summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/kernel')
-rw-r--r--pkg/sentry/kernel/BUILD3
-rw-r--r--pkg/sentry/kernel/fd_table.go49
-rw-r--r--pkg/sentry/kernel/fs_context.go120
-rw-r--r--pkg/sentry/kernel/kernel.go188
-rw-r--r--pkg/sentry/kernel/rseq.go16
-rw-r--r--pkg/sentry/kernel/task.go54
-rw-r--r--pkg/sentry/kernel/task_clone.go27
-rw-r--r--pkg/sentry/kernel/task_context.go4
-rw-r--r--pkg/sentry/kernel/task_exec.go2
-rw-r--r--pkg/sentry/kernel/task_exit.go7
-rw-r--r--pkg/sentry/kernel/task_log.go21
-rw-r--r--pkg/sentry/kernel/task_net.go19
-rw-r--r--pkg/sentry/kernel/task_start.go55
-rw-r--r--pkg/sentry/kernel/task_usermem.go2
-rw-r--r--pkg/sentry/kernel/thread_group.go6
15 files changed, 438 insertions, 135 deletions
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index 2231d6973..beba29a09 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -157,6 +157,7 @@ go_library(
"//pkg/context",
"//pkg/cpuid",
"//pkg/eventchannel",
+ "//pkg/fspath",
"//pkg/log",
"//pkg/metric",
"//pkg/refs",
@@ -167,6 +168,7 @@ go_library(
"//pkg/sentry/fs",
"//pkg/sentry/fs/lock",
"//pkg/sentry/fs/timerfd",
+ "//pkg/sentry/fsbridge",
"//pkg/sentry/hostcpu",
"//pkg/sentry/inet",
"//pkg/sentry/kernel/auth",
@@ -199,6 +201,7 @@ go_library(
"//pkg/tcpip/stack",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
],
)
diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go
index 23b88f7a6..58001d56c 100644
--- a/pkg/sentry/kernel/fd_table.go
+++ b/pkg/sentry/kernel/fd_table.go
@@ -296,6 +296,50 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags
return fds, nil
}
+// NewFDVFS2 allocates a file descriptor greater than or equal to minfd for
+// the given file description. If it succeeds, it takes a reference on file.
+func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) {
+ if minfd < 0 {
+ // Don't accept negative FDs.
+ return -1, syscall.EINVAL
+ }
+
+ // Default limit.
+ end := int32(math.MaxInt32)
+
+ // Ensure we don't get past the provided limit.
+ if limitSet := limits.FromContext(ctx); limitSet != nil {
+ lim := limitSet.Get(limits.NumberOfFiles)
+ if lim.Cur != limits.Infinity {
+ end = int32(lim.Cur)
+ }
+ if minfd >= end {
+ return -1, syscall.EMFILE
+ }
+ }
+
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ // From f.next to find available fd.
+ fd := minfd
+ if fd < f.next {
+ fd = f.next
+ }
+ for fd < end {
+ if d, _, _ := f.get(fd); d == nil {
+ f.setVFS2(fd, file, flags)
+ if fd == f.next {
+ // Update next search start position.
+ f.next = fd + 1
+ }
+ return fd, nil
+ }
+ fd++
+ }
+ return -1, syscall.EMFILE
+}
+
// NewFDAt sets the file reference for the given FD. If there is an active
// reference for that FD, the ref count for that existing reference is
// decremented.
@@ -316,9 +360,6 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2
return syscall.EBADF
}
- f.mu.Lock()
- defer f.mu.Unlock()
-
// Check the limit for the provided file.
if limitSet := limits.FromContext(ctx); limitSet != nil {
if lim := limitSet.Get(limits.NumberOfFiles); lim.Cur != limits.Infinity && uint64(fd) >= lim.Cur {
@@ -327,6 +368,8 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2
}
// Install the entry.
+ f.mu.Lock()
+ defer f.mu.Unlock()
f.setAll(fd, file, fileVFS2, flags)
return nil
}
diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go
index 2448c1d99..47f78df9a 100644
--- a/pkg/sentry/kernel/fs_context.go
+++ b/pkg/sentry/kernel/fs_context.go
@@ -19,6 +19,7 @@ import (
"gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/fs"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/sync"
)
@@ -37,10 +38,16 @@ type FSContext struct {
// destroyed.
root *fs.Dirent
+ // rootVFS2 is the filesystem root.
+ rootVFS2 vfs.VirtualDentry
+
// cwd is the current working directory. Will be nil iff the FSContext
// has been destroyed.
cwd *fs.Dirent
+ // cwdVFS2 is the current working directory.
+ cwdVFS2 vfs.VirtualDentry
+
// umask is the current file mode creation mask. When a thread using this
// context invokes a syscall that creates a file, bits set in umask are
// removed from the permissions that the file is created with.
@@ -60,6 +67,19 @@ func newFSContext(root, cwd *fs.Dirent, umask uint) *FSContext {
return &f
}
+// NewFSContextVFS2 returns a new filesystem context.
+func NewFSContextVFS2(root, cwd vfs.VirtualDentry, umask uint) *FSContext {
+ root.IncRef()
+ cwd.IncRef()
+ f := FSContext{
+ rootVFS2: root,
+ cwdVFS2: cwd,
+ umask: umask,
+ }
+ f.EnableLeakCheck("kernel.FSContext")
+ return &f
+}
+
// destroy is the destructor for an FSContext.
//
// This will call DecRef on both root and cwd Dirents. If either call to
@@ -75,11 +95,17 @@ func (f *FSContext) destroy() {
f.mu.Lock()
defer f.mu.Unlock()
- f.root.DecRef()
- f.root = nil
-
- f.cwd.DecRef()
- f.cwd = nil
+ if VFS2Enabled {
+ f.rootVFS2.DecRef()
+ f.rootVFS2 = vfs.VirtualDentry{}
+ f.cwdVFS2.DecRef()
+ f.cwdVFS2 = vfs.VirtualDentry{}
+ } else {
+ f.root.DecRef()
+ f.root = nil
+ f.cwd.DecRef()
+ f.cwd = nil
+ }
}
// DecRef implements RefCounter.DecRef with destructor f.destroy.
@@ -93,12 +119,21 @@ func (f *FSContext) DecRef() {
func (f *FSContext) Fork() *FSContext {
f.mu.Lock()
defer f.mu.Unlock()
- f.cwd.IncRef()
- f.root.IncRef()
+
+ if VFS2Enabled {
+ f.cwdVFS2.IncRef()
+ f.rootVFS2.IncRef()
+ } else {
+ f.cwd.IncRef()
+ f.root.IncRef()
+ }
+
return &FSContext{
- cwd: f.cwd,
- root: f.root,
- umask: f.umask,
+ cwd: f.cwd,
+ root: f.root,
+ cwdVFS2: f.cwdVFS2,
+ rootVFS2: f.rootVFS2,
+ umask: f.umask,
}
}
@@ -109,12 +144,23 @@ func (f *FSContext) Fork() *FSContext {
func (f *FSContext) WorkingDirectory() *fs.Dirent {
f.mu.Lock()
defer f.mu.Unlock()
- if f.cwd != nil {
- f.cwd.IncRef()
- }
+
+ f.cwd.IncRef()
return f.cwd
}
+// WorkingDirectoryVFS2 returns the current working directory.
+//
+// This will return nil if called after destroy(), otherwise it will return a
+// Dirent with a reference taken.
+func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry {
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ f.cwdVFS2.IncRef()
+ return f.cwdVFS2
+}
+
// SetWorkingDirectory sets the current working directory.
// This will take an extra reference on the Dirent.
//
@@ -137,6 +183,20 @@ func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) {
old.DecRef()
}
+// SetWorkingDirectoryVFS2 sets the current working directory.
+// This will take an extra reference on the VirtualDentry.
+//
+// This is not a valid call after destroy.
+func (f *FSContext) SetWorkingDirectoryVFS2(d vfs.VirtualDentry) {
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ old := f.cwdVFS2
+ f.cwdVFS2 = d
+ d.IncRef()
+ old.DecRef()
+}
+
// RootDirectory returns the current filesystem root.
//
// This will return nil if called after destroy(), otherwise it will return a
@@ -150,6 +210,18 @@ func (f *FSContext) RootDirectory() *fs.Dirent {
return f.root
}
+// RootDirectoryVFS2 returns the current filesystem root.
+//
+// This will return nil if called after destroy(), otherwise it will return a
+// Dirent with a reference taken.
+func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry {
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ f.rootVFS2.IncRef()
+ return f.rootVFS2
+}
+
// SetRootDirectory sets the root directory.
// This will take an extra reference on the Dirent.
//
@@ -172,6 +244,28 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) {
old.DecRef()
}
+// SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd.
+//
+// This is not a valid call after free.
+func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) {
+ if !vd.Ok() {
+ panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry")
+ }
+
+ f.mu.Lock()
+
+ if !f.rootVFS2.Ok() {
+ f.mu.Unlock()
+ panic(fmt.Sprintf("FSContext.SetRootDirectoryVFS2(%v)) called after destroy", vd))
+ }
+
+ old := f.rootVFS2
+ vd.IncRef()
+ f.rootVFS2 = vd
+ f.mu.Unlock()
+ old.DecRef()
+}
+
// Umask returns the current umask.
func (f *FSContext) Umask() uint {
f.mu.Lock()
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 3ee760ba2..8b76750e9 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -43,11 +43,13 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/cpuid"
"gvisor.dev/gvisor/pkg/eventchannel"
+ "gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/fs/timerfd"
+ "gvisor.dev/gvisor/pkg/sentry/fsbridge"
"gvisor.dev/gvisor/pkg/sentry/hostcpu"
"gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
@@ -71,6 +73,10 @@ import (
"gvisor.dev/gvisor/pkg/tcpip"
)
+// VFS2Enabled is set to true when VFS2 is enabled. Added as a global for allow
+// easy access everywhere. To be removed once VFS2 becomes the default.
+var VFS2Enabled = false
+
// Kernel represents an emulated Linux kernel. It must be initialized by calling
// Init() or LoadFrom().
//
@@ -105,7 +111,7 @@ type Kernel struct {
timekeeper *Timekeeper
tasks *TaskSet
rootUserNamespace *auth.UserNamespace
- networkStack inet.Stack `state:"nosave"`
+ rootNetworkNamespace *inet.Namespace
applicationCores uint
useHostCores bool
extraAuxv []arch.AuxEntry
@@ -238,6 +244,13 @@ type Kernel struct {
// SpecialOpts contains special kernel options.
SpecialOpts
+
+ // VFS keeps the filesystem state used across the kernel.
+ vfs vfs.VirtualFilesystem
+
+ // If set to true, report address space activation waits as if the task is in
+ // external wait so that the watchdog doesn't report the task stuck.
+ SleepForAddressSpaceActivation bool
}
// InitKernelArgs holds arguments to Init.
@@ -251,8 +264,9 @@ type InitKernelArgs struct {
// RootUserNamespace is the root user namespace.
RootUserNamespace *auth.UserNamespace
- // NetworkStack is the TCP/IP network stack. NetworkStack may be nil.
- NetworkStack inet.Stack
+ // RootNetworkNamespace is the root network namespace. If nil, no networking
+ // will be available.
+ RootNetworkNamespace *inet.Namespace
// ApplicationCores is the number of logical CPUs visible to sandboxed
// applications. The set of logical CPU IDs is [0, ApplicationCores); thus
@@ -311,7 +325,10 @@ func (k *Kernel) Init(args InitKernelArgs) error {
k.rootUTSNamespace = args.RootUTSNamespace
k.rootIPCNamespace = args.RootIPCNamespace
k.rootAbstractSocketNamespace = args.RootAbstractSocketNamespace
- k.networkStack = args.NetworkStack
+ k.rootNetworkNamespace = args.RootNetworkNamespace
+ if k.rootNetworkNamespace == nil {
+ k.rootNetworkNamespace = inet.NewRootNamespace(nil, nil)
+ }
k.applicationCores = args.ApplicationCores
if args.UseHostCores {
k.useHostCores = true
@@ -534,8 +551,6 @@ func (ts *TaskSet) unregisterEpollWaiters() {
func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) error {
loadStart := time.Now()
- k.networkStack = net
-
initAppCores := k.applicationCores
// Load the pre-saved CPUID FeatureSet.
@@ -566,6 +581,10 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks)
log.Infof("Kernel load stats: %s", &stats)
log.Infof("Kernel load took [%s].", time.Since(kernelStart))
+ // rootNetworkNamespace should be populated after loading the state file.
+ // Restore the root network stack.
+ k.rootNetworkNamespace.RestoreRootStack(net)
+
// Load the memory file's state.
memoryStart := time.Now()
if err := k.mf.LoadFrom(k.SupervisorContext(), r); err != nil {
@@ -624,7 +643,7 @@ type CreateProcessArgs struct {
// File is a passed host FD pointing to a file to load as the init binary.
//
// This is checked if and only if Filename is "".
- File *fs.File
+ File fsbridge.File
// Argvv is a list of arguments.
Argv []string
@@ -673,6 +692,13 @@ type CreateProcessArgs struct {
// increment it).
MountNamespace *fs.MountNamespace
+ // MountNamespaceVFS2 optionally contains the mount namespace for this
+ // process. If nil, the init process's mount namespace is used.
+ //
+ // Anyone setting MountNamespaceVFS2 must donate a reference (i.e.
+ // increment it).
+ MountNamespaceVFS2 *vfs.MountNamespace
+
// ContainerID is the container that the process belongs to.
ContainerID string
}
@@ -711,11 +737,22 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} {
return ctx.args.Credentials
case fs.CtxRoot:
if ctx.args.MountNamespace != nil {
- // MountNamespace.Root() will take a reference on the root
- // dirent for us.
+ // MountNamespace.Root() will take a reference on the root dirent for us.
return ctx.args.MountNamespace.Root()
}
return nil
+ case vfs.CtxRoot:
+ if ctx.args.MountNamespaceVFS2 == nil {
+ return nil
+ }
+ // MountNamespaceVFS2.Root() takes a reference on the root dirent for us.
+ return ctx.args.MountNamespaceVFS2.Root()
+ case vfs.CtxMountNamespace:
+ if ctx.k.globalInit == nil {
+ return nil
+ }
+ // MountNamespaceVFS2 takes a reference for us.
+ return ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
case fs.CtxDirentCacheLimiter:
return ctx.k.DirentCacheLimiter
case ktime.CtxRealtimeClock:
@@ -757,34 +794,77 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
defer k.extMu.Unlock()
log.Infof("EXEC: %v", args.Argv)
- // Grab the mount namespace.
- mounts := args.MountNamespace
- if mounts == nil {
- mounts = k.GlobalInit().Leader().MountNamespace()
- mounts.IncRef()
- }
-
- tg := k.NewThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits)
ctx := args.NewContext(k)
- // Get the root directory from the MountNamespace.
- root := mounts.Root()
- // The call to newFSContext below will take a reference on root, so we
- // don't need to hold this one.
- defer root.DecRef()
-
- // Grab the working directory.
- remainingTraversals := uint(args.MaxSymlinkTraversals)
- wd := root // Default.
- if args.WorkingDirectory != "" {
- var err error
- wd, err = mounts.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals)
- if err != nil {
- return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err)
+ var (
+ opener fsbridge.Lookup
+ fsContext *FSContext
+ mntns *fs.MountNamespace
+ )
+
+ if VFS2Enabled {
+ mntnsVFS2 := args.MountNamespaceVFS2
+ if mntnsVFS2 == nil {
+ // MountNamespaceVFS2 adds a reference to the namespace, which is
+ // transferred to the new process.
+ mntnsVFS2 = k.GlobalInit().Leader().MountNamespaceVFS2()
}
- defer wd.DecRef()
+ // Get the root directory from the MountNamespace.
+ root := args.MountNamespaceVFS2.Root()
+ // The call to newFSContext below will take a reference on root, so we
+ // don't need to hold this one.
+ defer root.DecRef()
+
+ // Grab the working directory.
+ wd := root // Default.
+ if args.WorkingDirectory != "" {
+ pop := vfs.PathOperation{
+ Root: root,
+ Start: wd,
+ Path: fspath.Parse(args.WorkingDirectory),
+ FollowFinalSymlink: true,
+ }
+ var err error
+ wd, err = k.VFS().GetDentryAt(ctx, args.Credentials, &pop, &vfs.GetDentryOptions{
+ CheckSearchable: true,
+ })
+ if err != nil {
+ return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err)
+ }
+ defer wd.DecRef()
+ }
+ opener = fsbridge.NewVFSLookup(mntnsVFS2, root, wd)
+ fsContext = NewFSContextVFS2(root, wd, args.Umask)
+
+ } else {
+ mntns = args.MountNamespace
+ if mntns == nil {
+ mntns = k.GlobalInit().Leader().MountNamespace()
+ mntns.IncRef()
+ }
+ // Get the root directory from the MountNamespace.
+ root := mntns.Root()
+ // The call to newFSContext below will take a reference on root, so we
+ // don't need to hold this one.
+ defer root.DecRef()
+
+ // Grab the working directory.
+ remainingTraversals := args.MaxSymlinkTraversals
+ wd := root // Default.
+ if args.WorkingDirectory != "" {
+ var err error
+ wd, err = mntns.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals)
+ if err != nil {
+ return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err)
+ }
+ defer wd.DecRef()
+ }
+ opener = fsbridge.NewFSLookup(mntns, root, wd)
+ fsContext = newFSContext(root, wd, args.Umask)
}
+ tg := k.NewThreadGroup(mntns, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits)
+
// Check which file to start from.
switch {
case args.Filename != "":
@@ -805,11 +885,9 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
}
// Create a fresh task context.
- remainingTraversals = uint(args.MaxSymlinkTraversals)
+ remainingTraversals := args.MaxSymlinkTraversals
loadArgs := loader.LoadArgs{
- Mounts: mounts,
- Root: root,
- WorkingDirectory: wd,
+ Opener: opener,
RemainingTraversals: &remainingTraversals,
ResolveFinal: true,
Filename: args.Filename,
@@ -834,13 +912,15 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
Kernel: k,
ThreadGroup: tg,
TaskContext: tc,
- FSContext: newFSContext(root, wd, args.Umask),
+ FSContext: fsContext,
FDTable: args.FDTable,
Credentials: args.Credentials,
+ NetworkNamespace: k.RootNetworkNamespace(),
AllowedCPUMask: sched.NewFullCPUSet(k.applicationCores),
UTSNamespace: args.UTSNamespace,
IPCNamespace: args.IPCNamespace,
AbstractSocketNamespace: args.AbstractSocketNamespace,
+ MountNamespaceVFS2: args.MountNamespaceVFS2,
ContainerID: args.ContainerID,
}
t, err := k.tasks.NewTask(config)
@@ -1100,6 +1180,14 @@ func (k *Kernel) SendExternalSignal(info *arch.SignalInfo, context string) {
k.sendExternalSignal(info, context)
}
+// SendExternalSignalThreadGroup injects a signal into an specific ThreadGroup.
+// This function doesn't skip signals like SendExternalSignal does.
+func (k *Kernel) SendExternalSignalThreadGroup(tg *ThreadGroup, info *arch.SignalInfo) error {
+ k.extMu.Lock()
+ defer k.extMu.Unlock()
+ return tg.SendSignal(info)
+}
+
// SendContainerSignal sends the given signal to all processes inside the
// namespace that match the given container ID.
func (k *Kernel) SendContainerSignal(cid string, info *arch.SignalInfo) error {
@@ -1178,10 +1266,9 @@ func (k *Kernel) RootAbstractSocketNamespace() *AbstractSocketNamespace {
return k.rootAbstractSocketNamespace
}
-// NetworkStack returns the network stack. NetworkStack may return nil if no
-// network stack is available.
-func (k *Kernel) NetworkStack() inet.Stack {
- return k.networkStack
+// RootNetworkNamespace returns the root network namespace, always non-nil.
+func (k *Kernel) RootNetworkNamespace() *inet.Namespace {
+ return k.rootNetworkNamespace
}
// GlobalInit returns the thread group with ID 1 in the root PID namespace, or
@@ -1378,6 +1465,20 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
return ctx.k.globalInit.mounts.Root()
}
return nil
+ case vfs.CtxRoot:
+ if ctx.k.globalInit == nil {
+ return vfs.VirtualDentry{}
+ }
+ mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
+ defer mntns.DecRef()
+ // Root() takes a reference on the root dirent for us.
+ return mntns.Root()
+ case vfs.CtxMountNamespace:
+ if ctx.k.globalInit == nil {
+ return nil
+ }
+ // MountNamespaceVFS2() takes a reference for us.
+ return ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
case fs.CtxDirentCacheLimiter:
return ctx.k.DirentCacheLimiter
case ktime.CtxRealtimeClock:
@@ -1423,3 +1524,8 @@ func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
Registers: t.Arch().StateData().Proto(),
})
}
+
+// VFS returns the virtual filesystem for the kernel.
+func (k *Kernel) VFS() *vfs.VirtualFilesystem {
+ return &k.vfs
+}
diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go
index efebfd872..ded95f532 100644
--- a/pkg/sentry/kernel/rseq.go
+++ b/pkg/sentry/kernel/rseq.go
@@ -303,26 +303,14 @@ func (t *Task) rseqAddrInterrupt() {
return
}
- buf = t.CopyScratchBuffer(linux.SizeOfRSeqCriticalSection)
- if _, err := t.CopyInBytes(critAddr, buf); err != nil {
+ var cs linux.RSeqCriticalSection
+ if err := cs.CopyIn(t, critAddr); err != nil {
t.Debugf("Failed to copy critical section from %#x for rseq: %v", critAddr, err)
t.forceSignal(linux.SIGSEGV, false /* unconditional */)
t.SendSignal(SignalInfoPriv(linux.SIGSEGV))
return
}
- // Manually marshal RSeqCriticalSection as this is in the hot path when
- // rseq is enabled. It must be as fast as possible.
- //
- // TODO(b/130243041): Replace with go_marshal.
- cs := linux.RSeqCriticalSection{
- Version: usermem.ByteOrder.Uint32(buf[0:4]),
- Flags: usermem.ByteOrder.Uint32(buf[4:8]),
- Start: usermem.ByteOrder.Uint64(buf[8:16]),
- PostCommitOffset: usermem.ByteOrder.Uint64(buf[16:24]),
- Abort: usermem.ByteOrder.Uint64(buf[24:32]),
- }
-
if cs.Version != 0 {
t.Debugf("Unknown version in %+v", cs)
t.forceSignal(linux.SIGSEGV, false /* unconditional */)
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 981e8c7fe..2cee2e6ed 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -424,6 +424,11 @@ type Task struct {
// abstractSockets is protected by mu.
abstractSockets *AbstractSocketNamespace
+ // mountNamespaceVFS2 is the task's mount namespace.
+ //
+ // It is protected by mu. It is owned by the task goroutine.
+ mountNamespaceVFS2 *vfs.MountNamespace
+
// parentDeathSignal is sent to this task's thread group when its parent exits.
//
// parentDeathSignal is protected by mu.
@@ -481,13 +486,10 @@ type Task struct {
numaPolicy int32
numaNodeMask uint64
- // If netns is true, the task is in a non-root network namespace. Network
- // namespaces aren't currently implemented in full; being in a network
- // namespace simply prevents the task from observing any network devices
- // (including loopback) or using abstract socket addresses (see unix(7)).
+ // netns is the task's network namespace. netns is never nil.
//
- // netns is protected by mu. netns is owned by the task goroutine.
- netns bool
+ // netns is protected by mu.
+ netns *inet.Namespace
// If rseqPreempted is true, before the next call to p.Switch(),
// interrupt rseq critical regions as defined by rseqAddr and
@@ -638,6 +640,11 @@ func (t *Task) Value(key interface{}) interface{} {
return int32(t.ThreadGroup().ID())
case fs.CtxRoot:
return t.fsContext.RootDirectory()
+ case vfs.CtxRoot:
+ return t.fsContext.RootDirectoryVFS2()
+ case vfs.CtxMountNamespace:
+ t.mountNamespaceVFS2.IncRef()
+ return t.mountNamespaceVFS2
case fs.CtxDirentCacheLimiter:
return t.k.DirentCacheLimiter
case inet.CtxStack:
@@ -701,6 +708,14 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock {
// Preconditions: The caller must be running on the task goroutine, or t.mu
// must be locked.
func (t *Task) IsChrooted() bool {
+ if VFS2Enabled {
+ realRoot := t.mountNamespaceVFS2.Root()
+ defer realRoot.DecRef()
+ root := t.fsContext.RootDirectoryVFS2()
+ defer root.DecRef()
+ return root != realRoot
+ }
+
realRoot := t.tg.mounts.Root()
defer realRoot.DecRef()
root := t.fsContext.RootDirectory()
@@ -774,6 +789,15 @@ func (t *Task) NewFDFrom(fd int32, file *fs.File, flags FDFlags) (int32, error)
return fds[0], nil
}
+// NewFDFromVFS2 is a convenience wrapper for t.FDTable().NewFDVFS2.
+//
+// This automatically passes the task as the context.
+//
+// Precondition: same as FDTable.Get.
+func (t *Task) NewFDFromVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) {
+ return t.fdTable.NewFDVFS2(t, fd, file, flags)
+}
+
// NewFDAt is a convenience wrapper for t.FDTable().NewFDAt.
//
// This automatically passes the task as the context.
@@ -783,6 +807,15 @@ func (t *Task) NewFDAt(fd int32, file *fs.File, flags FDFlags) error {
return t.fdTable.NewFDAt(t, fd, file, flags)
}
+// NewFDAtVFS2 is a convenience wrapper for t.FDTable().NewFDAtVFS2.
+//
+// This automatically passes the task as the context.
+//
+// Precondition: same as FDTable.
+func (t *Task) NewFDAtVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) error {
+ return t.fdTable.NewFDAtVFS2(t, fd, file, flags)
+}
+
// WithMuLocked executes f with t.mu locked.
func (t *Task) WithMuLocked(f func(*Task)) {
t.mu.Lock()
@@ -796,6 +829,15 @@ func (t *Task) MountNamespace() *fs.MountNamespace {
return t.tg.mounts
}
+// MountNamespaceVFS2 returns t's MountNamespace. A reference is taken on the
+// returned mount namespace.
+func (t *Task) MountNamespaceVFS2() *vfs.MountNamespace {
+ t.mu.Lock()
+ defer t.mu.Unlock()
+ t.mountNamespaceVFS2.IncRef()
+ return t.mountNamespaceVFS2
+}
+
// AbstractSockets returns t's AbstractSocketNamespace.
func (t *Task) AbstractSockets() *AbstractSocketNamespace {
return t.abstractSockets
diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go
index 53d4d211b..78866f280 100644
--- a/pkg/sentry/kernel/task_clone.go
+++ b/pkg/sentry/kernel/task_clone.go
@@ -17,6 +17,7 @@ package kernel
import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/bpf"
+ "gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
)
@@ -54,8 +55,7 @@ type SharingOptions struct {
NewUserNamespace bool
// If NewNetworkNamespace is true, the task should have an independent
- // network namespace. (Note that network namespaces are not really
- // implemented; see comment on Task.netns for details.)
+ // network namespace.
NewNetworkNamespace bool
// If NewFiles is true, the task should use an independent file descriptor
@@ -199,6 +199,17 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
ipcns = NewIPCNamespace(userns)
}
+ netns := t.NetworkNamespace()
+ if opts.NewNetworkNamespace {
+ netns = inet.NewNamespace(netns)
+ }
+
+ // TODO(b/63601033): Implement CLONE_NEWNS.
+ mntnsVFS2 := t.mountNamespaceVFS2
+ if mntnsVFS2 != nil {
+ mntnsVFS2.IncRef()
+ }
+
tc, err := t.tc.Fork(t, t.k, !opts.NewAddressSpace)
if err != nil {
return 0, nil, err
@@ -241,7 +252,9 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
rseqAddr := usermem.Addr(0)
rseqSignature := uint32(0)
if opts.NewThreadGroup {
- tg.mounts.IncRef()
+ if tg.mounts != nil {
+ tg.mounts.IncRef()
+ }
sh := t.tg.signalHandlers
if opts.NewSignalHandlers {
sh = sh.Fork()
@@ -260,11 +273,12 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
FDTable: fdTable,
Credentials: creds,
Niceness: t.Niceness(),
- NetworkNamespaced: t.netns,
+ NetworkNamespace: netns,
AllowedCPUMask: t.CPUMask(),
UTSNamespace: utsns,
IPCNamespace: ipcns,
AbstractSocketNamespace: t.abstractSockets,
+ MountNamespaceVFS2: mntnsVFS2,
RSeqAddr: rseqAddr,
RSeqSignature: rseqSignature,
ContainerID: t.ContainerID(),
@@ -274,9 +288,6 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
} else {
cfg.InheritParent = t
}
- if opts.NewNetworkNamespace {
- cfg.NetworkNamespaced = true
- }
nt, err := t.tg.pidns.owner.NewTask(cfg)
if err != nil {
if opts.NewThreadGroup {
@@ -473,7 +484,7 @@ func (t *Task) Unshare(opts *SharingOptions) error {
t.mu.Unlock()
return syserror.EPERM
}
- t.netns = true
+ t.netns = inet.NewNamespace(t.netns)
}
if opts.NewUTSNamespace {
if !haveCapSysAdmin {
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index 2d6e7733c..0158b1788 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -136,11 +136,11 @@ func (t *Task) Stack() *arch.Stack {
func (k *Kernel) LoadTaskImage(ctx context.Context, args loader.LoadArgs) (*TaskContext, *syserr.Error) {
// If File is not nil, we should load that instead of resolving Filename.
if args.File != nil {
- args.Filename = args.File.MappedName(ctx)
+ args.Filename = args.File.PathnameWithDeleted(ctx)
}
// Prepare a new user address space to load into.
- m := mm.NewMemoryManager(k, k)
+ m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation)
defer m.DecUsers(ctx)
args.MemoryManager = m
diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go
index 8f57a34a6..00c425cca 100644
--- a/pkg/sentry/kernel/task_exec.go
+++ b/pkg/sentry/kernel/task_exec.go
@@ -220,7 +220,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState {
t.mu.Unlock()
t.unstopVforkParent()
// NOTE(b/30316266): All locks must be dropped prior to calling Activate.
- t.MemoryManager().Activate()
+ t.MemoryManager().Activate(t)
t.ptraceExec(oldTID)
return (*runSyscallExit)(nil)
diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go
index 435761e5a..c4ade6e8e 100644
--- a/pkg/sentry/kernel/task_exit.go
+++ b/pkg/sentry/kernel/task_exit.go
@@ -269,6 +269,13 @@ func (*runExitMain) execute(t *Task) taskRunState {
t.fsContext.DecRef()
t.fdTable.DecRef()
+ t.mu.Lock()
+ if t.mountNamespaceVFS2 != nil {
+ t.mountNamespaceVFS2.DecRef()
+ t.mountNamespaceVFS2 = nil
+ }
+ t.mu.Unlock()
+
// If this is the last task to exit from the thread group, release the
// thread group's resources.
if lastExiter {
diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go
index 41259210c..eeccaa197 100644
--- a/pkg/sentry/kernel/task_log.go
+++ b/pkg/sentry/kernel/task_log.go
@@ -32,21 +32,21 @@ const (
// Infof logs an formatted info message by calling log.Infof.
func (t *Task) Infof(fmt string, v ...interface{}) {
if log.IsLogging(log.Info) {
- log.Infof(t.logPrefix.Load().(string)+fmt, v...)
+ log.InfofAtDepth(1, t.logPrefix.Load().(string)+fmt, v...)
}
}
// Warningf logs a warning string by calling log.Warningf.
func (t *Task) Warningf(fmt string, v ...interface{}) {
if log.IsLogging(log.Warning) {
- log.Warningf(t.logPrefix.Load().(string)+fmt, v...)
+ log.WarningfAtDepth(1, t.logPrefix.Load().(string)+fmt, v...)
}
}
// Debugf creates a debug string that includes the task ID.
func (t *Task) Debugf(fmt string, v ...interface{}) {
if log.IsLogging(log.Debug) {
- log.Debugf(t.logPrefix.Load().(string)+fmt, v...)
+ log.DebugfAtDepth(1, t.logPrefix.Load().(string)+fmt, v...)
}
}
@@ -198,18 +198,11 @@ func (t *Task) traceExecEvent(tc *TaskContext) {
if !trace.IsEnabled() {
return
}
- d := tc.MemoryManager.Executable()
- if d == nil {
+ file := tc.MemoryManager.Executable()
+ if file == nil {
trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>")
return
}
- defer d.DecRef()
- root := t.fsContext.RootDirectory()
- if root == nil {
- trace.Logf(t.traceContext, traceCategory, "exec: << no root directory >>")
- return
- }
- defer root.DecRef()
- n, _ := d.FullName(root)
- trace.Logf(t.traceContext, traceCategory, "exec: %s", n)
+ defer file.DecRef()
+ trace.Logf(t.traceContext, traceCategory, "exec: %s", file.PathnameWithDeleted(t))
}
diff --git a/pkg/sentry/kernel/task_net.go b/pkg/sentry/kernel/task_net.go
index 172a31e1d..f7711232c 100644
--- a/pkg/sentry/kernel/task_net.go
+++ b/pkg/sentry/kernel/task_net.go
@@ -22,14 +22,23 @@ import (
func (t *Task) IsNetworkNamespaced() bool {
t.mu.Lock()
defer t.mu.Unlock()
- return t.netns
+ return !t.netns.IsRoot()
}
// NetworkContext returns the network stack used by the task. NetworkContext
// may return nil if no network stack is available.
+//
+// TODO(gvisor.dev/issue/1833): Migrate callers of this method to
+// NetworkNamespace().
func (t *Task) NetworkContext() inet.Stack {
- if t.IsNetworkNamespaced() {
- return nil
- }
- return t.k.networkStack
+ t.mu.Lock()
+ defer t.mu.Unlock()
+ return t.netns.Stack()
+}
+
+// NetworkNamespace returns the network namespace observed by the task.
+func (t *Task) NetworkNamespace() *inet.Namespace {
+ t.mu.Lock()
+ defer t.mu.Unlock()
+ return t.netns
}
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index de838beef..a5035bb7f 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -17,10 +17,12 @@ package kernel
import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/kernel/futex"
"gvisor.dev/gvisor/pkg/sentry/kernel/sched"
"gvisor.dev/gvisor/pkg/sentry/usage"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
)
@@ -64,9 +66,8 @@ type TaskConfig struct {
// Niceness is the niceness of the new task.
Niceness int
- // If NetworkNamespaced is true, the new task should observe a non-root
- // network namespace.
- NetworkNamespaced bool
+ // NetworkNamespace is the network namespace to be used for the new task.
+ NetworkNamespace *inet.Namespace
// AllowedCPUMask contains the cpus that this task can run on.
AllowedCPUMask sched.CPUSet
@@ -80,6 +81,9 @@ type TaskConfig struct {
// AbstractSocketNamespace is the AbstractSocketNamespace of the new task.
AbstractSocketNamespace *AbstractSocketNamespace
+ // MountNamespaceVFS2 is the MountNamespace of the new task.
+ MountNamespaceVFS2 *vfs.MountNamespace
+
// RSeqAddr is a pointer to the the userspace linux.RSeq structure.
RSeqAddr usermem.Addr
@@ -116,28 +120,29 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) {
parent: cfg.Parent,
children: make(map[*Task]struct{}),
},
- runState: (*runApp)(nil),
- interruptChan: make(chan struct{}, 1),
- signalMask: cfg.SignalMask,
- signalStack: arch.SignalStack{Flags: arch.SignalStackFlagDisable},
- tc: *tc,
- fsContext: cfg.FSContext,
- fdTable: cfg.FDTable,
- p: cfg.Kernel.Platform.NewContext(),
- k: cfg.Kernel,
- ptraceTracees: make(map[*Task]struct{}),
- allowedCPUMask: cfg.AllowedCPUMask.Copy(),
- ioUsage: &usage.IO{},
- niceness: cfg.Niceness,
- netns: cfg.NetworkNamespaced,
- utsns: cfg.UTSNamespace,
- ipcns: cfg.IPCNamespace,
- abstractSockets: cfg.AbstractSocketNamespace,
- rseqCPU: -1,
- rseqAddr: cfg.RSeqAddr,
- rseqSignature: cfg.RSeqSignature,
- futexWaiter: futex.NewWaiter(),
- containerID: cfg.ContainerID,
+ runState: (*runApp)(nil),
+ interruptChan: make(chan struct{}, 1),
+ signalMask: cfg.SignalMask,
+ signalStack: arch.SignalStack{Flags: arch.SignalStackFlagDisable},
+ tc: *tc,
+ fsContext: cfg.FSContext,
+ fdTable: cfg.FDTable,
+ p: cfg.Kernel.Platform.NewContext(),
+ k: cfg.Kernel,
+ ptraceTracees: make(map[*Task]struct{}),
+ allowedCPUMask: cfg.AllowedCPUMask.Copy(),
+ ioUsage: &usage.IO{},
+ niceness: cfg.Niceness,
+ netns: cfg.NetworkNamespace,
+ utsns: cfg.UTSNamespace,
+ ipcns: cfg.IPCNamespace,
+ abstractSockets: cfg.AbstractSocketNamespace,
+ mountNamespaceVFS2: cfg.MountNamespaceVFS2,
+ rseqCPU: -1,
+ rseqAddr: cfg.RSeqAddr,
+ rseqSignature: cfg.RSeqSignature,
+ futexWaiter: futex.NewWaiter(),
+ containerID: cfg.ContainerID,
}
t.creds.Store(cfg.Credentials)
t.endStopCond.L = &t.tg.signalHandlers.mu
diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go
index 2bf3ce8a8..b02044ad2 100644
--- a/pkg/sentry/kernel/task_usermem.go
+++ b/pkg/sentry/kernel/task_usermem.go
@@ -30,7 +30,7 @@ var MAX_RW_COUNT = int(usermem.Addr(math.MaxInt32).RoundDown())
// Activate ensures that the task has an active address space.
func (t *Task) Activate() {
if mm := t.MemoryManager(); mm != nil {
- if err := mm.Activate(); err != nil {
+ if err := mm.Activate(t); err != nil {
panic("unable to activate mm: " + err.Error())
}
}
diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go
index 768e958d2..268f62e9d 100644
--- a/pkg/sentry/kernel/thread_group.go
+++ b/pkg/sentry/kernel/thread_group.go
@@ -256,7 +256,7 @@ type ThreadGroup struct {
tty *TTY
}
-// NewThreadGroup returns a new, empty thread group in PID namespace ns. The
+// NewThreadGroup returns a new, empty thread group in PID namespace pidns. The
// thread group leader will send its parent terminationSignal when it exits.
// The new thread group isn't visible to the system until a task has been
// created inside of it by a successful call to TaskSet.NewTask.
@@ -317,7 +317,9 @@ func (tg *ThreadGroup) release() {
for _, it := range its {
it.DestroyTimer()
}
- tg.mounts.DecRef()
+ if tg.mounts != nil {
+ tg.mounts.DecRef()
+ }
}
// forEachChildThreadGroupLocked indicates over all child ThreadGroups.
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-16 20:33:18 +0000