mirror of
https://github.com/moby/moby.git
synced 2022-11-09 12:21:53 -05:00
cc5c9013d9
This vendors in new spec/runc that supports
setting readonly and masked paths in the
configuration. Using this allows us to make an
exception for `—-privileged`.
Signed-off-by: Tonis Tiigi <tonistiigi@gmail.com>
(cherry picked from commit 3f81b49352
)
4310 lines
155 KiB
Go
4310 lines
155 KiB
Go
package main
|
|
|
|
import (
|
|
"bufio"
|
|
"bytes"
|
|
"fmt"
|
|
"io/ioutil"
|
|
"net"
|
|
"os"
|
|
"os/exec"
|
|
"path"
|
|
"path/filepath"
|
|
"reflect"
|
|
"regexp"
|
|
"sort"
|
|
"strconv"
|
|
"strings"
|
|
"sync"
|
|
"time"
|
|
|
|
"github.com/docker/docker/pkg/integration/checker"
|
|
"github.com/docker/docker/pkg/mount"
|
|
"github.com/docker/docker/pkg/stringutils"
|
|
"github.com/docker/docker/runconfig"
|
|
"github.com/docker/go-connections/nat"
|
|
"github.com/docker/libnetwork/netutils"
|
|
"github.com/docker/libnetwork/resolvconf"
|
|
"github.com/go-check/check"
|
|
)
|
|
|
|
// "test123" should be printed by docker run
|
|
func (s *DockerSuite) TestRunEchoStdout(c *check.C) {
|
|
out, _ := dockerCmd(c, "run", "busybox", "echo", "test123")
|
|
if out != "test123\n" {
|
|
c.Fatalf("container should've printed 'test123', got '%s'", out)
|
|
}
|
|
}
|
|
|
|
// "test" should be printed
|
|
func (s *DockerSuite) TestRunEchoNamedContainer(c *check.C) {
|
|
out, _ := dockerCmd(c, "run", "--name", "testfoonamedcontainer", "busybox", "echo", "test")
|
|
if out != "test\n" {
|
|
c.Errorf("container should've printed 'test'")
|
|
}
|
|
}
|
|
|
|
// docker run should not leak file descriptors. This test relies on Unix
|
|
// specific functionality and cannot run on Windows.
|
|
func (s *DockerSuite) TestRunLeakyFileDescriptors(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "busybox", "ls", "-C", "/proc/self/fd")
|
|
|
|
// normally, we should only get 0, 1, and 2, but 3 gets created by "ls" when it does "opendir" on the "fd" directory
|
|
if out != "0 1 2 3\n" {
|
|
c.Errorf("container should've printed '0 1 2 3', not: %s", out)
|
|
}
|
|
}
|
|
|
|
// it should be possible to lookup Google DNS
|
|
// this will fail when Internet access is unavailable
|
|
func (s *DockerSuite) TestRunLookupGoogleDns(c *check.C) {
|
|
testRequires(c, Network, NotArm)
|
|
image := DefaultImage
|
|
if daemonPlatform == "windows" {
|
|
// nslookup isn't present in Windows busybox. Is built-in.
|
|
image = WindowsBaseImage
|
|
}
|
|
dockerCmd(c, "run", image, "nslookup", "google.com")
|
|
}
|
|
|
|
// the exit code should be 0
|
|
func (s *DockerSuite) TestRunExitCodeZero(c *check.C) {
|
|
dockerCmd(c, "run", "busybox", "true")
|
|
}
|
|
|
|
// the exit code should be 1
|
|
func (s *DockerSuite) TestRunExitCodeOne(c *check.C) {
|
|
_, exitCode, err := dockerCmdWithError("run", "busybox", "false")
|
|
if err != nil && !strings.Contains("exit status 1", fmt.Sprintf("%s", err)) {
|
|
c.Fatal(err)
|
|
}
|
|
if exitCode != 1 {
|
|
c.Errorf("container should've exited with exit code 1. Got %d", exitCode)
|
|
}
|
|
}
|
|
|
|
// it should be possible to pipe in data via stdin to a process running in a container
|
|
func (s *DockerSuite) TestRunStdinPipe(c *check.C) {
|
|
// TODO Windows: This needs some work to make compatible.
|
|
testRequires(c, DaemonIsLinux)
|
|
runCmd := exec.Command(dockerBinary, "run", "-i", "-a", "stdin", "busybox", "cat")
|
|
runCmd.Stdin = strings.NewReader("blahblah")
|
|
out, _, _, err := runCommandWithStdoutStderr(runCmd)
|
|
if err != nil {
|
|
c.Fatalf("failed to run container: %v, output: %q", err, out)
|
|
}
|
|
|
|
out = strings.TrimSpace(out)
|
|
dockerCmd(c, "wait", out)
|
|
|
|
logsOut, _ := dockerCmd(c, "logs", out)
|
|
|
|
containerLogs := strings.TrimSpace(logsOut)
|
|
if containerLogs != "blahblah" {
|
|
c.Errorf("logs didn't print the container's logs %s", containerLogs)
|
|
}
|
|
|
|
dockerCmd(c, "rm", out)
|
|
}
|
|
|
|
// the container's ID should be printed when starting a container in detached mode
|
|
func (s *DockerSuite) TestRunDetachedContainerIDPrinting(c *check.C) {
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "true")
|
|
|
|
out = strings.TrimSpace(out)
|
|
dockerCmd(c, "wait", out)
|
|
|
|
rmOut, _ := dockerCmd(c, "rm", out)
|
|
|
|
rmOut = strings.TrimSpace(rmOut)
|
|
if rmOut != out {
|
|
c.Errorf("rm didn't print the container ID %s %s", out, rmOut)
|
|
}
|
|
}
|
|
|
|
// the working directory should be set correctly
|
|
func (s *DockerSuite) TestRunWorkingDirectory(c *check.C) {
|
|
// TODO Windows: There's a Windows bug stopping this from working.
|
|
testRequires(c, DaemonIsLinux)
|
|
dir := "/root"
|
|
image := "busybox"
|
|
if daemonPlatform == "windows" {
|
|
dir = `/windows`
|
|
image = WindowsBaseImage
|
|
}
|
|
|
|
// First with -w
|
|
out, _ := dockerCmd(c, "run", "-w", dir, image, "pwd")
|
|
out = strings.TrimSpace(out)
|
|
if out != dir {
|
|
c.Errorf("-w failed to set working directory")
|
|
}
|
|
|
|
// Then with --workdir
|
|
out, _ = dockerCmd(c, "run", "--workdir", dir, image, "pwd")
|
|
out = strings.TrimSpace(out)
|
|
if out != dir {
|
|
c.Errorf("--workdir failed to set working directory")
|
|
}
|
|
}
|
|
|
|
// pinging Google's DNS resolver should fail when we disable the networking
|
|
func (s *DockerSuite) TestRunWithoutNetworking(c *check.C) {
|
|
count := "-c"
|
|
image := "busybox"
|
|
if daemonPlatform == "windows" {
|
|
count = "-n"
|
|
image = WindowsBaseImage
|
|
}
|
|
|
|
// First using the long form --net
|
|
out, exitCode, err := dockerCmdWithError("run", "--net=none", image, "ping", count, "1", "8.8.8.8")
|
|
if err != nil && exitCode != 1 {
|
|
c.Fatal(out, err)
|
|
}
|
|
if exitCode != 1 {
|
|
c.Errorf("--net=none should've disabled the network; the container shouldn't have been able to ping 8.8.8.8")
|
|
}
|
|
}
|
|
|
|
//test --link use container name to link target
|
|
func (s *DockerSuite) TestRunLinksContainerWithContainerName(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as the networking
|
|
// settings are not populated back yet on inspect.
|
|
testRequires(c, DaemonIsLinux)
|
|
dockerCmd(c, "run", "-i", "-t", "-d", "--name", "parent", "busybox")
|
|
|
|
ip := inspectField(c, "parent", "NetworkSettings.Networks.bridge.IPAddress")
|
|
|
|
out, _ := dockerCmd(c, "run", "--link", "parent:test", "busybox", "/bin/cat", "/etc/hosts")
|
|
if !strings.Contains(out, ip+" test") {
|
|
c.Fatalf("use a container name to link target failed")
|
|
}
|
|
}
|
|
|
|
//test --link use container id to link target
|
|
func (s *DockerSuite) TestRunLinksContainerWithContainerId(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as the networking
|
|
// settings are not populated back yet on inspect.
|
|
testRequires(c, DaemonIsLinux)
|
|
cID, _ := dockerCmd(c, "run", "-i", "-t", "-d", "busybox")
|
|
|
|
cID = strings.TrimSpace(cID)
|
|
ip := inspectField(c, cID, "NetworkSettings.Networks.bridge.IPAddress")
|
|
|
|
out, _ := dockerCmd(c, "run", "--link", cID+":test", "busybox", "/bin/cat", "/etc/hosts")
|
|
if !strings.Contains(out, ip+" test") {
|
|
c.Fatalf("use a container id to link target failed")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestUserDefinedNetworkLinks(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "udlinkNet")
|
|
|
|
dockerCmd(c, "run", "-d", "--net=udlinkNet", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
// run a container in user-defined network udlinkNet with a link for an existing container
|
|
// and a link for a container that doesn't exist
|
|
dockerCmd(c, "run", "-d", "--net=udlinkNet", "--name=second", "--link=first:foo",
|
|
"--link=third:bar", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// ping to first and its alias foo must succeed
|
|
_, _, err := dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo")
|
|
c.Assert(err, check.IsNil)
|
|
|
|
// ping to third and its alias must fail
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "third")
|
|
c.Assert(err, check.NotNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "bar")
|
|
c.Assert(err, check.NotNil)
|
|
|
|
// start third container now
|
|
dockerCmd(c, "run", "-d", "--net=udlinkNet", "--name=third", "busybox", "top")
|
|
c.Assert(waitRun("third"), check.IsNil)
|
|
|
|
// ping to third and its alias must succeed now
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "third")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "bar")
|
|
c.Assert(err, check.IsNil)
|
|
}
|
|
|
|
func (s *DockerSuite) TestUserDefinedNetworkLinksWithRestart(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "udlinkNet")
|
|
|
|
dockerCmd(c, "run", "-d", "--net=udlinkNet", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
dockerCmd(c, "run", "-d", "--net=udlinkNet", "--name=second", "--link=first:foo",
|
|
"busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// ping to first and its alias foo must succeed
|
|
_, _, err := dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo")
|
|
c.Assert(err, check.IsNil)
|
|
|
|
// Restart first container
|
|
dockerCmd(c, "restart", "first")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
// ping to first and its alias foo must still succeed
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo")
|
|
c.Assert(err, check.IsNil)
|
|
|
|
// Restart second container
|
|
dockerCmd(c, "restart", "second")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// ping to first and its alias foo must still succeed
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo")
|
|
c.Assert(err, check.IsNil)
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithNetAliasOnDefaultNetworks(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
|
|
defaults := []string{"bridge", "host", "none"}
|
|
for _, net := range defaults {
|
|
out, _, err := dockerCmdWithError("run", "-d", "--net", net, "--net-alias", "alias_"+net, "busybox", "top")
|
|
c.Assert(err, checker.NotNil)
|
|
c.Assert(out, checker.Contains, runconfig.ErrUnsupportedNetworkAndAlias.Error())
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestUserDefinedNetworkAlias(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "net1")
|
|
|
|
dockerCmd(c, "run", "-d", "--net=net1", "--name=first", "--net-alias=foo1", "--net-alias=foo2", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
dockerCmd(c, "run", "-d", "--net=net1", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// ping to first and its network-scoped aliases
|
|
_, _, err := dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo1")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo2")
|
|
c.Assert(err, check.IsNil)
|
|
|
|
// Restart first container
|
|
dockerCmd(c, "restart", "first")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
// ping to first and its network-scoped aliases must succeed
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "first")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo1")
|
|
c.Assert(err, check.IsNil)
|
|
_, _, err = dockerCmdWithError("exec", "second", "ping", "-c", "1", "foo2")
|
|
c.Assert(err, check.IsNil)
|
|
}
|
|
|
|
// Issue 9677.
|
|
func (s *DockerSuite) TestRunWithDaemonFlags(c *check.C) {
|
|
out, _, err := dockerCmdWithError("--exec-opt", "foo=bar", "run", "-i", "busybox", "true")
|
|
if err != nil {
|
|
if !strings.Contains(out, "flag provided but not defined: --exec-opt") { // no daemon (client-only)
|
|
c.Fatal(err, out)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Regression test for #4979
|
|
func (s *DockerSuite) TestRunWithVolumesFromExited(c *check.C) {
|
|
|
|
var (
|
|
out string
|
|
exitCode int
|
|
)
|
|
|
|
// Create a file in a volume
|
|
if daemonPlatform == "windows" {
|
|
out, exitCode = dockerCmd(c, "run", "--name", "test-data", "--volume", `c:\some\dir`, WindowsBaseImage, `cmd /c echo hello > c:\some\dir\file`)
|
|
} else {
|
|
out, exitCode = dockerCmd(c, "run", "--name", "test-data", "--volume", "/some/dir", "busybox", "touch", "/some/dir/file")
|
|
}
|
|
if exitCode != 0 {
|
|
c.Fatal("1", out, exitCode)
|
|
}
|
|
|
|
// Read the file from another container using --volumes-from to access the volume in the second container
|
|
if daemonPlatform == "windows" {
|
|
out, exitCode = dockerCmd(c, "run", "--volumes-from", "test-data", WindowsBaseImage, `cmd /c type c:\some\dir\file`)
|
|
} else {
|
|
out, exitCode = dockerCmd(c, "run", "--volumes-from", "test-data", "busybox", "cat", "/some/dir/file")
|
|
}
|
|
if exitCode != 0 {
|
|
c.Fatal("2", out, exitCode)
|
|
}
|
|
}
|
|
|
|
// Volume path is a symlink which also exists on the host, and the host side is a file not a dir
|
|
// But the volume call is just a normal volume, not a bind mount
|
|
func (s *DockerSuite) TestRunCreateVolumesInSymlinkDir(c *check.C) {
|
|
var (
|
|
dockerFile string
|
|
containerPath string
|
|
cmd string
|
|
)
|
|
testRequires(c, SameHostDaemon)
|
|
name := "test-volume-symlink"
|
|
|
|
dir, err := ioutil.TempDir("", name)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(dir)
|
|
|
|
// In the case of Windows to Windows CI, if the machine is setup so that
|
|
// the temp directory is not the C: drive, this test is invalid and will
|
|
// not work.
|
|
if daemonPlatform == "windows" && strings.ToLower(dir[:1]) != "c" {
|
|
c.Skip("Requires TEMP to point to C: drive")
|
|
}
|
|
|
|
f, err := os.OpenFile(filepath.Join(dir, "test"), os.O_CREATE, 0700)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
f.Close()
|
|
|
|
if daemonPlatform == "windows" {
|
|
dockerFile = fmt.Sprintf("FROM %s\nRUN mkdir %s\nRUN mklink /D c:\\test %s", WindowsBaseImage, dir, dir)
|
|
containerPath = `c:\test\test`
|
|
cmd = "tasklist"
|
|
} else {
|
|
dockerFile = fmt.Sprintf("FROM busybox\nRUN mkdir -p %s\nRUN ln -s %s /test", dir, dir)
|
|
containerPath = "/test/test"
|
|
cmd = "true"
|
|
}
|
|
if _, err := buildImage(name, dockerFile, false); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-v", containerPath, name, cmd)
|
|
}
|
|
|
|
// Volume path is a symlink in the container
|
|
func (s *DockerSuite) TestRunCreateVolumesInSymlinkDir2(c *check.C) {
|
|
var (
|
|
dockerFile string
|
|
containerPath string
|
|
cmd string
|
|
)
|
|
testRequires(c, SameHostDaemon)
|
|
name := "test-volume-symlink2"
|
|
|
|
if daemonPlatform == "windows" {
|
|
dockerFile = fmt.Sprintf("FROM %s\nRUN mkdir c:\\%s\nRUN mklink /D c:\\test c:\\%s", WindowsBaseImage, name, name)
|
|
containerPath = `c:\test\test`
|
|
cmd = "tasklist"
|
|
} else {
|
|
dockerFile = fmt.Sprintf("FROM busybox\nRUN mkdir -p /%s\nRUN ln -s /%s /test", name, name)
|
|
containerPath = "/test/test"
|
|
cmd = "true"
|
|
}
|
|
if _, err := buildImage(name, dockerFile, false); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-v", containerPath, name, cmd)
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesMountedAsReadonly(c *check.C) {
|
|
// TODO Windows (Post TP4): This test cannot run on a Windows daemon as
|
|
// Windows does not support read-only bind mounts.
|
|
testRequires(c, DaemonIsLinux)
|
|
if _, code, err := dockerCmdWithError("run", "-v", "/test:/test:ro", "busybox", "touch", "/test/somefile"); err == nil || code == 0 {
|
|
c.Fatalf("run should fail because volume is ro: exit code %d", code)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesFromInReadonlyModeFails(c *check.C) {
|
|
// TODO Windows (Post TP4): This test cannot run on a Windows daemon as
|
|
// Windows does not support read-only bind mounts. Modified for when ro is supported.
|
|
testRequires(c, DaemonIsLinux)
|
|
var (
|
|
volumeDir string
|
|
fileInVol string
|
|
)
|
|
if daemonPlatform == "windows" {
|
|
volumeDir = `c:/test` // Forward-slash as using busybox
|
|
fileInVol = `c:/test/file`
|
|
} else {
|
|
testRequires(c, DaemonIsLinux)
|
|
volumeDir = "/test"
|
|
fileInVol = `/test/file`
|
|
}
|
|
dockerCmd(c, "run", "--name", "parent", "-v", volumeDir, "busybox", "true")
|
|
|
|
if _, code, err := dockerCmdWithError("run", "--volumes-from", "parent:ro", "busybox", "touch", fileInVol); err == nil || code == 0 {
|
|
c.Fatalf("run should fail because volume is ro: exit code %d", code)
|
|
}
|
|
}
|
|
|
|
// Regression test for #1201
|
|
func (s *DockerSuite) TestRunVolumesFromInReadWriteMode(c *check.C) {
|
|
var (
|
|
volumeDir string
|
|
fileInVol string
|
|
)
|
|
if daemonPlatform == "windows" {
|
|
volumeDir = `c:/test` // Forward-slash as using busybox
|
|
fileInVol = `c:/test/file`
|
|
} else {
|
|
volumeDir = "/test"
|
|
fileInVol = "/test/file"
|
|
}
|
|
|
|
dockerCmd(c, "run", "--name", "parent", "-v", volumeDir, "busybox", "true")
|
|
dockerCmd(c, "run", "--volumes-from", "parent:rw", "busybox", "touch", fileInVol)
|
|
|
|
if out, _, err := dockerCmdWithError("run", "--volumes-from", "parent:bar", "busybox", "touch", fileInVol); err == nil || !strings.Contains(out, `invalid mode: bar`) {
|
|
c.Fatalf("running --volumes-from parent:bar should have failed with invalid mode: %q", out)
|
|
}
|
|
|
|
dockerCmd(c, "run", "--volumes-from", "parent", "busybox", "touch", fileInVol)
|
|
}
|
|
|
|
func (s *DockerSuite) TestVolumesFromGetsProperMode(c *check.C) {
|
|
// TODO Windows: This test cannot yet run on a Windows daemon as Windows does
|
|
// not support read-only bind mounts as at TP4
|
|
testRequires(c, DaemonIsLinux)
|
|
dockerCmd(c, "run", "--name", "parent", "-v", "/test:/test:ro", "busybox", "true")
|
|
|
|
// Expect this "rw" mode to be be ignored since the inherited volume is "ro"
|
|
if _, _, err := dockerCmdWithError("run", "--volumes-from", "parent:rw", "busybox", "touch", "/test/file"); err == nil {
|
|
c.Fatal("Expected volumes-from to inherit read-only volume even when passing in `rw`")
|
|
}
|
|
|
|
dockerCmd(c, "run", "--name", "parent2", "-v", "/test:/test:ro", "busybox", "true")
|
|
|
|
// Expect this to be read-only since both are "ro"
|
|
if _, _, err := dockerCmdWithError("run", "--volumes-from", "parent2:ro", "busybox", "touch", "/test/file"); err == nil {
|
|
c.Fatal("Expected volumes-from to inherit read-only volume even when passing in `ro`")
|
|
}
|
|
}
|
|
|
|
// Test for GH#10618
|
|
func (s *DockerSuite) TestRunNoDupVolumes(c *check.C) {
|
|
path1 := randomTmpDirPath("test1", daemonPlatform)
|
|
path2 := randomTmpDirPath("test2", daemonPlatform)
|
|
|
|
someplace := ":/someplace"
|
|
if daemonPlatform == "windows" {
|
|
// Windows requires that the source directory exists before calling HCS
|
|
testRequires(c, SameHostDaemon)
|
|
someplace = `:c:\someplace`
|
|
if err := os.MkdirAll(path1, 0755); err != nil {
|
|
c.Fatalf("Failed to create %s: %q", path1, err)
|
|
}
|
|
defer os.RemoveAll(path1)
|
|
if err := os.MkdirAll(path2, 0755); err != nil {
|
|
c.Fatalf("Failed to create %s: %q", path1, err)
|
|
}
|
|
defer os.RemoveAll(path2)
|
|
}
|
|
mountstr1 := path1 + someplace
|
|
mountstr2 := path2 + someplace
|
|
|
|
if out, _, err := dockerCmdWithError("run", "-v", mountstr1, "-v", mountstr2, "busybox", "true"); err == nil {
|
|
c.Fatal("Expected error about duplicate mount definitions")
|
|
} else {
|
|
if !strings.Contains(out, "Duplicate mount point") {
|
|
c.Fatalf("Expected 'duplicate mount point' error, got %v", out)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Test for #1351
|
|
func (s *DockerSuite) TestRunApplyVolumesFromBeforeVolumes(c *check.C) {
|
|
prefix := ""
|
|
if daemonPlatform == "windows" {
|
|
prefix = `c:`
|
|
}
|
|
dockerCmd(c, "run", "--name", "parent", "-v", prefix+"/test", "busybox", "touch", prefix+"/test/foo")
|
|
dockerCmd(c, "run", "--volumes-from", "parent", "-v", prefix+"/test", "busybox", "cat", prefix+"/test/foo")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunMultipleVolumesFrom(c *check.C) {
|
|
prefix := ""
|
|
if daemonPlatform == "windows" {
|
|
prefix = `c:`
|
|
}
|
|
dockerCmd(c, "run", "--name", "parent1", "-v", prefix+"/test", "busybox", "touch", prefix+"/test/foo")
|
|
dockerCmd(c, "run", "--name", "parent2", "-v", prefix+"/other", "busybox", "touch", prefix+"/other/bar")
|
|
dockerCmd(c, "run", "--volumes-from", "parent1", "--volumes-from", "parent2", "busybox", "sh", "-c", "cat /test/foo && cat /other/bar")
|
|
}
|
|
|
|
// this tests verifies the ID format for the container
|
|
func (s *DockerSuite) TestRunVerifyContainerID(c *check.C) {
|
|
out, exit, err := dockerCmdWithError("run", "-d", "busybox", "true")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if exit != 0 {
|
|
c.Fatalf("expected exit code 0 received %d", exit)
|
|
}
|
|
|
|
match, err := regexp.MatchString("^[0-9a-f]{64}$", strings.TrimSuffix(out, "\n"))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if !match {
|
|
c.Fatalf("Invalid container ID: %s", out)
|
|
}
|
|
}
|
|
|
|
// Test that creating a container with a volume doesn't crash. Regression test for #995.
|
|
func (s *DockerSuite) TestRunCreateVolume(c *check.C) {
|
|
prefix := ""
|
|
if daemonPlatform == "windows" {
|
|
prefix = `c:`
|
|
}
|
|
dockerCmd(c, "run", "-v", prefix+"/var/lib/data", "busybox", "true")
|
|
}
|
|
|
|
// Test that creating a volume with a symlink in its path works correctly. Test for #5152.
|
|
// Note that this bug happens only with symlinks with a target that starts with '/'.
|
|
func (s *DockerSuite) TestRunCreateVolumeWithSymlink(c *check.C) {
|
|
// Cannot run on Windows as relies on Linux-specific functionality (sh -c mount...)
|
|
testRequires(c, DaemonIsLinux)
|
|
image := "docker-test-createvolumewithsymlink"
|
|
|
|
buildCmd := exec.Command(dockerBinary, "build", "-t", image, "-")
|
|
buildCmd.Stdin = strings.NewReader(`FROM busybox
|
|
RUN ln -s home /bar`)
|
|
buildCmd.Dir = workingDirectory
|
|
err := buildCmd.Run()
|
|
if err != nil {
|
|
c.Fatalf("could not build '%s': %v", image, err)
|
|
}
|
|
|
|
_, exitCode, err := dockerCmdWithError("run", "-v", "/bar/foo", "--name", "test-createvolumewithsymlink", image, "sh", "-c", "mount | grep -q /home/foo")
|
|
if err != nil || exitCode != 0 {
|
|
c.Fatalf("[run] err: %v, exitcode: %d", err, exitCode)
|
|
}
|
|
|
|
volPath, err := inspectMountSourceField("test-createvolumewithsymlink", "/bar/foo")
|
|
c.Assert(err, checker.IsNil)
|
|
|
|
_, exitCode, err = dockerCmdWithError("rm", "-v", "test-createvolumewithsymlink")
|
|
if err != nil || exitCode != 0 {
|
|
c.Fatalf("[rm] err: %v, exitcode: %d", err, exitCode)
|
|
}
|
|
|
|
_, err = os.Stat(volPath)
|
|
if !os.IsNotExist(err) {
|
|
c.Fatalf("[open] (expecting 'file does not exist' error) err: %v, volPath: %s", err, volPath)
|
|
}
|
|
}
|
|
|
|
// Tests that a volume path that has a symlink exists in a container mounting it with `--volumes-from`.
|
|
func (s *DockerSuite) TestRunVolumesFromSymlinkPath(c *check.C) {
|
|
name := "docker-test-volumesfromsymlinkpath"
|
|
prefix := ""
|
|
dfContents := `FROM busybox
|
|
RUN ln -s home /foo
|
|
VOLUME ["/foo/bar"]`
|
|
|
|
if daemonPlatform == "windows" {
|
|
prefix = `c:`
|
|
dfContents = `FROM ` + WindowsBaseImage + `
|
|
RUN mkdir c:\home
|
|
RUN mklink /D c:\foo c:\home
|
|
VOLUME ["c:/foo/bar"]
|
|
ENTRYPOINT c:\windows\system32\cmd.exe`
|
|
}
|
|
|
|
buildCmd := exec.Command(dockerBinary, "build", "-t", name, "-")
|
|
buildCmd.Stdin = strings.NewReader(dfContents)
|
|
buildCmd.Dir = workingDirectory
|
|
err := buildCmd.Run()
|
|
if err != nil {
|
|
c.Fatalf("could not build 'docker-test-volumesfromsymlinkpath': %v", err)
|
|
}
|
|
|
|
out, exitCode, err := dockerCmdWithError("run", "--name", "test-volumesfromsymlinkpath", name)
|
|
if err != nil || exitCode != 0 {
|
|
c.Fatalf("[run] (volume) err: %v, exitcode: %d, out: %s", err, exitCode, out)
|
|
}
|
|
|
|
_, exitCode, err = dockerCmdWithError("run", "--volumes-from", "test-volumesfromsymlinkpath", "busybox", "sh", "-c", "ls "+prefix+"/foo | grep -q bar")
|
|
if err != nil || exitCode != 0 {
|
|
c.Fatalf("[run] err: %v, exitcode: %d", err, exitCode)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunExitCode(c *check.C) {
|
|
var (
|
|
exit int
|
|
err error
|
|
)
|
|
|
|
_, exit, err = dockerCmdWithError("run", "busybox", "/bin/sh", "-c", "exit 72")
|
|
|
|
if err == nil {
|
|
c.Fatal("should not have a non nil error")
|
|
}
|
|
if exit != 72 {
|
|
c.Fatalf("expected exit code 72 received %d", exit)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserDefaults(c *check.C) {
|
|
expected := "uid=0(root) gid=0(root)"
|
|
if daemonPlatform == "windows" {
|
|
// TODO Windows: Remove this check once TP4 is no longer supported.
|
|
if windowsDaemonKV < 14250 {
|
|
expected = "uid=1000(SYSTEM) gid=1000(SYSTEM)"
|
|
} else {
|
|
expected = "uid=1000(ContainerAdministrator) gid=1000(ContainerAdministrator)"
|
|
}
|
|
}
|
|
out, _ := dockerCmd(c, "run", "busybox", "id")
|
|
if !strings.Contains(out, expected) {
|
|
c.Fatalf("expected '%s' got %s", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserByName(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-u", "root", "busybox", "id")
|
|
if !strings.Contains(out, "uid=0(root) gid=0(root)") {
|
|
c.Fatalf("expected root user got %s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserByID(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-u", "1", "busybox", "id")
|
|
if !strings.Contains(out, "uid=1(daemon) gid=1(daemon)") {
|
|
c.Fatalf("expected daemon user got %s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserByIDBig(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux, NotArm)
|
|
out, _, err := dockerCmdWithError("run", "-u", "2147483648", "busybox", "id")
|
|
if err == nil {
|
|
c.Fatal("No error, but must be.", out)
|
|
}
|
|
if !strings.Contains(out, "Uids and gids must be in range") {
|
|
c.Fatalf("expected error about uids range, got %s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserByIDNegative(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "-u", "-1", "busybox", "id")
|
|
if err == nil {
|
|
c.Fatal("No error, but must be.", out)
|
|
}
|
|
if !strings.Contains(out, "Uids and gids must be in range") {
|
|
c.Fatalf("expected error about uids range, got %s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserByIDZero(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "-u", "0", "busybox", "id")
|
|
if err != nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if !strings.Contains(out, "uid=0(root) gid=0(root) groups=10(wheel)") {
|
|
c.Fatalf("expected daemon user got %s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUserNotFound(c *check.C) {
|
|
// TODO Windows: This test cannot run on a Windows daemon as Windows does
|
|
// not support the use of -u
|
|
testRequires(c, DaemonIsLinux)
|
|
_, _, err := dockerCmdWithError("run", "-u", "notme", "busybox", "id")
|
|
if err == nil {
|
|
c.Fatal("unknown user should cause container to fail")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunTwoConcurrentContainers(c *check.C) {
|
|
// TODO Windows. There are two bugs in TP4 which means this test cannot
|
|
// be reliably enabled. The first is a race condition where sometimes
|
|
// HCS CreateComputeSystem() will fail "Invalid class string". #4985252 and
|
|
// #4493430.
|
|
//
|
|
// The second, which is seen more readily by increasing the number of concurrent
|
|
// containers to 5 or more, is that CSRSS hangs. This may fixed in the TP4 ZDP.
|
|
// #4898773.
|
|
testRequires(c, DaemonIsLinux)
|
|
sleepTime := "2"
|
|
if daemonPlatform == "windows" {
|
|
sleepTime = "5" // Make more reliable on Windows
|
|
}
|
|
group := sync.WaitGroup{}
|
|
group.Add(2)
|
|
|
|
errChan := make(chan error, 2)
|
|
for i := 0; i < 2; i++ {
|
|
go func() {
|
|
defer group.Done()
|
|
_, _, err := dockerCmdWithError("run", "busybox", "sleep", sleepTime)
|
|
errChan <- err
|
|
}()
|
|
}
|
|
|
|
group.Wait()
|
|
close(errChan)
|
|
|
|
for err := range errChan {
|
|
c.Assert(err, check.IsNil)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunEnvironment(c *check.C) {
|
|
// TODO Windows: Environment handling is different between Linux and
|
|
// Windows and this test relies currently on unix functionality.
|
|
testRequires(c, DaemonIsLinux)
|
|
cmd := exec.Command(dockerBinary, "run", "-h", "testing", "-e=FALSE=true", "-e=TRUE", "-e=TRICKY", "-e=HOME=", "busybox", "env")
|
|
cmd.Env = append(os.Environ(),
|
|
"TRUE=false",
|
|
"TRICKY=tri\ncky\n",
|
|
)
|
|
|
|
out, _, err := runCommandWithOutput(cmd)
|
|
if err != nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
|
|
actualEnv := strings.Split(strings.TrimSpace(out), "\n")
|
|
sort.Strings(actualEnv)
|
|
|
|
goodEnv := []string{
|
|
"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
|
|
"HOSTNAME=testing",
|
|
"FALSE=true",
|
|
"TRUE=false",
|
|
"TRICKY=tri",
|
|
"cky",
|
|
"",
|
|
"HOME=/root",
|
|
}
|
|
sort.Strings(goodEnv)
|
|
if len(goodEnv) != len(actualEnv) {
|
|
c.Fatalf("Wrong environment: should be %d variables, not: %q\n", len(goodEnv), strings.Join(actualEnv, ", "))
|
|
}
|
|
for i := range goodEnv {
|
|
if actualEnv[i] != goodEnv[i] {
|
|
c.Fatalf("Wrong environment variable: should be %s, not %s", goodEnv[i], actualEnv[i])
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunEnvironmentErase(c *check.C) {
|
|
// TODO Windows: Environment handling is different between Linux and
|
|
// Windows and this test relies currently on unix functionality.
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
// Test to make sure that when we use -e on env vars that are
|
|
// not set in our local env that they're removed (if present) in
|
|
// the container
|
|
|
|
cmd := exec.Command(dockerBinary, "run", "-e", "FOO", "-e", "HOSTNAME", "busybox", "env")
|
|
cmd.Env = appendBaseEnv(true)
|
|
|
|
out, _, err := runCommandWithOutput(cmd)
|
|
if err != nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
|
|
actualEnv := strings.Split(strings.TrimSpace(out), "\n")
|
|
sort.Strings(actualEnv)
|
|
|
|
goodEnv := []string{
|
|
"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
|
|
"HOME=/root",
|
|
}
|
|
sort.Strings(goodEnv)
|
|
if len(goodEnv) != len(actualEnv) {
|
|
c.Fatalf("Wrong environment: should be %d variables, not: %q\n", len(goodEnv), strings.Join(actualEnv, ", "))
|
|
}
|
|
for i := range goodEnv {
|
|
if actualEnv[i] != goodEnv[i] {
|
|
c.Fatalf("Wrong environment variable: should be %s, not %s", goodEnv[i], actualEnv[i])
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunEnvironmentOverride(c *check.C) {
|
|
// TODO Windows: Environment handling is different between Linux and
|
|
// Windows and this test relies currently on unix functionality.
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
// Test to make sure that when we use -e on env vars that are
|
|
// already in the env that we're overriding them
|
|
|
|
cmd := exec.Command(dockerBinary, "run", "-e", "HOSTNAME", "-e", "HOME=/root2", "busybox", "env")
|
|
cmd.Env = appendBaseEnv(true, "HOSTNAME=bar")
|
|
|
|
out, _, err := runCommandWithOutput(cmd)
|
|
if err != nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
|
|
actualEnv := strings.Split(strings.TrimSpace(out), "\n")
|
|
sort.Strings(actualEnv)
|
|
|
|
goodEnv := []string{
|
|
"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
|
|
"HOME=/root2",
|
|
"HOSTNAME=bar",
|
|
}
|
|
sort.Strings(goodEnv)
|
|
if len(goodEnv) != len(actualEnv) {
|
|
c.Fatalf("Wrong environment: should be %d variables, not: %q\n", len(goodEnv), strings.Join(actualEnv, ", "))
|
|
}
|
|
for i := range goodEnv {
|
|
if actualEnv[i] != goodEnv[i] {
|
|
c.Fatalf("Wrong environment variable: should be %s, not %s", goodEnv[i], actualEnv[i])
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerNetwork(c *check.C) {
|
|
if daemonPlatform == "windows" {
|
|
// Windows busybox does not have ping. Use built in ping instead.
|
|
dockerCmd(c, "run", WindowsBaseImage, "ping", "-n", "1", "127.0.0.1")
|
|
} else {
|
|
dockerCmd(c, "run", "busybox", "ping", "-c", "1", "127.0.0.1")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetHostNotAllowedWithLinks(c *check.C) {
|
|
// TODO Windows: This is Linux specific as --link is not supported and
|
|
// this will be deprecated in favor of container networking model.
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
dockerCmd(c, "run", "--name", "linked", "busybox", "true")
|
|
|
|
_, _, err := dockerCmdWithError("run", "--net=host", "--link", "linked:linked", "busybox", "true")
|
|
if err == nil {
|
|
c.Fatal("Expected error")
|
|
}
|
|
}
|
|
|
|
// #7851 hostname outside container shows FQDN, inside only shortname
|
|
// For testing purposes it is not required to set host's hostname directly
|
|
// and use "--net=host" (as the original issue submitter did), as the same
|
|
// codepath is executed with "docker run -h <hostname>". Both were manually
|
|
// tested, but this testcase takes the simpler path of using "run -h .."
|
|
func (s *DockerSuite) TestRunFullHostnameSet(c *check.C) {
|
|
// TODO Windows: -h is not yet functional.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-h", "foo.bar.baz", "busybox", "hostname")
|
|
if actual := strings.Trim(out, "\r\n"); actual != "foo.bar.baz" {
|
|
c.Fatalf("expected hostname 'foo.bar.baz', received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPrivilegedCanMknod(c *check.C) {
|
|
// Not applicable for Windows as Windows daemon does not support
|
|
// the concept of --privileged, and mknod is a Unix concept.
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "--privileged", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUnprivilegedCanMknod(c *check.C) {
|
|
// Not applicable for Windows as Windows daemon does not support
|
|
// the concept of --privileged, and mknod is a Unix concept.
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapDropInvalid(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-drop
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-drop=CHPASS", "busybox", "ls")
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapDropCannotMknod(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-drop or mknod
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-drop=MKNOD", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual == "ok" {
|
|
c.Fatalf("expected output not ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapDropCannotMknodLowerCase(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-drop or mknod
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-drop=mknod", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual == "ok" {
|
|
c.Fatalf("expected output not ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapDropALLCannotMknod(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-drop or mknod
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-drop=ALL", "--cap-add=SETGID", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual == "ok" {
|
|
c.Fatalf("expected output not ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapDropALLAddMknodCanMknod(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-drop or mknod
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "--cap-drop=ALL", "--cap-add=MKNOD", "--cap-add=SETGID", "busybox", "sh", "-c", "mknod /tmp/sda b 8 0 && echo ok")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddInvalid(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-add
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-add=CHPASS", "busybox", "ls")
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddCanDownInterface(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-add
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--cap-add=NET_ADMIN", "busybox", "sh", "-c", "ip link set eth0 down && echo ok")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddALLCanDownInterface(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-add
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--cap-add=ALL", "busybox", "sh", "-c", "ip link set eth0 down && echo ok")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddALLDropNetAdminCanDownInterface(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --cap-add
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "--cap-add=ALL", "--cap-drop=NET_ADMIN", "busybox", "sh", "-c", "ip link set eth0 down && echo ok")
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual == "ok" {
|
|
c.Fatalf("expected output not ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunGroupAdd(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --group-add
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--group-add=audio", "--group-add=staff", "--group-add=777", "busybox", "sh", "-c", "id")
|
|
|
|
groupsList := "uid=0(root) gid=0(root) groups=10(wheel),29(audio),50(staff),777"
|
|
if actual := strings.Trim(out, "\r\n"); actual != groupsList {
|
|
c.Fatalf("expected output %s received %s", groupsList, actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPrivilegedCanMount(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --privileged
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "--privileged", "busybox", "sh", "-c", "mount -t tmpfs none /tmp && echo ok")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUnprivilegedCannotMount(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of unprivileged
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, err := dockerCmdWithError("run", "busybox", "sh", "-c", "mount -t tmpfs none /tmp && echo ok")
|
|
|
|
if err == nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual == "ok" {
|
|
c.Fatalf("expected output not ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunSysNotWritableInNonPrivilegedContainers(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of unprivileged
|
|
testRequires(c, DaemonIsLinux, NotArm)
|
|
if _, code, err := dockerCmdWithError("run", "busybox", "touch", "/sys/kernel/profiling"); err == nil || code == 0 {
|
|
c.Fatal("sys should not be writable in a non privileged container")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunSysWritableInPrivilegedContainers(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of unprivileged
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
if _, code, err := dockerCmdWithError("run", "--privileged", "busybox", "touch", "/sys/kernel/profiling"); err != nil || code != 0 {
|
|
c.Fatalf("sys should be writable in privileged container")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunProcNotWritableInNonPrivilegedContainers(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of unprivileged
|
|
testRequires(c, DaemonIsLinux)
|
|
if _, code, err := dockerCmdWithError("run", "busybox", "touch", "/proc/sysrq-trigger"); err == nil || code == 0 {
|
|
c.Fatal("proc should not be writable in a non privileged container")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunProcWritableInPrivilegedContainers(c *check.C) {
|
|
// Not applicable for Windows as there is no concept of --privileged
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
if _, code := dockerCmd(c, "run", "--privileged", "busybox", "sh", "-c", "touch /proc/sysrq-trigger"); code != 0 {
|
|
c.Fatalf("proc should be writable in privileged container")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDeviceNumbers(c *check.C) {
|
|
// Not applicable on Windows as /dev/ is a Unix specific concept
|
|
// TODO: NotUserNamespace could be removed here if "root" "root" is replaced w user
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "busybox", "sh", "-c", "ls -l /dev/null")
|
|
deviceLineFields := strings.Fields(out)
|
|
deviceLineFields[6] = ""
|
|
deviceLineFields[7] = ""
|
|
deviceLineFields[8] = ""
|
|
expected := []string{"crw-rw-rw-", "1", "root", "root", "1,", "3", "", "", "", "/dev/null"}
|
|
|
|
if !(reflect.DeepEqual(deviceLineFields, expected)) {
|
|
c.Fatalf("expected output\ncrw-rw-rw- 1 root root 1, 3 May 24 13:29 /dev/null\n received\n %s\n", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunThatCharacterDevicesActLikeCharacterDevices(c *check.C) {
|
|
// Not applicable on Windows as /dev/ is a Unix specific concept
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "busybox", "sh", "-c", "dd if=/dev/zero of=/zero bs=1k count=5 2> /dev/null ; du -h /zero")
|
|
if actual := strings.Trim(out, "\r\n"); actual[0] == '0' {
|
|
c.Fatalf("expected a new file called /zero to be create that is greater than 0 bytes long, but du says: %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUnprivilegedWithChroot(c *check.C) {
|
|
// Not applicable on Windows as it does not support chroot
|
|
testRequires(c, DaemonIsLinux)
|
|
dockerCmd(c, "run", "busybox", "chroot", "/", "true")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAddingOptionalDevices(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support --device
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "--device", "/dev/zero:/dev/nulo", "busybox", "sh", "-c", "ls /dev/nulo")
|
|
if actual := strings.Trim(out, "\r\n"); actual != "/dev/nulo" {
|
|
c.Fatalf("expected output /dev/nulo, received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAddingOptionalDevicesNoSrc(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support --device
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _ := dockerCmd(c, "run", "--device", "/dev/zero:rw", "busybox", "sh", "-c", "ls /dev/zero")
|
|
if actual := strings.Trim(out, "\r\n"); actual != "/dev/zero" {
|
|
c.Fatalf("expected output /dev/zero, received %s", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAddingOptionalDevicesInvalidMode(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support --device
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
_, _, err := dockerCmdWithError("run", "--device", "/dev/zero:ro", "busybox", "sh", "-c", "ls /dev/zero")
|
|
if err == nil {
|
|
c.Fatalf("run container with device mode ro should fail")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeHostname(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support -h
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "run", "-h=testhostname", "busybox", "cat", "/etc/hostname")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "testhostname" {
|
|
c.Fatalf("expected 'testhostname', but says: %q", actual)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "--net=host", "busybox", "cat", "/etc/hostname")
|
|
|
|
hostname, err := os.Hostname()
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if actual := strings.Trim(out, "\r\n"); actual != hostname {
|
|
c.Fatalf("expected %q, but says: %q", hostname, actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunRootWorkdir(c *check.C) {
|
|
out, _ := dockerCmd(c, "run", "--workdir", "/", "busybox", "pwd")
|
|
expected := "/\n"
|
|
if daemonPlatform == "windows" {
|
|
expected = "C:" + expected
|
|
}
|
|
if out != expected {
|
|
c.Fatalf("pwd returned %q (expected %s)", s, expected)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAllowBindMountingRoot(c *check.C) {
|
|
if daemonPlatform == "windows" {
|
|
// Windows busybox will fail with Permission Denied on items such as pagefile.sys
|
|
dockerCmd(c, "run", "-v", `c:\:c:\host`, WindowsBaseImage, "cmd", "-c", "dir", `c:\host`)
|
|
} else {
|
|
dockerCmd(c, "run", "-v", "/:/host", "busybox", "ls", "/host")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDisallowBindMountingRootToRoot(c *check.C) {
|
|
mount := "/:/"
|
|
targetDir := "/host"
|
|
if daemonPlatform == "windows" {
|
|
mount = `c:\:c\`
|
|
targetDir = "c:/host" // Forward slash as using busybox
|
|
}
|
|
out, _, err := dockerCmdWithError("run", "-v", mount, "busybox", "ls", targetDir)
|
|
if err == nil {
|
|
c.Fatal(out, err)
|
|
}
|
|
}
|
|
|
|
// Verify that a container gets default DNS when only localhost resolvers exist
|
|
func (s *DockerSuite) TestRunDnsDefaultOptions(c *check.C) {
|
|
// Not applicable on Windows as this is testing Unix specific functionality
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
// preserve original resolv.conf for restoring after test
|
|
origResolvConf, err := ioutil.ReadFile("/etc/resolv.conf")
|
|
if os.IsNotExist(err) {
|
|
c.Fatalf("/etc/resolv.conf does not exist")
|
|
}
|
|
// defer restored original conf
|
|
defer func() {
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", origResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
}()
|
|
|
|
// test 3 cases: standard IPv4 localhost, commented out localhost, and IPv6 localhost
|
|
// 2 are removed from the file at container start, and the 3rd (commented out) one is ignored by
|
|
// GetNameservers(), leading to a replacement of nameservers with the default set
|
|
tmpResolvConf := []byte("nameserver 127.0.0.1\n#nameserver 127.0.2.1\nnameserver ::1")
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", tmpResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
actual, _ := dockerCmd(c, "run", "busybox", "cat", "/etc/resolv.conf")
|
|
// check that the actual defaults are appended to the commented out
|
|
// localhost resolver (which should be preserved)
|
|
// NOTE: if we ever change the defaults from google dns, this will break
|
|
expected := "#nameserver 127.0.2.1\n\nnameserver 8.8.8.8\nnameserver 8.8.4.4\n"
|
|
if actual != expected {
|
|
c.Fatalf("expected resolv.conf be: %q, but was: %q", expected, actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDnsOptions(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support --dns*, or
|
|
// the Unix-specific functionality of resolv.conf.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, stderr, _ := dockerCmdWithStdoutStderr(c, "run", "--dns=127.0.0.1", "--dns-search=mydomain", "--dns-opt=ndots:9", "busybox", "cat", "/etc/resolv.conf")
|
|
|
|
// The client will get a warning on stderr when setting DNS to a localhost address; verify this:
|
|
if !strings.Contains(stderr, "Localhost DNS setting") {
|
|
c.Fatalf("Expected warning on stderr about localhost resolver, but got %q", stderr)
|
|
}
|
|
|
|
actual := strings.Replace(strings.Trim(out, "\r\n"), "\n", " ", -1)
|
|
if actual != "search mydomain nameserver 127.0.0.1 options ndots:9" {
|
|
c.Fatalf("expected 'search mydomain nameserver 127.0.0.1 options ndots:9', but says: %q", actual)
|
|
}
|
|
|
|
out, stderr, _ = dockerCmdWithStdoutStderr(c, "run", "--dns=127.0.0.1", "--dns-search=.", "--dns-opt=ndots:3", "busybox", "cat", "/etc/resolv.conf")
|
|
|
|
actual = strings.Replace(strings.Trim(strings.Trim(out, "\r\n"), " "), "\n", " ", -1)
|
|
if actual != "nameserver 127.0.0.1 options ndots:3" {
|
|
c.Fatalf("expected 'nameserver 127.0.0.1 options ndots:3', but says: %q", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDnsRepeatOptions(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _, _ := dockerCmdWithStdoutStderr(c, "run", "--dns=1.1.1.1", "--dns=2.2.2.2", "--dns-search=mydomain", "--dns-search=mydomain2", "--dns-opt=ndots:9", "--dns-opt=timeout:3", "busybox", "cat", "/etc/resolv.conf")
|
|
|
|
actual := strings.Replace(strings.Trim(out, "\r\n"), "\n", " ", -1)
|
|
if actual != "search mydomain mydomain2 nameserver 1.1.1.1 nameserver 2.2.2.2 options ndots:9 timeout:3" {
|
|
c.Fatalf("expected 'search mydomain mydomain2 nameserver 1.1.1.1 nameserver 2.2.2.2 options ndots:9 timeout:3', but says: %q", actual)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDnsOptionsBasedOnHostResolvConf(c *check.C) {
|
|
// Not applicable on Windows as testing Unix specific functionality
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
origResolvConf, err := ioutil.ReadFile("/etc/resolv.conf")
|
|
if os.IsNotExist(err) {
|
|
c.Fatalf("/etc/resolv.conf does not exist")
|
|
}
|
|
|
|
hostNamservers := resolvconf.GetNameservers(origResolvConf, netutils.IP)
|
|
hostSearch := resolvconf.GetSearchDomains(origResolvConf)
|
|
|
|
var out string
|
|
out, _ = dockerCmd(c, "run", "--dns=127.0.0.1", "busybox", "cat", "/etc/resolv.conf")
|
|
|
|
if actualNameservers := resolvconf.GetNameservers([]byte(out), netutils.IP); string(actualNameservers[0]) != "127.0.0.1" {
|
|
c.Fatalf("expected '127.0.0.1', but says: %q", string(actualNameservers[0]))
|
|
}
|
|
|
|
actualSearch := resolvconf.GetSearchDomains([]byte(out))
|
|
if len(actualSearch) != len(hostSearch) {
|
|
c.Fatalf("expected %q search domain(s), but it has: %q", len(hostSearch), len(actualSearch))
|
|
}
|
|
for i := range actualSearch {
|
|
if actualSearch[i] != hostSearch[i] {
|
|
c.Fatalf("expected %q domain, but says: %q", actualSearch[i], hostSearch[i])
|
|
}
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "--dns-search=mydomain", "busybox", "cat", "/etc/resolv.conf")
|
|
|
|
actualNameservers := resolvconf.GetNameservers([]byte(out), netutils.IP)
|
|
if len(actualNameservers) != len(hostNamservers) {
|
|
c.Fatalf("expected %q nameserver(s), but it has: %q", len(hostNamservers), len(actualNameservers))
|
|
}
|
|
for i := range actualNameservers {
|
|
if actualNameservers[i] != hostNamservers[i] {
|
|
c.Fatalf("expected %q nameserver, but says: %q", actualNameservers[i], hostNamservers[i])
|
|
}
|
|
}
|
|
|
|
if actualSearch = resolvconf.GetSearchDomains([]byte(out)); string(actualSearch[0]) != "mydomain" {
|
|
c.Fatalf("expected 'mydomain', but says: %q", string(actualSearch[0]))
|
|
}
|
|
|
|
// test with file
|
|
tmpResolvConf := []byte("search example.com\nnameserver 12.34.56.78\nnameserver 127.0.0.1")
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", tmpResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
// put the old resolvconf back
|
|
defer func() {
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", origResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
}()
|
|
|
|
resolvConf, err := ioutil.ReadFile("/etc/resolv.conf")
|
|
if os.IsNotExist(err) {
|
|
c.Fatalf("/etc/resolv.conf does not exist")
|
|
}
|
|
|
|
hostNamservers = resolvconf.GetNameservers(resolvConf, netutils.IP)
|
|
hostSearch = resolvconf.GetSearchDomains(resolvConf)
|
|
|
|
out, _ = dockerCmd(c, "run", "busybox", "cat", "/etc/resolv.conf")
|
|
if actualNameservers = resolvconf.GetNameservers([]byte(out), netutils.IP); string(actualNameservers[0]) != "12.34.56.78" || len(actualNameservers) != 1 {
|
|
c.Fatalf("expected '12.34.56.78', but has: %v", actualNameservers)
|
|
}
|
|
|
|
actualSearch = resolvconf.GetSearchDomains([]byte(out))
|
|
if len(actualSearch) != len(hostSearch) {
|
|
c.Fatalf("expected %q search domain(s), but it has: %q", len(hostSearch), len(actualSearch))
|
|
}
|
|
for i := range actualSearch {
|
|
if actualSearch[i] != hostSearch[i] {
|
|
c.Fatalf("expected %q domain, but says: %q", actualSearch[i], hostSearch[i])
|
|
}
|
|
}
|
|
}
|
|
|
|
// Test to see if a non-root user can resolve a DNS name. Also
|
|
// check if the container resolv.conf file has at least 0644 perm.
|
|
func (s *DockerSuite) TestRunNonRootUserResolvName(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support --user
|
|
testRequires(c, SameHostDaemon, Network, DaemonIsLinux, NotArm)
|
|
|
|
dockerCmd(c, "run", "--name=testperm", "--user=nobody", "busybox", "nslookup", "apt.dockerproject.org")
|
|
|
|
cID, err := getIDByName("testperm")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
fmode := (os.FileMode)(0644)
|
|
finfo, err := os.Stat(containerStorageFile(cID, "resolv.conf"))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if (finfo.Mode() & fmode) != fmode {
|
|
c.Fatalf("Expected container resolv.conf mode to be at least %s, instead got %s", fmode.String(), finfo.Mode().String())
|
|
}
|
|
}
|
|
|
|
// Test if container resolv.conf gets updated the next time it restarts
|
|
// if host /etc/resolv.conf has changed. This only applies if the container
|
|
// uses the host's /etc/resolv.conf and does not have any dns options provided.
|
|
func (s *DockerSuite) TestRunResolvconfUpdate(c *check.C) {
|
|
// Not applicable on Windows as testing unix specific functionality
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
tmpResolvConf := []byte("search pommesfrites.fr\nnameserver 12.34.56.78\n")
|
|
tmpLocalhostResolvConf := []byte("nameserver 127.0.0.1")
|
|
|
|
//take a copy of resolv.conf for restoring after test completes
|
|
resolvConfSystem, err := ioutil.ReadFile("/etc/resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// This test case is meant to test monitoring resolv.conf when it is
|
|
// a regular file not a bind mounc. So we unmount resolv.conf and replace
|
|
// it with a file containing the original settings.
|
|
cmd := exec.Command("umount", "/etc/resolv.conf")
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
//cleanup
|
|
defer func() {
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", resolvConfSystem, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
}()
|
|
|
|
//1. test that a restarting container gets an updated resolv.conf
|
|
dockerCmd(c, "run", "--name='first'", "busybox", "true")
|
|
containerID1, err := getIDByName("first")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// replace resolv.conf with our temporary copy
|
|
bytesResolvConf := []byte(tmpResolvConf)
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", bytesResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// start the container again to pickup changes
|
|
dockerCmd(c, "start", "first")
|
|
|
|
// check for update in container
|
|
containerResolv, err := readContainerFile(containerID1, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if !bytes.Equal(containerResolv, bytesResolvConf) {
|
|
c.Fatalf("Restarted container does not have updated resolv.conf; expected %q, got %q", tmpResolvConf, string(containerResolv))
|
|
}
|
|
|
|
/* //make a change to resolv.conf (in this case replacing our tmp copy with orig copy)
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", resolvConfSystem, 0644); err != nil {
|
|
c.Fatal(err)
|
|
} */
|
|
//2. test that a restarting container does not receive resolv.conf updates
|
|
// if it modified the container copy of the starting point resolv.conf
|
|
dockerCmd(c, "run", "--name='second'", "busybox", "sh", "-c", "echo 'search mylittlepony.com' >>/etc/resolv.conf")
|
|
containerID2, err := getIDByName("second")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
//make a change to resolv.conf (in this case replacing our tmp copy with orig copy)
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", resolvConfSystem, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// start the container again
|
|
dockerCmd(c, "start", "second")
|
|
|
|
// check for update in container
|
|
containerResolv, err = readContainerFile(containerID2, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if bytes.Equal(containerResolv, resolvConfSystem) {
|
|
c.Fatalf("Container's resolv.conf should not have been updated with host resolv.conf: %q", string(containerResolv))
|
|
}
|
|
|
|
//3. test that a running container's resolv.conf is not modified while running
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "top")
|
|
runningContainerID := strings.TrimSpace(out)
|
|
|
|
// replace resolv.conf
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", bytesResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// check for update in container
|
|
containerResolv, err = readContainerFile(runningContainerID, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if bytes.Equal(containerResolv, bytesResolvConf) {
|
|
c.Fatalf("Running container should not have updated resolv.conf; expected %q, got %q", string(resolvConfSystem), string(containerResolv))
|
|
}
|
|
|
|
//4. test that a running container's resolv.conf is updated upon restart
|
|
// (the above container is still running..)
|
|
dockerCmd(c, "restart", runningContainerID)
|
|
|
|
// check for update in container
|
|
containerResolv, err = readContainerFile(runningContainerID, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if !bytes.Equal(containerResolv, bytesResolvConf) {
|
|
c.Fatalf("Restarted container should have updated resolv.conf; expected %q, got %q", string(bytesResolvConf), string(containerResolv))
|
|
}
|
|
|
|
//5. test that additions of a localhost resolver are cleaned from
|
|
// host resolv.conf before updating container's resolv.conf copies
|
|
|
|
// replace resolv.conf with a localhost-only nameserver copy
|
|
bytesResolvConf = []byte(tmpLocalhostResolvConf)
|
|
if err = ioutil.WriteFile("/etc/resolv.conf", bytesResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// start the container again to pickup changes
|
|
dockerCmd(c, "start", "first")
|
|
|
|
// our first exited container ID should have been updated, but with default DNS
|
|
// after the cleanup of resolv.conf found only a localhost nameserver:
|
|
containerResolv, err = readContainerFile(containerID1, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
expected := "\nnameserver 8.8.8.8\nnameserver 8.8.4.4\n"
|
|
if !bytes.Equal(containerResolv, []byte(expected)) {
|
|
c.Fatalf("Container does not have cleaned/replaced DNS in resolv.conf; expected %q, got %q", expected, string(containerResolv))
|
|
}
|
|
|
|
//6. Test that replacing (as opposed to modifying) resolv.conf triggers an update
|
|
// of containers' resolv.conf.
|
|
|
|
// Restore the original resolv.conf
|
|
if err := ioutil.WriteFile("/etc/resolv.conf", resolvConfSystem, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Run the container so it picks up the old settings
|
|
dockerCmd(c, "run", "--name='third'", "busybox", "true")
|
|
containerID3, err := getIDByName("third")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Create a modified resolv.conf.aside and override resolv.conf with it
|
|
bytesResolvConf = []byte(tmpResolvConf)
|
|
if err := ioutil.WriteFile("/etc/resolv.conf.aside", bytesResolvConf, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
err = os.Rename("/etc/resolv.conf.aside", "/etc/resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// start the container again to pickup changes
|
|
dockerCmd(c, "start", "third")
|
|
|
|
// check for update in container
|
|
containerResolv, err = readContainerFile(containerID3, "resolv.conf")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if !bytes.Equal(containerResolv, bytesResolvConf) {
|
|
c.Fatalf("Stopped container does not have updated resolv.conf; expected\n%q\n got\n%q", tmpResolvConf, string(containerResolv))
|
|
}
|
|
|
|
//cleanup, restore original resolv.conf happens in defer func()
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAddHost(c *check.C) {
|
|
// Not applicable on Windows as it does not support --add-host
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--add-host=extra:86.75.30.9", "busybox", "grep", "extra", "/etc/hosts")
|
|
|
|
actual := strings.Trim(out, "\r\n")
|
|
if actual != "86.75.30.9\textra" {
|
|
c.Fatalf("expected '86.75.30.9\textra', but says: %q", actual)
|
|
}
|
|
}
|
|
|
|
// Regression test for #6983
|
|
func (s *DockerSuite) TestRunAttachStdErrOnlyTTYMode(c *check.C) {
|
|
_, exitCode := dockerCmd(c, "run", "-t", "-a", "stderr", "busybox", "true")
|
|
if exitCode != 0 {
|
|
c.Fatalf("Container should have exited with error code 0")
|
|
}
|
|
}
|
|
|
|
// Regression test for #6983
|
|
func (s *DockerSuite) TestRunAttachStdOutOnlyTTYMode(c *check.C) {
|
|
_, exitCode := dockerCmd(c, "run", "-t", "-a", "stdout", "busybox", "true")
|
|
if exitCode != 0 {
|
|
c.Fatalf("Container should have exited with error code 0")
|
|
}
|
|
}
|
|
|
|
// Regression test for #6983
|
|
func (s *DockerSuite) TestRunAttachStdOutAndErrTTYMode(c *check.C) {
|
|
_, exitCode := dockerCmd(c, "run", "-t", "-a", "stdout", "-a", "stderr", "busybox", "true")
|
|
if exitCode != 0 {
|
|
c.Fatalf("Container should have exited with error code 0")
|
|
}
|
|
}
|
|
|
|
// Test for #10388 - this will run the same test as TestRunAttachStdOutAndErrTTYMode
|
|
// but using --attach instead of -a to make sure we read the flag correctly
|
|
func (s *DockerSuite) TestRunAttachWithDetach(c *check.C) {
|
|
cmd := exec.Command(dockerBinary, "run", "-d", "--attach", "stdout", "busybox", "true")
|
|
_, stderr, _, err := runCommandWithStdoutStderr(cmd)
|
|
if err == nil {
|
|
c.Fatal("Container should have exited with error code different than 0")
|
|
} else if !strings.Contains(stderr, "Conflicting options: -a and -d") {
|
|
c.Fatal("Should have been returned an error with conflicting options -a and -d")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunState(c *check.C) {
|
|
// TODO Windows: This needs some rework as Windows busybox does not support top
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
state := inspectField(c, id, "State.Running")
|
|
if state != "true" {
|
|
c.Fatal("Container state is 'not running'")
|
|
}
|
|
pid1 := inspectField(c, id, "State.Pid")
|
|
if pid1 == "0" {
|
|
c.Fatal("Container state Pid 0")
|
|
}
|
|
|
|
dockerCmd(c, "stop", id)
|
|
state = inspectField(c, id, "State.Running")
|
|
if state != "false" {
|
|
c.Fatal("Container state is 'running'")
|
|
}
|
|
pid2 := inspectField(c, id, "State.Pid")
|
|
if pid2 == pid1 {
|
|
c.Fatalf("Container state Pid %s, but expected %s", pid2, pid1)
|
|
}
|
|
|
|
dockerCmd(c, "start", id)
|
|
state = inspectField(c, id, "State.Running")
|
|
if state != "true" {
|
|
c.Fatal("Container state is 'not running'")
|
|
}
|
|
pid3 := inspectField(c, id, "State.Pid")
|
|
if pid3 == pid1 {
|
|
c.Fatalf("Container state Pid %s, but expected %s", pid2, pid1)
|
|
}
|
|
}
|
|
|
|
// Test for #1737
|
|
func (s *DockerSuite) TestRunCopyVolumeUidGid(c *check.C) {
|
|
// Not applicable on Windows as it does not support uid or gid in this way
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "testrunvolumesuidgid"
|
|
_, err := buildImage(name,
|
|
`FROM busybox
|
|
RUN echo 'dockerio:x:1001:1001::/bin:/bin/false' >> /etc/passwd
|
|
RUN echo 'dockerio:x:1001:' >> /etc/group
|
|
RUN mkdir -p /hello && touch /hello/test && chown dockerio.dockerio /hello`,
|
|
true)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Test that the uid and gid is copied from the image to the volume
|
|
out, _ := dockerCmd(c, "run", "--rm", "-v", "/hello", name, "sh", "-c", "ls -l / | grep hello | awk '{print $3\":\"$4}'")
|
|
out = strings.TrimSpace(out)
|
|
if out != "dockerio:dockerio" {
|
|
c.Fatalf("Wrong /hello ownership: %s, expected dockerio:dockerio", out)
|
|
}
|
|
}
|
|
|
|
// Test for #1582
|
|
func (s *DockerSuite) TestRunCopyVolumeContent(c *check.C) {
|
|
// TODO Windows, post TP4. Windows does not yet support volume functionality
|
|
// that copies from the image to the volume.
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "testruncopyvolumecontent"
|
|
_, err := buildImage(name,
|
|
`FROM busybox
|
|
RUN mkdir -p /hello/local && echo hello > /hello/local/world`,
|
|
true)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Test that the content is copied from the image to the volume
|
|
out, _ := dockerCmd(c, "run", "--rm", "-v", "/hello", name, "find", "/hello")
|
|
if !(strings.Contains(out, "/hello/local/world") && strings.Contains(out, "/hello/local")) {
|
|
c.Fatal("Container failed to transfer content to volume")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCleanupCmdOnEntrypoint(c *check.C) {
|
|
name := "testrunmdcleanuponentrypoint"
|
|
if _, err := buildImage(name,
|
|
`FROM busybox
|
|
ENTRYPOINT ["echo"]
|
|
CMD ["testingpoint"]`,
|
|
true); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, exit := dockerCmd(c, "run", "--entrypoint", "whoami", name)
|
|
if exit != 0 {
|
|
c.Fatalf("expected exit code 0 received %d, out: %q", exit, out)
|
|
}
|
|
out = strings.TrimSpace(out)
|
|
expected := "root"
|
|
if daemonPlatform == "windows" {
|
|
// TODO Windows: Remove this check once TP4 is no longer supported.
|
|
if windowsDaemonKV < 14250 {
|
|
expected = `nt authority\system`
|
|
} else {
|
|
expected = `user manager\containeradministrator`
|
|
}
|
|
}
|
|
if out != expected {
|
|
c.Fatalf("Expected output %s, got %q", expected, out)
|
|
}
|
|
}
|
|
|
|
// TestRunWorkdirExistsAndIsFile checks that if 'docker run -w' with existing file can be detected
|
|
func (s *DockerSuite) TestRunWorkdirExistsAndIsFile(c *check.C) {
|
|
existingFile := "/bin/cat"
|
|
expected := "not a directory"
|
|
if daemonPlatform == "windows" {
|
|
existingFile = `\windows\system32\ntdll.dll`
|
|
expected = `Cannot mkdir: \windows\system32\ntdll.dll is not a directory.`
|
|
}
|
|
|
|
out, exitCode, err := dockerCmdWithError("run", "-w", existingFile, "busybox")
|
|
if !(err != nil && exitCode == 125 && strings.Contains(out, expected)) {
|
|
c.Fatalf("Existing binary as a directory should error out with exitCode 125; we got: %s, exitCode: %d", out, exitCode)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunExitOnStdinClose(c *check.C) {
|
|
name := "testrunexitonstdinclose"
|
|
|
|
meow := "/bin/cat"
|
|
delay := 1
|
|
if daemonPlatform == "windows" {
|
|
meow = "cat"
|
|
delay = 60
|
|
}
|
|
runCmd := exec.Command(dockerBinary, "run", "--name", name, "-i", "busybox", meow)
|
|
|
|
stdin, err := runCmd.StdinPipe()
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
stdout, err := runCmd.StdoutPipe()
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if err := runCmd.Start(); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if _, err := stdin.Write([]byte("hello\n")); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
r := bufio.NewReader(stdout)
|
|
line, err := r.ReadString('\n')
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
line = strings.TrimSpace(line)
|
|
if line != "hello" {
|
|
c.Fatalf("Output should be 'hello', got '%q'", line)
|
|
}
|
|
if err := stdin.Close(); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
finish := make(chan error)
|
|
go func() {
|
|
finish <- runCmd.Wait()
|
|
close(finish)
|
|
}()
|
|
select {
|
|
case err := <-finish:
|
|
c.Assert(err, check.IsNil)
|
|
case <-time.After(time.Duration(delay) * time.Second):
|
|
c.Fatal("docker run failed to exit on stdin close")
|
|
}
|
|
state := inspectField(c, name, "State.Running")
|
|
|
|
if state != "false" {
|
|
c.Fatal("Container must be stopped after stdin closing")
|
|
}
|
|
}
|
|
|
|
// Test for #2267
|
|
func (s *DockerSuite) TestRunWriteHostsFileAndNotCommit(c *check.C) {
|
|
// Cannot run on Windows as Windows does not support diff.
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "writehosts"
|
|
out, _ := dockerCmd(c, "run", "--name", name, "busybox", "sh", "-c", "echo test2267 >> /etc/hosts && cat /etc/hosts")
|
|
if !strings.Contains(out, "test2267") {
|
|
c.Fatal("/etc/hosts should contain 'test2267'")
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "diff", name)
|
|
if len(strings.Trim(out, "\r\n")) != 0 && !eqToBaseDiff(out, c) {
|
|
c.Fatal("diff should be empty")
|
|
}
|
|
}
|
|
|
|
func eqToBaseDiff(out string, c *check.C) bool {
|
|
name := "eqToBaseDiff" + stringutils.GenerateRandomAlphaOnlyString(32)
|
|
dockerCmd(c, "run", "--name", name, "busybox", "echo", "hello")
|
|
cID, err := getIDByName(name)
|
|
c.Assert(err, check.IsNil)
|
|
|
|
baseDiff, _ := dockerCmd(c, "diff", cID)
|
|
baseArr := strings.Split(baseDiff, "\n")
|
|
sort.Strings(baseArr)
|
|
outArr := strings.Split(out, "\n")
|
|
sort.Strings(outArr)
|
|
return sliceEq(baseArr, outArr)
|
|
}
|
|
|
|
func sliceEq(a, b []string) bool {
|
|
if len(a) != len(b) {
|
|
return false
|
|
}
|
|
|
|
for i := range a {
|
|
if a[i] != b[i] {
|
|
return false
|
|
}
|
|
}
|
|
|
|
return true
|
|
}
|
|
|
|
// Test for #2267
|
|
func (s *DockerSuite) TestRunWriteHostnameFileAndNotCommit(c *check.C) {
|
|
// Cannot run on Windows as Windows does not support diff.
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "writehostname"
|
|
out, _ := dockerCmd(c, "run", "--name", name, "busybox", "sh", "-c", "echo test2267 >> /etc/hostname && cat /etc/hostname")
|
|
if !strings.Contains(out, "test2267") {
|
|
c.Fatal("/etc/hostname should contain 'test2267'")
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "diff", name)
|
|
if len(strings.Trim(out, "\r\n")) != 0 && !eqToBaseDiff(out, c) {
|
|
c.Fatal("diff should be empty")
|
|
}
|
|
}
|
|
|
|
// Test for #2267
|
|
func (s *DockerSuite) TestRunWriteResolvFileAndNotCommit(c *check.C) {
|
|
// Cannot run on Windows as Windows does not support diff.
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "writeresolv"
|
|
out, _ := dockerCmd(c, "run", "--name", name, "busybox", "sh", "-c", "echo test2267 >> /etc/resolv.conf && cat /etc/resolv.conf")
|
|
if !strings.Contains(out, "test2267") {
|
|
c.Fatal("/etc/resolv.conf should contain 'test2267'")
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "diff", name)
|
|
if len(strings.Trim(out, "\r\n")) != 0 && !eqToBaseDiff(out, c) {
|
|
c.Fatal("diff should be empty")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithBadDevice(c *check.C) {
|
|
// Cannot run on Windows as Windows does not support --device
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "baddevice"
|
|
out, _, err := dockerCmdWithError("run", "--name", name, "--device", "/etc", "busybox", "true")
|
|
|
|
if err == nil {
|
|
c.Fatal("Run should fail with bad device")
|
|
}
|
|
expected := `"/etc": not a device node`
|
|
if !strings.Contains(out, expected) {
|
|
c.Fatalf("Output should contain %q, actual out: %q", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunEntrypoint(c *check.C) {
|
|
name := "entrypoint"
|
|
|
|
// Note Windows does not have an echo.exe built in.
|
|
var out, expected string
|
|
if daemonPlatform == "windows" {
|
|
out, _ = dockerCmd(c, "run", "--name", name, "--entrypoint", "cmd /s /c echo", "busybox", "foobar")
|
|
expected = "foobar\r\n"
|
|
} else {
|
|
out, _ = dockerCmd(c, "run", "--name", name, "--entrypoint", "/bin/echo", "busybox", "-n", "foobar")
|
|
expected = "foobar"
|
|
}
|
|
|
|
if out != expected {
|
|
c.Fatalf("Output should be %q, actual out: %q", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunBindMounts(c *check.C) {
|
|
testRequires(c, SameHostDaemon)
|
|
if daemonPlatform == "linux" {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
}
|
|
|
|
tmpDir, err := ioutil.TempDir("", "docker-test-container")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
defer os.RemoveAll(tmpDir)
|
|
writeFile(path.Join(tmpDir, "touch-me"), "", c)
|
|
|
|
// TODO Windows Post TP4. Windows does not yet support :ro binds
|
|
if daemonPlatform != "windows" {
|
|
// Test reading from a read-only bind mount
|
|
out, _ := dockerCmd(c, "run", "-v", fmt.Sprintf("%s:/tmp:ro", tmpDir), "busybox", "ls", "/tmp")
|
|
if !strings.Contains(out, "touch-me") {
|
|
c.Fatal("Container failed to read from bind mount")
|
|
}
|
|
}
|
|
|
|
// test writing to bind mount
|
|
if daemonPlatform == "windows" {
|
|
dockerCmd(c, "run", "-v", fmt.Sprintf(`%s:c:\tmp:rw`, tmpDir), "busybox", "touch", "c:/tmp/holla")
|
|
} else {
|
|
dockerCmd(c, "run", "-v", fmt.Sprintf("%s:/tmp:rw", tmpDir), "busybox", "touch", "/tmp/holla")
|
|
}
|
|
|
|
readFile(path.Join(tmpDir, "holla"), c) // Will fail if the file doesn't exist
|
|
|
|
// test mounting to an illegal destination directory
|
|
_, _, err = dockerCmdWithError("run", "-v", fmt.Sprintf("%s:.", tmpDir), "busybox", "ls", ".")
|
|
if err == nil {
|
|
c.Fatal("Container bind mounted illegal directory")
|
|
}
|
|
|
|
// Windows does not (and likely never will) support mounting a single file
|
|
if daemonPlatform != "windows" {
|
|
// test mount a file
|
|
dockerCmd(c, "run", "-v", fmt.Sprintf("%s/holla:/tmp/holla:rw", tmpDir), "busybox", "sh", "-c", "echo -n 'yotta' > /tmp/holla")
|
|
content := readFile(path.Join(tmpDir, "holla"), c) // Will fail if the file doesn't exist
|
|
expected := "yotta"
|
|
if content != expected {
|
|
c.Fatalf("Output should be %q, actual out: %q", expected, content)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Ensure that CIDFile gets deleted if it's empty
|
|
// Perform this test by making `docker run` fail
|
|
func (s *DockerSuite) TestRunCidFileCleanupIfEmpty(c *check.C) {
|
|
tmpDir, err := ioutil.TempDir("", "TestRunCidFile")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir)
|
|
tmpCidFile := path.Join(tmpDir, "cid")
|
|
|
|
image := "emptyfs"
|
|
if daemonPlatform == "windows" {
|
|
// Windows can't support an emptyfs image. Just use the regular Windows image
|
|
image = WindowsBaseImage
|
|
}
|
|
out, _, err := dockerCmdWithError("run", "--cidfile", tmpCidFile, image)
|
|
if err == nil {
|
|
c.Fatalf("Run without command must fail. out=%s", out)
|
|
} else if !strings.Contains(out, "No command specified") {
|
|
c.Fatalf("Run without command failed with wrong output. out=%s\nerr=%v", out, err)
|
|
}
|
|
|
|
if _, err := os.Stat(tmpCidFile); err == nil {
|
|
c.Fatalf("empty CIDFile %q should've been deleted", tmpCidFile)
|
|
}
|
|
}
|
|
|
|
// #2098 - Docker cidFiles only contain short version of the containerId
|
|
//sudo docker run --cidfile /tmp/docker_tesc.cid ubuntu echo "test"
|
|
// TestRunCidFile tests that run --cidfile returns the longid
|
|
func (s *DockerSuite) TestRunCidFileCheckIDLength(c *check.C) {
|
|
tmpDir, err := ioutil.TempDir("", "TestRunCidFile")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
tmpCidFile := path.Join(tmpDir, "cid")
|
|
defer os.RemoveAll(tmpDir)
|
|
|
|
out, _ := dockerCmd(c, "run", "-d", "--cidfile", tmpCidFile, "busybox", "true")
|
|
|
|
id := strings.TrimSpace(out)
|
|
buffer, err := ioutil.ReadFile(tmpCidFile)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
cid := string(buffer)
|
|
if len(cid) != 64 {
|
|
c.Fatalf("--cidfile should be a long id, not %q", id)
|
|
}
|
|
if cid != id {
|
|
c.Fatalf("cid must be equal to %s, got %s", id, cid)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunSetMacAddress(c *check.C) {
|
|
mac := "12:34:56:78:9a:bc"
|
|
var out string
|
|
if daemonPlatform == "windows" {
|
|
out, _ = dockerCmd(c, "run", "-i", "--rm", fmt.Sprintf("--mac-address=%s", mac), "busybox", "sh", "-c", "ipconfig /all | grep 'Physical Address' | awk '{print $12}'")
|
|
mac = strings.Replace(strings.ToUpper(mac), ":", "-", -1) // To Windows-style MACs
|
|
} else {
|
|
out, _ = dockerCmd(c, "run", "-i", "--rm", fmt.Sprintf("--mac-address=%s", mac), "busybox", "/bin/sh", "-c", "ip link show eth0 | tail -1 | awk '{print $2}'")
|
|
}
|
|
|
|
actualMac := strings.TrimSpace(out)
|
|
if actualMac != mac {
|
|
c.Fatalf("Set MAC address with --mac-address failed. The container has an incorrect MAC address: %q, expected: %q", actualMac, mac)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunInspectMacAddress(c *check.C) {
|
|
// TODO Windows. Network settings are not propagated back to inspect.
|
|
testRequires(c, DaemonIsLinux)
|
|
mac := "12:34:56:78:9a:bc"
|
|
out, _ := dockerCmd(c, "run", "-d", "--mac-address="+mac, "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
inspectedMac := inspectField(c, id, "NetworkSettings.Networks.bridge.MacAddress")
|
|
if inspectedMac != mac {
|
|
c.Fatalf("docker inspect outputs wrong MAC address: %q, should be: %q", inspectedMac, mac)
|
|
}
|
|
}
|
|
|
|
// test docker run use an invalid mac address
|
|
func (s *DockerSuite) TestRunWithInvalidMacAddress(c *check.C) {
|
|
out, _, err := dockerCmdWithError("run", "--mac-address", "92:d0:c6:0a:29", "busybox")
|
|
//use an invalid mac address should with an error out
|
|
if err == nil || !strings.Contains(out, "is not a valid mac address") {
|
|
c.Fatalf("run with an invalid --mac-address should with error out")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunDeallocatePortOnMissingIptablesRule(c *check.C) {
|
|
// TODO Windows. Network settings are not propagated back to inspect.
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
out, _ := dockerCmd(c, "run", "-d", "-p", "23:23", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
ip := inspectField(c, id, "NetworkSettings.Networks.bridge.IPAddress")
|
|
iptCmd := exec.Command("iptables", "-D", "DOCKER", "-d", fmt.Sprintf("%s/32", ip),
|
|
"!", "-i", "docker0", "-o", "docker0", "-p", "tcp", "-m", "tcp", "--dport", "23", "-j", "ACCEPT")
|
|
out, _, err := runCommandWithOutput(iptCmd)
|
|
if err != nil {
|
|
c.Fatal(err, out)
|
|
}
|
|
if err := deleteContainer(id); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-d", "-p", "23:23", "busybox", "top")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPortInUse(c *check.C) {
|
|
// TODO Windows. The duplicate NAT message returned by Windows will be
|
|
// changing as is currently completely undecipherable. Does need modifying
|
|
// to run sh rather than top though as top isn't in Windows busybox.
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
port := "1234"
|
|
dockerCmd(c, "run", "-d", "-p", port+":80", "busybox", "top")
|
|
|
|
out, _, err := dockerCmdWithError("run", "-d", "-p", port+":80", "busybox", "top")
|
|
if err == nil {
|
|
c.Fatalf("Binding on used port must fail")
|
|
}
|
|
if !strings.Contains(out, "port is already allocated") {
|
|
c.Fatalf("Out must be about \"port is already allocated\", got %s", out)
|
|
}
|
|
}
|
|
|
|
// https://github.com/docker/docker/issues/12148
|
|
func (s *DockerSuite) TestRunAllocatePortInReservedRange(c *check.C) {
|
|
// TODO Windows. -P is not yet supported
|
|
testRequires(c, DaemonIsLinux)
|
|
// allocate a dynamic port to get the most recent
|
|
out, _ := dockerCmd(c, "run", "-d", "-P", "-p", "80", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
out, _ = dockerCmd(c, "port", id, "80")
|
|
|
|
strPort := strings.Split(strings.TrimSpace(out), ":")[1]
|
|
port, err := strconv.ParseInt(strPort, 10, 64)
|
|
if err != nil {
|
|
c.Fatalf("invalid port, got: %s, error: %s", strPort, err)
|
|
}
|
|
|
|
// allocate a static port and a dynamic port together, with static port
|
|
// takes the next recent port in dynamic port range.
|
|
dockerCmd(c, "run", "-d", "-P", "-p", "80", "-p", fmt.Sprintf("%d:8080", port+1), "busybox", "top")
|
|
}
|
|
|
|
// Regression test for #7792
|
|
func (s *DockerSuite) TestRunMountOrdering(c *check.C) {
|
|
// TODO Windows: Post TP4. Updated, but Windows does not support nested mounts currently.
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
|
|
tmpDir, err := ioutil.TempDir("", "docker_nested_mount_test")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir)
|
|
|
|
tmpDir2, err := ioutil.TempDir("", "docker_nested_mount_test2")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir2)
|
|
|
|
// Create a temporary tmpfs mounc.
|
|
fooDir := filepath.Join(tmpDir, "foo")
|
|
if err := os.MkdirAll(filepath.Join(tmpDir, "foo"), 0755); err != nil {
|
|
c.Fatalf("failed to mkdir at %s - %s", fooDir, err)
|
|
}
|
|
|
|
if err := ioutil.WriteFile(fmt.Sprintf("%s/touch-me", fooDir), []byte{}, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if err := ioutil.WriteFile(fmt.Sprintf("%s/touch-me", tmpDir), []byte{}, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if err := ioutil.WriteFile(fmt.Sprintf("%s/touch-me", tmpDir2), []byte{}, 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run",
|
|
"-v", fmt.Sprintf("%s:"+prefix+"/tmp", tmpDir),
|
|
"-v", fmt.Sprintf("%s:"+prefix+"/tmp/foo", fooDir),
|
|
"-v", fmt.Sprintf("%s:"+prefix+"/tmp/tmp2", tmpDir2),
|
|
"-v", fmt.Sprintf("%s:"+prefix+"/tmp/tmp2/foo", fooDir),
|
|
"busybox:latest", "sh", "-c",
|
|
"ls "+prefix+"/tmp/touch-me && ls "+prefix+"/tmp/foo/touch-me && ls "+prefix+"/tmp/tmp2/touch-me && ls "+prefix+"/tmp/tmp2/foo/touch-me")
|
|
}
|
|
|
|
// Regression test for https://github.com/docker/docker/issues/8259
|
|
func (s *DockerSuite) TestRunReuseBindVolumeThatIsSymlink(c *check.C) {
|
|
// Not applicable on Windows as Windows does not support volumes
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
|
|
tmpDir, err := ioutil.TempDir(os.TempDir(), "testlink")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir)
|
|
|
|
linkPath := os.TempDir() + "/testlink2"
|
|
if err := os.Symlink(tmpDir, linkPath); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(linkPath)
|
|
|
|
// Create first container
|
|
dockerCmd(c, "run", "-v", fmt.Sprintf("%s:"+prefix+"/tmp/test", linkPath), "busybox", "ls", prefix+"/tmp/test")
|
|
|
|
// Create second container with same symlinked path
|
|
// This will fail if the referenced issue is hit with a "Volume exists" error
|
|
dockerCmd(c, "run", "-v", fmt.Sprintf("%s:"+prefix+"/tmp/test", linkPath), "busybox", "ls", prefix+"/tmp/test")
|
|
}
|
|
|
|
//GH#10604: Test an "/etc" volume doesn't overlay special bind mounts in container
|
|
func (s *DockerSuite) TestRunCreateVolumeEtc(c *check.C) {
|
|
// While Windows supports volumes, it does not support --add-host hence
|
|
// this test is not applicable on Windows.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--dns=127.0.0.1", "-v", "/etc", "busybox", "cat", "/etc/resolv.conf")
|
|
if !strings.Contains(out, "nameserver 127.0.0.1") {
|
|
c.Fatal("/etc volume mount hides /etc/resolv.conf")
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "-h=test123", "-v", "/etc", "busybox", "cat", "/etc/hostname")
|
|
if !strings.Contains(out, "test123") {
|
|
c.Fatal("/etc volume mount hides /etc/hostname")
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "--add-host=test:192.168.0.1", "-v", "/etc", "busybox", "cat", "/etc/hosts")
|
|
out = strings.Replace(out, "\n", " ", -1)
|
|
if !strings.Contains(out, "192.168.0.1\ttest") || !strings.Contains(out, "127.0.0.1\tlocalhost") {
|
|
c.Fatal("/etc volume mount hides /etc/hosts")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestVolumesNoCopyData(c *check.C) {
|
|
// TODO Windows (Post TP4). Windows does not support volumes which
|
|
// are pre-populated such as is built in the dockerfile used in this test.
|
|
testRequires(c, DaemonIsLinux)
|
|
if _, err := buildImage("dataimage",
|
|
`FROM busybox
|
|
RUN mkdir -p /foo
|
|
RUN touch /foo/bar`,
|
|
true); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "--name", "test", "-v", "/foo", "busybox")
|
|
|
|
if out, _, err := dockerCmdWithError("run", "--volumes-from", "test", "dataimage", "ls", "-lh", "/foo/bar"); err == nil || !strings.Contains(out, "No such file or directory") {
|
|
c.Fatalf("Data was copied on volumes-from but shouldn't be:\n%q", out)
|
|
}
|
|
|
|
tmpDir := randomTmpDirPath("docker_test_bind_mount_copy_data", daemonPlatform)
|
|
if out, _, err := dockerCmdWithError("run", "-v", tmpDir+":/foo", "dataimage", "ls", "-lh", "/foo/bar"); err == nil || !strings.Contains(out, "No such file or directory") {
|
|
c.Fatalf("Data was copied on bind-mount but shouldn't be:\n%q", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNoOutputFromPullInStdout(c *check.C) {
|
|
// just run with unknown image
|
|
cmd := exec.Command(dockerBinary, "run", "asdfsg")
|
|
stdout := bytes.NewBuffer(nil)
|
|
cmd.Stdout = stdout
|
|
if err := cmd.Run(); err == nil {
|
|
c.Fatal("Run with unknown image should fail")
|
|
}
|
|
if stdout.Len() != 0 {
|
|
c.Fatalf("Stdout contains output from pull: %s", stdout)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesCleanPaths(c *check.C) {
|
|
testRequires(c, SameHostDaemon)
|
|
prefix, slash := getPrefixAndSlashFromDaemonPlatform()
|
|
if _, err := buildImage("run_volumes_clean_paths",
|
|
`FROM busybox
|
|
VOLUME `+prefix+`/foo/`,
|
|
true); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-v", prefix+"/foo", "-v", prefix+"/bar/", "--name", "dark_helmet", "run_volumes_clean_paths")
|
|
|
|
out, err := inspectMountSourceField("dark_helmet", prefix+slash+"foo"+slash)
|
|
if err != errMountNotFound {
|
|
c.Fatalf("Found unexpected volume entry for '%s/foo/' in volumes\n%q", prefix, out)
|
|
}
|
|
|
|
out, err = inspectMountSourceField("dark_helmet", prefix+slash+`foo`)
|
|
c.Assert(err, check.IsNil)
|
|
if !strings.Contains(strings.ToLower(out), strings.ToLower(volumesConfigPath)) {
|
|
c.Fatalf("Volume was not defined for %s/foo\n%q", prefix, out)
|
|
}
|
|
|
|
out, err = inspectMountSourceField("dark_helmet", prefix+slash+"bar"+slash)
|
|
if err != errMountNotFound {
|
|
c.Fatalf("Found unexpected volume entry for '%s/bar/' in volumes\n%q", prefix, out)
|
|
}
|
|
|
|
out, err = inspectMountSourceField("dark_helmet", prefix+slash+"bar")
|
|
c.Assert(err, check.IsNil)
|
|
if !strings.Contains(strings.ToLower(out), strings.ToLower(volumesConfigPath)) {
|
|
c.Fatalf("Volume was not defined for %s/bar\n%q", prefix, out)
|
|
}
|
|
}
|
|
|
|
// Regression test for #3631
|
|
func (s *DockerSuite) TestRunSlowStdoutConsumer(c *check.C) {
|
|
// TODO Windows: This should be able to run on Windows if can find an
|
|
// alternate to /dev/zero and /dev/stdout.
|
|
testRequires(c, DaemonIsLinux)
|
|
cont := exec.Command(dockerBinary, "run", "--rm", "busybox", "/bin/sh", "-c", "dd if=/dev/zero of=/dev/stdout bs=1024 count=2000 | catv")
|
|
|
|
stdout, err := cont.StdoutPipe()
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
if err := cont.Start(); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
n, err := consumeWithSpeed(stdout, 10000, 5*time.Millisecond, nil)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
expected := 2 * 1024 * 2000
|
|
if n != expected {
|
|
c.Fatalf("Expected %d, got %d", expected, n)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAllowPortRangeThroughExpose(c *check.C) {
|
|
// TODO Windows: -P is not currently supported. Also network
|
|
// settings are not propagated back.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-d", "--expose", "3000-3003", "-P", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
portstr := inspectFieldJSON(c, id, "NetworkSettings.Ports")
|
|
var ports nat.PortMap
|
|
if err := unmarshalJSON([]byte(portstr), &ports); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
for port, binding := range ports {
|
|
portnum, _ := strconv.Atoi(strings.Split(string(port), "/")[0])
|
|
if portnum < 3000 || portnum > 3003 {
|
|
c.Fatalf("Port %d is out of range ", portnum)
|
|
}
|
|
if binding == nil || len(binding) != 1 || len(binding[0].HostPort) == 0 {
|
|
c.Fatalf("Port is not mapped for the port %s", port)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunExposePort(c *check.C) {
|
|
out, _, err := dockerCmdWithError("run", "--expose", "80000", "busybox")
|
|
c.Assert(err, checker.NotNil, check.Commentf("--expose with an invalid port should error out"))
|
|
c.Assert(out, checker.Contains, "invalid range format for --expose")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUnknownCommand(c *check.C) {
|
|
out, _, _ := dockerCmdWithStdoutStderr(c, "create", "busybox", "/bin/nada")
|
|
|
|
cID := strings.TrimSpace(out)
|
|
_, _, err := dockerCmdWithError("start", cID)
|
|
|
|
// Windows and Linux are different here by architectural design. Linux will
|
|
// fail to start the container, so an error is expected. Windows will
|
|
// successfully start the container, and once started attempt to execute
|
|
// the command which will fail.
|
|
if daemonPlatform == "windows" {
|
|
// Wait for it to exit.
|
|
waitExited(cID, 30*time.Second)
|
|
c.Assert(err, check.IsNil)
|
|
} else {
|
|
c.Assert(err, check.NotNil)
|
|
}
|
|
|
|
rc := inspectField(c, cID, "State.ExitCode")
|
|
if rc == "0" {
|
|
c.Fatalf("ExitCode(%v) cannot be 0", rc)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeIpcHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
hostIpc, err := os.Readlink("/proc/1/ns/ipc")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "run", "--ipc=host", "busybox", "readlink", "/proc/self/ns/ipc")
|
|
out = strings.Trim(out, "\n")
|
|
if hostIpc != out {
|
|
c.Fatalf("IPC different with --ipc=host %s != %s\n", hostIpc, out)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "busybox", "readlink", "/proc/self/ns/ipc")
|
|
out = strings.Trim(out, "\n")
|
|
if hostIpc == out {
|
|
c.Fatalf("IPC should be different without --ipc=host %s == %s\n", hostIpc, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeIpcContainer(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "sh", "-c", "echo -n test > /dev/shm/test && touch /dev/mqueue/toto && top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
state := inspectField(c, id, "State.Running")
|
|
if state != "true" {
|
|
c.Fatal("Container state is 'not running'")
|
|
}
|
|
pid1 := inspectField(c, id, "State.Pid")
|
|
|
|
parentContainerIpc, err := os.Readlink(fmt.Sprintf("/proc/%s/ns/ipc", pid1))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", fmt.Sprintf("--ipc=container:%s", id), "busybox", "readlink", "/proc/self/ns/ipc")
|
|
out = strings.Trim(out, "\n")
|
|
if parentContainerIpc != out {
|
|
c.Fatalf("IPC different with --ipc=container:%s %s != %s\n", id, parentContainerIpc, out)
|
|
}
|
|
|
|
catOutput, _ := dockerCmd(c, "run", fmt.Sprintf("--ipc=container:%s", id), "busybox", "cat", "/dev/shm/test")
|
|
if catOutput != "test" {
|
|
c.Fatalf("Output of /dev/shm/test expected test but found: %s", catOutput)
|
|
}
|
|
|
|
// check that /dev/mqueue is actually of mqueue type
|
|
grepOutput, _ := dockerCmd(c, "run", fmt.Sprintf("--ipc=container:%s", id), "busybox", "grep", "/dev/mqueue", "/proc/mounts")
|
|
if !strings.HasPrefix(grepOutput, "mqueue /dev/mqueue mqueue rw") {
|
|
c.Fatalf("Output of 'grep /proc/mounts' expected 'mqueue /dev/mqueue mqueue rw' but found: %s", grepOutput)
|
|
}
|
|
|
|
lsOutput, _ := dockerCmd(c, "run", fmt.Sprintf("--ipc=container:%s", id), "busybox", "ls", "/dev/mqueue")
|
|
lsOutput = strings.Trim(lsOutput, "\n")
|
|
if lsOutput != "toto" {
|
|
c.Fatalf("Output of 'ls /dev/mqueue' expected 'toto' but found: %s", lsOutput)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeIpcContainerNotExists(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _, err := dockerCmdWithError("run", "-d", "--ipc", "container:abcd1234", "busybox", "top")
|
|
if !strings.Contains(out, "abcd1234") || err == nil {
|
|
c.Fatalf("run IPC from a non exists container should with correct error out")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeIpcContainerNotRunning(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "create", "busybox")
|
|
|
|
id := strings.TrimSpace(out)
|
|
out, _, err := dockerCmdWithError("run", fmt.Sprintf("--ipc=container:%s", id), "busybox")
|
|
if err == nil {
|
|
c.Fatalf("Run container with ipc mode container should fail with non running container: %s\n%s", out, err)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunMountShmMqueueFromHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
dockerCmd(c, "run", "-d", "--name", "shmfromhost", "-v", "/dev/shm:/dev/shm", "-v", "/dev/mqueue:/dev/mqueue", "busybox", "sh", "-c", "echo -n test > /dev/shm/test && touch /dev/mqueue/toto && top")
|
|
defer os.Remove("/dev/mqueue/toto")
|
|
defer os.Remove("/dev/shm/test")
|
|
volPath, err := inspectMountSourceField("shmfromhost", "/dev/shm")
|
|
c.Assert(err, checker.IsNil)
|
|
if volPath != "/dev/shm" {
|
|
c.Fatalf("volumePath should have been /dev/shm, was %s", volPath)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "run", "--name", "ipchost", "--ipc", "host", "busybox", "cat", "/dev/shm/test")
|
|
if out != "test" {
|
|
c.Fatalf("Output of /dev/shm/test expected test but found: %s", out)
|
|
}
|
|
|
|
// Check that the mq was created
|
|
if _, err := os.Stat("/dev/mqueue/toto"); err != nil {
|
|
c.Fatalf("Failed to confirm '/dev/mqueue/toto' presence on host: %s", err.Error())
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainerNetworkMode(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "top")
|
|
id := strings.TrimSpace(out)
|
|
c.Assert(waitRun(id), check.IsNil)
|
|
pid1 := inspectField(c, id, "State.Pid")
|
|
|
|
parentContainerNet, err := os.Readlink(fmt.Sprintf("/proc/%s/ns/net", pid1))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", fmt.Sprintf("--net=container:%s", id), "busybox", "readlink", "/proc/self/ns/net")
|
|
out = strings.Trim(out, "\n")
|
|
if parentContainerNet != out {
|
|
c.Fatalf("NET different with --net=container:%s %s != %s\n", id, parentContainerNet, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModePidHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
hostPid, err := os.Readlink("/proc/1/ns/pid")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "run", "--pid=host", "busybox", "readlink", "/proc/self/ns/pid")
|
|
out = strings.Trim(out, "\n")
|
|
if hostPid != out {
|
|
c.Fatalf("PID different with --pid=host %s != %s\n", hostPid, out)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "busybox", "readlink", "/proc/self/ns/pid")
|
|
out = strings.Trim(out, "\n")
|
|
if hostPid == out {
|
|
c.Fatalf("PID should be different without --pid=host %s == %s\n", hostPid, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeUTSHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
hostUTS, err := os.Readlink("/proc/1/ns/uts")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "run", "--uts=host", "busybox", "readlink", "/proc/self/ns/uts")
|
|
out = strings.Trim(out, "\n")
|
|
if hostUTS != out {
|
|
c.Fatalf("UTS different with --uts=host %s != %s\n", hostUTS, out)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "busybox", "readlink", "/proc/self/ns/uts")
|
|
out = strings.Trim(out, "\n")
|
|
if hostUTS == out {
|
|
c.Fatalf("UTS should be different without --uts=host %s == %s\n", hostUTS, out)
|
|
}
|
|
|
|
out, _ = dockerCmdWithFail(c, "run", "-h=name", "--uts=host", "busybox", "ps")
|
|
c.Assert(out, checker.Contains, runconfig.ErrConflictUTSHostname.Error())
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunTLSverify(c *check.C) {
|
|
// Remote daemons use TLS and this test is not applicable when TLS is required.
|
|
testRequires(c, SameHostDaemon)
|
|
if out, code, err := dockerCmdWithError("ps"); err != nil || code != 0 {
|
|
c.Fatalf("Should have worked: %v:\n%v", err, out)
|
|
}
|
|
|
|
// Regardless of whether we specify true or false we need to
|
|
// test to make sure tls is turned on if --tlsverify is specified at all
|
|
out, code, err := dockerCmdWithError("--tlsverify=false", "ps")
|
|
if err == nil || code == 0 || !strings.Contains(out, "trying to connect") {
|
|
c.Fatalf("Should have failed: \net:%v\nout:%v\nerr:%v", code, out, err)
|
|
}
|
|
|
|
out, code, err = dockerCmdWithError("--tlsverify=true", "ps")
|
|
if err == nil || code == 0 || !strings.Contains(out, "cert") {
|
|
c.Fatalf("Should have failed: \net:%v\nout:%v\nerr:%v", code, out, err)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPortFromDockerRangeInUse(c *check.C) {
|
|
// TODO Windows. Once moved to libnetwork/CNM, this may be able to be
|
|
// re-instated.
|
|
testRequires(c, DaemonIsLinux)
|
|
// first find allocator current position
|
|
out, _ := dockerCmd(c, "run", "-d", "-p", ":80", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
out, _ = dockerCmd(c, "port", id)
|
|
|
|
out = strings.TrimSpace(out)
|
|
if out == "" {
|
|
c.Fatal("docker port command output is empty")
|
|
}
|
|
out = strings.Split(out, ":")[1]
|
|
lastPort, err := strconv.Atoi(out)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
port := lastPort + 1
|
|
l, err := net.Listen("tcp", ":"+strconv.Itoa(port))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer l.Close()
|
|
|
|
out, _ = dockerCmd(c, "run", "-d", "-p", ":80", "busybox", "top")
|
|
|
|
id = strings.TrimSpace(out)
|
|
dockerCmd(c, "port", id)
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunTTYWithPipe(c *check.C) {
|
|
errChan := make(chan error)
|
|
go func() {
|
|
defer close(errChan)
|
|
|
|
cmd := exec.Command(dockerBinary, "run", "-ti", "busybox", "true")
|
|
if _, err := cmd.StdinPipe(); err != nil {
|
|
errChan <- err
|
|
return
|
|
}
|
|
|
|
expected := "cannot enable tty mode"
|
|
if out, _, err := runCommandWithOutput(cmd); err == nil {
|
|
errChan <- fmt.Errorf("run should have failed")
|
|
return
|
|
} else if !strings.Contains(out, expected) {
|
|
errChan <- fmt.Errorf("run failed with error %q: expected %q", out, expected)
|
|
return
|
|
}
|
|
}()
|
|
|
|
select {
|
|
case err := <-errChan:
|
|
c.Assert(err, check.IsNil)
|
|
case <-time.After(6 * time.Second):
|
|
c.Fatal("container is running but should have failed")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNonLocalMacAddress(c *check.C) {
|
|
addr := "00:16:3E:08:00:50"
|
|
cmd := "ifconfig"
|
|
image := "busybox"
|
|
expected := addr
|
|
|
|
if daemonPlatform == "windows" {
|
|
cmd = "ipconfig /all"
|
|
image = WindowsBaseImage
|
|
expected = strings.Replace(strings.ToUpper(addr), ":", "-", -1)
|
|
|
|
}
|
|
|
|
if out, _ := dockerCmd(c, "run", "--mac-address", addr, image, cmd); !strings.Contains(out, expected) {
|
|
c.Fatalf("Output should have contained %q: %s", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
hostNet, err := os.Readlink("/proc/1/ns/net")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "run", "--net=host", "busybox", "readlink", "/proc/self/ns/net")
|
|
out = strings.Trim(out, "\n")
|
|
if hostNet != out {
|
|
c.Fatalf("Net namespace different with --net=host %s != %s\n", hostNet, out)
|
|
}
|
|
|
|
out, _ = dockerCmd(c, "run", "busybox", "readlink", "/proc/self/ns/net")
|
|
out = strings.Trim(out, "\n")
|
|
if hostNet == out {
|
|
c.Fatalf("Net namespace should be different without --net=host %s == %s\n", hostNet, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetHostTwiceSameName(c *check.C) {
|
|
// TODO Windows. As Windows networking evolves and converges towards
|
|
// CNM, this test may be possible to enable on Windows.
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
dockerCmd(c, "run", "--rm", "--name=thost", "--net=host", "busybox", "true")
|
|
dockerCmd(c, "run", "--rm", "--name=thost", "--net=host", "busybox", "true")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetContainerWhichHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix-specific capabilities
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
hostNet, err := os.Readlink("/proc/1/ns/net")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-d", "--net=host", "--name=test", "busybox", "top")
|
|
|
|
out, _ := dockerCmd(c, "run", "--net=container:test", "busybox", "readlink", "/proc/self/ns/net")
|
|
out = strings.Trim(out, "\n")
|
|
if hostNet != out {
|
|
c.Fatalf("Container should have host network namespace")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAllowPortRangeThroughPublish(c *check.C) {
|
|
// TODO Windows. This may be possible to enable in the future. However,
|
|
// Windows does not currently support --expose, or populate the network
|
|
// settings seen through inspect.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-d", "--expose", "3000-3003", "-p", "3000-3003", "busybox", "top")
|
|
|
|
id := strings.TrimSpace(out)
|
|
portstr := inspectFieldJSON(c, id, "NetworkSettings.Ports")
|
|
|
|
var ports nat.PortMap
|
|
err := unmarshalJSON([]byte(portstr), &ports)
|
|
c.Assert(err, checker.IsNil, check.Commentf("failed to unmarshal: %v", portstr))
|
|
for port, binding := range ports {
|
|
portnum, _ := strconv.Atoi(strings.Split(string(port), "/")[0])
|
|
if portnum < 3000 || portnum > 3003 {
|
|
c.Fatalf("Port %d is out of range ", portnum)
|
|
}
|
|
if binding == nil || len(binding) != 1 || len(binding[0].HostPort) == 0 {
|
|
c.Fatal("Port is not mapped for the port "+port, out)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunSetDefaultRestartPolicy(c *check.C) {
|
|
dockerCmd(c, "run", "-d", "--name", "test", "busybox", "sleep", "30")
|
|
out := inspectField(c, "test", "HostConfig.RestartPolicy.Name")
|
|
if out != "no" {
|
|
c.Fatalf("Set default restart policy failed")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunRestartMaxRetries(c *check.C) {
|
|
out, _ := dockerCmd(c, "run", "-d", "--restart=on-failure:3", "busybox", "false")
|
|
timeout := 10 * time.Second
|
|
if daemonPlatform == "windows" {
|
|
timeout = 120 * time.Second
|
|
}
|
|
|
|
id := strings.TrimSpace(string(out))
|
|
if err := waitInspect(id, "{{ .State.Restarting }} {{ .State.Running }}", "false false", timeout); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
count := inspectField(c, id, "RestartCount")
|
|
if count != "3" {
|
|
c.Fatalf("Container was restarted %s times, expected %d", count, 3)
|
|
}
|
|
|
|
MaximumRetryCount := inspectField(c, id, "HostConfig.RestartPolicy.MaximumRetryCount")
|
|
if MaximumRetryCount != "3" {
|
|
c.Fatalf("Container Maximum Retry Count is %s, expected %s", MaximumRetryCount, "3")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithWritableRootfs(c *check.C) {
|
|
dockerCmd(c, "run", "--rm", "busybox", "touch", "/file")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithReadonlyRootfs(c *check.C) {
|
|
// Not applicable on Windows which does not support --read-only
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
testReadOnlyFile(c, "/file", "/etc/hosts", "/etc/resolv.conf", "/etc/hostname", "/sys/kernel", "/dev/.dont.touch.me")
|
|
}
|
|
|
|
func (s *DockerSuite) TestPermissionsPtsReadonlyRootfs(c *check.C) {
|
|
// Not applicable on Windows due to use of Unix specific functionality, plus
|
|
// the use of --read-only which is not supported.
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
|
|
// Ensure we have not broken writing /dev/pts
|
|
out, status := dockerCmd(c, "run", "--read-only", "--rm", "busybox", "mount")
|
|
if status != 0 {
|
|
c.Fatal("Could not obtain mounts when checking /dev/pts mntpnt.")
|
|
}
|
|
expected := "type devpts (rw,"
|
|
if !strings.Contains(string(out), expected) {
|
|
c.Fatalf("expected output to contain %s but contains %s", expected, out)
|
|
}
|
|
}
|
|
|
|
func testReadOnlyFile(c *check.C, filenames ...string) {
|
|
// Not applicable on Windows which does not support --read-only
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
touch := "touch " + strings.Join(filenames, " ")
|
|
out, _, err := dockerCmdWithError("run", "--read-only", "--rm", "busybox", "sh", "-c", touch)
|
|
c.Assert(err, checker.NotNil)
|
|
|
|
for _, f := range filenames {
|
|
expected := "touch: " + f + ": Read-only file system"
|
|
c.Assert(out, checker.Contains, expected)
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "--read-only", "--privileged", "--rm", "busybox", "sh", "-c", touch)
|
|
c.Assert(err, checker.NotNil)
|
|
|
|
for _, f := range filenames {
|
|
expected := "touch: " + f + ": Read-only file system"
|
|
c.Assert(out, checker.Contains, expected)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithReadonlyEtcHostsAndLinkedContainer(c *check.C) {
|
|
// Not applicable on Windows which does not support --link
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
|
|
dockerCmd(c, "run", "-d", "--name", "test-etc-hosts-ro-linked", "busybox", "top")
|
|
|
|
out, _ := dockerCmd(c, "run", "--read-only", "--link", "test-etc-hosts-ro-linked:testlinked", "busybox", "cat", "/etc/hosts")
|
|
if !strings.Contains(string(out), "testlinked") {
|
|
c.Fatal("Expected /etc/hosts to be updated even if --read-only enabled")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithReadonlyRootfsWithDnsFlag(c *check.C) {
|
|
// Not applicable on Windows which does not support either --read-only or --dns.
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "run", "--read-only", "--dns", "1.1.1.1", "busybox", "/bin/cat", "/etc/resolv.conf")
|
|
if !strings.Contains(string(out), "1.1.1.1") {
|
|
c.Fatal("Expected /etc/resolv.conf to be updated even if --read-only enabled and --dns flag used")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithReadonlyRootfsWithAddHostFlag(c *check.C) {
|
|
// Not applicable on Windows which does not support --read-only
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
|
|
out, _ := dockerCmd(c, "run", "--read-only", "--add-host", "testreadonly:127.0.0.1", "busybox", "/bin/cat", "/etc/hosts")
|
|
if !strings.Contains(string(out), "testreadonly") {
|
|
c.Fatal("Expected /etc/hosts to be updated even if --read-only enabled and --add-host flag used")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesFromRestartAfterRemoved(c *check.C) {
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
dockerCmd(c, "run", "-d", "--name", "voltest", "-v", prefix+"/foo", "busybox", "sleep", "60")
|
|
dockerCmd(c, "run", "-d", "--name", "restarter", "--volumes-from", "voltest", "busybox", "sleep", "60")
|
|
|
|
// Remove the main volume container and restart the consuming container
|
|
dockerCmd(c, "rm", "-f", "voltest")
|
|
|
|
// This should not fail since the volumes-from were already applied
|
|
dockerCmd(c, "restart", "restarter")
|
|
}
|
|
|
|
// run container with --rm should remove container if exit code != 0
|
|
func (s *DockerSuite) TestRunContainerWithRmFlagExitCodeNotEqualToZero(c *check.C) {
|
|
name := "flowers"
|
|
out, _, err := dockerCmdWithError("run", "--name", name, "--rm", "busybox", "ls", "/notexists")
|
|
if err == nil {
|
|
c.Fatal("Expected docker run to fail", out, err)
|
|
}
|
|
|
|
out, err = getAllContainers()
|
|
if err != nil {
|
|
c.Fatal(out, err)
|
|
}
|
|
|
|
if out != "" {
|
|
c.Fatal("Expected not to have containers", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithRmFlagCannotStartContainer(c *check.C) {
|
|
name := "sparkles"
|
|
out, _, err := dockerCmdWithError("run", "--name", name, "--rm", "busybox", "commandNotFound")
|
|
if err == nil {
|
|
c.Fatal("Expected docker run to fail", out, err)
|
|
}
|
|
|
|
out, err = getAllContainers()
|
|
if err != nil {
|
|
c.Fatal(out, err)
|
|
}
|
|
|
|
if out != "" {
|
|
c.Fatal("Expected not to have containers", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPidHostWithChildIsKillable(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
name := "ibuildthecloud"
|
|
dockerCmd(c, "run", "-d", "--pid=host", "--name", name, "busybox", "sh", "-c", "sleep 30; echo hi")
|
|
|
|
c.Assert(waitRun(name), check.IsNil)
|
|
|
|
errchan := make(chan error)
|
|
go func() {
|
|
if out, _, err := dockerCmdWithError("kill", name); err != nil {
|
|
errchan <- fmt.Errorf("%v:\n%s", err, out)
|
|
}
|
|
close(errchan)
|
|
}()
|
|
select {
|
|
case err := <-errchan:
|
|
c.Assert(err, check.IsNil)
|
|
case <-time.After(5 * time.Second):
|
|
c.Fatal("Kill container timed out")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithTooSmallMemoryLimit(c *check.C) {
|
|
// TODO Windows. This may be possible to enable once Windows supports
|
|
// memory limits on containers
|
|
testRequires(c, DaemonIsLinux)
|
|
// this memory limit is 1 byte less than the min, which is 4MB
|
|
// https://github.com/docker/docker/blob/v1.5.0/daemon/create.go#L22
|
|
out, _, err := dockerCmdWithError("run", "-m", "4194303", "busybox")
|
|
if err == nil || !strings.Contains(out, "Minimum memory limit allowed is 4MB") {
|
|
c.Fatalf("expected run to fail when using too low a memory limit: %q", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWriteToProcAsound(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
_, code, err := dockerCmdWithError("run", "busybox", "sh", "-c", "echo 111 >> /proc/asound/version")
|
|
if err == nil || code == 0 {
|
|
c.Fatal("standard container should not be able to write to /proc/asound")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunReadProcTimer(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
out, code, err := dockerCmdWithError("run", "busybox", "cat", "/proc/timer_stats")
|
|
if code != 0 {
|
|
return
|
|
}
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if strings.Trim(out, "\n ") != "" {
|
|
c.Fatalf("expected to receive no output from /proc/timer_stats but received %q", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunReadProcLatency(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
// some kernels don't have this configured so skip the test if this file is not found
|
|
// on the host running the tests.
|
|
if _, err := os.Stat("/proc/latency_stats"); err != nil {
|
|
c.Skip("kernel doesn't have latency_stats configured")
|
|
return
|
|
}
|
|
out, code, err := dockerCmdWithError("run", "busybox", "cat", "/proc/latency_stats")
|
|
if code != 0 {
|
|
return
|
|
}
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
if strings.Trim(out, "\n ") != "" {
|
|
c.Fatalf("expected to receive no output from /proc/latency_stats but received %q", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunReadFilteredProc(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, Apparmor, DaemonIsLinux, NotUserNamespace)
|
|
|
|
testReadPaths := []string{
|
|
"/proc/latency_stats",
|
|
"/proc/timer_stats",
|
|
"/proc/kcore",
|
|
}
|
|
for i, filePath := range testReadPaths {
|
|
name := fmt.Sprintf("procsieve-%d", i)
|
|
shellCmd := fmt.Sprintf("exec 3<%s", filePath)
|
|
|
|
out, exitCode, err := dockerCmdWithError("run", "--privileged", "--security-opt", "apparmor=docker-default", "--name", name, "busybox", "sh", "-c", shellCmd)
|
|
if exitCode != 0 {
|
|
return
|
|
}
|
|
if err != nil {
|
|
c.Fatalf("Open FD for read should have failed with permission denied, got: %s, %v", out, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestMountIntoProc(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
_, code, err := dockerCmdWithError("run", "-v", "/proc//sys", "busybox", "true")
|
|
if err == nil || code == 0 {
|
|
c.Fatal("container should not be able to mount into /proc")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestMountIntoSys(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
testRequires(c, NotUserNamespace)
|
|
dockerCmd(c, "run", "-v", "/sys/fs/cgroup", "busybox", "true")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunUnshareProc(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, Apparmor, DaemonIsLinux, NotUserNamespace)
|
|
|
|
// In this test goroutines are used to run test cases in parallel to prevent the test from taking a long time to run.
|
|
errChan := make(chan error)
|
|
|
|
go func() {
|
|
name := "acidburn"
|
|
out, _, err := dockerCmdWithError("run", "--name", name, "--security-opt", "seccomp=unconfined", "debian:jessie", "unshare", "-p", "-m", "-f", "-r", "--mount-proc=/proc", "mount")
|
|
if err == nil ||
|
|
!(strings.Contains(strings.ToLower(out), "permission denied") ||
|
|
strings.Contains(strings.ToLower(out), "operation not permitted")) {
|
|
errChan <- fmt.Errorf("unshare with --mount-proc should have failed with 'permission denied' or 'operation not permitted', got: %s, %v", out, err)
|
|
} else {
|
|
errChan <- nil
|
|
}
|
|
}()
|
|
|
|
go func() {
|
|
name := "cereal"
|
|
out, _, err := dockerCmdWithError("run", "--name", name, "--security-opt", "seccomp=unconfined", "debian:jessie", "unshare", "-p", "-m", "-f", "-r", "mount", "-t", "proc", "none", "/proc")
|
|
if err == nil ||
|
|
!(strings.Contains(strings.ToLower(out), "mount: cannot mount none") ||
|
|
strings.Contains(strings.ToLower(out), "permission denied") ||
|
|
strings.Contains(strings.ToLower(out), "operation not permitted")) {
|
|
errChan <- fmt.Errorf("unshare and mount of /proc should have failed with 'mount: cannot mount none' or 'permission denied', got: %s, %v", out, err)
|
|
} else {
|
|
errChan <- nil
|
|
}
|
|
}()
|
|
|
|
/* Ensure still fails if running privileged with the default policy */
|
|
go func() {
|
|
name := "crashoverride"
|
|
out, _, err := dockerCmdWithError("run", "--privileged", "--security-opt", "seccomp=unconfined", "--security-opt", "apparmor=docker-default", "--name", name, "debian:jessie", "unshare", "-p", "-m", "-f", "-r", "mount", "-t", "proc", "none", "/proc")
|
|
if err == nil ||
|
|
!(strings.Contains(strings.ToLower(out), "mount: cannot mount none") ||
|
|
strings.Contains(strings.ToLower(out), "permission denied") ||
|
|
strings.Contains(strings.ToLower(out), "operation not permitted")) {
|
|
errChan <- fmt.Errorf("privileged unshare with apparmor should have failed with 'mount: cannot mount none' or 'permission denied', got: %s, %v", out, err)
|
|
} else {
|
|
errChan <- nil
|
|
}
|
|
}()
|
|
|
|
for i := 0; i < 3; i++ {
|
|
err := <-errChan
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunPublishPort(c *check.C) {
|
|
// TODO Windows: This may be possible once Windows moves to libnetwork and CNM
|
|
testRequires(c, DaemonIsLinux)
|
|
dockerCmd(c, "run", "-d", "--name", "test", "--expose", "8080", "busybox", "top")
|
|
out, _ := dockerCmd(c, "port", "test")
|
|
out = strings.Trim(out, "\r\n")
|
|
if out != "" {
|
|
c.Fatalf("run without --publish-all should not publish port, out should be nil, but got: %s", out)
|
|
}
|
|
}
|
|
|
|
// Issue #10184.
|
|
func (s *DockerSuite) TestDevicePermissions(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
const permissions = "crw-rw-rw-"
|
|
out, status := dockerCmd(c, "run", "--device", "/dev/fuse:/dev/fuse:mrw", "busybox:latest", "ls", "-l", "/dev/fuse")
|
|
if status != 0 {
|
|
c.Fatalf("expected status 0, got %d", status)
|
|
}
|
|
if !strings.HasPrefix(out, permissions) {
|
|
c.Fatalf("output should begin with %q, got %q", permissions, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddCHOWN(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--cap-drop=ALL", "--cap-add=CHOWN", "busybox", "sh", "-c", "adduser -D -H newuser && chown newuser /home && echo ok")
|
|
|
|
if actual := strings.Trim(out, "\r\n"); actual != "ok" {
|
|
c.Fatalf("expected output ok received %s", actual)
|
|
}
|
|
}
|
|
|
|
// https://github.com/docker/docker/pull/14498
|
|
func (s *DockerSuite) TestVolumeFromMixedRWOptions(c *check.C) {
|
|
// TODO Windows post TP4. Enable the read-only bits once they are
|
|
// supported on the platform.
|
|
prefix, slash := getPrefixAndSlashFromDaemonPlatform()
|
|
|
|
dockerCmd(c, "run", "--name", "parent", "-v", prefix+"/test", "busybox", "true")
|
|
if daemonPlatform != "windows" {
|
|
dockerCmd(c, "run", "--volumes-from", "parent:ro", "--name", "test-volumes-1", "busybox", "true")
|
|
}
|
|
dockerCmd(c, "run", "--volumes-from", "parent:rw", "--name", "test-volumes-2", "busybox", "true")
|
|
|
|
if daemonPlatform != "windows" {
|
|
mRO, err := inspectMountPoint("test-volumes-1", prefix+slash+"test")
|
|
c.Assert(err, checker.IsNil, check.Commentf("failed to inspect mount point"))
|
|
if mRO.RW {
|
|
c.Fatalf("Expected RO volume was RW")
|
|
}
|
|
}
|
|
|
|
mRW, err := inspectMountPoint("test-volumes-2", prefix+slash+"test")
|
|
c.Assert(err, checker.IsNil, check.Commentf("failed to inspect mount point"))
|
|
if !mRW.RW {
|
|
c.Fatalf("Expected RW volume was RO")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWriteFilteredProc(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, Apparmor, DaemonIsLinux, NotUserNamespace)
|
|
|
|
testWritePaths := []string{
|
|
/* modprobe and core_pattern should both be denied by generic
|
|
* policy of denials for /proc/sys/kernel. These files have been
|
|
* picked to be checked as they are particularly sensitive to writes */
|
|
"/proc/sys/kernel/modprobe",
|
|
"/proc/sys/kernel/core_pattern",
|
|
"/proc/sysrq-trigger",
|
|
"/proc/kcore",
|
|
}
|
|
for i, filePath := range testWritePaths {
|
|
name := fmt.Sprintf("writeprocsieve-%d", i)
|
|
|
|
shellCmd := fmt.Sprintf("exec 3>%s", filePath)
|
|
out, code, err := dockerCmdWithError("run", "--privileged", "--security-opt", "apparmor=docker-default", "--name", name, "busybox", "sh", "-c", shellCmd)
|
|
if code != 0 {
|
|
return
|
|
}
|
|
if err != nil {
|
|
c.Fatalf("Open FD for write should have failed with permission denied, got: %s, %v", out, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetworkFilesBindMount(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
expected := "test123"
|
|
|
|
filename := createTmpFile(c, expected)
|
|
defer os.Remove(filename)
|
|
|
|
nwfiles := []string{"/etc/resolv.conf", "/etc/hosts", "/etc/hostname"}
|
|
|
|
for i := range nwfiles {
|
|
actual, _ := dockerCmd(c, "run", "-v", filename+":"+nwfiles[i], "busybox", "cat", nwfiles[i])
|
|
if actual != expected {
|
|
c.Fatalf("expected %s be: %q, but was: %q", nwfiles[i], expected, actual)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetworkFilesBindMountRO(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux)
|
|
|
|
filename := createTmpFile(c, "test123")
|
|
defer os.Remove(filename)
|
|
|
|
nwfiles := []string{"/etc/resolv.conf", "/etc/hosts", "/etc/hostname"}
|
|
|
|
for i := range nwfiles {
|
|
_, exitCode, err := dockerCmdWithError("run", "-v", filename+":"+nwfiles[i]+":ro", "busybox", "touch", nwfiles[i])
|
|
if err == nil || exitCode == 0 {
|
|
c.Fatalf("run should fail because bind mount of %s is ro: exit code %d", nwfiles[i], exitCode)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetworkFilesBindMountROFilesystem(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, SameHostDaemon, DaemonIsLinux, NotUserNamespace)
|
|
|
|
filename := createTmpFile(c, "test123")
|
|
defer os.Remove(filename)
|
|
|
|
nwfiles := []string{"/etc/resolv.conf", "/etc/hosts", "/etc/hostname"}
|
|
|
|
for i := range nwfiles {
|
|
_, exitCode := dockerCmd(c, "run", "-v", filename+":"+nwfiles[i], "--read-only", "busybox", "touch", nwfiles[i])
|
|
if exitCode != 0 {
|
|
c.Fatalf("run should not fail because %s is mounted writable on read-only root filesystem: exit code %d", nwfiles[i], exitCode)
|
|
}
|
|
}
|
|
|
|
for i := range nwfiles {
|
|
_, exitCode, err := dockerCmdWithError("run", "-v", filename+":"+nwfiles[i]+":ro", "--read-only", "busybox", "touch", nwfiles[i])
|
|
if err == nil || exitCode == 0 {
|
|
c.Fatalf("run should fail because %s is mounted read-only on read-only root filesystem: exit code %d", nwfiles[i], exitCode)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerTrustSuite) TestTrustedRun(c *check.C) {
|
|
// Windows does not support this functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
repoName := s.setupTrustedImage(c, "trusted-run")
|
|
|
|
// Try run
|
|
runCmd := exec.Command(dockerBinary, "run", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err := runCommandWithOutput(runCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running trusted run: %s\n%s\n", err, out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "Tagging") {
|
|
c.Fatalf("Missing expected output on trusted push:\n%s", out)
|
|
}
|
|
|
|
dockerCmd(c, "rmi", repoName)
|
|
|
|
// Try untrusted run to ensure we pushed the tag to the registry
|
|
runCmd = exec.Command(dockerBinary, "run", "--disable-content-trust=true", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err = runCommandWithOutput(runCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running trusted run: %s\n%s", err, out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "Status: Downloaded") {
|
|
c.Fatalf("Missing expected output on trusted run with --disable-content-trust:\n%s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerTrustSuite) TestUntrustedRun(c *check.C) {
|
|
// Windows does not support this functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
repoName := fmt.Sprintf("%v/dockercliuntrusted/runtest:latest", privateRegistryURL)
|
|
// tag the image and upload it to the private registry
|
|
dockerCmd(c, "tag", "busybox", repoName)
|
|
dockerCmd(c, "push", repoName)
|
|
dockerCmd(c, "rmi", repoName)
|
|
|
|
// Try trusted run on untrusted tag
|
|
runCmd := exec.Command(dockerBinary, "run", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err := runCommandWithOutput(runCmd)
|
|
if err == nil {
|
|
c.Fatalf("Error expected when running trusted run with:\n%s", out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "does not have trust data for") {
|
|
c.Fatalf("Missing expected output on trusted run:\n%s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerTrustSuite) TestRunWhenCertExpired(c *check.C) {
|
|
// Windows does not support this functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
c.Skip("Currently changes system time, causing instability")
|
|
repoName := s.setupTrustedImage(c, "trusted-run-expired")
|
|
|
|
// Certificates have 10 years of expiration
|
|
elevenYearsFromNow := time.Now().Add(time.Hour * 24 * 365 * 11)
|
|
|
|
runAtDifferentDate(elevenYearsFromNow, func() {
|
|
// Try run
|
|
runCmd := exec.Command(dockerBinary, "run", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err := runCommandWithOutput(runCmd)
|
|
if err == nil {
|
|
c.Fatalf("Error running trusted run in the distant future: %s\n%s", err, out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "could not validate the path to a trusted root") {
|
|
c.Fatalf("Missing expected output on trusted run in the distant future:\n%s", out)
|
|
}
|
|
})
|
|
|
|
runAtDifferentDate(elevenYearsFromNow, func() {
|
|
// Try run
|
|
runCmd := exec.Command(dockerBinary, "run", "--disable-content-trust", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err := runCommandWithOutput(runCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running untrusted run in the distant future: %s\n%s", err, out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "Status: Downloaded") {
|
|
c.Fatalf("Missing expected output on untrusted run in the distant future:\n%s", out)
|
|
}
|
|
})
|
|
}
|
|
|
|
func (s *DockerTrustSuite) TestTrustedRunFromBadTrustServer(c *check.C) {
|
|
// Windows does not support this functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
repoName := fmt.Sprintf("%v/dockerclievilrun/trusted:latest", privateRegistryURL)
|
|
evilLocalConfigDir, err := ioutil.TempDir("", "evilrun-local-config-dir")
|
|
if err != nil {
|
|
c.Fatalf("Failed to create local temp dir")
|
|
}
|
|
|
|
// tag the image and upload it to the private registry
|
|
dockerCmd(c, "tag", "busybox", repoName)
|
|
|
|
pushCmd := exec.Command(dockerBinary, "push", repoName)
|
|
s.trustedCmd(pushCmd)
|
|
out, _, err := runCommandWithOutput(pushCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running trusted push: %s\n%s", err, out)
|
|
}
|
|
if !strings.Contains(string(out), "Signing and pushing trust metadata") {
|
|
c.Fatalf("Missing expected output on trusted push:\n%s", out)
|
|
}
|
|
|
|
dockerCmd(c, "rmi", repoName)
|
|
|
|
// Try run
|
|
runCmd := exec.Command(dockerBinary, "run", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err = runCommandWithOutput(runCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running trusted run: %s\n%s", err, out)
|
|
}
|
|
|
|
if !strings.Contains(string(out), "Tagging") {
|
|
c.Fatalf("Missing expected output on trusted push:\n%s", out)
|
|
}
|
|
|
|
dockerCmd(c, "rmi", repoName)
|
|
|
|
// Kill the notary server, start a new "evil" one.
|
|
s.not.Close()
|
|
s.not, err = newTestNotary(c)
|
|
if err != nil {
|
|
c.Fatalf("Restarting notary server failed.")
|
|
}
|
|
|
|
// In order to make an evil server, lets re-init a client (with a different trust dir) and push new data.
|
|
// tag an image and upload it to the private registry
|
|
dockerCmd(c, "--config", evilLocalConfigDir, "tag", "busybox", repoName)
|
|
|
|
// Push up to the new server
|
|
pushCmd = exec.Command(dockerBinary, "--config", evilLocalConfigDir, "push", repoName)
|
|
s.trustedCmd(pushCmd)
|
|
out, _, err = runCommandWithOutput(pushCmd)
|
|
if err != nil {
|
|
c.Fatalf("Error running trusted push: %s\n%s", err, out)
|
|
}
|
|
if !strings.Contains(string(out), "Signing and pushing trust metadata") {
|
|
c.Fatalf("Missing expected output on trusted push:\n%s", out)
|
|
}
|
|
|
|
// Now, try running with the original client from this new trust server. This should fallback to our cached timestamp and metadata.
|
|
runCmd = exec.Command(dockerBinary, "run", repoName)
|
|
s.trustedCmd(runCmd)
|
|
out, _, err = runCommandWithOutput(runCmd)
|
|
|
|
if err != nil {
|
|
c.Fatalf("Error falling back to cached trust data: %s\n%s", err, out)
|
|
}
|
|
if !strings.Contains(string(out), "Error while downloading remote metadata, using cached timestamp") {
|
|
c.Fatalf("Missing expected output on trusted push:\n%s", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestPtraceContainerProcsFromHost(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux, SameHostDaemon)
|
|
|
|
out, _ := dockerCmd(c, "run", "-d", "busybox", "top")
|
|
id := strings.TrimSpace(out)
|
|
c.Assert(waitRun(id), check.IsNil)
|
|
pid1 := inspectField(c, id, "State.Pid")
|
|
|
|
_, err := os.Readlink(fmt.Sprintf("/proc/%s/ns/net", pid1))
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestAppArmorDeniesPtrace(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, SameHostDaemon, Apparmor, DaemonIsLinux, NotGCCGO)
|
|
|
|
// Run through 'sh' so we are NOT pid 1. Pid 1 may be able to trace
|
|
// itself, but pid>1 should not be able to trace pid1.
|
|
_, exitCode, _ := dockerCmdWithError("run", "busybox", "sh", "-c", "sh -c readlink /proc/1/ns/net")
|
|
if exitCode == 0 {
|
|
c.Fatal("ptrace was not successfully restricted by AppArmor")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestAppArmorTraceSelf(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux, SameHostDaemon, Apparmor)
|
|
|
|
_, exitCode, _ := dockerCmdWithError("run", "busybox", "readlink", "/proc/1/ns/net")
|
|
if exitCode != 0 {
|
|
c.Fatal("ptrace of self failed.")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestAppArmorDeniesChmodProc(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, SameHostDaemon, Apparmor, DaemonIsLinux, NotUserNamespace)
|
|
_, exitCode, _ := dockerCmdWithError("run", "busybox", "chmod", "744", "/proc/cpuinfo")
|
|
if exitCode == 0 {
|
|
// If our test failed, attempt to repair the host system...
|
|
_, exitCode, _ := dockerCmdWithError("run", "busybox", "chmod", "444", "/proc/cpuinfo")
|
|
if exitCode == 0 {
|
|
c.Fatal("AppArmor was unsuccessful in prohibiting chmod of /proc/* files.")
|
|
}
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunCapAddSYSTIME(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
dockerCmd(c, "run", "--cap-drop=ALL", "--cap-add=SYS_TIME", "busybox", "sh", "-c", "grep ^CapEff /proc/self/status | sed 's/^CapEff:\t//' | grep ^0000000002000000$")
|
|
}
|
|
|
|
// run create container failed should clean up the container
|
|
func (s *DockerSuite) TestRunCreateContainerFailedCleanUp(c *check.C) {
|
|
// TODO Windows. This may be possible to enable once link is supported
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "unique_name"
|
|
_, _, err := dockerCmdWithError("run", "--name", name, "--link", "nothing:nothing", "busybox")
|
|
c.Assert(err, check.NotNil, check.Commentf("Expected docker run to fail!"))
|
|
|
|
containerID, err := inspectFieldWithError(name, "Id")
|
|
c.Assert(err, checker.NotNil, check.Commentf("Expected not to have this container: %s!", containerID))
|
|
c.Assert(containerID, check.Equals, "", check.Commentf("Expected not to have this container: %s!", containerID))
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNamedVolume(c *check.C) {
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
testRequires(c, DaemonIsLinux)
|
|
dockerCmd(c, "run", "--name=test", "-v", "testing:"+prefix+"/foo", "busybox", "sh", "-c", "echo hello > "+prefix+"/foo/bar")
|
|
|
|
out, _ := dockerCmd(c, "run", "--volumes-from", "test", "busybox", "sh", "-c", "cat "+prefix+"/foo/bar")
|
|
c.Assert(strings.TrimSpace(out), check.Equals, "hello")
|
|
|
|
out, _ = dockerCmd(c, "run", "-v", "testing:"+prefix+"/foo", "busybox", "sh", "-c", "cat "+prefix+"/foo/bar")
|
|
c.Assert(strings.TrimSpace(out), check.Equals, "hello")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithUlimits(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
out, _ := dockerCmd(c, "run", "--name=testulimits", "--ulimit", "nofile=42", "busybox", "/bin/sh", "-c", "ulimit -n")
|
|
ul := strings.TrimSpace(out)
|
|
if ul != "42" {
|
|
c.Fatalf("expected `ulimit -n` to be 42, got %s", ul)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithCgroupParent(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
cgroupParent := "test"
|
|
name := "cgroup-test"
|
|
|
|
out, _, err := dockerCmdWithError("run", "--cgroup-parent", cgroupParent, "--name", name, "busybox", "cat", "/proc/self/cgroup")
|
|
if err != nil {
|
|
c.Fatalf("unexpected failure when running container with --cgroup-parent option - %s\n%v", string(out), err)
|
|
}
|
|
cgroupPaths := parseCgroupPaths(string(out))
|
|
if len(cgroupPaths) == 0 {
|
|
c.Fatalf("unexpected output - %q", string(out))
|
|
}
|
|
id, err := getIDByName(name)
|
|
c.Assert(err, check.IsNil)
|
|
expectedCgroup := path.Join(cgroupParent, id)
|
|
found := false
|
|
for _, path := range cgroupPaths {
|
|
if strings.HasSuffix(path, expectedCgroup) {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
if !found {
|
|
c.Fatalf("unexpected cgroup paths. Expected at least one cgroup path to have suffix %q. Cgroup Paths: %v", expectedCgroup, cgroupPaths)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithCgroupParentAbsPath(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
cgroupParent := "/cgroup-parent/test"
|
|
name := "cgroup-test"
|
|
out, _, err := dockerCmdWithError("run", "--cgroup-parent", cgroupParent, "--name", name, "busybox", "cat", "/proc/self/cgroup")
|
|
if err != nil {
|
|
c.Fatalf("unexpected failure when running container with --cgroup-parent option - %s\n%v", string(out), err)
|
|
}
|
|
cgroupPaths := parseCgroupPaths(string(out))
|
|
if len(cgroupPaths) == 0 {
|
|
c.Fatalf("unexpected output - %q", string(out))
|
|
}
|
|
id, err := getIDByName(name)
|
|
c.Assert(err, check.IsNil)
|
|
expectedCgroup := path.Join(cgroupParent, id)
|
|
found := false
|
|
for _, path := range cgroupPaths {
|
|
if strings.HasSuffix(path, expectedCgroup) {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
if !found {
|
|
c.Fatalf("unexpected cgroup paths. Expected at least one cgroup path to have suffix %q. Cgroup Paths: %v", expectedCgroup, cgroupPaths)
|
|
}
|
|
}
|
|
|
|
// TestRunInvalidCgroupParent checks that a specially-crafted cgroup parent doesn't cause Docker to crash or start modifying /.
|
|
func (s *DockerSuite) TestRunInvalidCgroupParent(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
cgroupParent := "../../../../../../../../SHOULD_NOT_EXIST"
|
|
cleanCgroupParent := "SHOULD_NOT_EXIST"
|
|
name := "cgroup-invalid-test"
|
|
|
|
out, _, err := dockerCmdWithError("run", "--cgroup-parent", cgroupParent, "--name", name, "busybox", "cat", "/proc/self/cgroup")
|
|
if err != nil {
|
|
// XXX: This may include a daemon crash.
|
|
c.Fatalf("unexpected failure when running container with --cgroup-parent option - %s\n%v", string(out), err)
|
|
}
|
|
|
|
// We expect "/SHOULD_NOT_EXIST" to not exist. If not, we have a security issue.
|
|
if _, err := os.Stat("/SHOULD_NOT_EXIST"); err == nil || !os.IsNotExist(err) {
|
|
c.Fatalf("SECURITY: --cgroup-parent with ../../ relative paths cause files to be created in the host (this is bad) !!")
|
|
}
|
|
|
|
cgroupPaths := parseCgroupPaths(string(out))
|
|
if len(cgroupPaths) == 0 {
|
|
c.Fatalf("unexpected output - %q", string(out))
|
|
}
|
|
id, err := getIDByName(name)
|
|
c.Assert(err, check.IsNil)
|
|
expectedCgroup := path.Join(cleanCgroupParent, id)
|
|
found := false
|
|
for _, path := range cgroupPaths {
|
|
if strings.HasSuffix(path, expectedCgroup) {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
if !found {
|
|
c.Fatalf("unexpected cgroup paths. Expected at least one cgroup path to have suffix %q. Cgroup Paths: %v", expectedCgroup, cgroupPaths)
|
|
}
|
|
}
|
|
|
|
// TestRunInvalidCgroupParent checks that a specially-crafted cgroup parent doesn't cause Docker to crash or start modifying /.
|
|
func (s *DockerSuite) TestRunAbsoluteInvalidCgroupParent(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
cgroupParent := "/../../../../../../../../SHOULD_NOT_EXIST"
|
|
cleanCgroupParent := "/SHOULD_NOT_EXIST"
|
|
name := "cgroup-absolute-invalid-test"
|
|
|
|
out, _, err := dockerCmdWithError("run", "--cgroup-parent", cgroupParent, "--name", name, "busybox", "cat", "/proc/self/cgroup")
|
|
if err != nil {
|
|
// XXX: This may include a daemon crash.
|
|
c.Fatalf("unexpected failure when running container with --cgroup-parent option - %s\n%v", string(out), err)
|
|
}
|
|
|
|
// We expect "/SHOULD_NOT_EXIST" to not exist. If not, we have a security issue.
|
|
if _, err := os.Stat("/SHOULD_NOT_EXIST"); err == nil || !os.IsNotExist(err) {
|
|
c.Fatalf("SECURITY: --cgroup-parent with /../../ garbage paths cause files to be created in the host (this is bad) !!")
|
|
}
|
|
|
|
cgroupPaths := parseCgroupPaths(string(out))
|
|
if len(cgroupPaths) == 0 {
|
|
c.Fatalf("unexpected output - %q", string(out))
|
|
}
|
|
id, err := getIDByName(name)
|
|
c.Assert(err, check.IsNil)
|
|
expectedCgroup := path.Join(cleanCgroupParent, id)
|
|
found := false
|
|
for _, path := range cgroupPaths {
|
|
if strings.HasSuffix(path, expectedCgroup) {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
if !found {
|
|
c.Fatalf("unexpected cgroup paths. Expected at least one cgroup path to have suffix %q. Cgroup Paths: %v", expectedCgroup, cgroupPaths)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerWithCgroupMountRO(c *check.C) {
|
|
// Not applicable on Windows as uses Unix specific functionality
|
|
// --read-only + userns has remount issues
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
|
|
filename := "/sys/fs/cgroup/devices/test123"
|
|
out, _, err := dockerCmdWithError("run", "busybox", "touch", filename)
|
|
if err == nil {
|
|
c.Fatal("expected cgroup mount point to be read-only, touch file should fail")
|
|
}
|
|
expected := "Read-only file system"
|
|
if !strings.Contains(out, expected) {
|
|
c.Fatalf("expected output from failure to contain %s but contains %s", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerNetworkModeToSelf(c *check.C) {
|
|
// Not applicable on Windows which does not support --net=container
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _, err := dockerCmdWithError("run", "--name=me", "--net=container:me", "busybox", "true")
|
|
if err == nil || !strings.Contains(out, "cannot join own network") {
|
|
c.Fatalf("using container net mode to self should result in an error\nerr: %q\nout: %s", err, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerNetModeWithDnsMacHosts(c *check.C) {
|
|
// Not applicable on Windows which does not support --net=container
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, _, err := dockerCmdWithError("run", "-d", "--name", "parent", "busybox", "top")
|
|
if err != nil {
|
|
c.Fatalf("failed to run container: %v, output: %q", err, out)
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "--dns", "1.2.3.4", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictNetworkAndDNS.Error()) {
|
|
c.Fatalf("run --net=container with --dns should error out")
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "--mac-address", "92:d0:c6:0a:29:33", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictContainerNetworkAndMac.Error()) {
|
|
c.Fatalf("run --net=container with --mac-address should error out")
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "--add-host", "test:192.168.2.109", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictNetworkHosts.Error()) {
|
|
c.Fatalf("run --net=container with --add-host should error out")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunContainerNetModeWithExposePort(c *check.C) {
|
|
// Not applicable on Windows which does not support --net=container
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
dockerCmd(c, "run", "-d", "--name", "parent", "busybox", "top")
|
|
|
|
out, _, err := dockerCmdWithError("run", "-p", "5000:5000", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictNetworkPublishPorts.Error()) {
|
|
c.Fatalf("run --net=container with -p should error out")
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "-P", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictNetworkPublishPorts.Error()) {
|
|
c.Fatalf("run --net=container with -P should error out")
|
|
}
|
|
|
|
out, _, err = dockerCmdWithError("run", "--expose", "5000", "--net=container:parent", "busybox")
|
|
if err == nil || !strings.Contains(out, runconfig.ErrConflictNetworkExposePorts.Error()) {
|
|
c.Fatalf("run --net=container with --expose should error out")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunLinkToContainerNetMode(c *check.C) {
|
|
// Not applicable on Windows which does not support --net=container or --link
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
dockerCmd(c, "run", "--name", "test", "-d", "busybox", "top")
|
|
dockerCmd(c, "run", "--name", "parent", "-d", "--net=container:test", "busybox", "top")
|
|
dockerCmd(c, "run", "-d", "--link=parent:parent", "busybox", "top")
|
|
dockerCmd(c, "run", "--name", "child", "-d", "--net=container:parent", "busybox", "top")
|
|
dockerCmd(c, "run", "-d", "--link=child:child", "busybox", "top")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunLoopbackOnlyExistsWhenNetworkingDisabled(c *check.C) {
|
|
// TODO Windows: This may be possible to convert.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "--net=none", "busybox", "ip", "-o", "-4", "a", "show", "up")
|
|
|
|
var (
|
|
count = 0
|
|
parts = strings.Split(out, "\n")
|
|
)
|
|
|
|
for _, l := range parts {
|
|
if l != "" {
|
|
count++
|
|
}
|
|
}
|
|
|
|
if count != 1 {
|
|
c.Fatalf("Wrong interface count in container %d", count)
|
|
}
|
|
|
|
if !strings.HasPrefix(out, "1: lo") {
|
|
c.Fatalf("Wrong interface in test container: expected [1: lo], got %s", out)
|
|
}
|
|
}
|
|
|
|
// Issue #4681
|
|
func (s *DockerSuite) TestRunLoopbackWhenNetworkDisabled(c *check.C) {
|
|
if daemonPlatform == "windows" {
|
|
dockerCmd(c, "run", "--net=none", WindowsBaseImage, "ping", "-n", "1", "127.0.0.1")
|
|
} else {
|
|
dockerCmd(c, "run", "--net=none", "busybox", "ping", "-c", "1", "127.0.0.1")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunModeNetContainerHostname(c *check.C) {
|
|
// Windows does not support --net=container
|
|
testRequires(c, DaemonIsLinux, ExecSupport, NotUserNamespace)
|
|
|
|
dockerCmd(c, "run", "-i", "-d", "--name", "parent", "busybox", "top")
|
|
out, _ := dockerCmd(c, "exec", "parent", "cat", "/etc/hostname")
|
|
out1, _ := dockerCmd(c, "run", "--net=container:parent", "busybox", "cat", "/etc/hostname")
|
|
|
|
if out1 != out {
|
|
c.Fatal("containers with shared net namespace should have same hostname")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNetworkNotInitializedNoneMode(c *check.C) {
|
|
// TODO Windows: Network settings are not currently propagated. This may
|
|
// be resolved in the future with the move to libnetwork and CNM.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, _ := dockerCmd(c, "run", "-d", "--net=none", "busybox", "top")
|
|
id := strings.TrimSpace(out)
|
|
res := inspectField(c, id, "NetworkSettings.Networks.none.IPAddress")
|
|
if res != "" {
|
|
c.Fatalf("For 'none' mode network must not be initialized, but container got IP: %s", res)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestTwoContainersInNetHost(c *check.C) {
|
|
// Not applicable as Windows does not support --net=host
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotUserNamespace)
|
|
dockerCmd(c, "run", "-d", "--net=host", "--name=first", "busybox", "top")
|
|
dockerCmd(c, "run", "-d", "--net=host", "--name=second", "busybox", "top")
|
|
dockerCmd(c, "stop", "first")
|
|
dockerCmd(c, "stop", "second")
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainersInUserDefinedNetwork(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork")
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
dockerCmd(c, "run", "-t", "--net=testnetwork", "--name=second", "busybox", "ping", "-c", "1", "first")
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainersInMultipleNetworks(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
// Create 2 networks using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork2")
|
|
// Run and connect containers to testnetwork1
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
// Check connectivity between containers in testnetwork2
|
|
dockerCmd(c, "exec", "first", "ping", "-c", "1", "second.testnetwork1")
|
|
// Connect containers to testnetwork2
|
|
dockerCmd(c, "network", "connect", "testnetwork2", "first")
|
|
dockerCmd(c, "network", "connect", "testnetwork2", "second")
|
|
// Check connectivity between containers
|
|
dockerCmd(c, "exec", "second", "ping", "-c", "1", "first.testnetwork2")
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainersNetworkIsolation(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
// Create 2 networks using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork2")
|
|
// Run 1 container in testnetwork1 and another in testnetwork2
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork2", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// Check Isolation between containers : ping must fail
|
|
_, _, err := dockerCmdWithError("exec", "first", "ping", "-c", "1", "second")
|
|
c.Assert(err, check.NotNil)
|
|
// Connect first container to testnetwork2
|
|
dockerCmd(c, "network", "connect", "testnetwork2", "first")
|
|
// ping must succeed now
|
|
_, _, err = dockerCmdWithError("exec", "first", "ping", "-c", "1", "second")
|
|
c.Assert(err, check.IsNil)
|
|
|
|
// Disconnect first container from testnetwork2
|
|
dockerCmd(c, "network", "disconnect", "testnetwork2", "first")
|
|
// ping must fail again
|
|
_, _, err = dockerCmdWithError("exec", "first", "ping", "-c", "1", "second")
|
|
c.Assert(err, check.NotNil)
|
|
}
|
|
|
|
func (s *DockerSuite) TestNetworkRmWithActiveContainers(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
// Create 2 networks using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
// Run and connect containers to testnetwork1
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
// Network delete with active containers must fail
|
|
_, _, err := dockerCmdWithError("network", "rm", "testnetwork1")
|
|
c.Assert(err, check.NotNil)
|
|
|
|
dockerCmd(c, "stop", "first")
|
|
_, _, err = dockerCmdWithError("network", "rm", "testnetwork1")
|
|
c.Assert(err, check.NotNil)
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainerRestartInMultipleNetworks(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace, NotArm)
|
|
// Create 2 networks using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork2")
|
|
|
|
// Run and connect containers to testnetwork1
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
// Check connectivity between containers in testnetwork2
|
|
dockerCmd(c, "exec", "first", "ping", "-c", "1", "second.testnetwork1")
|
|
// Connect containers to testnetwork2
|
|
dockerCmd(c, "network", "connect", "testnetwork2", "first")
|
|
dockerCmd(c, "network", "connect", "testnetwork2", "second")
|
|
// Check connectivity between containers
|
|
dockerCmd(c, "exec", "second", "ping", "-c", "1", "first.testnetwork2")
|
|
|
|
// Stop second container and test ping failures on both networks
|
|
dockerCmd(c, "stop", "second")
|
|
_, _, err := dockerCmdWithError("exec", "first", "ping", "-c", "1", "second.testnetwork1")
|
|
c.Assert(err, check.NotNil)
|
|
_, _, err = dockerCmdWithError("exec", "first", "ping", "-c", "1", "second.testnetwork2")
|
|
c.Assert(err, check.NotNil)
|
|
|
|
// Start second container and connectivity must be restored on both networks
|
|
dockerCmd(c, "start", "second")
|
|
dockerCmd(c, "exec", "first", "ping", "-c", "1", "second.testnetwork1")
|
|
dockerCmd(c, "exec", "second", "ping", "-c", "1", "first.testnetwork2")
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainerWithConflictingHostNetworks(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
// Run a container with --net=host
|
|
dockerCmd(c, "run", "-d", "--net=host", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
// Create a network using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
|
|
// Connecting to the user defined network must fail
|
|
_, _, err := dockerCmdWithError("network", "connect", "testnetwork1", "first")
|
|
c.Assert(err, check.NotNil)
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainerWithConflictingSharedNetwork(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
dockerCmd(c, "run", "-d", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
// Run second container in first container's network namespace
|
|
dockerCmd(c, "run", "-d", "--net=container:first", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// Create a network using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
|
|
// Connecting to the user defined network must fail
|
|
out, _, err := dockerCmdWithError("network", "connect", "testnetwork1", "second")
|
|
c.Assert(err, check.NotNil)
|
|
c.Assert(out, checker.Contains, runconfig.ErrConflictSharedNetwork.Error())
|
|
}
|
|
|
|
func (s *DockerSuite) TestContainerWithConflictingNoneNetwork(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
dockerCmd(c, "run", "-d", "--net=none", "--name=first", "busybox", "top")
|
|
c.Assert(waitRun("first"), check.IsNil)
|
|
|
|
// Create a network using bridge driver
|
|
dockerCmd(c, "network", "create", "-d", "bridge", "testnetwork1")
|
|
|
|
// Connecting to the user defined network must fail
|
|
out, _, err := dockerCmdWithError("network", "connect", "testnetwork1", "first")
|
|
c.Assert(err, check.NotNil)
|
|
c.Assert(out, checker.Contains, runconfig.ErrConflictNoNetwork.Error())
|
|
|
|
// create a container connected to testnetwork1
|
|
dockerCmd(c, "run", "-d", "--net=testnetwork1", "--name=second", "busybox", "top")
|
|
c.Assert(waitRun("second"), check.IsNil)
|
|
|
|
// Connect second container to none network. it must fail as well
|
|
_, _, err = dockerCmdWithError("network", "connect", "none", "second")
|
|
c.Assert(err, check.NotNil)
|
|
}
|
|
|
|
// #11957 - stdin with no tty does not exit if stdin is not closed even though container exited
|
|
func (s *DockerSuite) TestRunStdinBlockedAfterContainerExit(c *check.C) {
|
|
cmd := exec.Command(dockerBinary, "run", "-i", "--name=test", "busybox", "true")
|
|
in, err := cmd.StdinPipe()
|
|
c.Assert(err, check.IsNil)
|
|
defer in.Close()
|
|
c.Assert(cmd.Start(), check.IsNil)
|
|
|
|
waitChan := make(chan error)
|
|
go func() {
|
|
waitChan <- cmd.Wait()
|
|
}()
|
|
|
|
select {
|
|
case err := <-waitChan:
|
|
c.Assert(err, check.IsNil)
|
|
case <-time.After(30 * time.Second):
|
|
c.Fatal("timeout waiting for command to exit")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWrongCpusetCpusFlagValue(c *check.C) {
|
|
// TODO Windows: This needs validation (error out) in the daemon.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, exitCode, err := dockerCmdWithError("run", "--cpuset-cpus", "1-10,11--", "busybox", "true")
|
|
c.Assert(err, check.NotNil)
|
|
expected := "Error response from daemon: Invalid value 1-10,11-- for cpuset cpus.\n"
|
|
if !(strings.Contains(out, expected) || exitCode == 125) {
|
|
c.Fatalf("Expected output to contain %q with exitCode 125, got out: %q exitCode: %v", expected, out, exitCode)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWrongCpusetMemsFlagValue(c *check.C) {
|
|
// TODO Windows: This needs validation (error out) in the daemon.
|
|
testRequires(c, DaemonIsLinux)
|
|
out, exitCode, err := dockerCmdWithError("run", "--cpuset-mems", "1-42--", "busybox", "true")
|
|
c.Assert(err, check.NotNil)
|
|
expected := "Error response from daemon: Invalid value 1-42-- for cpuset mems.\n"
|
|
if !(strings.Contains(out, expected) || exitCode == 125) {
|
|
c.Fatalf("Expected output to contain %q with exitCode 125, got out: %q exitCode: %v", expected, out, exitCode)
|
|
}
|
|
}
|
|
|
|
// TestRunNonExecutableCmd checks that 'docker run busybox foo' exits with error code 127'
|
|
func (s *DockerSuite) TestRunNonExecutableCmd(c *check.C) {
|
|
name := "testNonExecutableCmd"
|
|
runCmd := exec.Command(dockerBinary, "run", "--name", name, "busybox", "foo")
|
|
_, exit, _ := runCommandWithOutput(runCmd)
|
|
stateExitCode := findContainerExitCode(c, name)
|
|
if !(exit == 127 && strings.Contains(stateExitCode, "127")) {
|
|
c.Fatalf("Run non-executable command should have errored with exit code 127, but we got exit: %d, State.ExitCode: %s", exit, stateExitCode)
|
|
}
|
|
}
|
|
|
|
// TestRunNonExistingCmd checks that 'docker run busybox /bin/foo' exits with code 127.
|
|
func (s *DockerSuite) TestRunNonExistingCmd(c *check.C) {
|
|
name := "testNonExistingCmd"
|
|
runCmd := exec.Command(dockerBinary, "run", "--name", name, "busybox", "/bin/foo")
|
|
_, exit, _ := runCommandWithOutput(runCmd)
|
|
stateExitCode := findContainerExitCode(c, name)
|
|
if !(exit == 127 && strings.Contains(stateExitCode, "127")) {
|
|
c.Fatalf("Run non-existing command should have errored with exit code 127, but we got exit: %d, State.ExitCode: %s", exit, stateExitCode)
|
|
}
|
|
}
|
|
|
|
// TestCmdCannotBeInvoked checks that 'docker run busybox /etc' exits with 126, or
|
|
// 127 on Windows. The difference is that in Windows, the container must be started
|
|
// as that's when the check is made (and yes, by it's design...)
|
|
func (s *DockerSuite) TestCmdCannotBeInvoked(c *check.C) {
|
|
expected := 126
|
|
if daemonPlatform == "windows" {
|
|
expected = 127
|
|
}
|
|
name := "testCmdCannotBeInvoked"
|
|
runCmd := exec.Command(dockerBinary, "run", "--name", name, "busybox", "/etc")
|
|
_, exit, _ := runCommandWithOutput(runCmd)
|
|
stateExitCode := findContainerExitCode(c, name)
|
|
if !(exit == expected && strings.Contains(stateExitCode, strconv.Itoa(expected))) {
|
|
c.Fatalf("Run cmd that cannot be invoked should have errored with code %d, but we got exit: %d, State.ExitCode: %s", expected, exit, stateExitCode)
|
|
}
|
|
}
|
|
|
|
// TestRunNonExistingImage checks that 'docker run foo' exits with error msg 125 and contains 'Unable to find image'
|
|
func (s *DockerSuite) TestRunNonExistingImage(c *check.C) {
|
|
runCmd := exec.Command(dockerBinary, "run", "foo")
|
|
out, exit, err := runCommandWithOutput(runCmd)
|
|
if !(err != nil && exit == 125 && strings.Contains(out, "Unable to find image")) {
|
|
c.Fatalf("Run non-existing image should have errored with 'Unable to find image' code 125, but we got out: %s, exit: %d, err: %s", out, exit, err)
|
|
}
|
|
}
|
|
|
|
// TestDockerFails checks that 'docker run -foo busybox' exits with 125 to signal docker run failed
|
|
func (s *DockerSuite) TestDockerFails(c *check.C) {
|
|
runCmd := exec.Command(dockerBinary, "run", "-foo", "busybox")
|
|
out, exit, err := runCommandWithOutput(runCmd)
|
|
if !(err != nil && exit == 125) {
|
|
c.Fatalf("Docker run with flag not defined should exit with 125, but we got out: %s, exit: %d, err: %s", out, exit, err)
|
|
}
|
|
}
|
|
|
|
// TestRunInvalidReference invokes docker run with a bad reference.
|
|
func (s *DockerSuite) TestRunInvalidReference(c *check.C) {
|
|
out, exit, _ := dockerCmdWithError("run", "busybox@foo")
|
|
if exit == 0 {
|
|
c.Fatalf("expected non-zero exist code; received %d", exit)
|
|
}
|
|
|
|
if !strings.Contains(out, "Error parsing reference") {
|
|
c.Fatalf(`Expected "Error parsing reference" in output; got: %s`, out)
|
|
}
|
|
}
|
|
|
|
// Test fix for issue #17854
|
|
func (s *DockerSuite) TestRunInitLayerPathOwnership(c *check.C) {
|
|
// Not applicable on Windows as it does not support Linux uid/gid ownership
|
|
testRequires(c, DaemonIsLinux)
|
|
name := "testetcfileownership"
|
|
_, err := buildImage(name,
|
|
`FROM busybox
|
|
RUN echo 'dockerio:x:1001:1001::/bin:/bin/false' >> /etc/passwd
|
|
RUN echo 'dockerio:x:1001:' >> /etc/group
|
|
RUN chown dockerio:dockerio /etc`,
|
|
true)
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Test that dockerio ownership of /etc is retained at runtime
|
|
out, _ := dockerCmd(c, "run", "--rm", name, "stat", "-c", "%U:%G", "/etc")
|
|
out = strings.TrimSpace(out)
|
|
if out != "dockerio:dockerio" {
|
|
c.Fatalf("Wrong /etc ownership: expected dockerio:dockerio, got %q", out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithOomScoreAdj(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
expected := "642"
|
|
out, _ := dockerCmd(c, "run", "--oom-score-adj", expected, "busybox", "cat", "/proc/self/oom_score_adj")
|
|
oomScoreAdj := strings.TrimSpace(out)
|
|
if oomScoreAdj != "642" {
|
|
c.Fatalf("Expected oom_score_adj set to %q, got %q instead", expected, oomScoreAdj)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunWithOomScoreAdjInvalidRange(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
out, _, err := dockerCmdWithError("run", "--oom-score-adj", "1001", "busybox", "true")
|
|
c.Assert(err, check.NotNil)
|
|
expected := "Invalid value 1001, range for oom score adj is [-1000, 1000]."
|
|
if !strings.Contains(out, expected) {
|
|
c.Fatalf("Expected output to contain %q, got %q instead", expected, out)
|
|
}
|
|
out, _, err = dockerCmdWithError("run", "--oom-score-adj", "-1001", "busybox", "true")
|
|
c.Assert(err, check.NotNil)
|
|
expected = "Invalid value -1001, range for oom score adj is [-1000, 1000]."
|
|
if !strings.Contains(out, expected) {
|
|
c.Fatalf("Expected output to contain %q, got %q instead", expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesMountedAsShared(c *check.C) {
|
|
// Volume propagation is linux only. Also it creates directories for
|
|
// bind mounting, so needs to be same host.
|
|
testRequires(c, DaemonIsLinux, SameHostDaemon, NotUserNamespace)
|
|
|
|
// Prepare a source directory to bind mount
|
|
tmpDir, err := ioutil.TempDir("", "volume-source")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir)
|
|
|
|
if err := os.Mkdir(path.Join(tmpDir, "mnt1"), 0755); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Convert this directory into a shared mount point so that we do
|
|
// not rely on propagation properties of parent mount.
|
|
cmd := exec.Command("mount", "--bind", tmpDir, tmpDir)
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
cmd = exec.Command("mount", "--make-private", "--make-shared", tmpDir)
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "--privileged", "-v", fmt.Sprintf("%s:/volume-dest:shared", tmpDir), "busybox", "mount", "--bind", "/volume-dest/mnt1", "/volume-dest/mnt1")
|
|
|
|
// Make sure a bind mount under a shared volume propagated to host.
|
|
if mounted, _ := mount.Mounted(path.Join(tmpDir, "mnt1")); !mounted {
|
|
c.Fatalf("Bind mount under shared volume did not propagate to host")
|
|
}
|
|
|
|
mount.Unmount(path.Join(tmpDir, "mnt1"))
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumesMountedAsSlave(c *check.C) {
|
|
// Volume propagation is linux only. Also it creates directories for
|
|
// bind mounting, so needs to be same host.
|
|
testRequires(c, DaemonIsLinux, SameHostDaemon, NotUserNamespace)
|
|
|
|
// Prepare a source directory to bind mount
|
|
tmpDir, err := ioutil.TempDir("", "volume-source")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir)
|
|
|
|
if err := os.Mkdir(path.Join(tmpDir, "mnt1"), 0755); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Prepare a source directory with file in it. We will bind mount this
|
|
// direcotry and see if file shows up.
|
|
tmpDir2, err := ioutil.TempDir("", "volume-source2")
|
|
if err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
defer os.RemoveAll(tmpDir2)
|
|
|
|
if err := ioutil.WriteFile(path.Join(tmpDir2, "slave-testfile"), []byte("Test"), 0644); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
// Convert this directory into a shared mount point so that we do
|
|
// not rely on propagation properties of parent mount.
|
|
cmd := exec.Command("mount", "--bind", tmpDir, tmpDir)
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
cmd = exec.Command("mount", "--make-private", "--make-shared", tmpDir)
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
dockerCmd(c, "run", "-i", "-d", "--name", "parent", "-v", fmt.Sprintf("%s:/volume-dest:slave", tmpDir), "busybox", "top")
|
|
|
|
// Bind mount tmpDir2/ onto tmpDir/mnt1. If mount propagates inside
|
|
// container then contents of tmpDir2/slave-testfile should become
|
|
// visible at "/volume-dest/mnt1/slave-testfile"
|
|
cmd = exec.Command("mount", "--bind", tmpDir2, path.Join(tmpDir, "mnt1"))
|
|
if _, err = runCommand(cmd); err != nil {
|
|
c.Fatal(err)
|
|
}
|
|
|
|
out, _ := dockerCmd(c, "exec", "parent", "cat", "/volume-dest/mnt1/slave-testfile")
|
|
|
|
mount.Unmount(path.Join(tmpDir, "mnt1"))
|
|
|
|
if out != "Test" {
|
|
c.Fatalf("Bind mount under slave volume did not propagate to container")
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNamedVolumesMountedAsShared(c *check.C) {
|
|
testRequires(c, DaemonIsLinux, NotUserNamespace)
|
|
out, exitcode, _ := dockerCmdWithError("run", "-v", "foo:/test:shared", "busybox", "touch", "/test/somefile")
|
|
|
|
if exitcode == 0 {
|
|
c.Fatalf("expected non-zero exit code; received %d", exitcode)
|
|
}
|
|
|
|
if expected := "Invalid volume specification"; !strings.Contains(out, expected) {
|
|
c.Fatalf(`Expected %q in output; got: %s`, expected, out)
|
|
}
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNamedVolumeCopyImageData(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
testImg := "testvolumecopy"
|
|
_, err := buildImage(testImg, `
|
|
FROM busybox
|
|
RUN mkdir -p /foo && echo hello > /foo/hello
|
|
`, true)
|
|
c.Assert(err, check.IsNil)
|
|
|
|
dockerCmd(c, "run", "-v", "foo:/foo", testImg)
|
|
out, _ := dockerCmd(c, "run", "-v", "foo:/foo", "busybox", "cat", "/foo/hello")
|
|
c.Assert(strings.TrimSpace(out), check.Equals, "hello")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNamedVolumeNotRemoved(c *check.C) {
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
|
|
dockerCmd(c, "volume", "create", "--name", "test")
|
|
|
|
dockerCmd(c, "run", "--rm", "-v", "test:"+prefix+"/foo", "-v", prefix+"/bar", "busybox", "true")
|
|
dockerCmd(c, "volume", "inspect", "test")
|
|
out, _ := dockerCmd(c, "volume", "ls", "-q")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "test")
|
|
|
|
dockerCmd(c, "run", "--name=test", "-v", "test:"+prefix+"/foo", "-v", prefix+"/bar", "busybox", "true")
|
|
dockerCmd(c, "rm", "-fv", "test")
|
|
dockerCmd(c, "volume", "inspect", "test")
|
|
out, _ = dockerCmd(c, "volume", "ls", "-q")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "test")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunNamedVolumesFromNotRemoved(c *check.C) {
|
|
prefix, _ := getPrefixAndSlashFromDaemonPlatform()
|
|
|
|
dockerCmd(c, "volume", "create", "--name", "test")
|
|
dockerCmd(c, "run", "--name=parent", "-v", "test:"+prefix+"/foo", "-v", prefix+"/bar", "busybox", "true")
|
|
dockerCmd(c, "run", "--name=child", "--volumes-from=parent", "busybox", "true")
|
|
|
|
// Remove the parent so there are not other references to the volumes
|
|
dockerCmd(c, "rm", "-f", "parent")
|
|
// now remove the child and ensure the named volume (and only the named volume) still exists
|
|
dockerCmd(c, "rm", "-fv", "child")
|
|
dockerCmd(c, "volume", "inspect", "test")
|
|
out, _ := dockerCmd(c, "volume", "ls", "-q")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "test")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunAttachFailedNoLeak(c *check.C) {
|
|
nroutines, err := getGoroutineNumber()
|
|
c.Assert(err, checker.IsNil)
|
|
|
|
runSleepingContainer(c, "--name=test", "-p", "8000:8000")
|
|
|
|
// Wait until container is fully up and running
|
|
c.Assert(waitRun("test"), check.IsNil)
|
|
|
|
out, _, err := dockerCmdWithError("run", "-p", "8000:8000", "busybox", "true")
|
|
c.Assert(err, checker.NotNil)
|
|
// check for windows error as well
|
|
// TODO Windows Post TP5. Fix the error message string
|
|
c.Assert(strings.Contains(string(out), "port is already allocated") ||
|
|
strings.Contains(string(out), "were not connected because a duplicate name exists") ||
|
|
strings.Contains(string(out), "HNS failed with error : Failed to create endpoint"), checker.Equals, true, check.Commentf("Output: %s", out))
|
|
dockerCmd(c, "rm", "-f", "test")
|
|
|
|
// NGoroutines is not updated right away, so we need to wait before failing
|
|
c.Assert(waitForGoroutines(nroutines), checker.IsNil)
|
|
}
|
|
|
|
// Test for one character directory name case (#20122)
|
|
func (s *DockerSuite) TestRunVolumeWithOneCharacter(c *check.C) {
|
|
testRequires(c, DaemonIsLinux)
|
|
|
|
out, _ := dockerCmd(c, "run", "-v", "/tmp/q:/foo", "busybox", "sh", "-c", "find /foo")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "/foo")
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunVolumeCopyFlag(c *check.C) {
|
|
testRequires(c, DaemonIsLinux) // Windows does not support copying data from image to the volume
|
|
_, err := buildImage("volumecopy",
|
|
`FROM busybox
|
|
RUN mkdir /foo && echo hello > /foo/bar
|
|
CMD cat /foo/bar`,
|
|
true,
|
|
)
|
|
c.Assert(err, checker.IsNil)
|
|
|
|
dockerCmd(c, "volume", "create", "--name=test")
|
|
|
|
// test with the nocopy flag
|
|
out, _, err := dockerCmdWithError("run", "-v", "test:/foo:nocopy", "volumecopy")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
// test default behavior which is to copy for non-binds
|
|
out, _ = dockerCmd(c, "run", "-v", "test:/foo", "volumecopy")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "hello")
|
|
// error out when the volume is already populated
|
|
out, _, err = dockerCmdWithError("run", "-v", "test:/foo:copy", "volumecopy")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
// do not error out when copy isn't explicitly set even though it's already populated
|
|
out, _ = dockerCmd(c, "run", "-v", "test:/foo", "volumecopy")
|
|
c.Assert(strings.TrimSpace(out), checker.Equals, "hello")
|
|
|
|
// do not allow copy modes on volumes-from
|
|
dockerCmd(c, "run", "--name=test", "-v", "/foo", "busybox", "true")
|
|
out, _, err = dockerCmdWithError("run", "--volumes-from=test:copy", "busybox", "true")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
out, _, err = dockerCmdWithError("run", "--volumes-from=test:nocopy", "busybox", "true")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
|
|
// do not allow copy modes on binds
|
|
out, _, err = dockerCmdWithError("run", "-v", "/foo:/bar:copy", "busybox", "true")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
out, _, err = dockerCmdWithError("run", "-v", "/foo:/bar:nocopy", "busybox", "true")
|
|
c.Assert(err, checker.NotNil, check.Commentf(out))
|
|
}
|
|
|
|
func (s *DockerSuite) TestRunTooLongHostname(c *check.C) {
|
|
// Test case in #21445
|
|
hostname1 := "this-is-a-way-too-long-hostname-but-it-should-give-a-nice-error.local"
|
|
out, _, err := dockerCmdWithError("run", "--hostname", hostname1, "busybox", "echo", "test")
|
|
c.Assert(err, checker.NotNil, check.Commentf("Expected docker run to fail!"))
|
|
c.Assert(out, checker.Contains, "invalid hostname format for --hostname:", check.Commentf("Expected to have 'invalid hostname format for --hostname:' in the output, get: %s!", out))
|
|
|
|
// HOST_NAME_MAX=64 so 65 bytes will fail
|
|
hostname2 := "this-is-a-hostname-with-65-bytes-so-it-should-give-an-error.local"
|
|
out, _, err = dockerCmdWithError("run", "--hostname", hostname2, "busybox", "echo", "test")
|
|
c.Assert(err, checker.NotNil, check.Commentf("Expected docker run to fail!"))
|
|
c.Assert(out, checker.Contains, "invalid hostname format for --hostname:", check.Commentf("Expected to have 'invalid hostname format for --hostname:' in the output, get: %s!", out))
|
|
|
|
// 64 bytes will be OK
|
|
hostname3 := "this-is-a-hostname-with-64-bytes-so-will-not-give-an-error.local"
|
|
dockerCmd(c, "run", "--hostname", hostname3, "busybox", "echo", "test")
|
|
}
|