + clientFD, clientPort := dut.CreateBoundSocket(unix.SOCK_STREAM|unix.SOCK_NONBLOCK, unix.IPPROTO_TCP, net.ParseIP(testbench.RemoteIPv6)) + conn := testbench.NewTCPIPv6(t, testbench.TCP{DstPort: &clientPort}, testbench.TCP{SrcPort: &clientPort}) + defer conn.Close() + + // Bring the DUT to SYN-SENT state with a non-blocking connect. + ctx, cancel := context.WithTimeout(context.Background(), testbench.RPCTimeout) + defer cancel() + sa := unix.SockaddrInet6{ + Port: int(conn.SrcPort()), + ZoneId: uint32(testbench.RemoteInterfaceID), + } + copy(sa.Addr[:], net.IP(net.ParseIP(testbench.LocalIPv6)).To16()) + if _, err := dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EINPROGRESS) { + t.Errorf("expected connect to fail with EINPROGRESS, but got %v", err) + } + + // Get the SYN. + tcpLayers, err := conn.ExpectData(&testbench.TCP{Flags: testbench.Uint8(header.TCPFlagSyn)}, nil, time.Second) + if err != nil { + t.Fatalf("expected SYN: %s", err) + } + + // Send a host unreachable message. + rawConn := (*testbench.Connection)(&conn) + layers := rawConn.CreateFrame(nil) + layers = layers[:len(layers)-1] + const ipLayer = 1 + const tcpLayer = ipLayer + 1 + ip, ok := tcpLayers[ipLayer].(*testbench.IPv6) + if !ok { + t.Fatalf("expected %s to be IPv6", tcpLayers[ipLayer]) + } + tcp, ok := tcpLayers[tcpLayer].(*testbench.TCP) + if !ok { + t.Fatalf("expected %s to be TCP", tcpLayers[tcpLayer]) + } + var icmpv6 testbench.ICMPv6 = testbench.ICMPv6{ + Type: testbench.ICMPv6Type(header.ICMPv6DstUnreachable), + Code: testbench.Uint8(header.ICMPv6NetworkUnreachable), + // Per RFC 4443 3.1, the payload contains 4 zeroed bytes. + Payload: []byte{0, 0, 0, 0}, + } + layers = append(layers, &icmpv6, ip, tcp) + rawConn.SendFrameStateless(layers) + + if _, err = dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.ENETUNREACH) { + t.Errorf("expected connect to fail with ENETUNREACH, but got %v", err) + } +} diff --git a/test/packetimpact/tests/udp_discard_mcast_source_addr_test.go b/test/packetimpact/tests/udp_discard_mcast_source_addr_test.go new file mode 100644 index 000000000..b0315e67c --- /dev/null +++ b/test/packetimpact/tests/udp_discard_mcast_source_addr_test.go @@ -0,0 +1,92 @@ +// 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 udp_discard_mcast_source_addr_test + +import ( + "context" + "flag" + "fmt" + "net" + "syscall" + "testing" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/test/packetimpact/testbench" +) + +var oneSecond = unix.Timeval{Sec: 1, Usec: 0} + +func init() { + testbench.RegisterFlags(flag.CommandLine) +} + +func TestDiscardsUDPPacketsWithMcastSourceAddressV4(t *testing.T) { + dut := testbench.NewDUT(t) + defer dut.TearDown() + remoteFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP(testbench.RemoteIPv4)) + defer dut.Close(remoteFD) + dut.SetSockOptTimeval(remoteFD, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &oneSecond) + conn := testbench.NewUDPIPv4(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) + defer conn.Close() + + for _, mcastAddr := range []net.IP{ + net.IPv4allsys, + net.IPv4allrouter, + net.IPv4(224, 0, 1, 42), + net.IPv4(232, 1, 2, 3), + } { + t.Run(fmt.Sprintf("srcaddr=%s", mcastAddr), func(t *testing.T) { + conn.SendIP( + testbench.IPv4{SrcAddr: testbench.Address(tcpip.Address(mcastAddr.To4()))}, + testbench.UDP{}, + ) + + ret, payload, errno := dut.RecvWithErrno(context.Background(), remoteFD, 100, 0) + if errno != syscall.EAGAIN || errno != syscall.EWOULDBLOCK { + t.Errorf("Recv got unexpected result, ret=%d, payload=%q, errno=%s", ret, payload, errno) + } + }) + } +} + +func TestDiscardsUDPPacketsWithMcastSourceAddressV6(t *testing.T) { + dut := testbench.NewDUT(t) + defer dut.TearDown() + remoteFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP(testbench.RemoteIPv6)) + defer dut.Close(remoteFD) + dut.SetSockOptTimeval(remoteFD, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &oneSecond) + conn := testbench.NewUDPIPv6(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) + defer conn.Close() + + for _, mcastAddr := range []net.IP{ + net.IPv6interfacelocalallnodes, + net.IPv6linklocalallnodes, + net.IPv6linklocalallrouters, + net.ParseIP("fe01::42"), + net.ParseIP("fe02::4242"), + } { + t.Run(fmt.Sprintf("srcaddr=%s", mcastAddr), func(t *testing.T) { + conn.SendIPv6( + testbench.IPv6{SrcAddr: testbench.Address(tcpip.Address(mcastAddr.To16()))}, + testbench.UDP{}, + ) + ret, payload, errno := dut.RecvWithErrno(context.Background(), remoteFD, 100, 0) + if errno != syscall.EAGAIN || errno != syscall.EWOULDBLOCK { + t.Errorf("Recv got unexpected result, ret=%d, payload=%q, errno=%s", ret, payload, errno) + } + }) + } +} diff --git a/test/packetimpact/tests/udp_recv_multicast_test.go b/test/packetimpact/tests/udp_recv_mcast_bcast_test.go index 77a9bfa1d..263a54291 100644 --- a/test/packetimpact/tests/udp_recv_multicast_test.go +++ b/test/packetimpact/tests/udp_recv_mcast_bcast_test.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package udp_recv_multicast_test +package udp_recv_mcast_bcast_test import ( "flag" @@ -28,13 +28,36 @@ func init() { testbench.RegisterFlags(flag.CommandLine) } -func TestUDPRecvMulticast(t *testing.T) { +func TestUDPRecvMulticastBroadcast(t *testing.T) { dut := testbench.NewDUT(t) defer dut.TearDown() - boundFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP("0.0.0.0")) + boundFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.IPv4(0, 0, 0, 0)) defer dut.Close(boundFD) conn := testbench.NewUDPIPv4(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) defer conn.Close() - conn.SendIP(testbench.IPv4{DstAddr: testbench.Address(tcpip.Address(net.ParseIP("224.0.0.1").To4()))}, testbench.UDP{}) - dut.Recv(boundFD, 100, 0) + + for _, bcastAddr := range []net.IP{ + broadcastAddr(net.ParseIP(testbench.RemoteIPv4), net.CIDRMask(testbench.IPv4PrefixLength, 32)), + net.IPv4(255, 255, 255, 255), + net.IPv4(224, 0, 0, 1), + } { + payload := testbench.GenerateRandomPayload(t, 1<<10) + conn.SendIP( + testbench.IPv4{DstAddr: testbench.Address(tcpip.Address(bcastAddr.To4()))}, + testbench.UDP{}, + &testbench.Payload{Bytes: payload}, + ) + t.Logf("Receiving packet sent to address: %s", bcastAddr) + if got, want := string(dut.Recv(boundFD, int32(len(payload)), 0)), string(payload); got != want { + t.Errorf("received payload does not match sent payload got: %s, want: %s", got, want) + } + } +} + +func broadcastAddr(ip net.IP, mask net.IPMask) net.IP { + ip4 := ip.To4() + for i := range ip4 { + ip4[i] |= ^mask[i] + } + return ip4 } diff --git a/test/packetimpact/tests/udp_send_recv_dgram_test.go b/test/packetimpact/tests/udp_send_recv_dgram_test.go index 224feef85..bd53ad90b 100644 --- a/test/packetimpact/tests/udp_send_recv_dgram_test.go +++ b/test/packetimpact/tests/udp_send_recv_dgram_test.go @@ -28,62 +28,74 @@ func init() { testbench.RegisterFlags(flag.CommandLine) } -func TestUDPRecv(t *testing.T) { +type udpConn interface { + Send(testbench.UDP, ...testbench.Layer) + ExpectData(testbench.UDP, testbench.Payload, time.Duration) (testbench.Layers, error) + Drain() + Close() +} + +func TestUDP(t *testing.T) { dut := testbench.NewDUT(t) defer dut.TearDown() - boundFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP("0.0.0.0")) - defer dut.Close(boundFD) - conn := testbench.NewUDPIPv4(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) - defer conn.Close() - testCases := []struct { - name string - payload []byte - }{ - {"emptypayload", nil}, - {"small payload", []byte("hello world")}, - {"1kPayload", testbench.GenerateRandomPayload(t, 1<<10)}, - // Even though UDP allows larger dgrams we don't test it here as - // they need to be fragmented and written out as individual - // frames. - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - conn.Send(testbench.UDP{}, &testbench.Payload{Bytes: tc.payload}) - if got, want := string(dut.Recv(boundFD, int32(len(tc.payload)), 0)), string(tc.payload); got != want { - t.Fatalf("received payload does not match sent payload got: %s, want: %s", got, want) + for _, isIPv4 := range []bool{true, false} { + ipVersionName := "IPv6" + if isIPv4 { + ipVersionName = "IPv4" + } + t.Run(ipVersionName, func(t *testing.T) { + var addr string + if isIPv4 { + addr = testbench.RemoteIPv4 + } else { + addr = testbench.RemoteIPv6 } - }) - } -} + boundFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP(addr)) + defer dut.Close(boundFD) -func TestUDPSend(t *testing.T) { - dut := testbench.NewDUT(t) - defer dut.TearDown() - boundFD, remotePort := dut.CreateBoundSocket(unix.SOCK_DGRAM, unix.IPPROTO_UDP, net.ParseIP("0.0.0.0")) - defer dut.Close(boundFD) - conn := testbench.NewUDPIPv4(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) - defer conn.Close() + var conn udpConn + var localAddr unix.Sockaddr + if isIPv4 { + v4Conn := testbench.NewUDPIPv4(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) + localAddr = v4Conn.LocalAddr() + conn = &v4Conn + } else { + v6Conn := testbench.NewUDPIPv6(t, testbench.UDP{DstPort: &remotePort}, testbench.UDP{SrcPort: &remotePort}) + localAddr = v6Conn.LocalAddr() + conn = &v6Conn + } + defer conn.Close() - testCases := []struct { - name string - payload []byte - }{ - {"emptypayload", nil}, - {"small payload", []byte("hello world")}, - {"1kPayload", testbench.GenerateRandomPayload(t, 1<<10)}, - // Even though UDP allows larger dgrams we don't test it here as - // they need to be fragmented and written out as individual - // frames. - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - conn.Drain() - if got, want := int(dut.SendTo(boundFD, tc.payload, 0, conn.LocalAddr())), len(tc.payload); got != want { - t.Fatalf("short write got: %d, want: %d", got, want) + testCases := []struct { + name string + payload []byte + }{ + {"emptypayload", nil}, + {"small payload", []byte("hello world")}, + {"1kPayload", testbench.GenerateRandomPayload(t, 1<<10)}, + // Even though UDP allows larger dgrams we don't test it here as + // they need to be fragmented and written out as individual + // frames. } - if _, err := conn.ExpectData(testbench.UDP{SrcPort: &remotePort}, testbench.Payload{Bytes: tc.payload}, 1*time.Second); err != nil { - t.Fatal(err) + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + t.Run("Send", func(t *testing.T) { + conn.Send(testbench.UDP{}, &testbench.Payload{Bytes: tc.payload}) + if got, want := string(dut.Recv(boundFD, int32(len(tc.payload)), 0)), string(tc.payload); got != want { + t.Fatalf("received payload does not match sent payload got: %s, want: %s", got, want) + } + }) + t.Run("Recv", func(t *testing.T) { + conn.Drain() + if got, want := int(dut.SendTo(boundFD, tc.payload, 0, localAddr)), len(tc.payload); got != want { + t.Fatalf("short write got: %d, want: %d", got, want) + } + if _, err := conn.ExpectData(testbench.UDP{SrcPort: &remotePort}, testbench.Payload{Bytes: tc.payload}, time.Second); err != nil { + t.Fatal(err) + } + }) + }) } }) } diff --git a/test/root/BUILD b/test/root/BUILD index a9e91ccd6..a9130b34f 100644 --- a/test/root/BUILD +++ b/test/root/BUILD @@ -41,7 +41,7 @@ go_test( "//runsc/container", "//runsc/specutils", "@com_github_cenkalti_backoff//:go_default_library", - "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", + "@com_github_opencontainers_runtime_spec//specs-go:go_default_library", "@com_github_syndtr_gocapability//capability:go_default_library", "@org_golang_x_sys//unix:go_default_library", ], @@ -51,8 +51,5 @@ vm_test( name = "root_vm_test", size = "large", shard_count = 1, - targets = [ - "//tools/installers:shim", - ":root_test", - ], + targets = [":root_test"], ) diff --git a/test/root/cgroup_test.go b/test/root/cgroup_test.go index d0634b5c3..a26b83081 100644 --- a/test/root/cgroup_test.go +++ b/test/root/cgroup_test.go @@ -16,6 +16,7 @@ package root import ( "bufio" + "context" "fmt" "io/ioutil" "os" @@ -56,25 +57,24 @@ func verifyPid(pid int, path string) error { return fmt.Errorf("got: %v, want: %d", gots, pid) } -func TestMemCGroup(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() +func TestMemCgroup(t *testing.T) { + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Start a new container and allocate the specified about of memory. 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 { + + if err := d.Spawn(ctx, dockerutil.RunOpts{ + Image: "basic/ubuntu", + Memory: allocMemLimit, // Must be in bytes. + }, "python3", "-c", fmt.Sprintf("import time; s = 'a' * %d; time.sleep(100)", allocMemSize)); err != nil { t.Fatalf("docker run failed: %v", err) } // Extract the ID to lookup the cgroup. - gid, err := d.ID() - if err != nil { - t.Fatalf("Docker.ID() failed: %v", err) - } + gid := d.ID() t.Logf("cgroup ID: %s", gid) // Wait when the container will allocate memory. @@ -127,8 +127,9 @@ func TestMemCGroup(t *testing.T) { // TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestCgroup(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // This is not a comprehensive list of attributes. // @@ -137,94 +138,133 @@ func TestCgroup(t *testing.T) { // are often run on a single core virtual machine, and there is only a single // CPU available in our current set, and every container's set. attrs := []struct { - arg string + field string + value int64 ctrl string file string want string skipIfNotFound bool }{ { - arg: "--cpu-shares=1000", - ctrl: "cpu", - file: "cpu.shares", - want: "1000", + field: "cpu-shares", + value: 1000, + ctrl: "cpu", + file: "cpu.shares", + want: "1000", }, { - arg: "--cpu-period=2000", - ctrl: "cpu", - file: "cpu.cfs_period_us", - want: "2000", + field: "cpu-period", + value: 2000, + ctrl: "cpu", + file: "cpu.cfs_period_us", + want: "2000", }, { - arg: "--cpu-quota=3000", - ctrl: "cpu", - file: "cpu.cfs_quota_us", - want: "3000", + field: "cpu-quota", + value: 3000, + ctrl: "cpu", + file: "cpu.cfs_quota_us", + want: "3000", }, { - arg: "--kernel-memory=100MB", - ctrl: "memory", - file: "memory.kmem.limit_in_bytes", - want: "104857600", + field: "kernel-memory", + value: 100 << 20, + ctrl: "memory", + file: "memory.kmem.limit_in_bytes", + want: "104857600", }, { - arg: "--memory=1GB", - ctrl: "memory", - file: "memory.limit_in_bytes", - want: "1073741824", + field: "memory", + value: 1 << 30, + ctrl: "memory", + file: "memory.limit_in_bytes", + want: "1073741824", }, { - arg: "--memory-reservation=500MB", - ctrl: "memory", - file: "memory.soft_limit_in_bytes", - want: "524288000", + field: "memory-reservation", + value: 500 << 20, + ctrl: "memory", + file: "memory.soft_limit_in_bytes", + want: "524288000", }, { - arg: "--memory-swap=2GB", + field: "memory-swap", + value: 2 << 30, ctrl: "memory", file: "memory.memsw.limit_in_bytes", want: "2147483648", skipIfNotFound: true, // swap may be disabled on the machine. }, { - arg: "--memory-swappiness=5", - ctrl: "memory", - file: "memory.swappiness", - want: "5", + field: "memory-swappiness", + value: 5, + ctrl: "memory", + file: "memory.swappiness", + want: "5", }, { - arg: "--blkio-weight=750", + field: "blkio-weight", + value: 750, ctrl: "blkio", file: "blkio.weight", want: "750", skipIfNotFound: true, // blkio groups may not be available. }, { - arg: "--pids-limit=1000", - ctrl: "pids", - file: "pids.max", - want: "1000", + field: "pids-limit", + value: 1000, + ctrl: "pids", + file: "pids.max", + want: "1000", }, } - args := make([]string, 0, len(attrs)) + // Make configs. + conf, hostconf, _ := d.ConfigsFrom(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "10000") + + // Add Cgroup arguments to configs. for _, attr := range attrs { - args = append(args, attr.arg) + switch attr.field { + case "cpu-shares": + hostconf.Resources.CPUShares = attr.value + case "cpu-period": + hostconf.Resources.CPUPeriod = attr.value + case "cpu-quota": + hostconf.Resources.CPUQuota = attr.value + case "kernel-memory": + hostconf.Resources.KernelMemory = attr.value + case "memory": + hostconf.Resources.Memory = attr.value + case "memory-reservation": + hostconf.Resources.MemoryReservation = attr.value + case "memory-swap": + hostconf.Resources.MemorySwap = attr.value + case "memory-swappiness": + val := attr.value + hostconf.Resources.MemorySwappiness = &val + case "blkio-weight": + hostconf.Resources.BlkioWeight = uint16(attr.value) + case "pids-limit": + val := attr.value + hostconf.Resources.PidsLimit = &val + + } } - // 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) + // Create container. + if err := d.CreateFrom(ctx, conf, hostconf, nil); err != nil { + t.Fatalf("create failed with: %v", err) } - // Lookup the relevant cgroup ID. - gid, err := d.ID() - if err != nil { - t.Fatalf("Docker.ID() failed: %v", err) + // Start container. + if err := d.Start(ctx); err != nil { + t.Fatalf("start failed with: %v", err) } + + // Lookup the relevant cgroup ID. + gid := d.ID() t.Logf("cgroup ID: %s", gid) // Check list of attributes defined above. @@ -239,7 +279,7 @@ func TestCgroup(t *testing.T) { t.Fatalf("failed to read %q: %v", path, err) } if got := strings.TrimSpace(string(out)); got != attr.want { - t.Errorf("arg: %q, cgroup attribute %s/%s, got: %q, want: %q", attr.arg, attr.ctrl, attr.file, got, attr.want) + t.Errorf("field: %q, cgroup attribute %s/%s, got: %q, want: %q", attr.field, attr.ctrl, attr.file, got, attr.want) } } @@ -257,7 +297,7 @@ func TestCgroup(t *testing.T) { "pids", "systemd", } - pid, err := d.SandboxPid() + pid, err := d.SandboxPid(ctx) if err != nil { t.Fatalf("SandboxPid: %v", err) } @@ -269,29 +309,34 @@ func TestCgroup(t *testing.T) { } } -// TestCgroup sets cgroup options and checks that cgroup was properly configured. +// TestCgroupParent sets the "CgroupParent" option and checks that the child and parent's +// cgroups are created correctly relative to each other. func TestCgroupParent(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) // Construct a known cgroup name. parent := testutil.RandomID("runsc-") - if err := d.Spawn(dockerutil.RunOpts{ + conf, hostconf, _ := d.ConfigsFrom(dockerutil.RunOpts{ Image: "basic/alpine", - Extra: []string{fmt.Sprintf("--cgroup-parent=%s", parent)}, - }, "sleep", "10000"); err != nil { - t.Fatalf("docker run failed: %v", err) + }, "sleep", "10000") + hostconf.Resources.CgroupParent = parent + + if err := d.CreateFrom(ctx, conf, hostconf, nil); err != nil { + t.Fatalf("create failed with: %v", err) } - // Extract the ID to look up the cgroup. - gid, err := d.ID() - if err != nil { - t.Fatalf("Docker.ID() failed: %v", err) + if err := d.Start(ctx); err != nil { + t.Fatalf("start failed with: %v", err) } + + // Extract the ID to look up the cgroup. + gid := d.ID() t.Logf("cgroup ID: %s", gid) // Check that sandbox is inside cgroup. - pid, err := d.SandboxPid() + pid, err := d.SandboxPid(ctx) if err != nil { t.Fatalf("SandboxPid: %v", err) } diff --git a/test/root/chroot_test.go b/test/root/chroot_test.go index a306132a4..58fcd6f08 100644 --- a/test/root/chroot_test.go +++ b/test/root/chroot_test.go @@ -16,6 +16,7 @@ package root import ( + "context" "fmt" "io/ioutil" "os/exec" @@ -30,16 +31,17 @@ import ( // 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(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) } - pid, err := d.SandboxPid() + pid, err := d.SandboxPid(ctx) if err != nil { t.Fatalf("Docker.SandboxPid(): %v", err) } @@ -75,14 +77,15 @@ func TestChroot(t *testing.T) { t.Errorf("chroot got children %v, want %v", fi[0].Name(), "proc") } - d.CleanUp() + d.CleanUp(ctx) } func TestChrootGofer(t *testing.T) { - d := dockerutil.MakeDocker(t) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) - if err := d.Spawn(dockerutil.RunOpts{ + if err := d.Spawn(ctx, dockerutil.RunOpts{ Image: "basic/alpine", }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) @@ -91,7 +94,7 @@ func TestChrootGofer(t *testing.T) { // 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 // are gofers. - sandPID, err := d.SandboxPid() + sandPID, err := d.SandboxPid(ctx) if err != nil { t.Fatalf("Docker.SandboxPid(): %v", err) } diff --git a/test/root/crictl_test.go b/test/root/crictl_test.go index 732fae821..df91fa0fe 100644 --- a/test/root/crictl_test.go +++ b/test/root/crictl_test.go @@ -20,13 +20,14 @@ import ( "fmt" "io" "io/ioutil" - "log" "net/http" "os" "os/exec" "path" - "path/filepath" + "regexp" + "strconv" "strings" + "sync" "testing" "time" @@ -75,6 +76,8 @@ func SimpleSpec(name, image string, cmd []string, extra map[string]interface{}) // Log files are not deleted after root tests are run. Log to random // paths to ensure logs are fresh. "log_path": fmt.Sprintf("%s.log", testutil.RandomID(name)), + "stdin": false, + "tty": false, } if len(cmd) > 0 { // Omit if empty. s["command"] = cmd @@ -95,25 +98,29 @@ var Httpd = SimpleSpec("httpd", "basic/httpd", nil, nil) // TestCrictlSanity refers to b/112433158. func TestCrictlSanity(t *testing.T) { - // Setup containerd and crictl. - crictl, cleanup, err := setup(t) - if err != nil { - t.Fatalf("failed to setup crictl: %v", err) - } - defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox("default"), Httpd) - if err != nil { - t.Fatalf("start failed: %v", err) - } - - // Look for the httpd page. - if err = httpGet(crictl, podID, "index.html"); err != nil { - t.Fatalf("failed to get page: %v", err) - } - - // Stop everything. - if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatalf("stop failed: %v", err) + for _, version := range allVersions { + t.Run(version, func(t *testing.T) { + // Setup containerd and crictl. + crictl, cleanup, err := setup(t, version) + if err != nil { + t.Fatalf("failed to setup crictl: %v", err) + } + defer cleanup() + podID, contID, err := crictl.StartPodAndContainer(containerdRuntime, "basic/httpd", Sandbox("default"), Httpd) + if err != nil { + t.Fatalf("start failed: %v", err) + } + + // Look for the httpd page. + if err = httpGet(crictl, podID, "index.html"); err != nil { + t.Fatalf("failed to get page: %v", err) + } + + // Stop everything. + if err := crictl.StopPodAndContainer(podID, contID); err != nil { + t.Fatalf("stop failed: %v", err) + } + }) } } @@ -147,146 +154,179 @@ var HttpdMountPaths = SimpleSpec("httpd", "basic/httpd", nil, map[string]interfa // TestMountPaths refers to b/117635704. func TestMountPaths(t *testing.T) { - // Setup containerd and crictl. - crictl, cleanup, err := setup(t) - if err != nil { - t.Fatalf("failed to setup crictl: %v", err) - } - defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox("default"), HttpdMountPaths) - if err != nil { - t.Fatalf("start failed: %v", err) - } - - // Look for the directory available at /test. - if err = httpGet(crictl, podID, "test"); err != nil { - t.Fatalf("failed to get page: %v", err) - } - - // Stop everything. - if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatalf("stop failed: %v", err) + for _, version := range allVersions { + t.Run(version, func(t *testing.T) { + // Setup containerd and crictl. + crictl, cleanup, err := setup(t, version) + if err != nil { + t.Fatalf("failed to setup crictl: %v", err) + } + defer cleanup() + podID, contID, err := crictl.StartPodAndContainer(containerdRuntime, "basic/httpd", Sandbox("default"), HttpdMountPaths) + if err != nil { + t.Fatalf("start failed: %v", err) + } + + // Look for the directory available at /test. + if err = httpGet(crictl, podID, "test"); err != nil { + t.Fatalf("failed to get page: %v", err) + } + + // Stop everything. + if err := crictl.StopPodAndContainer(podID, contID); err != nil { + t.Fatalf("stop failed: %v", err) + } + }) } } // TestMountPaths refers to b/118728671. func TestMountOverSymlinks(t *testing.T) { - // Setup containerd and crictl. - crictl, cleanup, err := setup(t) - if err != nil { - t.Fatalf("failed to setup crictl: %v", err) - } - defer cleanup() - - spec := SimpleSpec("busybox", "basic/resolv", []string{"sleep", "1000"}, nil) - podID, contID, err := crictl.StartPodAndContainer("basic/resolv", Sandbox("default"), spec) - if err != nil { - t.Fatalf("start failed: %v", err) - } - - out, err := crictl.Exec(contID, "readlink", "/etc/resolv.conf") - if err != nil { - 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)) - } - - etc, err := crictl.Exec(contID, "cat", "/etc/resolv.conf") - if err != nil { - t.Fatalf("cat failed: %v, out: %s", err, etc) - } - tmp, err := crictl.Exec(contID, "cat", "/tmp/resolv.conf") - if err != nil { - 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)) - } - - // Stop everything. - if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatalf("stop failed: %v", err) + for _, version := range allVersions { + t.Run(version, func(t *testing.T) { + // Setup containerd and crictl. + crictl, cleanup, err := setup(t, version) + if err != nil { + t.Fatalf("failed to setup crictl: %v", err) + } + defer cleanup() + + spec := SimpleSpec("busybox", "basic/resolv", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer(containerdRuntime, "basic/resolv", Sandbox("default"), spec) + if err != nil { + t.Fatalf("start failed: %v", err) + } + + out, err := crictl.Exec(contID, "readlink", "/etc/resolv.conf") + if err != nil { + 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)) + } + + etc, err := crictl.Exec(contID, "cat", "/etc/resolv.conf") + if err != nil { + t.Fatalf("cat failed: %v, out: %s", err, etc) + } + tmp, err := crictl.Exec(contID, "cat", "/tmp/resolv.conf") + if err != nil { + 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)) + } + + // Stop everything. + if err := crictl.StopPodAndContainer(podID, contID); err != nil { + t.Fatalf("stop failed: %v", err) + } + }) } } // TestHomeDir tests that the HOME environment variable is set for // Pod containers. func TestHomeDir(t *testing.T) { - // Setup containerd and crictl. - crictl, cleanup, err := setup(t) - if err != nil { - t.Fatalf("failed to setup crictl: %v", err) + for _, version := range allVersions { + t.Run(version, func(t *testing.T) { + // Setup containerd and crictl. + crictl, cleanup, err := setup(t, version) + if err != nil { + t.Fatalf("failed to setup crictl: %v", err) + } + defer cleanup() + + // Note that container ID returned here is a sub-container. All Pod + // containers are sub-containers. The root container of the sandbox is the + // pause container. + t.Run("sub-container", func(t *testing.T) { + contSpec := SimpleSpec("subcontainer", "basic/busybox", []string{"sh", "-c", "echo $HOME"}, nil) + podID, contID, err := crictl.StartPodAndContainer(containerdRuntime, "basic/busybox", Sandbox("subcont-sandbox"), contSpec) + if err != nil { + t.Fatalf("start failed: %v", err) + } + + out, err := crictl.Logs(contID) + if err != nil { + t.Fatalf("failed retrieving container logs: %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) + } + + // Stop everything; note that the pod may have already stopped. + crictl.StopPodAndContainer(podID, contID) + }) + + // Tests that HOME is set for the exec process. + t.Run("exec", func(t *testing.T) { + contSpec := SimpleSpec("exec", "basic/busybox", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer(containerdRuntime, "basic/busybox", Sandbox("exec-sandbox"), contSpec) + if err != nil { + t.Fatalf("start failed: %v", err) + } + + out, err := crictl.Exec(contID, "sh", "-c", "echo $HOME") + if err != nil { + t.Fatalf("failed retrieving container logs: %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) + } + + // Stop everything. + if err := crictl.StopPodAndContainer(podID, contID); err != nil { + t.Fatalf("stop failed: %v", err) + } + }) + }) } - defer cleanup() - - // Note that container ID returned here is a sub-container. All Pod - // containers are sub-containers. The root container of the sandbox is the - // pause container. - t.Run("sub-container", func(t *testing.T) { - contSpec := SimpleSpec("subcontainer", "basic/busybox", []string{"sh", "-c", "echo $HOME"}, nil) - podID, contID, err := crictl.StartPodAndContainer("basic/busybox", Sandbox("subcont-sandbox"), contSpec) - if err != nil { - t.Fatalf("start failed: %v", err) - } - - out, err := crictl.Logs(contID) - if err != nil { - t.Fatalf("failed retrieving container logs: %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) - } - - // Stop everything. - if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatalf("stop failed: %v", err) - } - }) - - // Tests that HOME is set for the exec process. - t.Run("exec", func(t *testing.T) { - contSpec := SimpleSpec("exec", "basic/busybox", []string{"sleep", "1000"}, nil) - podID, contID, err := crictl.StartPodAndContainer("basic/busybox", Sandbox("exec-sandbox"), contSpec) - if err != nil { - t.Fatalf("start failed: %v", err) - } - - out, err := crictl.Exec(contID, "sh", "-c", "echo $HOME") - if err != nil { - t.Fatalf("failed retrieving container logs: %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) - } - - // Stop everything. - if err := crictl.StopPodAndContainer(podID, contID); err != nil { - 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 = ` +const containerdRuntime = "runsc" + +const v1Template = ` disabled_plugins = ["restart"] +[plugins.cri] + disable_tcp_service = true [plugins.linux] - runtime = "%s" - runtime_root = "/tmp/test-containerd/runsc" - shim = "/usr/local/bin/gvisor-containerd-shim" + shim = "%s" shim_debug = true - -[plugins.cri.containerd.runtimes.runsc] +[plugins.cri.containerd.runtimes.` + containerdRuntime + `] runtime_type = "io.containerd.runtime.v1.linux" runtime_engine = "%s" + runtime_root = "%s/root/runsc" ` +const v2Template = ` +disabled_plugins = ["restart"] +[plugins.cri] + disable_tcp_service = true +[plugins.linux] + shim_debug = true +[plugins.cri.containerd.runtimes.` + containerdRuntime + `] + runtime_type = "io.containerd.` + containerdRuntime + `.v1" +[plugins.cri.containerd.runtimes.` + containerdRuntime + `.options] + TypeUrl = "io.containerd.` + containerdRuntime + `.v1.options" +` + +const ( + // v1 is the containerd API v1. + v1 string = "v1" + + // v1 is the containerd API v21. + v2 string = "v2" +) + +// allVersions is the set of known versions. +var allVersions = []string{v1, v2} + // 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. // * Returns a cleanup function that should be called at the end of the test. -func setup(t *testing.T) (*criutil.Crictl, func(), error) { +func setup(t *testing.T, version string) (*criutil.Crictl, func(), error) { // Create temporary containerd root and state directories, and a socket // via which crictl and containerd communicate. containerdRoot, err := ioutil.TempDir(testutil.TmpDir(), "containerd-root") @@ -295,13 +335,43 @@ func setup(t *testing.T) (*criutil.Crictl, func(), error) { } cu := cleanup.Make(func() { os.RemoveAll(containerdRoot) }) defer cu.Clean() + t.Logf("Using containerd root: %s", containerdRoot) containerdState, err := ioutil.TempDir(testutil.TmpDir(), "containerd-state") if err != nil { t.Fatalf("failed to create containerd state: %v", err) } cu.Add(func() { os.RemoveAll(containerdState) }) - sockAddr := filepath.Join(testutil.TmpDir(), "containerd-test.sock") + t.Logf("Using containerd state: %s", containerdState) + + sockDir, err := ioutil.TempDir(testutil.TmpDir(), "containerd-sock") + if err != nil { + t.Fatalf("failed to create containerd socket directory: %v", err) + } + cu.Add(func() { os.RemoveAll(sockDir) }) + sockAddr := path.Join(sockDir, "test.sock") + t.Logf("Using containerd socket: %s", sockAddr) + + // Extract the containerd version. + versionCmd := exec.Command(getContainerd(), "-v") + out, err := versionCmd.CombinedOutput() + if err != nil { + t.Fatalf("error extracting containerd version: %v (%s)", err, string(out)) + } + r := regexp.MustCompile(" v([0-9]+)\\.([0-9]+)\\.([0-9+])") + vs := r.FindStringSubmatch(string(out)) + if len(vs) != 4 { + t.Fatalf("error unexpected version string: %s", string(out)) + } + major, err := strconv.ParseUint(vs[1], 10, 64) + if err != nil { + t.Fatalf("error parsing containerd major version: %v (%s)", err, string(out)) + } + minor, err := strconv.ParseUint(vs[2], 10, 64) + if err != nil { + t.Fatalf("error parsing containerd minor version: %v (%s)", err, string(out)) + } + t.Logf("Using containerd version: %d.%d", major, minor) // We rewrite a configuration. This is based on the current docker // configuration for the runtime under test. @@ -309,28 +379,97 @@ func setup(t *testing.T) (*criutil.Crictl, func(), error) { if err != nil { t.Fatalf("error discovering runtime path: %v", err) } - config, configCleanup, err := testutil.WriteTmpFile("containerd-config", fmt.Sprintf(containerdConfigTemplate, runtime, runtime)) + t.Logf("Using runtime: %v", runtime) + + // Construct a PATH that includes the runtime directory. This is + // because the shims will be installed there, and containerd may infer + // the binary name and search the PATH. + runtimeDir := path.Dir(runtime) + modifiedPath := os.Getenv("PATH") + if modifiedPath != "" { + modifiedPath = ":" + modifiedPath // We prepend below. + } + modifiedPath = path.Dir(getContainerd()) + modifiedPath + modifiedPath = runtimeDir + ":" + modifiedPath + t.Logf("Using PATH: %v", modifiedPath) + + var ( + config string + runpArgs []string + ) + switch version { + case v1: + // This is only supported less than 1.3. + if major > 1 || (major == 1 && minor >= 3) { + t.Skipf("skipping unsupported containerd (want less than 1.3, got %d.%d)", major, minor) + } + + // We provide the shim, followed by the runtime, and then a + // temporary root directory. + config = fmt.Sprintf(v1Template, criutil.ResolvePath("gvisor-containerd-shim"), runtime, containerdRoot) + case v2: + // This is only supported past 1.2. + if major < 1 || (major == 1 && minor <= 1) { + t.Skipf("skipping incompatible containerd (want at least 1.2, got %d.%d)", major, minor) + } + + // The runtime is provided via parameter. Note that the v2 shim + // binary name is always containerd-shim-* so we don't actually + // care about the docker runtime name. + config = v2Template + default: + t.Fatalf("unknown version: %d", version) + } + t.Logf("Using config: %s", config) + + // Generate the configuration for the test. + configFile, configCleanup, err := testutil.WriteTmpFile("containerd-config", config) if err != nil { t.Fatalf("failed to write containerd config") } cu.Add(configCleanup) // Start containerd. - cmd := exec.Command(getContainerd(), - "--config", config, + args := []string{ + getContainerd(), + "--config", configFile, "--log-level", "debug", "--root", containerdRoot, "--state", containerdState, - "--address", sockAddr) + "--address", sockAddr, + } + t.Logf("Using args: %s", strings.Join(args, " ")) + cmd := exec.Command(args[0], args[1:]...) + cmd.Env = append(os.Environ(), "PATH="+modifiedPath) + + // Include output in logs. + stderrPipe, err := cmd.StderrPipe() + if err != nil { + t.Fatalf("failed to create stderr pipe: %v", err) + } + cu.Add(func() { stderrPipe.Close() }) + stdoutPipe, err := cmd.StdoutPipe() + if err != nil { + t.Fatalf("failed to create stdout pipe: %v", err) + } + cu.Add(func() { stdoutPipe.Close() }) + var ( + wg sync.WaitGroup + stderr bytes.Buffer + stdout bytes.Buffer + ) 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) + wg.Add(2) + go func() { + defer wg.Done() + io.Copy(io.MultiWriter(startupW, &stderr), stderrPipe) + }() + go func() { + defer wg.Done() + io.Copy(io.MultiWriter(startupW, &stdout), stdoutPipe) + }() cu.Add(func() { - // Log output in case of failure. + wg.Wait() t.Logf("containerd stdout: %s", stdout.String()) t.Logf("containerd stderr: %s", stderr.String()) }) @@ -345,13 +484,17 @@ func setup(t *testing.T) (*criutil.Crictl, func(), error) { t.Fatalf("failed to start containerd: %v", err) } + // Discard all subsequent data. + go io.Copy(ioutil.Discard, startupR) + + // Create the crictl interface. + cc := criutil.NewCrictl(t, sockAddr, runpArgs) + cu.Add(cc.CleanUp) + // Kill must be the last cleanup (as it will be executed first). - cc := criutil.NewCrictl(t, sockAddr) cu.Add(func() { - cc.CleanUp() // Remove tmp files, etc. - if err := testutil.KillCommand(cmd); err != nil { - log.Printf("error killing containerd: %v", err) - } + // Best effort: ignore errors. + testutil.KillCommand(cmd) }) return cc, cu.Release(), nil diff --git a/test/runner/BUILD b/test/runner/BUILD index 6833c9986..1f45a6922 100644 --- a/test/runner/BUILD +++ b/test/runner/BUILD @@ -16,7 +16,7 @@ go_binary( "//runsc/specutils", "//test/runner/gtest", "//test/uds", - "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", + "@com_github_opencontainers_runtime_spec//specs-go:go_default_library", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/test/runner/defs.bzl b/test/runner/defs.bzl index 921e499be..600cb5192 100644 --- a/test/runner/defs.bzl +++ b/test/runner/defs.bzl @@ -61,7 +61,8 @@ def _syscall_test( file_access = "exclusive", overlay = False, add_uds_tree = False, - vfs2 = False): + vfs2 = False, + fuse = False): # Prepend "runsc" to non-native platform names. full_platform = platform if platform == "native" else "runsc_" + platform @@ -73,6 +74,8 @@ def _syscall_test( name += "_overlay" if vfs2: name += "_vfs2" + if fuse: + name += "_fuse" if network != "none": name += "_" + network + "net" @@ -107,6 +110,7 @@ def _syscall_test( "--overlay=" + str(overlay), "--add-uds-tree=" + str(add_uds_tree), "--vfs2=" + str(vfs2), + "--fuse=" + str(fuse), ] # Call the rule above. @@ -129,6 +133,7 @@ def syscall_test( add_uds_tree = False, add_hostinet = False, vfs2 = False, + fuse = False, tags = None): """syscall_test is a macro that will create targets for all platforms. @@ -188,6 +193,19 @@ def syscall_test( vfs2 = True, ) + if vfs2 and fuse: + _syscall_test( + test = test, + shard_count = shard_count, + size = size, + platform = default_platform, + use_tmpfs = use_tmpfs, + add_uds_tree = add_uds_tree, + tags = platforms[default_platform] + vfs2_tags, + vfs2 = True, + fuse = True, + ) + # TODO(gvisor.dev/issue/1487): Enable VFS2 overlay tests. if add_overlay: _syscall_test( diff --git a/test/runner/runner.go b/test/runner/runner.go index 5456e46a6..2296f3a46 100644 --- a/test/runner/runner.go +++ b/test/runner/runner.go @@ -47,6 +47,7 @@ var ( fileAccess = flag.String("file-access", "exclusive", "mounts root in exclusive or shared mode") overlay = flag.Bool("overlay", false, "wrap filesystem mounts with writable tmpfs overlay") vfs2 = flag.Bool("vfs2", false, "enable VFS2") + fuse = flag.Bool("fuse", false, "enable FUSE") parallel = flag.Bool("parallel", false, "run tests in parallel") runscPath = flag.String("runsc", "", "path to runsc binary") @@ -149,6 +150,9 @@ func runRunsc(tc gtest.TestCase, spec *specs.Spec) error { } if *vfs2 { args = append(args, "-vfs2") + if *fuse { + args = append(args, "-fuse") + } } if *debug { args = append(args, "-debug", "-log-packets=true") @@ -358,6 +362,12 @@ func runTestCaseRunsc(testBin string, tc gtest.TestCase, t *testing.T) { vfsVar := "GVISOR_VFS" if *vfs2 { env = append(env, vfsVar+"=VFS2") + fuseVar := "FUSE_ENABLED" + if *fuse { + env = append(env, fuseVar+"=TRUE") + } else { + env = append(env, fuseVar+"=FALSE") + } } else { env = append(env, vfsVar+"=VFS1") } diff --git a/test/runtimes/BUILD b/test/runtimes/BUILD index 022de5ff7..f98d02e00 100644 --- a/test/runtimes/BUILD +++ b/test/runtimes/BUILD @@ -6,28 +6,33 @@ runtime_test( name = "go1.12", exclude_file = "exclude_go1.12.csv", lang = "go", + shard_count = 5, ) runtime_test( name = "java11", exclude_file = "exclude_java11.csv", lang = "java", + shard_count = 10, ) runtime_test( name = "nodejs12.4.0", exclude_file = "exclude_nodejs12.4.0.csv", lang = "nodejs", + shard_count = 5, ) runtime_test( name = "php7.3.6", exclude_file = "exclude_php7.3.6.csv", lang = "php", + shard_count = 5, ) runtime_test( name = "python3.7.3", exclude_file = "exclude_python3.7.3.csv", lang = "python", + shard_count = 5, ) diff --git a/test/runtimes/defs.bzl b/test/runtimes/defs.bzl index dc3667f05..5779b9591 100644 --- a/test/runtimes/defs.bzl +++ b/test/runtimes/defs.bzl @@ -20,7 +20,7 @@ def _runtime_test_impl(ctx): 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)), + "%s %s $@\n" % (ctx.files._runner[0].short_path, " ".join(args)), ]) ctx.actions.write(runner, runner_content, is_executable = True) @@ -65,6 +65,7 @@ def runtime_test(name, **kwargs): "local", "manual", ], + size = "enormous", **kwargs ) diff --git a/test/runtimes/exclude_nodejs12.4.0.csv b/test/runtimes/exclude_nodejs12.4.0.csv index 4ab4e2927..e7edfa0a5 100644 --- a/test/runtimes/exclude_nodejs12.4.0.csv +++ b/test/runtimes/exclude_nodejs12.4.0.csv @@ -9,6 +9,8 @@ fixtures/test-fs-stat-sync-overflow.js,, internet/test-dgram-broadcast-multi-process.js,, internet/test-dgram-multicast-multi-process.js,, internet/test-dgram-multicast-set-interface-lo.js,, +internet/test-doctool-versions.js,, +internet/test-uv-threadpool-schedule.js,, parallel/test-cluster-dgram-reuse.js,b/64024294, parallel/test-dgram-bind-fd.js,b/132447356, parallel/test-dgram-create-socket-handle-fd.js,b/132447238, @@ -45,3 +47,4 @@ pseudo-tty/test-tty-window-size.js,, pseudo-tty/test-tty-wrap.js,, pummel/test-net-pingpong.js,, pummel/test-vm-memleak.js,, +tick-processor/test-tick-processor-builtin.js,, diff --git a/test/runtimes/exclude_php7.3.6.csv b/test/runtimes/exclude_php7.3.6.csv index 456bf7487..f3606bfe8 100644 --- a/test/runtimes/exclude_php7.3.6.csv +++ b/test/runtimes/exclude_php7.3.6.csv @@ -8,6 +8,9 @@ ext/mbstring/tests/bug77165.phpt,, ext/mbstring/tests/bug77454.phpt,, ext/mbstring/tests/mb_convert_encoding_leak.phpt,, ext/mbstring/tests/mb_strrpos_encoding_3rd_param.phpt,, +ext/session/tests/session_set_save_handler_class_018.phpt,, +ext/session/tests/session_set_save_handler_iface_003.phpt,, +ext/session/tests/session_set_save_handler_variation4.phpt,, ext/standard/tests/file/filetype_variation.phpt,, ext/standard/tests/file/fopen_variation19.phpt,, ext/standard/tests/file/php_fd_wrapper_01.phpt,, @@ -21,9 +24,12 @@ ext/standard/tests/file/symlink_link_linkinfo_is_link_variation8.phpt,, ext/standard/tests/general_functions/escapeshellarg_bug71270.phpt,, ext/standard/tests/general_functions/escapeshellcmd_bug71270.phpt,, ext/standard/tests/network/bug20134.phpt,, +ext/standard/tests/streams/stream_socket_sendto.phpt,, +ext/standard/tests/strings/007.phpt,, tests/output/stream_isatty_err.phpt,b/68720279, tests/output/stream_isatty_in-err.phpt,b/68720282, tests/output/stream_isatty_in-out-err.phpt,, tests/output/stream_isatty_in-out.phpt,b/68720299, tests/output/stream_isatty_out-err.phpt,b/68720311, tests/output/stream_isatty_out.phpt,b/68720325, +Zend/tests/concat_003.phpt,, diff --git a/test/runtimes/proctor/go.go b/test/runtimes/proctor/go.go index 3e2d5d8db..073c2959d 100644 --- a/test/runtimes/proctor/go.go +++ b/test/runtimes/proctor/go.go @@ -74,17 +74,26 @@ func (goRunner) ListTests() ([]string, error) { return append(toolSlice, diskFiltered...), nil } -// TestCmd implements TestRunner.TestCmd. -func (goRunner) TestCmd(test string) *exec.Cmd { - // Check if test exists on disk by searching for file of the same name. - // This will determine whether or not it is a Go test on disk. - if strings.HasSuffix(test, ".go") { - // Test has suffix ".go" which indicates a disk test, run it as such. - cmd := exec.Command("go", "run", "run.go", "-v", "--", test) +// TestCmds implements TestRunner.TestCmds. +func (goRunner) TestCmds(tests []string) []*exec.Cmd { + var toolTests, onDiskTests []string + for _, test := range tests { + if strings.HasSuffix(test, ".go") { + onDiskTests = append(onDiskTests, test) + } else { + toolTests = append(toolTests, test) + } + } + + var cmds []*exec.Cmd + if len(toolTests) > 0 { + cmds = append(cmds, exec.Command("go", "tool", "dist", "test", "-run", strings.Join(toolTests, "\\|"))) + } + if len(onDiskTests) > 0 { + cmd := exec.Command("go", append([]string{"run", "run.go", "-v", "--"}, onDiskTests...)...) cmd.Dir = goTestDir - return cmd + cmds = append(cmds, cmd) } - // No ".go" suffix, run as a tool test. - return exec.Command("go", "tool", "dist", "test", "-run", test) + return cmds } diff --git a/test/runtimes/proctor/java.go b/test/runtimes/proctor/java.go index 8b362029d..737fbe23e 100644 --- a/test/runtimes/proctor/java.go +++ b/test/runtimes/proctor/java.go @@ -60,12 +60,14 @@ func (javaRunner) ListTests() ([]string, error) { return testSlice, nil } -// TestCmd implements TestRunner.TestCmd. -func (javaRunner) TestCmd(test string) *exec.Cmd { - args := []string{ - "-noreport", - "-dir:" + javaTestDir, - test, - } - return exec.Command("jtreg", args...) +// TestCmds implements TestRunner.TestCmds. +func (javaRunner) TestCmds(tests []string) []*exec.Cmd { + args := append( + []string{ + "-noreport", + "-dir:" + javaTestDir, + }, + tests..., + ) + return []*exec.Cmd{exec.Command("jtreg", args...)} } diff --git a/test/runtimes/proctor/nodejs.go b/test/runtimes/proctor/nodejs.go index bd57db444..23d6edc72 100644 --- a/test/runtimes/proctor/nodejs.go +++ b/test/runtimes/proctor/nodejs.go @@ -39,8 +39,8 @@ func (nodejsRunner) ListTests() ([]string, error) { return testSlice, nil } -// TestCmd implements TestRunner.TestCmd. -func (nodejsRunner) TestCmd(test string) *exec.Cmd { - args := []string{filepath.Join("tools", "test.py"), test} - return exec.Command("/usr/bin/python", args...) +// TestCmds implements TestRunner.TestCmds. +func (nodejsRunner) TestCmds(tests []string) []*exec.Cmd { + args := append([]string{filepath.Join("tools", "test.py")}, tests...) + return []*exec.Cmd{exec.Command("/usr/bin/python", args...)} } diff --git a/test/runtimes/proctor/php.go b/test/runtimes/proctor/php.go index 9115040e1..6a83d64e3 100644 --- a/test/runtimes/proctor/php.go +++ b/test/runtimes/proctor/php.go @@ -17,6 +17,7 @@ package main import ( "os/exec" "regexp" + "strings" ) var phpTestRegEx = regexp.MustCompile(`^.+\.phpt$`) @@ -35,8 +36,8 @@ func (phpRunner) ListTests() ([]string, error) { return testSlice, nil } -// TestCmd implements TestRunner.TestCmd. -func (phpRunner) TestCmd(test string) *exec.Cmd { - args := []string{"test", "TESTS=" + test} - return exec.Command("make", args...) +// TestCmds implements TestRunner.TestCmds. +func (phpRunner) TestCmds(tests []string) []*exec.Cmd { + args := []string{"test", "TESTS=" + strings.Join(tests, " ")} + return []*exec.Cmd{exec.Command("make", args...)} } diff --git a/test/runtimes/proctor/proctor.go b/test/runtimes/proctor/proctor.go index b54abe434..9e0642424 100644 --- a/test/runtimes/proctor/proctor.go +++ b/test/runtimes/proctor/proctor.go @@ -25,6 +25,7 @@ import ( "os/signal" "path/filepath" "regexp" + "strings" "syscall" ) @@ -34,15 +35,18 @@ type TestRunner interface { // ListTests returns a string slice of tests available to run. ListTests() ([]string, error) - // TestCmd returns an *exec.Cmd that will run the given test. - TestCmd(test string) *exec.Cmd + // TestCmds returns a slice of *exec.Cmd that will run the given tests. + // There is no correlation between the number of exec.Cmds returned and the + // number of tests. It could return one command to run all tests or a few + // commands that collectively run all. + TestCmds(tests []string) []*exec.Cmd } var ( - runtime = flag.String("runtime", "", "name of runtime") - list = flag.Bool("list", false, "list all available tests") - testName = flag.String("test", "", "run a single test from the list of available tests") - pause = flag.Bool("pause", false, "cause container to pause indefinitely, reaping any zombie children") + runtime = flag.String("runtime", "", "name of runtime") + list = flag.Bool("list", false, "list all available tests") + testNames = flag.String("tests", "", "run a subset of the available tests") + pause = flag.Bool("pause", false, "cause container to pause indefinitely, reaping any zombie children") ) func main() { @@ -75,18 +79,20 @@ func main() { } var tests []string - if *testName == "" { + if *testNames == "" { // Run every test. tests, err = tr.ListTests() if err != nil { log.Fatalf("failed to get all tests: %v", err) } } else { - // Run a single test. - tests = []string{*testName} + // Run subset of test. + tests = strings.Split(*testNames, ",") } - for _, test := range tests { - cmd := tr.TestCmd(test) + + // Run tests. + cmds := tr.TestCmds(tests) + for _, cmd := range cmds { cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr if err := cmd.Run(); err != nil { log.Fatalf("FAIL: %v", err) diff --git a/test/runtimes/proctor/python.go b/test/runtimes/proctor/python.go index b9e0fbe6f..7c598801b 100644 --- a/test/runtimes/proctor/python.go +++ b/test/runtimes/proctor/python.go @@ -42,8 +42,8 @@ func (pythonRunner) ListTests() ([]string, error) { return toolSlice, nil } -// TestCmd implements TestRunner.TestCmd. -func (pythonRunner) TestCmd(test string) *exec.Cmd { - args := []string{"-m", "test", test} - return exec.Command("./python", args...) +// TestCmds implements TestRunner.TestCmds. +func (pythonRunner) TestCmds(tests []string) []*exec.Cmd { + args := append([]string{"-m", "test"}, tests...) + return []*exec.Cmd{exec.Command("./python", args...)} } diff --git a/test/runtimes/runner/BUILD b/test/runtimes/runner/BUILD index 3972244b9..dc0d5d5b4 100644 --- a/test/runtimes/runner/BUILD +++ b/test/runtimes/runner/BUILD @@ -8,6 +8,7 @@ go_binary( srcs = ["main.go"], visibility = ["//test/runtimes:__pkg__"], deps = [ + "//pkg/log", "//pkg/test/dockerutil", "//pkg/test/testutil", ], diff --git a/test/runtimes/runner/exclude_test.go b/test/runtimes/runner/exclude_test.go index c08755894..67c2170c8 100644 --- a/test/runtimes/runner/exclude_test.go +++ b/test/runtimes/runner/exclude_test.go @@ -26,7 +26,7 @@ func TestMain(m *testing.M) { } // Test that the exclude file parses without error. -func TestBlacklists(t *testing.T) { +func TestExcludelist(t *testing.T) { ex, err := getExcludes() if err != nil { t.Fatalf("error parsing exclude file: %v", err) diff --git a/test/runtimes/runner/main.go b/test/runtimes/runner/main.go index 54d1169ef..e230912c9 100644 --- a/test/runtimes/runner/main.go +++ b/test/runtimes/runner/main.go @@ -16,6 +16,7 @@ package main import ( + "context" "encoding/csv" "flag" "fmt" @@ -26,6 +27,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/pkg/test/testutil" ) @@ -34,10 +36,11 @@ var ( lang = flag.String("lang", "", "language runtime to test") image = flag.String("image", "", "docker image with runtime tests") excludeFile = flag.String("exclude_file", "", "file containing list of tests to exclude, in CSV format with fields: test name, bug id, comment") + batchSize = flag.Int("batch", 50, "number of test cases run in one command") ) // Wait time for each test to run. -const timeout = 5 * time.Minute +const timeout = 45 * time.Minute func main() { flag.Parse() @@ -60,13 +63,19 @@ func runTests() int { } // Construct the shared docker instance. - d := dockerutil.MakeDocker(testutil.DefaultLogger(*lang)) - defer d.CleanUp() + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, testutil.DefaultLogger(*lang)) + defer d.CleanUp(ctx) + + if err := testutil.TouchShardStatusFile(); err != nil { + fmt.Fprintf(os.Stderr, "error touching status shard file: %v\n", err) + return 1 + } // Get a slice of tests to run. This will also start a single Docker // container that will be used to run each test. The final test will // stop the Docker container. - tests, err := getTests(d, excludes) + tests, err := getTests(ctx, d, excludes) if err != nil { fmt.Fprintf(os.Stderr, "%s\n", err.Error()) return 1 @@ -77,18 +86,18 @@ func runTests() int { } // getTests executes all tests as table tests. -func getTests(d *dockerutil.Docker, excludes map[string]struct{}) ([]testing.InternalTest, error) { +func getTests(ctx context.Context, d *dockerutil.Container, excludes map[string]struct{}) ([]testing.InternalTest, error) { // Start the container. opts := dockerutil.RunOpts{ Image: fmt.Sprintf("runtimes/%s", *image), } d.CopyFiles(&opts, "/proctor", "test/runtimes/proctor/proctor") - if err := d.Spawn(opts, "/proctor/proctor", "--pause"); err != nil { + if err := d.Spawn(ctx, opts, "/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(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--list") + list, err := d.Exec(ctx, dockerutil.ExecOpts{}, "/proctor/proctor", "--runtime", *lang, "--list") if err != nil { return nil, fmt.Errorf("docker exec failed: %v", err) } @@ -103,17 +112,23 @@ func getTests(d *dockerutil.Docker, excludes map[string]struct{}) ([]testing.Int } var itests []testing.InternalTest - for _, tci := range indices { - // Capture tc in this scope. - tc := tests[tci] + for i := 0; i < len(indices); i += *batchSize { + var tcs []string + end := i + *batchSize + if end > len(indices) { + end = len(indices) + } + for _, tc := range indices[i:end] { + // Add test if not excluded. + if _, ok := excludes[tests[tc]]; ok { + log.Infof("Skipping test case %s\n", tests[tc]) + continue + } + tcs = append(tcs, tests[tc]) + } itests = append(itests, testing.InternalTest{ - Name: tc, + Name: strings.Join(tcs, ", "), F: func(t *testing.T) { - // Is the test excluded? - if _, ok := excludes[tc]; ok { - t.Skipf("SKIP: excluded test %q", tc) - } - var ( now = time.Now() done = make(chan struct{}) @@ -122,20 +137,20 @@ func getTests(d *dockerutil.Docker, excludes map[string]struct{}) ([]testing.Int ) go func() { - fmt.Printf("RUNNING %s...\n", tc) - output, err = d.Exec(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--test", tc) + fmt.Printf("RUNNING the following in a batch\n%s\n", strings.Join(tcs, "\n")) + output, err = d.Exec(ctx, dockerutil.ExecOpts{}, "/proctor/proctor", "--runtime", *lang, "--tests", strings.Join(tcs, ",")) close(done) }() select { case <-done: if err == nil { - fmt.Printf("PASS: %s (%v)\n\n", tc, time.Since(now)) + fmt.Printf("PASS: (%v)\n\n", time.Since(now)) return } - t.Errorf("FAIL: %s (%v):\n%s\n", tc, time.Since(now), output) + t.Errorf("FAIL: (%v):\n%s\n", time.Since(now), output) case <-time.After(timeout): - t.Errorf("TIMEOUT: %s (%v):\n%s\n", tc, time.Since(now), output) + t.Errorf("TIMEOUT: (%v):\n%s\n", time.Since(now), output) } }, }) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 7c4cd8192..15e5fd223 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -146,6 +146,7 @@ syscall_test( ) syscall_test( + fuse = "True", test = "//test/syscalls/linux:dev_test", vfs2 = "True", ) @@ -283,6 +284,7 @@ syscall_test( size = "medium", add_overlay = False, # TODO(gvisor.dev/issue/317): enable when fixed. test = "//test/syscalls/linux:inotify_test", + vfs2 = "True", ) syscall_test( @@ -351,6 +353,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:mknod_test", + vfs2 = "True", ) syscall_test( @@ -637,11 +640,13 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_socket_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 9e097c888..66a31cd28 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -943,6 +943,7 @@ cc_binary( "//test/util:eventfd_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:node_hash_set", "@com_google_absl//absl/strings", gtest, "//test/util:posix_error", @@ -1330,6 +1331,7 @@ cc_binary( name = "packet_socket_raw_test", testonly = 1, srcs = ["packet_socket_raw.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -1809,6 +1811,7 @@ cc_binary( name = "raw_socket_test", testonly = 1, srcs = ["raw_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -3407,6 +3410,7 @@ cc_binary( name = "tcp_socket_test", testonly = 1, srcs = ["tcp_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 3c88c4cbd..1d0d584cd 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -156,11 +156,24 @@ TEST(DevTest, TTYExists) { TEST(DevTest, OpenDevFuse) { // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new // device registration is complete. - SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor() || !IsFUSEEnabled()); ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); } +TEST(DevTest, ReadDevFuseWithoutMount) { + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); + + std::vector<char> buf(1); + EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), + SyscallFailsWithErrno(EPERM)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/exec.cc b/test/syscalls/linux/exec.cc index e09afafe9..c5acfc794 100644 --- a/test/syscalls/linux/exec.cc +++ b/test/syscalls/linux/exec.cc @@ -553,7 +553,12 @@ TEST(ExecTest, SymlinkLimitRefreshedForInterpreter) { // Hold onto TempPath objects so they are not destructed prematurely. std::vector<TempPath> interpreter_symlinks; std::vector<TempPath> script_symlinks; - for (int i = 0; i < kLinuxMaxSymlinks; i++) { + // Replace both the interpreter and script paths with symlink chains of just + // over half the symlink limit each; this is the minimum required to test that + // the symlink limit applies separately to each traversal, while tolerating + // some symlinks in the resolution of (the original) interpreter_path and + // script_path. + for (int i = 0; i < (kLinuxMaxSymlinks / 2) + 1; i++) { interpreter_symlinks.push_back(ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(tmp_dir, interpreter_path))); interpreter_path = interpreter_symlinks[i].path(); @@ -679,18 +684,16 @@ TEST(ExecveatTest, UnshareFiles) { const FileDescriptor fd_closed_on_exec = ASSERT_NO_ERRNO_AND_VALUE(Open(tempFile.path(), O_RDONLY | O_CLOEXEC)); - pid_t child; - EXPECT_THAT(child = syscall(__NR_clone, SIGCHLD | CLONE_VFORK | CLONE_FILES, - 0, 0, 0, 0), - SyscallSucceeds()); + ExecveArray argv = {"test"}; + ExecveArray envp; + std::string child_path = RunfilePath(kBasicWorkload); + pid_t child = + syscall(__NR_clone, SIGCHLD | CLONE_VFORK | CLONE_FILES, 0, 0, 0, 0); if (child == 0) { - ExecveArray argv = {"test"}; - ExecveArray envp; - ASSERT_THAT( - execve(RunfilePath(kBasicWorkload).c_str(), argv.get(), envp.get()), - SyscallSucceeds()); + execve(child_path.c_str(), argv.get(), envp.get()); _exit(1); } + ASSERT_THAT(child, SyscallSucceeds()); int status; ASSERT_THAT(RetryEINTR(waitpid)(child, &status, 0), SyscallSucceeds()); diff --git a/test/syscalls/linux/futex.cc b/test/syscalls/linux/futex.cc index 40c80a6e1..90b1f0508 100644 --- a/test/syscalls/linux/futex.cc +++ b/test/syscalls/linux/futex.cc @@ -18,6 +18,7 @@ #include <sys/syscall.h> #include <sys/time.h> #include <sys/types.h> +#include <syscall.h> #include <unistd.h> #include <algorithm> @@ -737,6 +738,97 @@ TEST_P(PrivateAndSharedFutexTest, PITryLockConcurrency_NoRandomSave) { } } +int get_robust_list(int pid, struct robust_list_head** head_ptr, + size_t* len_ptr) { + return syscall(__NR_get_robust_list, pid, head_ptr, len_ptr); +} + +int set_robust_list(struct robust_list_head* head, size_t len) { + return syscall(__NR_set_robust_list, head, len); +} + +TEST(RobustFutexTest, BasicSetGet) { + struct robust_list_head hd = {}; + struct robust_list_head* hd_ptr = &hd; + + // Set! + EXPECT_THAT(set_robust_list(hd_ptr, sizeof(hd)), SyscallSucceedsWithValue(0)); + + // Get! + struct robust_list_head* new_hd_ptr = hd_ptr; + size_t len; + EXPECT_THAT(get_robust_list(0, &new_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(new_hd_ptr, hd_ptr); + EXPECT_EQ(len, sizeof(hd)); +} + +TEST(RobustFutexTest, GetFromOtherTid) { + // Get the current tid and list head. + pid_t tid = gettid(); + struct robust_list_head* hd_ptr = {}; + size_t len; + EXPECT_THAT(get_robust_list(0, &hd_ptr, &len), SyscallSucceedsWithValue(0)); + + // Create a new thread. + ScopedThread t([&] { + // Current tid list head should be different from parent tid. + struct robust_list_head* got_hd_ptr = {}; + EXPECT_THAT(get_robust_list(0, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_NE(hd_ptr, got_hd_ptr); + + // Get the parent list head by passing its tid. + EXPECT_THAT(get_robust_list(tid, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(hd_ptr, got_hd_ptr); + }); + + // Wait for thread. + t.Join(); +} + +TEST(RobustFutexTest, InvalidSize) { + struct robust_list_head* hd = {}; + EXPECT_THAT(set_robust_list(hd, sizeof(*hd) + 1), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(RobustFutexTest, PthreadMutexAttr) { + constexpr int kNumMutexes = 3; + + // Create a bunch of robust mutexes. + pthread_mutexattr_t attrs[kNumMutexes]; + pthread_mutex_t mtxs[kNumMutexes]; + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutexattr_init(&attrs[i]) == 0); + TEST_PCHECK(pthread_mutexattr_setrobust(&attrs[i], PTHREAD_MUTEX_ROBUST) == + 0); + TEST_PCHECK(pthread_mutex_init(&mtxs[i], &attrs[i]) == 0); + } + + // Start thread to lock the mutexes and then exit. + ScopedThread t([&] { + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutex_lock(&mtxs[i]) == 0); + } + pthread_exit(NULL); + }); + + // Wait for thread. + t.Join(); + + // Now try to take the mutexes. + for (int i = 0; i < kNumMutexes; i++) { + // Should get EOWNERDEAD. + EXPECT_EQ(pthread_mutex_lock(&mtxs[i]), EOWNERDEAD); + // Make the mutex consistent. + EXPECT_EQ(pthread_mutex_consistent(&mtxs[i]), 0); + // Unlock. + EXPECT_EQ(pthread_mutex_unlock(&mtxs[i]), 0); + } +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/getdents.cc b/test/syscalls/linux/getdents.cc index b147d6181..b040cdcf7 100644 --- a/test/syscalls/linux/getdents.cc +++ b/test/syscalls/linux/getdents.cc @@ -32,6 +32,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/container/node_hash_set.h" #include "absl/strings/numbers.h" #include "absl/strings/str_cat.h" #include "test/util/eventfd_util.h" @@ -393,7 +394,7 @@ TYPED_TEST(GetdentsTest, ProcSelfFd) { // Make the buffer very small since we want to iterate. typename TestFixture::DirentBufferType dirents( 2 * sizeof(typename TestFixture::LinuxDirentType)); - std::unordered_set<int> prev_fds; + absl::node_hash_set<int> prev_fds; while (true) { dirents.Reset(); int rv; diff --git a/test/syscalls/linux/mknod.cc b/test/syscalls/linux/mknod.cc index 4c45766c7..05dfb375a 100644 --- a/test/syscalls/linux/mknod.cc +++ b/test/syscalls/linux/mknod.cc @@ -15,6 +15,7 @@ #include <errno.h> #include <fcntl.h> #include <sys/stat.h> +#include <sys/types.h> #include <sys/un.h> #include <unistd.h> @@ -39,7 +40,28 @@ TEST(MknodTest, RegularFile) { EXPECT_THAT(mknod(node1.c_str(), 0, 0), SyscallSucceeds()); } -TEST(MknodTest, MknodAtRegularFile) { +TEST(MknodTest, RegularFilePermissions) { + const std::string node = NewTempAbsPath(); + mode_t newUmask = 0077; + umask(newUmask); + + // Attempt to open file with mode 0777. Not specifying file type should create + // a regualar file. + mode_t perms = S_IRWXU | S_IRWXG | S_IRWXO; + EXPECT_THAT(mknod(node.c_str(), perms, 0), SyscallSucceeds()); + + // In the absence of a default ACL, the permissions of the created node are + // (mode & ~umask). -- mknod(2) + mode_t wantPerms = perms & ~newUmask; + struct stat st; + ASSERT_THAT(stat(node.c_str(), &st), SyscallSucceeds()); + ASSERT_EQ(st.st_mode & 0777, wantPerms); + + // "Zero file type is equivalent to type S_IFREG." - mknod(2) + ASSERT_EQ(st.st_mode & S_IFMT, S_IFREG); +} + +TEST(MknodTest, MknodAtFIFO) { const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const std::string fifo_relpath = NewTempRelPath(); const std::string fifo = JoinPath(dir.path(), fifo_relpath); @@ -72,7 +94,7 @@ TEST(MknodTest, MknodOnExistingPathFails) { TEST(MknodTest, UnimplementedTypesReturnError) { const std::string path = NewTempAbsPath(); - if (IsRunningOnGvisor()) { + if (IsRunningWithVFS1()) { ASSERT_THAT(mknod(path.c_str(), S_IFSOCK, 0), SyscallFailsWithErrno(EOPNOTSUPP)); } diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc index a3e9745cf..7664fa73d 100644 --- a/test/syscalls/linux/mount.cc +++ b/test/syscalls/linux/mount.cc @@ -321,6 +321,34 @@ TEST(MountTest, RenameRemoveMountPoint) { ASSERT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EBUSY)); } +TEST(MountTest, MountFuseFilesystemNoDevice) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(mount("", dir.path().c_str(), "fuse", 0, ""), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(MountTest, MountFuseFilesystem) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY)); + std::string mopts = "fd=" + std::to_string(fd.get()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + auto const mount = + ASSERT_NO_ERRNO_AND_VALUE(Mount("", dir.path(), "fuse", 0, mopts, 0)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index bb7d108e8..bf350946b 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -235,7 +235,7 @@ TEST_F(OpenTest, AppendOnly) { ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_, O_RDWR | O_APPEND)); EXPECT_THAT(lseek(fd2.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(0)); - // Then try to write to the first file and make sure the bytes are appended. + // Then try to write to the first fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd1.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); @@ -247,7 +247,7 @@ TEST_F(OpenTest, AppendOnly) { EXPECT_THAT(lseek(fd1.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(kBufSize * 2)); - // Then try to write to the second file and make sure the bytes are appended. + // Then try to write to the second fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd2.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 5ac68feb4..40aa9326d 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -193,6 +193,7 @@ void ReceiveMessage(int sock, int ifindex) { EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, ifindex); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); @@ -343,7 +344,7 @@ TEST_P(CookedPacketTest, BindReceive) { } // Double Bind socket. -TEST_P(CookedPacketTest, DoubleBind) { +TEST_P(CookedPacketTest, DoubleBindSucceeds) { struct sockaddr_ll bind_addr = {}; bind_addr.sll_family = AF_PACKET; bind_addr.sll_protocol = htons(GetParam()); @@ -354,12 +355,11 @@ TEST_P(CookedPacketTest, DoubleBind) { SyscallSucceeds()); // Binding socket again should fail. - ASSERT_THAT( - bind(socket_, reinterpret_cast<struct sockaddr*>(&bind_addr), - sizeof(bind_addr)), - // Linux 4.09 returns EINVAL here, but some time before 4.19 it switched - // to EADDRINUSE. - AnyOf(SyscallFailsWithErrno(EADDRINUSE), SyscallFailsWithErrno(EINVAL))); + ASSERT_THAT(bind(socket_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + // Linux 4.09 returns EINVAL here, but some time before 4.19 it + // switched to EADDRINUSE. + SyscallSucceeds()); } // Bind and verify we do not receive data on interface which is not bound @@ -417,6 +417,122 @@ TEST_P(CookedPacketTest, BindDrop) { EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); } +// Verify that we receive outbound packets. This test requires at least one +// non loopback interface so that we can actually capture an outgoing packet. +TEST_P(CookedPacketTest, ReceiveOutbound) { + // Only ETH_P_ALL sockets can receive outbound packets on linux. + SKIP_IF(GetParam() != ETH_P_ALL); + + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ifaddrs* if_addr_list = nullptr; + auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); }); + + ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds()); + + // Get interface other than loopback. + struct ifreq ifr = {}; + for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) { + if (strcmp(i->ifa_name, "lo") != 0) { + strncpy(ifr.ifr_name, i->ifa_name, sizeof(ifr.ifr_name)); + break; + } + } + + // Skip if no interface is available other than loopback. + if (strlen(ifr.ifr_name) == 0) { + GTEST_SKIP(); + } + + // Get interface index and name. + EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_NE(ifr.ifr_ifindex, 0); + int ifindex = ifr.ifr_ifindex; + + constexpr int kMACSize = 6; + char hwaddr[kMACSize]; + // Get interface address. + ASSERT_THAT(ioctl(socket_, SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + ASSERT_THAT(ifr.ifr_hwaddr.sa_family, + AnyOf(Eq(ARPHRD_NONE), Eq(ARPHRD_ETHER))); + memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, kMACSize); + + // Just send it to the google dns server 8.8.8.8. It's UDP we don't care + // if it actually gets to the DNS Server we just want to see that we receive + // it on our AF_PACKET socket. + // + // NOTE: We just want to pick an IP that is non-local to avoid having to + // handle ARP as this should cause the UDP packet to be sent to the default + // gateway configured for the system under test. Otherwise the only packet we + // will see is the ARP query unless we picked an IP which will actually + // resolve. The test is a bit brittle but this was the best compromise for + // now. + struct sockaddr_in dest = {}; + ASSERT_EQ(inet_pton(AF_INET, "8.8.8.8", &dest.sin_addr.s_addr), 1); + dest.sin_family = AF_INET; + dest.sin_port = kPort; + EXPECT_THAT(sendto(udp_sock.get(), kMessage, sizeof(kMessage), 0, + reinterpret_cast<struct sockaddr*>(&dest), sizeof(dest)), + SyscallSucceedsWithValue(sizeof(kMessage))); + + // Wait and make sure the socket receives the data. + struct pollfd pfd = {}; + pfd.fd = socket_; + pfd.events = POLLIN; + EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(1)); + + // Now read and check that the packet is the one we just sent. + // Read and verify the data. + constexpr size_t packet_size = + sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kMessage); + char buf[64]; + struct sockaddr_ll src = {}; + socklen_t src_len = sizeof(src); + ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + reinterpret_cast<struct sockaddr*>(&src), &src_len), + SyscallSucceedsWithValue(packet_size)); + + // sockaddr_ll ends with an 8 byte physical address field, but ethernet + // addresses only use 6 bytes. Linux used to return sizeof(sockaddr_ll)-2 + // here, but since commit b2cf86e1563e33a14a1c69b3e508d15dc12f804c returns + // sizeof(sockaddr_ll). + ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); + + // Verify the source address. + EXPECT_EQ(src.sll_family, AF_PACKET); + EXPECT_EQ(src.sll_ifindex, ifindex); + EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); + EXPECT_EQ(src.sll_pkttype, PACKET_OUTGOING); + // Verify the link address of the interface matches that of the non + // non loopback interface address we stored above. + for (int i = 0; i < src.sll_halen; i++) { + EXPECT_EQ(src.sll_addr[i], hwaddr[i]); + } + + // Verify the IP header. + struct iphdr ip = {}; + memcpy(&ip, buf, sizeof(ip)); + EXPECT_EQ(ip.ihl, 5); + EXPECT_EQ(ip.version, 4); + EXPECT_EQ(ip.tot_len, htons(packet_size)); + EXPECT_EQ(ip.protocol, IPPROTO_UDP); + EXPECT_EQ(ip.daddr, dest.sin_addr.s_addr); + EXPECT_NE(ip.saddr, htonl(INADDR_LOOPBACK)); + + // Verify the UDP header. + struct udphdr udp = {}; + memcpy(&udp, buf + sizeof(iphdr), sizeof(udp)); + EXPECT_EQ(udp.dest, kPort); + EXPECT_EQ(udp.len, htons(sizeof(udphdr) + sizeof(kMessage))); + + // Verify the payload. + char* payload = reinterpret_cast<char*>(buf + sizeof(iphdr) + sizeof(udphdr)); + EXPECT_EQ(strncmp(payload, kMessage, sizeof(kMessage)), 0); +} + // Bind with invalid address. TEST_P(CookedPacketTest, BindFail) { // Null address. diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index d258d353c..2fca9fe4d 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -14,6 +14,9 @@ #include <arpa/inet.h> #include <linux/capability.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <linux/if_arp.h> #include <linux/if_packet.h> #include <net/ethernet.h> @@ -97,7 +100,7 @@ class RawPacketTest : public ::testing::TestWithParam<int> { int GetLoopbackIndex(); // The socket used for both reading and writing. - int socket_; + int s_; }; void RawPacketTest::SetUp() { @@ -108,34 +111,58 @@ void RawPacketTest::SetUp() { } if (!IsRunningOnGvisor()) { + // Ensure that looped back packets aren't rejected by the kernel. FileDescriptor acceptLocal = ASSERT_NO_ERRNO_AND_VALUE( - Open("/proc/sys/net/ipv4/conf/lo/accept_local", O_RDONLY)); + Open("/proc/sys/net/ipv4/conf/lo/accept_local", O_RDWR)); FileDescriptor routeLocalnet = ASSERT_NO_ERRNO_AND_VALUE( - Open("/proc/sys/net/ipv4/conf/lo/route_localnet", O_RDONLY)); + Open("/proc/sys/net/ipv4/conf/lo/route_localnet", O_RDWR)); char enabled; ASSERT_THAT(read(acceptLocal.get(), &enabled, 1), SyscallSucceedsWithValue(1)); - ASSERT_EQ(enabled, '1'); + if (enabled != '1') { + enabled = '1'; + ASSERT_THAT(lseek(acceptLocal.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(write(acceptLocal.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_THAT(lseek(acceptLocal.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(read(acceptLocal.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_EQ(enabled, '1'); + } + ASSERT_THAT(read(routeLocalnet.get(), &enabled, 1), SyscallSucceedsWithValue(1)); - ASSERT_EQ(enabled, '1'); + if (enabled != '1') { + enabled = '1'; + ASSERT_THAT(lseek(routeLocalnet.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(write(routeLocalnet.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_THAT(lseek(routeLocalnet.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(read(routeLocalnet.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_EQ(enabled, '1'); + } } - ASSERT_THAT(socket_ = socket(AF_PACKET, SOCK_RAW, htons(GetParam())), + ASSERT_THAT(s_ = socket(AF_PACKET, SOCK_RAW, htons(GetParam())), SyscallSucceeds()); } void RawPacketTest::TearDown() { // TearDown will be run even if we skip the test. if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { - EXPECT_THAT(close(socket_), SyscallSucceeds()); + EXPECT_THAT(close(s_), SyscallSucceeds()); } } int RawPacketTest::GetLoopbackIndex() { struct ifreq ifr; snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); - EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_THAT(ioctl(s_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); EXPECT_NE(ifr.ifr_ifindex, 0); return ifr.ifr_ifindex; } @@ -149,7 +176,7 @@ TEST_P(RawPacketTest, Receive) { // Wait for the socket to become readable. struct pollfd pfd = {}; - pfd.fd = socket_; + pfd.fd = s_; pfd.events = POLLIN; EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 2000), SyscallSucceedsWithValue(1)); @@ -159,7 +186,7 @@ TEST_P(RawPacketTest, Receive) { char buf[64]; struct sockaddr_ll src = {}; socklen_t src_len = sizeof(src); - ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + ASSERT_THAT(recvfrom(s_, buf, sizeof(buf), 0, reinterpret_cast<struct sockaddr*>(&src), &src_len), SyscallSucceedsWithValue(packet_size)); // sockaddr_ll ends with an 8 byte physical address field, but ethernet @@ -173,6 +200,7 @@ TEST_P(RawPacketTest, Receive) { EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); @@ -277,7 +305,7 @@ TEST_P(RawPacketTest, Send) { sizeof(kMessage)); // Send it. - ASSERT_THAT(sendto(socket_, send_buf, sizeof(send_buf), 0, + ASSERT_THAT(sendto(s_, send_buf, sizeof(send_buf), 0, reinterpret_cast<struct sockaddr*>(&dest), sizeof(dest)), SyscallSucceedsWithValue(sizeof(send_buf))); @@ -286,13 +314,13 @@ TEST_P(RawPacketTest, Send) { pfd.fd = udp_sock.get(); pfd.events = POLLIN; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 5000), SyscallSucceedsWithValue(1)); - pfd.fd = socket_; + pfd.fd = s_; pfd.events = POLLIN; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 5000), SyscallSucceedsWithValue(1)); // Receive on the packet socket. char recv_buf[sizeof(send_buf)]; - ASSERT_THAT(recv(socket_, recv_buf, sizeof(recv_buf), 0), + ASSERT_THAT(recv(s_, recv_buf, sizeof(recv_buf), 0), SyscallSucceedsWithValue(sizeof(recv_buf))); ASSERT_EQ(memcmp(recv_buf, send_buf, sizeof(send_buf)), 0); @@ -309,6 +337,318 @@ TEST_P(RawPacketTest, Send) { EXPECT_EQ(src.sin_addr.s_addr, htonl(INADDR_LOOPBACK)); } +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketRecvBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum receive buf size by trying to set it to zero. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketRecvBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover max buf size by trying to set the largest possible buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= kRcvBufSz <= max is honored. +TEST_P(RawPacketTest, SetSocketRecvBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover max buf size by trying to set a really large buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set a zero size receive buffer + // size. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove when Netstack matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketSendBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(RawPacketTest, SetSocketSendBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(RawPacketTest, SetSocketSendBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove the gvisor special casing when Netstack + // matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} + +TEST_P(RawPacketTest, GetSocketError) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); +} + +TEST_P(RawPacketTest, GetSocketErrorBind) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + { + // Bind to the loopback device. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // SO_ERROR should return no errors. + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); + } + + { + // Now try binding to an invalid interface. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = 0xffff; // Just pick a really large number. + + // Binding should fail with EINVAL + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + SyscallFailsWithErrno(ENODEV)); + + // SO_ERROR does not return error when the device is invalid. + // On Linux there is just one odd ball condition where this can return + // an error where the device was valid and then removed or disabled + // between the first check for index and the actual registration of + // the packet endpoint. On Netstack this is not possible as the stack + // global mutex is held during registration and check. + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); + } +} + +#ifndef __fuchsia__ + +TEST_P(RawPacketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + // + // gVisor returns no error on SO_DETACH_FILTER even if there is no filter + // attached unlike linux which does return ENOENT in such cases. This is + // because gVisor doesn't support SO_ATTACH_FILTER and just silently returns + // success. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawPacketTest, GetSocketDetachFilter) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, RawPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index aabfa6955..f9392b9e0 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -634,6 +634,11 @@ TEST_F(PtyTest, TermiosAffectsSlave) { // Verify this by setting ICRNL (which rewrites input \r to \n) and verify that // it has no effect on the master. TEST_F(PtyTest, MasterTermiosUnchangable) { + struct kernel_termios master_termios = {}; + EXPECT_THAT(ioctl(master_.get(), TCGETS, &master_termios), SyscallSucceeds()); + master_termios.c_lflag |= ICRNL; + EXPECT_THAT(ioctl(master_.get(), TCSETS, &master_termios), SyscallSucceeds()); + char c = '\r'; ASSERT_THAT(WriteFd(slave_.get(), &c, 1), SyscallSucceedsWithValue(1)); diff --git a/test/syscalls/linux/raw_socket.cc b/test/syscalls/linux/raw_socket.cc index 05c4ed03f..8d6e5c913 100644 --- a/test/syscalls/linux/raw_socket.cc +++ b/test/syscalls/linux/raw_socket.cc @@ -13,6 +13,9 @@ // limitations under the License. #include <linux/capability.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <netinet/in.h> #include <netinet/ip.h> #include <netinet/ip6.h> @@ -21,6 +24,7 @@ #include <sys/socket.h> #include <sys/types.h> #include <unistd.h> + #include <algorithm> #include "gtest/gtest.h" @@ -258,6 +262,27 @@ TEST_P(RawSocketTest, SendWithoutConnectFails) { SyscallFailsWithErrno(EDESTADDRREQ)); } +// Wildcard Bind. +TEST_P(RawSocketTest, BindToWildcard) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + struct sockaddr_storage addr; + addr = {}; + + // We don't set ports because raw sockets don't have a notion of ports. + if (Family() == AF_INET) { + struct sockaddr_in* sin = reinterpret_cast<struct sockaddr_in*>(&addr); + sin->sin_family = AF_INET; + sin->sin_addr.s_addr = htonl(INADDR_ANY); + } else { + struct sockaddr_in6* sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + sin6->sin6_family = AF_INET6; + sin6->sin6_addr = in6addr_any; + } + + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); +} + // Bind to localhost. TEST_P(RawSocketTest, BindToLocalhost) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); @@ -790,10 +815,30 @@ void RawSocketTest::ReceiveBufFrom(int sock, char* recv_buf, ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(sock, recv_buf, recv_buf_len)); } -INSTANTIATE_TEST_SUITE_P(AllInetTests, RawSocketTest, - ::testing::Combine( - ::testing::Values(IPPROTO_TCP, IPPROTO_UDP), - ::testing::Values(AF_INET, AF_INET6))); +#ifndef __fuchsia__ + +TEST_P(RawSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ // AF_INET6+SOCK_RAW+IPPROTO_RAW sockets can be created, but not written to. TEST(RawSocketTest, IPv6ProtoRaw) { @@ -813,6 +858,11 @@ TEST(RawSocketTest, IPv6ProtoRaw) { SyscallFailsWithErrno(EINVAL)); } +INSTANTIATE_TEST_SUITE_P( + AllInetTests, RawSocketTest, + ::testing::Combine(::testing::Values(IPPROTO_TCP, IPPROTO_UDP), + ::testing::Values(AF_INET, AF_INET6))); + } // namespace } // namespace testing diff --git a/test/syscalls/linux/raw_socket_hdrincl.cc b/test/syscalls/linux/raw_socket_hdrincl.cc index 0a27506aa..97f0467aa 100644 --- a/test/syscalls/linux/raw_socket_hdrincl.cc +++ b/test/syscalls/linux/raw_socket_hdrincl.cc @@ -167,7 +167,7 @@ TEST_F(RawHDRINCL, NotReadable) { // nothing to be read. char buf[117]; ASSERT_THAT(RetryEINTR(recv)(socket_, buf, sizeof(buf), MSG_DONTWAIT), - SyscallFailsWithErrno(EINVAL)); + SyscallFailsWithErrno(EAGAIN)); } // Test that we can connect() to a valid IP (loopback). @@ -178,6 +178,9 @@ TEST_F(RawHDRINCL, ConnectToLoopback) { } TEST_F(RawHDRINCL, SendWithoutConnectSucceeds) { + // FIXME(github.dev/issue/3159): Test currently flaky. + SKIP_IF(true); + struct iphdr hdr = LoopbackHeader(); ASSERT_THAT(send(socket_, &hdr, sizeof(hdr), 0), SyscallSucceedsWithValue(sizeof(hdr))); @@ -273,14 +276,17 @@ TEST_F(RawHDRINCL, SendAndReceive) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should be 0, as the packet is less than 68 bytes. - struct iphdr iphdr = {}; - memcpy(&iphdr, recv_buf, sizeof(iphdr)); - EXPECT_EQ(iphdr.id, 0); + // The packet ID should not be 0, as the packet has DF=0. + struct iphdr* iphdr = reinterpret_cast<struct iphdr*>(recv_buf); + EXPECT_NE(iphdr->id, 0); } -// Send and receive a packet with nonzero IP ID. -TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { +// Send and receive a packet where the sendto address is not the same as the +// provided destination. +TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { + // FIXME(github.dev/issue/3160): Test currently flaky. + SKIP_IF(true); + int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( @@ -292,19 +298,24 @@ TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); - // Construct a packet with an IP header, UDP header, and payload. Make the - // payload large enough to force an IP ID to be assigned. - constexpr char kPayload[128] = {}; + // Construct a packet with an IP header, UDP header, and payload. + constexpr char kPayload[] = "toto"; char packet[sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kPayload)]; ASSERT_TRUE( FillPacket(packet, sizeof(packet), port, kPayload, sizeof(kPayload))); + // Overwrite the IP destination address with an IP we can't get to. + struct iphdr iphdr = {}; + memcpy(&iphdr, packet, sizeof(iphdr)); + iphdr.daddr = 42; + memcpy(packet, &iphdr, sizeof(iphdr)); socklen_t addrlen = sizeof(addr_); ASSERT_NO_FATAL_FAILURE(sendto(socket_, &packet, sizeof(packet), 0, reinterpret_cast<struct sockaddr*>(&addr_), addrlen)); - // Receive the payload. + // Receive the payload, since sendto should replace the bad destination with + // localhost. char recv_buf[sizeof(packet)]; struct sockaddr_in src; socklen_t src_size = sizeof(src); @@ -318,47 +329,58 @@ TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should not be 0, as the packet was more than 68 bytes. - struct iphdr* iphdr = reinterpret_cast<struct iphdr*>(recv_buf); - EXPECT_NE(iphdr->id, 0); + // The packet ID should not be 0, as the packet has DF=0. + struct iphdr recv_iphdr = {}; + memcpy(&recv_iphdr, recv_buf, sizeof(recv_iphdr)); + EXPECT_NE(recv_iphdr.id, 0); + // The destination address should be localhost, not the bad IP we set + // initially. + EXPECT_EQ(absl::gbswap_32(recv_iphdr.daddr), INADDR_LOOPBACK); } -// Send and receive a packet where the sendto address is not the same as the -// provided destination. -TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { +// Send and receive a packet w/ the IP_HDRINCL option set. +TEST_F(RawHDRINCL, SendAndReceiveIPHdrIncl) { int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( PortAvailable(0, AddressFamily::kIpv4, SocketType::kUdp, false))); } - // IPPROTO_RAW sockets are write-only. We'll have to open another socket to - // read what we write. - FileDescriptor udp_sock = + FileDescriptor recv_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + + FileDescriptor send_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + // Enable IP_HDRINCL option so that we can build and send w/ an IP + // header. + constexpr int kSockOptOn = 1; + ASSERT_THAT(setsockopt(send_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + // This is not strictly required but we do it to make sure that setting + // IP_HDRINCL on a non IPPROTO_RAW socket does not prevent it from receiving + // packets. + ASSERT_THAT(setsockopt(recv_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + // Construct a packet with an IP header, UDP header, and payload. constexpr char kPayload[] = "toto"; char packet[sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kPayload)]; ASSERT_TRUE( FillPacket(packet, sizeof(packet), port, kPayload, sizeof(kPayload))); - // Overwrite the IP destination address with an IP we can't get to. - struct iphdr iphdr = {}; - memcpy(&iphdr, packet, sizeof(iphdr)); - iphdr.daddr = 42; - memcpy(packet, &iphdr, sizeof(iphdr)); socklen_t addrlen = sizeof(addr_); - ASSERT_NO_FATAL_FAILURE(sendto(socket_, &packet, sizeof(packet), 0, + ASSERT_NO_FATAL_FAILURE(sendto(send_sock.get(), &packet, sizeof(packet), 0, reinterpret_cast<struct sockaddr*>(&addr_), addrlen)); - // Receive the payload, since sendto should replace the bad destination with - // localhost. + // Receive the payload. char recv_buf[sizeof(packet)]; struct sockaddr_in src; socklen_t src_size = sizeof(src); - ASSERT_THAT(recvfrom(udp_sock.get(), recv_buf, sizeof(recv_buf), 0, + ASSERT_THAT(recvfrom(recv_sock.get(), recv_buf, sizeof(recv_buf), 0, reinterpret_cast<struct sockaddr*>(&src), &src_size), SyscallSucceedsWithValue(sizeof(packet))); EXPECT_EQ( @@ -368,13 +390,20 @@ TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should be 0, as the packet is less than 68 bytes. - struct iphdr recv_iphdr = {}; - memcpy(&recv_iphdr, recv_buf, sizeof(recv_iphdr)); - EXPECT_EQ(recv_iphdr.id, 0); - // The destination address should be localhost, not the bad IP we set - // initially. - EXPECT_EQ(absl::gbswap_32(recv_iphdr.daddr), INADDR_LOOPBACK); + struct iphdr iphdr = {}; + memcpy(&iphdr, recv_buf, sizeof(iphdr)); + EXPECT_NE(iphdr.id, 0); + + // Also verify that the packet we just sent was not delivered to the + // IPPROTO_RAW socket. + { + char recv_buf[sizeof(packet)]; + struct sockaddr_in src; + socklen_t src_size = sizeof(src); + ASSERT_THAT(recvfrom(socket_, recv_buf, sizeof(recv_buf), MSG_DONTWAIT, + reinterpret_cast<struct sockaddr*>(&src), &src_size), + SyscallFailsWithErrno(EAGAIN)); + } } } // namespace diff --git a/test/syscalls/linux/socket_netdevice.cc b/test/syscalls/linux/socket_netdevice.cc index 15d4b85a7..5f8d7f981 100644 --- a/test/syscalls/linux/socket_netdevice.cc +++ b/test/syscalls/linux/socket_netdevice.cc @@ -12,6 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include <linux/ethtool.h> #include <linux/netlink.h> #include <linux/rtnetlink.h> #include <linux/sockios.h> @@ -49,6 +50,7 @@ TEST(NetdeviceTest, Loopback) { // Check that the loopback is zero hardware address. ASSERT_THAT(ioctl(sock.get(), SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + EXPECT_EQ(ifr.ifr_hwaddr.sa_family, ARPHRD_LOOPBACK); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[0], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[1], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[2], 0); @@ -178,6 +180,27 @@ TEST(NetdeviceTest, InterfaceMTU) { EXPECT_GT(ifr.ifr_mtu, 0); } +TEST(NetdeviceTest, EthtoolGetTSInfo) { + FileDescriptor sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ethtool_ts_info tsi = {}; + tsi.cmd = ETHTOOL_GET_TS_INFO; // Get NIC's Timestamping capabilities. + + // Prepare the request. + struct ifreq ifr = {}; + snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); + ifr.ifr_data = (void*)&tsi; + + // Check that SIOCGIFMTU returns a nonzero MTU. + if (IsRunningOnGvisor()) { + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), + SyscallFailsWithErrno(EOPNOTSUPP)); + return; + } + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), SyscallSucceeds()); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index a4d2953e1..0cea7d11f 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -13,6 +13,9 @@ // limitations under the License. #include <fcntl.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <netinet/in.h> #include <netinet/tcp.h> #include <poll.h> @@ -1559,6 +1562,63 @@ TEST_P(SimpleTcpSocketTest, SetTCPWindowClampAboveHalfMinRcvBuf) { } } +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(SimpleTcpSocketTest, SetSocketAttachDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // Program generated using sudo tcpdump -i lo tcp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000006}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000006}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(SimpleTcpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(SimpleTcpSocketTest, GetSocketDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 9cc6be4fb..60c48ed6e 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -16,6 +16,9 @@ #include <arpa/inet.h> #include <fcntl.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <netinet/in.h> #include <poll.h> #include <sys/ioctl.h> @@ -1723,5 +1726,56 @@ TEST_P(UdpSocketTest, RecvBufLimits) { } } +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(UdpSocketTest, SetSocketDetachFilter) { + // Program generated using sudo tcpdump -i lo udp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000011}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000011}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(UdpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(UdpSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT( + getsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + } // namespace testing } // namespace gvisor diff --git a/test/util/fs_util.cc b/test/util/fs_util.cc index 052781445..5418948fe 100644 --- a/test/util/fs_util.cc +++ b/test/util/fs_util.cc @@ -125,12 +125,12 @@ PosixErrorOr<struct stat> Fstat(int fd) { PosixErrorOr<bool> Exists(absl::string_view path) { struct stat stat_buf; - int res = stat(std::string(path).c_str(), &stat_buf); + int res = lstat(std::string(path).c_str(), &stat_buf); if (res < 0) { if (errno == ENOENT) { return false; } - return PosixError(errno, absl::StrCat("stat ", path)); + return PosixError(errno, absl::StrCat("lstat ", path)); } return true; } diff --git a/test/util/fs_util.h b/test/util/fs_util.h index caf19b24d..8cdac23a1 100644 --- a/test/util/fs_util.h +++ b/test/util/fs_util.h @@ -44,9 +44,14 @@ PosixErrorOr<std::string> GetCWD(); // can't be determined. PosixErrorOr<bool> Exists(absl::string_view path); -// Returns a stat structure for the given path or an error. +// Returns a stat structure for the given path or an error. If the path +// represents a symlink, it will be traversed. PosixErrorOr<struct stat> Stat(absl::string_view path); +// Returns a stat structure for the given path or an error. If the path +// represents a symlink, it will not be traversed. +PosixErrorOr<struct stat> Lstat(absl::string_view path); + // Returns a stat struct for the given fd. PosixErrorOr<struct stat> Fstat(int fd); diff --git a/test/util/temp_path.cc b/test/util/temp_path.cc index 9c10b6674..e1bdee7fd 100644 --- a/test/util/temp_path.cc +++ b/test/util/temp_path.cc @@ -56,7 +56,7 @@ void TryDeleteRecursively(std::string const& path) { if (undeleted_dirs || undeleted_files || !status.ok()) { std::cerr << path << ": failed to delete " << undeleted_dirs << " directories and " << undeleted_files - << " files: " << status; + << " files: " << status << std::endl; } } } diff --git a/test/util/test_util.cc b/test/util/test_util.cc index 8a037f45f..d0c1d6426 100644 --- a/test/util/test_util.cc +++ b/test/util/test_util.cc @@ -42,6 +42,7 @@ namespace testing { constexpr char kGvisorNetwork[] = "GVISOR_NETWORK"; constexpr char kGvisorVfs[] = "GVISOR_VFS"; +constexpr char kFuseEnabled[] = "FUSE_ENABLED"; bool IsRunningOnGvisor() { return GvisorPlatform() != Platform::kNative; } @@ -68,6 +69,11 @@ bool IsRunningWithVFS1() { return strcmp(env, "VFS1") == 0; } +bool IsFUSEEnabled() { + const char* env = getenv(kFuseEnabled); + return env && strcmp(env, "TRUE") == 0; +} + // Inline cpuid instruction. Preserve %ebx/%rbx register. In PIC compilations // %ebx contains the address of the global offset table. %rbx is occasionally // used to address stack variables in presence of dynamic allocas. diff --git a/test/util/test_util.h b/test/util/test_util.h index 109078fc7..373c54f32 100644 --- a/test/util/test_util.h +++ b/test/util/test_util.h @@ -225,6 +225,7 @@ const std::string GvisorPlatform(); bool IsRunningWithHostinet(); // TODO(gvisor.dev/issue/1624): Delete once VFS1 is gone. bool IsRunningWithVFS1(); +bool IsFUSEEnabled(); #ifdef __linux__ void SetupGvisorDeathTest(); @@ -567,6 +568,25 @@ ssize_t ApplyFileIoSyscall(F const& f, size_t const count) { } // namespace internal +inline PosixErrorOr<std::string> ReadAllFd(int fd) { + std::string all; + all.reserve(128 * 1024); // arbitrary. + + std::vector<char> buffer(16 * 1024); + for (;;) { + auto const bytes = RetryEINTR(read)(fd, buffer.data(), buffer.size()); + if (bytes < 0) { + return PosixError(errno, "file read"); + } + if (bytes == 0) { + return std::move(all); + } + if (bytes > 0) { + all.append(buffer.data(), bytes); + } + } +} + inline ssize_t ReadFd(int fd, void* buf, size_t count) { return internal::ApplyFileIoSyscall( [&](size_t completed) { |