summaryrefslogtreecommitdiffhomepage
path: root/runsc
diff options
context:
space:
mode:
Diffstat (limited to 'runsc')
-rw-r--r--runsc/boot/BUILD1
-rw-r--r--runsc/boot/filter/config.go21
-rw-r--r--runsc/boot/filter/config_amd64.go41
-rw-r--r--runsc/boot/filter/config_arm64.go25
-rw-r--r--runsc/boot/loader.go39
-rw-r--r--runsc/boot/loader_test.go24
-rw-r--r--runsc/boot/vfs.go184
-rw-r--r--runsc/cmd/gofer.go15
-rw-r--r--runsc/container/container.go3
-rw-r--r--runsc/container/container_test.go55
-rw-r--r--runsc/container/multi_container_test.go13
-rw-r--r--runsc/fsgofer/filter/config.go17
-rw-r--r--runsc/fsgofer/filter/config_amd64.go35
-rw-r--r--runsc/fsgofer/filter/config_arm64.go21
-rw-r--r--runsc/fsgofer/filter/extra_filters_race.go1
-rw-r--r--runsc/fsgofer/fsgofer.go3
-rw-r--r--runsc/fsgofer/fsgofer_test.go23
-rw-r--r--runsc/sandbox/sandbox.go39
-rw-r--r--runsc/specutils/seccomp/seccomp_test.go2
19 files changed, 367 insertions, 195 deletions
diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD
index 01f62d50a..2d9517f4a 100644
--- a/runsc/boot/BUILD
+++ b/runsc/boot/BUILD
@@ -27,6 +27,7 @@ go_library(
"//pkg/abi",
"//pkg/abi/linux",
"//pkg/bpf",
+ "//pkg/cleanup",
"//pkg/context",
"//pkg/control/server",
"//pkg/cpuid",
diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go
index 4ed28b5cd..a7c4ebb0c 100644
--- a/runsc/boot/filter/config.go
+++ b/runsc/boot/filter/config.go
@@ -27,19 +27,8 @@ import (
// allowedSyscalls is the set of syscalls executed by the Sentry to the host OS.
var allowedSyscalls = seccomp.SyscallRules{
syscall.SYS_CLOCK_GETTIME: {},
- syscall.SYS_CLONE: []seccomp.Rule{
- {
- seccomp.EqualTo(
- syscall.CLONE_VM |
- syscall.CLONE_FS |
- syscall.CLONE_FILES |
- syscall.CLONE_SIGHAND |
- syscall.CLONE_SYSVSEM |
- syscall.CLONE_THREAD),
- },
- },
- syscall.SYS_CLOSE: {},
- syscall.SYS_DUP: {},
+ syscall.SYS_CLOSE: {},
+ syscall.SYS_DUP: {},
syscall.SYS_DUP3: []seccomp.Rule{
{
seccomp.MatchAny{},
@@ -173,6 +162,12 @@ var allowedSyscalls = seccomp.SyscallRules{
},
syscall.SYS_LSEEK: {},
syscall.SYS_MADVISE: {},
+ unix.SYS_MEMBARRIER: []seccomp.Rule{
+ {
+ seccomp.EqualTo(linux.MEMBARRIER_CMD_GLOBAL),
+ seccomp.EqualTo(0),
+ },
+ },
syscall.SYS_MINCORE: {},
// Used by the Go runtime as a temporarily workaround for a Linux
// 5.2-5.4 bug.
diff --git a/runsc/boot/filter/config_amd64.go b/runsc/boot/filter/config_amd64.go
index 9b1799416..cea5613b8 100644
--- a/runsc/boot/filter/config_amd64.go
+++ b/runsc/boot/filter/config_amd64.go
@@ -24,8 +24,41 @@ import (
)
func init() {
- allowedSyscalls[syscall.SYS_ARCH_PRCTL] = append(allowedSyscalls[syscall.SYS_ARCH_PRCTL],
- seccomp.Rule{seccomp.EqualTo(linux.ARCH_GET_FS)},
- seccomp.Rule{seccomp.EqualTo(linux.ARCH_SET_FS)},
- )
+ allowedSyscalls[syscall.SYS_ARCH_PRCTL] = []seccomp.Rule{
+ // TODO(b/168828518): No longer used in Go 1.16+.
+ {seccomp.EqualTo(linux.ARCH_SET_FS)},
+ }
+
+ allowedSyscalls[syscall.SYS_CLONE] = []seccomp.Rule{
+ // parent_tidptr and child_tidptr are always 0 because neither
+ // CLONE_PARENT_SETTID nor CLONE_CHILD_SETTID are used.
+ {
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SETTLS |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ seccomp.EqualTo(0), // parent_tidptr
+ seccomp.EqualTo(0), // child_tidptr
+ seccomp.MatchAny{}, // tls
+ },
+ {
+ // TODO(b/168828518): No longer used in Go 1.16+ (on amd64).
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ seccomp.EqualTo(0), // parent_tidptr
+ seccomp.EqualTo(0), // child_tidptr
+ seccomp.MatchAny{}, // tls
+ },
+ }
}
diff --git a/runsc/boot/filter/config_arm64.go b/runsc/boot/filter/config_arm64.go
index 7fa9bbda3..37313f97f 100644
--- a/runsc/boot/filter/config_arm64.go
+++ b/runsc/boot/filter/config_arm64.go
@@ -16,6 +16,29 @@
package filter
-// Reserve for future customization.
+import (
+ "syscall"
+
+ "gvisor.dev/gvisor/pkg/seccomp"
+)
+
func init() {
+ allowedSyscalls[syscall.SYS_CLONE] = []seccomp.Rule{
+ {
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ // These arguments are left uninitialized by the Go
+ // runtime, so they may be anything (and are unused by
+ // the host).
+ seccomp.MatchAny{}, // parent_tidptr
+ seccomp.MatchAny{}, // tls
+ seccomp.MatchAny{}, // child_tidptr
+ },
+ }
}
diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go
index 4940ea96a..dee2c4fbb 100644
--- a/runsc/boot/loader.go
+++ b/runsc/boot/loader.go
@@ -282,6 +282,7 @@ func New(args Args) (*Loader, error) {
args.NumCPU = runtime.NumCPU()
}
log.Infof("CPUs: %d", args.NumCPU)
+ runtime.GOMAXPROCS(args.NumCPU)
if args.TotalMem > 0 {
// Adjust the total memory returned by the Sentry so that applications that
@@ -797,7 +798,7 @@ func (l *Loader) createContainerProcess(root bool, cid string, info *containerIn
}
// startGoferMonitor runs a goroutine to monitor gofer's health. It polls on
-// the gofer FDs looking for disconnects, and destroys the container if a
+// the gofer FDs looking for disconnects, and kills the container processes if a
// disconnect occurs in any of the gofer FDs.
func (l *Loader) startGoferMonitor(cid string, goferFDs []*fd.FD) {
go func() {
@@ -818,18 +819,15 @@ func (l *Loader) startGoferMonitor(cid string, goferFDs []*fd.FD) {
panic(fmt.Sprintf("Error monitoring gofer FDs: %v", err))
}
- // Check if the gofer has stopped as part of normal container destruction.
- // This is done just to avoid sending an annoying error message to the log.
- // Note that there is a small race window in between mu.Unlock() and the
- // lock being reacquired in destroyContainer(), but it's harmless to call
- // destroyContainer() multiple times.
l.mu.Lock()
- _, ok := l.processes[execID{cid: cid}]
- l.mu.Unlock()
- if ok {
- log.Infof("Gofer socket disconnected, destroying container %q", cid)
- if err := l.destroyContainer(cid); err != nil {
- log.Warningf("Error destroying container %q after gofer stopped: %v", cid, err)
+ defer l.mu.Unlock()
+
+ // The gofer could have been stopped due to a normal container shutdown.
+ // Check if the container has not stopped yet.
+ if tg, _ := l.tryThreadGroupFromIDLocked(execID{cid: cid}); tg != nil {
+ log.Infof("Gofer socket disconnected, killing container %q", cid)
+ if err := l.signalAllProcesses(cid, int32(linux.SIGKILL)); err != nil {
+ log.Warningf("Error killing container %q after gofer stopped: %v", cid, err)
}
}
}()
@@ -898,17 +896,24 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
return 0, fmt.Errorf("container %q not started", args.ContainerID)
}
- // Get the container MountNamespace from the Task.
+ // Get the container MountNamespace from the Task. Try to acquire ref may fail
+ // in case it raced with task exit.
if kernel.VFS2Enabled {
// task.MountNamespace() does not take a ref, so we must do so ourselves.
args.MountNamespaceVFS2 = tg.Leader().MountNamespaceVFS2()
- args.MountNamespaceVFS2.IncRef()
+ if !args.MountNamespaceVFS2.TryIncRef() {
+ return 0, fmt.Errorf("container %q has stopped", args.ContainerID)
+ }
} else {
+ var reffed bool
tg.Leader().WithMuLocked(func(t *kernel.Task) {
// task.MountNamespace() does not take a ref, so we must do so ourselves.
args.MountNamespace = t.MountNamespace()
- args.MountNamespace.IncRef()
+ reffed = args.MountNamespace.TryIncRef()
})
+ if !reffed {
+ return 0, fmt.Errorf("container %q has stopped", args.ContainerID)
+ }
}
// Add the HOME environment variable if it is not already set.
@@ -1059,8 +1064,8 @@ func newRootNetworkNamespace(conf *config.Config, clock tcpip.Clock, uniqueID st
}
func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) {
- netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}
- transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()}
+ netProtos := []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol, arp.NewProtocol}
+ transProtos := []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol, icmp.NewProtocol4}
s := netstack.Stack{stack.New(stack.Options{
NetworkProtocols: netProtos,
TransportProtocols: transProtos,
diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go
index bf9ec5d38..1f49431a2 100644
--- a/runsc/boot/loader_test.go
+++ b/runsc/boot/loader_test.go
@@ -264,7 +264,7 @@ type CreateMountTestcase struct {
expectedPaths []string
}
-func createMountTestcases(vfs2 bool) []*CreateMountTestcase {
+func createMountTestcases() []*CreateMountTestcase {
testCases := []*CreateMountTestcase{
&CreateMountTestcase{
// Only proc.
@@ -409,32 +409,26 @@ func createMountTestcases(vfs2 bool) []*CreateMountTestcase {
Destination: "/proc",
Type: "tmpfs",
},
- // TODO (gvisor.dev/issue/1487): Re-add this case when sysfs supports
- // MkDirAt in VFS2 (and remove the reduntant append).
- // {
- // Destination: "/sys/bar",
- // Type: "tmpfs",
- // },
- //
+ {
+ Destination: "/sys/bar",
+ Type: "tmpfs",
+ },
+
{
Destination: "/tmp/baz",
Type: "tmpfs",
},
},
},
- expectedPaths: []string{"/proc", "/sys" /* "/sys/bar" ,*/, "/tmp", "/tmp/baz"},
+ expectedPaths: []string{"/proc", "/sys", "/sys/bar", "/tmp", "/tmp/baz"},
}
- if !vfs2 {
- vfsCase.spec.Mounts = append(vfsCase.spec.Mounts, specs.Mount{Destination: "/sys/bar", Type: "tmpfs"})
- vfsCase.expectedPaths = append(vfsCase.expectedPaths, "/sys/bar")
- }
return append(testCases, vfsCase)
}
// Test that MountNamespace can be created with various specs.
func TestCreateMountNamespace(t *testing.T) {
- for _, tc := range createMountTestcases(false /* vfs2 */) {
+ for _, tc := range createMountTestcases() {
t.Run(tc.name, func(t *testing.T) {
conf := testConfig()
ctx := contexttest.Context(t)
@@ -471,7 +465,7 @@ func TestCreateMountNamespace(t *testing.T) {
// Test that MountNamespace can be created with various specs.
func TestCreateMountNamespaceVFS2(t *testing.T) {
- for _, tc := range createMountTestcases(true /* vfs2 */) {
+ for _, tc := range createMountTestcases() {
t.Run(tc.name, func(t *testing.T) {
spec := testSpec()
spec.Mounts = tc.spec.Mounts
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index 7844ea28c..e36664938 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -21,6 +21,7 @@ import (
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/log"
@@ -168,26 +169,30 @@ func (c *containerMounter) mountAll(conf *config.Config, procArgs *kernel.Create
}
rootProcArgs.MountNamespaceVFS2 = mns
+ root := mns.Root()
+ defer root.DecRef(rootCtx)
+ if root.Mount().ReadOnly() {
+ // Switch to ReadWrite while we setup submounts.
+ if err := c.k.VFS().SetMountReadOnly(root.Mount(), false); err != nil {
+ return nil, fmt.Errorf(`failed to set mount at "/" readwrite: %w`, err)
+ }
+ // Restore back to ReadOnly at the end.
+ defer func() {
+ if err := c.k.VFS().SetMountReadOnly(root.Mount(), true); err != nil {
+ panic(fmt.Sprintf(`failed to restore mount at "/" back to readonly: %v`, err))
+ }
+ }()
+ }
+
// Mount submounts.
if err := c.mountSubmountsVFS2(rootCtx, conf, mns, rootCreds); err != nil {
return nil, fmt.Errorf("mounting submounts vfs2: %w", err)
}
- if c.root.Readonly || conf.Overlay {
- // Switch to ReadOnly after all submounts were setup.
- root := mns.Root()
- defer root.DecRef(rootCtx)
- if err := c.k.VFS().SetMountReadOnly(root.Mount(), true); err != nil {
- return nil, fmt.Errorf(`failed to set mount at "/" readonly: %v`, err)
- }
- }
-
return mns, nil
}
// createMountNamespaceVFS2 creates the container's root mount and namespace.
-// The mount is created ReadWrite to allow mount point for submounts to be
-// created. ** The caller is responsible to switch to ReadOnly if needed **
func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *config.Config, creds *auth.Credentials) (*vfs.MountNamespace, error) {
fd := c.fds.remove()
data := p9MountData(fd, conf.FileAccess, true /* vfs2 */)
@@ -201,21 +206,71 @@ func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *c
log.Infof("Mounting root over 9P, ioFD: %d", fd)
opts := &vfs.MountOptions{
- // Always mount as ReadWrite to allow other mounts on top of it. It'll be
- // made ReadOnly in the caller (if needed).
- ReadOnly: false,
+ ReadOnly: c.root.Readonly,
GetFilesystemOptions: vfs.GetFilesystemOptions{
Data: strings.Join(data, ","),
},
InternalMount: true,
}
- mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", gofer.Name, opts)
+
+ fsName := gofer.Name
+ if conf.Overlay && !c.root.Readonly {
+ log.Infof("Adding overlay on top of root")
+ var err error
+ var cleanup func()
+ opts, cleanup, err = c.configureOverlay(ctx, creds, opts, fsName)
+ if err != nil {
+ return nil, fmt.Errorf("mounting root with overlay: %w", err)
+ }
+ defer cleanup()
+ fsName = overlay.Name
+ }
+
+ mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", fsName, opts)
if err != nil {
return nil, fmt.Errorf("setting up mount namespace: %w", err)
}
return mns, nil
}
+// configureOverlay mounts the lower layer using "lowerOpts", mounts the upper
+// layer using tmpfs, and return overlay mount options. "cleanup" must be called
+// after the options have been used to mount the overlay, to release refs on
+// lower and upper mounts.
+func (c *containerMounter) configureOverlay(ctx context.Context, creds *auth.Credentials, lowerOpts *vfs.MountOptions, lowerFSName string) (*vfs.MountOptions, func(), error) {
+ // First copy options from lower layer to upper layer and overlay. Clear
+ // filesystem specific options.
+ upperOpts := *lowerOpts
+ upperOpts.GetFilesystemOptions = vfs.GetFilesystemOptions{}
+
+ overlayOpts := *lowerOpts
+ overlayOpts.GetFilesystemOptions = vfs.GetFilesystemOptions{}
+
+ // Next mount upper and lower. Upper is a tmpfs mount to keep all
+ // modifications inside the sandbox.
+ upper, err := c.k.VFS().MountDisconnected(ctx, creds, "" /* source */, tmpfs.Name, &upperOpts)
+ if err != nil {
+ return nil, nil, fmt.Errorf("failed to create upper layer for overlay, opts: %+v: %v", upperOpts, err)
+ }
+ cu := cleanup.Make(func() { upper.DecRef(ctx) })
+ defer cu.Clean()
+
+ // All writes go to the upper layer, be paranoid and make lower readonly.
+ lowerOpts.ReadOnly = true
+ lower, err := c.k.VFS().MountDisconnected(ctx, creds, "" /* source */, lowerFSName, lowerOpts)
+ if err != nil {
+ return nil, nil, err
+ }
+ cu.Add(func() { lower.DecRef(ctx) })
+
+ // Configure overlay with both layers.
+ overlayOpts.GetFilesystemOptions.InternalData = overlay.FilesystemOptions{
+ UpperRoot: vfs.MakeVirtualDentry(upper, upper.Root()),
+ LowerRoots: []vfs.VirtualDentry{vfs.MakeVirtualDentry(lower, lower.Root())},
+ }
+ return &overlayOpts, cu.Release(), nil
+}
+
func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials) error {
mounts, err := c.prepareMountsVFS2()
if err != nil {
@@ -245,7 +300,7 @@ func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *config.
if mnt != nil && mnt.ReadOnly() {
// Switch to ReadWrite while we setup submounts.
if err := c.k.VFS().SetMountReadOnly(mnt, false); err != nil {
- return fmt.Errorf("failed to set mount at %q readwrite: %v", submount.Destination, err)
+ return fmt.Errorf("failed to set mount at %q readwrite: %w", submount.Destination, err)
}
// Restore back to ReadOnly at the end.
defer func() {
@@ -297,14 +352,7 @@ func (c *containerMounter) prepareMountsVFS2() ([]mountAndFD, error) {
}
func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *mountAndFD) (*vfs.Mount, error) {
- root := mns.Root()
- defer root.DecRef(ctx)
- target := &vfs.PathOperation{
- Root: root,
- Start: root,
- Path: fspath.Parse(submount.Destination),
- }
- fsName, opts, err := c.getMountNameAndOptionsVFS2(conf, submount)
+ fsName, opts, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, submount)
if err != nil {
return nil, fmt.Errorf("mountOptions failed: %w", err)
}
@@ -313,8 +361,27 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.C
return nil, nil
}
- if err := c.k.VFS().MakeSyntheticMountpoint(ctx, submount.Destination, root, creds); err != nil {
- return nil, err
+ if err := c.makeMountPoint(ctx, creds, mns, submount.Destination); err != nil {
+ return nil, fmt.Errorf("creating mount point %q: %w", submount.Destination, err)
+ }
+
+ if useOverlay {
+ log.Infof("Adding overlay on top of mount %q", submount.Destination)
+ var cleanup func()
+ opts, cleanup, err = c.configureOverlay(ctx, creds, opts, fsName)
+ if err != nil {
+ return nil, fmt.Errorf("mounting volume with overlay at %q: %w", submount.Destination, err)
+ }
+ defer cleanup()
+ fsName = overlay.Name
+ }
+
+ root := mns.Root()
+ defer root.DecRef(ctx)
+ target := &vfs.PathOperation{
+ Root: root,
+ Start: root,
+ Path: fspath.Parse(submount.Destination),
}
mnt, err := c.k.VFS().MountAt(ctx, creds, "", target, fsName, opts)
if err != nil {
@@ -326,8 +393,9 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.C
// getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values
// used for mounts.
-func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mountAndFD) (string, *vfs.MountOptions, error) {
+func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mountAndFD) (string, *vfs.MountOptions, bool, error) {
fsName := m.Type
+ useOverlay := false
var data []string
// Find filesystem name and FS specific data field.
@@ -342,7 +410,7 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
var err error
data, err = parseAndFilterOptions(m.Options, tmpfsAllowedData...)
if err != nil {
- return "", nil, err
+ return "", nil, false, err
}
case bind:
@@ -350,13 +418,16 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
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")
+ return "", nil, false, fmt.Errorf("9P mount requires a connection FD")
}
data = p9MountData(m.fd, c.getMountAccessType(m.Mount), true /* vfs2 */)
+ // If configured, add overlay to all writable mounts.
+ useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly
+
default:
log.Warningf("ignoring unknown filesystem type %q", m.Type)
- return "", nil, nil
+ return "", nil, false, nil
}
opts := &vfs.MountOptions{
@@ -381,11 +452,7 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
}
}
- if conf.Overlay {
- // All writes go to upper, be paranoid and make lower readonly.
- opts.ReadOnly = true
- }
- return fsName, opts, nil
+ return fsName, opts, useOverlay, nil
}
// mountTmpVFS2 mounts an internal tmpfs at '/tmp' if it's safe to do so.
@@ -488,13 +555,25 @@ func (c *containerMounter) mountSharedMasterVFS2(ctx context.Context, conf *conf
// 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)
+ fsName, opts, useOverlay, 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)
}
+
+ if useOverlay {
+ log.Infof("Adding overlay on top of shared mount %q", mntFD.Destination)
+ var cleanup func()
+ opts, cleanup, err = c.configureOverlay(ctx, creds, opts, fsName)
+ if err != nil {
+ return nil, fmt.Errorf("mounting shared volume with overlay at %q: %w", mntFD.Destination, err)
+ }
+ defer cleanup()
+ fsName = overlay.Name
+ }
+
return c.k.VFS().MountDisconnected(ctx, creds, "", fsName, opts)
}
@@ -505,7 +584,9 @@ func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *co
return nil, err
}
- _, opts, err := c.getMountNameAndOptionsVFS2(conf, &mountAndFD{Mount: mount})
+ // Ignore data and useOverlay because these were already applied to
+ // the master mount.
+ _, opts, _, err := c.getMountNameAndOptionsVFS2(conf, &mountAndFD{Mount: mount})
if err != nil {
return nil, err
}
@@ -517,18 +598,39 @@ func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *co
root := mns.Root()
defer root.DecRef(ctx)
- if err := c.k.VFS().MakeSyntheticMountpoint(ctx, mount.Destination, root, creds); err != nil {
- return nil, err
- }
-
target := &vfs.PathOperation{
Root: root,
Start: root,
Path: fspath.Parse(mount.Destination),
}
+
+ if err := c.makeMountPoint(ctx, creds, mns, mount.Destination); err != nil {
+ return nil, fmt.Errorf("creating mount point %q: %w", mount.Destination, err)
+ }
+
if err := c.k.VFS().ConnectMountAt(ctx, creds, newMnt, target); err != nil {
return nil, err
}
log.Infof("Mounted %q type shared bind to %q", mount.Destination, source.name)
return newMnt, nil
}
+
+func (c *containerMounter) makeMountPoint(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, dest string) error {
+ root := mns.Root()
+ defer root.DecRef(ctx)
+ target := &vfs.PathOperation{
+ Root: root,
+ Start: root,
+ Path: fspath.Parse(dest),
+ }
+ // First check if mount point exists. When overlay is enabled, gofer doesn't
+ // allow changes to the FS, making MakeSytheticMountpoint() ineffective
+ // because MkdirAt fails with EROFS even if file exists.
+ vd, err := c.k.VFS().GetDentryAt(ctx, creds, target, &vfs.GetDentryOptions{})
+ if err == nil {
+ // File exists, we're done.
+ vd.DecRef(ctx)
+ return nil
+ }
+ return c.k.VFS().MakeSyntheticMountpoint(ctx, dest, root, creds)
+}
diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go
index bba00d551..371fcc0ae 100644
--- a/runsc/cmd/gofer.go
+++ b/runsc/cmd/gofer.go
@@ -62,9 +62,8 @@ type Gofer struct {
applyCaps bool
setUpRoot bool
- panicOnWrite bool
- specFD int
- mountsFD int
+ specFD int
+ mountsFD int
}
// Name implements subcommands.Command.
@@ -87,7 +86,6 @@ func (g *Gofer) SetFlags(f *flag.FlagSet) {
f.StringVar(&g.bundleDir, "bundle", "", "path to the root of the bundle directory, defaults to the current directory")
f.Var(&g.ioFDs, "io-fds", "list of FDs to connect 9P servers. They must follow this order: root first, then mounts as defined in the spec")
f.BoolVar(&g.applyCaps, "apply-caps", true, "if true, apply capabilities to restrict what the Gofer process can do")
- f.BoolVar(&g.panicOnWrite, "panic-on-write", false, "if true, panics on attempts to write to RO mounts. RW mounts are unnaffected")
f.BoolVar(&g.setUpRoot, "setup-root", true, "if true, set up an empty root for the process")
f.IntVar(&g.specFD, "spec-fd", -1, "required fd with the container spec")
f.IntVar(&g.mountsFD, "mounts-fd", -1, "mountsFD is the file descriptor to write list of mounts after they have been resolved (direct paths, no symlinks).")
@@ -168,8 +166,7 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
// Start with root mount, then add any other additional mount as needed.
ats := make([]p9.Attacher, 0, len(spec.Mounts)+1)
ap, err := fsgofer.NewAttachPoint("/", fsgofer.Config{
- ROMount: spec.Root.Readonly || conf.Overlay,
- PanicOnWrite: g.panicOnWrite,
+ ROMount: spec.Root.Readonly || conf.Overlay,
})
if err != nil {
Fatalf("creating attach point: %v", err)
@@ -181,9 +178,8 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
for _, m := range spec.Mounts {
if specutils.Is9PMount(m) {
cfg := fsgofer.Config{
- ROMount: isReadonlyMount(m.Options) || conf.Overlay,
- PanicOnWrite: g.panicOnWrite,
- HostUDS: conf.FSGoferHostUDS,
+ ROMount: isReadonlyMount(m.Options) || conf.Overlay,
+ HostUDS: conf.FSGoferHostUDS,
}
ap, err := fsgofer.NewAttachPoint(m.Destination, cfg)
if err != nil {
@@ -316,6 +312,7 @@ func setupRootFS(spec *specs.Spec, conf *config.Config) error {
if err != nil {
return fmt.Errorf("resolving symlinks to %q: %v", spec.Process.Cwd, err)
}
+ log.Infof("Create working directory %q if needed", spec.Process.Cwd)
if err := os.MkdirAll(dst, 0755); err != nil {
return fmt.Errorf("creating working directory %q: %v", spec.Process.Cwd, err)
}
diff --git a/runsc/container/container.go b/runsc/container/container.go
index 6e1d6a568..63478ba8c 100644
--- a/runsc/container/container.go
+++ b/runsc/container/container.go
@@ -902,9 +902,6 @@ func (c *Container) createGoferProcess(spec *specs.Spec, conf *config.Config, bu
}
args = append(args, "gofer", "--bundle", bundleDir)
- if conf.Overlay {
- args = append(args, "--panic-on-write=true")
- }
// Open the spec file to donate to the sandbox.
specFile, err := specutils.OpenSpec(bundleDir)
diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go
index 33ada5bb9..1f8e277cc 100644
--- a/runsc/container/container_test.go
+++ b/runsc/container/container_test.go
@@ -293,22 +293,20 @@ var (
func configs(t *testing.T, opts ...configOption) map[string]*config.Config {
// Always load the default config.
cs := make(map[string]*config.Config)
+ testutil.TestConfig(t)
for _, o := range opts {
+ c := testutil.TestConfig(t)
switch o {
case overlay:
- c := testutil.TestConfig(t)
c.Overlay = true
cs["overlay"] = c
case ptrace:
- c := testutil.TestConfig(t)
c.Platform = platforms.Ptrace
cs["ptrace"] = c
case kvm:
- c := testutil.TestConfig(t)
c.Platform = platforms.KVM
cs["kvm"] = c
case nonExclusiveFS:
- c := testutil.TestConfig(t)
c.FileAccess = config.FileAccessShared
cs["non-exclusive"] = c
default:
@@ -318,23 +316,14 @@ func configs(t *testing.T, opts ...configOption) map[string]*config.Config {
return cs
}
+// TODO(gvisor.dev/issue/1624): Merge with configs when VFS2 is the default.
func configsWithVFS2(t *testing.T, opts ...configOption) map[string]*config.Config {
- vfs1 := configs(t, opts...)
-
- var optsVFS2 []configOption
- for _, opt := range opts {
- // TODO(gvisor.dev/issue/1487): Enable overlay tests.
- if opt != overlay {
- optsVFS2 = append(optsVFS2, opt)
- }
- }
-
- for key, value := range configs(t, optsVFS2...) {
+ all := configs(t, opts...)
+ for key, value := range configs(t, opts...) {
value.VFS2 = true
- vfs1[key+"VFS2"] = value
+ all[key+"VFS2"] = value
}
-
- return vfs1
+ return all
}
// TestLifecycle tests the basic Create/Start/Signal/Destroy container lifecycle.
@@ -513,7 +502,7 @@ func TestExePath(t *testing.T) {
t.Fatalf("error making directory: %v", err)
}
- for name, conf := range configsWithVFS2(t, overlay) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
for _, test := range []struct {
path string
@@ -838,7 +827,7 @@ func TestExecProcList(t *testing.T) {
// TestKillPid verifies that we can signal individual exec'd processes.
func TestKillPid(t *testing.T) {
- for name, conf := range configsWithVFS2(t, overlay) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
app, err := testutil.FindFile("test/cmd/test_app/test_app")
if err != nil {
@@ -906,13 +895,15 @@ func TestKillPid(t *testing.T) {
}
}
-// TestCheckpointRestore creates a container that continuously writes successive integers
-// to a file. To test checkpoint and restore functionality, the container is
-// checkpointed and the last number printed to the file is recorded. Then, it is restored in two
-// new containers and the first number printed from these containers is checked. Both should
-// be the next consecutive number after the last number from the checkpointed container.
+// TestCheckpointRestore creates a container that continuously writes successive
+// integers to a file. To test checkpoint and restore functionality, the
+// container is checkpointed and the last number printed to the file is
+// recorded. Then, it is restored in two new containers and the first number
+// printed from these containers is checked. Both should be the next consecutive
+// number after the last number from the checkpointed container.
func TestCheckpointRestore(t *testing.T) {
// Skip overlay because test requires writing to host file.
+ // TODO(gvisor.dev/issue/1663): Add VFS when S/R support is added.
for name, conf := range configs(t, noOverlay...) {
t.Run(name, func(t *testing.T) {
dir, err := ioutil.TempDir(testutil.TmpDir(), "checkpoint-test")
@@ -1074,6 +1065,7 @@ func TestCheckpointRestore(t *testing.T) {
// with filesystem Unix Domain Socket use.
func TestUnixDomainSockets(t *testing.T) {
// Skip overlay because test requires writing to host file.
+ // TODO(gvisor.dev/issue/1663): Add VFS when S/R support is added.
for name, conf := range configs(t, noOverlay...) {
t.Run(name, func(t *testing.T) {
// UDS path is limited to 108 chars for compatibility with older systems.
@@ -1211,7 +1203,7 @@ func TestUnixDomainSockets(t *testing.T) {
// recreated. Then it resumes the container, verify that the file gets created
// again.
func TestPauseResume(t *testing.T) {
- for name, conf := range configs(t, noOverlay...) {
+ for name, conf := range configsWithVFS2(t, noOverlay...) {
t.Run(name, func(t *testing.T) {
tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "lock")
if err != nil {
@@ -1471,7 +1463,7 @@ func TestRunNonRoot(t *testing.T) {
// TestMountNewDir checks that runsc will create destination directory if it
// doesn't exit.
func TestMountNewDir(t *testing.T) {
- for name, conf := range configsWithVFS2(t, overlay) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
root, err := ioutil.TempDir(testutil.TmpDir(), "root")
if err != nil {
@@ -1855,8 +1847,9 @@ func TestUserLog(t *testing.T) {
t.Fatal("error finding test_app:", err)
}
- // sched_rr_get_interval = 148 - not implemented in gvisor.
- spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall=148")
+ // sched_rr_get_interval - not implemented in gvisor.
+ num := strconv.Itoa(syscall.SYS_SCHED_RR_GET_INTERVAL)
+ spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall="+num)
conf := testutil.TestConfig(t)
_, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf)
if err != nil {
@@ -2038,7 +2031,7 @@ func doDestroyStartingTest(t *testing.T, vfs2 bool) {
}
func TestCreateWorkingDir(t *testing.T) {
- for name, conf := range configsWithVFS2(t, overlay) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "cwd-create")
if err != nil {
@@ -2153,7 +2146,7 @@ func TestMountPropagation(t *testing.T) {
}
func TestMountSymlink(t *testing.T) {
- for name, conf := range configsWithVFS2(t, overlay) {
+ for name, conf := range configsWithVFS2(t, all...) {
t.Run(name, func(t *testing.T) {
dir, err := ioutil.TempDir(testutil.TmpDir(), "mount-symlink")
if err != nil {
diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go
index 952215ec1..850e80290 100644
--- a/runsc/container/multi_container_test.go
+++ b/runsc/container/multi_container_test.go
@@ -480,7 +480,7 @@ func TestMultiContainerMount(t *testing.T) {
// TestMultiContainerSignal checks that it is possible to signal individual
// containers without killing the entire sandbox.
func TestMultiContainerSignal(t *testing.T) {
- for 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 {
@@ -1691,12 +1691,11 @@ func TestMultiContainerRunNonRoot(t *testing.T) {
}
// TestMultiContainerHomeEnvDir tests that the HOME environment variable is set
-// for root containers, sub-containers, and execed processes.
+// for root containers, sub-containers, and exec'ed processes.
func TestMultiContainerHomeEnvDir(t *testing.T) {
- // TODO(gvisor.dev/issue/1487): VFSv2 configs failing.
// NOTE: Don't use overlay since we need changes to persist to the temp dir
// outside the sandbox.
- for testName, conf := range configs(t, noOverlay...) {
+ for testName, conf := range configsWithVFS2(t, noOverlay...) {
t.Run(testName, func(t *testing.T) {
rootDir, cleanup, err := testutil.SetupRootDir()
@@ -1718,9 +1717,9 @@ func TestMultiContainerHomeEnvDir(t *testing.T) {
// We will sleep in the root container in order to ensure that the root
//container doesn't terminate before sub containers can be created.
- rootCmd := []string{"/bin/sh", "-c", fmt.Sprintf("printf \"$HOME\" > %s; sleep 1000", homeDirs["root"].Name())}
- subCmd := []string{"/bin/sh", "-c", fmt.Sprintf("printf \"$HOME\" > %s", homeDirs["sub"].Name())}
- execCmd := fmt.Sprintf("printf \"$HOME\" > %s", homeDirs["exec"].Name())
+ rootCmd := []string{"/bin/sh", "-c", fmt.Sprintf(`printf "$HOME" > %s; sleep 1000`, homeDirs["root"].Name())}
+ subCmd := []string{"/bin/sh", "-c", fmt.Sprintf(`printf "$HOME" > %s`, homeDirs["sub"].Name())}
+ execCmd := fmt.Sprintf(`printf "$HOME" > %s`, homeDirs["exec"].Name())
// Setup the containers, a root container and sub container.
specConfig, ids := createSpecs(rootCmd, subCmd)
diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go
index 0cb9b1cae..39b8a0b1e 100644
--- a/runsc/fsgofer/filter/config.go
+++ b/runsc/fsgofer/filter/config.go
@@ -27,20 +27,9 @@ import (
var allowedSyscalls = seccomp.SyscallRules{
syscall.SYS_ACCEPT: {},
syscall.SYS_CLOCK_GETTIME: {},
- syscall.SYS_CLONE: []seccomp.Rule{
- {
- seccomp.EqualTo(
- syscall.CLONE_VM |
- syscall.CLONE_FS |
- syscall.CLONE_FILES |
- syscall.CLONE_SIGHAND |
- syscall.CLONE_SYSVSEM |
- syscall.CLONE_THREAD),
- },
- },
- syscall.SYS_CLOSE: {},
- syscall.SYS_DUP: {},
- syscall.SYS_EPOLL_CTL: {},
+ syscall.SYS_CLOSE: {},
+ syscall.SYS_DUP: {},
+ syscall.SYS_EPOLL_CTL: {},
syscall.SYS_EPOLL_PWAIT: []seccomp.Rule{
{
seccomp.MatchAny{},
diff --git a/runsc/fsgofer/filter/config_amd64.go b/runsc/fsgofer/filter/config_amd64.go
index 53506b5e1..686753d96 100644
--- a/runsc/fsgofer/filter/config_amd64.go
+++ b/runsc/fsgofer/filter/config_amd64.go
@@ -25,9 +25,42 @@ import (
func init() {
allowedSyscalls[syscall.SYS_ARCH_PRCTL] = []seccomp.Rule{
- {seccomp.EqualTo(linux.ARCH_GET_FS)},
+ // TODO(b/168828518): No longer used in Go 1.16+.
{seccomp.EqualTo(linux.ARCH_SET_FS)},
}
+ allowedSyscalls[syscall.SYS_CLONE] = []seccomp.Rule{
+ // parent_tidptr and child_tidptr are always 0 because neither
+ // CLONE_PARENT_SETTID nor CLONE_CHILD_SETTID are used.
+ {
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SETTLS |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ seccomp.EqualTo(0), // parent_tidptr
+ seccomp.EqualTo(0), // child_tidptr
+ seccomp.MatchAny{}, // tls
+ },
+ {
+ // TODO(b/168828518): No longer used in Go 1.16+ (on amd64).
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ seccomp.EqualTo(0), // parent_tidptr
+ seccomp.EqualTo(0), // child_tidptr
+ seccomp.MatchAny{}, // tls
+ },
+ }
+
allowedSyscalls[syscall.SYS_NEWFSTATAT] = []seccomp.Rule{}
}
diff --git a/runsc/fsgofer/filter/config_arm64.go b/runsc/fsgofer/filter/config_arm64.go
index d2697deb7..ff0cf77a0 100644
--- a/runsc/fsgofer/filter/config_arm64.go
+++ b/runsc/fsgofer/filter/config_arm64.go
@@ -23,5 +23,26 @@ import (
)
func init() {
+ allowedSyscalls[syscall.SYS_CLONE] = []seccomp.Rule{
+ // parent_tidptr and child_tidptr are always 0 because neither
+ // CLONE_PARENT_SETTID nor CLONE_CHILD_SETTID are used.
+ {
+ seccomp.EqualTo(
+ syscall.CLONE_VM |
+ syscall.CLONE_FS |
+ syscall.CLONE_FILES |
+ syscall.CLONE_SIGHAND |
+ syscall.CLONE_SYSVSEM |
+ syscall.CLONE_THREAD),
+ seccomp.MatchAny{}, // newsp
+ // These arguments are left uninitialized by the Go
+ // runtime, so they may be anything (and are unused by
+ // the host).
+ seccomp.MatchAny{}, // parent_tidptr
+ seccomp.MatchAny{}, // tls
+ seccomp.MatchAny{}, // child_tidptr
+ },
+ }
+
allowedSyscalls[syscall.SYS_FSTATAT] = []seccomp.Rule{}
}
diff --git a/runsc/fsgofer/filter/extra_filters_race.go b/runsc/fsgofer/filter/extra_filters_race.go
index 885c92f7a..20a0732be 100644
--- a/runsc/fsgofer/filter/extra_filters_race.go
+++ b/runsc/fsgofer/filter/extra_filters_race.go
@@ -35,6 +35,7 @@ func instrumentationFilters() seccomp.SyscallRules {
syscall.SYS_MUNLOCK: {},
syscall.SYS_NANOSLEEP: {},
syscall.SYS_OPEN: {},
+ syscall.SYS_OPENAT: {},
syscall.SYS_SET_ROBUST_LIST: {},
// Used within glibc's malloc.
syscall.SYS_TIME: {},
diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go
index 4268d97a1..0b628c8ce 100644
--- a/runsc/fsgofer/fsgofer.go
+++ b/runsc/fsgofer/fsgofer.go
@@ -1181,9 +1181,6 @@ func extractErrno(err error) unix.Errno {
func (l *localFile) checkROMount() error {
if conf := l.attachPoint.conf; conf.ROMount {
- if conf.PanicOnWrite {
- panic("attempt to write to RO mount")
- }
return unix.EROFS
}
return nil
diff --git a/runsc/fsgofer/fsgofer_test.go b/runsc/fsgofer/fsgofer_test.go
index 0e4945b3d..a84206686 100644
--- a/runsc/fsgofer/fsgofer_test.go
+++ b/runsc/fsgofer/fsgofer_test.go
@@ -553,29 +553,6 @@ func TestROMountChecks(t *testing.T) {
})
}
-func TestROMountPanics(t *testing.T) {
- conf := Config{ROMount: true, PanicOnWrite: true}
- uid := p9.UID(os.Getuid())
- gid := p9.GID(os.Getgid())
-
- runCustom(t, allTypes, []Config{conf}, func(t *testing.T, s state) {
- if s.fileType != unix.S_IFLNK {
- assertPanic(t, func() { s.file.Open(p9.WriteOnly) })
- }
- assertPanic(t, func() { s.file.Create("some_file", p9.ReadWrite, 0777, uid, gid) })
- assertPanic(t, func() { s.file.Mkdir("some_dir", 0777, uid, gid) })
- assertPanic(t, func() { s.file.RenameAt("some_file", s.file, "other_file") })
- assertPanic(t, func() { s.file.Symlink("some_place", "some_symlink", uid, gid) })
- assertPanic(t, func() { s.file.UnlinkAt("some_file", 0) })
- assertPanic(t, func() { s.file.Link(s.file, "some_link") })
- assertPanic(t, func() { s.file.Mknod("some-nod", 0777, 1, 2, uid, gid) })
-
- valid := p9.SetAttrMask{Size: true}
- attr := p9.SetAttr{Size: 0}
- assertPanic(t, func() { s.file.SetAttr(valid, attr) })
- })
-}
-
func TestWalkNotFound(t *testing.T) {
runCustom(t, []uint32{unix.S_IFDIR}, allConfs, func(t *testing.T, s state) {
if _, _, err := s.file.Walk([]string{"nobody-here"}); err != unix.ENOENT {
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index a8f4f64a5..c4309feb3 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -72,11 +72,14 @@ type Sandbox struct {
// will have it as a child process.
child bool
- // status is an exit status of a sandbox process.
- status syscall.WaitStatus
-
// statusMu protects status.
statusMu sync.Mutex
+
+ // status is the exit status of a sandbox process. It's only set if the
+ // child==true and the sandbox was waited on. This field allows for multiple
+ // threads to wait on sandbox and get the exit code, since Linux will return
+ // WaitStatus to one of the waiters only.
+ status syscall.WaitStatus
}
// Args is used to configure a new sandbox.
@@ -746,35 +749,47 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
// Wait waits for the containerized process to exit, and returns its WaitStatus.
func (s *Sandbox) Wait(cid string) (syscall.WaitStatus, error) {
log.Debugf("Waiting for container %q in sandbox %q", cid, s.ID)
- var ws syscall.WaitStatus
if conn, err := s.sandboxConnect(); err != nil {
- // The sandbox may have exited while before we had a chance to
- // wait on it.
+ // The sandbox may have exited while before we had a chance to wait on it.
+ // There is nothing we can do for subcontainers. For the init container, we
+ // can try to get the sandbox exit code.
+ if !s.IsRootContainer(cid) {
+ return syscall.WaitStatus(0), err
+ }
log.Warningf("Wait on container %q failed: %v. Will try waiting on the sandbox process instead.", cid, err)
} else {
defer conn.Close()
+
// Try the Wait RPC to the sandbox.
+ var ws syscall.WaitStatus
err = conn.Call(boot.ContainerWait, &cid, &ws)
if err == nil {
// It worked!
return ws, nil
}
+ // See comment above.
+ if !s.IsRootContainer(cid) {
+ return syscall.WaitStatus(0), err
+ }
+
// The sandbox may have exited after we connected, but before
// or during the Wait RPC.
log.Warningf("Wait RPC to container %q failed: %v. Will try waiting on the sandbox process instead.", cid, err)
}
- // The sandbox may have already exited, or exited while handling the
- // Wait RPC. The best we can do is ask Linux what the sandbox exit
- // status was, since in most cases that will be the same as the
- // container exit status.
+ // The sandbox may have already exited, or exited while handling the Wait RPC.
+ // The best we can do is ask Linux what the sandbox exit status was, since in
+ // most cases that will be the same as the container exit status.
if err := s.waitForStopped(); err != nil {
- return ws, err
+ return syscall.WaitStatus(0), err
}
if !s.child {
- return ws, fmt.Errorf("sandbox no longer running and its exit status is unavailable")
+ return syscall.WaitStatus(0), fmt.Errorf("sandbox no longer running and its exit status is unavailable")
}
+
+ s.statusMu.Lock()
+ defer s.statusMu.Unlock()
return s.status, nil
}
diff --git a/runsc/specutils/seccomp/seccomp_test.go b/runsc/specutils/seccomp/seccomp_test.go
index 2079cd2e9..850c237ba 100644
--- a/runsc/specutils/seccomp/seccomp_test.go
+++ b/runsc/specutils/seccomp/seccomp_test.go
@@ -169,7 +169,7 @@ var (
},
},
},
- input: testInput(nativeArchAuditNo, "open", nil),
+ input: testInput(nativeArchAuditNo, "openat", nil),
expected: uint32(allowAction),
},
{
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-27 06:51:23 +0000