summaryrefslogtreecommitdiffhomepage
path: root/runsc
diff options
context:
space:
mode:
Diffstat (limited to 'runsc')
-rw-r--r--runsc/boot/fs.go9
-rw-r--r--runsc/boot/loader.go12
-rw-r--r--runsc/boot/loader_test.go2
-rw-r--r--runsc/boot/vfs.go116
-rw-r--r--runsc/cmd/spec.go220
-rw-r--r--runsc/container/multi_container_test.go184
-rw-r--r--runsc/fsgofer/filter/config.go1
-rw-r--r--runsc/fsgofer/fsgofer.go60
-rw-r--r--runsc/sandbox/network.go1
9 files changed, 368 insertions, 237 deletions
diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go
index e83584b82..59639ba19 100644
--- a/runsc/boot/fs.go
+++ b/runsc/boot/fs.go
@@ -29,6 +29,7 @@ import (
_ "gvisor.dev/gvisor/pkg/sentry/fs/sys"
_ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs"
_ "gvisor.dev/gvisor/pkg/sentry/fs/tty"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
@@ -390,6 +391,10 @@ type mountHint struct {
// root is the inode where the volume is mounted. For mounts with 'pod' share
// the volume is mounted once and then bind mounted inside the containers.
root *fs.Inode
+
+ // vfsMount is the master mount for the volume. For mounts with 'pod' share
+ // the master volume is bind mounted inside the containers.
+ vfsMount *vfs.Mount
}
func (m *mountHint) setField(key, val string) error {
@@ -571,9 +576,9 @@ func newContainerMounter(spec *specs.Spec, goferFDs []int, k *kernel.Kernel, hin
// processHints processes annotations that container hints about how volumes
// should be mounted (e.g. a volume shared between containers). It must be
// called for the root container only.
-func (c *containerMounter) processHints(conf *Config) error {
+func (c *containerMounter) processHints(conf *Config, creds *auth.Credentials) error {
if conf.VFS2 {
- return nil
+ return c.processHintsVFS2(conf, creds)
}
ctx := c.k.SupervisorContext()
for _, hint := range c.hints.mounts {
diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go
index b5df1deb9..0c0423ab2 100644
--- a/runsc/boot/loader.go
+++ b/runsc/boot/loader.go
@@ -227,9 +227,7 @@ func New(args Args) (*Loader, error) {
// Create VDSO.
//
// Pass k as the platform since it is savable, unlike the actual platform.
- //
- // FIXME(b/109889800): Use non-nil context.
- vdso, err := loader.PrepareVDSO(nil, k)
+ vdso, err := loader.PrepareVDSO(k)
if err != nil {
return nil, fmt.Errorf("creating vdso: %v", err)
}
@@ -300,6 +298,12 @@ func New(args Args) (*Loader, error) {
return nil, fmt.Errorf("initializing kernel: %v", err)
}
+ if kernel.VFS2Enabled {
+ if err := registerFilesystems(k); err != nil {
+ return nil, fmt.Errorf("registering filesystems: %w", err)
+ }
+ }
+
if err := adjustDirentCache(k); err != nil {
return nil, err
}
@@ -561,7 +565,7 @@ func (l *Loader) run() error {
l.startGoferMonitor(l.sandboxID, l.goferFDs)
mntr := newContainerMounter(l.spec, l.goferFDs, l.k, l.mountHints)
- if err := mntr.processHints(l.conf); err != nil {
+ if err := mntr.processHints(l.conf, l.rootProcArgs.Credentials); err != nil {
return err
}
if err := setupContainerFS(ctx, l.conf, mntr, &l.rootProcArgs); err != nil {
diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go
index e448fd773..b723e4335 100644
--- a/runsc/boot/loader_test.go
+++ b/runsc/boot/loader_test.go
@@ -480,7 +480,7 @@ func TestCreateMountNamespaceVFS2(t *testing.T) {
defer loaderCleanup()
mntr := newContainerMounter(l.spec, l.goferFDs, l.k, l.mountHints)
- if err := mntr.processHints(l.conf); err != nil {
+ if err := mntr.processHints(l.conf, l.rootProcArgs.Credentials); err != nil {
t.Fatalf("failed process hints: %v", err)
}
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index b68117867..6ee6fae04 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -43,7 +43,11 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
-func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error {
+func registerFilesystems(k *kernel.Kernel) error {
+ ctx := k.SupervisorContext()
+ creds := auth.NewRootCredentials(k.RootUserNamespace())
+ vfsObj := k.VFS()
+
vfsObj.MustRegisterFilesystemType(devpts.Name, &devpts.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserList: true,
// TODO(b/29356795): Users may mount this once the terminals are in a
@@ -113,9 +117,6 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre
}
func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error {
- if err := mntr.k.VFS().Init(); err != nil {
- return fmt.Errorf("failed to initialize VFS: %w", err)
- }
mns, err := mntr.setupVFS2(ctx, conf, procArgs)
if err != nil {
return fmt.Errorf("failed to setupFS: %w", err)
@@ -144,10 +145,6 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs
rootProcArgs.MaxSymlinkTraversals = linux.MaxSymlinkTraversals
rootCtx := procArgs.NewContext(c.k)
- if err := registerFilesystems(rootCtx, c.k.VFS(), rootCreds); err != nil {
- return nil, fmt.Errorf("register filesystems: %w", err)
- }
-
mns, err := c.createMountNamespaceVFS2(rootCtx, conf, rootCreds)
if err != nil {
return nil, fmt.Errorf("creating mount namespace: %w", err)
@@ -182,8 +179,14 @@ func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config,
for i := range mounts {
submount := &mounts[i]
log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options)
- if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil {
- return err
+ if hint := c.hints.findMount(submount.Mount); hint != nil && hint.isSupported() {
+ if err := c.mountSharedSubmountVFS2(ctx, conf, mns, creds, submount.Mount, hint); err != nil {
+ return fmt.Errorf("mount shared mount %q to %q: %v", hint.name, submount.Destination, err)
+ }
+ } else {
+ if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil {
+ return fmt.Errorf("mount submount %q: %w", submount.Destination, err)
+ }
}
}
@@ -257,20 +260,18 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config,
// getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values
// used for mounts.
func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndFD) (string, *vfs.MountOptions, error) {
- var (
- fsName string
- data []string
- )
+ fsName := m.Type
+ var data []string
// Find filesystem name and FS specific data field.
switch m.Type {
case devpts.Name, devtmpfs.Name, proc.Name, sys.Name:
- fsName = m.Type
+ // Nothing to do.
+
case nonefs:
fsName = sys.Name
- case tmpfs.Name:
- fsName = m.Type
+ case tmpfs.Name:
var err error
data, err = parseAndFilterOptions(m.Options, tmpfsAllowedData...)
if err != nil {
@@ -279,10 +280,16 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndF
case bind:
fsName = gofer.Name
+ if m.fd == 0 {
+ // Check that an FD was provided to fails fast. Technically FD=0 is valid,
+ // but unlikely to be correct in this context.
+ return "", nil, fmt.Errorf("9P mount requires a connection FD")
+ }
data = p9MountData(m.fd, c.getMountAccessType(m.Mount), true /* vfs2 */)
default:
log.Warningf("ignoring unknown filesystem type %q", m.Type)
+ return "", nil, nil
}
opts := &vfs.MountOptions{
@@ -322,7 +329,7 @@ func (c *containerMounter) makeSyntheticMount(ctx context.Context, currentPath s
}
_, err := c.k.VFS().StatAt(ctx, creds, target, &vfs.StatOptions{})
if err == nil {
- // Mount point exists, nothing else to do.
+ log.Debugf("Mount point %q already exists", currentPath)
return nil
}
if err != syserror.ENOENT {
@@ -400,3 +407,76 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *Config, creds
return fmt.Errorf(`stating "/tmp" inside container: %w`, err)
}
}
+
+// processHintsVFS2 processes annotations that container hints about how volumes
+// should be mounted (e.g. a volume shared between containers). It must be
+// called for the root container only.
+func (c *containerMounter) processHintsVFS2(conf *Config, creds *auth.Credentials) error {
+ ctx := c.k.SupervisorContext()
+ for _, hint := range c.hints.mounts {
+ // TODO(b/142076984): Only support tmpfs for now. Bind mounts require a
+ // common gofer to mount all shared volumes.
+ if hint.mount.Type != tmpfs.Name {
+ continue
+ }
+
+ log.Infof("Mounting master of shared mount %q from %q type %q", hint.name, hint.mount.Source, hint.mount.Type)
+ mnt, err := c.mountSharedMasterVFS2(ctx, conf, hint, creds)
+ if err != nil {
+ return fmt.Errorf("mounting shared master %q: %v", hint.name, err)
+ }
+ hint.vfsMount = mnt
+ }
+ return nil
+}
+
+// mountSharedMasterVFS2 mounts the master of a volume that is shared among
+// containers in a pod.
+func (c *containerMounter) mountSharedMasterVFS2(ctx context.Context, conf *Config, hint *mountHint, creds *auth.Credentials) (*vfs.Mount, error) {
+ // Map mount type to filesystem name, and parse out the options that we are
+ // capable of dealing with.
+ mntFD := &mountAndFD{Mount: hint.mount}
+ fsName, opts, err := c.getMountNameAndOptionsVFS2(conf, mntFD)
+ if err != nil {
+ return nil, err
+ }
+ if len(fsName) == 0 {
+ return nil, fmt.Errorf("mount type not supported %q", hint.mount.Type)
+ }
+ return c.k.VFS().MountDisconnected(ctx, creds, "", fsName, opts)
+}
+
+// mountSharedSubmount binds mount to a previously mounted volume that is shared
+// among containers in the same pod.
+func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, mount specs.Mount, source *mountHint) error {
+ if err := source.checkCompatible(mount); err != nil {
+ return err
+ }
+
+ _, opts, err := c.getMountNameAndOptionsVFS2(conf, &mountAndFD{Mount: mount})
+ if err != nil {
+ return err
+ }
+ newMnt, err := c.k.VFS().NewDisconnectedMount(source.vfsMount.Filesystem(), source.vfsMount.Root(), opts)
+ if err != nil {
+ return err
+ }
+ defer newMnt.DecRef()
+
+ root := mns.Root()
+ defer root.DecRef()
+ if err := c.makeSyntheticMount(ctx, mount.Destination, root, creds); err != nil {
+ return err
+ }
+
+ target := &vfs.PathOperation{
+ Root: root,
+ Start: root,
+ Path: fspath.Parse(mount.Destination),
+ }
+ if err := c.k.VFS().ConnectMountAt(ctx, creds, newMnt, target); err != nil {
+ return err
+ }
+ log.Infof("Mounted %q type shared bind to %q", mount.Destination, source.name)
+ return nil
+}
diff --git a/runsc/cmd/spec.go b/runsc/cmd/spec.go
index a2b0a4b14..55194e641 100644
--- a/runsc/cmd/spec.go
+++ b/runsc/cmd/spec.go
@@ -16,124 +16,122 @@ package cmd
import (
"context"
- "fmt"
- "io/ioutil"
+ "encoding/json"
+ "io"
"os"
"path/filepath"
"github.com/google/subcommands"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/runsc/flag"
)
-func genSpec(cwd string) []byte {
- var template = fmt.Sprintf(`{
- "ociVersion": "1.0.0",
- "process": {
- "terminal": true,
- "user": {
- "uid": 0,
- "gid": 0
- },
- "args": [
- "sh"
- ],
- "env": [
- "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
- "TERM=xterm"
- ],
- "cwd": "%s",
- "capabilities": {
- "bounding": [
- "CAP_AUDIT_WRITE",
- "CAP_KILL",
- "CAP_NET_BIND_SERVICE"
- ],
- "effective": [
- "CAP_AUDIT_WRITE",
- "CAP_KILL",
- "CAP_NET_BIND_SERVICE"
- ],
- "inheritable": [
- "CAP_AUDIT_WRITE",
- "CAP_KILL",
- "CAP_NET_BIND_SERVICE"
- ],
- "permitted": [
- "CAP_AUDIT_WRITE",
- "CAP_KILL",
- "CAP_NET_BIND_SERVICE"
- ],
- "ambient": [
- "CAP_AUDIT_WRITE",
- "CAP_KILL",
- "CAP_NET_BIND_SERVICE"
- ]
- },
- "rlimits": [
- {
- "type": "RLIMIT_NOFILE",
- "hard": 1024,
- "soft": 1024
- }
- ]
- },
- "root": {
- "path": "rootfs",
- "readonly": true
- },
- "hostname": "runsc",
- "mounts": [
- {
- "destination": "/proc",
- "type": "proc",
- "source": "proc"
+func writeSpec(w io.Writer, cwd string, netns string, args []string) error {
+ spec := &specs.Spec{
+ Version: "1.0.0",
+ Process: &specs.Process{
+ Terminal: true,
+ User: specs.User{
+ UID: 0,
+ GID: 0,
+ },
+ Args: args,
+ Env: []string{
+ "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
+ "TERM=xterm",
+ },
+ Cwd: cwd,
+ Capabilities: &specs.LinuxCapabilities{
+ Bounding: []string{
+ "CAP_AUDIT_WRITE",
+ "CAP_KILL",
+ "CAP_NET_BIND_SERVICE",
+ },
+ Effective: []string{
+ "CAP_AUDIT_WRITE",
+ "CAP_KILL",
+ "CAP_NET_BIND_SERVICE",
+ },
+ Inheritable: []string{
+ "CAP_AUDIT_WRITE",
+ "CAP_KILL",
+ "CAP_NET_BIND_SERVICE",
+ },
+ Permitted: []string{
+ "CAP_AUDIT_WRITE",
+ "CAP_KILL",
+ "CAP_NET_BIND_SERVICE",
+ },
+ // TODO(gvisor.dev/issue/3166): support ambient capabilities
+ },
+ Rlimits: []specs.POSIXRlimit{
+ {
+ Type: "RLIMIT_NOFILE",
+ Hard: 1024,
+ Soft: 1024,
+ },
+ },
},
- {
- "destination": "/dev",
- "type": "tmpfs",
- "source": "tmpfs",
- "options": []
+ Root: &specs.Root{
+ Path: "rootfs",
+ Readonly: true,
},
- {
- "destination": "/sys",
- "type": "sysfs",
- "source": "sysfs",
- "options": [
- "nosuid",
- "noexec",
- "nodev",
- "ro"
- ]
- }
- ],
- "linux": {
- "namespaces": [
+ Hostname: "runsc",
+ Mounts: []specs.Mount{
{
- "type": "pid"
+ Destination: "/proc",
+ Type: "proc",
+ Source: "proc",
},
{
- "type": "network"
+ Destination: "/dev",
+ Type: "tmpfs",
+ Source: "tmpfs",
},
{
- "type": "ipc"
+ Destination: "/sys",
+ Type: "sysfs",
+ Source: "sysfs",
+ Options: []string{
+ "nosuid",
+ "noexec",
+ "nodev",
+ "ro",
+ },
},
- {
- "type": "uts"
+ },
+ Linux: &specs.Linux{
+ Namespaces: []specs.LinuxNamespace{
+ {
+ Type: "pid",
+ },
+ {
+ Type: "network",
+ Path: netns,
+ },
+ {
+ Type: "ipc",
+ },
+ {
+ Type: "uts",
+ },
+ {
+ Type: "mount",
+ },
},
- {
- "type": "mount"
- }
- ]
+ },
}
-}`, cwd)
- return []byte(template)
+ e := json.NewEncoder(w)
+ e.SetIndent("", " ")
+ return e.Encode(spec)
}
// Spec implements subcommands.Command for the "spec" command.
type Spec struct {
bundle string
cwd string
+ netns string
}
// Name implements subcommands.Command.Name.
@@ -148,21 +146,26 @@ func (*Spec) Synopsis() string {
// Usage implements subcommands.Command.Usage.
func (*Spec) Usage() string {
- return `spec [options] - create a new OCI bundle specification file.
+ return `spec [options] [-- args...] - create a new OCI bundle specification file.
+
+The spec command creates a new specification file (config.json) for a new OCI
+bundle.
-The spec command creates a new specification file (config.json) for a new OCI bundle.
+The specification file is a starter file that runs the command specified by
+'args' in the container. If 'args' is not specified the default is to run the
+'sh' program.
-The specification file is a starter file that runs the "sh" command in the container. You
-should edit the file to suit your needs. You can find out more about the format of the
-specification file by visiting the OCI runtime spec repository:
+While a number of flags are provided to change values in the specification, you
+can examine the file and edit it to suit your needs after this command runs.
+You can find out more about the format of the specification file by visiting
+the OCI runtime spec repository:
https://github.com/opencontainers/runtime-spec/
EXAMPLE:
$ mkdir -p bundle/rootfs
$ cd bundle
- $ runsc spec
+ $ runsc spec -- /hello
$ docker export $(docker create hello-world) | tar -xf - -C rootfs
- $ sed -i 's;"sh";"/hello";' config.json
$ sudo runsc run hello
`
@@ -173,18 +176,29 @@ func (s *Spec) SetFlags(f *flag.FlagSet) {
f.StringVar(&s.bundle, "bundle", ".", "path to the root of the OCI bundle")
f.StringVar(&s.cwd, "cwd", "/", "working directory that will be set for the executable, "+
"this value MUST be an absolute path")
+ f.StringVar(&s.netns, "netns", "", "network namespace path")
}
// Execute implements subcommands.Command.Execute.
func (s *Spec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus {
+ // Grab the arguments.
+ containerArgs := f.Args()
+ if len(containerArgs) == 0 {
+ containerArgs = []string{"sh"}
+ }
+
confPath := filepath.Join(s.bundle, "config.json")
if _, err := os.Stat(confPath); !os.IsNotExist(err) {
Fatalf("file %q already exists", confPath)
}
- var spec = genSpec(s.cwd)
+ configFile, err := os.OpenFile(confPath, os.O_WRONLY|os.O_CREATE, 0664)
+ if err != nil {
+ Fatalf("opening file %q: %v", confPath, err)
+ }
- if err := ioutil.WriteFile(confPath, spec, 0664); err != nil {
+ err = writeSpec(configFile, s.cwd, s.netns, containerArgs)
+ if err != nil {
Fatalf("writing to %q: %v", confPath, err)
}
diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go
index a27a01942..e189648f4 100644
--- a/runsc/container/multi_container_test.go
+++ b/runsc/container/multi_container_test.go
@@ -100,19 +100,20 @@ type execDesc struct {
c *Container
cmd []string
want int
- desc string
+ name string
}
-func execMany(execs []execDesc) error {
+func execMany(t *testing.T, execs []execDesc) {
for _, exec := range execs {
- args := &control.ExecArgs{Argv: exec.cmd}
- if ws, err := exec.c.executeSync(args); err != nil {
- return fmt.Errorf("error executing %+v: %v", args, err)
- } else if ws.ExitStatus() != exec.want {
- return fmt.Errorf("%q: exec %q got exit status: %d, want: %d", exec.desc, exec.cmd, ws.ExitStatus(), exec.want)
- }
+ t.Run(exec.name, func(t *testing.T) {
+ args := &control.ExecArgs{Argv: exec.cmd}
+ if ws, err := exec.c.executeSync(args); err != nil {
+ t.Errorf("error executing %+v: %v", args, err)
+ } else if ws.ExitStatus() != exec.want {
+ t.Errorf("%q: exec %q got exit status: %d, want: %d", exec.name, exec.cmd, ws.ExitStatus(), exec.want)
+ }
+ })
}
- return nil
}
func createSharedMount(mount specs.Mount, name string, pod ...*specs.Spec) {
@@ -1072,7 +1073,7 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) {
// Test that pod shared mounts are properly mounted in 2 containers and that
// changes from one container is reflected in the other.
func TestMultiContainerSharedMount(t *testing.T) {
- for name, conf := range configs(t, all...) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
rootDir, cleanup, err := testutil.SetupRootDir()
if err != nil {
@@ -1110,84 +1111,82 @@ func TestMultiContainerSharedMount(t *testing.T) {
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-d", mnt0.Destination},
- desc: "directory is mounted in container0",
+ name: "directory is mounted in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-d", mnt1.Destination},
- desc: "directory is mounted in container1",
+ name: "directory is mounted in container1",
},
{
c: containers[0],
- cmd: []string{"/usr/bin/touch", file0},
- desc: "create file in container0",
+ cmd: []string{"/bin/touch", file0},
+ name: "create file in container0",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-f", file0},
- desc: "file appears in container0",
+ name: "file appears in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-f", file1},
- desc: "file appears in container1",
+ name: "file appears in container1",
},
{
c: containers[1],
cmd: []string{"/bin/rm", file1},
- desc: "file removed from container1",
+ name: "remove file from container1",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "!", "-f", file0},
- desc: "file removed from container0",
+ name: "file removed from container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "!", "-f", file1},
- desc: "file removed from container1",
+ name: "file removed from container1",
},
{
c: containers[1],
cmd: []string{"/bin/mkdir", file1},
- desc: "create directory in container1",
+ name: "create directory in container1",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-d", file0},
- desc: "dir appears in container0",
+ name: "dir appears in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-d", file1},
- desc: "dir appears in container1",
+ name: "dir appears in container1",
},
{
c: containers[0],
cmd: []string{"/bin/rmdir", file0},
- desc: "create directory in container0",
+ name: "remove directory from container0",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "!", "-d", file0},
- desc: "dir removed from container0",
+ name: "dir removed from container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "!", "-d", file1},
- desc: "dir removed from container1",
+ name: "dir removed from container1",
},
}
- if err := execMany(execs); err != nil {
- t.Fatal(err.Error())
- }
+ execMany(t, execs)
})
}
}
// Test that pod mounts are mounted as readonly when requested.
func TestMultiContainerSharedMountReadonly(t *testing.T) {
- for name, conf := range configs(t, all...) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
rootDir, cleanup, err := testutil.SetupRootDir()
if err != nil {
@@ -1225,35 +1224,34 @@ func TestMultiContainerSharedMountReadonly(t *testing.T) {
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-d", mnt0.Destination},
- desc: "directory is mounted in container0",
+ name: "directory is mounted in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-d", mnt1.Destination},
- desc: "directory is mounted in container1",
+ name: "directory is mounted in container1",
},
{
c: containers[0],
- cmd: []string{"/usr/bin/touch", file0},
+ cmd: []string{"/bin/touch", file0},
want: 1,
- desc: "fails to write to container0",
+ name: "fails to write to container0",
},
{
c: containers[1],
- cmd: []string{"/usr/bin/touch", file1},
+ cmd: []string{"/bin/touch", file1},
want: 1,
- desc: "fails to write to container1",
+ name: "fails to write to container1",
},
}
- if err := execMany(execs); err != nil {
- t.Fatal(err.Error())
- }
+ execMany(t, execs)
})
}
}
// Test that shared pod mounts continue to work after container is restarted.
func TestMultiContainerSharedMountRestart(t *testing.T) {
+ //TODO(gvisor.dev/issue/1487): This is failing with VFS2.
for name, conf := range configs(t, all...) {
t.Run(name, func(t *testing.T) {
rootDir, cleanup, err := testutil.SetupRootDir()
@@ -1291,23 +1289,21 @@ func TestMultiContainerSharedMountRestart(t *testing.T) {
execs := []execDesc{
{
c: containers[0],
- cmd: []string{"/usr/bin/touch", file0},
- desc: "create file in container0",
+ cmd: []string{"/bin/touch", file0},
+ name: "create file in container0",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-f", file0},
- desc: "file appears in container0",
+ name: "file appears in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-f", file1},
- desc: "file appears in container1",
+ name: "file appears in container1",
},
}
- if err := execMany(execs); err != nil {
- t.Fatal(err.Error())
- }
+ execMany(t, execs)
containers[1].Destroy()
@@ -1334,32 +1330,30 @@ func TestMultiContainerSharedMountRestart(t *testing.T) {
{
c: containers[0],
cmd: []string{"/usr/bin/test", "-f", file0},
- desc: "file is still in container0",
+ name: "file is still in container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "-f", file1},
- desc: "file is still in container1",
+ name: "file is still in container1",
},
{
c: containers[1],
cmd: []string{"/bin/rm", file1},
- desc: "file removed from container1",
+ name: "file removed from container1",
},
{
c: containers[0],
cmd: []string{"/usr/bin/test", "!", "-f", file0},
- desc: "file removed from container0",
+ name: "file removed from container0",
},
{
c: containers[1],
cmd: []string{"/usr/bin/test", "!", "-f", file1},
- desc: "file removed from container1",
+ name: "file removed from container1",
},
}
- if err := execMany(execs); err != nil {
- t.Fatal(err.Error())
- }
+ execMany(t, execs)
})
}
}
@@ -1367,53 +1361,53 @@ func TestMultiContainerSharedMountRestart(t *testing.T) {
// Test that unsupported pod mounts options are ignored when matching master and
// slave mounts.
func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) {
- rootDir, cleanup, err := testutil.SetupRootDir()
- if err != nil {
- t.Fatalf("error creating root dir: %v", err)
- }
- defer cleanup()
-
- conf := testutil.TestConfig(t)
- conf.RootDir = rootDir
+ for name, conf := range configsWithVFS2(t, all...) {
+ t.Run(name, func(t *testing.T) {
+ rootDir, cleanup, err := testutil.SetupRootDir()
+ if err != nil {
+ t.Fatalf("error creating root dir: %v", err)
+ }
+ defer cleanup()
+ conf.RootDir = rootDir
- // Setup the containers.
- sleep := []string{"/bin/sleep", "100"}
- podSpec, ids := createSpecs(sleep, sleep)
- mnt0 := specs.Mount{
- Destination: "/mydir/test",
- Source: "/some/dir",
- Type: "tmpfs",
- Options: []string{"rw", "rbind", "relatime"},
- }
- podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0)
+ // Setup the containers.
+ sleep := []string{"/bin/sleep", "100"}
+ podSpec, ids := createSpecs(sleep, sleep)
+ mnt0 := specs.Mount{
+ Destination: "/mydir/test",
+ Source: "/some/dir",
+ Type: "tmpfs",
+ Options: []string{"rw", "rbind", "relatime"},
+ }
+ podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0)
- mnt1 := mnt0
- mnt1.Destination = "/mydir2/test2"
- mnt1.Options = []string{"rw", "nosuid"}
- podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1)
+ mnt1 := mnt0
+ mnt1.Destination = "/mydir2/test2"
+ mnt1.Options = []string{"rw", "nosuid"}
+ podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1)
- createSharedMount(mnt0, "test-mount", podSpec...)
+ createSharedMount(mnt0, "test-mount", podSpec...)
- containers, cleanup, err := startContainers(conf, podSpec, ids)
- if err != nil {
- t.Fatalf("error starting containers: %v", err)
- }
- defer cleanup()
+ containers, cleanup, err := startContainers(conf, podSpec, ids)
+ if err != nil {
+ t.Fatalf("error starting containers: %v", err)
+ }
+ defer cleanup()
- execs := []execDesc{
- {
- c: containers[0],
- cmd: []string{"/usr/bin/test", "-d", mnt0.Destination},
- desc: "directory is mounted in container0",
- },
- {
- c: containers[1],
- cmd: []string{"/usr/bin/test", "-d", mnt1.Destination},
- desc: "directory is mounted in container1",
- },
- }
- if err := execMany(execs); err != nil {
- t.Fatal(err.Error())
+ execs := []execDesc{
+ {
+ c: containers[0],
+ cmd: []string{"/usr/bin/test", "-d", mnt0.Destination},
+ name: "directory is mounted in container0",
+ },
+ {
+ c: containers[1],
+ cmd: []string{"/usr/bin/test", "-d", mnt1.Destination},
+ name: "directory is mounted in container1",
+ },
+ }
+ execMany(t, execs)
+ })
}
}
diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go
index 1dce36965..88814b83c 100644
--- a/runsc/fsgofer/filter/config.go
+++ b/runsc/fsgofer/filter/config.go
@@ -128,6 +128,7 @@ var allowedSyscalls = seccomp.SyscallRules{
syscall.SYS_MADVISE: {},
unix.SYS_MEMFD_CREATE: {}, /// Used by flipcall.PacketWindowAllocator.Init().
syscall.SYS_MKDIRAT: {},
+ syscall.SYS_MKNODAT: {},
// Used by the Go runtime as a temporarily workaround for a Linux
// 5.2-5.4 bug.
//
diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go
index 74977c313..b7521bda7 100644
--- a/runsc/fsgofer/fsgofer.go
+++ b/runsc/fsgofer/fsgofer.go
@@ -139,7 +139,7 @@ func (a *attachPoint) Attach() (p9.File, error) {
return nil, fmt.Errorf("unable to open %q: %v", a.prefix, err)
}
- stat, err := stat(f.FD())
+ stat, err := fstat(f.FD())
if err != nil {
return nil, fmt.Errorf("unable to stat %q: %v", a.prefix, err)
}
@@ -352,7 +352,7 @@ func newFDMaybe(file *fd.FD) *fd.FD {
return dup
}
-func stat(fd int) (syscall.Stat_t, error) {
+func fstat(fd int) (syscall.Stat_t, error) {
var stat syscall.Stat_t
if err := syscall.Fstat(fd, &stat); err != nil {
return syscall.Stat_t{}, err
@@ -360,6 +360,14 @@ func stat(fd int) (syscall.Stat_t, error) {
return stat, nil
}
+func stat(path string) (syscall.Stat_t, error) {
+ var stat syscall.Stat_t
+ if err := syscall.Stat(path, &stat); err != nil {
+ return syscall.Stat_t{}, err
+ }
+ return stat, nil
+}
+
func fchown(fd int, uid p9.UID, gid p9.GID) error {
return syscall.Fchownat(fd, "", int(uid), int(gid), linux.AT_EMPTY_PATH|unix.AT_SYMLINK_NOFOLLOW)
}
@@ -388,7 +396,7 @@ func (l *localFile) Open(flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) {
}
}
- stat, err := stat(newFile.FD())
+ stat, err := fstat(newFile.FD())
if err != nil {
if newFile != l.file {
newFile.Close()
@@ -449,7 +457,7 @@ func (l *localFile) Create(name string, mode p9.OpenFlags, perm p9.FileMode, uid
if err := fchown(child.FD(), uid, gid); err != nil {
return nil, nil, p9.QID{}, 0, extractErrno(err)
}
- stat, err := stat(child.FD())
+ stat, err := fstat(child.FD())
if err != nil {
return nil, nil, p9.QID{}, 0, extractErrno(err)
}
@@ -497,7 +505,7 @@ func (l *localFile) Mkdir(name string, perm p9.FileMode, uid p9.UID, gid p9.GID)
if err := fchown(f.FD(), uid, gid); err != nil {
return p9.QID{}, extractErrno(err)
}
- stat, err := stat(f.FD())
+ stat, err := fstat(f.FD())
if err != nil {
return p9.QID{}, extractErrno(err)
}
@@ -517,7 +525,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) {
return nil, nil, extractErrno(err)
}
- stat, err := stat(newFile.FD())
+ stat, err := fstat(newFile.FD())
if err != nil {
newFile.Close()
return nil, nil, extractErrno(err)
@@ -542,7 +550,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) {
if err != nil {
return nil, nil, extractErrno(err)
}
- stat, err := stat(f.FD())
+ stat, err := fstat(f.FD())
if err != nil {
f.Close()
return nil, nil, extractErrno(err)
@@ -592,7 +600,7 @@ func (l *localFile) FSync() error {
// GetAttr implements p9.File.
func (l *localFile) GetAttr(_ p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) {
- stat, err := stat(l.file.FD())
+ stat, err := fstat(l.file.FD())
if err != nil {
return p9.QID{}, p9.AttrMask{}, p9.Attr{}, extractErrno(err)
}
@@ -880,7 +888,7 @@ func (l *localFile) Symlink(target, newName string, uid p9.UID, gid p9.GID) (p9.
if err := fchown(f.FD(), uid, gid); err != nil {
return p9.QID{}, extractErrno(err)
}
- stat, err := stat(f.FD())
+ stat, err := fstat(f.FD())
if err != nil {
return p9.QID{}, extractErrno(err)
}
@@ -907,13 +915,39 @@ func (l *localFile) Link(target p9.File, newName string) error {
}
// Mknod implements p9.File.
-//
-// Not implemented.
-func (*localFile) Mknod(_ string, _ p9.FileMode, _ uint32, _ uint32, _ p9.UID, _ p9.GID) (p9.QID, error) {
+func (l *localFile) Mknod(name string, mode p9.FileMode, _ uint32, _ uint32, uid p9.UID, gid p9.GID) (p9.QID, error) {
+ conf := l.attachPoint.conf
+ if conf.ROMount {
+ if conf.PanicOnWrite {
+ panic("attempt to write to RO mount")
+ }
+ return p9.QID{}, syscall.EROFS
+ }
+
+ hostPath := path.Join(l.hostPath, name)
+
+ // Return EEXIST if the file already exists.
+ if _, err := stat(hostPath); err == nil {
+ return p9.QID{}, syscall.EEXIST
+ }
+
// From mknod(2) man page:
// "EPERM: [...] if the filesystem containing pathname does not support
// the type of node requested."
- return p9.QID{}, syscall.EPERM
+ if mode.FileType() != p9.ModeRegular {
+ return p9.QID{}, syscall.EPERM
+ }
+
+ // Allow Mknod to create regular files.
+ if err := syscall.Mknod(hostPath, uint32(mode), 0); err != nil {
+ return p9.QID{}, err
+ }
+
+ stat, err := stat(hostPath)
+ if err != nil {
+ return p9.QID{}, extractErrno(err)
+ }
+ return l.attachPoint.makeQID(stat), nil
}
// UnlinkAt implements p9.File.
diff --git a/runsc/sandbox/network.go b/runsc/sandbox/network.go
index deee619f3..817a923ad 100644
--- a/runsc/sandbox/network.go
+++ b/runsc/sandbox/network.go
@@ -134,7 +134,6 @@ func createInterfacesAndRoutesFromNS(conn *urpc.Client, nsPath string, hardwareG
return err
}
if isRoot {
-
return fmt.Errorf("cannot run with network enabled in root network namespace")
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-23 19:13:28 +0000