summaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
Diffstat
-rw-r--r--pkg/sentry/control/proc.go2
-rw-r--r--pkg/sentry/fdimport/BUILD1
-rw-r--r--pkg/sentry/fdimport/fdimport.go22
-rw-r--r--pkg/sentry/fsimpl/host/host.go163
-rw-r--r--runsc/boot/loader.go16
-rw-r--r--runsc/cmd/exec.go6
-rw-r--r--runsc/container/console_test.go2
-rw-r--r--runsc/container/container.go9
-rw-r--r--runsc/container/container_test.go42
-rw-r--r--runsc/container/multi_container_test.go42
-rw-r--r--runsc/container/shared_volume_test.go28
-rw-r--r--runsc/sandbox/sandbox.go135
-rw-r--r--test/e2e/integration_test.go113
13 files changed, 414 insertions, 167 deletions
diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go
index b7ee5425d..6352ea71a 100644
--- a/pkg/sentry/control/proc.go
+++ b/pkg/sentry/control/proc.go
@@ -223,7 +223,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
_ = fd.Close()
}
}()
- ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, args.StdioIsPty, fds)
+ ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, args.StdioIsPty, args.KUID, args.KGID, fds)
if err != nil {
return nil, 0, nil, nil, err
}
diff --git a/pkg/sentry/fdimport/BUILD b/pkg/sentry/fdimport/BUILD
index 6b4f8b0ed..563e96e0d 100644
--- a/pkg/sentry/fdimport/BUILD
+++ b/pkg/sentry/fdimport/BUILD
@@ -15,6 +15,7 @@ go_library(
"//pkg/sentry/fs/host",
"//pkg/sentry/fsimpl/host",
"//pkg/sentry/kernel",
+ "//pkg/sentry/kernel/auth",
"//pkg/sentry/vfs",
],
)
diff --git a/pkg/sentry/fdimport/fdimport.go b/pkg/sentry/fdimport/fdimport.go
index badd5b073..f2b9630eb 100644
--- a/pkg/sentry/fdimport/fdimport.go
+++ b/pkg/sentry/fdimport/fdimport.go
@@ -24,6 +24,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/fs/host"
hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
"gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/vfs"
)
@@ -31,9 +32,9 @@ import (
// sets up TTY for the first 3 FDs in the slice representing stdin, stdout,
// stderr. Used FDs are either closed or released. It's safe for the caller to
// close any remaining files upon return.
-func Import(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []*fd.FD) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
+func Import(ctx context.Context, fdTable *kernel.FDTable, console bool, uid auth.KUID, gid auth.KGID, fds []*fd.FD) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
if kernel.VFS2Enabled {
- ttyFile, err := importVFS2(ctx, fdTable, console, fds)
+ ttyFile, err := importVFS2(ctx, fdTable, console, uid, gid, fds)
return nil, ttyFile, err
}
ttyFile, err := importFS(ctx, fdTable, console, fds)
@@ -89,7 +90,7 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []
return ttyFile.FileOperations.(*host.TTYFileOperations), nil
}
-func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdioFDs []*fd.FD) (*hostvfs2.TTYFileDescription, error) {
+func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, uid auth.KUID, gid auth.KGID, stdioFDs []*fd.FD) (*hostvfs2.TTYFileDescription, error) {
k := kernel.KernelFromContext(ctx)
if k == nil {
return nil, fmt.Errorf("cannot find kernel from context")
@@ -103,7 +104,13 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi
// Import the file as a host TTY file.
if ttyFile == nil {
var err error
- appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD.FD(), true /* isTTY */)
+ appFile, err = hostvfs2.NewFD(ctx, k.HostMount(), hostFD.FD(), &hostvfs2.NewFDOptions{
+ Savable: true,
+ IsTTY: true,
+ VirtualOwner: true,
+ UID: uid,
+ GID: gid,
+ })
if err != nil {
return nil, err
}
@@ -121,7 +128,12 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi
}
} else {
var err error
- appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD.FD(), false /* isTTY */)
+ appFile, err = hostvfs2.NewFD(ctx, k.HostMount(), hostFD.FD(), &hostvfs2.NewFDOptions{
+ Savable: true,
+ VirtualOwner: true,
+ UID: uid,
+ GID: gid,
+ })
if err != nil {
return nil, err
}
diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go
index 9998bd0c3..89aa7b3d9 100644
--- a/pkg/sentry/fsimpl/host/host.go
+++ b/pkg/sentry/fsimpl/host/host.go
@@ -42,6 +42,36 @@ import (
"gvisor.dev/gvisor/pkg/waiter"
)
+// These are the modes that are stored with virtualOwner.
+const virtualOwnerModes = linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID
+
+// +stateify savable
+type virtualOwner struct {
+ // This field is initialized at creation time and is immutable.
+ enabled bool
+
+ // mu protects the fields below and they can be accessed using atomic memory
+ // operations.
+ mu sync.Mutex `state:"nosave"`
+ uid uint32
+ gid uint32
+ // mode is also stored, otherwise setting the host file to `0000` could remove
+ // access to the file.
+ mode uint32
+}
+
+func (v *virtualOwner) atomicUID() uint32 {
+ return atomic.LoadUint32(&v.uid)
+}
+
+func (v *virtualOwner) atomicGID() uint32 {
+ return atomic.LoadUint32(&v.gid)
+}
+
+func (v *virtualOwner) atomicMode() uint32 {
+ return atomic.LoadUint32(&v.mode)
+}
+
// inode implements kernfs.Inode.
//
// +stateify savable
@@ -98,6 +128,11 @@ type inode struct {
// Event queue for blocking operations.
queue waiter.Queue
+ // virtualOwner caches ownership and permission information to override the
+ // underlying file owner and permission. This is used to allow the unstrusted
+ // application to change these fields without affecting the host.
+ virtualOwner virtualOwner
+
// If haveBuf is non-zero, hostFD represents a pipe, and buf contains data
// read from the pipe from previous calls to inode.beforeSave(). haveBuf
// and buf are protected by bufMu. haveBuf is accessed using atomic memory
@@ -147,7 +182,7 @@ func newInode(ctx context.Context, fs *filesystem, hostFD int, savable bool, fil
type NewFDOptions struct {
// If Savable is true, the host file descriptor may be saved/restored by
// numeric value; the sandbox API requires a corresponding host FD with the
- // same numeric value to be provieded at time of restore.
+ // same numeric value to be provided at time of restore.
Savable bool
// If IsTTY is true, the file descriptor is a TTY.
@@ -157,6 +192,12 @@ type NewFDOptions struct {
// the new file description will inherit flags from hostFD.
HaveFlags bool
Flags uint32
+
+ // VirtualOwner allow the host file to have owner and permissions different
+ // than the underlying host file.
+ VirtualOwner bool
+ UID auth.KUID
+ GID auth.KGID
}
// NewFD returns a vfs.FileDescription representing the given host file
@@ -168,8 +209,8 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions)
}
// Retrieve metadata.
- var s unix.Stat_t
- if err := unix.Fstat(hostFD, &s); err != nil {
+ var stat unix.Stat_t
+ if err := unix.Fstat(hostFD, &stat); err != nil {
return nil, err
}
@@ -183,11 +224,19 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions)
flags = uint32(flagsInt)
}
- d := &kernfs.Dentry{}
- i, err := newInode(ctx, fs, hostFD, opts.Savable, linux.FileMode(s.Mode).FileType(), opts.IsTTY)
+ fileType := linux.FileMode(stat.Mode).FileType()
+ i, err := newInode(ctx, fs, hostFD, opts.Savable, fileType, opts.IsTTY)
if err != nil {
return nil, err
}
+ if opts.VirtualOwner {
+ i.virtualOwner.enabled = true
+ i.virtualOwner.uid = uint32(opts.UID)
+ i.virtualOwner.gid = uint32(opts.GID)
+ i.virtualOwner.mode = stat.Mode
+ }
+
+ d := &kernfs.Dentry{}
d.Init(&fs.Filesystem, i)
// i.open will take a reference on d.
@@ -196,15 +245,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions)
// For simplicity, fileDescription.offset is set to 0. Technically, we
// should only set to 0 on files that are not seekable (sockets, pipes,
// etc.), and use the offset from the host fd otherwise when importing.
- return i.open(ctx, d, mnt, flags)
-}
-
-// ImportFD sets up and returns a vfs.FileDescription from a donated fd.
-func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) {
- return NewFD(ctx, mnt, hostFD, &NewFDOptions{
- Savable: true,
- IsTTY: isTTY,
- })
+ return i.open(ctx, d, mnt, fileType, flags)
}
// filesystemType implements vfs.FilesystemType.
@@ -270,7 +311,7 @@ func (fs *filesystem) MountOptions() string {
// CheckPermissions implements kernfs.Inode.CheckPermissions.
func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error {
var s unix.Stat_t
- if err := unix.Fstat(i.hostFD, &s); err != nil {
+ if err := i.stat(&s); err != nil {
return err
}
return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid))
@@ -279,7 +320,7 @@ func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, a
// Mode implements kernfs.Inode.Mode.
func (i *inode) Mode() linux.FileMode {
var s unix.Stat_t
- if err := unix.Fstat(i.hostFD, &s); err != nil {
+ if err := i.stat(&s); err != nil {
// Retrieving the mode from the host fd using fstat(2) should not fail.
// If the syscall does not succeed, something is fundamentally wrong.
panic(fmt.Sprintf("failed to retrieve mode from host fd %d: %v", i.hostFD, err))
@@ -306,7 +347,7 @@ func (i *inode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOp
// Fallback to fstat(2), if statx(2) is not supported on the host.
//
// TODO(b/151263641): Remove fallback.
- return i.fstat(fs)
+ return i.statxFromStat(fs)
}
if err != nil {
return linux.Statx{}, err
@@ -330,19 +371,35 @@ func (i *inode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOp
// device numbers.
ls.Mask |= s.Mask & linux.STATX_ALL
if s.Mask&linux.STATX_TYPE != 0 {
- ls.Mode |= s.Mode & linux.S_IFMT
+ if i.virtualOwner.enabled {
+ ls.Mode |= uint16(i.virtualOwner.atomicMode()) & linux.S_IFMT
+ } else {
+ ls.Mode |= s.Mode & linux.S_IFMT
+ }
}
if s.Mask&linux.STATX_MODE != 0 {
- ls.Mode |= s.Mode &^ linux.S_IFMT
+ if i.virtualOwner.enabled {
+ ls.Mode |= uint16(i.virtualOwner.atomicMode()) &^ linux.S_IFMT
+ } else {
+ ls.Mode |= s.Mode &^ linux.S_IFMT
+ }
}
if s.Mask&linux.STATX_NLINK != 0 {
ls.Nlink = s.Nlink
}
if s.Mask&linux.STATX_UID != 0 {
- ls.UID = s.Uid
+ if i.virtualOwner.enabled {
+ ls.UID = i.virtualOwner.atomicUID()
+ } else {
+ ls.UID = s.Uid
+ }
}
if s.Mask&linux.STATX_GID != 0 {
- ls.GID = s.Gid
+ if i.virtualOwner.enabled {
+ ls.GID = i.virtualOwner.atomicGID()
+ } else {
+ ls.GID = s.Gid
+ }
}
if s.Mask&linux.STATX_ATIME != 0 {
ls.Atime = unixToLinuxStatxTimestamp(s.Atime)
@@ -366,7 +423,7 @@ func (i *inode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOp
return ls, nil
}
-// fstat is a best-effort fallback for inode.Stat() if the host does not
+// statxFromStat is a best-effort fallback for inode.Stat() if the host does not
// support statx(2).
//
// We ignore the mask and sync flags in opts and simply supply
@@ -374,9 +431,9 @@ func (i *inode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOp
// of a mask or sync flags. fstat(2) does not provide any metadata
// equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so
// those fields remain empty.
-func (i *inode) fstat(fs *filesystem) (linux.Statx, error) {
+func (i *inode) statxFromStat(fs *filesystem) (linux.Statx, error) {
var s unix.Stat_t
- if err := unix.Fstat(i.hostFD, &s); err != nil {
+ if err := i.stat(&s); err != nil {
return linux.Statx{}, err
}
@@ -400,7 +457,21 @@ func (i *inode) fstat(fs *filesystem) (linux.Statx, error) {
}, nil
}
+func (i *inode) stat(stat *unix.Stat_t) error {
+ if err := unix.Fstat(i.hostFD, stat); err != nil {
+ return err
+ }
+ if i.virtualOwner.enabled {
+ stat.Uid = i.virtualOwner.atomicUID()
+ stat.Gid = i.virtualOwner.atomicGID()
+ stat.Mode = i.virtualOwner.atomicMode()
+ }
+ return nil
+}
+
// SetStat implements kernfs.Inode.SetStat.
+//
+// +checklocksignore
func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error {
s := &opts.Stat
@@ -408,11 +479,22 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre
if m == 0 {
return nil
}
- if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 {
+ supportedModes := uint32(linux.STATX_MODE | linux.STATX_SIZE | linux.STATX_ATIME | linux.STATX_MTIME)
+ if i.virtualOwner.enabled {
+ if m&virtualOwnerModes != 0 {
+ // Take lock if any of the virtual owner fields will be updated.
+ i.virtualOwner.mu.Lock()
+ defer i.virtualOwner.mu.Unlock()
+ }
+
+ supportedModes |= virtualOwnerModes
+ }
+ if m&^supportedModes != 0 {
return linuxerr.EPERM
}
+
var hostStat unix.Stat_t
- if err := unix.Fstat(i.hostFD, &hostStat); err != nil {
+ if err := i.stat(&hostStat); err != nil {
return err
}
if err := vfs.CheckSetStat(ctx, creds, &opts, linux.FileMode(hostStat.Mode), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil {
@@ -420,8 +502,12 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre
}
if m&linux.STATX_MODE != 0 {
- if err := unix.Fchmod(i.hostFD, uint32(s.Mode)); err != nil {
- return err
+ if i.virtualOwner.enabled {
+ i.virtualOwner.mode = uint32(opts.Stat.Mode)
+ } else {
+ if err := unix.Fchmod(i.hostFD, uint32(s.Mode)); err != nil {
+ return err
+ }
}
}
if m&linux.STATX_SIZE != 0 {
@@ -449,6 +535,14 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre
return err
}
}
+ if i.virtualOwner.enabled {
+ if m&linux.STATX_UID != 0 {
+ i.virtualOwner.uid = opts.Stat.UID
+ }
+ if m&linux.STATX_GID != 0 {
+ i.virtualOwner.gid = opts.Stat.GID
+ }
+ }
return nil
}
@@ -473,16 +567,15 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentr
if i.Mode().FileType() == linux.S_IFSOCK {
return nil, linuxerr.ENXIO
}
- return i.open(ctx, d, rp.Mount(), opts.Flags)
-}
-
-func (i *inode) open(ctx context.Context, d *kernfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) {
- var s unix.Stat_t
- if err := unix.Fstat(i.hostFD, &s); err != nil {
+ var stat unix.Stat_t
+ if err := i.stat(&stat); err != nil {
return nil, err
}
- fileType := s.Mode & linux.FileTypeMask
+ fileType := linux.FileMode(stat.Mode).FileType()
+ return i.open(ctx, d, rp.Mount(), fileType, opts.Flags)
+}
+func (i *inode) open(ctx context.Context, d *kernfs.Dentry, mnt *vfs.Mount, fileType linux.FileMode, flags uint32) (*vfs.FileDescription, error) {
// Constrain flags to a subset we can handle.
//
// TODO(gvisor.dev/issue/2601): Support O_NONBLOCK by adding RWF_NOWAIT to pread/pwrite calls.
diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go
index 165fb2ebb..5bfb15971 100644
--- a/runsc/boot/loader.go
+++ b/runsc/boot/loader.go
@@ -715,7 +715,7 @@ func (l *Loader) startContainer(spec *specs.Spec, conf *config.Config, cid strin
return fmt.Errorf("using TTY, stdios not expected: %d", l)
}
if ep.hostTTY == nil {
- return fmt.Errorf("terminal enabled but no TTY provided (--console-socket possibly passed)")
+ return fmt.Errorf("terminal enabled but no TTY provided. Did you set --console-socket on create?")
}
info.stdioFDs = []*fd.FD{ep.hostTTY, ep.hostTTY, ep.hostTTY}
ep.hostTTY = nil
@@ -734,7 +734,7 @@ func (l *Loader) startContainer(spec *specs.Spec, conf *config.Config, cid strin
func (l *Loader) createContainerProcess(root bool, cid string, info *containerInfo) (*kernel.ThreadGroup, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
// Create the FD map, which will set stdin, stdout, and stderr.
ctx := info.procArgs.NewContext(l.k)
- fdTable, ttyFile, ttyFileVFS2, err := createFDTable(ctx, info.spec.Process.Terminal, info.stdioFDs)
+ fdTable, ttyFile, ttyFileVFS2, err := createFDTable(ctx, info.spec.Process.Terminal, info.stdioFDs, info.spec.Process.User)
if err != nil {
return nil, nil, nil, fmt.Errorf("importing fds: %w", err)
}
@@ -980,7 +980,7 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
tty: ttyFile,
ttyVFS2: ttyFileVFS2,
}
- log.Debugf("updated processes: %s", l.processes)
+ log.Debugf("updated processes: %v", l.processes)
return tgid, nil
}
@@ -1024,7 +1024,7 @@ func (l *Loader) waitPID(tgid kernel.ThreadID, cid string, waitStatus *uint32) e
l.mu.Lock()
delete(l.processes, eid)
- log.Debugf("updated processes (removal): %s", l.processes)
+ log.Debugf("updated processes (removal): %v", l.processes)
l.mu.Unlock()
return nil
}
@@ -1092,7 +1092,7 @@ func newRootNetworkNamespace(conf *config.Config, clock tcpip.Clock, uniqueID st
return inet.NewRootNamespace(s, creator), nil
default:
- panic(fmt.Sprintf("invalid network configuration: %d", conf.Network))
+ panic(fmt.Sprintf("invalid network configuration: %v", conf.Network))
}
}
@@ -1212,7 +1212,7 @@ func (l *Loader) signal(cid string, pid, signo int32, mode SignalDeliveryMode) e
return nil
default:
- panic(fmt.Sprintf("unknown signal delivery mode %s", mode))
+ panic(fmt.Sprintf("unknown signal delivery mode %v", mode))
}
}
@@ -1337,14 +1337,14 @@ func (l *Loader) ttyFromIDLocked(key execID) (*host.TTYFileOperations, *hostvfs2
return ep.tty, ep.ttyVFS2, nil
}
-func createFDTable(ctx context.Context, console bool, stdioFDs []*fd.FD) (*kernel.FDTable, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
+func createFDTable(ctx context.Context, console bool, stdioFDs []*fd.FD, user specs.User) (*kernel.FDTable, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
if len(stdioFDs) != 3 {
return nil, nil, nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs))
}
k := kernel.KernelFromContext(ctx)
fdTable := k.NewFDTable()
- ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, console, stdioFDs)
+ ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, console, auth.KUID(user.UID), auth.KGID(user.GID), stdioFDs)
if err != nil {
fdTable.DecRef(ctx)
return nil, nil, nil, err
diff --git a/runsc/cmd/exec.go b/runsc/cmd/exec.go
index 242d474b8..2139fdf53 100644
--- a/runsc/cmd/exec.go
+++ b/runsc/cmd/exec.go
@@ -146,12 +146,12 @@ func (ex *Exec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
if ex.detach {
return ex.execChildAndWait(waitStatus)
}
- return ex.exec(c, e, waitStatus)
+ return ex.exec(conf, c, e, waitStatus)
}
-func (ex *Exec) exec(c *container.Container, e *control.ExecArgs, waitStatus *unix.WaitStatus) subcommands.ExitStatus {
+func (ex *Exec) exec(conf *config.Config, c *container.Container, e *control.ExecArgs, waitStatus *unix.WaitStatus) subcommands.ExitStatus {
// Start the new process and get its pid.
- pid, err := c.Execute(e)
+ pid, err := c.Execute(conf, e)
if err != nil {
return Errorf("executing processes for container: %v", err)
}
diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go
index 8c65b9cd4..9d36086c3 100644
--- a/runsc/container/console_test.go
+++ b/runsc/container/console_test.go
@@ -288,7 +288,7 @@ func TestJobControlSignalExec(t *testing.T) {
StdioIsPty: true,
}
- pid, err := c.Execute(execArgs)
+ pid, err := c.Execute(conf, execArgs)
if err != nil {
t.Fatalf("error executing: %v", err)
}
diff --git a/runsc/container/container.go b/runsc/container/container.go
index b789bc7da..213fbc771 100644
--- a/runsc/container/container.go
+++ b/runsc/container/container.go
@@ -310,7 +310,7 @@ func New(conf *config.Config, args Args) (*Container, error) {
defer tty.Close()
}
- if err := c.Sandbox.CreateContainer(c.ID, tty); err != nil {
+ if err := c.Sandbox.CreateContainer(conf, c.ID, tty); err != nil {
return nil, err
}
}
@@ -480,13 +480,13 @@ func Run(conf *config.Config, args Args) (unix.WaitStatus, error) {
// Execute runs the specified command in the container. It returns the PID of
// the newly created process.
-func (c *Container) Execute(args *control.ExecArgs) (int32, error) {
+func (c *Container) Execute(conf *config.Config, args *control.ExecArgs) (int32, error) {
log.Debugf("Execute in container, cid: %s, args: %+v", c.ID, args)
if err := c.requireStatus("execute in", Created, Running); err != nil {
return 0, err
}
args.ContainerID = c.ID
- return c.Sandbox.Execute(args)
+ return c.Sandbox.Execute(conf, args)
}
// Event returns events for the container.
@@ -910,6 +910,9 @@ func (c *Container) createGoferProcess(spec *specs.Spec, conf *config.Config, bu
binPath := specutils.ExePath
cmd := exec.Command(binPath, args...)
cmd.ExtraFiles = goferEnds
+
+ // Set Args[0] to make easier to spot the gofer process. Otherwise it's
+ // shown as `exe`.
cmd.Args[0] = "runsc-gofer"
if attached {
diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go
index 7360eae35..5fb4a3672 100644
--- a/runsc/container/container_test.go
+++ b/runsc/container/container_test.go
@@ -60,15 +60,15 @@ func TestMain(m *testing.M) {
os.Exit(m.Run())
}
-func execute(cont *Container, name string, arg ...string) (unix.WaitStatus, error) {
+func execute(conf *config.Config, cont *Container, name string, arg ...string) (unix.WaitStatus, error) {
args := &control.ExecArgs{
Filename: name,
Argv: append([]string{name}, arg...),
}
- return cont.executeSync(args)
+ return cont.executeSync(conf, args)
}
-func executeCombinedOutput(cont *Container, name string, arg ...string) ([]byte, error) {
+func executeCombinedOutput(conf *config.Config, cont *Container, name string, arg ...string) ([]byte, error) {
r, w, err := os.Pipe()
if err != nil {
return nil, err
@@ -80,7 +80,7 @@ func executeCombinedOutput(cont *Container, name string, arg ...string) ([]byte,
Argv: append([]string{name}, arg...),
FilePayload: urpc.FilePayload{Files: []*os.File{os.Stdin, w, w}},
}
- ws, err := cont.executeSync(args)
+ ws, err := cont.executeSync(conf, args)
w.Close()
if err != nil {
return nil, err
@@ -94,8 +94,8 @@ func executeCombinedOutput(cont *Container, name string, arg ...string) ([]byte,
}
// executeSync synchronously executes a new process.
-func (c *Container) executeSync(args *control.ExecArgs) (unix.WaitStatus, error) {
- pid, err := c.Execute(args)
+func (c *Container) executeSync(conf *config.Config, args *control.ExecArgs) (unix.WaitStatus, error) {
+ pid, err := c.Execute(conf, args)
if err != nil {
return 0, fmt.Errorf("error executing: %v", err)
}
@@ -172,8 +172,8 @@ func blockUntilWaitable(pid int) error {
}
// execPS executes `ps` inside the container and return the processes.
-func execPS(c *Container) ([]*control.Process, error) {
- out, err := executeCombinedOutput(c, "/bin/ps", "-e")
+func execPS(conf *config.Config, c *Container) ([]*control.Process, error) {
+ out, err := executeCombinedOutput(conf, c, "/bin/ps", "-e")
if err != nil {
return nil, err
}
@@ -864,7 +864,7 @@ func TestExec(t *testing.T) {
} {
t.Run(tc.name, func(t *testing.T) {
// t.Parallel()
- if ws, err := cont.executeSync(&tc.args); err != nil {
+ if ws, err := cont.executeSync(conf, &tc.args); err != nil {
t.Fatalf("executeAsync(%+v): %v", tc.args, err)
} else if ws != 0 {
t.Fatalf("executeAsync(%+v) failed with exit: %v", tc.args, ws)
@@ -882,7 +882,7 @@ func TestExec(t *testing.T) {
}
defer unix.Close(fds[0])
- _, err = cont.executeSync(&control.ExecArgs{
+ _, err = cont.executeSync(conf, &control.ExecArgs{
Argv: []string{"/nonexist"},
FilePayload: urpc.FilePayload{
Files: []*os.File{os.NewFile(uintptr(fds[1]), "sock")},
@@ -937,7 +937,7 @@ func TestExecProcList(t *testing.T) {
// start running exec (which blocks).
ch := make(chan error)
go func() {
- exitStatus, err := cont.executeSync(execArgs)
+ exitStatus, err := cont.executeSync(conf, execArgs)
if err != nil {
ch <- err
} else if exitStatus != 0 {
@@ -1544,7 +1544,7 @@ func TestCapabilities(t *testing.T) {
}
// "exe" should fail because we don't have the necessary permissions.
- if _, err := cont.executeSync(execArgs); err == nil {
+ if _, err := cont.executeSync(conf, execArgs); err == nil {
t.Fatalf("container executed without error, but an error was expected")
}
@@ -1553,7 +1553,7 @@ func TestCapabilities(t *testing.T) {
EffectiveCaps: auth.CapabilitySetOf(linux.CAP_DAC_OVERRIDE),
}
// "exe" should not fail this time.
- if _, err := cont.executeSync(execArgs); err != nil {
+ if _, err := cont.executeSync(conf, execArgs); err != nil {
t.Fatalf("container failed to exec %v: %v", args, err)
}
})
@@ -1664,7 +1664,7 @@ func TestReadonlyRoot(t *testing.T) {
}
// Read mounts to check that root is readonly.
- out, err := executeCombinedOutput(c, "/bin/sh", "-c", "mount | grep ' / ' | grep -o -e '(.*)'")
+ out, err := executeCombinedOutput(conf, c, "/bin/sh", "-c", "mount | grep ' / ' | grep -o -e '(.*)'")
if err != nil {
t.Fatalf("exec failed: %v", err)
}
@@ -1674,7 +1674,7 @@ func TestReadonlyRoot(t *testing.T) {
}
// Check that file cannot be created.
- ws, err := execute(c, "/bin/touch", "/foo")
+ ws, err := execute(conf, c, "/bin/touch", "/foo")
if err != nil {
t.Fatalf("touch file in ro mount: %v", err)
}
@@ -1723,7 +1723,7 @@ func TestReadonlyMount(t *testing.T) {
// Read mounts to check that volume is readonly.
cmd := fmt.Sprintf("mount | grep ' %s ' | grep -o -e '(.*)'", dir)
- out, err := executeCombinedOutput(c, "/bin/sh", "-c", cmd)
+ out, err := executeCombinedOutput(conf, c, "/bin/sh", "-c", cmd)
if err != nil {
t.Fatalf("exec failed, err: %v", err)
}
@@ -1733,7 +1733,7 @@ func TestReadonlyMount(t *testing.T) {
}
// Check that file cannot be created.
- ws, err := execute(c, "/bin/touch", path.Join(dir, "file"))
+ ws, err := execute(conf, c, "/bin/touch", path.Join(dir, "file"))
if err != nil {
t.Fatalf("touch file in ro mount: %v", err)
}
@@ -2278,13 +2278,13 @@ func TestMountPropagation(t *testing.T) {
// Check that mount didn't propagate to private mount.
privFile := filepath.Join(priv, "mnt", "file")
- if ws, err := execute(cont, "/usr/bin/test", "!", "-f", privFile); err != nil || ws != 0 {
+ if ws, err := execute(conf, cont, "/usr/bin/test", "!", "-f", privFile); err != nil || ws != 0 {
t.Fatalf("exec: test ! -f %q, ws: %v, err: %v", privFile, ws, err)
}
// Check that mount propagated to slave mount.
slaveFile := filepath.Join(slave, "mnt", "file")
- if ws, err := execute(cont, "/usr/bin/test", "-f", slaveFile); err != nil || ws != 0 {
+ if ws, err := execute(conf, cont, "/usr/bin/test", "-f", slaveFile); err != nil || ws != 0 {
t.Fatalf("exec: test -f %q, ws: %v, err: %v", privFile, ws, err)
}
}
@@ -2350,7 +2350,7 @@ func TestMountSymlink(t *testing.T) {
// Check that symlink was resolved and mount was created where the symlink
// is pointing to.
file := path.Join(target, "file")
- if ws, err := execute(cont, "/usr/bin/test", "-f", file); err != nil || ws != 0 {
+ if ws, err := execute(conf, cont, "/usr/bin/test", "-f", file); err != nil || ws != 0 {
t.Fatalf("exec: test -f %q, ws: %v, err: %v", file, ws, err)
}
})
@@ -2589,7 +2589,7 @@ func TestRlimitsExec(t *testing.T) {
t.Fatalf("error starting container: %v", err)
}
- got, err := executeCombinedOutput(cont, "/bin/sh", "-c", "ulimit -n")
+ got, err := executeCombinedOutput(conf, cont, "/bin/sh", "-c", "ulimit -n")
if err != nil {
t.Fatal(err)
}
diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go
index 58ae18232..9d8022e50 100644
--- a/runsc/container/multi_container_test.go
+++ b/runsc/container/multi_container_test.go
@@ -105,11 +105,11 @@ type execDesc struct {
name string
}
-func execMany(t *testing.T, execs []execDesc) {
+func execMany(t *testing.T, conf *config.Config, execs []execDesc) {
for _, exec := range execs {
t.Run(exec.name, func(t *testing.T) {
args := &control.ExecArgs{Argv: exec.cmd}
- if ws, err := exec.c.executeSync(args); err != nil {
+ if ws, err := exec.c.executeSync(conf, 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)
@@ -217,7 +217,7 @@ func TestMultiPIDNS(t *testing.T) {
newProcessBuilder().PID(2).Cmd("sleep").Process(),
newProcessBuilder().Cmd("ps").Process(),
}
- got, err := execPS(containers[0])
+ got, err := execPS(conf, containers[0])
if err != nil {
t.Fatal(err)
}
@@ -229,7 +229,7 @@ func TestMultiPIDNS(t *testing.T) {
newProcessBuilder().PID(1).Cmd("sleep").Process(),
newProcessBuilder().Cmd("ps").Process(),
}
- got, err = execPS(containers[1])
+ got, err = execPS(conf, containers[1])
if err != nil {
t.Fatal(err)
}
@@ -313,7 +313,7 @@ func TestMultiPIDNSPath(t *testing.T) {
newProcessBuilder().PID(3).Cmd("sleep").Process(),
newProcessBuilder().Cmd("ps").Process(),
}
- got, err := execPS(containers[0])
+ got, err := execPS(conf, containers[0])
if err != nil {
t.Fatal(err)
}
@@ -328,7 +328,7 @@ func TestMultiPIDNSPath(t *testing.T) {
newProcessBuilder().PID(3).Cmd("sleep").Process(),
newProcessBuilder().Cmd("ps").Process(),
}
- got, err = execPS(containers[1])
+ got, err = execPS(conf, containers[1])
if err != nil {
t.Fatal(err)
}
@@ -341,7 +341,7 @@ func TestMultiPIDNSPath(t *testing.T) {
newProcessBuilder().PID(1).Cmd("sleep").Process(),
newProcessBuilder().Cmd("ps").Process(),
}
- got, err = execPS(containers[2])
+ got, err = execPS(conf, containers[2])
if err != nil {
t.Fatal(err)
}
@@ -541,7 +541,7 @@ func TestExecWait(t *testing.T) {
WorkingDirectory: "/",
KUID: 0,
}
- pid, err := containers[0].Execute(args)
+ pid, err := containers[0].Execute(conf, args)
if err != nil {
t.Fatalf("error executing: %v", err)
}
@@ -744,7 +744,7 @@ func TestMultiContainerDestroy(t *testing.T) {
Filename: app,
Argv: []string{app, "fork-bomb"},
}
- if _, err := containers[1].Execute(args); err != nil {
+ if _, err := containers[1].Execute(conf, args); err != nil {
t.Fatalf("error exec'ing: %v", err)
}
@@ -821,7 +821,7 @@ func TestMultiContainerProcesses(t *testing.T) {
Filename: "/bin/sleep",
Argv: []string{"/bin/sleep", "100"},
}
- if _, err := containers[1].Execute(args); err != nil {
+ if _, err := containers[1].Execute(conf, args); err != nil {
t.Fatalf("error exec'ing: %v", err)
}
expectedPL1 = append(expectedPL1, newProcessBuilder().PID(4).Cmd("sleep").Process())
@@ -882,7 +882,7 @@ func TestMultiContainerKillAll(t *testing.T) {
Filename: app,
Argv: []string{app, "task-tree", "--depth=2", "--width=2"},
}
- if _, err := containers[1].Execute(args); err != nil {
+ if _, err := containers[1].Execute(conf, args); err != nil {
t.Fatalf("error exec'ing: %v", err)
}
// Wait for these new processes to start.
@@ -1317,7 +1317,7 @@ func TestMultiContainerSharedMount(t *testing.T) {
name: "dir removed from container1",
},
}
- execMany(t, execs)
+ execMany(t, conf, execs)
})
}
}
@@ -1382,7 +1382,7 @@ func TestMultiContainerSharedMountReadonly(t *testing.T) {
name: "fails to write to container1",
},
}
- execMany(t, execs)
+ execMany(t, conf, execs)
})
}
}
@@ -1440,7 +1440,7 @@ func TestMultiContainerSharedMountRestart(t *testing.T) {
name: "file appears in container1",
},
}
- execMany(t, execs)
+ execMany(t, conf, execs)
containers[1].Destroy()
@@ -1490,7 +1490,7 @@ func TestMultiContainerSharedMountRestart(t *testing.T) {
name: "file removed from container1",
},
}
- execMany(t, execs)
+ execMany(t, conf, execs)
})
}
}
@@ -1543,7 +1543,7 @@ func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) {
name: "directory is mounted in container1",
},
}
- execMany(t, execs)
+ execMany(t, conf, execs)
})
}
}
@@ -1654,7 +1654,7 @@ func TestMultiContainerGoferKilled(t *testing.T) {
}
// Check that container isn't running anymore.
- if _, err := execute(c, "/bin/true"); err == nil {
+ if _, err := execute(conf, c, "/bin/true"); err == nil {
t.Fatalf("Container %q was not stopped after gofer death", c.ID)
}
@@ -1669,7 +1669,7 @@ func TestMultiContainerGoferKilled(t *testing.T) {
if err := waitForProcessList(c, pl); err != nil {
t.Errorf("Container %q was affected by another container: %v", c.ID, err)
}
- if _, err := execute(c, "/bin/true"); err != nil {
+ if _, err := execute(conf, c, "/bin/true"); err != nil {
t.Fatalf("Container %q was affected by another container: %v", c.ID, err)
}
}
@@ -1691,7 +1691,7 @@ func TestMultiContainerGoferKilled(t *testing.T) {
// Check that entire sandbox isn't running anymore.
for _, c := range containers {
- if _, err := execute(c, "/bin/true"); err == nil {
+ if _, err := execute(conf, c, "/bin/true"); err == nil {
t.Fatalf("Container %q was not stopped after gofer death", c.ID)
}
}
@@ -1867,7 +1867,7 @@ func TestMultiContainerHomeEnvDir(t *testing.T) {
defer cleanup()
// Exec into the root container synchronously.
- if _, err := execute(containers[0], "/bin/sh", "-c", execCmd); err != nil {
+ if _, err := execute(conf, containers[0], "/bin/sh", "-c", execCmd); err != nil {
t.Errorf("error executing %+v: %v", execCmd, err)
}
@@ -2056,7 +2056,7 @@ func TestDuplicateEnvVariable(t *testing.T) {
Argv: []string{"/bin/sh", "-c", cmdExec},
Envv: []string{"VAR=foo", "VAR=bar"},
}
- if ws, err := containers[0].executeSync(execArgs); err != nil || ws.ExitStatus() != 0 {
+ if ws, err := containers[0].executeSync(conf, execArgs); err != nil || ws.ExitStatus() != 0 {
t.Fatalf("exec failed, ws: %v, err: %v", ws, err)
}
diff --git a/runsc/container/shared_volume_test.go b/runsc/container/shared_volume_test.go
index 7d05ee16b..f16b2bd02 100644
--- a/runsc/container/shared_volume_test.go
+++ b/runsc/container/shared_volume_test.go
@@ -72,7 +72,7 @@ func TestSharedVolume(t *testing.T) {
Filename: "/usr/bin/test",
Argv: []string{"test", "-f", filename},
}
- if ws, err := c.executeSync(argsTestFile); err != nil {
+ if ws, err := c.executeSync(conf, argsTestFile); err != nil {
t.Fatalf("unexpected error testing file %q: %v", filename, err)
} else if ws.ExitStatus() == 0 {
t.Errorf("test %q exited with code %v, wanted not zero", ws.ExitStatus(), err)
@@ -84,7 +84,7 @@ func TestSharedVolume(t *testing.T) {
}
// Now we should be able to test the file from within the sandbox.
- if ws, err := c.executeSync(argsTestFile); err != nil {
+ if ws, err := c.executeSync(conf, argsTestFile); err != nil {
t.Fatalf("unexpected error testing file %q: %v", filename, err)
} else if ws.ExitStatus() != 0 {
t.Errorf("test %q exited with code %v, wanted zero", filename, ws.ExitStatus())
@@ -97,7 +97,7 @@ func TestSharedVolume(t *testing.T) {
}
// File should no longer exist at the old path within the sandbox.
- if ws, err := c.executeSync(argsTestFile); err != nil {
+ if ws, err := c.executeSync(conf, argsTestFile); err != nil {
t.Fatalf("unexpected error testing file %q: %v", filename, err)
} else if ws.ExitStatus() == 0 {
t.Errorf("test %q exited with code %v, wanted not zero", filename, ws.ExitStatus())
@@ -108,7 +108,7 @@ func TestSharedVolume(t *testing.T) {
Filename: "/usr/bin/test",
Argv: []string{"test", "-f", newFilename},
}
- if ws, err := c.executeSync(argsTestNewFile); err != nil {
+ if ws, err := c.executeSync(conf, argsTestNewFile); err != nil {
t.Fatalf("unexpected error testing file %q: %v", newFilename, err)
} else if ws.ExitStatus() != 0 {
t.Errorf("test %q exited with code %v, wanted zero", newFilename, ws.ExitStatus())
@@ -120,7 +120,7 @@ func TestSharedVolume(t *testing.T) {
}
// Renamed file should no longer exist at the old path within the sandbox.
- if ws, err := c.executeSync(argsTestNewFile); err != nil {
+ if ws, err := c.executeSync(conf, argsTestNewFile); err != nil {
t.Fatalf("unexpected error testing file %q: %v", newFilename, err)
} else if ws.ExitStatus() == 0 {
t.Errorf("test %q exited with code %v, wanted not zero", newFilename, ws.ExitStatus())
@@ -133,7 +133,7 @@ func TestSharedVolume(t *testing.T) {
KUID: auth.KUID(os.Getuid()),
KGID: auth.KGID(os.Getgid()),
}
- if ws, err := c.executeSync(argsTouch); err != nil {
+ if ws, err := c.executeSync(conf, argsTouch); err != nil {
t.Fatalf("unexpected error touching file %q: %v", filename, err)
} else if ws.ExitStatus() != 0 {
t.Errorf("touch %q exited with code %v, wanted zero", filename, ws.ExitStatus())
@@ -154,7 +154,7 @@ func TestSharedVolume(t *testing.T) {
Filename: "/bin/rm",
Argv: []string{"rm", filename},
}
- if ws, err := c.executeSync(argsRemove); err != nil {
+ if ws, err := c.executeSync(conf, argsRemove); err != nil {
t.Fatalf("unexpected error removing file %q: %v", filename, err)
} else if ws.ExitStatus() != 0 {
t.Errorf("remove %q exited with code %v, wanted zero", filename, ws.ExitStatus())
@@ -166,9 +166,9 @@ func TestSharedVolume(t *testing.T) {
}
}
-func checkFile(c *Container, filename string, want []byte) error {
+func checkFile(conf *config.Config, c *Container, filename string, want []byte) error {
cpy := filename + ".copy"
- if _, err := execute(c, "/bin/cp", "-f", filename, cpy); err != nil {
+ if _, err := execute(conf, c, "/bin/cp", "-f", filename, cpy); err != nil {
return fmt.Errorf("unexpected error copying file %q to %q: %v", filename, cpy, err)
}
got, err := ioutil.ReadFile(cpy)
@@ -226,16 +226,16 @@ func TestSharedVolumeFile(t *testing.T) {
if err := ioutil.WriteFile(filename, []byte(want), 0666); err != nil {
t.Fatalf("Error writing to %q: %v", filename, err)
}
- if err := checkFile(c, filename, want); err != nil {
+ if err := checkFile(conf, c, filename, want); err != nil {
t.Fatal(err.Error())
}
// Append to file inside the container and check that content is not lost.
- if _, err := execute(c, "/bin/bash", "-c", "echo -n sandbox- >> "+filename); err != nil {
+ if _, err := execute(conf, c, "/bin/bash", "-c", "echo -n sandbox- >> "+filename); err != nil {
t.Fatalf("unexpected error appending file %q: %v", filename, err)
}
want = []byte("host-sandbox-")
- if err := checkFile(c, filename, want); err != nil {
+ if err := checkFile(conf, c, filename, want); err != nil {
t.Fatal(err.Error())
}
@@ -250,7 +250,7 @@ func TestSharedVolumeFile(t *testing.T) {
t.Fatalf("Error writing to file %q: %v", filename, err)
}
want = []byte("host-sandbox-host")
- if err := checkFile(c, filename, want); err != nil {
+ if err := checkFile(conf, c, filename, want); err != nil {
t.Fatal(err.Error())
}
@@ -259,7 +259,7 @@ func TestSharedVolumeFile(t *testing.T) {
t.Fatalf("Error truncating file %q: %v", filename, err)
}
want = want[:5]
- if err := checkFile(c, filename, want); err != nil {
+ if err := checkFile(conf, c, filename, want); err != nil {
t.Fatal(err.Error())
}
}
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index 9dea7c4d2..95b5d9615 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -65,6 +65,11 @@ type Sandbox struct {
// is not running.
Pid int `json:"pid"`
+ // UID is the user ID in the parent namespace that the sandbox is running as.
+ UID int `json:"uid"`
+ // GID is the group ID in the parent namespace that the sandbox is running as.
+ GID int `json:"gid"`
+
// Cgroup has the cgroup configuration for the sandbox.
Cgroup *cgroup.Cgroup `json:"cgroup"`
@@ -176,18 +181,22 @@ func New(conf *config.Config, args *Args) (*Sandbox, error) {
}
// CreateContainer creates a non-root container inside the sandbox.
-func (s *Sandbox) CreateContainer(cid string, tty *os.File) error {
+func (s *Sandbox) CreateContainer(conf *config.Config, cid string, tty *os.File) error {
log.Debugf("Create non-root container %q in sandbox %q, PID: %d", cid, s.ID, s.Pid)
- sandboxConn, err := s.sandboxConnect()
- if err != nil {
- return fmt.Errorf("couldn't connect to sandbox: %v", err)
- }
- defer sandboxConn.Close()
var files []*os.File
if tty != nil {
files = []*os.File{tty}
}
+ if err := s.configureStdios(conf, files); err != nil {
+ return err
+ }
+
+ sandboxConn, err := s.sandboxConnect()
+ if err != nil {
+ return fmt.Errorf("couldn't connect to sandbox: %v", err)
+ }
+ defer sandboxConn.Close()
args := boot.CreateArgs{
CID: cid,
@@ -225,6 +234,11 @@ func (s *Sandbox) StartRoot(spec *specs.Spec, conf *config.Config) error {
// StartContainer starts running a non-root container inside the sandbox.
func (s *Sandbox) StartContainer(spec *specs.Spec, conf *config.Config, cid string, stdios, goferFiles []*os.File) error {
log.Debugf("Start non-root container %q in sandbox %q, PID: %d", cid, s.ID, s.Pid)
+
+ if err := s.configureStdios(conf, stdios); err != nil {
+ return err
+ }
+
sandboxConn, err := s.sandboxConnect()
if err != nil {
return fmt.Errorf("couldn't connect to sandbox: %v", err)
@@ -318,8 +332,13 @@ func (s *Sandbox) NewCGroup() (*cgroup.Cgroup, error) {
// Execute runs the specified command in the container. It returns the PID of
// the newly created process.
-func (s *Sandbox) Execute(args *control.ExecArgs) (int32, error) {
+func (s *Sandbox) Execute(conf *config.Config, args *control.ExecArgs) (int32, error) {
log.Debugf("Executing new process in container %q in sandbox %q", args.ContainerID, s.ID)
+
+ if err := s.configureStdios(conf, args.Files); err != nil {
+ return 0, err
+ }
+
conn, err := s.sandboxConnect()
if err != nil {
return 0, s.connError(err)
@@ -505,6 +524,7 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
cmd.Stdin = nil
cmd.Stdout = nil
cmd.Stderr = nil
+ var stdios [3]*os.File
// If the console control socket file is provided, then create a new
// pty master/replica pair and set the TTY on the sandbox process.
@@ -525,11 +545,9 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
cmd.SysProcAttr.Ctty = nextFD
// Pass the tty as all stdio fds to sandbox.
- for i := 0; i < 3; i++ {
- cmd.ExtraFiles = append(cmd.ExtraFiles, tty)
- cmd.Args = append(cmd.Args, "--stdio-fds="+strconv.Itoa(nextFD))
- nextFD++
- }
+ stdios[0] = tty
+ stdios[1] = tty
+ stdios[2] = tty
if conf.Debug {
// If debugging, send the boot process stdio to the
@@ -541,11 +559,9 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
} else {
// If not using a console, pass our current stdio as the
// container stdio via flags.
- for _, f := range []*os.File{os.Stdin, os.Stdout, os.Stderr} {
- cmd.ExtraFiles = append(cmd.ExtraFiles, f)
- cmd.Args = append(cmd.Args, "--stdio-fds="+strconv.Itoa(nextFD))
- nextFD++
- }
+ stdios[0] = os.Stdin
+ stdios[1] = os.Stdout
+ stdios[2] = os.Stderr
if conf.Debug {
// If debugging, send the boot process stdio to the
@@ -595,6 +611,10 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
nss = append(nss, specs.LinuxNamespace{Type: specs.NetworkNamespace})
}
+ // These are set to the uid/gid that the sandbox process will use.
+ s.UID = os.Getuid()
+ s.GID = os.Getgid()
+
// User namespace depends on the network type. Host network requires to run
// inside the user namespace specified in the spec or the current namespace
// if none is configured.
@@ -636,51 +656,49 @@ func (s *Sandbox) createSandboxProcess(conf *config.Config, args *Args, startSyn
const nobody = 65534
if conf.Rootless {
log.Infof("Rootless mode: sandbox will run as nobody inside user namespace, mapped to the current user, uid: %d, gid: %d", os.Getuid(), os.Getgid())
- cmd.SysProcAttr.UidMappings = []syscall.SysProcIDMap{
- {
- ContainerID: nobody,
- HostID: os.Getuid(),
- Size: 1,
- },
- }
- cmd.SysProcAttr.GidMappings = []syscall.SysProcIDMap{
- {
- ContainerID: nobody,
- HostID: os.Getgid(),
- Size: 1,
- },
- }
-
} else {
// Map nobody in the new namespace to nobody in the parent namespace.
- //
- // A sandbox process will construct an empty
- // root for itself, so it has to have
- // CAP_SYS_ADMIN and CAP_SYS_CHROOT capabilities.
- cmd.SysProcAttr.UidMappings = []syscall.SysProcIDMap{
- {
- ContainerID: nobody,
- HostID: nobody,
- Size: 1,
- },
- }
- cmd.SysProcAttr.GidMappings = []syscall.SysProcIDMap{
- {
- ContainerID: nobody,
- HostID: nobody,
- Size: 1,
- },
- }
+ s.UID = nobody
+ s.GID = nobody
}
// Set credentials to run as user and group nobody.
cmd.SysProcAttr.Credential = &syscall.Credential{Uid: nobody, Gid: nobody}
+ cmd.SysProcAttr.UidMappings = []syscall.SysProcIDMap{
+ {
+ ContainerID: nobody,
+ HostID: s.UID,
+ Size: 1,
+ },
+ }
+ cmd.SysProcAttr.GidMappings = []syscall.SysProcIDMap{
+ {
+ ContainerID: nobody,
+ HostID: s.GID,
+ Size: 1,
+ },
+ }
+
+ // A sandbox process will construct an empty root for itself, so it has
+ // to have CAP_SYS_ADMIN and CAP_SYS_CHROOT capabilities.
cmd.SysProcAttr.AmbientCaps = append(cmd.SysProcAttr.AmbientCaps, uintptr(capability.CAP_SYS_ADMIN), uintptr(capability.CAP_SYS_CHROOT))
+
} else {
return fmt.Errorf("can't run sandbox process as user nobody since we don't have CAP_SETUID or CAP_SETGID")
}
}
+ if err := s.configureStdios(conf, stdios[:]); err != nil {
+ return fmt.Errorf("configuring stdios: %w", err)
+ }
+ for _, file := range stdios {
+ cmd.ExtraFiles = append(cmd.ExtraFiles, file)
+ cmd.Args = append(cmd.Args, "--stdio-fds="+strconv.Itoa(nextFD))
+ nextFD++
+ }
+
+ // Set Args[0] to make easier to spot the sandbox process. Otherwise it's
+ // shown as `exe`.
cmd.Args[0] = "runsc-sandbox"
if s.Cgroup != nil {
@@ -1167,6 +1185,23 @@ func (s *Sandbox) waitForStopped() error {
return backoff.Retry(op, b)
}
+// configureStdios change stdios ownership to give access to the sandbox
+// process. This may be skipped depending on the configuration.
+func (s *Sandbox) configureStdios(conf *config.Config, stdios []*os.File) error {
+ if conf.Rootless || conf.TestOnlyAllowRunAsCurrentUserWithoutChroot {
+ // Cannot change ownership without CAP_CHOWN.
+ return nil
+ }
+
+ for _, file := range stdios {
+ log.Debugf("Changing %q ownership to %d/%d", file.Name(), s.UID, s.GID)
+ if err := file.Chown(s.UID, s.GID); err != nil {
+ return err
+ }
+ }
+ return nil
+}
+
// deviceFileForPlatform opens the device file for the given platform. If the
// platform does not need a device file, then nil is returned.
func deviceFileForPlatform(name string) (*os.File, error) {
diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go
index 1accc3b3b..f53417cab 100644
--- a/test/e2e/integration_test.go
+++ b/test/e2e/integration_test.go
@@ -30,6 +30,7 @@ import (
"net/http"
"os"
"path/filepath"
+ "regexp"
"strconv"
"strings"
"testing"
@@ -426,10 +427,10 @@ func TestTmpMount(t *testing.T) {
// Test that it is allowed to mount a file on top of /dev files, e.g.
// /dev/random.
func TestMountOverDev(t *testing.T) {
- if usingVFS2, err := dockerutil.UsingVFS2(); !usingVFS2 {
- t.Skip("VFS1 doesn't allow /dev/random to be mounted.")
- } else if err != nil {
+ if vfs2, err := dockerutil.UsingVFS2(); err != nil {
t.Fatalf("Failed to read config for runtime %s: %v", dockerutil.Runtime(), err)
+ } else if !vfs2 {
+ t.Skip("VFS1 doesn't allow /dev/random to be mounted.")
}
random, err := ioutil.TempFile(testutil.TmpDir(), "random")
@@ -574,11 +575,12 @@ func runIntegrationTest(t *testing.T, capAdd []string, args ...string) {
d := dockerutil.MakeContainer(ctx, t)
defer d.CleanUp(ctx)
- if got, err := d.Run(ctx, dockerutil.RunOpts{
+ opts := dockerutil.RunOpts{
Image: "basic/integrationtest",
WorkDir: "/root",
CapAdd: capAdd,
- }, args...); err != nil {
+ }
+ if got, err := d.Run(ctx, opts, args...); err != nil {
t.Fatalf("docker run failed: %v", err)
} else if got != "" {
t.Errorf("test failed:\n%s", got)
@@ -609,6 +611,107 @@ func TestBindOverlay(t *testing.T) {
}
}
+func TestStdios(t *testing.T) {
+ if vfs2, err := dockerutil.UsingVFS2(); err != nil {
+ t.Fatalf("Failed to read config for runtime %s: %v", dockerutil.Runtime(), err)
+ } else if !vfs2 {
+ t.Skip("VFS1 doesn't adjust stdios user")
+ }
+
+ ctx := context.Background()
+ d := dockerutil.MakeContainer(ctx, t)
+ defer d.CleanUp(ctx)
+
+ testStdios(t, func(user string, args ...string) (string, error) {
+ defer d.CleanUp(ctx)
+ opts := dockerutil.RunOpts{
+ Image: "basic/alpine",
+ User: user,
+ }
+ return d.Run(ctx, opts, args...)
+ })
+}
+
+func TestStdiosExec(t *testing.T) {
+ if vfs2, err := dockerutil.UsingVFS2(); err != nil {
+ t.Fatalf("Failed to read config for runtime %s: %v", dockerutil.Runtime(), err)
+ } else if !vfs2 {
+ t.Skip("VFS1 doesn't adjust stdios user")
+ }
+
+ ctx := context.Background()
+ d := dockerutil.MakeContainer(ctx, t)
+ defer d.CleanUp(ctx)
+
+ runOpts := dockerutil.RunOpts{Image: "basic/alpine"}
+ if err := d.Spawn(ctx, runOpts, "sleep", "100"); err != nil {
+ t.Fatalf("docker run failed: %v", err)
+ }
+
+ testStdios(t, func(user string, args ...string) (string, error) {
+ opts := dockerutil.ExecOpts{User: user}
+ return d.Exec(ctx, opts, args...)
+ })
+}
+
+func testStdios(t *testing.T, run func(string, ...string) (string, error)) {
+ const cmd = "stat -L /proc/self/fd/0 /proc/self/fd/1 /proc/self/fd/2 | grep 'Uid:'"
+ got, err := run("123", "/bin/sh", "-c", cmd)
+ if err != nil {
+ t.Fatalf("docker exec failed: %v", err)
+ }
+ if len(got) == 0 {
+ t.Errorf("Unexpected empty output from %q", cmd)
+ }
+ re := regexp.MustCompile(`Uid: \(\s*(\w+)\/.*\)`)
+ for _, line := range strings.SplitN(got, "\n", 3) {
+ t.Logf("stat -L: %s", line)
+ matches := re.FindSubmatch([]byte(line))
+ if len(matches) != 2 {
+ t.Fatalf("wrong output format: %q: matches: %v", line, matches)
+ }
+ if want, got := "123", string(matches[1]); want != got {
+ t.Errorf("wrong user, want: %q, got: %q", want, got)
+ }
+ }
+
+ // Check that stdout and stderr can be open and written to. This checks
+ // that ownership and permissions are correct inside gVisor.
+ got, err = run("456", "/bin/sh", "-c", "echo foobar | tee /proc/self/fd/1 > /proc/self/fd/2")
+ if err != nil {
+ t.Fatalf("docker run failed: %v", err)
+ }
+ t.Logf("echo foobar: %q", got)
+ // Check it repeats twice, once for stdout and once for stderr.
+ if want := "foobar\nfoobar\n"; want != got {
+ t.Errorf("Wrong echo output, want: %q, got: %q", want, got)
+ }
+
+ // Check that timestamps can be changed. Setting timestamps require an extra
+ // write check _after_ the file was opened, and may fail if the underlying
+ // host file is not setup correctly.
+ if _, err := run("789", "touch", "/proc/self/fd/0", "/proc/self/fd/1", "/proc/self/fd/2"); err != nil {
+ t.Fatalf("docker run failed: %v", err)
+ }
+}
+
+func TestStdiosChown(t *testing.T) {
+ if vfs2, err := dockerutil.UsingVFS2(); err != nil {
+ t.Fatalf("Failed to read config for runtime %s: %v", dockerutil.Runtime(), err)
+ } else if !vfs2 {
+ t.Skip("VFS1 doesn't adjust stdios user")
+ }
+
+ ctx := context.Background()
+ d := dockerutil.MakeContainer(ctx, t)
+ defer d.CleanUp(ctx)
+
+ opts := dockerutil.RunOpts{Image: "basic/alpine"}
+ if _, err := d.Run(ctx, opts, "chown", "123", "/proc/self/fd/0", "/proc/self/fd/1", "/proc/self/fd/2"); err != nil {
+ t.Fatalf("docker run failed: %v", err)
+ }
+}
+
func TestMain(m *testing.M) {
dockerutil.EnsureSupportedDockerVersion()
flag.Parse()
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-07 11:35:47 +0000