summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/vfs
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/vfs')
-rw-r--r--pkg/sentry/vfs/BUILD2
-rw-r--r--pkg/sentry/vfs/context.go7
-rw-r--r--pkg/sentry/vfs/dentry.go4
-rw-r--r--pkg/sentry/vfs/device.go3
-rw-r--r--pkg/sentry/vfs/epoll.go22
-rw-r--r--pkg/sentry/vfs/file_description.go10
-rw-r--r--pkg/sentry/vfs/file_description_impl_util_test.go10
-rw-r--r--pkg/sentry/vfs/filesystem.go2
-rw-r--r--pkg/sentry/vfs/filesystem_type.go1
-rw-r--r--pkg/sentry/vfs/mount.go14
-rw-r--r--pkg/sentry/vfs/mount_unsafe.go20
-rw-r--r--pkg/sentry/vfs/options.go7
-rw-r--r--pkg/sentry/vfs/permissions.go19
-rw-r--r--pkg/sentry/vfs/resolving_path.go2
-rw-r--r--pkg/sentry/vfs/vfs.go53
15 files changed, 112 insertions, 64 deletions
diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD
index 14b39eb9d..07c8383e6 100644
--- a/pkg/sentry/vfs/BUILD
+++ b/pkg/sentry/vfs/BUILD
@@ -43,6 +43,8 @@ go_library(
"//pkg/abi/linux",
"//pkg/context",
"//pkg/fspath",
+ "//pkg/gohacks",
+ "//pkg/log",
"//pkg/sentry/arch",
"//pkg/sentry/fs/lock",
"//pkg/sentry/kernel/auth",
diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go
index d97362b9a..82781e6d3 100644
--- a/pkg/sentry/vfs/context.go
+++ b/pkg/sentry/vfs/context.go
@@ -29,9 +29,10 @@ const (
CtxRoot
)
-// MountNamespaceFromContext returns the MountNamespace used by ctx. It does
-// not take a reference on the returned MountNamespace. If ctx is not
-// associated with a MountNamespace, MountNamespaceFromContext returns nil.
+// MountNamespaceFromContext returns the MountNamespace used by ctx. If ctx is
+// not associated with a MountNamespace, MountNamespaceFromContext returns nil.
+//
+// A reference is taken on the returned MountNamespace.
func MountNamespaceFromContext(ctx context.Context) *MountNamespace {
if v := ctx.Value(CtxMountNamespace); v != nil {
return v.(*MountNamespace)
diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go
index 486a76475..35b208721 100644
--- a/pkg/sentry/vfs/dentry.go
+++ b/pkg/sentry/vfs/dentry.go
@@ -71,6 +71,8 @@ import (
// lifetime. Dentry reference counts only indicate the extent to which VFS
// requires Dentries to exist; Filesystems may elect to cache or discard
// Dentries with zero references.
+//
+// +stateify savable
type Dentry struct {
// parent is this Dentry's parent in this Filesystem. If this Dentry is
// independent, parent is nil.
@@ -89,7 +91,7 @@ type Dentry struct {
children map[string]*Dentry
// mu synchronizes disowning and mounting over this Dentry.
- mu sync.Mutex
+ mu sync.Mutex `state:"nosave"`
// impl is the DentryImpl associated with this Dentry. impl is immutable.
// This should be the last field in Dentry.
diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go
index 3af2aa58d..bda5576fa 100644
--- a/pkg/sentry/vfs/device.go
+++ b/pkg/sentry/vfs/device.go
@@ -56,6 +56,7 @@ type Device interface {
Open(ctx context.Context, mnt *Mount, d *Dentry, opts OpenOptions) (*FileDescription, error)
}
+// +stateify savable
type registeredDevice struct {
dev Device
opts RegisterDeviceOptions
@@ -63,6 +64,8 @@ type registeredDevice struct {
// RegisterDeviceOptions contains options to
// VirtualFilesystem.RegisterDevice().
+//
+// +stateify savable
type RegisterDeviceOptions struct {
// GroupName is the name shown for this device registration in
// /proc/devices. If GroupName is empty, this registration will not be
diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go
index 7c83f9a5a..3da45d744 100644
--- a/pkg/sentry/vfs/epoll.go
+++ b/pkg/sentry/vfs/epoll.go
@@ -85,8 +85,8 @@ type epollInterest struct {
ready bool
epollInterestEntry
- // userData is the epoll_data_t associated with this epollInterest.
- // userData is protected by epoll.mu.
+ // userData is the struct epoll_event::data associated with this
+ // epollInterest. userData is protected by epoll.mu.
userData [2]int32
}
@@ -157,7 +157,7 @@ func (ep *EpollInstance) Seek(ctx context.Context, offset int64, whence int32) (
// AddInterest implements the semantics of EPOLL_CTL_ADD.
//
// Preconditions: A reference must be held on file.
-func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error {
+func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event linux.EpollEvent) error {
// Check for cyclic polling if necessary.
subep, _ := file.impl.(*EpollInstance)
if subep != nil {
@@ -183,12 +183,12 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint
}
// Register interest in file.
- mask |= linux.EPOLLERR | linux.EPOLLRDHUP
+ mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP
epi := &epollInterest{
epoll: ep,
key: key,
mask: mask,
- userData: userData,
+ userData: event.Data,
}
ep.interest[key] = epi
wmask := waiter.EventMaskFromLinux(mask)
@@ -202,6 +202,9 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint
// Add epi to file.epolls so that it is removed when the last
// FileDescription reference is dropped.
file.epollMu.Lock()
+ if file.epolls == nil {
+ file.epolls = make(map[*epollInterest]struct{})
+ }
file.epolls[epi] = struct{}{}
file.epollMu.Unlock()
@@ -236,7 +239,7 @@ func (ep *EpollInstance) mightPollRecursive(ep2 *EpollInstance, remainingRecursi
// ModifyInterest implements the semantics of EPOLL_CTL_MOD.
//
// Preconditions: A reference must be held on file.
-func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error {
+func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, event linux.EpollEvent) error {
ep.interestMu.Lock()
defer ep.interestMu.Unlock()
@@ -250,13 +253,13 @@ func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, mask u
}
// Update epi for the next call to ep.ReadEvents().
+ mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP
ep.mu.Lock()
epi.mask = mask
- epi.userData = userData
+ epi.userData = event.Data
ep.mu.Unlock()
// Re-register with the new mask.
- mask |= linux.EPOLLERR | linux.EPOLLRDHUP
file.EventUnregister(&epi.waiter)
wmask := waiter.EventMaskFromLinux(mask)
file.EventRegister(&epi.waiter, wmask)
@@ -363,8 +366,7 @@ func (ep *EpollInstance) ReadEvents(events []linux.EpollEvent) int {
// Report ievents.
events[i] = linux.EpollEvent{
Events: ievents.ToLinux(),
- Fd: epi.userData[0],
- Data: epi.userData[1],
+ Data: epi.userData,
}
i++
if i == len(events) {
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index 5bac660c7..9a1ad630c 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -435,11 +435,11 @@ type Dirent struct {
// IterDirentsCallback receives Dirents from FileDescriptionImpl.IterDirents.
type IterDirentsCallback interface {
- // Handle handles the given iterated Dirent. It returns true if iteration
- // should continue, and false if FileDescriptionImpl.IterDirents should
- // terminate now and restart with the same Dirent the next time it is
- // called.
- Handle(dirent Dirent) bool
+ // Handle handles the given iterated Dirent. If Handle returns a non-nil
+ // error, FileDescriptionImpl.IterDirents must stop iteration and return
+ // the error; the next call to FileDescriptionImpl.IterDirents should
+ // restart with the same Dirent.
+ Handle(dirent Dirent) error
}
// OnClose is called when a file descriptor representing the FileDescription is
diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go
index 8fa26418e..3a75d4d62 100644
--- a/pkg/sentry/vfs/file_description_impl_util_test.go
+++ b/pkg/sentry/vfs/file_description_impl_util_test.go
@@ -107,7 +107,10 @@ func (fd *testFD) SetStat(ctx context.Context, opts SetStatOptions) error {
func TestGenCountFD(t *testing.T) {
ctx := contexttest.Context(t)
- vfsObj := New() // vfs.New()
+ vfsObj := &VirtualFilesystem{}
+ if err := vfsObj.Init(); err != nil {
+ t.Fatalf("VFS init: %v", err)
+ }
fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{})
defer fd.DecRef()
@@ -162,7 +165,10 @@ func TestGenCountFD(t *testing.T) {
func TestWritable(t *testing.T) {
ctx := contexttest.Context(t)
- vfsObj := New() // vfs.New()
+ vfsObj := &VirtualFilesystem{}
+ if err := vfsObj.Init(); err != nil {
+ t.Fatalf("VFS init: %v", err)
+ }
fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"})
defer fd.DecRef()
diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go
index a06a6caf3..556976d0b 100644
--- a/pkg/sentry/vfs/filesystem.go
+++ b/pkg/sentry/vfs/filesystem.go
@@ -29,6 +29,8 @@ import (
// Filesystem methods require that a reference is held.
//
// Filesystem is analogous to Linux's struct super_block.
+//
+// +stateify savable
type Filesystem struct {
// refs is the reference count. refs is accessed using atomic memory
// operations.
diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go
index c58b70728..bb9cada81 100644
--- a/pkg/sentry/vfs/filesystem_type.go
+++ b/pkg/sentry/vfs/filesystem_type.go
@@ -44,6 +44,7 @@ type GetFilesystemOptions struct {
InternalData interface{}
}
+// +stateify savable
type registeredFilesystemType struct {
fsType FilesystemType
opts RegisterFilesystemTypeOptions
diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go
index 1fbb420f9..9912df799 100644
--- a/pkg/sentry/vfs/mount.go
+++ b/pkg/sentry/vfs/mount.go
@@ -38,6 +38,8 @@ import (
//
// Mount is analogous to Linux's struct mount. (gVisor does not distinguish
// between struct mount and struct vfsmount.)
+//
+// +stateify savable
type Mount struct {
// vfs, fs, and root are immutable. References are held on fs and root.
//
@@ -85,6 +87,8 @@ type Mount struct {
// MountNamespace methods require that a reference is held.
//
// MountNamespace is analogous to Linux's struct mnt_namespace.
+//
+// +stateify savable
type MountNamespace struct {
// root is the MountNamespace's root mount. root is immutable.
root *Mount
@@ -114,6 +118,7 @@ type MountNamespace struct {
func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *GetFilesystemOptions) (*MountNamespace, error) {
rft := vfs.getFilesystemType(fsTypeName)
if rft == nil {
+ ctx.Warningf("Unknown filesystem: %s", fsTypeName)
return nil, syserror.ENODEV
}
fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, *opts)
@@ -231,9 +236,12 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti
return syserror.EINVAL
}
vfs.mountMu.Lock()
- if mntns := MountNamespaceFromContext(ctx); mntns != nil && mntns != vd.mount.ns {
- vfs.mountMu.Unlock()
- return syserror.EINVAL
+ if mntns := MountNamespaceFromContext(ctx); mntns != nil {
+ defer mntns.DecRef()
+ if mntns != vd.mount.ns {
+ vfs.mountMu.Unlock()
+ return syserror.EINVAL
+ }
}
// TODO(jamieliu): Linux special-cases umount of the caller's root, which
diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go
index bd90d36c4..bc7581698 100644
--- a/pkg/sentry/vfs/mount_unsafe.go
+++ b/pkg/sentry/vfs/mount_unsafe.go
@@ -26,6 +26,7 @@ import (
"sync/atomic"
"unsafe"
+ "gvisor.dev/gvisor/pkg/gohacks"
"gvisor.dev/gvisor/pkg/sync"
)
@@ -64,6 +65,8 @@ func (mnt *Mount) storeKey(vd VirtualDentry) {
// (provided mutation is sufficiently uncommon).
//
// mountTable.Init() must be called on new mountTables before use.
+//
+// +stateify savable
type mountTable struct {
// mountTable is implemented as a seqcount-protected hash table that
// resolves collisions with linear probing, featuring Robin Hood insertion
@@ -75,8 +78,8 @@ type mountTable struct {
// intrinsics and inline assembly, limiting the performance of this
// approach.)
- seq sync.SeqCount
- seed uint32 // for hashing keys
+ seq sync.SeqCount `state:"nosave"`
+ seed uint32 // for hashing keys
// size holds both length (number of elements) and capacity (number of
// slots): capacity is stored as its base-2 log (referred to as order) in
@@ -89,7 +92,7 @@ type mountTable struct {
// length and cap in separate uint32s) for ~free.
size uint64
- slots unsafe.Pointer // []mountSlot; never nil after Init
+ slots unsafe.Pointer `state:"nosave"` // []mountSlot; never nil after Init
}
type mountSlot struct {
@@ -158,7 +161,7 @@ func newMountTableSlots(cap uintptr) unsafe.Pointer {
// Lookup may be called even if there are concurrent mutators of mt.
func (mt *mountTable) Lookup(parent *Mount, point *Dentry) *Mount {
key := mountKey{parent: unsafe.Pointer(parent), point: unsafe.Pointer(point)}
- hash := memhash(noescape(unsafe.Pointer(&key)), uintptr(mt.seed), mountKeyBytes)
+ hash := memhash(gohacks.Noescape(unsafe.Pointer(&key)), uintptr(mt.seed), mountKeyBytes)
loop:
for {
@@ -359,12 +362,3 @@ func memhash(p unsafe.Pointer, seed, s uintptr) uintptr
//go:linkname rand32 runtime.fastrand
func rand32() uint32
-
-// This is copy/pasted from runtime.noescape(), and is needed because arguments
-// apparently escape from all functions defined by linkname.
-//
-//go:nosplit
-func noescape(p unsafe.Pointer) unsafe.Pointer {
- x := uintptr(p)
- return unsafe.Pointer(x ^ 0)
-}
diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go
index b7774bf28..6af7fdac1 100644
--- a/pkg/sentry/vfs/options.go
+++ b/pkg/sentry/vfs/options.go
@@ -61,7 +61,7 @@ type MountOptions struct {
type OpenOptions struct {
// Flags contains access mode and flags as specified for open(2).
//
- // FilesystemImpls is reponsible for implementing the following flags:
+ // FilesystemImpls are responsible for implementing the following flags:
// O_RDONLY, O_WRONLY, O_RDWR, O_APPEND, O_CREAT, O_DIRECT, O_DSYNC,
// O_EXCL, O_NOATIME, O_NOCTTY, O_NONBLOCK, O_PATH, O_SYNC, O_TMPFILE, and
// O_TRUNC. VFS is responsible for handling O_DIRECTORY, O_LARGEFILE, and
@@ -72,6 +72,11 @@ type OpenOptions struct {
// If FilesystemImpl.OpenAt() creates a file, Mode is the file mode for the
// created file.
Mode linux.FileMode
+
+ // FileExec is set when the file is being opened to be executed.
+ // VirtualFilesystem.OpenAt() checks that the caller has execute permissions
+ // on the file, and that the file is a regular file.
+ FileExec bool
}
// ReadOptions contains options to FileDescription.PRead(),
diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go
index f664581f4..8e250998a 100644
--- a/pkg/sentry/vfs/permissions.go
+++ b/pkg/sentry/vfs/permissions.go
@@ -103,17 +103,22 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo
// AccessTypesForOpenFlags returns MayRead|MayWrite in this case.
//
// Use May{Read,Write}FileWithOpenFlags() for these checks instead.
-func AccessTypesForOpenFlags(flags uint32) AccessTypes {
- switch flags & linux.O_ACCMODE {
+func AccessTypesForOpenFlags(opts *OpenOptions) AccessTypes {
+ ats := AccessTypes(0)
+ if opts.FileExec {
+ ats |= MayExec
+ }
+
+ switch opts.Flags & linux.O_ACCMODE {
case linux.O_RDONLY:
- if flags&linux.O_TRUNC != 0 {
- return MayRead | MayWrite
+ if opts.Flags&linux.O_TRUNC != 0 {
+ return ats | MayRead | MayWrite
}
- return MayRead
+ return ats | MayRead
case linux.O_WRONLY:
- return MayWrite
+ return ats | MayWrite
default:
- return MayRead | MayWrite
+ return ats | MayRead | MayWrite
}
}
diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go
index 8a0b382f6..eb4ebb511 100644
--- a/pkg/sentry/vfs/resolving_path.go
+++ b/pkg/sentry/vfs/resolving_path.go
@@ -228,7 +228,7 @@ func (rp *ResolvingPath) Advance() {
rp.pit = next
} else { // at end of path segment, continue with next one
rp.curPart--
- rp.pit = rp.parts[rp.curPart-1]
+ rp.pit = rp.parts[rp.curPart]
}
}
diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go
index 908c69f91..73f8043be 100644
--- a/pkg/sentry/vfs/vfs.go
+++ b/pkg/sentry/vfs/vfs.go
@@ -46,11 +46,13 @@ import (
//
// There is no analogue to the VirtualFilesystem type in Linux, as the
// equivalent state in Linux is global.
+//
+// +stateify savable
type VirtualFilesystem struct {
// mountMu serializes mount mutations.
//
// mountMu is analogous to Linux's namespace_sem.
- mountMu sync.Mutex
+ mountMu sync.Mutex `state:"nosave"`
// mounts maps (mount parent, mount point) pairs to mounts. (Since mounts
// are uniquely namespaced, including mount parent in the key correctly
@@ -89,44 +91,42 @@ type VirtualFilesystem struct {
// devices contains all registered Devices. devices is protected by
// devicesMu.
- devicesMu sync.RWMutex
+ devicesMu sync.RWMutex `state:"nosave"`
devices map[devTuple]*registeredDevice
// anonBlockDevMinor contains all allocated anonymous block device minor
// numbers. anonBlockDevMinorNext is a lower bound for the smallest
// unallocated anonymous block device number. anonBlockDevMinorNext and
// anonBlockDevMinor are protected by anonBlockDevMinorMu.
- anonBlockDevMinorMu sync.Mutex
+ anonBlockDevMinorMu sync.Mutex `state:"nosave"`
anonBlockDevMinorNext uint32
anonBlockDevMinor map[uint32]struct{}
// fsTypes contains all registered FilesystemTypes. fsTypes is protected by
// fsTypesMu.
- fsTypesMu sync.RWMutex
+ fsTypesMu sync.RWMutex `state:"nosave"`
fsTypes map[string]*registeredFilesystemType
// filesystems contains all Filesystems. filesystems is protected by
// filesystemsMu.
- filesystemsMu sync.Mutex
+ filesystemsMu sync.Mutex `state:"nosave"`
filesystems map[*Filesystem]struct{}
}
-// New returns a new VirtualFilesystem with no mounts or FilesystemTypes.
-func New() *VirtualFilesystem {
- vfs := &VirtualFilesystem{
- mountpoints: make(map[*Dentry]map[*Mount]struct{}),
- devices: make(map[devTuple]*registeredDevice),
- anonBlockDevMinorNext: 1,
- anonBlockDevMinor: make(map[uint32]struct{}),
- fsTypes: make(map[string]*registeredFilesystemType),
- filesystems: make(map[*Filesystem]struct{}),
- }
+// Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes.
+func (vfs *VirtualFilesystem) Init() error {
+ vfs.mountpoints = make(map[*Dentry]map[*Mount]struct{})
+ vfs.devices = make(map[devTuple]*registeredDevice)
+ vfs.anonBlockDevMinorNext = 1
+ vfs.anonBlockDevMinor = make(map[uint32]struct{})
+ vfs.fsTypes = make(map[string]*registeredFilesystemType)
+ vfs.filesystems = make(map[*Filesystem]struct{})
vfs.mounts.Init()
// Construct vfs.anonMount.
anonfsDevMinor, err := vfs.GetAnonBlockDevMinor()
if err != nil {
- panic(fmt.Sprintf("VirtualFilesystem.GetAnonBlockDevMinor() failed during VirtualFilesystem construction: %v", err))
+ return err
}
anonfs := anonFilesystem{
devMinor: anonfsDevMinor,
@@ -137,8 +137,7 @@ func New() *VirtualFilesystem {
fs: &anonfs.vfsfs,
refs: 1,
}
-
- return vfs
+ return nil
}
// PathOperation specifies the path operated on by a VFS method.
@@ -379,6 +378,22 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential
fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts)
if err == nil {
vfs.putResolvingPath(rp)
+
+ // TODO(gvisor.dev/issue/1193): Move inside fsimpl to avoid another call
+ // to FileDescription.Stat().
+ if opts.FileExec {
+ // Only a regular file can be executed.
+ stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE})
+ if err != nil {
+ fd.DecRef()
+ return nil, err
+ }
+ if stat.Mask&linux.STATX_TYPE == 0 || stat.Mode&linux.S_IFMT != linux.S_IFREG {
+ fd.DecRef()
+ return nil, syserror.EACCES
+ }
+ }
+
return fd, nil
}
if !rp.handleError(err) {
@@ -724,6 +739,8 @@ func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error {
// VirtualDentry methods require that a reference is held on the VirtualDentry.
//
// VirtualDentry is analogous to Linux's struct path.
+//
+// +stateify savable
type VirtualDentry struct {
mount *Mount
dentry *Dentry
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-30 20:22:07 +0000