summaryrefslogtreecommitdiffhomepage
path: root/runsc
diff options
context:
space:
mode:
Diffstat (limited to 'runsc')
-rw-r--r--runsc/boot/compat.go8
-rw-r--r--runsc/boot/compat_amd64.go22
-rw-r--r--runsc/boot/compat_arm64.go22
-rw-r--r--runsc/boot/config.go6
-rw-r--r--runsc/boot/fs.go89
-rw-r--r--runsc/boot/vfs.go187
-rw-r--r--runsc/cgroup/BUILD2
-rw-r--r--runsc/cgroup/cgroup.go54
-rw-r--r--runsc/cmd/gofer.go4
-rw-r--r--runsc/cmd/help.go12
-rw-r--r--runsc/cmd/spec.go18
-rw-r--r--runsc/container/BUILD2
-rw-r--r--runsc/container/container.go7
-rw-r--r--runsc/container/container_test.go61
-rw-r--r--runsc/container/multi_container_test.go28
-rw-r--r--runsc/fsgofer/BUILD2
-rw-r--r--runsc/fsgofer/fsgofer.go8
-rw-r--r--runsc/main.go2
-rw-r--r--runsc/sandbox/BUILD1
-rw-r--r--runsc/sandbox/sandbox.go3
-rw-r--r--runsc/specutils/specutils.go44
21 files changed, 227 insertions, 355 deletions
diff --git a/runsc/boot/compat.go b/runsc/boot/compat.go
index b7cfb35bf..84c67cbc2 100644
--- a/runsc/boot/compat.go
+++ b/runsc/boot/compat.go
@@ -119,7 +119,13 @@ func (c *compatEmitter) emitUnimplementedSyscall(us *spb.UnimplementedSyscall) {
}
if tr.shouldReport(regs) {
- c.sink.Infof("Unsupported syscall: %s, regs: %+v", c.nameMap.Name(uintptr(sysnr)), regs)
+ name := c.nameMap.Name(uintptr(sysnr))
+ c.sink.Infof("Unsupported syscall %s(%#x,%#x,%#x,%#x,%#x,%#x). It is "+
+ "likely that you can safely ignore this message and that this is not "+
+ "the cause of any error. Please, refer to %s/%s for more information.",
+ name, argVal(0, regs), argVal(1, regs), argVal(2, regs), argVal(3, regs),
+ argVal(4, regs), argVal(5, regs), syscallLink, name)
+
tr.onReported(regs)
}
}
diff --git a/runsc/boot/compat_amd64.go b/runsc/boot/compat_amd64.go
index 42b0ca8b0..8eb76b2ba 100644
--- a/runsc/boot/compat_amd64.go
+++ b/runsc/boot/compat_amd64.go
@@ -24,8 +24,12 @@ import (
"gvisor.dev/gvisor/pkg/sentry/strace"
)
-// reportLimit is the max number of events that should be reported per tracker.
-const reportLimit = 100
+const (
+ // reportLimit is the max number of events that should be reported per
+ // tracker.
+ reportLimit = 100
+ syscallLink = "https://gvisor.dev/c/linux/amd64"
+)
// newRegs create a empty Registers instance.
func newRegs() *rpb.Registers {
@@ -36,22 +40,22 @@ func newRegs() *rpb.Registers {
}
}
-func argVal(argIdx int, regs *rpb.Registers) uint32 {
+func argVal(argIdx int, regs *rpb.Registers) uint64 {
amd64Regs := regs.GetArch().(*rpb.Registers_Amd64).Amd64
switch argIdx {
case 0:
- return uint32(amd64Regs.Rdi)
+ return amd64Regs.Rdi
case 1:
- return uint32(amd64Regs.Rsi)
+ return amd64Regs.Rsi
case 2:
- return uint32(amd64Regs.Rdx)
+ return amd64Regs.Rdx
case 3:
- return uint32(amd64Regs.R10)
+ return amd64Regs.R10
case 4:
- return uint32(amd64Regs.R8)
+ return amd64Regs.R8
case 5:
- return uint32(amd64Regs.R9)
+ return amd64Regs.R9
}
panic(fmt.Sprintf("invalid syscall argument index %d", argIdx))
}
diff --git a/runsc/boot/compat_arm64.go b/runsc/boot/compat_arm64.go
index f784cd237..bce9d95b3 100644
--- a/runsc/boot/compat_arm64.go
+++ b/runsc/boot/compat_arm64.go
@@ -23,8 +23,12 @@ import (
"gvisor.dev/gvisor/pkg/sentry/strace"
)
-// reportLimit is the max number of events that should be reported per tracker.
-const reportLimit = 100
+const (
+ // reportLimit is the max number of events that should be reported per
+ // tracker.
+ reportLimit = 100
+ syscallLink = "https://gvisor.dev/c/linux/arm64"
+)
// newRegs create a empty Registers instance.
func newRegs() *rpb.Registers {
@@ -35,22 +39,22 @@ func newRegs() *rpb.Registers {
}
}
-func argVal(argIdx int, regs *rpb.Registers) uint32 {
+func argVal(argIdx int, regs *rpb.Registers) uint64 {
arm64Regs := regs.GetArch().(*rpb.Registers_Arm64).Arm64
switch argIdx {
case 0:
- return uint32(arm64Regs.R0)
+ return arm64Regs.R0
case 1:
- return uint32(arm64Regs.R1)
+ return arm64Regs.R1
case 2:
- return uint32(arm64Regs.R2)
+ return arm64Regs.R2
case 3:
- return uint32(arm64Regs.R3)
+ return arm64Regs.R3
case 4:
- return uint32(arm64Regs.R4)
+ return arm64Regs.R4
case 5:
- return uint32(arm64Regs.R5)
+ return arm64Regs.R5
}
panic(fmt.Sprintf("invalid syscall argument index %d", argIdx))
}
diff --git a/runsc/boot/config.go b/runsc/boot/config.go
index 6d6a705f8..bcec7e4db 100644
--- a/runsc/boot/config.go
+++ b/runsc/boot/config.go
@@ -241,8 +241,10 @@ type Config struct {
// ReferenceLeakMode sets reference leak check mode
ReferenceLeakMode refs.LeakMode
- // OverlayfsStaleRead causes cached FDs to reopen after a file is opened for
- // write to workaround overlayfs limitation on kernels before 4.19.
+ // OverlayfsStaleRead instructs the sandbox to assume that the root mount
+ // is on a Linux overlayfs mount, which does not necessarily preserve
+ // coherence between read-only and subsequent writable file descriptors
+ // representing the "same" file.
OverlayfsStaleRead bool
// TestOnlyAllowRunAsCurrentUserWithoutChroot should only be used in
diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go
index 4875452e2..52f8344ca 100644
--- a/runsc/boot/fs.go
+++ b/runsc/boot/fs.go
@@ -37,6 +37,13 @@ import (
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/fs/gofer"
"gvisor.dev/gvisor/pkg/sentry/fs/ramfs"
+ "gvisor.dev/gvisor/pkg/sentry/fs/user"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs"
+ gofervfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer"
+ procvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc"
+ sysvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys"
+ tmpfsvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/syserror"
@@ -44,23 +51,15 @@ import (
)
const (
- // Filesystem name for 9p gofer mounts.
- rootFsName = "9p"
-
// Device name for root mount.
rootDevice = "9pfs-/"
// MountPrefix is the annotation prefix for mount hints.
MountPrefix = "dev.gvisor.spec.mount."
- // Filesystems that runsc supports.
- bind = "bind"
- devpts = "devpts"
- devtmpfs = "devtmpfs"
- proc = "proc"
- sysfs = "sysfs"
- tmpfs = "tmpfs"
- nonefs = "none"
+ // Supported filesystems that map to different internal filesystem.
+ bind = "bind"
+ nonefs = "none"
)
// tmpfs has some extra supported options that we must pass through.
@@ -108,12 +107,12 @@ func compileMounts(spec *specs.Spec) []specs.Mount {
// Always mount /dev.
mounts = append(mounts, specs.Mount{
- Type: devtmpfs,
+ Type: devtmpfs.Name,
Destination: "/dev",
})
mounts = append(mounts, specs.Mount{
- Type: devpts,
+ Type: devpts.Name,
Destination: "/dev/pts",
})
@@ -137,13 +136,13 @@ func compileMounts(spec *specs.Spec) []specs.Mount {
var mandatoryMounts []specs.Mount
if !procMounted {
mandatoryMounts = append(mandatoryMounts, specs.Mount{
- Type: proc,
+ Type: procvfs2.Name,
Destination: "/proc",
})
}
if !sysMounted {
mandatoryMounts = append(mandatoryMounts, specs.Mount{
- Type: sysfs,
+ Type: sysvfs2.Name,
Destination: "/sys",
})
}
@@ -156,12 +155,16 @@ func compileMounts(spec *specs.Spec) []specs.Mount {
}
// p9MountOptions creates a slice of options for a p9 mount.
-func p9MountOptions(fd int, fa FileAccessType) []string {
+func p9MountOptions(fd int, fa FileAccessType, vfs2 bool) []string {
opts := []string{
"trans=fd",
"rfdno=" + strconv.Itoa(fd),
"wfdno=" + strconv.Itoa(fd),
- "privateunixsocket=true",
+ }
+ if !vfs2 {
+ // privateunixsocket is always enabled in VFS2. VFS1 requires explicit
+ // enablement.
+ opts = append(opts, "privateunixsocket=true")
}
if fa == FileAccessShared {
opts = append(opts, "cache=remote_revalidating")
@@ -219,9 +222,6 @@ func mountFlags(opts []string) fs.MountSourceFlags {
mf.NoAtime = true
case "noexec":
mf.NoExec = true
- case "bind", "rbind":
- // When options include either "bind" or "rbind",
- // it's converted to a 9P mount.
default:
log.Warningf("ignoring unknown mount option %q", o)
}
@@ -234,7 +234,7 @@ func isSupportedMountFlag(fstype, opt string) bool {
case "rw", "ro", "noatime", "noexec":
return true
}
- if fstype == tmpfs {
+ if fstype == tmpfsvfs2.Name {
ok, err := parseMountOption(opt, tmpfsAllowedOptions...)
return ok && err == nil
}
@@ -292,17 +292,10 @@ func setupContainerFS(ctx context.Context, conf *Config, mntr *containerMounter,
// Set namespace here so that it can be found in ctx.
procArgs.MountNamespace = mns
- return setExecutablePath(ctx, procArgs)
-}
-
-// setExecutablePath sets the procArgs.Filename by searching the PATH for an
-// executable matching the procArgs.Argv[0].
-func setExecutablePath(ctx context.Context, procArgs *kernel.CreateProcessArgs) error {
- paths := fs.GetPath(procArgs.Envv)
- exe := procArgs.Argv[0]
- f, err := procArgs.MountNamespace.ResolveExecutablePath(ctx, procArgs.WorkingDirectory, exe, paths)
+ // Resolve the executable path from working dir and environment.
+ f, err := user.ResolveExecutablePath(ctx, procArgs.Credentials, procArgs.MountNamespace, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0])
if err != nil {
- return fmt.Errorf("searching for executable %q, cwd: %q, $PATH=%q: %v", exe, procArgs.WorkingDirectory, strings.Join(paths, ":"), err)
+ return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err)
}
procArgs.Filename = f
return nil
@@ -444,7 +437,7 @@ func (m *mountHint) setOptions(val string) error {
}
func (m *mountHint) isSupported() bool {
- return m.mount.Type == tmpfs && m.share == pod
+ return m.mount.Type == tmpfsvfs2.Name && m.share == pod
}
// checkCompatible verifies that shared mount is compatible with master.
@@ -586,7 +579,7 @@ func (c *containerMounter) processHints(conf *Config) error {
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 {
+ if hint.mount.Type != tmpfsvfs2.Name {
continue
}
log.Infof("Mounting master of shared mount %q from %q type %q", hint.name, hint.mount.Source, hint.mount.Type)
@@ -723,7 +716,7 @@ func (c *containerMounter) createRootMount(ctx context.Context, conf *Config) (*
fd := c.fds.remove()
log.Infof("Mounting root over 9P, ioFD: %d", fd)
p9FS := mustFindFilesystem("9p")
- opts := p9MountOptions(fd, conf.FileAccess)
+ opts := p9MountOptions(fd, conf.FileAccess, false /* vfs2 */)
if conf.OverlayfsStaleRead {
// We can't check for overlayfs here because sandbox is chroot'ed and gofer
@@ -768,22 +761,12 @@ func (c *containerMounter) getMountNameAndOptions(conf *Config, m specs.Mount) (
useOverlay bool
)
- for _, opt := range m.Options {
- // When options include either "bind" or "rbind", this behaves as
- // bind mount even if the mount type is equal to a filesystem supported
- // on runsc.
- if opt == "bind" || opt == "rbind" {
- m.Type = bind
- break
- }
- }
-
switch m.Type {
- case devpts, devtmpfs, proc, sysfs:
+ case devpts.Name, devtmpfs.Name, procvfs2.Name, sysvfs2.Name:
fsName = m.Type
case nonefs:
- fsName = sysfs
- case tmpfs:
+ fsName = sysvfs2.Name
+ case tmpfsvfs2.Name:
fsName = m.Type
var err error
@@ -794,8 +777,8 @@ func (c *containerMounter) getMountNameAndOptions(conf *Config, m specs.Mount) (
case bind:
fd := c.fds.remove()
- fsName = "9p"
- opts = p9MountOptions(fd, c.getMountAccessType(m))
+ fsName = gofervfs2.Name
+ opts = p9MountOptions(fd, c.getMountAccessType(m), conf.VFS2)
// If configured, add overlay to all writable mounts.
useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly
@@ -948,7 +931,7 @@ func (c *containerMounter) createRestoreEnvironment(conf *Config) (*fs.RestoreEn
// Add root mount.
fd := c.fds.remove()
- opts := p9MountOptions(fd, conf.FileAccess)
+ opts := p9MountOptions(fd, conf.FileAccess, false /* vfs2 */)
mf := fs.MountSourceFlags{}
if c.root.Readonly || conf.Overlay {
@@ -960,7 +943,7 @@ func (c *containerMounter) createRestoreEnvironment(conf *Config) (*fs.RestoreEn
Flags: mf,
DataString: strings.Join(opts, ","),
}
- renv.MountSources[rootFsName] = append(renv.MountSources[rootFsName], rootMount)
+ renv.MountSources[gofervfs2.Name] = append(renv.MountSources[gofervfs2.Name], rootMount)
// Add submounts.
var tmpMounted bool
@@ -976,7 +959,7 @@ func (c *containerMounter) createRestoreEnvironment(conf *Config) (*fs.RestoreEn
// TODO(b/67958150): handle '/tmp' properly (see mountTmp()).
if !tmpMounted {
tmpMount := specs.Mount{
- Type: tmpfs,
+ Type: tmpfsvfs2.Name,
Destination: "/tmp",
}
if err := c.addRestoreMount(conf, renv, tmpMount); err != nil {
@@ -1032,7 +1015,7 @@ func (c *containerMounter) mountTmp(ctx context.Context, conf *Config, mns *fs.M
// No '/tmp' found (or fallthrough from above). Safe to mount internal
// tmpfs.
tmpMount := specs.Mount{
- Type: tmpfs,
+ Type: tmpfsvfs2.Name,
Destination: "/tmp",
// Sticky bit is added to prevent accidental deletion of files from
// another user. This is normally done for /tmp.
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index ef95d0c65..f48e6b0f1 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -18,52 +18,50 @@ import (
"fmt"
"path"
"sort"
- "strconv"
"strings"
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/fspath"
- "gvisor.dev/gvisor/pkg/sentry/devices/memdev"
- "gvisor.dev/gvisor/pkg/sentry/fs"
- devpts2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts"
- devtmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs"
- goferimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer"
- procimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc"
- sysimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys"
- tmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs"
- "gvisor.dev/gvisor/pkg/syserror"
-
"gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/log"
+ "gvisor.dev/gvisor/pkg/sentry/devices/memdev"
+ "gvisor.dev/gvisor/pkg/sentry/fs/user"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
)
func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error {
- vfsObj.MustRegisterFilesystemType(devpts2.Name, &devpts2.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(devpts.Name, &devpts.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserList: true,
// TODO(b/29356795): Users may mount this once the terminals are in a
// usable state.
AllowUserMount: false,
})
- vfsObj.MustRegisterFilesystemType(devtmpfsimpl.Name, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(devtmpfs.Name, &devtmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
AllowUserList: true,
})
- vfsObj.MustRegisterFilesystemType(goferimpl.Name, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(gofer.Name, &gofer.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserList: true,
})
- vfsObj.MustRegisterFilesystemType(procimpl.Name, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(proc.Name, &proc.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
AllowUserList: true,
})
- vfsObj.MustRegisterFilesystemType(sysimpl.Name, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(sys.Name, &sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
AllowUserList: true,
})
- vfsObj.MustRegisterFilesystemType(tmpfsimpl.Name, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
+ vfsObj.MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{
AllowUserMount: true,
AllowUserList: true,
})
@@ -72,7 +70,7 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre
if err := memdev.Register(vfsObj); err != nil {
return fmt.Errorf("registering memdev: %w", err)
}
- a, err := devtmpfsimpl.NewAccessor(ctx, vfsObj, creds, devtmpfsimpl.Name)
+ a, err := devtmpfs.NewAccessor(ctx, vfsObj, creds, devtmpfs.Name)
if err != nil {
return fmt.Errorf("creating devtmpfs accessor: %w", err)
}
@@ -96,69 +94,14 @@ func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounte
return fmt.Errorf("failed to setupFS: %w", err)
}
procArgs.MountNamespaceVFS2 = mns
- return setExecutablePathVFS2(ctx, procArgs)
-}
-
-func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error {
- exe := procArgs.Argv[0]
-
- // Absolute paths can be used directly.
- if path.IsAbs(exe) {
- procArgs.Filename = exe
- return nil
- }
- // Paths with '/' in them should be joined to the working directory, or
- // to the root if working directory is not set.
- if strings.IndexByte(exe, '/') > 0 {
- if !path.IsAbs(procArgs.WorkingDirectory) {
- return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory)
- }
- procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe)
- return nil
- }
-
- // Paths with a '/' are relative to the CWD.
- if strings.IndexByte(exe, '/') > 0 {
- procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe)
- return nil
- }
-
- // Otherwise, We must lookup the name in the paths, starting from the
- // root directory.
- root := procArgs.MountNamespaceVFS2.Root()
- defer root.DecRef()
-
- paths := fs.GetPath(procArgs.Envv)
- creds := procArgs.Credentials
-
- for _, p := range paths {
- binPath := path.Join(p, exe)
- pop := &vfs.PathOperation{
- Root: root,
- Start: root,
- Path: fspath.Parse(binPath),
- FollowFinalSymlink: true,
- }
- opts := &vfs.OpenOptions{
- FileExec: true,
- Flags: linux.O_RDONLY,
- }
- dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts)
- if err == syserror.ENOENT || err == syserror.EACCES {
- // Didn't find it here.
- continue
- }
- if err != nil {
- return err
- }
- dentry.DecRef()
-
- procArgs.Filename = binPath
- return nil
+ // Resolve the executable path from working dir and environment.
+ f, err := user.ResolveExecutablePathVFS2(ctx, procArgs.Credentials, procArgs.MountNamespaceVFS2, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0])
+ if err != nil {
+ return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err)
}
-
- return fmt.Errorf("executable %q not found in $PATH=%q", exe, strings.Join(paths, ":"))
+ procArgs.Filename = f
+ return nil
}
func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs *kernel.CreateProcessArgs) (*vfs.MountNamespace, error) {
@@ -193,10 +136,10 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs
func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *Config, creds *auth.Credentials) (*vfs.MountNamespace, error) {
fd := c.fds.remove()
- opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",")
+ opts := strings.Join(p9MountOptions(fd, conf.FileAccess, true /* vfs2 */), ",")
log.Infof("Mounting root over 9P, ioFD: %d", fd)
- mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", rootFsName, &vfs.GetFilesystemOptions{Data: opts})
+ mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", gofer.Name, &vfs.GetFilesystemOptions{Data: opts})
if err != nil {
return nil, fmt.Errorf("setting up mount namespace: %w", err)
}
@@ -204,27 +147,61 @@ func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *C
}
func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error {
- c.prepareMountsVFS2()
+ mounts, err := c.prepareMountsVFS2()
+ if err != nil {
+ return err
+ }
- for _, submount := range c.mounts {
+ 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 {
+ if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil {
return err
}
}
// TODO(gvisor.dev/issue/1487): implement mountTmp from fs.go.
- return c.checkDispenser()
+ return nil
+}
+
+type mountAndFD struct {
+ specs.Mount
+ fd int
}
-func (c *containerMounter) prepareMountsVFS2() {
+func (c *containerMounter) prepareMountsVFS2() ([]mountAndFD, error) {
+ // Associate bind mounts with their FDs before sorting since there is an
+ // undocumented assumption that FDs are dispensed in the order in which
+ // they are required by mounts.
+ var mounts []mountAndFD
+ for _, m := range c.mounts {
+ fd := -1
+ // Only bind mounts use host FDs; see
+ // containerMounter.getMountNameAndOptionsVFS2.
+ if m.Type == bind {
+ fd = c.fds.remove()
+ }
+ mounts = append(mounts, mountAndFD{
+ Mount: m,
+ fd: fd,
+ })
+ }
+ if err := c.checkDispenser(); err != nil {
+ return nil, err
+ }
+
// Sort the mounts so that we don't place children before parents.
- sort.Slice(c.mounts, func(i, j int) bool { return len(c.mounts[i].Destination) < len(c.mounts[j].Destination) })
+ sort.Slice(mounts, func(i, j int) bool {
+ return len(mounts[i].Destination) < len(mounts[j].Destination)
+ })
+
+ return mounts, nil
}
-// TODO(gvisor.dev/issue/1487): Implement submount options similar to the VFS1 version.
-func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *specs.Mount) error {
+// TODO(gvisor.dev/issue/1487): Implement submount options similar to the VFS1
+// version.
+func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *mountAndFD) error {
root := mns.Root()
defer root.DecRef()
target := &vfs.PathOperation{
@@ -233,7 +210,7 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config,
Path: fspath.Parse(submount.Destination),
}
- fsName, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, *submount)
+ fsName, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, submount)
if err != nil {
return fmt.Errorf("mountOptions failed: %w", err)
}
@@ -265,7 +242,7 @@ 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 specs.Mount) (string, []string, bool, error) {
+func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndFD) (string, []string, bool, error) {
var (
fsName string
opts []string
@@ -273,11 +250,11 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Moun
)
switch m.Type {
- case devpts, devtmpfs, proc, sysfs:
+ case devpts.Name, devtmpfs.Name, proc.Name, sys.Name:
fsName = m.Type
case nonefs:
- fsName = sysfs
- case tmpfs:
+ fsName = sys.Name
+ case tmpfs.Name:
fsName = m.Type
var err error
@@ -287,9 +264,8 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Moun
}
case bind:
- fd := c.fds.remove()
- fsName = "9p"
- opts = p9MountOptionsVFS2(fd, c.getMountAccessType(m))
+ fsName = gofer.Name
+ opts = p9MountOptions(m.fd, c.getMountAccessType(m.Mount), true /* vfs2 */)
// If configured, add overlay to all writable mounts.
useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly
@@ -299,21 +275,6 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Moun
return fsName, opts, useOverlay, nil
}
-// p9MountOptions creates a slice of options for a p9 mount.
-// TODO(gvisor.dev/issue/1624): Remove this version once privateunixsocket is
-// deleted, along with the rest of VFS1.
-func p9MountOptionsVFS2(fd int, fa FileAccessType) []string {
- opts := []string{
- "trans=fd",
- "rfdno=" + strconv.Itoa(fd),
- "wfdno=" + strconv.Itoa(fd),
- }
- if fa == FileAccessShared {
- opts = append(opts, "cache=remote_revalidating")
- }
- return opts
-}
-
func (c *containerMounter) makeSyntheticMount(ctx context.Context, currentPath string, root vfs.VirtualDentry, creds *auth.Credentials) error {
target := &vfs.PathOperation{
Root: root,
diff --git a/runsc/cgroup/BUILD b/runsc/cgroup/BUILD
index d4c7bdfbb..c087e1a3c 100644
--- a/runsc/cgroup/BUILD
+++ b/runsc/cgroup/BUILD
@@ -7,8 +7,8 @@ go_library(
srcs = ["cgroup.go"],
visibility = ["//:sandbox"],
deps = [
+ "//pkg/cleanup",
"//pkg/log",
- "//runsc/specutils",
"@com_github_cenkalti_backoff//:go_default_library",
"@com_github_opencontainers_runtime-spec//specs-go:go_default_library",
],
diff --git a/runsc/cgroup/cgroup.go b/runsc/cgroup/cgroup.go
index fa40ee509..ef01820ef 100644
--- a/runsc/cgroup/cgroup.go
+++ b/runsc/cgroup/cgroup.go
@@ -19,6 +19,7 @@ package cgroup
import (
"bufio"
"context"
+ "errors"
"fmt"
"io/ioutil"
"os"
@@ -30,29 +31,31 @@ import (
"github.com/cenkalti/backoff"
specs "github.com/opencontainers/runtime-spec/specs-go"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/log"
- "gvisor.dev/gvisor/runsc/specutils"
)
const (
cgroupRoot = "/sys/fs/cgroup"
)
-var controllers = map[string]controller{
- "blkio": &blockIO{},
- "cpu": &cpu{},
- "cpuset": &cpuSet{},
- "memory": &memory{},
- "net_cls": &networkClass{},
- "net_prio": &networkPrio{},
- "pids": &pids{},
+var controllers = map[string]config{
+ "blkio": config{ctrlr: &blockIO{}},
+ "cpu": config{ctrlr: &cpu{}},
+ "cpuset": config{ctrlr: &cpuSet{}},
+ "memory": config{ctrlr: &memory{}},
+ "net_cls": config{ctrlr: &networkClass{}},
+ "net_prio": config{ctrlr: &networkPrio{}},
+ "pids": config{ctrlr: &pids{}},
// These controllers either don't have anything in the OCI spec or is
// irrelevant for a sandbox.
- "devices": &noop{},
- "freezer": &noop{},
- "perf_event": &noop{},
- "systemd": &noop{},
+ "devices": config{ctrlr: &noop{}},
+ "freezer": config{ctrlr: &noop{}},
+ "hugetlb": config{ctrlr: &noop{}, optional: true},
+ "perf_event": config{ctrlr: &noop{}},
+ "rdma": config{ctrlr: &noop{}, optional: true},
+ "systemd": config{ctrlr: &noop{}},
}
func setOptionalValueInt(path, name string, val *int64) error {
@@ -196,8 +199,9 @@ func LoadPaths(pid string) (map[string]string, error) {
return paths, nil
}
-// Cgroup represents a group inside all controllers. For example: Name='/foo/bar'
-// maps to /sys/fs/cgroup/<controller>/foo/bar on all controllers.
+// Cgroup represents a group inside all controllers. For example:
+// Name='/foo/bar' maps to /sys/fs/cgroup/<controller>/foo/bar on
+// all controllers.
type Cgroup struct {
Name string `json:"name"`
Parents map[string]string `json:"parents"`
@@ -242,16 +246,20 @@ func (c *Cgroup) Install(res *specs.LinuxResources) error {
// The Cleanup object cleans up partially created cgroups when an error occurs.
// Errors occuring during cleanup itself are ignored.
- clean := specutils.MakeCleanup(func() { _ = c.Uninstall() })
+ clean := cleanup.Make(func() { _ = c.Uninstall() })
defer clean.Clean()
- for key, ctrl := range controllers {
+ for key, cfg := range controllers {
path := c.makePath(key)
if err := os.MkdirAll(path, 0755); err != nil {
+ if cfg.optional && errors.Is(err, syscall.EROFS) {
+ log.Infof("Skipping cgroup %q", key)
+ continue
+ }
return err
}
if res != nil {
- if err := ctrl.set(res, path); err != nil {
+ if err := cfg.ctrlr.set(res, path); err != nil {
return err
}
}
@@ -321,10 +329,13 @@ func (c *Cgroup) Join() (func(), error) {
}
// Now join the cgroups.
- for key := range controllers {
+ for key, cfg := range controllers {
path := c.makePath(key)
log.Debugf("Joining cgroup %q", path)
if err := setValue(path, "cgroup.procs", "0"); err != nil {
+ if cfg.optional && os.IsNotExist(err) {
+ continue
+ }
return undo, err
}
}
@@ -375,6 +386,11 @@ func (c *Cgroup) makePath(controllerName string) string {
return filepath.Join(cgroupRoot, controllerName, path)
}
+type config struct {
+ ctrlr controller
+ optional bool
+}
+
type controller interface {
set(*specs.LinuxResources, string) error
}
diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go
index 28f0d54b9..10448a759 100644
--- a/runsc/cmd/gofer.go
+++ b/runsc/cmd/gofer.go
@@ -168,7 +168,7 @@ 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,
+ ROMount: spec.Root.Readonly || conf.Overlay,
PanicOnWrite: g.panicOnWrite,
})
if err != nil {
@@ -181,7 +181,7 @@ 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),
+ ROMount: isReadonlyMount(m.Options) || conf.Overlay,
PanicOnWrite: g.panicOnWrite,
HostUDS: conf.FSGoferHostUDS,
}
diff --git a/runsc/cmd/help.go b/runsc/cmd/help.go
index c7d210140..cd85dabbb 100644
--- a/runsc/cmd/help.go
+++ b/runsc/cmd/help.go
@@ -65,16 +65,10 @@ func (h *Help) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}
switch f.NArg() {
case 0:
fmt.Fprintf(h.cdr.Output, "Usage: %s <flags> <subcommand> <subcommand args>\n\n", h.cdr.Name())
- fmt.Fprintf(h.cdr.Output, `runsc is a command line client for running applications packaged in the Open
-Container Initiative (OCI) format. Applications run by runsc are run in an
-isolated gVisor sandbox that emulates a Linux environment.
+ fmt.Fprintf(h.cdr.Output, `runsc is the gVisor container runtime.
-gVisor is a user-space kernel, written in Go, that implements a substantial
-portion of the Linux system call interface. It provides an additional layer
-of isolation between running applications and the host operating system.
-
-Functionality is provided by subcommands. For additonal help on individual
-subcommands use "%s %s <subcommand>".
+Functionality is provided by subcommands. For help with a specific subcommand,
+use "%s %s <subcommand>".
`, h.cdr.Name(), h.Name())
h.cdr.VisitGroups(func(g *subcommands.CommandGroup) {
diff --git a/runsc/cmd/spec.go b/runsc/cmd/spec.go
index 8e2b36e85..a2b0a4b14 100644
--- a/runsc/cmd/spec.go
+++ b/runsc/cmd/spec.go
@@ -16,6 +16,7 @@ package cmd
import (
"context"
+ "fmt"
"io/ioutil"
"os"
"path/filepath"
@@ -24,7 +25,8 @@ import (
"gvisor.dev/gvisor/runsc/flag"
)
-var specTemplate = []byte(`{
+func genSpec(cwd string) []byte {
+ var template = fmt.Sprintf(`{
"ociVersion": "1.0.0",
"process": {
"terminal": true,
@@ -39,7 +41,7 @@ var specTemplate = []byte(`{
"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
"TERM=xterm"
],
- "cwd": "/",
+ "cwd": "%s",
"capabilities": {
"bounding": [
"CAP_AUDIT_WRITE",
@@ -123,11 +125,15 @@ var specTemplate = []byte(`{
}
]
}
-}`)
+}`, cwd)
+
+ return []byte(template)
+}
// Spec implements subcommands.Command for the "spec" command.
type Spec struct {
bundle string
+ cwd string
}
// Name implements subcommands.Command.Name.
@@ -165,6 +171,8 @@ EXAMPLE:
// SetFlags implements subcommands.Command.SetFlags.
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")
}
// Execute implements subcommands.Command.Execute.
@@ -174,7 +182,9 @@ func (s *Spec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
Fatalf("file %q already exists", confPath)
}
- if err := ioutil.WriteFile(confPath, specTemplate, 0664); err != nil {
+ var spec = genSpec(s.cwd)
+
+ if err := ioutil.WriteFile(confPath, spec, 0664); err != nil {
Fatalf("writing to %q: %v", confPath, err)
}
diff --git a/runsc/container/BUILD b/runsc/container/BUILD
index 46154df60..9a856d65c 100644
--- a/runsc/container/BUILD
+++ b/runsc/container/BUILD
@@ -16,6 +16,7 @@ go_library(
],
deps = [
"//pkg/abi/linux",
+ "//pkg/cleanup",
"//pkg/log",
"//pkg/sentry/control",
"//pkg/sentry/sighandling",
@@ -53,6 +54,7 @@ go_test(
deps = [
"//pkg/abi/linux",
"//pkg/bits",
+ "//pkg/cleanup",
"//pkg/log",
"//pkg/sentry/control",
"//pkg/sentry/kernel",
diff --git a/runsc/container/container.go b/runsc/container/container.go
index 8539f252d..6d297d0df 100644
--- a/runsc/container/container.go
+++ b/runsc/container/container.go
@@ -31,6 +31,7 @@ import (
"github.com/cenkalti/backoff"
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/control"
"gvisor.dev/gvisor/pkg/sentry/sighandling"
@@ -293,7 +294,7 @@ func New(conf *boot.Config, args Args) (*Container, error) {
}
// The Cleanup object cleans up partially created containers when an error
// occurs. Any errors occurring during cleanup itself are ignored.
- cu := specutils.MakeCleanup(func() { _ = c.Destroy() })
+ cu := cleanup.Make(func() { _ = c.Destroy() })
defer cu.Clean()
// Lock the container metadata file to prevent concurrent creations of
@@ -402,7 +403,7 @@ func (c *Container) Start(conf *boot.Config) error {
if err := c.Saver.lock(); err != nil {
return err
}
- unlock := specutils.MakeCleanup(func() { c.Saver.unlock() })
+ unlock := cleanup.Make(func() { c.Saver.unlock() })
defer unlock.Clean()
if err := c.requireStatus("start", Created); err != nil {
@@ -506,7 +507,7 @@ func Run(conf *boot.Config, args Args) (syscall.WaitStatus, error) {
}
// Clean up partially created container if an error occurs.
// Any errors returned by Destroy() itself are ignored.
- cu := specutils.MakeCleanup(func() {
+ cu := cleanup.Make(func() {
c.Destroy()
})
defer cu.Clean()
diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go
index acf988aa0..d59a1d97e 100644
--- a/runsc/container/container_test.go
+++ b/runsc/container/container_test.go
@@ -1537,28 +1537,6 @@ func TestReadonlyMount(t *testing.T) {
}
}
-func TestBindMountByOption(t *testing.T) {
- for _, conf := range configs(t, overlay) {
- t.Logf("Running test with conf: %+v", conf)
-
- dir, err := ioutil.TempDir(testutil.TmpDir(), "bind-mount")
- spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file"))
- if err != nil {
- t.Fatalf("ioutil.TempDir() failed: %v", err)
- }
- spec.Mounts = append(spec.Mounts, specs.Mount{
- Destination: dir,
- Source: dir,
- Type: "none",
- Options: []string{"rw", "bind"},
- })
-
- if err := run(spec, conf); err != nil {
- t.Fatalf("error running sandbox: %v", err)
- }
- }
-}
-
// TestAbbreviatedIDs checks that runsc supports using abbreviated container
// IDs in place of full IDs.
func TestAbbreviatedIDs(t *testing.T) {
@@ -1760,7 +1738,7 @@ func TestUserLog(t *testing.T) {
if err != nil {
t.Fatalf("error opening user log file %q: %v", userLog, err)
}
- if want := "Unsupported syscall: sched_rr_get_interval"; !strings.Contains(string(out), want) {
+ if want := "Unsupported syscall sched_rr_get_interval("; !strings.Contains(string(out), want) {
t.Errorf("user log file doesn't contain %q, out: %s", want, string(out))
}
}
@@ -2126,43 +2104,6 @@ func TestNetRaw(t *testing.T) {
}
}
-// TestOverlayfsStaleRead most basic test that '--overlayfs-stale-read' works.
-func TestOverlayfsStaleRead(t *testing.T) {
- conf := testutil.TestConfig(t)
- conf.OverlayfsStaleRead = true
-
- in, err := ioutil.TempFile(testutil.TmpDir(), "stale-read.in")
- if err != nil {
- t.Fatalf("ioutil.TempFile() failed: %v", err)
- }
- defer in.Close()
- if _, err := in.WriteString("stale data"); err != nil {
- t.Fatalf("in.Write() failed: %v", err)
- }
-
- out, err := ioutil.TempFile(testutil.TmpDir(), "stale-read.out")
- if err != nil {
- t.Fatalf("ioutil.TempFile() failed: %v", err)
- }
- defer out.Close()
-
- const want = "foobar"
- cmd := fmt.Sprintf("cat %q >&2 && echo %q> %q && cp %q %q", in.Name(), want, in.Name(), in.Name(), out.Name())
- spec := testutil.NewSpecWithArgs("/bin/bash", "-c", cmd)
- if err := run(spec, conf); err != nil {
- t.Fatalf("Error running container: %v", err)
- }
-
- gotBytes, err := ioutil.ReadAll(out)
- if err != nil {
- t.Fatalf("out.Read() failed: %v", err)
- }
- got := strings.TrimSpace(string(gotBytes))
- if want != got {
- t.Errorf("Wrong content in out file, got: %q. want: %q", got, want)
- }
-}
-
// TestTTYField checks TTY field returned by container.Processes().
func TestTTYField(t *testing.T) {
stop := testutil.StartReaper()
diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go
index f6861b1dd..dc825abd9 100644
--- a/runsc/container/multi_container_test.go
+++ b/runsc/container/multi_container_test.go
@@ -27,6 +27,7 @@ import (
"time"
specs "github.com/opencontainers/runtime-spec/specs-go"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/sentry/control"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sync"
@@ -64,29 +65,16 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C
panic("conf.RootDir not set. Call testutil.SetupRootDir() to set.")
}
- var (
- containers []*Container
- cleanups []func()
- )
- cleanups = append(cleanups, func() {
- for _, c := range containers {
- c.Destroy()
- }
- })
- cleanupAll := func() {
- for _, c := range cleanups {
- c()
- }
- }
- localClean := specutils.MakeCleanup(cleanupAll)
- defer localClean.Clean()
+ cu := cleanup.Cleanup{}
+ defer cu.Clean()
+ var containers []*Container
for i, spec := range specs {
bundleDir, cleanup, err := testutil.SetupBundleDir(spec)
if err != nil {
return nil, nil, fmt.Errorf("error setting up container: %v", err)
}
- cleanups = append(cleanups, cleanup)
+ cu.Add(cleanup)
args := Args{
ID: ids[i],
@@ -97,6 +85,7 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C
if err != nil {
return nil, nil, fmt.Errorf("error creating container: %v", err)
}
+ cu.Add(func() { cont.Destroy() })
containers = append(containers, cont)
if err := cont.Start(conf); err != nil {
@@ -104,8 +93,7 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C
}
}
- localClean.Release()
- return containers, cleanupAll, nil
+ return containers, cu.Release(), nil
}
type execDesc struct {
@@ -1394,7 +1382,7 @@ func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) {
Destination: "/mydir/test",
Source: "/some/dir",
Type: "tmpfs",
- Options: []string{"rw", "relatime"},
+ Options: []string{"rw", "rbind", "relatime"},
}
podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0)
diff --git a/runsc/fsgofer/BUILD b/runsc/fsgofer/BUILD
index 64a406ae2..1036b0630 100644
--- a/runsc/fsgofer/BUILD
+++ b/runsc/fsgofer/BUILD
@@ -13,12 +13,12 @@ go_library(
visibility = ["//runsc:__subpackages__"],
deps = [
"//pkg/abi/linux",
+ "//pkg/cleanup",
"//pkg/fd",
"//pkg/log",
"//pkg/p9",
"//pkg/sync",
"//pkg/syserr",
- "//runsc/specutils",
"@org_golang_x_sys//unix:go_default_library",
],
)
diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go
index 1942f50d7..edc239013 100644
--- a/runsc/fsgofer/fsgofer.go
+++ b/runsc/fsgofer/fsgofer.go
@@ -33,11 +33,11 @@ import (
"golang.org/x/sys/unix"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/fd"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/sync"
- "gvisor.dev/gvisor/runsc/specutils"
)
const (
@@ -439,7 +439,7 @@ func (l *localFile) Create(name string, mode p9.OpenFlags, perm p9.FileMode, uid
if err != nil {
return nil, nil, p9.QID{}, 0, extractErrno(err)
}
- cu := specutils.MakeCleanup(func() {
+ cu := cleanup.Make(func() {
child.Close()
// Best effort attempt to remove the file in case of failure.
if err := syscall.Unlinkat(l.file.FD(), name); err != nil {
@@ -480,7 +480,7 @@ func (l *localFile) Mkdir(name string, perm p9.FileMode, uid p9.UID, gid p9.GID)
if err := syscall.Mkdirat(l.file.FD(), name, uint32(perm.Permissions())); err != nil {
return p9.QID{}, extractErrno(err)
}
- cu := specutils.MakeCleanup(func() {
+ cu := cleanup.Make(func() {
// Best effort attempt to remove the dir in case of failure.
if err := unix.Unlinkat(l.file.FD(), name, unix.AT_REMOVEDIR); err != nil {
log.Warningf("error unlinking dir %q after failure: %v", path.Join(l.hostPath, name), err)
@@ -864,7 +864,7 @@ func (l *localFile) Symlink(target, newName string, uid p9.UID, gid p9.GID) (p9.
if err := unix.Symlinkat(target, l.file.FD(), newName); err != nil {
return p9.QID{}, extractErrno(err)
}
- cu := specutils.MakeCleanup(func() {
+ cu := cleanup.Make(func() {
// Best effort attempt to remove the symlink in case of failure.
if err := syscall.Unlinkat(l.file.FD(), newName); err != nil {
log.Warningf("error unlinking file %q after failure: %v", path.Join(l.hostPath, newName), err)
diff --git a/runsc/main.go b/runsc/main.go
index 0625a06e0..920ed84a5 100644
--- a/runsc/main.go
+++ b/runsc/main.go
@@ -76,7 +76,7 @@ var (
fileAccess = flag.String("file-access", "exclusive", "specifies which filesystem to use for the root mount: exclusive (default), shared. Volume mounts are always shared.")
fsGoferHostUDS = flag.Bool("fsgofer-host-uds", false, "allow the gofer to mount Unix Domain Sockets.")
overlay = flag.Bool("overlay", false, "wrap filesystem mounts with writable overlay. All modifications are stored in memory inside the sandbox.")
- overlayfsStaleRead = flag.Bool("overlayfs-stale-read", false, "reopen cached FDs after a file is opened for write to workaround overlayfs limitation on kernels before 4.19.")
+ overlayfsStaleRead = flag.Bool("overlayfs-stale-read", true, "assume root mount is an overlay filesystem")
watchdogAction = flag.String("watchdog-action", "log", "sets what action the watchdog takes when triggered: log (default), panic.")
panicSignal = flag.Int("panic-signal", -1, "register signal handling that panics. Usually set to SIGUSR2(12) to troubleshoot hangs. -1 disables it.")
profile = flag.Bool("profile", false, "prepares the sandbox to use Golang profiler. Note that enabling profiler loosens the seccomp protection added to the sandbox (DO NOT USE IN PRODUCTION).")
diff --git a/runsc/sandbox/BUILD b/runsc/sandbox/BUILD
index c95d50294..035dcd3e3 100644
--- a/runsc/sandbox/BUILD
+++ b/runsc/sandbox/BUILD
@@ -13,6 +13,7 @@ go_library(
"//runsc:__subpackages__",
],
deps = [
+ "//pkg/cleanup",
"//pkg/control/client",
"//pkg/control/server",
"//pkg/log",
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index e4ec16e2f..6e1a2af25 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -30,6 +30,7 @@ import (
"github.com/cenkalti/backoff"
specs "github.com/opencontainers/runtime-spec/specs-go"
"github.com/syndtr/gocapability/capability"
+ "gvisor.dev/gvisor/pkg/cleanup"
"gvisor.dev/gvisor/pkg/control/client"
"gvisor.dev/gvisor/pkg/control/server"
"gvisor.dev/gvisor/pkg/log"
@@ -119,7 +120,7 @@ func New(conf *boot.Config, args *Args) (*Sandbox, error) {
s := &Sandbox{ID: args.ID, Cgroup: args.Cgroup}
// The Cleanup object cleans up partially created sandboxes when an error
// occurs. Any errors occurring during cleanup itself are ignored.
- c := specutils.MakeCleanup(func() {
+ c := cleanup.Make(func() {
err := s.destroy()
log.Warningf("error destroying sandbox: %v", err)
})
diff --git a/runsc/specutils/specutils.go b/runsc/specutils/specutils.go
index 202518b58..f1fa573c5 100644
--- a/runsc/specutils/specutils.go
+++ b/runsc/specutils/specutils.go
@@ -311,19 +311,7 @@ func capsFromNames(names []string, skipSet map[linux.Capability]struct{}) (auth.
// Is9PMount returns true if the given mount can be mounted as an external gofer.
func Is9PMount(m specs.Mount) bool {
- var isBind bool
- switch m.Type {
- case "bind":
- isBind = true
- default:
- for _, opt := range m.Options {
- if opt == "bind" || opt == "rbind" {
- isBind = true
- break
- }
- }
- }
- return isBind && m.Source != "" && IsSupportedDevMount(m)
+ return m.Type == "bind" && m.Source != "" && IsSupportedDevMount(m)
}
// IsSupportedDevMount returns true if the mount is a supported /dev mount.
@@ -456,36 +444,6 @@ func ContainsStr(strs []string, str string) bool {
return false
}
-// Cleanup allows defers to be aborted when cleanup needs to happen
-// conditionally. Usage:
-// c := MakeCleanup(func() { f.Close() })
-// defer c.Clean() // any failure before release is called will close the file.
-// ...
-// c.Release() // on success, aborts closing the file and return it.
-// return f
-type Cleanup struct {
- clean func()
-}
-
-// MakeCleanup creates a new Cleanup object.
-func MakeCleanup(f func()) Cleanup {
- return Cleanup{clean: f}
-}
-
-// Clean calls the cleanup function.
-func (c *Cleanup) Clean() {
- if c.clean != nil {
- c.clean()
- c.clean = nil
- }
-}
-
-// Release releases the cleanup from its duties, i.e. cleanup function is not
-// called after this point.
-func (c *Cleanup) Release() {
- c.clean = nil
-}
-
// RetryEintr retries the function until an error different than EINTR is
// returned.
func RetryEintr(f func() (uintptr, uintptr, error)) (uintptr, uintptr, error) {
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-15 22:27:23 +0000