summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry')
-rw-r--r--pkg/sentry/control/proc.go6
-rw-r--r--pkg/sentry/fs/host/BUILD1
-rw-r--r--pkg/sentry/fs/host/control.go2
-rw-r--r--pkg/sentry/fs/host/file.go10
-rw-r--r--pkg/sentry/fs/host/inode_test.go3
-rw-r--r--pkg/sentry/fs/host/wait_test.go3
-rw-r--r--pkg/sentry/fsimpl/devtmpfs/devtmpfs.go5
-rw-r--r--pkg/sentry/fsimpl/ext/BUILD1
-rw-r--r--pkg/sentry/fsimpl/ext/ext.go12
-rw-r--r--pkg/sentry/fsimpl/ext/filesystem.go12
-rw-r--r--pkg/sentry/fsimpl/gofer/BUILD1
-rw-r--r--pkg/sentry/fsimpl/gofer/filesystem.go8
-rw-r--r--pkg/sentry/fsimpl/gofer/gofer.go7
-rw-r--r--pkg/sentry/fsimpl/host/host.go125
-rw-r--r--pkg/sentry/fsimpl/kernfs/BUILD1
-rw-r--r--pkg/sentry/fsimpl/kernfs/filesystem.go13
-rw-r--r--pkg/sentry/fsimpl/kernfs/kernfs.go7
-rw-r--r--pkg/sentry/fsimpl/kernfs/kernfs_test.go8
-rw-r--r--pkg/sentry/fsimpl/proc/filesystem.go11
-rw-r--r--pkg/sentry/fsimpl/sys/sys.go9
-rw-r--r--pkg/sentry/fsimpl/tmpfs/BUILD1
-rw-r--r--pkg/sentry/fsimpl/tmpfs/filesystem.go8
-rw-r--r--pkg/sentry/fsimpl/tmpfs/tmpfs.go7
-rw-r--r--pkg/sentry/kernel/task.go12
-rw-r--r--pkg/sentry/socket/netfilter/BUILD1
-rw-r--r--pkg/sentry/socket/netfilter/netfilter.go7
-rw-r--r--pkg/sentry/socket/netfilter/owner_matcher.go128
-rw-r--r--pkg/sentry/socket/netstack/provider.go6
-rw-r--r--pkg/sentry/vfs/BUILD1
-rw-r--r--pkg/sentry/vfs/anonfs.go22
-rw-r--r--pkg/sentry/vfs/filesystem.go19
-rw-r--r--pkg/sentry/vfs/filesystem_type.go3
-rw-r--r--pkg/sentry/vfs/mount.go7
-rw-r--r--pkg/sentry/vfs/pathname.go43
-rw-r--r--pkg/sentry/vfs/vfs.go46
35 files changed, 474 insertions, 82 deletions
diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go
index 5457ba5e7..b51fb3959 100644
--- a/pkg/sentry/control/proc.go
+++ b/pkg/sentry/control/proc.go
@@ -224,8 +224,6 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
}
}
- mounter := fs.FileOwnerFromContext(ctx)
-
// TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2.
var ttyFile *fs.File
for appFD, hostFile := range args.FilePayload.Files {
@@ -235,7 +233,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
// Import the file as a host TTY file.
if ttyFile == nil {
var err error
- appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, true /* isTTY */)
+ appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), true /* isTTY */)
if err != nil {
return nil, 0, nil, err
}
@@ -254,7 +252,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
} else {
// Import the file as a regular host file.
var err error
- appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, false /* isTTY */)
+ appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), false /* isTTY */)
if err != nil {
return nil, 0, nil, err
}
diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD
index 011625c80..aabce6cc9 100644
--- a/pkg/sentry/fs/host/BUILD
+++ b/pkg/sentry/fs/host/BUILD
@@ -71,7 +71,6 @@ go_test(
"//pkg/fd",
"//pkg/fdnotifier",
"//pkg/sentry/contexttest",
- "//pkg/sentry/fs",
"//pkg/sentry/kernel/time",
"//pkg/sentry/socket",
"//pkg/sentry/socket/unix/transport",
diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go
index cd84e1337..52c0504b6 100644
--- a/pkg/sentry/fs/host/control.go
+++ b/pkg/sentry/fs/host/control.go
@@ -78,7 +78,7 @@ func fdsToFiles(ctx context.Context, fds []int) []*fs.File {
}
// Create the file backed by hostFD.
- file, err := NewFile(ctx, fd, fs.FileOwnerFromContext(ctx))
+ file, err := NewFile(ctx, fd)
if err != nil {
ctx.Warningf("Error creating file from host FD: %v", err)
break
diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go
index 034862694..3e48b8b2c 100644
--- a/pkg/sentry/fs/host/file.go
+++ b/pkg/sentry/fs/host/file.go
@@ -60,8 +60,8 @@ var _ fs.FileOperations = (*fileOperations)(nil)
// The returned File cannot be saved, since there is no guarantee that the same
// FD will exist or represent the same file at time of restore. If such a
// guarantee does exist, use ImportFile instead.
-func NewFile(ctx context.Context, fd int, mounter fs.FileOwner) (*fs.File, error) {
- return newFileFromDonatedFD(ctx, fd, mounter, false, false)
+func NewFile(ctx context.Context, fd int) (*fs.File, error) {
+ return newFileFromDonatedFD(ctx, fd, false, false)
}
// ImportFile creates a new File backed by the provided host file descriptor.
@@ -71,13 +71,13 @@ func NewFile(ctx context.Context, fd int, mounter fs.FileOwner) (*fs.File, error
// If the returned file is saved, it will be restored by re-importing the FD
// originally passed to ImportFile. It is the restorer's responsibility to
// ensure that the FD represents the same file.
-func ImportFile(ctx context.Context, fd int, mounter fs.FileOwner, isTTY bool) (*fs.File, error) {
- return newFileFromDonatedFD(ctx, fd, mounter, true, isTTY)
+func ImportFile(ctx context.Context, fd int, isTTY bool) (*fs.File, error) {
+ return newFileFromDonatedFD(ctx, fd, true, isTTY)
}
// newFileFromDonatedFD returns an fs.File from a donated FD. If the FD is
// saveable, then saveable is true.
-func newFileFromDonatedFD(ctx context.Context, donated int, mounter fs.FileOwner, saveable, isTTY bool) (*fs.File, error) {
+func newFileFromDonatedFD(ctx context.Context, donated int, saveable, isTTY bool) (*fs.File, error) {
var s syscall.Stat_t
if err := syscall.Fstat(donated, &s); err != nil {
return nil, err
diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go
index 4c374681c..c507f57eb 100644
--- a/pkg/sentry/fs/host/inode_test.go
+++ b/pkg/sentry/fs/host/inode_test.go
@@ -19,7 +19,6 @@ import (
"testing"
"gvisor.dev/gvisor/pkg/sentry/contexttest"
- "gvisor.dev/gvisor/pkg/sentry/fs"
)
// TestCloseFD verifies fds will be closed.
@@ -33,7 +32,7 @@ func TestCloseFD(t *testing.T) {
// Use the write-end because we will detect if it's closed on the read end.
ctx := contexttest.Context(t)
- file, err := NewFile(ctx, p[1], fs.RootOwner)
+ file, err := NewFile(ctx, p[1])
if err != nil {
t.Fatalf("Failed to create File: %v", err)
}
diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go
index d49c3a635..ce397a5e3 100644
--- a/pkg/sentry/fs/host/wait_test.go
+++ b/pkg/sentry/fs/host/wait_test.go
@@ -20,7 +20,6 @@ import (
"time"
"gvisor.dev/gvisor/pkg/sentry/contexttest"
- "gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/waiter"
)
@@ -34,7 +33,7 @@ func TestWait(t *testing.T) {
defer syscall.Close(fds[1])
ctx := contexttest.Context(t)
- file, err := NewFile(ctx, fds[0], fs.RootOwner)
+ file, err := NewFile(ctx, fds[0])
if err != nil {
syscall.Close(fds[0])
t.Fatalf("NewFile failed: %v", err)
diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go
index abd4f24e7..64f1b142c 100644
--- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go
+++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go
@@ -42,6 +42,11 @@ type FilesystemType struct {
root *vfs.Dentry
}
+// Name implements vfs.FilesystemType.Name.
+func (*FilesystemType) Name() string {
+ return Name
+}
+
// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
func (fst *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
fst.initOnce.Do(func() {
diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD
index 6f78f478f..d83d75b3d 100644
--- a/pkg/sentry/fsimpl/ext/BUILD
+++ b/pkg/sentry/fsimpl/ext/BUILD
@@ -45,6 +45,7 @@ go_library(
"//pkg/sentry/fsimpl/ext/disklayout",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/memmap",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sentry/syscalls/linux",
"//pkg/sentry/vfs",
"//pkg/sync",
diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go
index 373d23b74..7176af6d1 100644
--- a/pkg/sentry/fsimpl/ext/ext.go
+++ b/pkg/sentry/fsimpl/ext/ext.go
@@ -30,6 +30,9 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
+// Name is the name of this filesystem.
+const Name = "ext"
+
// FilesystemType implements vfs.FilesystemType.
type FilesystemType struct{}
@@ -91,8 +94,13 @@ func isCompatible(sb disklayout.SuperBlock) bool {
return true
}
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+ return Name
+}
+
// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
-func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
// TODO(b/134676337): Ensure that the user is mounting readonly. If not,
// EACCESS should be returned according to mount(2). Filesystem independent
// flags (like readonly) are currently not available in pkg/sentry/vfs.
@@ -103,7 +111,7 @@ func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile
}
fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)}
- fs.vfsfs.Init(vfsObj, &fs)
+ fs.vfsfs.Init(vfsObj, &fsType, &fs)
fs.sb, err = readSuperBlock(dev)
if err != nil {
return nil, nil, err
diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go
index 8497be615..48eaccdbc 100644
--- a/pkg/sentry/fsimpl/ext/filesystem.go
+++ b/pkg/sentry/fsimpl/ext/filesystem.go
@@ -23,6 +23,7 @@ import (
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
@@ -463,6 +464,17 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
return syserror.EROFS
}
+// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+ _, _, err := fs.walk(rp, false)
+ if err != nil {
+ return nil, err
+ }
+
+ // TODO(b/134676337): Support sockets.
+ return nil, syserror.ECONNREFUSED
+}
+
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
_, _, err := fs.walk(rp, false)
diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD
index 4ba76a1e8..d15a36709 100644
--- a/pkg/sentry/fsimpl/gofer/BUILD
+++ b/pkg/sentry/fsimpl/gofer/BUILD
@@ -46,6 +46,7 @@ go_library(
"//pkg/sentry/memmap",
"//pkg/sentry/pgalloc",
"//pkg/sentry/platform",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sentry/usage",
"//pkg/sentry/vfs",
"//pkg/syserror",
diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go
index 1e43df9ec..269624362 100644
--- a/pkg/sentry/fsimpl/gofer/filesystem.go
+++ b/pkg/sentry/fsimpl/gofer/filesystem.go
@@ -22,6 +22,7 @@ import (
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -1059,6 +1060,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
return fs.unlinkAt(ctx, rp, false /* dir */)
}
+// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+//
+// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt.
+func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+ return nil, syserror.ECONNREFUSED
+}
+
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
var ds *[]*dentry
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index cf276a417..8e41b6b1c 100644
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -199,6 +199,11 @@ const (
InteropModeShared
)
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+ return Name
+}
+
// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
mfp := pgalloc.MemoryFileProviderFromContext(ctx)
@@ -374,7 +379,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
dentries: make(map[*dentry]struct{}),
specialFileFDs: make(map[*specialFileFD]struct{}),
}
- fs.vfsfs.Init(vfsObj, fs)
+ fs.vfsfs.Init(vfsObj, &fstype, fs)
// Construct the root dentry.
root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr)
diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go
index a54985ef5..7d9dcd4c9 100644
--- a/pkg/sentry/fsimpl/host/host.go
+++ b/pkg/sentry/fsimpl/host/host.go
@@ -38,6 +38,19 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
+// filesystemType implements vfs.FilesystemType.
+type filesystemType struct{}
+
+// GetFilesystem implements FilesystemType.GetFilesystem.
+func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+ panic("cannot instaniate a host filesystem")
+}
+
+// Name implements FilesystemType.Name.
+func (filesystemType) Name() string {
+ return "none"
+}
+
// filesystem implements vfs.FilesystemImpl.
type filesystem struct {
kernfs.Filesystem
@@ -46,7 +59,7 @@ type filesystem struct {
// NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD.
func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) {
fs := &filesystem{}
- fs.Init(vfsObj)
+ fs.Init(vfsObj, &filesystemType{})
vfsfs := fs.VFSFilesystem()
// NewDisconnectedMount will take an additional reference on vfsfs.
defer vfsfs.DecRef()
@@ -54,7 +67,7 @@ func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) {
}
// ImportFD sets up and returns a vfs.FileDescription from a donated fd.
-func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID, isTTY bool) (*vfs.FileDescription, error) {
+func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) {
fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem)
if !ok {
return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl())
@@ -78,8 +91,6 @@ func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID
canMap: canMap(uint32(fileType)),
ino: fs.NextIno(),
mode: fileMode,
- uid: ownerUID,
- gid: ownerGID,
// For simplicity, set offset 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.
@@ -135,17 +146,20 @@ type inode struct {
// This field is initialized at creation time and is immutable.
ino uint64
- // TODO(gvisor.dev/issue/1672): protect mode, uid, and gid with mutex.
+ // modeMu protects mode.
+ modeMu sync.Mutex
- // mode is the file mode of this inode. Note that this value may become out
- // of date if the mode is changed on the host, e.g. with chmod.
+ // mode is a cached version of the file mode on the host. Note that it may
+ // become out of date if the mode is changed on the host, e.g. with chmod.
+ //
+ // Generally, it is better to retrieve the mode from the host through an
+ // fstat syscall. We only use this value in inode.Mode(), which cannot
+ // return an error, if the syscall to host fails.
+ //
+ // FIXME(b/152294168): Plumb error into Inode.Mode() return value so we
+ // can get rid of this.
mode linux.FileMode
- // uid and gid of the file owner. Note that these refer to the owner of the
- // file created on import, not the fd on the host.
- uid auth.KUID
- gid auth.KGID
-
// offsetMu protects offset.
offsetMu sync.Mutex
@@ -168,12 +182,35 @@ func fileFlagsFromHostFD(fd int) (int, error) {
// CheckPermissions implements kernfs.Inode.
func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error {
- return vfs.GenericCheckPermissions(creds, ats, i.mode, i.uid, i.gid)
+ mode, uid, gid, err := i.getPermissions()
+ if err != nil {
+ return err
+ }
+ return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid)
}
// Mode implements kernfs.Inode.
func (i *inode) Mode() linux.FileMode {
- return i.mode
+ mode, _, _, err := i.getPermissions()
+ if err != nil {
+ return i.mode
+ }
+
+ return linux.FileMode(mode)
+}
+
+func (i *inode) getPermissions() (linux.FileMode, auth.KUID, auth.KGID, error) {
+ // Retrieve metadata.
+ var s syscall.Stat_t
+ if err := syscall.Fstat(i.hostFD, &s); err != nil {
+ return 0, 0, 0, err
+ }
+
+ // Update cached mode.
+ i.modeMu.Lock()
+ i.mode = linux.FileMode(s.Mode)
+ i.modeMu.Unlock()
+ return linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid), nil
}
// Stat implements kernfs.Inode.
@@ -213,45 +250,51 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro
ls.Attributes = s.Attributes
ls.AttributesMask = s.Attributes_mask
- if mask|linux.STATX_TYPE != 0 {
+ if mask&linux.STATX_TYPE != 0 {
ls.Mode |= s.Mode & linux.S_IFMT
}
- if mask|linux.STATX_MODE != 0 {
+ if mask&linux.STATX_MODE != 0 {
ls.Mode |= s.Mode &^ linux.S_IFMT
}
- if mask|linux.STATX_NLINK != 0 {
+ if mask&linux.STATX_NLINK != 0 {
ls.Nlink = s.Nlink
}
- if mask|linux.STATX_ATIME != 0 {
+ if mask&linux.STATX_UID != 0 {
+ ls.UID = s.Uid
+ }
+ if mask&linux.STATX_GID != 0 {
+ ls.GID = s.Gid
+ }
+ if mask&linux.STATX_ATIME != 0 {
ls.Atime = unixToLinuxStatxTimestamp(s.Atime)
}
- if mask|linux.STATX_BTIME != 0 {
+ if mask&linux.STATX_BTIME != 0 {
ls.Btime = unixToLinuxStatxTimestamp(s.Btime)
}
- if mask|linux.STATX_CTIME != 0 {
+ if mask&linux.STATX_CTIME != 0 {
ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime)
}
- if mask|linux.STATX_MTIME != 0 {
+ if mask&linux.STATX_MTIME != 0 {
ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime)
}
- if mask|linux.STATX_SIZE != 0 {
+ if mask&linux.STATX_SIZE != 0 {
ls.Size = s.Size
}
- if mask|linux.STATX_BLOCKS != 0 {
+ if mask&linux.STATX_BLOCKS != 0 {
ls.Blocks = s.Blocks
}
- // Use our own internal inode number and file owner.
- if mask|linux.STATX_INO != 0 {
+ // Use our own internal inode number.
+ if mask&linux.STATX_INO != 0 {
ls.Ino = i.ino
}
- if mask|linux.STATX_UID != 0 {
- ls.UID = uint32(i.uid)
- }
- if mask|linux.STATX_GID != 0 {
- ls.GID = uint32(i.gid)
- }
+ // Update cached mode.
+ if (mask&linux.STATX_TYPE != 0) && (mask&linux.STATX_MODE != 0) {
+ i.modeMu.Lock()
+ i.mode = linux.FileMode(s.Mode)
+ i.modeMu.Unlock()
+ }
return ls, nil
}
@@ -274,6 +317,8 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) {
Mask: linux.STATX_BASIC_STATS,
Blksize: uint32(s.Blksize),
Nlink: uint32(s.Nlink),
+ UID: s.Uid,
+ GID: s.Gid,
Mode: uint16(s.Mode),
Size: uint64(s.Size),
Blocks: uint64(s.Blocks),
@@ -282,15 +327,13 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) {
Mtime: timespecToStatxTimestamp(s.Mtim),
}
- // Use our own internal inode number and file owner.
+ // Use our own internal inode number.
//
// TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well.
// If we use the device number from the host, it may collide with another
// sentry-internal device number. We handle device/inode numbers without
// relying on the host to prevent collisions.
ls.Ino = i.ino
- ls.UID = uint32(i.uid)
- ls.GID = uint32(i.gid)
return ls, nil
}
@@ -306,7 +349,11 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre
if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 {
return syserror.EPERM
}
- if err := vfs.CheckSetStat(ctx, creds, &s, i.Mode(), i.uid, i.gid); err != nil {
+ mode, uid, gid, err := i.getPermissions()
+ if err != nil {
+ return err
+ }
+ if err := vfs.CheckSetStat(ctx, creds, &s, mode.Permissions(), uid, gid); err != nil {
return err
}
@@ -314,7 +361,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre
if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil {
return err
}
+ i.modeMu.Lock()
i.mode = linux.FileMode(s.Mode)
+ i.modeMu.Unlock()
}
if m&linux.STATX_SIZE != 0 {
if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil {
@@ -351,7 +400,11 @@ func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptio
}
func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) {
- fileType := i.mode.FileType()
+ mode, _, _, err := i.getPermissions()
+ if err != nil {
+ return nil, err
+ }
+ fileType := mode.FileType()
if fileType == syscall.S_IFSOCK {
if i.isTTY {
return nil, errors.New("cannot use host socket as TTY")
diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD
index e73f1f857..b3d6299d0 100644
--- a/pkg/sentry/fsimpl/kernfs/BUILD
+++ b/pkg/sentry/fsimpl/kernfs/BUILD
@@ -35,6 +35,7 @@ go_library(
"//pkg/refs",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/memmap",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sentry/vfs",
"//pkg/sync",
"//pkg/syserror",
diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go
index 31da8b511..a429fa23d 100644
--- a/pkg/sentry/fsimpl/kernfs/filesystem.go
+++ b/pkg/sentry/fsimpl/kernfs/filesystem.go
@@ -23,6 +23,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -728,6 +729,18 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
return nil
}
+// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+ fs.mu.RLock()
+ _, _, err := fs.walkExistingLocked(ctx, rp)
+ fs.mu.RUnlock()
+ fs.processDeferredDecRefs()
+ if err != nil {
+ return nil, err
+ }
+ return nil, syserror.ECONNREFUSED
+}
+
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
fs.mu.RLock()
diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go
index 794e38908..2cefef020 100644
--- a/pkg/sentry/fsimpl/kernfs/kernfs.go
+++ b/pkg/sentry/fsimpl/kernfs/kernfs.go
@@ -63,9 +63,6 @@ import (
"gvisor.dev/gvisor/pkg/sync"
)
-// FilesystemType implements vfs.FilesystemType.
-type FilesystemType struct{}
-
// Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory
// filesystem. Concrete implementations are expected to embed this in their own
// Filesystem type.
@@ -138,8 +135,8 @@ func (fs *Filesystem) processDeferredDecRefsLocked() {
// Init initializes a kernfs filesystem. This should be called from during
// vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding
// kernfs.
-func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem) {
- fs.vfsfs.Init(vfsObj, fs)
+func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem, fsType vfs.FilesystemType) {
+ fs.vfsfs.Init(vfsObj, fsType, fs)
}
// VFSFilesystem returns the generic vfs filesystem object.
diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go
index fb0d25ad7..465451f35 100644
--- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go
+++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go
@@ -187,9 +187,13 @@ func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, err
return nil, syserror.EPERM
}
-func (fst *fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+func (fsType) Name() string {
+ return "kernfs"
+}
+
+func (fst fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
fs := &filesystem{}
- fs.Init(vfsObj)
+ fs.Init(vfsObj, &fst)
root := fst.rootFn(creds, fs)
return fs.VFSFilesystem(), root.VFSDentry(), nil
}
diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go
index 5c19d5522..104fc9030 100644
--- a/pkg/sentry/fsimpl/proc/filesystem.go
+++ b/pkg/sentry/fsimpl/proc/filesystem.go
@@ -36,8 +36,13 @@ type FilesystemType struct{}
var _ vfs.FilesystemType = (*FilesystemType)(nil)
-// GetFilesystem implements vfs.FilesystemType.
-func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+ return Name
+}
+
+// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
+func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
k := kernel.KernelFromContext(ctx)
if k == nil {
return nil, nil, fmt.Errorf("procfs requires a kernel")
@@ -48,7 +53,7 @@ func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virtual
}
procfs := &kernfs.Filesystem{}
- procfs.VFSFilesystem().Init(vfsObj, procfs)
+ procfs.VFSFilesystem().Init(vfsObj, &ft, procfs)
var cgroups map[string]string
if opts.InternalData != nil {
diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go
index 7abfd62f2..5c617270e 100644
--- a/pkg/sentry/fsimpl/sys/sys.go
+++ b/pkg/sentry/fsimpl/sys/sys.go
@@ -39,10 +39,15 @@ type filesystem struct {
kernfs.Filesystem
}
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+ return Name
+}
+
// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
-func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
+func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
fs := &filesystem{}
- fs.Filesystem.Init(vfsObj)
+ fs.Filesystem.Init(vfsObj, &fsType)
k := kernel.KernelFromContext(ctx)
maxCPUCores := k.ApplicationCores()
defaultSysDirMode := linux.FileMode(0755)
diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD
index 57abd5583..6ea35affb 100644
--- a/pkg/sentry/fsimpl/tmpfs/BUILD
+++ b/pkg/sentry/fsimpl/tmpfs/BUILD
@@ -46,6 +46,7 @@ go_library(
"//pkg/sentry/memmap",
"//pkg/sentry/pgalloc",
"//pkg/sentry/platform",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sentry/usage",
"//pkg/sentry/vfs",
"//pkg/sentry/vfs/lock",
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index 12cc64385..e678ecc37 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -656,6 +657,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
return nil
}
+// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+//
+// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt.
+func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) {
+ return nil, syserror.ECONNREFUSED
+}
+
// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt.
func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) {
fs.mu.RLock()
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index 2f9e6c876..b07b0dbae 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -63,6 +63,11 @@ type filesystem struct {
nextInoMinusOne uint64 // accessed using atomic memory operations
}
+// Name implements vfs.FilesystemType.Name.
+func (FilesystemType) Name() string {
+ return Name
+}
+
// GetFilesystem implements vfs.FilesystemType.GetFilesystem.
func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) {
memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx)
@@ -74,7 +79,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
memFile: memFileProvider.MemoryFile(),
clock: clock,
}
- fs.vfsfs.Init(vfsObj, &fs)
+ fs.vfsfs.Init(vfsObj, &fstype, &fs)
root := fs.newDentry(fs.newDirectory(creds, 01777))
return &fs.vfsfs, &root.vfsd, nil
}
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 8452ddf5b..d6546735e 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -863,3 +863,15 @@ func (t *Task) SetOOMScoreAdj(adj int32) error {
atomic.StoreInt32(&t.tg.oomScoreAdj, adj)
return nil
}
+
+// UID returns t's uid.
+// TODO(gvisor.dev/issue/170): This method is not namespaced yet.
+func (t *Task) UID() uint32 {
+ return uint32(t.Credentials().EffectiveKUID)
+}
+
+// GID returns t's gid.
+// TODO(gvisor.dev/issue/170): This method is not namespaced yet.
+func (t *Task) GID() uint32 {
+ return uint32(t.Credentials().EffectiveKGID)
+}
diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD
index e801abeb8..721094bbf 100644
--- a/pkg/sentry/socket/netfilter/BUILD
+++ b/pkg/sentry/socket/netfilter/BUILD
@@ -7,6 +7,7 @@ go_library(
srcs = [
"extensions.go",
"netfilter.go",
+ "owner_matcher.go",
"targets.go",
"tcp_matcher.go",
"udp_matcher.go",
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index 55bcc3ace..878f81fd5 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -517,11 +517,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
}
// TODO(gvisor.dev/issue/170): Support other chains.
- // Since we only support modifying the INPUT chain and redirect for
- // PREROUTING chain right now, make sure all other chains point to
- // ACCEPT rules.
+ // Since we only support modifying the INPUT, PREROUTING and OUTPUT chain right now,
+ // make sure all other chains point to ACCEPT rules.
for hook, ruleIdx := range table.BuiltinChains {
- if hook != stack.Input && hook != stack.Prerouting {
+ if hook == stack.Forward || hook == stack.Postrouting {
if _, ok := table.Rules[ruleIdx].Target.(stack.AcceptTarget); !ok {
nflog("hook %d is unsupported.", hook)
return syserr.ErrInvalidArgument
diff --git a/pkg/sentry/socket/netfilter/owner_matcher.go b/pkg/sentry/socket/netfilter/owner_matcher.go
new file mode 100644
index 000000000..5949a7c29
--- /dev/null
+++ b/pkg/sentry/socket/netfilter/owner_matcher.go
@@ -0,0 +1,128 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package netfilter
+
+import (
+ "fmt"
+
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/binary"
+ "gvisor.dev/gvisor/pkg/tcpip/stack"
+ "gvisor.dev/gvisor/pkg/usermem"
+)
+
+const matcherNameOwner = "owner"
+
+func init() {
+ registerMatchMaker(ownerMarshaler{})
+}
+
+// ownerMarshaler implements matchMaker for owner matching.
+type ownerMarshaler struct{}
+
+// name implements matchMaker.name.
+func (ownerMarshaler) name() string {
+ return matcherNameOwner
+}
+
+// marshal implements matchMaker.marshal.
+func (ownerMarshaler) marshal(mr stack.Matcher) []byte {
+ matcher := mr.(*OwnerMatcher)
+ iptOwnerInfo := linux.IPTOwnerInfo{
+ UID: matcher.uid,
+ GID: matcher.gid,
+ }
+
+ // Support for UID match.
+ // TODO(gvisor.dev/issue/170): Need to support gid match.
+ if matcher.matchUID {
+ iptOwnerInfo.Match = linux.XT_OWNER_UID
+ } else if matcher.matchGID {
+ panic("GID match is not supported.")
+ } else {
+ panic("UID match is not set.")
+ }
+
+ buf := make([]byte, 0, linux.SizeOfIPTOwnerInfo)
+ return marshalEntryMatch(matcherNameOwner, binary.Marshal(buf, usermem.ByteOrder, iptOwnerInfo))
+}
+
+// unmarshal implements matchMaker.unmarshal.
+func (ownerMarshaler) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Matcher, error) {
+ if len(buf) < linux.SizeOfIPTOwnerInfo {
+ return nil, fmt.Errorf("buf has insufficient size for owner match: %d", len(buf))
+ }
+
+ // For alignment reasons, the match's total size may
+ // exceed what's strictly necessary to hold matchData.
+ var matchData linux.IPTOwnerInfo
+ binary.Unmarshal(buf[:linux.SizeOfIPTOwnerInfo], usermem.ByteOrder, &matchData)
+ nflog("parseMatchers: parsed IPTOwnerInfo: %+v", matchData)
+
+ if matchData.Invert != 0 {
+ return nil, fmt.Errorf("invert flag is not supported for owner match")
+ }
+
+ // Support for UID match.
+ // TODO(gvisor.dev/issue/170): Need to support gid match.
+ if matchData.Match&linux.XT_OWNER_UID != linux.XT_OWNER_UID {
+ return nil, fmt.Errorf("owner match is only supported for uid")
+ }
+
+ // Check Flags.
+ var owner OwnerMatcher
+ owner.uid = matchData.UID
+ owner.gid = matchData.GID
+ owner.matchUID = true
+
+ return &owner, nil
+}
+
+type OwnerMatcher struct {
+ uid uint32
+ gid uint32
+ matchUID bool
+ matchGID bool
+ invert uint8
+}
+
+// Name implements Matcher.Name.
+func (*OwnerMatcher) Name() string {
+ return matcherNameOwner
+}
+
+// Match implements Matcher.Match.
+func (om *OwnerMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) {
+ // Support only for OUTPUT chain.
+ // TODO(gvisor.dev/issue/170): Need to support for POSTROUTING chain also.
+ if hook != stack.Output {
+ return false, true
+ }
+
+ // If the packet owner is not set, drop the packet.
+ // Support for uid match.
+ // TODO(gvisor.dev/issue/170): Need to support gid match.
+ if pkt.Owner == nil || !om.matchUID {
+ return false, true
+ }
+
+ // TODO(gvisor.dev/issue/170): Need to add tests to verify
+ // drop rule when packet UID does not match owner matcher UID.
+ if pkt.Owner.UID() != om.uid {
+ return false, false
+ }
+
+ return true, false
+}
diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go
index 5f181f017..eb090e79b 100644
--- a/pkg/sentry/socket/netstack/provider.go
+++ b/pkg/sentry/socket/netstack/provider.go
@@ -126,6 +126,12 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*
ep, e = eps.Stack.NewRawEndpoint(transProto, p.netProto, wq, associated)
} else {
ep, e = eps.Stack.NewEndpoint(transProto, p.netProto, wq)
+
+ // Assign task to PacketOwner interface to get the UID and GID for
+ // iptables owner matching.
+ if e == nil {
+ ep.SetOwner(t)
+ }
}
if e != nil {
return nil, syserr.TranslateNetstackError(e)
diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD
index a2a06fc8f..bf4d27c7d 100644
--- a/pkg/sentry/vfs/BUILD
+++ b/pkg/sentry/vfs/BUILD
@@ -53,6 +53,7 @@ go_library(
"//pkg/sentry/kernel/auth",
"//pkg/sentry/limits",
"//pkg/sentry/memmap",
+ "//pkg/sentry/socket/unix/transport",
"//pkg/sync",
"//pkg/syserror",
"//pkg/usermem",
diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go
index a62e43589..d1f6dfb45 100644
--- a/pkg/sentry/vfs/anonfs.go
+++ b/pkg/sentry/vfs/anonfs.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
)
@@ -50,6 +51,19 @@ const (
anonFileGID = auth.RootKGID
)
+// anonFilesystemType implements FilesystemType.
+type anonFilesystemType struct{}
+
+// GetFilesystem implements FilesystemType.GetFilesystem.
+func (anonFilesystemType) GetFilesystem(context.Context, *VirtualFilesystem, *auth.Credentials, string, GetFilesystemOptions) (*Filesystem, *Dentry, error) {
+ panic("cannot instaniate an anon filesystem")
+}
+
+// Name implemenents FilesystemType.Name.
+func (anonFilesystemType) Name() string {
+ return "none"
+}
+
// anonFilesystem is the implementation of FilesystemImpl that backs
// VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry().
//
@@ -222,6 +236,14 @@ func (fs *anonFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) error
return syserror.EPERM
}
+// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt.
+func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error) {
+ if !rp.Final() {
+ return nil, syserror.ENOTDIR
+ }
+ return nil, syserror.ECONNREFUSED
+}
+
// ListxattrAt implements FilesystemImpl.ListxattrAt.
func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) {
if !rp.Done() {
diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go
index 332decce6..cd34782ff 100644
--- a/pkg/sentry/vfs/filesystem.go
+++ b/pkg/sentry/vfs/filesystem.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
)
// A Filesystem is a tree of nodes represented by Dentries, which forms part of
@@ -41,21 +42,30 @@ type Filesystem struct {
// immutable.
vfs *VirtualFilesystem
+ // fsType is the FilesystemType of this Filesystem.
+ fsType FilesystemType
+
// impl is the FilesystemImpl associated with this Filesystem. impl is
// immutable. This should be the last field in Dentry.
impl FilesystemImpl
}
// Init must be called before first use of fs.
-func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, impl FilesystemImpl) {
+func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, fsType FilesystemType, impl FilesystemImpl) {
fs.refs = 1
fs.vfs = vfsObj
+ fs.fsType = fsType
fs.impl = impl
vfsObj.filesystemsMu.Lock()
vfsObj.filesystems[fs] = struct{}{}
vfsObj.filesystemsMu.Unlock()
}
+// FilesystemType returns the FilesystemType for this Filesystem.
+func (fs *Filesystem) FilesystemType() FilesystemType {
+ return fs.fsType
+}
+
// VirtualFilesystem returns the containing VirtualFilesystem.
func (fs *Filesystem) VirtualFilesystem() *VirtualFilesystem {
return fs.vfs
@@ -460,6 +470,11 @@ type FilesystemImpl interface {
// RemovexattrAt returns ENOTSUP.
RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error
+ // BoundEndpointAt returns the Unix socket endpoint bound at the path rp.
+ //
+ // - If a non-socket file exists at rp, then BoundEndpointAt returns ECONNREFUSED.
+ BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error)
+
// PrependPath prepends a path from vd to vd.Mount().Root() to b.
//
// If vfsroot.Ok(), it is the contextual VFS root; if it is encountered
@@ -482,7 +497,7 @@ type FilesystemImpl interface {
// Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl.
PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error
- // TODO: inotify_add_watch(); bind()
+ // TODO: inotify_add_watch()
}
// PrependPathAtVFSRootError is returned by implementations of
diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go
index bb9cada81..f2298f7f6 100644
--- a/pkg/sentry/vfs/filesystem_type.go
+++ b/pkg/sentry/vfs/filesystem_type.go
@@ -30,6 +30,9 @@ type FilesystemType interface {
// along with its mount root. A reference is taken on the returned
// Filesystem and Dentry.
GetFilesystem(ctx context.Context, vfsObj *VirtualFilesystem, creds *auth.Credentials, source string, opts GetFilesystemOptions) (*Filesystem, *Dentry, error)
+
+ // Name returns the name of this FilesystemType.
+ Name() string
}
// GetFilesystemOptions contains options to FilesystemType.GetFilesystem.
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index 05f6233f9..4b68cabda 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -24,6 +24,9 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
)
+// lastMountID is used to allocate mount ids. Must be accessed atomically.
+var lastMountID uint64
+
// A Mount is a replacement of a Dentry (Mount.key.point) from one Filesystem
// (Mount.key.parent.fs) with a Dentry (Mount.root) from another Filesystem
// (Mount.fs), which applies to path resolution in the context of a particular
@@ -48,6 +51,9 @@ type Mount struct {
fs *Filesystem
root *Dentry
+ // ID is the immutable mount ID.
+ ID uint64
+
// key is protected by VirtualFilesystem.mountMu and
// VirtualFilesystem.mounts.seq, and may be nil. References are held on
// key.parent and key.point if they are not nil.
@@ -87,6 +93,7 @@ type Mount struct {
func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount {
mnt := &Mount{
+ ID: atomic.AddUint64(&lastMountID, 1),
vfs: vfs,
fs: fs,
root: root,
diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go
index b318c681a..f21a88034 100644
--- a/pkg/sentry/vfs/pathname.go
+++ b/pkg/sentry/vfs/pathname.go
@@ -90,6 +90,49 @@ loop:
return b.String(), nil
}
+// PathnameReachable returns an absolute pathname to vd, consistent with
+// Linux's __d_path() (as used by seq_path_root()). If vfsroot.Ok() and vd is
+// not reachable from vfsroot, such that seq_path_root() would return SEQ_SKIP
+// (causing the entire containing entry to be skipped), PathnameReachable
+// returns ("", nil).
+func (vfs *VirtualFilesystem) PathnameReachable(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) {
+ b := getFSPathBuilder()
+ defer putFSPathBuilder(b)
+ haveRef := false
+ defer func() {
+ if haveRef {
+ vd.DecRef()
+ }
+ }()
+loop:
+ for {
+ err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b)
+ switch err.(type) {
+ case nil:
+ if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry {
+ break loop
+ }
+ nextVD := vfs.getMountpointAt(vd.mount, vfsroot)
+ if !nextVD.Ok() {
+ return "", nil
+ }
+ if haveRef {
+ vd.DecRef()
+ }
+ vd = nextVD
+ haveRef = true
+ case PrependPathAtVFSRootError:
+ break loop
+ case PrependPathAtNonMountRootError, PrependPathSyntheticError:
+ return "", nil
+ default:
+ return "", err
+ }
+ }
+ b.PrependByte('/')
+ return b.String(), nil
+}
+
// PathnameForGetcwd returns an absolute pathname to vd, consistent with
// Linux's sys_getcwd().
func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) {
diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go
index 03d1fb943..720b90d8f 100644
--- a/pkg/sentry/vfs/vfs.go
+++ b/pkg/sentry/vfs/vfs.go
@@ -38,6 +38,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -133,7 +134,7 @@ func (vfs *VirtualFilesystem) Init() error {
anonfs := anonFilesystem{
devMinor: anonfsDevMinor,
}
- anonfs.vfsfs.Init(vfs, &anonfs)
+ anonfs.vfsfs.Init(vfs, &anonFilesystemType{}, &anonfs)
defer anonfs.vfsfs.DecRef()
anonMount, err := vfs.NewDisconnectedMount(&anonfs.vfsfs, nil, &MountOptions{})
if err != nil {
@@ -230,7 +231,7 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -271,7 +272,7 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -307,7 +308,7 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -340,7 +341,7 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -350,6 +351,33 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia
}
}
+// BoundEndpointAt gets the bound endpoint at the given path, if one exists.
+func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (transport.BoundEndpoint, error) {
+ if !pop.Path.Begin.Ok() {
+ if pop.Path.Absolute {
+ return nil, syserror.ECONNREFUSED
+ }
+ return nil, syserror.ENOENT
+ }
+ rp := vfs.getResolvingPath(creds, pop)
+ for {
+ bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp)
+ if err == nil {
+ vfs.putResolvingPath(rp)
+ return bep, nil
+ }
+ if checkInvariants {
+ if rp.canHandleError(err) && rp.Done() {
+ panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
+ }
+ }
+ if !rp.handleError(err) {
+ vfs.putResolvingPath(rp)
+ return nil, err
+ }
+ }
+}
+
// OpenAt returns a FileDescription providing access to the file at the given
// path. A reference is taken on the returned FileDescription.
func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) {
@@ -494,7 +522,7 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -527,7 +555,7 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -608,7 +636,7 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {
@@ -640,7 +668,7 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti
}
if checkInvariants {
if rp.canHandleError(err) && rp.Done() {
- panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err))
+ panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err))
}
}
if !rp.handleError(err) {