summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/kernel')
-rw-r--r--pkg/sentry/kernel/BUILD1
-rw-r--r--pkg/sentry/kernel/kcov.go40
-rw-r--r--pkg/sentry/kernel/kernel.go46
-rw-r--r--pkg/sentry/kernel/seccomp.go46
-rw-r--r--pkg/sentry/kernel/signalfd/BUILD1
-rw-r--r--pkg/sentry/kernel/signalfd/signalfd.go14
-rw-r--r--pkg/sentry/kernel/task.go4
-rw-r--r--pkg/sentry/kernel/threads.go7
-rw-r--r--pkg/sentry/kernel/vdso.go19
9 files changed, 108 insertions, 70 deletions
diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index 083071b5e..5de70aecb 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -204,7 +204,6 @@ go_library(
"//pkg/abi",
"//pkg/abi/linux",
"//pkg/amutex",
- "//pkg/binary",
"//pkg/bits",
"//pkg/bpf",
"//pkg/context",
diff --git a/pkg/sentry/kernel/kcov.go b/pkg/sentry/kernel/kcov.go
index d3e76ca7b..060c056df 100644
--- a/pkg/sentry/kernel/kcov.go
+++ b/pkg/sentry/kernel/kcov.go
@@ -89,7 +89,7 @@ func (kcov *Kcov) TaskWork(t *Task) {
kcov.mu.Lock()
defer kcov.mu.Unlock()
- if kcov.mode != linux.KCOV_TRACE_PC {
+ if kcov.mode != linux.KCOV_MODE_TRACE_PC {
return
}
@@ -146,7 +146,7 @@ func (kcov *Kcov) InitTrace(size uint64) error {
}
// EnableTrace performs the KCOV_ENABLE_TRACE ioctl.
-func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error {
+func (kcov *Kcov) EnableTrace(ctx context.Context, traceKind uint8) error {
t := TaskFromContext(ctx)
if t == nil {
panic("kcovInode.EnableTrace() cannot be used outside of a task goroutine")
@@ -160,9 +160,9 @@ func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error {
return syserror.EINVAL
}
- switch traceMode {
+ switch traceKind {
case linux.KCOV_TRACE_PC:
- kcov.mode = traceMode
+ kcov.mode = linux.KCOV_MODE_TRACE_PC
case linux.KCOV_TRACE_CMP:
// We do not support KCOV_MODE_TRACE_CMP.
return syserror.ENOTSUP
@@ -175,6 +175,7 @@ func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error {
}
kcov.owningTask = t
+ t.SetKcov(kcov)
t.RegisterWork(kcov)
// Clear existing coverage data; the task expects to read only coverage data
@@ -196,26 +197,35 @@ func (kcov *Kcov) DisableTrace(ctx context.Context) error {
if t != kcov.owningTask {
return syserror.EINVAL
}
- kcov.owningTask = nil
kcov.mode = linux.KCOV_MODE_INIT
- kcov.resetLocked()
+ kcov.owningTask = nil
+ kcov.mappable = nil
return nil
}
-// Reset is called when the owning task exits.
-func (kcov *Kcov) Reset() {
+// Clear resets the mode and clears the owning task and memory mapping for kcov.
+// It is called when the fd corresponding to kcov is closed. Note that the mode
+// needs to be set so that the next call to kcov.TaskWork() will exit early.
+func (kcov *Kcov) Clear() {
kcov.mu.Lock()
- kcov.resetLocked()
+ kcov.clearLocked()
kcov.mu.Unlock()
}
-// The kcov instance is reset when the owning task exits or when tracing is
-// disabled.
-func (kcov *Kcov) resetLocked() {
+func (kcov *Kcov) clearLocked() {
+ kcov.mode = linux.KCOV_MODE_INIT
kcov.owningTask = nil
- if kcov.mappable != nil {
- kcov.mappable = nil
- }
+ kcov.mappable = nil
+}
+
+// OnTaskExit is called when the owning task exits. It is similar to
+// kcov.Clear(), except the memory mapping is not cleared, so that the same
+// mapping can be used in the future if kcov is enabled again by another task.
+func (kcov *Kcov) OnTaskExit() {
+ kcov.mu.Lock()
+ kcov.mode = linux.KCOV_MODE_INIT
+ kcov.owningTask = nil
+ kcov.mu.Unlock()
}
// ConfigureMMap is called by the vfs.FileDescription for this kcov instance to
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index d6c21adb7..675506269 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -841,14 +841,16 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} {
if ctx.args.MountNamespaceVFS2 == nil {
return nil
}
- // MountNamespaceVFS2.Root() takes a reference on the root dirent for us.
- return ctx.args.MountNamespaceVFS2.Root()
+ root := ctx.args.MountNamespaceVFS2.Root()
+ root.IncRef()
+ return root
case vfs.CtxMountNamespace:
if ctx.k.globalInit == nil {
return nil
}
- // MountNamespaceVFS2 takes a reference for us.
- return ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
+ mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
+ mntns.IncRef()
+ return mntns
case fs.CtxDirentCacheLimiter:
return ctx.k.DirentCacheLimiter
case inet.CtxStack:
@@ -904,14 +906,13 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
if VFS2Enabled {
mntnsVFS2 = args.MountNamespaceVFS2
if mntnsVFS2 == nil {
- // MountNamespaceVFS2 adds a reference to the namespace, which is
- // transferred to the new process.
+ // Add a reference to the namespace, which is transferred to the new process.
mntnsVFS2 = k.globalInit.Leader().MountNamespaceVFS2()
+ mntnsVFS2.IncRef()
}
// Get the root directory from the MountNamespace.
root := mntnsVFS2.Root()
- // The call to newFSContext below will take a reference on root, so we
- // don't need to hold this one.
+ root.IncRef()
defer root.DecRef(ctx)
// Grab the working directory.
@@ -1648,16 +1649,16 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
if ctx.k.globalInit == nil {
return vfs.VirtualDentry{}
}
- mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
- defer mntns.DecRef(ctx)
- // Root() takes a reference on the root dirent for us.
- return mntns.Root()
+ root := ctx.k.GlobalInit().Leader().MountNamespaceVFS2().Root()
+ root.IncRef()
+ return root
case vfs.CtxMountNamespace:
if ctx.k.globalInit == nil {
return nil
}
- // MountNamespaceVFS2() takes a reference for us.
- return ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
+ mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2()
+ mntns.IncRef()
+ return mntns
case fs.CtxDirentCacheLimiter:
return ctx.k.DirentCacheLimiter
case inet.CtxStack:
@@ -1738,3 +1739,20 @@ func (k *Kernel) ShmMount() *vfs.Mount {
func (k *Kernel) SocketMount() *vfs.Mount {
return k.socketMount
}
+
+// Release releases resources owned by k.
+//
+// Precondition: This should only be called after the kernel is fully
+// initialized, e.g. after k.Start() has been called.
+func (k *Kernel) Release() {
+ ctx := k.SupervisorContext()
+ if VFS2Enabled {
+ k.hostMount.DecRef(ctx)
+ k.pipeMount.DecRef(ctx)
+ k.shmMount.DecRef(ctx)
+ k.socketMount.DecRef(ctx)
+ k.vfs.Release(ctx)
+ }
+ k.timekeeper.Destroy()
+ k.vdso.Release(ctx)
+}
diff --git a/pkg/sentry/kernel/seccomp.go b/pkg/sentry/kernel/seccomp.go
index c38c5a40c..387edfa91 100644
--- a/pkg/sentry/kernel/seccomp.go
+++ b/pkg/sentry/kernel/seccomp.go
@@ -18,7 +18,6 @@ import (
"syscall"
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/binary"
"gvisor.dev/gvisor/pkg/bpf"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/syserror"
@@ -27,25 +26,18 @@ import (
const maxSyscallFilterInstructions = 1 << 15
-// seccompData is equivalent to struct seccomp_data, which contains the data
-// passed to seccomp-bpf filters.
-type seccompData struct {
- // nr is the system call number.
- nr int32
-
- // arch is an AUDIT_ARCH_* value indicating the system call convention.
- arch uint32
-
- // instructionPointer is the value of the instruction pointer at the time
- // of the system call.
- instructionPointer uint64
-
- // args contains the first 6 system call arguments.
- args [6]uint64
-}
-
-func (d *seccompData) asBPFInput() bpf.Input {
- return bpf.InputBytes{binary.Marshal(nil, usermem.ByteOrder, d), usermem.ByteOrder}
+// dataAsBPFInput returns a serialized BPF program, only valid on the current task
+// goroutine.
+//
+// Note: this is called for every syscall, which is a very hot path.
+func dataAsBPFInput(t *Task, d *linux.SeccompData) bpf.Input {
+ buf := t.CopyScratchBuffer(d.SizeBytes())
+ d.MarshalUnsafe(buf)
+ return bpf.InputBytes{
+ Data: buf,
+ // Go-marshal always uses the native byte order.
+ Order: usermem.ByteOrder,
+ }
}
func seccompSiginfo(t *Task, errno, sysno int32, ip usermem.Addr) *arch.SignalInfo {
@@ -112,20 +104,20 @@ func (t *Task) checkSeccompSyscall(sysno int32, args arch.SyscallArguments, ip u
}
func (t *Task) evaluateSyscallFilters(sysno int32, args arch.SyscallArguments, ip usermem.Addr) uint32 {
- data := seccompData{
- nr: sysno,
- arch: t.tc.st.AuditNumber,
- instructionPointer: uint64(ip),
+ data := linux.SeccompData{
+ Nr: sysno,
+ Arch: t.tc.st.AuditNumber,
+ InstructionPointer: uint64(ip),
}
// data.args is []uint64 and args is []arch.SyscallArgument (uintptr), so
// we can't do any slicing tricks or even use copy/append here.
for i, arg := range args {
- if i >= len(data.args) {
+ if i >= len(data.Args) {
break
}
- data.args[i] = arg.Uint64()
+ data.Args[i] = arg.Uint64()
}
- input := data.asBPFInput()
+ input := dataAsBPFInput(t, &data)
ret := uint32(linux.SECCOMP_RET_ALLOW)
f := t.syscallFilters.Load()
diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD
index 3eb78e91b..76d472292 100644
--- a/pkg/sentry/kernel/signalfd/BUILD
+++ b/pkg/sentry/kernel/signalfd/BUILD
@@ -8,7 +8,6 @@ go_library(
visibility = ["//pkg/sentry:internal"],
deps = [
"//pkg/abi/linux",
- "//pkg/binary",
"//pkg/context",
"//pkg/sentry/fs",
"//pkg/sentry/fs/anon",
diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go
index b07e1c1bd..78f718cfe 100644
--- a/pkg/sentry/kernel/signalfd/signalfd.go
+++ b/pkg/sentry/kernel/signalfd/signalfd.go
@@ -17,7 +17,6 @@ package signalfd
import (
"gvisor.dev/gvisor/pkg/abi/linux"
- "gvisor.dev/gvisor/pkg/binary"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/fs/anon"
@@ -103,8 +102,7 @@ func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS
}
// Copy out the signal info using the specified format.
- var buf [128]byte
- binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{
+ infoNative := linux.SignalfdSiginfo{
Signo: uint32(info.Signo),
Errno: info.Errno,
Code: info.Code,
@@ -113,9 +111,13 @@ func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS
Status: info.Status(),
Overrun: uint32(info.Overrun()),
Addr: info.Addr(),
- })
- n, err := dst.CopyOut(ctx, buf[:])
- return int64(n), err
+ }
+ n, err := infoNative.WriteTo(dst.Writer(ctx))
+ if err == usermem.ErrEndOfIOSequence {
+ // Partial copy-out ok.
+ err = nil
+ }
+ return n, err
}
// Readiness implements waiter.Waitable.Readiness.
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index a436610c9..e90a19cfb 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -735,7 +735,6 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock {
func (t *Task) IsChrooted() bool {
if VFS2Enabled {
realRoot := t.mountNamespaceVFS2.Root()
- defer realRoot.DecRef(t)
root := t.fsContext.RootDirectoryVFS2()
defer root.DecRef(t)
return root != realRoot
@@ -868,7 +867,6 @@ func (t *Task) MountNamespace() *fs.MountNamespace {
func (t *Task) MountNamespaceVFS2() *vfs.MountNamespace {
t.mu.Lock()
defer t.mu.Unlock()
- t.mountNamespaceVFS2.IncRef()
return t.mountNamespaceVFS2
}
@@ -917,7 +915,7 @@ func (t *Task) SetKcov(k *Kcov) {
// ResetKcov clears the kcov instance associated with t.
func (t *Task) ResetKcov() {
if t.kcov != nil {
- t.kcov.Reset()
+ t.kcov.OnTaskExit()
t.kcov = nil
}
}
diff --git a/pkg/sentry/kernel/threads.go b/pkg/sentry/kernel/threads.go
index 5ae5906e8..fdadb52c0 100644
--- a/pkg/sentry/kernel/threads.go
+++ b/pkg/sentry/kernel/threads.go
@@ -265,6 +265,13 @@ func (ns *PIDNamespace) Tasks() []*Task {
return tasks
}
+// NumTasks returns the number of tasks in ns.
+func (ns *PIDNamespace) NumTasks() int {
+ ns.owner.mu.RLock()
+ defer ns.owner.mu.RUnlock()
+ return len(ns.tids)
+}
+
// ThreadGroups returns a snapshot of the thread groups in ns.
func (ns *PIDNamespace) ThreadGroups() []*ThreadGroup {
return ns.ThreadGroupsAppend(nil)
diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go
index e44a139b3..9bc452e67 100644
--- a/pkg/sentry/kernel/vdso.go
+++ b/pkg/sentry/kernel/vdso.go
@@ -17,7 +17,6 @@ package kernel
import (
"fmt"
- "gvisor.dev/gvisor/pkg/binary"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
@@ -28,6 +27,8 @@ import (
//
// They are exposed to the VDSO via a parameter page managed by VDSOParamPage,
// which also includes a sequence counter.
+//
+// +marshal
type vdsoParams struct {
monotonicReady uint64
monotonicBaseCycles int64
@@ -68,6 +69,13 @@ type VDSOParamPage struct {
// checked in state_test_util tests, causing this field to change across
// save / restore.
seq uint64
+
+ // copyScratchBuffer is a temporary buffer used to marshal the params before
+ // copying it to the real parameter page. The parameter page is typically
+ // updated at a moderate frequency of ~O(seconds) throughout the lifetime of
+ // the sentry, so reusing this buffer is a good tradeoff between memory
+ // usage and the cost of allocation.
+ copyScratchBuffer []byte
}
// NewVDSOParamPage returns a VDSOParamPage.
@@ -79,7 +87,11 @@ type VDSOParamPage struct {
// * VDSOParamPage must be the only writer to fr.
// * mfp.MemoryFile().MapInternal(fr) must return a single safemem.Block.
func NewVDSOParamPage(mfp pgalloc.MemoryFileProvider, fr memmap.FileRange) *VDSOParamPage {
- return &VDSOParamPage{mfp: mfp, fr: fr}
+ return &VDSOParamPage{
+ mfp: mfp,
+ fr: fr,
+ copyScratchBuffer: make([]byte, (*vdsoParams)(nil).SizeBytes()),
+ }
}
// access returns a mapping of the param page.
@@ -133,7 +145,8 @@ func (v *VDSOParamPage) Write(f func() vdsoParams) error {
// Get the new params.
p := f()
- buf := binary.Marshal(nil, usermem.ByteOrder, p)
+ buf := v.copyScratchBuffer[:p.SizeBytes()]
+ p.MarshalUnsafe(buf)
// Skip the sequence counter.
if _, err := safemem.Copy(paramPage.DropFirst(8), safemem.BlockFromSafeSlice(buf)); err != nil {
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-22 05:03:21 +0000