Merge release-20200810.0-83-g5ec3d4ed3 (automated)

9 files changed, 99 insertions, 39 deletions

diff --git a/pkg/abi/linux/linux_abi_autogen_unsafe.go b/pkg/abi/linux/linux_abi_autogen_unsafe.go
index 93a6b4cd4..64994692a 100644
--- a/pkg/abi/linux/linux_abi_autogen_unsafe.go
+++ b/pkg/abi/linux/linux_abi_autogen_unsafe.go
@@ -167,7 +167,7 @@ func (s *Statx) MarshalUnsafe(dst []byte) {
 
 // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.
 func (s *Statx) UnmarshalUnsafe(src []byte) {
-    if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() {
+    if s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() {
         safecopy.CopyOut(unsafe.Pointer(s), src)
     } else {
         // Type Statx doesn't have a packed layout in memory, fallback to UnmarshalBytes.
@@ -208,7 +208,7 @@ func (s *Statx) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {
 // CopyIn implements marshal.Marshallable.CopyIn.
 //go:nosplit
 func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
-    if !s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() {
+    if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() {
         // Type Statx doesn't have a packed layout in memory, fall back to UnmarshalBytes.
         buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay.
         length, err := task.CopyInBytes(addr, buf) // escapes: okay.
@@ -642,7 +642,7 @@ func (f *FUSEHeaderIn) MarshalUnsafe(dst []byte) {
 
 // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.
 func (f *FUSEHeaderIn) UnmarshalUnsafe(src []byte) {
-    if f.Opcode.Packed() && f.Unique.Packed() {
+    if f.Unique.Packed() && f.Opcode.Packed() {
         safecopy.CopyOut(unsafe.Pointer(f), src)
     } else {
         // Type FUSEHeaderIn doesn't have a packed layout in memory, fallback to UnmarshalBytes.
@@ -683,7 +683,7 @@ func (f *FUSEHeaderIn) CopyOut(task marshal.Task, addr usermem.Addr) (int, error
 // CopyIn implements marshal.Marshallable.CopyIn.
 //go:nosplit
 func (f *FUSEHeaderIn) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
-    if !f.Opcode.Packed() && f.Unique.Packed() {
+    if !f.Unique.Packed() && f.Opcode.Packed() {
         // Type FUSEHeaderIn doesn't have a packed layout in memory, fall back to UnmarshalBytes.
         buf := task.CopyScratchBuffer(f.SizeBytes()) // escapes: okay.
         length, err := task.CopyInBytes(addr, buf) // escapes: okay.
@@ -2020,7 +2020,7 @@ func (i *IPTEntry) UnmarshalBytes(src []byte) {
 // Packed implements marshal.Marshallable.Packed.
 //go:nosplit
 func (i *IPTEntry) Packed() bool {
-    return i.Counters.Packed() && i.IP.Packed()
+    return i.IP.Packed() && i.Counters.Packed()
 }
 
 // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe.
@@ -2035,7 +2035,7 @@ func (i *IPTEntry) MarshalUnsafe(dst []byte) {
 
 // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.
 func (i *IPTEntry) UnmarshalUnsafe(src []byte) {
-    if i.IP.Packed() && i.Counters.Packed() {
+    if i.Counters.Packed() && i.IP.Packed() {
         safecopy.CopyOut(unsafe.Pointer(i), src)
     } else {
         // Type IPTEntry doesn't have a packed layout in memory, fallback to UnmarshalBytes.
@@ -2208,7 +2208,7 @@ func (i *IPTIP) UnmarshalBytes(src []byte) {
 // Packed implements marshal.Marshallable.Packed.
 //go:nosplit
 func (i *IPTIP) Packed() bool {
-    return i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed() && i.Src.Packed()
+    return i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed()
 }
 
 // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe.
@@ -2223,7 +2223,7 @@ func (i *IPTIP) MarshalUnsafe(dst []byte) {
 
 // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.
 func (i *IPTIP) UnmarshalUnsafe(src []byte) {
-    if i.DstMask.Packed() && i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() {
+    if i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed() {
         safecopy.CopyOut(unsafe.Pointer(i), src)
     } else {
         // Type IPTIP doesn't have a packed layout in memory, fallback to UnmarshalBytes.
@@ -2290,7 +2290,7 @@ func (i *IPTIP) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
 
 // WriteTo implements io.WriterTo.WriteTo.
 func (i *IPTIP) WriteTo(w io.Writer) (int64, error) {
-    if !i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed() {
+    if !i.DstMask.Packed() && i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() {
         // Type IPTIP doesn't have a packed layout in memory, fall back to MarshalBytes.
         buf := make([]byte, i.SizeBytes())
         i.MarshalBytes(buf)
@@ -3055,7 +3055,7 @@ func (i *IP6TEntry) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {
 // CopyIn implements marshal.Marshallable.CopyIn.
 //go:nosplit
 func (i *IP6TEntry) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
-    if !i.IPv6.Packed() && i.Counters.Packed() {
+    if !i.Counters.Packed() && i.IPv6.Packed() {
         // Type IP6TEntry doesn't have a packed layout in memory, fall back to UnmarshalBytes.
         buf := task.CopyScratchBuffer(i.SizeBytes()) // escapes: okay.
         length, err := task.CopyInBytes(addr, buf) // escapes: okay.
@@ -3196,7 +3196,7 @@ func (i *IP6TIP) UnmarshalBytes(src []byte) {
 // Packed implements marshal.Marshallable.Packed.
 //go:nosplit
 func (i *IP6TIP) Packed() bool {
-    return i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed()
+    return i.DstMask.Packed() && i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed()
 }
 
 // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe.
@@ -3278,7 +3278,7 @@ func (i *IP6TIP) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
 
 // WriteTo implements io.WriterTo.WriteTo.
 func (i *IP6TIP) WriteTo(w io.Writer) (int64, error) {
-    if !i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() && i.DstMask.Packed() {
+    if !i.DstMask.Packed() && i.Src.Packed() && i.Dst.Packed() && i.SrcMask.Packed() {
         // Type IP6TIP doesn't have a packed layout in memory, fall back to MarshalBytes.
         buf := make([]byte, i.SizeBytes())
         i.MarshalBytes(buf)
diff --git a/pkg/sentry/fsimpl/host/connected_endpoint_refs.go b/pkg/sentry/fsimpl/host/connected_endpoint_refs.go
index 1fcf8ab73..3ef6500fa 100644
--- a/pkg/sentry/fsimpl/host/connected_endpoint_refs.go
+++ b/pkg/sentry/fsimpl/host/connected_endpoint_refs.go
@@ -1,10 +1,11 @@
 package host
 
 import (
-	"gvisor.dev/gvisor/pkg/log"
 	refs_vfs1 "gvisor.dev/gvisor/pkg/refs"
 	"runtime"
 	"sync/atomic"
+
+	"gvisor.dev/gvisor/pkg/log"
 )
 
 // ownerType is used to customize logging. Note that we use a pointer to T so
diff --git a/pkg/sentry/fsimpl/host/inode_refs.go b/pkg/sentry/fsimpl/host/inode_refs.go
index c920a3c4e..fc13b3cc7 100644
--- a/pkg/sentry/fsimpl/host/inode_refs.go
+++ b/pkg/sentry/fsimpl/host/inode_refs.go
@@ -1,10 +1,11 @@
 package host
 
 import (
-	"gvisor.dev/gvisor/pkg/log"
 	refs_vfs1 "gvisor.dev/gvisor/pkg/refs"
 	"runtime"
 	"sync/atomic"
+
+	"gvisor.dev/gvisor/pkg/log"
 )
 
 // ownerType is used to customize logging. Note that we use a pointer to T so
diff --git a/pkg/sentry/fsimpl/tmpfs/inode_refs.go b/pkg/sentry/fsimpl/tmpfs/inode_refs.go
index 38eddde7e..175dc3409 100644
--- a/pkg/sentry/fsimpl/tmpfs/inode_refs.go
+++ b/pkg/sentry/fsimpl/tmpfs/inode_refs.go
@@ -1,10 +1,11 @@
 package tmpfs
 
 import (
-	"gvisor.dev/gvisor/pkg/log"
 	refs_vfs1 "gvisor.dev/gvisor/pkg/refs"
 	"runtime"
 	"sync/atomic"
+
+	"gvisor.dev/gvisor/pkg/log"
 )
 
 // ownerType is used to customize logging. Note that we use a pointer to T so
diff --git a/pkg/sentry/socket/unix/socket_refs.go b/pkg/sentry/socket/unix/socket_refs.go
index 8eb0c9327..090395057 100644
--- a/pkg/sentry/socket/unix/socket_refs.go
+++ b/pkg/sentry/socket/unix/socket_refs.go
@@ -1,10 +1,11 @@
 package unix
 
 import (
-	"gvisor.dev/gvisor/pkg/log"
 	refs_vfs1 "gvisor.dev/gvisor/pkg/refs"
 	"runtime"
 	"sync/atomic"
+
+	"gvisor.dev/gvisor/pkg/log"
 )
 
 // ownerType is used to customize logging. Note that we use a pointer to T so
diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go
index 4bd5c7ca2..769c9b92f 100644
--- a/pkg/sentry/syscalls/linux/vfs2/mount.go
+++ b/pkg/sentry/syscalls/linux/vfs2/mount.go
@@ -109,8 +109,8 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 		return 0, nil, err
 	}
 	defer target.Release(t)
-
-	return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts)
+	_, err = t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts)
+	return 0, nil, err
 }
 
 // Umount2 implements Linux syscall umount2(2).
diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go
index c9e724fef..97018651f 100644
--- a/pkg/sentry/vfs/context.go
+++ b/pkg/sentry/vfs/context.go
@@ -40,6 +40,30 @@ func MountNamespaceFromContext(ctx context.Context) *MountNamespace {
 	return nil
 }
 
+type mountNamespaceContext struct {
+	context.Context
+	mntns *MountNamespace
+}
+
+// Value implements Context.Value.
+func (mc mountNamespaceContext) Value(key interface{}) interface{} {
+	switch key {
+	case CtxMountNamespace:
+		mc.mntns.IncRef()
+		return mc.mntns
+	default:
+		return mc.Context.Value(key)
+	}
+}
+
+// WithMountNamespace returns a copy of ctx with the given MountNamespace.
+func WithMountNamespace(ctx context.Context, mntns *MountNamespace) context.Context {
+	return &mountNamespaceContext{
+		Context: ctx,
+		mntns:   mntns,
+	}
+}
+
 // RootFromContext returns the VFS root used by ctx. It takes a reference on
 // the returned VirtualDentry. If ctx does not have a specific VFS root,
 // RootFromContext returns a zero-value VirtualDentry.
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index 714af6907..09fea3628 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -263,16 +263,20 @@ func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Cr
 }
 
 // MountAt creates and mounts a Filesystem configured by the given arguments.
-func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error {
+// The VirtualFilesystem will hold a reference to the Mount until it is unmounted.
+//
+// This method returns the mounted Mount without a reference, for convenience
+// during VFS setup when there is no chance of racing with unmount.
+func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) (*Mount, error) {
 	mnt, err := vfs.MountDisconnected(ctx, creds, source, fsTypeName, opts)
 	if err != nil {
-		return err
+		return nil, err
 	}
 	defer mnt.DecRef(ctx)
 	if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil {
-		return err
+		return nil, err
 	}
-	return nil
+	return mnt, nil
 }
 
 // UmountAt removes the Mount at the given path.
@@ -657,6 +661,13 @@ retryFirst:
 	return VirtualDentry{mnt, d}
 }
 
+// SetMountReadOnly sets the mount as ReadOnly.
+func (vfs *VirtualFilesystem) SetMountReadOnly(mnt *Mount, ro bool) error {
+	vfs.mountMu.Lock()
+	defer vfs.mountMu.Unlock()
+	return mnt.setReadOnlyLocked(ro)
+}
+
 // CheckBeginWrite increments the counter of in-progress write operations on
 // mnt. If mnt is mounted MS_RDONLY, CheckBeginWrite does nothing and returns
 // EROFS.
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index f27a6ff6b..fb200e988 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -205,15 +205,34 @@ func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *config.
 	for i := range mounts {
 		submount := &mounts[i]
 		log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options)
+		var (
+			mnt *vfs.Mount
+			err error
+		)
+
 		if hint := c.hints.findMount(submount.Mount); hint != nil && hint.isSupported() {
-			if err := c.mountSharedSubmountVFS2(ctx, conf, mns, creds, submount.Mount, hint); err != nil {
+			mnt, err = c.mountSharedSubmountVFS2(ctx, conf, mns, creds, submount.Mount, hint)
+			if err != nil {
 				return fmt.Errorf("mount shared mount %q to %q: %v", hint.name, submount.Destination, err)
 			}
 		} else {
-			if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil {
+			mnt, err = c.mountSubmountVFS2(ctx, conf, mns, creds, submount)
+			if err != nil {
 				return fmt.Errorf("mount submount %q: %w", submount.Destination, err)
 			}
 		}
+
+		if mnt != nil && mnt.ReadOnly() {
+			// Switch to ReadWrite while we setup submounts.
+			if err := c.k.VFS().SetMountReadOnly(mnt, false); err != nil {
+				return fmt.Errorf("failed to set mount at %q readwrite: %v", submount.Destination, err)
+			}
+			defer func() {
+				if err := c.k.VFS().SetMountReadOnly(mnt, true); err != nil {
+					panic(fmt.Sprintf("failed to restore mount at %q back to readonly: %v", submount.Destination, err))
+				}
+			}()
+		}
 	}
 
 	if err := c.mountTmpVFS2(ctx, conf, creds, mns); err != nil {
@@ -256,7 +275,7 @@ func (c *containerMounter) prepareMountsVFS2() ([]mountAndFD, error) {
 	return mounts, nil
 }
 
-func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *mountAndFD) error {
+func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *mountAndFD) (*vfs.Mount, error) {
 	root := mns.Root()
 	defer root.DecRef(ctx)
 	target := &vfs.PathOperation{
@@ -266,21 +285,22 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *config.C
 	}
 	fsName, opts, err := c.getMountNameAndOptionsVFS2(conf, submount)
 	if err != nil {
-		return fmt.Errorf("mountOptions failed: %w", err)
+		return nil, fmt.Errorf("mountOptions failed: %w", err)
 	}
 	if len(fsName) == 0 {
 		// Filesystem is not supported (e.g. cgroup), just skip it.
-		return nil
+		return nil, nil
 	}
 
 	if err := c.k.VFS().MkdirAllAt(ctx, submount.Destination, root, creds, &vfs.MkdirOptions{Mode: 0777, ForSyntheticMountpoint: true}); err != nil {
-		return err
+		return nil, err
 	}
-	if err := c.k.VFS().MountAt(ctx, creds, "", target, fsName, opts); err != nil {
-		return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts)
+	mnt, err := c.k.VFS().MountAt(ctx, creds, "", target, fsName, opts)
+	if err != nil {
+		return nil, fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts)
 	}
 	log.Infof("Mounted %q to %q type: %s, internal-options: %q", submount.Source, submount.Destination, submount.Type, opts.GetFilesystemOptions.Data)
-	return nil
+	return mnt, nil
 }
 
 // getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values
@@ -407,7 +427,8 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *config.Config
 			// another user. This is normally done for /tmp.
 			Options: []string{"mode=01777"},
 		}
-		return c.mountSubmountVFS2(ctx, conf, mns, creds, &mountAndFD{Mount: tmpMount})
+		_, err := c.mountSubmountVFS2(ctx, conf, mns, creds, &mountAndFD{Mount: tmpMount})
+		return err
 
 	case syserror.ENOTDIR:
 		// Not a dir?! Let it be.
@@ -458,25 +479,25 @@ func (c *containerMounter) mountSharedMasterVFS2(ctx context.Context, conf *conf
 
 // mountSharedSubmount binds mount to a previously mounted volume that is shared
 // among containers in the same pod.
-func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials, mount specs.Mount, source *mountHint) error {
+func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *config.Config, mns *vfs.MountNamespace, creds *auth.Credentials, mount specs.Mount, source *mountHint) (*vfs.Mount, error) {
 	if err := source.checkCompatible(mount); err != nil {
-		return err
+		return nil, err
 	}
 
 	_, opts, err := c.getMountNameAndOptionsVFS2(conf, &mountAndFD{Mount: mount})
 	if err != nil {
-		return err
+		return nil, err
 	}
 	newMnt, err := c.k.VFS().NewDisconnectedMount(source.vfsMount.Filesystem(), source.vfsMount.Root(), opts)
 	if err != nil {
-		return err
+		return nil, err
 	}
 	defer newMnt.DecRef(ctx)
 
 	root := mns.Root()
 	defer root.DecRef(ctx)
 	if err := c.k.VFS().MkdirAllAt(ctx, mount.Destination, root, creds, &vfs.MkdirOptions{Mode: 0777, ForSyntheticMountpoint: true}); err != nil {
-		return err
+		return nil, err
 	}
 
 	target := &vfs.PathOperation{
@@ -485,8 +506,8 @@ func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *co
 		Path:  fspath.Parse(mount.Destination),
 	}
 	if err := c.k.VFS().ConnectMountAt(ctx, creds, newMnt, target); err != nil {
-		return err
+		return nil, err
 	}
 	log.Infof("Mounted %q type shared bind to %q", mount.Destination, source.name)
-	return nil
+	return newMnt, nil
 }