summaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
-rw-r--r--WORKSPACE4
-rw-r--r--benchmarks/workloads/ruby/Gemfile.lock2
-rw-r--r--pkg/ilist/list.go8
-rw-r--r--pkg/safemem/seq_test.go21
-rw-r--r--pkg/safemem/seq_unsafe.go3
-rw-r--r--pkg/seccomp/seccomp_test.go2
-rw-r--r--pkg/sentry/control/pprof.go34
-rw-r--r--pkg/sentry/fs/dirent_cache.go2
-rw-r--r--pkg/sentry/fs/host/util.go12
-rw-r--r--pkg/sentry/fs/inotify.go5
-rw-r--r--pkg/sentry/fs/proc/task.go126
-rw-r--r--pkg/sentry/fsimpl/host/BUILD27
-rw-r--r--pkg/sentry/fsimpl/host/default_file.go233
-rw-r--r--pkg/sentry/fsimpl/host/host.go286
-rw-r--r--pkg/sentry/fsimpl/host/util.go86
-rw-r--r--pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go2
-rw-r--r--pkg/sentry/fsimpl/kernfs/fd_impl_util.go18
-rw-r--r--pkg/sentry/fsimpl/kernfs/filesystem.go6
-rw-r--r--pkg/sentry/fsimpl/kernfs/inode_impl_util.go54
-rw-r--r--pkg/sentry/fsimpl/kernfs/kernfs.go7
-rw-r--r--pkg/sentry/fsimpl/proc/subtasks.go13
-rw-r--r--pkg/sentry/fsimpl/proc/task.go35
-rw-r--r--pkg/sentry/fsimpl/proc/task_files.go43
-rw-r--r--pkg/sentry/fsimpl/proc/tasks.go19
-rw-r--r--pkg/sentry/fsimpl/proc/tasks_test.go32
-rw-r--r--pkg/sentry/inet/namespace.go5
-rw-r--r--pkg/sentry/kernel/epoll/epoll_state.go13
-rw-r--r--pkg/sentry/kernel/task.go33
-rw-r--r--pkg/sentry/kernel/task_clone.go6
-rw-r--r--pkg/sentry/kernel/task_start.go4
-rw-r--r--pkg/sentry/platform/kvm/bluepill_arm64.s8
-rw-r--r--pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go23
-rw-r--r--pkg/sentry/platform/kvm/kvm.go32
-rw-r--r--pkg/sentry/platform/kvm/kvm_amd64.go32
-rw-r--r--pkg/sentry/platform/kvm/kvm_arm64.go32
-rw-r--r--pkg/sentry/platform/ring0/aarch64.go27
-rw-r--r--pkg/sentry/vfs/BUILD4
-rw-r--r--pkg/sentry/vfs/file_description_impl_util.go4
-rw-r--r--pkg/tcpip/checker/checker.go106
-rw-r--r--pkg/tcpip/network/fragmentation/fragmentation.go8
-rw-r--r--pkg/tcpip/stack/BUILD1
-rw-r--r--pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go39
-rw-r--r--pkg/tcpip/stack/ndp.go29
-rw-r--r--pkg/tcpip/stack/ndp_test.go39
-rw-r--r--pkg/tcpip/transport/tcp/BUILD1
-rw-r--r--pkg/tcpip/transport/tcp/connect.go6
-rw-r--r--pkg/tcpip/transport/tcp/connect_unsafe.go30
-rw-r--r--pkg/tcpip/transport/tcp/endpoint.go5
-rw-r--r--runsc/boot/controller.go13
-rw-r--r--runsc/cmd/debug.go64
-rw-r--r--runsc/container/console_test.go4
-rw-r--r--runsc/container/container_test.go2
-rw-r--r--runsc/sandbox/sandbox.go60
-rwxr-xr-xscripts/dev.sh1
-rw-r--r--test/packetdrill/BUILD44
-rw-r--r--test/packetdrill/accept_ack_drop.pkt27
-rw-r--r--test/packetdrill/defs.bzl10
-rw-r--r--test/packetdrill/linux/tcp_user_timeout.pkt39
-rw-r--r--test/packetdrill/listen_close_before_handshake_complete.pkt31
-rw-r--r--test/packetdrill/netstack/tcp_user_timeout.pkt38
-rw-r--r--test/packetdrill/no_rst_to_rst.pkt36
-rwxr-xr-xtest/packetdrill/packetdrill_test.sh4
-rw-r--r--test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt9
-rw-r--r--test/packetdrill/sanity_test.pkt7
-rw-r--r--test/packetdrill/tcp_defer_accept.pkt48
-rw-r--r--test/packetdrill/tcp_defer_accept_timeout.pkt48
-rw-r--r--test/runner/gtest/gtest.go7
-rw-r--r--test/syscalls/BUILD8
-rw-r--r--test/syscalls/linux/BUILD28
-rw-r--r--test/syscalls/linux/proc.cc21
-rw-r--r--test/syscalls/linux/proc_pid_oomscore.cc72
-rw-r--r--test/syscalls/linux/tcp_socket.cc14
-rw-r--r--test/syscalls/linux/tuntap_hostinet.cc1
-rw-r--r--vdso/vdso.cc4
74 files changed, 1915 insertions, 292 deletions
diff --git a/WORKSPACE b/WORKSPACE
index d1c2d8a24..d2bbadc63 100644
--- a/WORKSPACE
+++ b/WORKSPACE
@@ -78,9 +78,9 @@ rules_proto_toolchains()
# Load python dependencies.
git_repository(
name = "rules_python",
- commit = "94677401bc56ed5d756f50b441a6a5c7f735a6d4",
+ commit = "abc4869e02fe9b3866942e89f07b7341f830e805",
remote = "https://github.com/bazelbuild/rules_python.git",
- shallow_since = "1573842889 -0500",
+ shallow_since = "1583341286 -0500",
)
load("@rules_python//python:pip.bzl", "pip_import")
diff --git a/benchmarks/workloads/ruby/Gemfile.lock b/benchmarks/workloads/ruby/Gemfile.lock
index 855edf17f..17ebcbec3 100644
--- a/benchmarks/workloads/ruby/Gemfile.lock
+++ b/benchmarks/workloads/ruby/Gemfile.lock
@@ -33,7 +33,7 @@ GEM
rack (2.2.2)
rack-protection (2.0.5)
rack
- rake (12.3.2)
+ rake (12.3.3)
rbnacl (7.1.1)
ffi
redis (4.1.1)
diff --git a/pkg/ilist/list.go b/pkg/ilist/list.go
index f3a609b57..8f93e4d6d 100644
--- a/pkg/ilist/list.go
+++ b/pkg/ilist/list.go
@@ -169,8 +169,9 @@ func (l *List) InsertBefore(a, e Element) {
// Remove removes e from l.
func (l *List) Remove(e Element) {
- prev := ElementMapper{}.linkerFor(e).Prev()
- next := ElementMapper{}.linkerFor(e).Next()
+ linker := ElementMapper{}.linkerFor(e)
+ prev := linker.Prev()
+ next := linker.Next()
if prev != nil {
ElementMapper{}.linkerFor(prev).SetNext(next)
@@ -183,6 +184,9 @@ func (l *List) Remove(e Element) {
} else {
l.tail = prev
}
+
+ linker.SetNext(nil)
+ linker.SetPrev(nil)
}
// Entry is a default implementation of Linker. Users can add anonymous fields
diff --git a/pkg/safemem/seq_test.go b/pkg/safemem/seq_test.go
index eba4bb535..de34005e9 100644
--- a/pkg/safemem/seq_test.go
+++ b/pkg/safemem/seq_test.go
@@ -20,6 +20,27 @@ import (
"testing"
)
+func TestBlockSeqOfEmptyBlock(t *testing.T) {
+ bs := BlockSeqOf(Block{})
+ if !bs.IsEmpty() {
+ t.Errorf("BlockSeqOf(Block{}).IsEmpty(): got false, wanted true; BlockSeq is %v", bs)
+ }
+}
+
+func TestBlockSeqOfNonemptyBlock(t *testing.T) {
+ b := BlockFromSafeSlice(make([]byte, 1))
+ bs := BlockSeqOf(b)
+ if bs.IsEmpty() {
+ t.Fatalf("BlockSeqOf(non-empty Block).IsEmpty(): got true, wanted false; BlockSeq is %v", bs)
+ }
+ if head := bs.Head(); head != b {
+ t.Fatalf("BlockSeqOf(non-empty Block).Head(): got %v, wanted %v", head, b)
+ }
+ if tail := bs.Tail(); !tail.IsEmpty() {
+ t.Fatalf("BlockSeqOf(non-empty Block).Tail().IsEmpty(): got false, wanted true: tail is %v", tail)
+ }
+}
+
type blockSeqTest struct {
desc string
diff --git a/pkg/safemem/seq_unsafe.go b/pkg/safemem/seq_unsafe.go
index dcdfc9600..f5f0574f8 100644
--- a/pkg/safemem/seq_unsafe.go
+++ b/pkg/safemem/seq_unsafe.go
@@ -56,6 +56,9 @@ type BlockSeq struct {
// BlockSeqOf returns a BlockSeq representing the single Block b.
func BlockSeqOf(b Block) BlockSeq {
+ if b.length == 0 {
+ return BlockSeq{}
+ }
bs := BlockSeq{
data: b.start,
length: -1,
diff --git a/pkg/seccomp/seccomp_test.go b/pkg/seccomp/seccomp_test.go
index da5a5e4b2..88766f33b 100644
--- a/pkg/seccomp/seccomp_test.go
+++ b/pkg/seccomp/seccomp_test.go
@@ -451,7 +451,7 @@ func TestRandom(t *testing.T) {
}
}
- fmt.Printf("Testing filters: %v", syscallRules)
+ t.Logf("Testing filters: %v", syscallRules)
instrs, err := BuildProgram([]RuleSet{
RuleSet{
Rules: syscallRules,
diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go
index 151808911..663e51989 100644
--- a/pkg/sentry/control/pprof.go
+++ b/pkg/sentry/control/pprof.go
@@ -117,9 +117,9 @@ func (p *Profile) HeapProfile(o *ProfileOpts, _ *struct{}) error {
return nil
}
-// Goroutine is an RPC stub which dumps out the stack trace for all running
-// goroutines.
-func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error {
+// GoroutineProfile is an RPC stub which dumps out the stack trace for all
+// running goroutines.
+func (p *Profile) GoroutineProfile(o *ProfileOpts, _ *struct{}) error {
if len(o.FilePayload.Files) < 1 {
return errNoOutput
}
@@ -131,6 +131,34 @@ func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error {
return nil
}
+// BlockProfile is an RPC stub which dumps out the stack trace that led to
+// blocking on synchronization primitives.
+func (p *Profile) BlockProfile(o *ProfileOpts, _ *struct{}) error {
+ if len(o.FilePayload.Files) < 1 {
+ return errNoOutput
+ }
+ output := o.FilePayload.Files[0]
+ defer output.Close()
+ if err := pprof.Lookup("block").WriteTo(output, 0); err != nil {
+ return err
+ }
+ return nil
+}
+
+// MutexProfile is an RPC stub which dumps out the stack trace of holders of
+// contended mutexes.
+func (p *Profile) MutexProfile(o *ProfileOpts, _ *struct{}) error {
+ if len(o.FilePayload.Files) < 1 {
+ return errNoOutput
+ }
+ output := o.FilePayload.Files[0]
+ defer output.Close()
+ if err := pprof.Lookup("mutex").WriteTo(output, 0); err != nil {
+ return err
+ }
+ return nil
+}
+
// StartTrace is an RPC stub which starts collection of an execution trace.
func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error {
if len(o.FilePayload.Files) < 1 {
diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go
index 25514ace4..33de32c69 100644
--- a/pkg/sentry/fs/dirent_cache.go
+++ b/pkg/sentry/fs/dirent_cache.go
@@ -101,8 +101,6 @@ func (c *DirentCache) remove(d *Dirent) {
panic(fmt.Sprintf("trying to remove %v, which is not in the dirent cache", d))
}
c.list.Remove(d)
- d.SetPrev(nil)
- d.SetNext(nil)
d.DecRef()
c.currentSize--
if c.limit != nil {
diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go
index e37e687c6..7c60dc1db 100644
--- a/pkg/sentry/fs/host/util.go
+++ b/pkg/sentry/fs/host/util.go
@@ -24,7 +24,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/device"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
- ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/time"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -152,9 +152,9 @@ func unstableAttr(mo *superOperations, s *syscall.Stat_t) fs.UnstableAttr {
Usage: s.Blocks * 512,
Perms: fs.FilePermsFromMode(linux.FileMode(s.Mode)),
Owner: owner(mo, s),
- AccessTime: ktime.FromUnix(s.Atim.Sec, s.Atim.Nsec),
- ModificationTime: ktime.FromUnix(s.Mtim.Sec, s.Mtim.Nsec),
- StatusChangeTime: ktime.FromUnix(s.Ctim.Sec, s.Ctim.Nsec),
+ AccessTime: time.FromUnix(s.Atim.Sec, s.Atim.Nsec),
+ ModificationTime: time.FromUnix(s.Mtim.Sec, s.Mtim.Nsec),
+ StatusChangeTime: time.FromUnix(s.Ctim.Sec, s.Ctim.Nsec),
Links: uint64(s.Nlink),
}
}
@@ -165,6 +165,8 @@ type dirInfo struct {
bufp int // location of next record in buf.
}
+// LINT.IfChange
+
// isBlockError unwraps os errors and checks if they are caused by EAGAIN or
// EWOULDBLOCK. This is so they can be transformed into syserror.ErrWouldBlock.
func isBlockError(err error) bool {
@@ -177,6 +179,8 @@ func isBlockError(err error) bool {
return false
}
+// LINT.ThenChange(../../fsimpl/host/util.go)
+
func hostEffectiveKIDs() (uint32, []uint32, error) {
gids, err := os.Getgroups()
if err != nil {
diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go
index 928c90aa0..e3a715c1f 100644
--- a/pkg/sentry/fs/inotify.go
+++ b/pkg/sentry/fs/inotify.go
@@ -143,7 +143,10 @@ func (i *Inotify) Read(ctx context.Context, _ *File, dst usermem.IOSequence, _ i
}
var writeLen int64
- for event := i.events.Front(); event != nil; event = event.Next() {
+ for it := i.events.Front(); it != nil; {
+ event := it
+ it = it.Next()
+
// Does the buffer have enough remaining space to hold the event we're
// about to write out?
if dst.NumBytes() < int64(event.sizeOf()) {
diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go
index 8ab8d8a02..4e9b0fc00 100644
--- a/pkg/sentry/fs/proc/task.go
+++ b/pkg/sentry/fs/proc/task.go
@@ -72,24 +72,26 @@ var _ fs.InodeOperations = (*taskDir)(nil)
// newTaskDir creates a new proc task entry.
func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode {
contents := map[string]*fs.Inode{
- "auxv": newAuxvec(t, msrc),
- "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
- "comm": newComm(t, msrc),
- "environ": newExecArgInode(t, msrc, environExecArg),
- "exe": newExe(t, msrc),
- "fd": newFdDir(t, msrc),
- "fdinfo": newFdInfoDir(t, msrc),
- "gid_map": newGIDMap(t, msrc),
- "io": newIO(t, msrc, isThreadGroup),
- "maps": newMaps(t, msrc),
- "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
- "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
- "ns": newNamespaceDir(t, msrc),
- "smaps": newSmaps(t, msrc),
- "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
- "statm": newStatm(t, msrc),
- "status": newStatus(t, msrc, p.pidns),
- "uid_map": newUIDMap(t, msrc),
+ "auxv": newAuxvec(t, msrc),
+ "cmdline": newExecArgInode(t, msrc, cmdlineExecArg),
+ "comm": newComm(t, msrc),
+ "environ": newExecArgInode(t, msrc, environExecArg),
+ "exe": newExe(t, msrc),
+ "fd": newFdDir(t, msrc),
+ "fdinfo": newFdInfoDir(t, msrc),
+ "gid_map": newGIDMap(t, msrc),
+ "io": newIO(t, msrc, isThreadGroup),
+ "maps": newMaps(t, msrc),
+ "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
+ "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
+ "ns": newNamespaceDir(t, msrc),
+ "oom_score": newOOMScore(t, msrc),
+ "oom_score_adj": newOOMScoreAdj(t, msrc),
+ "smaps": newSmaps(t, msrc),
+ "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns),
+ "statm": newStatm(t, msrc),
+ "status": newStatus(t, msrc, p.pidns),
+ "uid_map": newUIDMap(t, msrc),
}
if isThreadGroup {
contents["task"] = p.newSubtasks(t, msrc)
@@ -796,4 +798,92 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc
return int64(n), err
}
+// newOOMScore returns a oom_score file. It is a stub that always returns 0.
+// TODO(gvisor.dev/issue/1967)
+func newOOMScore(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ return newStaticProcInode(t, msrc, []byte("0\n"))
+}
+
+// oomScoreAdj is a file containing the oom_score adjustment for a task.
+//
+// +stateify savable
+type oomScoreAdj struct {
+ fsutil.SimpleFileInode
+
+ t *kernel.Task
+}
+
+// +stateify savable
+type oomScoreAdjFile struct {
+ fsutil.FileGenericSeek `state:"nosave"`
+ fsutil.FileNoIoctl `state:"nosave"`
+ fsutil.FileNoMMap `state:"nosave"`
+ fsutil.FileNoSplice `state:"nosave"`
+ fsutil.FileNoopFlush `state:"nosave"`
+ fsutil.FileNoopFsync `state:"nosave"`
+ fsutil.FileNoopRelease `state:"nosave"`
+ fsutil.FileNotDirReaddir `state:"nosave"`
+ fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+ waiter.AlwaysReady `state:"nosave"`
+
+ t *kernel.Task
+}
+
+// newOOMScoreAdj returns a oom_score_adj file.
+func newOOMScoreAdj(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+ i := &oomScoreAdj{
+ SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC),
+ t: t,
+ }
+ return newProcInode(t, i, msrc, fs.SpecialFile, t)
+}
+
+// Truncate implements fs.InodeOperations.Truncate. Truncate is called when
+// O_TRUNC is specified for any kind of existing Dirent but is not called via
+// (f)truncate for proc files.
+func (*oomScoreAdj) Truncate(context.Context, *fs.Inode, int64) error {
+ return nil
+}
+
+// GetFile implements fs.InodeOperations.GetFile.
+func (o *oomScoreAdj) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) {
+ return fs.NewFile(ctx, dirent, flags, &oomScoreAdjFile{t: o.t}), nil
+}
+
+// Read implements fs.FileOperations.Read.
+func (f *oomScoreAdjFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) {
+ if offset != 0 {
+ return 0, io.EOF
+ }
+ adj, err := f.t.OOMScoreAdj()
+ if err != nil {
+ return 0, err
+ }
+ adjBytes := []byte(strconv.FormatInt(int64(adj), 10) + "\n")
+ n, err := dst.CopyOut(ctx, adjBytes)
+ return int64(n), err
+}
+
+// Write implements fs.FileOperations.Write.
+func (f *oomScoreAdjFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) {
+ if src.NumBytes() == 0 {
+ return 0, nil
+ }
+
+ // Limit input size so as not to impact performance if input size is large.
+ src = src.TakeFirst(usermem.PageSize - 1)
+
+ var v int32
+ n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts)
+ if err != nil {
+ return 0, err
+ }
+
+ if err := f.t.SetOOMScoreAdj(v); err != nil {
+ return 0, err
+ }
+
+ return n, nil
+}
+
// LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go)
diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD
new file mode 100644
index 000000000..731f192b3
--- /dev/null
+++ b/pkg/sentry/fsimpl/host/BUILD
@@ -0,0 +1,27 @@
+load("//tools:defs.bzl", "go_library")
+
+licenses(["notice"])
+
+go_library(
+ name = "host",
+ srcs = [
+ "default_file.go",
+ "host.go",
+ "util.go",
+ ],
+ deps = [
+ "//pkg/abi/linux",
+ "//pkg/context",
+ "//pkg/log",
+ "//pkg/refs",
+ "//pkg/safemem",
+ "//pkg/sentry/fsimpl/kernfs",
+ "//pkg/sentry/kernel/auth",
+ "//pkg/sentry/memmap",
+ "//pkg/sentry/vfs",
+ "//pkg/sync",
+ "//pkg/syserror",
+ "//pkg/usermem",
+ "@org_golang_x_sys//unix:go_default_library",
+ ],
+)
diff --git a/pkg/sentry/fsimpl/host/default_file.go b/pkg/sentry/fsimpl/host/default_file.go
new file mode 100644
index 000000000..172cdb161
--- /dev/null
+++ b/pkg/sentry/fsimpl/host/default_file.go
@@ -0,0 +1,233 @@
+// 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 host
+
+import (
+ "math"
+ "syscall"
+
+ "golang.org/x/sys/unix"
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/safemem"
+ "gvisor.dev/gvisor/pkg/sentry/memmap"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+)
+
+// defaultFileFD implements FileDescriptionImpl for non-socket, non-TTY files.
+type defaultFileFD struct {
+ fileDescription
+
+ // canMap specifies whether we allow the file to be memory mapped.
+ canMap bool
+
+ // mu protects the fields below.
+ mu sync.Mutex
+
+ // offset specifies the current file offset.
+ offset int64
+}
+
+// TODO(gvisor.dev/issue/1672): Implement Waitable interface.
+
+// PRead implements FileDescriptionImpl.
+func (f *defaultFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
+ // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
+ if f.inode.isStream {
+ return 0, syserror.ESPIPE
+ }
+
+ return readFromHostFD(ctx, f.inode.hostFD, dst, offset, int(opts.Flags))
+}
+
+// Read implements FileDescriptionImpl.
+func (f *defaultFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+ // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
+ if f.inode.isStream {
+ // These files can't be memory mapped, assert this.
+ if f.canMap {
+ panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped")
+ }
+
+ f.mu.Lock()
+ n, err := readFromHostFD(ctx, f.inode.hostFD, dst, -1, int(opts.Flags))
+ f.mu.Unlock()
+ if isBlockError(err) {
+ // If we got any data at all, return it as a "completed" partial read
+ // rather than retrying until complete.
+ if n != 0 {
+ err = nil
+ } else {
+ err = syserror.ErrWouldBlock
+ }
+ }
+ return n, err
+ }
+ // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so.
+ f.mu.Lock()
+ n, err := readFromHostFD(ctx, f.inode.hostFD, dst, f.offset, int(opts.Flags))
+ f.offset += n
+ f.mu.Unlock()
+ return n, err
+}
+
+func readFromHostFD(ctx context.Context, fd int, dst usermem.IOSequence, offset int64, flags int) (int64, error) {
+ if flags&^(linux.RWF_VALID) != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ reader := safemem.FromVecReaderFunc{
+ func(srcs [][]byte) (int64, error) {
+ n, err := unix.Preadv2(fd, srcs, offset, flags)
+ return int64(n), err
+ },
+ }
+ n, err := dst.CopyOutFrom(ctx, reader)
+ return int64(n), err
+}
+
+// PWrite implements FileDescriptionImpl.
+func (f *defaultFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+ // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
+ if f.inode.isStream {
+ return 0, syserror.ESPIPE
+ }
+
+ return writeToHostFD(ctx, f.inode.hostFD, src, offset, int(opts.Flags))
+}
+
+// Write implements FileDescriptionImpl.
+func (f *defaultFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+ // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null.
+ if f.inode.isStream {
+ // These files can't be memory mapped, assert this.
+ if f.canMap {
+ panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped")
+ }
+
+ f.mu.Lock()
+ n, err := writeToHostFD(ctx, f.inode.hostFD, src, -1, int(opts.Flags))
+ f.mu.Unlock()
+ if isBlockError(err) {
+ err = syserror.ErrWouldBlock
+ }
+ return n, err
+ }
+ // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so.
+ // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file.
+ f.mu.Lock()
+ n, err := writeToHostFD(ctx, f.inode.hostFD, src, f.offset, int(opts.Flags))
+ f.offset += n
+ f.mu.Unlock()
+ return n, err
+}
+
+func writeToHostFD(ctx context.Context, fd int, src usermem.IOSequence, offset int64, flags int) (int64, error) {
+ if flags&^(linux.RWF_VALID) != 0 {
+ return 0, syserror.EOPNOTSUPP
+ }
+
+ writer := safemem.FromVecWriterFunc{
+ func(srcs [][]byte) (int64, error) {
+ n, err := unix.Pwritev2(fd, srcs, offset, flags)
+ return int64(n), err
+ },
+ }
+ n, err := src.CopyInTo(ctx, writer)
+ return int64(n), err
+}
+
+// Seek implements FileDescriptionImpl.
+//
+// Note that we do not support seeking on directories, since we do not even
+// allow directory fds to be imported at all.
+func (f *defaultFileFD) Seek(_ context.Context, offset int64, whence int32) (int64, error) {
+ // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null.
+ if f.inode.isStream {
+ return 0, syserror.ESPIPE
+ }
+
+ f.mu.Lock()
+ defer f.mu.Unlock()
+
+ switch whence {
+ case linux.SEEK_SET:
+ if offset < 0 {
+ return f.offset, syserror.EINVAL
+ }
+ f.offset = offset
+
+ case linux.SEEK_CUR:
+ // Check for overflow. Note that underflow cannot occur, since f.offset >= 0.
+ if offset > math.MaxInt64-f.offset {
+ return f.offset, syserror.EOVERFLOW
+ }
+ if f.offset+offset < 0 {
+ return f.offset, syserror.EINVAL
+ }
+ f.offset += offset
+
+ case linux.SEEK_END:
+ var s syscall.Stat_t
+ if err := syscall.Fstat(f.inode.hostFD, &s); err != nil {
+ return f.offset, err
+ }
+ size := s.Size
+
+ // Check for overflow. Note that underflow cannot occur, since size >= 0.
+ if offset > math.MaxInt64-size {
+ return f.offset, syserror.EOVERFLOW
+ }
+ if size+offset < 0 {
+ return f.offset, syserror.EINVAL
+ }
+ f.offset = size + offset
+
+ case linux.SEEK_DATA, linux.SEEK_HOLE:
+ // Modifying the offset in the host file table should not matter, since
+ // this is the only place where we use it.
+ //
+ // For reading and writing, we always rely on our internal offset.
+ n, err := unix.Seek(f.inode.hostFD, offset, int(whence))
+ if err != nil {
+ return f.offset, err
+ }
+ f.offset = n
+
+ default:
+ // Invalid whence.
+ return f.offset, syserror.EINVAL
+ }
+
+ return f.offset, nil
+}
+
+// Sync implements FileDescriptionImpl.
+func (f *defaultFileFD) Sync(context.Context) error {
+ // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything.
+ return unix.Fsync(f.inode.hostFD)
+}
+
+// ConfigureMMap implements FileDescriptionImpl.
+func (f *defaultFileFD) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error {
+ if !f.canMap {
+ return syserror.ENODEV
+ }
+ // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface.
+ return syserror.ENODEV
+}
diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go
new file mode 100644
index 000000000..c205e6a0b
--- /dev/null
+++ b/pkg/sentry/fsimpl/host/host.go
@@ -0,0 +1,286 @@
+// 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 host provides a filesystem implementation for host files imported as
+// file descriptors.
+package host
+
+import (
+ "errors"
+ "fmt"
+ "syscall"
+
+ "golang.org/x/sys/unix"
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/log"
+ "gvisor.dev/gvisor/pkg/refs"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+// filesystem implements vfs.FilesystemImpl.
+type filesystem struct {
+ kernfs.Filesystem
+}
+
+// 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) {
+ // Must be importing to a mount of host.filesystem.
+ fs, ok := mnt.Filesystem().Impl().(*filesystem)
+ if !ok {
+ return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl())
+ }
+
+ // Retrieve metadata.
+ var s syscall.Stat_t
+ if err := syscall.Fstat(hostFD, &s); err != nil {
+ return nil, err
+ }
+
+ fileMode := linux.FileMode(s.Mode)
+ fileType := fileMode.FileType()
+ // Pipes, character devices, and sockets can return EWOULDBLOCK for
+ // operations that would block.
+ isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK
+
+ i := &inode{
+ hostFD: hostFD,
+ isStream: isStream,
+ isTTY: isTTY,
+ ino: fs.NextIno(),
+ mode: fileMode,
+ uid: ownerUID,
+ gid: ownerGID,
+ }
+
+ d := &kernfs.Dentry{}
+ d.Init(i)
+ // i.open will take a reference on d.
+ defer d.DecRef()
+
+ return i.open(d.VFSDentry(), mnt)
+}
+
+// inode implements kernfs.Inode.
+type inode struct {
+ kernfs.InodeNotDirectory
+ kernfs.InodeNotSymlink
+
+ // When the reference count reaches zero, the host fd is closed.
+ refs.AtomicRefCount
+
+ // hostFD contains the host fd that this file was originally created from,
+ // which must be available at time of restore.
+ //
+ // This field is initialized at creation time and is immutable.
+ hostFD int
+
+ // isStream is true if the host fd points to a file representing a stream,
+ // e.g. a socket or a pipe. Such files are not seekable and can return
+ // EWOULDBLOCK for I/O operations.
+ //
+ // This field is initialized at creation time and is immutable.
+ isStream bool
+
+ // isTTY is true if this file represents a TTY.
+ //
+ // This field is initialized at creation time and is immutable.
+ isTTY bool
+
+ // ino is an inode number unique within this filesystem.
+ ino uint64
+
+ // mu protects the inode metadata below.
+ mu 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 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
+}
+
+// Note that these flags may become out of date, since they can be modified
+// on the host, e.g. with fcntl.
+func fileFlagsFromHostFD(fd int) (int, error) {
+ flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0)
+ if err != nil {
+ log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err)
+ return 0, err
+ }
+ // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags.
+ flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND
+ return flags, nil
+}
+
+// CheckPermissions implements kernfs.Inode.
+func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error {
+ return vfs.GenericCheckPermissions(creds, atx, false /* isDir */, uint16(i.mode), i.uid, i.gid)
+}
+
+// Mode implements kernfs.Inode.
+func (i *inode) Mode() linux.FileMode {
+ return i.mode
+}
+
+// Stat implements kernfs.Inode.
+func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) {
+ var s unix.Statx_t
+ if err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(opts.Mask), &s); err != nil {
+ return linux.Statx{}, err
+ }
+ ls := unixToLinuxStatx(s)
+
+ // Use our own internal inode number and file owner.
+ //
+ // 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)
+
+ // Update file mode from the host.
+ i.mode = linux.FileMode(ls.Mode)
+
+ return ls, nil
+}
+
+// SetStat implements kernfs.Inode.
+func (i *inode) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error {
+ s := opts.Stat
+
+ m := s.Mask
+ if m == 0 {
+ return nil
+ }
+ if m&(linux.STATX_UID|linux.STATX_GID) != 0 {
+ return syserror.EPERM
+ }
+ if m&linux.STATX_MODE != 0 {
+ if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil {
+ return err
+ }
+ i.mode = linux.FileMode(s.Mode)
+ }
+ if m&linux.STATX_SIZE != 0 {
+ if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil {
+ return err
+ }
+ }
+ if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 {
+ timestamps := []unix.Timespec{
+ toTimespec(s.Atime, m&linux.STATX_ATIME == 0),
+ toTimespec(s.Mtime, m&linux.STATX_MTIME == 0),
+ }
+ if err := unix.UtimesNanoAt(i.hostFD, "", timestamps, unix.AT_EMPTY_PATH); err != nil {
+ return err
+ }
+ }
+ return nil
+}
+
+// DecRef implements kernfs.Inode.
+func (i *inode) DecRef() {
+ i.AtomicRefCount.DecRefWithDestructor(i.Destroy)
+}
+
+// Destroy implements kernfs.Inode.
+func (i *inode) Destroy() {
+ if err := unix.Close(i.hostFD); err != nil {
+ log.Warningf("failed to close host fd %d: %v", i.hostFD, err)
+ }
+}
+
+// Open implements kernfs.Inode.
+func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
+ return i.open(vfsd, rp.Mount())
+}
+
+func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) {
+
+ fileType := i.mode.FileType()
+ if fileType == syscall.S_IFSOCK {
+ if i.isTTY {
+ return nil, errors.New("cannot use host socket as TTY")
+ }
+ // TODO(gvisor.dev/issue/1672): support importing sockets.
+ return nil, errors.New("importing host sockets not supported")
+ }
+
+ if i.isTTY {
+ // TODO(gvisor.dev/issue/1672): support importing host fd as TTY.
+ return nil, errors.New("importing host fd as TTY not supported")
+ }
+
+ // 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.
+ fd := &defaultFileFD{
+ fileDescription: fileDescription{
+ inode: i,
+ },
+ canMap: canMap(uint32(fileType)),
+ mu: sync.Mutex{},
+ offset: 0,
+ }
+
+ vfsfd := &fd.vfsfd
+ flags, err := fileFlagsFromHostFD(i.hostFD)
+ if err != nil {
+ return nil, err
+ }
+
+ if err := vfsfd.Init(fd, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil {
+ return nil, err
+ }
+ return vfsfd, nil
+}
+
+// fileDescription is embedded by host fd implementations of FileDescriptionImpl.
+type fileDescription struct {
+ vfsfd vfs.FileDescription
+ vfs.FileDescriptionDefaultImpl
+
+ // inode is vfsfd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode), but
+ // cached to reduce indirections and casting. fileDescription does not hold
+ // a reference on the inode through the inode field (since one is already
+ // held via the Dentry).
+ //
+ // inode is immutable after fileDescription creation.
+ inode *inode
+}
+
+// SetStat implements vfs.FileDescriptionImpl.
+func (f *fileDescription) SetStat(_ context.Context, opts vfs.SetStatOptions) error {
+ return f.inode.SetStat(nil, opts)
+}
+
+// Stat implements vfs.FileDescriptionImpl.
+func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) {
+ return f.inode.Stat(nil, opts)
+}
+
+// Release implements vfs.FileDescriptionImpl.
+func (f *fileDescription) Release() {
+ // noop
+}
diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go
new file mode 100644
index 000000000..e1ccacb4d
--- /dev/null
+++ b/pkg/sentry/fsimpl/host/util.go
@@ -0,0 +1,86 @@
+// 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 host
+
+import (
+ "syscall"
+
+ "golang.org/x/sys/unix"
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+func toTimespec(ts linux.StatxTimestamp, omit bool) unix.Timespec {
+ if omit {
+ return unix.Timespec{
+ Sec: 0,
+ Nsec: unix.UTIME_OMIT,
+ }
+ }
+ return unix.Timespec{
+ Sec: int64(ts.Sec),
+ Nsec: int64(ts.Nsec),
+ }
+}
+
+func unixToLinuxStatx(s unix.Statx_t) linux.Statx {
+ return linux.Statx{
+ Mask: s.Mask,
+ Blksize: s.Blksize,
+ Attributes: s.Attributes,
+ Nlink: s.Nlink,
+ UID: s.Uid,
+ GID: s.Gid,
+ Mode: s.Mode,
+ Ino: s.Ino,
+ Size: s.Size,
+ Blocks: s.Blocks,
+ AttributesMask: s.Attributes_mask,
+ Atime: unixToLinuxStatxTimestamp(s.Atime),
+ Btime: unixToLinuxStatxTimestamp(s.Btime),
+ Ctime: unixToLinuxStatxTimestamp(s.Ctime),
+ Mtime: unixToLinuxStatxTimestamp(s.Mtime),
+ RdevMajor: s.Rdev_major,
+ RdevMinor: s.Rdev_minor,
+ DevMajor: s.Dev_major,
+ DevMinor: s.Dev_minor,
+ }
+}
+
+func unixToLinuxStatxTimestamp(ts unix.StatxTimestamp) linux.StatxTimestamp {
+ return linux.StatxTimestamp{Sec: ts.Sec, Nsec: ts.Nsec}
+}
+
+// wouldBlock returns true for file types that can return EWOULDBLOCK
+// for blocking operations, e.g. pipes, character devices, and sockets.
+func wouldBlock(fileType uint32) bool {
+ return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK
+}
+
+// canMap returns true if a file with fileType is allowed to be memory mapped.
+// This is ported over from VFS1, but it's probably not the best way for us
+// to check if a file can be memory mapped.
+func canMap(fileType uint32) bool {
+ // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()).
+ //
+ // TODO(b/38213152): Some obscure character devices can be mapped.
+ return fileType == syscall.S_IFREG
+}
+
+// isBlockError checks if an error is EAGAIN or EWOULDBLOCK.
+// If so, they can be transformed into syserror.ErrWouldBlock.
+func isBlockError(err error) bool {
+ return err == syserror.EAGAIN || err == syserror.EWOULDBLOCK
+}
diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
index d092ccb2a..1c026f4d8 100644
--- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
+++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go
@@ -122,7 +122,7 @@ func (fd *DynamicBytesFD) Release() {}
// Stat implements vfs.FileDescriptionImpl.Stat.
func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
fs := fd.vfsfd.VirtualDentry().Mount().Filesystem()
- return fd.inode.Stat(fs), nil
+ return fd.inode.Stat(fs, opts)
}
// SetStat implements vfs.FileDescriptionImpl.SetStat.
diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
index 5650512e0..da821d524 100644
--- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
+++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
@@ -107,9 +107,13 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent
fs.mu.Lock()
defer fs.mu.Unlock()
+ opts := vfs.StatOptions{Mask: linux.STATX_INO}
// Handle ".".
if fd.off == 0 {
- stat := fd.inode().Stat(vfsFS)
+ stat, err := fd.inode().Stat(vfsFS, opts)
+ if err != nil {
+ return err
+ }
dirent := vfs.Dirent{
Name: ".",
Type: linux.DT_DIR,
@@ -125,7 +129,10 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent
// Handle "..".
if fd.off == 1 {
parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode
- stat := parentInode.Stat(vfsFS)
+ stat, err := parentInode.Stat(vfsFS, opts)
+ if err != nil {
+ return err
+ }
dirent := vfs.Dirent{
Name: "..",
Type: linux.FileMode(stat.Mode).DirentType(),
@@ -146,7 +153,10 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent
childIdx := fd.off - 2
for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() {
inode := it.Dentry.Impl().(*Dentry).inode
- stat := inode.Stat(vfsFS)
+ stat, err := inode.Stat(vfsFS, opts)
+ if err != nil {
+ return err
+ }
dirent := vfs.Dirent{
Name: it.Name,
Type: linux.FileMode(stat.Mode).DirentType(),
@@ -190,7 +200,7 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int
func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) {
fs := fd.filesystem()
inode := fd.inode()
- return inode.Stat(fs), nil
+ return inode.Stat(fs, opts)
}
// SetStat implements vfs.FileDescriptionImpl.SetStat.
diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go
index 292f58afd..1d7e04ad4 100644
--- a/pkg/sentry/fsimpl/kernfs/filesystem.go
+++ b/pkg/sentry/fsimpl/kernfs/filesystem.go
@@ -12,10 +12,10 @@
// See the License for the specific language governing permissions and
// limitations under the License.
-// This file implements vfs.FilesystemImpl for kernfs.
-
package kernfs
+// This file implements vfs.FilesystemImpl for kernfs.
+
import (
"fmt"
@@ -634,7 +634,7 @@ func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf
if err != nil {
return linux.Statx{}, err
}
- return inode.Stat(fs.VFSFilesystem()), nil
+ return inode.Stat(fs.VFSFilesystem(), opts)
}
// StatFSAt implements vfs.FilesystemImpl.StatFSAt.
diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
index 099d70a16..d50018b18 100644
--- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
+++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go
@@ -36,20 +36,20 @@ type InodeNoopRefCount struct {
}
// IncRef implements Inode.IncRef.
-func (n *InodeNoopRefCount) IncRef() {
+func (InodeNoopRefCount) IncRef() {
}
// DecRef implements Inode.DecRef.
-func (n *InodeNoopRefCount) DecRef() {
+func (InodeNoopRefCount) DecRef() {
}
// TryIncRef implements Inode.TryIncRef.
-func (n *InodeNoopRefCount) TryIncRef() bool {
+func (InodeNoopRefCount) TryIncRef() bool {
return true
}
// Destroy implements Inode.Destroy.
-func (n *InodeNoopRefCount) Destroy() {
+func (InodeNoopRefCount) Destroy() {
}
// InodeDirectoryNoNewChildren partially implements the Inode interface.
@@ -58,27 +58,27 @@ func (n *InodeNoopRefCount) Destroy() {
type InodeDirectoryNoNewChildren struct{}
// NewFile implements Inode.NewFile.
-func (*InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) {
+func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) {
return nil, syserror.EPERM
}
// NewDir implements Inode.NewDir.
-func (*InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) {
+func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) {
return nil, syserror.EPERM
}
// NewLink implements Inode.NewLink.
-func (*InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) {
+func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) {
return nil, syserror.EPERM
}
// NewSymlink implements Inode.NewSymlink.
-func (*InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) {
+func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) {
return nil, syserror.EPERM
}
// NewNode implements Inode.NewNode.
-func (*InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) {
+func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) {
return nil, syserror.EPERM
}
@@ -90,62 +90,62 @@ type InodeNotDirectory struct {
}
// HasChildren implements Inode.HasChildren.
-func (*InodeNotDirectory) HasChildren() bool {
+func (InodeNotDirectory) HasChildren() bool {
return false
}
// NewFile implements Inode.NewFile.
-func (*InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) {
+func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) {
panic("NewFile called on non-directory inode")
}
// NewDir implements Inode.NewDir.
-func (*InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) {
+func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) {
panic("NewDir called on non-directory inode")
}
// NewLink implements Inode.NewLinkink.
-func (*InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) {
+func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) {
panic("NewLink called on non-directory inode")
}
// NewSymlink implements Inode.NewSymlink.
-func (*InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) {
+func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) {
panic("NewSymlink called on non-directory inode")
}
// NewNode implements Inode.NewNode.
-func (*InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) {
+func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) {
panic("NewNode called on non-directory inode")
}
// Unlink implements Inode.Unlink.
-func (*InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error {
+func (InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error {
panic("Unlink called on non-directory inode")
}
// RmDir implements Inode.RmDir.
-func (*InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error {
+func (InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error {
panic("RmDir called on non-directory inode")
}
// Rename implements Inode.Rename.
-func (*InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) {
+func (InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) {
panic("Rename called on non-directory inode")
}
// Lookup implements Inode.Lookup.
-func (*InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
+func (InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
panic("Lookup called on non-directory inode")
}
// IterDirents implements Inode.IterDirents.
-func (*InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) {
+func (InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) {
panic("IterDirents called on non-directory inode")
}
// Valid implements Inode.Valid.
-func (*InodeNotDirectory) Valid(context.Context) bool {
+func (InodeNotDirectory) Valid(context.Context) bool {
return true
}
@@ -157,17 +157,17 @@ func (*InodeNotDirectory) Valid(context.Context) bool {
type InodeNoDynamicLookup struct{}
// Lookup implements Inode.Lookup.
-func (*InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
+func (InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) {
return nil, syserror.ENOENT
}
// IterDirents implements Inode.IterDirents.
-func (*InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) {
+func (InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) {
return offset, nil
}
// Valid implements Inode.Valid.
-func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool {
+func (InodeNoDynamicLookup) Valid(ctx context.Context) bool {
return true
}
@@ -177,7 +177,7 @@ func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool {
type InodeNotSymlink struct{}
// Readlink implements Inode.Readlink.
-func (*InodeNotSymlink) Readlink(context.Context) (string, error) {
+func (InodeNotSymlink) Readlink(context.Context) (string, error) {
return "", syserror.EINVAL
}
@@ -219,7 +219,7 @@ func (a *InodeAttrs) Mode() linux.FileMode {
// Stat partially implements Inode.Stat. Note that this function doesn't provide
// all the stat fields, and the embedder should consider extending the result
// with filesystem-specific fields.
-func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx {
+func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) {
var stat linux.Statx
stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK
stat.Ino = atomic.LoadUint64(&a.ino)
@@ -230,7 +230,7 @@ func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx {
// TODO: Implement other stat fields like timestamps.
- return stat
+ return stat, nil
}
// SetStat implements Inode.SetStat.
diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go
index c74fa999b..a8ab2a2ba 100644
--- a/pkg/sentry/fsimpl/kernfs/kernfs.go
+++ b/pkg/sentry/fsimpl/kernfs/kernfs.go
@@ -176,8 +176,6 @@ type Dentry struct {
vfsd vfs.Dentry
inode Inode
- refs uint64
-
// flags caches useful information about the dentry from the inode. See the
// dflags* consts above. Must be accessed by atomic ops.
flags uint32
@@ -302,7 +300,8 @@ type Inode interface {
// this inode. The returned file description should hold a reference on the
// inode for its lifetime.
//
- // Precondition: !rp.Done(). vfsd.Impl() must be a kernfs Dentry.
+ // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing
+ // the inode on which Open() is being called.
Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error)
}
@@ -328,7 +327,7 @@ type inodeMetadata interface {
// Stat returns the metadata for this inode. This corresponds to
// vfs.FilesystemImpl.StatAt.
- Stat(fs *vfs.Filesystem) linux.Statx
+ Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error)
// SetStat updates the metadata for this inode. This corresponds to
// vfs.FilesystemImpl.SetStatAt.
diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go
index f3f4e49b4..611645f3f 100644
--- a/pkg/sentry/fsimpl/proc/subtasks.go
+++ b/pkg/sentry/fsimpl/proc/subtasks.go
@@ -121,8 +121,13 @@ func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.O
}
// Stat implements kernfs.Inode.
-func (i *subtasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx {
- stat := i.InodeAttrs.Stat(vsfs)
- stat.Nlink += uint32(i.task.ThreadGroup().Count())
- return stat
+func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) {
+ stat, err := i.InodeAttrs.Stat(vsfs, opts)
+ if err != nil {
+ return linux.Statx{}, err
+ }
+ if opts.Mask&linux.STATX_NLINK != 0 {
+ stat.Nlink += uint32(i.task.ThreadGroup().Count())
+ }
+ return stat, nil
}
diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go
index 2d814668a..c0d643f51 100644
--- a/pkg/sentry/fsimpl/proc/task.go
+++ b/pkg/sentry/fsimpl/proc/task.go
@@ -62,11 +62,13 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames
"pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"),
"user": newNamespaceSymlink(task, inoGen.NextIno(), "user"),
}),
- "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}),
- "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}),
- "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}),
- "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}),
- "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}),
+ "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")),
+ "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}),
+ "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}),
+ "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}),
+ "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}),
+ "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}),
+ "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}),
}
if isThreadGroup {
contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers)
@@ -152,12 +154,21 @@ func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, childre
}
// Stat implements kernfs.Inode.
-func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx {
- stat := i.Inode.Stat(fs)
- uid, gid := i.getOwner(linux.FileMode(stat.Mode))
- stat.UID = uint32(uid)
- stat.GID = uint32(gid)
- return stat
+func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) {
+ stat, err := i.Inode.Stat(fs, opts)
+ if err != nil {
+ return linux.Statx{}, err
+ }
+ if opts.Mask&(linux.STATX_UID|linux.STATX_GID) != 0 {
+ uid, gid := i.getOwner(linux.FileMode(stat.Mode))
+ if opts.Mask&linux.STATX_UID != 0 {
+ stat.UID = uint32(uid)
+ }
+ if opts.Mask&linux.STATX_GID != 0 {
+ stat.GID = uint32(gid)
+ }
+ }
+ return stat, nil
}
// CheckPermissions implements kernfs.Inode.
@@ -234,7 +245,7 @@ func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentr
// member, there is one entry containing three colon-separated fields:
// hierarchy-ID:controller-list:cgroup-path"
func newCgroupData(controllers map[string]string) dynamicInode {
- buf := bytes.Buffer{}
+ var buf bytes.Buffer
// The hierarchy ids must be positive integers (for cgroup v1), but the
// exact number does not matter, so long as they are unique. We can
diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go
index efd3b3453..5a231ac86 100644
--- a/pkg/sentry/fsimpl/proc/task_files.go
+++ b/pkg/sentry/fsimpl/proc/task_files.go
@@ -525,3 +525,46 @@ func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error {
fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled)
return nil
}
+
+// oomScoreAdj is a stub of the /proc/<pid>/oom_score_adj file.
+//
+// +stateify savable
+type oomScoreAdj struct {
+ kernfs.DynamicBytesFile
+
+ task *kernel.Task
+}
+
+var _ vfs.WritableDynamicBytesSource = (*oomScoreAdj)(nil)
+
+// Generate implements vfs.DynamicBytesSource.Generate.
+func (o *oomScoreAdj) Generate(ctx context.Context, buf *bytes.Buffer) error {
+ adj, err := o.task.OOMScoreAdj()
+ if err != nil {
+ return err
+ }
+ fmt.Fprintf(buf, "%d\n", adj)
+ return nil
+}
+
+// Write implements vfs.WritableDynamicBytesSource.Write.
+func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) {
+ if src.NumBytes() == 0 {
+ return 0, nil
+ }
+
+ // Limit input size so as not to impact performance if input size is large.
+ src = src.TakeFirst(usermem.PageSize - 1)
+
+ var v int32
+ n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts)
+ if err != nil {
+ return 0, err
+ }
+
+ if err := o.task.SetOOMScoreAdj(v); err != nil {
+ return 0, err
+ }
+
+ return n, nil
+}
diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go
index 10c08fa90..b1e39c82f 100644
--- a/pkg/sentry/fsimpl/proc/tasks.go
+++ b/pkg/sentry/fsimpl/proc/tasks.go
@@ -211,17 +211,22 @@ func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.Open
return fd.VFSFileDescription(), nil
}
-func (i *tasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx {
- stat := i.InodeAttrs.Stat(vsfs)
+func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) {
+ stat, err := i.InodeAttrs.Stat(vsfs, opts)
+ if err != nil {
+ return linux.Statx{}, err
+ }
- // Add dynamic children to link count.
- for _, tg := range i.pidns.ThreadGroups() {
- if leader := tg.Leader(); leader != nil {
- stat.Nlink++
+ if opts.Mask&linux.STATX_NLINK != 0 {
+ // Add dynamic children to link count.
+ for _, tg := range i.pidns.ThreadGroups() {
+ if leader := tg.Leader(); leader != nil {
+ stat.Nlink++
+ }
}
}
- return stat
+ return stat, nil
}
func cpuInfoData(k *kernel.Kernel) string {
diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go
index c5d531fe0..0eb401619 100644
--- a/pkg/sentry/fsimpl/proc/tasks_test.go
+++ b/pkg/sentry/fsimpl/proc/tasks_test.go
@@ -63,21 +63,23 @@ var (
"thread-self": threadSelfLink.NextOff,
}
taskStaticFiles = map[string]testutil.DirentType{
- "auxv": linux.DT_REG,
- "cgroup": linux.DT_REG,
- "cmdline": linux.DT_REG,
- "comm": linux.DT_REG,
- "environ": linux.DT_REG,
- "gid_map": linux.DT_REG,
- "io": linux.DT_REG,
- "maps": linux.DT_REG,
- "ns": linux.DT_DIR,
- "smaps": linux.DT_REG,
- "stat": linux.DT_REG,
- "statm": linux.DT_REG,
- "status": linux.DT_REG,
- "task": linux.DT_DIR,
- "uid_map": linux.DT_REG,
+ "auxv": linux.DT_REG,
+ "cgroup": linux.DT_REG,
+ "cmdline": linux.DT_REG,
+ "comm": linux.DT_REG,
+ "environ": linux.DT_REG,
+ "gid_map": linux.DT_REG,
+ "io": linux.DT_REG,
+ "maps": linux.DT_REG,
+ "ns": linux.DT_DIR,
+ "oom_score": linux.DT_REG,
+ "oom_score_adj": linux.DT_REG,
+ "smaps": linux.DT_REG,
+ "stat": linux.DT_REG,
+ "statm": linux.DT_REG,
+ "status": linux.DT_REG,
+ "task": linux.DT_DIR,
+ "uid_map": linux.DT_REG,
}
)
diff --git a/pkg/sentry/inet/namespace.go b/pkg/sentry/inet/namespace.go
index c16667e7f..029af3025 100644
--- a/pkg/sentry/inet/namespace.go
+++ b/pkg/sentry/inet/namespace.go
@@ -23,7 +23,10 @@ type Namespace struct {
// creator allows kernel to create new network stack for network namespaces.
// If nil, no networking will function if network is namespaced.
- creator NetworkStackCreator
+ //
+ // At afterLoad(), creator will be used to create network stack. Stateify
+ // needs to wait for this field to be loaded before calling afterLoad().
+ creator NetworkStackCreator `state:"wait"`
// isRoot indicates whether this is the root network namespace.
isRoot bool
diff --git a/pkg/sentry/kernel/epoll/epoll_state.go b/pkg/sentry/kernel/epoll/epoll_state.go
index a0d35d350..8e9f200d0 100644
--- a/pkg/sentry/kernel/epoll/epoll_state.go
+++ b/pkg/sentry/kernel/epoll/epoll_state.go
@@ -38,11 +38,14 @@ func (e *EventPoll) afterLoad() {
}
}
- for it := e.waitingList.Front(); it != nil; it = it.Next() {
- if it.id.File.Readiness(it.mask) != 0 {
- e.waitingList.Remove(it)
- e.readyList.PushBack(it)
- it.curList = &e.readyList
+ for it := e.waitingList.Front(); it != nil; {
+ entry := it
+ it = it.Next()
+
+ if entry.id.File.Readiness(entry.mask) != 0 {
+ e.waitingList.Remove(entry)
+ e.readyList.PushBack(entry)
+ entry.curList = &e.readyList
e.Notify(waiter.EventIn)
}
}
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 2cee2e6ed..c0dbbe890 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -37,6 +37,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/sync"
+ "gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
)
@@ -554,6 +555,13 @@ type Task struct {
//
// startTime is protected by mu.
startTime ktime.Time
+
+ // oomScoreAdj is the task's OOM score adjustment. This is currently not
+ // used but is maintained for consistency.
+ // TODO(gvisor.dev/issue/1967)
+ //
+ // oomScoreAdj is protected by mu, and is owned by the task goroutine.
+ oomScoreAdj int32
}
func (t *Task) savePtraceTracer() *Task {
@@ -847,3 +855,28 @@ func (t *Task) AbstractSockets() *AbstractSocketNamespace {
func (t *Task) ContainerID() string {
return t.containerID
}
+
+// OOMScoreAdj gets the task's OOM score adjustment.
+func (t *Task) OOMScoreAdj() (int32, error) {
+ t.mu.Lock()
+ defer t.mu.Unlock()
+ if t.ExitState() == TaskExitDead {
+ return 0, syserror.ESRCH
+ }
+ return t.oomScoreAdj, nil
+}
+
+// SetOOMScoreAdj sets the task's OOM score adjustment. The value should be
+// between -1000 and 1000 inclusive.
+func (t *Task) SetOOMScoreAdj(adj int32) error {
+ t.mu.Lock()
+ defer t.mu.Unlock()
+ if t.ExitState() == TaskExitDead {
+ return syserror.ESRCH
+ }
+ if adj > 1000 || adj < -1000 {
+ return syserror.EINVAL
+ }
+ t.oomScoreAdj = adj
+ return nil
+}
diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go
index 78866f280..dda502bb8 100644
--- a/pkg/sentry/kernel/task_clone.go
+++ b/pkg/sentry/kernel/task_clone.go
@@ -264,6 +264,11 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
rseqSignature = t.rseqSignature
}
+ adj, err := t.OOMScoreAdj()
+ if err != nil {
+ return 0, nil, err
+ }
+
cfg := &TaskConfig{
Kernel: t.k,
ThreadGroup: tg,
@@ -282,6 +287,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
RSeqAddr: rseqAddr,
RSeqSignature: rseqSignature,
ContainerID: t.ContainerID(),
+ OOMScoreAdj: adj,
}
if opts.NewThreadGroup {
cfg.Parent = t
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index a5035bb7f..2bbf48bb8 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -93,6 +93,9 @@ type TaskConfig struct {
// ContainerID is the container the new task belongs to.
ContainerID string
+
+ // oomScoreAdj is the task's OOM score adjustment.
+ OOMScoreAdj int32
}
// NewTask creates a new task defined by cfg.
@@ -143,6 +146,7 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) {
rseqSignature: cfg.RSeqSignature,
futexWaiter: futex.NewWaiter(),
containerID: cfg.ContainerID,
+ oomScoreAdj: cfg.OOMScoreAdj,
}
t.creds.Store(cfg.Credentials)
t.endStopCond.L = &t.tg.signalHandlers.mu
diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.s b/pkg/sentry/platform/kvm/bluepill_arm64.s
index c61700892..04efa0147 100644
--- a/pkg/sentry/platform/kvm/bluepill_arm64.s
+++ b/pkg/sentry/platform/kvm/bluepill_arm64.s
@@ -82,6 +82,8 @@ fallback:
// dieTrampoline: see bluepill.go, bluepill_arm64_unsafe.go for documentation.
TEXT ·dieTrampoline(SB),NOSPLIT,$0
- // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64.
- MOVD R9, 8(RSP)
- BL ·dieHandler(SB)
+ // R0: Fake the old PC as caller
+ // R1: First argument (vCPU)
+ MOVD.P R1, 8(RSP) // R1: First argument (vCPU)
+ MOVD.P R0, 8(RSP) // R0: Fake the old PC as caller
+ B ·dieHandler(SB)
diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go
index 2f02c03cf..195331383 100644
--- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go
+++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go
@@ -18,9 +18,30 @@ package kvm
import (
"gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/platform/ring0"
)
+// dieArchSetup initialies the state for dieTrampoline.
+//
+// The arm64 dieTrampoline requires the vCPU to be set in R1, and the last PC
+// to be in R0. The trampoline then simulates a call to dieHandler from the
+// provided PC.
+//
//go:nosplit
func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) {
- // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64.
+ // If the vCPU is in user mode, we set the stack to the stored stack
+ // value in the vCPU itself. We don't want to unwind the user stack.
+ if guestRegs.Regs.Pstate&ring0.PSR_MODE_MASK == ring0.PSR_MODE_EL0t {
+ regs := c.CPU.Registers()
+ context.Regs[0] = regs.Regs[0]
+ context.Sp = regs.Sp
+ context.Regs[29] = regs.Regs[29] // stack base address
+ } else {
+ context.Regs[0] = guestRegs.Regs.Pc
+ context.Sp = guestRegs.Regs.Sp
+ context.Regs[29] = guestRegs.Regs.Regs[29]
+ context.Pstate = guestRegs.Regs.Pstate
+ }
+ context.Regs[1] = uint64(uintptr(unsafe.Pointer(c)))
+ context.Pc = uint64(dieTrampolineAddr)
}
diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go
index 972ba85c3..a9b4af43e 100644
--- a/pkg/sentry/platform/kvm/kvm.go
+++ b/pkg/sentry/platform/kvm/kvm.go
@@ -27,6 +27,38 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
+// userMemoryRegion is a region of physical memory.
+//
+// This mirrors kvm_memory_region.
+type userMemoryRegion struct {
+ slot uint32
+ flags uint32
+ guestPhysAddr uint64
+ memorySize uint64
+ userspaceAddr uint64
+}
+
+// runData is the run structure. This may be mapped for synchronous register
+// access (although that doesn't appear to be supported by my kernel at least).
+//
+// This mirrors kvm_run.
+type runData struct {
+ requestInterruptWindow uint8
+ _ [7]uint8
+
+ exitReason uint32
+ readyForInterruptInjection uint8
+ ifFlag uint8
+ _ [2]uint8
+
+ cr8 uint64
+ apicBase uint64
+
+ // This is the union data for exits. Interpretation depends entirely on
+ // the exitReason above (see vCPU code for more information).
+ data [32]uint64
+}
+
// KVM represents a lightweight VM context.
type KVM struct {
platform.NoCPUPreemptionDetection
diff --git a/pkg/sentry/platform/kvm/kvm_amd64.go b/pkg/sentry/platform/kvm/kvm_amd64.go
index c5a6f9c7d..093497bc4 100644
--- a/pkg/sentry/platform/kvm/kvm_amd64.go
+++ b/pkg/sentry/platform/kvm/kvm_amd64.go
@@ -21,17 +21,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/platform/ring0"
)
-// userMemoryRegion is a region of physical memory.
-//
-// This mirrors kvm_memory_region.
-type userMemoryRegion struct {
- slot uint32
- flags uint32
- guestPhysAddr uint64
- memorySize uint64
- userspaceAddr uint64
-}
-
// userRegs represents KVM user registers.
//
// This mirrors kvm_regs.
@@ -169,27 +158,6 @@ type modelControlRegisters struct {
entries [16]modelControlRegister
}
-// runData is the run structure. This may be mapped for synchronous register
-// access (although that doesn't appear to be supported by my kernel at least).
-//
-// This mirrors kvm_run.
-type runData struct {
- requestInterruptWindow uint8
- _ [7]uint8
-
- exitReason uint32
- readyForInterruptInjection uint8
- ifFlag uint8
- _ [2]uint8
-
- cr8 uint64
- apicBase uint64
-
- // This is the union data for exits. Interpretation depends entirely on
- // the exitReason above (see vCPU code for more information).
- data [32]uint64
-}
-
// cpuidEntry is a single CPUID entry.
//
// This mirrors kvm_cpuid_entry2.
diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go
index 2319c86d3..79045651e 100644
--- a/pkg/sentry/platform/kvm/kvm_arm64.go
+++ b/pkg/sentry/platform/kvm/kvm_arm64.go
@@ -20,17 +20,6 @@ import (
"syscall"
)
-// userMemoryRegion is a region of physical memory.
-//
-// This mirrors kvm_memory_region.
-type userMemoryRegion struct {
- slot uint32
- flags uint32
- guestPhysAddr uint64
- memorySize uint64
- userspaceAddr uint64
-}
-
type kvmOneReg struct {
id uint64
addr uint64
@@ -53,27 +42,6 @@ type userRegs struct {
fpRegs userFpsimdState
}
-// runData is the run structure. This may be mapped for synchronous register
-// access (although that doesn't appear to be supported by my kernel at least).
-//
-// This mirrors kvm_run.
-type runData struct {
- requestInterruptWindow uint8
- _ [7]uint8
-
- exitReason uint32
- readyForInterruptInjection uint8
- ifFlag uint8
- _ [2]uint8
-
- cr8 uint64
- apicBase uint64
-
- // This is the union data for exits. Interpretation depends entirely on
- // the exitReason above (see vCPU code for more information).
- data [32]uint64
-}
-
// updateGlobalOnce does global initialization. It has to be called only once.
func updateGlobalOnce(fd int) error {
physicalInit()
diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go
index f6da41c27..8122ac6e2 100644
--- a/pkg/sentry/platform/ring0/aarch64.go
+++ b/pkg/sentry/platform/ring0/aarch64.go
@@ -27,26 +27,27 @@ const (
_PTE_PGT_BASE = 0x7000
_PTE_PGT_SIZE = 0x1000
- _PSR_MODE_EL0t = 0x0
- _PSR_MODE_EL1t = 0x4
- _PSR_MODE_EL1h = 0x5
- _PSR_EL_MASK = 0xf
-
- _PSR_D_BIT = 0x200
- _PSR_A_BIT = 0x100
- _PSR_I_BIT = 0x80
- _PSR_F_BIT = 0x40
+ _PSR_D_BIT = 0x00000200
+ _PSR_A_BIT = 0x00000100
+ _PSR_I_BIT = 0x00000080
+ _PSR_F_BIT = 0x00000040
)
const (
+ // PSR bits
+ PSR_MODE_EL0t = 0x00000000
+ PSR_MODE_EL1t = 0x00000004
+ PSR_MODE_EL1h = 0x00000005
+ PSR_MODE_MASK = 0x0000000f
+
// KernelFlagsSet should always be set in the kernel.
- KernelFlagsSet = _PSR_MODE_EL1h
+ KernelFlagsSet = PSR_MODE_EL1h
// UserFlagsSet are always set in userspace.
- UserFlagsSet = _PSR_MODE_EL0t
+ UserFlagsSet = PSR_MODE_EL0t
- KernelFlagsClear = _PSR_EL_MASK
- UserFlagsClear = _PSR_EL_MASK
+ KernelFlagsClear = PSR_MODE_MASK
+ UserFlagsClear = PSR_MODE_MASK
PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT
)
diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD
index 07c8383e6..cb4deb068 100644
--- a/pkg/sentry/vfs/BUILD
+++ b/pkg/sentry/vfs/BUILD
@@ -42,10 +42,13 @@ go_library(
deps = [
"//pkg/abi/linux",
"//pkg/context",
+ "//pkg/fd",
"//pkg/fspath",
"//pkg/gohacks",
"//pkg/log",
+ "//pkg/safemem",
"//pkg/sentry/arch",
+ "//pkg/sentry/fs",
"//pkg/sentry/fs/lock",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/memmap",
@@ -53,6 +56,7 @@ go_library(
"//pkg/syserror",
"//pkg/usermem",
"//pkg/waiter",
+ "@org_golang_x_sys//unix:go_default_library",
],
)
diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go
index c2a52ec1b..45191d1c3 100644
--- a/pkg/sentry/vfs/file_description_impl_util.go
+++ b/pkg/sentry/vfs/file_description_impl_util.go
@@ -33,8 +33,8 @@ import (
// implementations to adapt:
// - Have a local fileDescription struct (containing FileDescription) which
// embeds FileDescriptionDefaultImpl and overrides the default methods
-// which are common to all fd implementations for that for that filesystem
-// like StatusFlags, SetStatusFlags, Stat, SetStat, StatFS, etc.
+// which are common to all fd implementations for that filesystem like
+// StatusFlags, SetStatusFlags, Stat, SetStat, StatFS, etc.
// - This should be embedded in all file description implementations as the
// first field by value.
// - Directory FDs would also embed DirectoryFileDescriptionDefaultImpl.
diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go
index c6c160dfc..8dc0f7c0e 100644
--- a/pkg/tcpip/checker/checker.go
+++ b/pkg/tcpip/checker/checker.go
@@ -785,6 +785,52 @@ func NDPNSTargetAddress(want tcpip.Address) TransportChecker {
}
}
+// ndpOptions checks that optsBuf only contains opts.
+func ndpOptions(t *testing.T, optsBuf header.NDPOptions, opts []header.NDPOption) {
+ t.Helper()
+
+ it, err := optsBuf.Iter(true)
+ if err != nil {
+ t.Errorf("optsBuf.Iter(true): %s", err)
+ return
+ }
+
+ i := 0
+ for {
+ opt, done, err := it.Next()
+ if err != nil {
+ // This should never happen as Iter(true) above did not return an error.
+ t.Fatalf("unexpected error when iterating over NDP options: %s", err)
+ }
+ if done {
+ break
+ }
+
+ if i >= len(opts) {
+ t.Errorf("got unexpected option: %s", opt)
+ continue
+ }
+
+ switch wantOpt := opts[i].(type) {
+ case header.NDPSourceLinkLayerAddressOption:
+ gotOpt, ok := opt.(header.NDPSourceLinkLayerAddressOption)
+ if !ok {
+ t.Errorf("got type = %T at index = %d; want = %T", opt, i, wantOpt)
+ } else if got, want := gotOpt.EthernetAddress(), wantOpt.EthernetAddress(); got != want {
+ t.Errorf("got EthernetAddress() = %s at index %d, want = %s", got, i, want)
+ }
+ default:
+ t.Fatalf("checker not implemented for expected NDP option: %T", wantOpt)
+ }
+
+ i++
+ }
+
+ if missing := opts[i:]; len(missing) > 0 {
+ t.Errorf("missing options: %s", missing)
+ }
+}
+
// NDPNSOptions creates a checker that checks that the packet contains the
// provided NDP options within an NDP Neighbor Solicitation message.
//
@@ -796,47 +842,31 @@ func NDPNSOptions(opts []header.NDPOption) TransportChecker {
icmp := h.(header.ICMPv6)
ns := header.NDPNeighborSolicit(icmp.NDPPayload())
- it, err := ns.Options().Iter(true)
- if err != nil {
- t.Errorf("opts.Iter(true): %s", err)
- return
- }
-
- i := 0
- for {
- opt, done, _ := it.Next()
- if done {
- break
- }
-
- if i >= len(opts) {
- t.Errorf("got unexpected option: %s", opt)
- continue
- }
-
- switch wantOpt := opts[i].(type) {
- case header.NDPSourceLinkLayerAddressOption:
- gotOpt, ok := opt.(header.NDPSourceLinkLayerAddressOption)
- if !ok {
- t.Errorf("got type = %T at index = %d; want = %T", opt, i, wantOpt)
- } else if got, want := gotOpt.EthernetAddress(), wantOpt.EthernetAddress(); got != want {
- t.Errorf("got EthernetAddress() = %s at index %d, want = %s", got, i, want)
- }
- default:
- panic("not implemented")
- }
-
- i++
- }
-
- if missing := opts[i:]; len(missing) > 0 {
- t.Errorf("missing options: %s", missing)
- }
+ ndpOptions(t, ns.Options(), opts)
}
}
// NDPRS creates a checker that checks that the packet contains a valid NDP
// Router Solicitation message (as per the raw wire format).
-func NDPRS() NetworkChecker {
- return NDP(header.ICMPv6RouterSolicit, header.NDPRSMinimumSize)
+//
+// checkers may assume that a valid ICMPv6 is passed to it containing a valid
+// NDPRS as far as the size of the message is concerned. The values within the
+// message are up to checkers to validate.
+func NDPRS(checkers ...TransportChecker) NetworkChecker {
+ return NDP(header.ICMPv6RouterSolicit, header.NDPRSMinimumSize, checkers...)
+}
+
+// NDPRSOptions creates a checker that checks that the packet contains the
+// provided NDP options within an NDP Router Solicitation message.
+//
+// The returned TransportChecker assumes that a valid ICMPv6 is passed to it
+// containing a valid NDPRS message as far as the size is concerned.
+func NDPRSOptions(opts []header.NDPOption) TransportChecker {
+ return func(t *testing.T, h header.Transport) {
+ t.Helper()
+
+ icmp := h.(header.ICMPv6)
+ rs := header.NDPRouterSolicit(icmp.NDPPayload())
+ ndpOptions(t, rs.Options(), opts)
+ }
}
diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go
index 92f2aa13a..f42abc4bb 100644
--- a/pkg/tcpip/network/fragmentation/fragmentation.go
+++ b/pkg/tcpip/network/fragmentation/fragmentation.go
@@ -115,10 +115,12 @@ func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buf
// Evict reassemblers if we are consuming more memory than highLimit until
// we reach lowLimit.
if f.size > f.highLimit {
- tail := f.rList.Back()
- for f.size > f.lowLimit && tail != nil {
+ for f.size > f.lowLimit {
+ tail := f.rList.Back()
+ if tail == nil {
+ break
+ }
f.release(tail)
- tail = tail.Prev()
}
}
f.mu.Unlock()
diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD
index 705cf01ee..8febd54c8 100644
--- a/pkg/tcpip/stack/BUILD
+++ b/pkg/tcpip/stack/BUILD
@@ -18,6 +18,7 @@ go_template_instance(
go_library(
name = "stack",
srcs = [
+ "dhcpv6configurationfromndpra_string.go",
"icmp_rate_limit.go",
"linkaddrcache.go",
"linkaddrentry_list.go",
diff --git a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
new file mode 100644
index 000000000..8b4213eec
--- /dev/null
+++ b/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go
@@ -0,0 +1,39 @@
+// 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.
+
+// Code generated by "stringer -type=DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT.
+
+package stack
+
+import "strconv"
+
+func _() {
+ // An "invalid array index" compiler error signifies that the constant values have changed.
+ // Re-run the stringer command to generate them again.
+ var x [1]struct{}
+ _ = x[DHCPv6NoConfiguration-0]
+ _ = x[DHCPv6ManagedAddress-1]
+ _ = x[DHCPv6OtherConfigurations-2]
+}
+
+const _DHCPv6ConfigurationFromNDPRA_name = "DHCPv6NoConfigurationDHCPv6ManagedAddressDHCPv6OtherConfigurations"
+
+var _DHCPv6ConfigurationFromNDPRA_index = [...]uint8{0, 21, 41, 66}
+
+func (i DHCPv6ConfigurationFromNDPRA) String() string {
+ if i < 0 || i >= DHCPv6ConfigurationFromNDPRA(len(_DHCPv6ConfigurationFromNDPRA_index)-1) {
+ return "DHCPv6ConfigurationFromNDPRA(" + strconv.FormatInt(int64(i), 10) + ")"
+ }
+ return _DHCPv6ConfigurationFromNDPRA_name[_DHCPv6ConfigurationFromNDPRA_index[i]:_DHCPv6ConfigurationFromNDPRA_index[i+1]]
+}
diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go
index f651871ce..a9f4d5dad 100644
--- a/pkg/tcpip/stack/ndp.go
+++ b/pkg/tcpip/stack/ndp.go
@@ -1220,9 +1220,15 @@ func (ndp *ndpState) startSolicitingRouters() {
}
ndp.rtrSolicitTimer = time.AfterFunc(delay, func() {
- // Send an RS message with the unspecified source address.
- ref := ndp.nic.getRefOrCreateTemp(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint, forceSpoofing)
- r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false)
+ // As per RFC 4861 section 4.1, the source of the RS is an address assigned
+ // to the sending interface, or the unspecified address if no address is
+ // assigned to the sending interface.
+ ref := ndp.nic.primaryIPv6Endpoint(header.IPv6AllRoutersMulticastAddress)
+ if ref == nil {
+ ref = ndp.nic.getRefOrCreateTemp(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint, forceSpoofing)
+ }
+ localAddr := ref.ep.ID().LocalAddress
+ r := makeRoute(header.IPv6ProtocolNumber, localAddr, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false)
defer r.Release()
// Route should resolve immediately since
@@ -1234,10 +1240,25 @@ func (ndp *ndpState) startSolicitingRouters() {
log.Fatalf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID())
}
- payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize
+ // As per RFC 4861 section 4.1, an NDP RS SHOULD include the source
+ // link-layer address option if the source address of the NDP RS is
+ // specified. This option MUST NOT be included if the source address is
+ // unspecified.
+ //
+ // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by
+ // LinkEndpoint.LinkAddress) before reaching this point.
+ var optsSerializer header.NDPOptionsSerializer
+ if localAddr != header.IPv6Any && header.IsValidUnicastEthernetAddress(r.LocalLinkAddress) {
+ optsSerializer = header.NDPOptionsSerializer{
+ header.NDPSourceLinkLayerAddressOption(r.LocalLinkAddress),
+ }
+ }
+ payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length())
hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + payloadSize)
pkt := header.ICMPv6(hdr.Prepend(payloadSize))
pkt.SetType(header.ICMPv6RouterSolicit)
+ rs := header.NDPRouterSolicit(pkt.NDPPayload())
+ rs.Options().Serialize(optsSerializer)
pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{}))
sent := r.Stats().ICMP.V6PacketsSent
diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go
index 6e9306d09..98b1c807c 100644
--- a/pkg/tcpip/stack/ndp_test.go
+++ b/pkg/tcpip/stack/ndp_test.go
@@ -3384,6 +3384,10 @@ func TestRouterSolicitation(t *testing.T) {
tests := []struct {
name string
linkHeaderLen uint16
+ linkAddr tcpip.LinkAddress
+ nicAddr tcpip.Address
+ expectedSrcAddr tcpip.Address
+ expectedNDPOpts []header.NDPOption
maxRtrSolicit uint8
rtrSolicitInt time.Duration
effectiveRtrSolicitInt time.Duration
@@ -3392,6 +3396,7 @@ func TestRouterSolicitation(t *testing.T) {
}{
{
name: "Single RS with delay",
+ expectedSrcAddr: header.IPv6Any,
maxRtrSolicit: 1,
rtrSolicitInt: time.Second,
effectiveRtrSolicitInt: time.Second,
@@ -3401,6 +3406,8 @@ func TestRouterSolicitation(t *testing.T) {
{
name: "Two RS with delay",
linkHeaderLen: 1,
+ nicAddr: llAddr1,
+ expectedSrcAddr: llAddr1,
maxRtrSolicit: 2,
rtrSolicitInt: time.Second,
effectiveRtrSolicitInt: time.Second,
@@ -3408,8 +3415,14 @@ func TestRouterSolicitation(t *testing.T) {
effectiveMaxRtrSolicitDelay: 500 * time.Millisecond,
},
{
- name: "Single RS without delay",
- linkHeaderLen: 2,
+ name: "Single RS without delay",
+ linkHeaderLen: 2,
+ linkAddr: linkAddr1,
+ nicAddr: llAddr1,
+ expectedSrcAddr: llAddr1,
+ expectedNDPOpts: []header.NDPOption{
+ header.NDPSourceLinkLayerAddressOption(linkAddr1),
+ },
maxRtrSolicit: 1,
rtrSolicitInt: time.Second,
effectiveRtrSolicitInt: time.Second,
@@ -3419,6 +3432,8 @@ func TestRouterSolicitation(t *testing.T) {
{
name: "Two RS without delay and invalid zero interval",
linkHeaderLen: 3,
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
maxRtrSolicit: 2,
rtrSolicitInt: 0,
effectiveRtrSolicitInt: 4 * time.Second,
@@ -3427,6 +3442,8 @@ func TestRouterSolicitation(t *testing.T) {
},
{
name: "Three RS without delay",
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
maxRtrSolicit: 3,
rtrSolicitInt: 500 * time.Millisecond,
effectiveRtrSolicitInt: 500 * time.Millisecond,
@@ -3435,6 +3452,8 @@ func TestRouterSolicitation(t *testing.T) {
},
{
name: "Two RS with invalid negative delay",
+ linkAddr: linkAddr1,
+ expectedSrcAddr: header.IPv6Any,
maxRtrSolicit: 2,
rtrSolicitInt: time.Second,
effectiveRtrSolicitInt: time.Second,
@@ -3457,7 +3476,7 @@ func TestRouterSolicitation(t *testing.T) {
t.Run(test.name, func(t *testing.T) {
t.Parallel()
e := channelLinkWithHeaderLength{
- Endpoint: channel.New(int(test.maxRtrSolicit), 1280, linkAddr1),
+ Endpoint: channel.New(int(test.maxRtrSolicit), 1280, test.linkAddr),
headerLength: test.linkHeaderLen,
}
e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired
@@ -3481,10 +3500,10 @@ func TestRouterSolicitation(t *testing.T) {
checker.IPv6(t,
p.Pkt.Header.View(),
- checker.SrcAddr(header.IPv6Any),
+ checker.SrcAddr(test.expectedSrcAddr),
checker.DstAddr(header.IPv6AllRoutersMulticastAddress),
checker.TTL(header.NDPHopLimit),
- checker.NDPRS(),
+ checker.NDPRS(checker.NDPRSOptions(test.expectedNDPOpts)),
)
if l, want := p.Pkt.Header.AvailableLength(), int(test.linkHeaderLen); l != want {
@@ -3510,13 +3529,19 @@ func TestRouterSolicitation(t *testing.T) {
t.Fatalf("CreateNIC(%d, _) = %s", nicID, err)
}
- // Make sure each RS got sent at the right
- // times.
+ if addr := test.nicAddr; addr != "" {
+ if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr); err != nil {
+ t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr, err)
+ }
+ }
+
+ // Make sure each RS is sent at the right time.
remaining := test.maxRtrSolicit
if remaining > 0 {
waitForPkt(test.effectiveMaxRtrSolicitDelay + defaultAsyncEventTimeout)
remaining--
}
+
for ; remaining > 0; remaining-- {
waitForNothing(test.effectiveRtrSolicitInt - defaultTimeout)
waitForPkt(defaultAsyncEventTimeout)
diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD
index 272e8f570..a32f9eacf 100644
--- a/pkg/tcpip/transport/tcp/BUILD
+++ b/pkg/tcpip/transport/tcp/BUILD
@@ -32,6 +32,7 @@ go_library(
srcs = [
"accept.go",
"connect.go",
+ "connect_unsafe.go",
"cubic.go",
"cubic_state.go",
"dispatcher.go",
diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go
index ae4f3f3a9..c0f73ef16 100644
--- a/pkg/tcpip/transport/tcp/connect.go
+++ b/pkg/tcpip/transport/tcp/connect.go
@@ -624,17 +624,17 @@ func parseSynSegmentOptions(s *segment) header.TCPSynOptions {
var optionPool = sync.Pool{
New: func() interface{} {
- return make([]byte, maxOptionSize)
+ return &[maxOptionSize]byte{}
},
}
func getOptions() []byte {
- return optionPool.Get().([]byte)
+ return (*optionPool.Get().(*[maxOptionSize]byte))[:]
}
func putOptions(options []byte) {
// Reslice to full capacity.
- optionPool.Put(options[0:cap(options)])
+ optionPool.Put(optionsToArray(options))
}
func makeSynOptions(opts header.TCPSynOptions) []byte {
diff --git a/pkg/tcpip/transport/tcp/connect_unsafe.go b/pkg/tcpip/transport/tcp/connect_unsafe.go
new file mode 100644
index 000000000..cfc304616
--- /dev/null
+++ b/pkg/tcpip/transport/tcp/connect_unsafe.go
@@ -0,0 +1,30 @@
+// Copyright 2018 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 tcp
+
+import (
+ "reflect"
+ "unsafe"
+)
+
+// optionsToArray converts a slice of capacity >-= maxOptionSize to an array.
+//
+// optionsToArray panics if the capacity of options is smaller than
+// maxOptionSize.
+func optionsToArray(options []byte) *[maxOptionSize]byte {
+ // Reslice to full capacity.
+ options = options[0:maxOptionSize]
+ return (*[maxOptionSize]byte)(unsafe.Pointer((*reflect.SliceHeader)(unsafe.Pointer(&options)).Data))
+}
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index 40cc664c0..dc9c18b6f 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -2117,10 +2117,13 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error {
// Close for write.
if (e.shutdownFlags & tcpip.ShutdownWrite) != 0 {
e.sndBufMu.Lock()
-
if e.sndClosed {
// Already closed.
e.sndBufMu.Unlock()
+ if e.EndpointState() == StateTimeWait {
+ e.mu.Unlock()
+ return tcpip.ErrNotConnected
+ }
break
}
diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go
index 17e774e0c..8125d5061 100644
--- a/runsc/boot/controller.go
+++ b/runsc/boot/controller.go
@@ -101,11 +101,14 @@ const (
// Profiling related commands (see pprof.go for more details).
const (
- StartCPUProfile = "Profile.StartCPUProfile"
- StopCPUProfile = "Profile.StopCPUProfile"
- HeapProfile = "Profile.HeapProfile"
- StartTrace = "Profile.StartTrace"
- StopTrace = "Profile.StopTrace"
+ StartCPUProfile = "Profile.StartCPUProfile"
+ StopCPUProfile = "Profile.StopCPUProfile"
+ HeapProfile = "Profile.HeapProfile"
+ GoroutineProfile = "Profile.GoroutineProfile"
+ BlockProfile = "Profile.BlockProfile"
+ MutexProfile = "Profile.MutexProfile"
+ StartTrace = "Profile.StartTrace"
+ StopTrace = "Profile.StopTrace"
)
// Logging related commands (see logging.go for more details).
diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go
index 79965460e..b5de2588b 100644
--- a/runsc/cmd/debug.go
+++ b/runsc/cmd/debug.go
@@ -32,17 +32,20 @@ import (
// Debug implements subcommands.Command for the "debug" command.
type Debug struct {
- pid int
- stacks bool
- signal int
- profileHeap string
- profileCPU string
- trace string
- strace string
- logLevel string
- logPackets string
- duration time.Duration
- ps bool
+ pid int
+ stacks bool
+ signal int
+ profileHeap string
+ profileCPU string
+ profileGoroutine string
+ profileBlock string
+ profileMutex string
+ trace string
+ strace string
+ logLevel string
+ logPackets string
+ duration time.Duration
+ ps bool
}
// Name implements subcommands.Command.
@@ -66,6 +69,9 @@ func (d *Debug) SetFlags(f *flag.FlagSet) {
f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log")
f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.")
f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.")
+ f.StringVar(&d.profileGoroutine, "profile-goroutine", "", "writes goroutine profile to the given file.")
+ f.StringVar(&d.profileBlock, "profile-block", "", "writes block profile to the given file.")
+ f.StringVar(&d.profileMutex, "profile-mutex", "", "writes mutex profile to the given file.")
f.DurationVar(&d.duration, "duration", time.Second, "amount of time to wait for CPU and trace profiles")
f.StringVar(&d.trace, "trace", "", "writes an execution trace to the given file.")
f.IntVar(&d.signal, "signal", -1, "sends signal to the sandbox")
@@ -147,6 +153,42 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
}
log.Infof("Heap profile written to %q", d.profileHeap)
}
+ if d.profileGoroutine != "" {
+ f, err := os.Create(d.profileGoroutine)
+ if err != nil {
+ return Errorf(err.Error())
+ }
+ defer f.Close()
+
+ if err := c.Sandbox.GoroutineProfile(f); err != nil {
+ return Errorf(err.Error())
+ }
+ log.Infof("Goroutine profile written to %q", d.profileGoroutine)
+ }
+ if d.profileBlock != "" {
+ f, err := os.Create(d.profileBlock)
+ if err != nil {
+ return Errorf(err.Error())
+ }
+ defer f.Close()
+
+ if err := c.Sandbox.BlockProfile(f); err != nil {
+ return Errorf(err.Error())
+ }
+ log.Infof("Block profile written to %q", d.profileBlock)
+ }
+ if d.profileMutex != "" {
+ f, err := os.Create(d.profileMutex)
+ if err != nil {
+ return Errorf(err.Error())
+ }
+ defer f.Close()
+
+ if err := c.Sandbox.MutexProfile(f); err != nil {
+ return Errorf(err.Error())
+ }
+ log.Infof("Mutex profile written to %q", d.profileMutex)
+ }
delay := false
if d.profileCPU != "" {
diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go
index c2518d52b..651615d4c 100644
--- a/runsc/container/console_test.go
+++ b/runsc/container/console_test.go
@@ -333,13 +333,13 @@ func TestJobControlSignalRootContainer(t *testing.T) {
// file. Writes after a certain point will block unless we drain the
// PTY, so we must continually copy from it.
//
- // We log the output to stdout for debugabilitly, and also to a buffer,
+ // We log the output to stderr for debugabilitly, and also to a buffer,
// since we wait on particular output from bash below. We use a custom
// blockingBuffer which is thread-safe and also blocks on Read calls,
// which makes this a suitable Reader for WaitUntilRead.
ptyBuf := newBlockingBuffer()
tee := io.TeeReader(ptyMaster, ptyBuf)
- go io.Copy(os.Stdout, tee)
+ go io.Copy(os.Stderr, tee)
// Start the container.
if err := c.Start(conf); err != nil {
diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go
index bdd65b498..c7eea85b3 100644
--- a/runsc/container/container_test.go
+++ b/runsc/container/container_test.go
@@ -2092,7 +2092,7 @@ func TestOverlayfsStaleRead(t *testing.T) {
defer out.Close()
const want = "foobar"
- cmd := fmt.Sprintf("cat %q && echo %q> %q && cp %q %q", in.Name(), want, in.Name(), in.Name(), out.Name())
+ cmd := fmt.Sprintf("cat %q >&2 && echo %q> %q && cp %q %q", in.Name(), want, in.Name(), in.Name(), out.Name())
spec := testutil.NewSpecWithArgs("/bin/bash", "-c", cmd)
if err := run(spec, conf); err != nil {
t.Fatalf("Error running container: %v", err)
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index 67e27df4d..192bde40c 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -990,6 +990,66 @@ func (s *Sandbox) StopCPUProfile() error {
return nil
}
+// GoroutineProfile writes a goroutine profile to the given file.
+func (s *Sandbox) GoroutineProfile(f *os.File) error {
+ log.Debugf("Goroutine profile %q", s.ID)
+ conn, err := s.sandboxConnect()
+ if err != nil {
+ return err
+ }
+ defer conn.Close()
+
+ opts := control.ProfileOpts{
+ FilePayload: urpc.FilePayload{
+ Files: []*os.File{f},
+ },
+ }
+ if err := conn.Call(boot.GoroutineProfile, &opts, nil); err != nil {
+ return fmt.Errorf("getting sandbox %q goroutine profile: %v", s.ID, err)
+ }
+ return nil
+}
+
+// BlockProfile writes a block profile to the given file.
+func (s *Sandbox) BlockProfile(f *os.File) error {
+ log.Debugf("Block profile %q", s.ID)
+ conn, err := s.sandboxConnect()
+ if err != nil {
+ return err
+ }
+ defer conn.Close()
+
+ opts := control.ProfileOpts{
+ FilePayload: urpc.FilePayload{
+ Files: []*os.File{f},
+ },
+ }
+ if err := conn.Call(boot.BlockProfile, &opts, nil); err != nil {
+ return fmt.Errorf("getting sandbox %q block profile: %v", s.ID, err)
+ }
+ return nil
+}
+
+// MutexProfile writes a mutex profile to the given file.
+func (s *Sandbox) MutexProfile(f *os.File) error {
+ log.Debugf("Mutex profile %q", s.ID)
+ conn, err := s.sandboxConnect()
+ if err != nil {
+ return err
+ }
+ defer conn.Close()
+
+ opts := control.ProfileOpts{
+ FilePayload: urpc.FilePayload{
+ Files: []*os.File{f},
+ },
+ }
+ if err := conn.Call(boot.MutexProfile, &opts, nil); err != nil {
+ return fmt.Errorf("getting sandbox %q mutex profile: %v", s.ID, err)
+ }
+ return nil
+}
+
// StartTrace start trace writing to the given file.
func (s *Sandbox) StartTrace(f *os.File) error {
log.Debugf("Trace start %q", s.ID)
diff --git a/scripts/dev.sh b/scripts/dev.sh
index 6238b4d0b..a9107f33e 100755
--- a/scripts/dev.sh
+++ b/scripts/dev.sh
@@ -66,6 +66,7 @@ if [[ ${REFRESH} -eq 0 ]]; then
else
mkdir -p "$(dirname ${RUNSC_BIN})"
cp -f ${OUTPUT} "${RUNSC_BIN}"
+ chmod a+rx "${RUNSC_BIN}"
echo
echo "Runtime ${RUNTIME} refreshed."
diff --git a/test/packetdrill/BUILD b/test/packetdrill/BUILD
index d113555b1..fb0b2db41 100644
--- a/test/packetdrill/BUILD
+++ b/test/packetdrill/BUILD
@@ -1,8 +1,48 @@
-load("defs.bzl", "packetdrill_test")
+load("defs.bzl", "packetdrill_linux_test", "packetdrill_netstack_test", "packetdrill_test")
package(licenses = ["notice"])
packetdrill_test(
- name = "fin_wait2_timeout",
+ name = "packetdrill_sanity_test",
+ scripts = ["sanity_test.pkt"],
+)
+
+packetdrill_test(
+ name = "accept_ack_drop_test",
+ scripts = ["accept_ack_drop.pkt"],
+)
+
+packetdrill_test(
+ name = "fin_wait2_timeout_test",
scripts = ["fin_wait2_timeout.pkt"],
)
+
+packetdrill_linux_test(
+ name = "tcp_user_timeout_test_linux_test",
+ scripts = ["linux/tcp_user_timeout.pkt"],
+)
+
+packetdrill_netstack_test(
+ name = "tcp_user_timeout_test_netstack_test",
+ scripts = ["netstack/tcp_user_timeout.pkt"],
+)
+
+packetdrill_test(
+ name = "listen_close_before_handshake_complete_test",
+ scripts = ["listen_close_before_handshake_complete.pkt"],
+)
+
+packetdrill_test(
+ name = "no_rst_to_rst_test",
+ scripts = ["no_rst_to_rst.pkt"],
+)
+
+packetdrill_test(
+ name = "tcp_defer_accept_test",
+ scripts = ["tcp_defer_accept.pkt"],
+)
+
+packetdrill_test(
+ name = "tcp_defer_accept_timeout_test",
+ scripts = ["tcp_defer_accept_timeout.pkt"],
+)
diff --git a/test/packetdrill/accept_ack_drop.pkt b/test/packetdrill/accept_ack_drop.pkt
new file mode 100644
index 000000000..76e638fd4
--- /dev/null
+++ b/test/packetdrill/accept_ack_drop.pkt
@@ -0,0 +1,27 @@
+// Test that the accept works if the final ACK is dropped and an ack with data
+// follows the dropped ack.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1 <...>
+
++0.0 < . 1:5(4) ack 1 win 257
++0.0 > . 1:1(0) ack 5 <...>
+
+// This should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
+// Now read the data and we should get 4 bytes.
++0.000 read(4,..., 4) = 4
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 5 <...>
++0.0 < F. 5:5(0) ack 2 win 257
++0.01 > . 2:2(0) ack 6 <...> \ No newline at end of file
diff --git a/test/packetdrill/defs.bzl b/test/packetdrill/defs.bzl
index 8623ce7b1..f499c177b 100644
--- a/test/packetdrill/defs.bzl
+++ b/test/packetdrill/defs.bzl
@@ -66,7 +66,7 @@ def packetdrill_linux_test(name, **kwargs):
if "tags" not in kwargs:
kwargs["tags"] = _PACKETDRILL_TAGS
_packetdrill_test(
- name = name + "_linux_test",
+ name = name,
flags = ["--dut_platform", "linux"],
**kwargs
)
@@ -75,13 +75,13 @@ def packetdrill_netstack_test(name, **kwargs):
if "tags" not in kwargs:
kwargs["tags"] = _PACKETDRILL_TAGS
_packetdrill_test(
- name = name + "_netstack_test",
+ name = name,
# This is the default runtime unless
# "--test_arg=--runtime=OTHER_RUNTIME" is used to override the value.
flags = ["--dut_platform", "netstack", "--runtime", "runsc-d"],
**kwargs
)
-def packetdrill_test(**kwargs):
- packetdrill_linux_test(**kwargs)
- packetdrill_netstack_test(**kwargs)
+def packetdrill_test(name, **kwargs):
+ packetdrill_linux_test(name + "_linux_test", **kwargs)
+ packetdrill_netstack_test(name + "_netstack_test", **kwargs)
diff --git a/test/packetdrill/linux/tcp_user_timeout.pkt b/test/packetdrill/linux/tcp_user_timeout.pkt
new file mode 100644
index 000000000..38018cb42
--- /dev/null
+++ b/test/packetdrill/linux/tcp_user_timeout.pkt
@@ -0,0 +1,39 @@
+// Test that a socket w/ TCP_USER_TIMEOUT set aborts the connection
+// if there is pending unacked data after the user specified timeout.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
++0.1 < . 1:1(0) ack 1 win 32792
+
++0.100 accept(3, ..., ...) = 4
+
+// Okay, we received nothing, and decide to close this idle socket.
+// We set TCP_USER_TIMEOUT to 3 seconds because really it is not worth
+// trying hard to cleanly close this flow, at the price of keeping
+// a TCP structure in kernel for about 1 minute!
++2 setsockopt(4, SOL_TCP, TCP_USER_TIMEOUT, [3000], 4) = 0
+
+// The write/ack is required mainly for netstack as netstack does
+// not update its RTO during the handshake.
++0 write(4, ..., 100) = 100
++0 > P. 1:101(100) ack 1 <...>
++0 < . 1:1(0) ack 101 win 32792
+
++0 close(4) = 0
+
++0 > F. 101:101(0) ack 1 <...>
++.3~+.400 > F. 101:101(0) ack 1 <...>
++.3~+.400 > F. 101:101(0) ack 1 <...>
++.6~+.800 > F. 101:101(0) ack 1 <...>
++1.2~+1.300 > F. 101:101(0) ack 1 <...>
+
+// We finally receive something from the peer, but it is way too late
+// Our socket vanished because TCP_USER_TIMEOUT was really small.
++.1 < . 1:2(1) ack 102 win 32792
++0 > R 102:102(0) win 0
diff --git a/test/packetdrill/listen_close_before_handshake_complete.pkt b/test/packetdrill/listen_close_before_handshake_complete.pkt
new file mode 100644
index 000000000..51c3f1a32
--- /dev/null
+++ b/test/packetdrill/listen_close_before_handshake_complete.pkt
@@ -0,0 +1,31 @@
+// Test that closing a listening socket closes any connections in SYN-RCVD
+// state and any packets bound for these connections generate a RESET.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
+
++0.100 close(3) = 0
++0.1 < P. 1:1(0) ack 1 win 257
+
+// Linux generates a reset with no ack number/bit set. This is contradictory to
+// what is specified in Rule 1 under Reset Generation in
+// https://tools.ietf.org/html/rfc793#section-3.4.
+// "1. If the connection does not exist (CLOSED) then a reset is sent
+// in response to any incoming segment except another reset. In
+// particular, SYNs addressed to a non-existent connection are rejected
+// by this means.
+//
+// If the incoming segment has an ACK field, the reset takes its
+// sequence number from the ACK field of the segment, otherwise the
+// reset has sequence number zero and the ACK field is set to the sum
+// of the sequence number and segment length of the incoming segment.
+// The connection remains in the CLOSED state."
+
++0.0 > R 1:1(0) win 0 \ No newline at end of file
diff --git a/test/packetdrill/netstack/tcp_user_timeout.pkt b/test/packetdrill/netstack/tcp_user_timeout.pkt
new file mode 100644
index 000000000..60103adba
--- /dev/null
+++ b/test/packetdrill/netstack/tcp_user_timeout.pkt
@@ -0,0 +1,38 @@
+// Test that a socket w/ TCP_USER_TIMEOUT set aborts the connection
+// if there is pending unacked data after the user specified timeout.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
++0.1 < . 1:1(0) ack 1 win 32792
+
++0.100 accept(3, ..., ...) = 4
+
+// Okay, we received nothing, and decide to close this idle socket.
+// We set TCP_USER_TIMEOUT to 3 seconds because really it is not worth
+// trying hard to cleanly close this flow, at the price of keeping
+// a TCP structure in kernel for about 1 minute!
++2 setsockopt(4, SOL_TCP, TCP_USER_TIMEOUT, [3000], 4) = 0
+
+// The write/ack is required mainly for netstack as netstack does
+// not update its RTO during the handshake.
++0 write(4, ..., 100) = 100
++0 > P. 1:101(100) ack 1 <...>
++0 < . 1:1(0) ack 101 win 32792
+
++0 close(4) = 0
+
++0 > F. 101:101(0) ack 1 <...>
++.2~+.300 > F. 101:101(0) ack 1 <...>
++.4~+.500 > F. 101:101(0) ack 1 <...>
++.8~+.900 > F. 101:101(0) ack 1 <...>
+
+// We finally receive something from the peer, but it is way too late
+// Our socket vanished because TCP_USER_TIMEOUT was really small.
++1.61 < . 1:2(1) ack 102 win 32792
++0 > R 102:102(0) win 0
diff --git a/test/packetdrill/no_rst_to_rst.pkt b/test/packetdrill/no_rst_to_rst.pkt
new file mode 100644
index 000000000..612747827
--- /dev/null
+++ b/test/packetdrill/no_rst_to_rst.pkt
@@ -0,0 +1,36 @@
+// Test a RST is not generated in response to a RST and a RST is correctly
+// generated when an accepted endpoint is RST due to an incoming RST.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
++0 < P. 1:1(0) ack 1 win 257
+
++0.100 accept(3, ..., ...) = 4
+
++0.200 < R 1:1(0) win 0
+
++0.300 read(4,..., 4) = -1 ECONNRESET (Connection Reset by Peer)
+
++0.00 < . 1:1(0) ack 1 win 257
+
+// Linux generates a reset with no ack number/bit set. This is contradictory to
+// what is specified in Rule 1 under Reset Generation in
+// https://tools.ietf.org/html/rfc793#section-3.4.
+// "1. If the connection does not exist (CLOSED) then a reset is sent
+// in response to any incoming segment except another reset. In
+// particular, SYNs addressed to a non-existent connection are rejected
+// by this means.
+//
+// If the incoming segment has an ACK field, the reset takes its
+// sequence number from the ACK field of the segment, otherwise the
+// reset has sequence number zero and the ACK field is set to the sum
+// of the sequence number and segment length of the incoming segment.
+// The connection remains in the CLOSED state."
+
++0.00 > R 1:1(0) win 0 \ No newline at end of file
diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh
index 0b22dfd5c..c8268170f 100755
--- a/test/packetdrill/packetdrill_test.sh
+++ b/test/packetdrill/packetdrill_test.sh
@@ -91,8 +91,8 @@ fi
# Variables specific to the test runner start with TEST_RUNNER_.
declare -r PACKETDRILL="/packetdrill/gtests/net/packetdrill/packetdrill"
# Use random numbers so that test networks don't collide.
-declare -r CTRL_NET="ctrl_net-${RANDOM}${RANDOM}"
-declare -r TEST_NET="test_net-${RANDOM}${RANDOM}"
+declare -r CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)"
+declare -r TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)"
declare -r tolerance_usecs=100000
# On both DUT and test runner, testing packets are on the eth2 interface.
declare -r TEST_DEVICE="eth2"
diff --git a/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt b/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt
new file mode 100644
index 000000000..a86b90ce6
--- /dev/null
+++ b/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt
@@ -0,0 +1,9 @@
+// Test that a listening socket generates a RST when it receives an
+// ACK and syn cookies are not in use.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
++0.1 < . 1:1(0) ack 1 win 32792
++0 > R 1:1(0) ack 0 win 0 \ No newline at end of file
diff --git a/test/packetdrill/sanity_test.pkt b/test/packetdrill/sanity_test.pkt
new file mode 100644
index 000000000..b3b58c366
--- /dev/null
+++ b/test/packetdrill/sanity_test.pkt
@@ -0,0 +1,7 @@
+// Basic sanity test. One system call.
+//
+// All of the plumbing has to be working however, and the packetdrill wire
+// client needs to be able to connect to the wire server and send the script,
+// probe local interfaces, run through the test w/ timings, etc.
+
+0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
diff --git a/test/packetdrill/tcp_defer_accept.pkt b/test/packetdrill/tcp_defer_accept.pkt
new file mode 100644
index 000000000..a17f946db
--- /dev/null
+++ b/test/packetdrill/tcp_defer_accept.pkt
@@ -0,0 +1,48 @@
+// Test that a bare ACK does not complete a connection when TCP_DEFER_ACCEPT
+// timeout is not hit but an ACK w/ data does complete and deliver the
+// connection to the accept queue.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 setsockopt(3, SOL_TCP, TCP_DEFER_ACCEPT, [5], 4) = 0
++0.000 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1 <...>
+
+// Send a bare ACK this should not complete the connection as we
+// set the TCP_DEFER_ACCEPT above.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The bare ACK should be dropped and no connection should be delivered
+// to the accept queue.
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// Send another bare ACK and it should still fail we set TCP_DEFER_ACCEPT
+// to 5 seconds above.
++2.5 < . 1:1(0) ack 1 win 257
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// set accept socket back to blocking.
++0.000 fcntl(3, F_SETFL, O_RDWR) = 0
+
+// Now send an ACK w/ data. This should complete the connection
+// and deliver the socket to the accept queue.
++0.1 < . 1:5(4) ack 1 win 257
++0.0 > . 1:1(0) ack 5 <...>
+
+// This should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
+// Now read the data and we should get 4 bytes.
++0.000 read(4,..., 4) = 4
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 5 <...>
++0.0 < F. 5:5(0) ack 2 win 257
++0.01 > . 2:2(0) ack 6 <...> \ No newline at end of file
diff --git a/test/packetdrill/tcp_defer_accept_timeout.pkt b/test/packetdrill/tcp_defer_accept_timeout.pkt
new file mode 100644
index 000000000..201fdeb14
--- /dev/null
+++ b/test/packetdrill/tcp_defer_accept_timeout.pkt
@@ -0,0 +1,48 @@
+// Test that a bare ACK is accepted after TCP_DEFER_ACCEPT timeout
+// is hit and a connection is delivered.
+
+0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 setsockopt(3, SOL_TCP, TCP_DEFER_ACCEPT, [3], 4) = 0
++0.000 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1 <...>
+
+// Send a bare ACK this should not complete the connection as we
+// set the TCP_DEFER_ACCEPT above.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The bare ACK should be dropped and no connection should be delivered
+// to the accept queue.
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// Send another bare ACK and it should still fail we set TCP_DEFER_ACCEPT
+// to 5 seconds above.
++2.5 < . 1:1(0) ack 1 win 257
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// set accept socket back to blocking.
++0.000 fcntl(3, F_SETFL, O_RDWR) = 0
+
+// We should see one more retransmit of the SYN-ACK as a last ditch
+// attempt when TCP_DEFER_ACCEPT timeout is hit to trigger another
+// ACK or a packet with data.
++.35~+2.35 > S. 0:0(0) ack 1 <...>
+
+// Now send another bare ACK after TCP_DEFER_ACCEPT time has been passed.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The ACK above should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 1 <...>
++0.0 < F. 1:1(0) ack 2 win 257
++0.01 > . 2:2(0) ack 2 <...>
diff --git a/test/runner/gtest/gtest.go b/test/runner/gtest/gtest.go
index f96e2415e..869169ad5 100644
--- a/test/runner/gtest/gtest.go
+++ b/test/runner/gtest/gtest.go
@@ -66,13 +66,12 @@ func (tc TestCase) Args() []string {
}
if tc.benchmark {
return []string{
- fmt.Sprintf("%s=^$", filterTestFlag),
fmt.Sprintf("%s=^%s$", filterBenchmarkFlag, tc.Name),
+ fmt.Sprintf("%s=", filterTestFlag),
}
}
return []string{
- fmt.Sprintf("%s=^%s$", filterTestFlag, tc.FullName()),
- fmt.Sprintf("%s=^$", filterBenchmarkFlag),
+ fmt.Sprintf("%s=%s", filterTestFlag, tc.FullName()),
}
}
@@ -147,6 +146,8 @@ func ParseTestCases(testBin string, benchmarks bool, extraArgs ...string) ([]Tes
return nil, fmt.Errorf("could not enumerate gtest benchmarks: %v\nstderr\n%s", err, exitErr.Stderr)
}
+ out = []byte(strings.Trim(string(out), "\n"))
+
// Parse benchmark output.
for _, line := range strings.Split(string(out), "\n") {
// Strip comments.
diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD
index a69b0ce13..9800a0cdf 100644
--- a/test/syscalls/BUILD
+++ b/test/syscalls/BUILD
@@ -318,10 +318,14 @@ syscall_test(
test = "//test/syscalls/linux:proc_test",
)
-syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test")
-
syscall_test(test = "//test/syscalls/linux:proc_net_test")
+syscall_test(test = "//test/syscalls/linux:proc_pid_oomscore_test")
+
+syscall_test(test = "//test/syscalls/linux:proc_pid_smaps_test")
+
+syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test")
+
syscall_test(
size = "medium",
test = "//test/syscalls/linux:pselect_test",
diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD
index dae2b1077..43455f1a3 100644
--- a/test/syscalls/linux/BUILD
+++ b/test/syscalls/linux/BUILD
@@ -603,7 +603,10 @@ cc_binary(
cc_binary(
name = "exceptions_test",
testonly = 1,
- srcs = ["exceptions.cc"],
+ srcs = select_arch(
+ amd64 = ["exceptions.cc"],
+ arm64 = [],
+ ),
linkstatic = 1,
deps = [
gtest,
@@ -1470,7 +1473,10 @@ cc_binary(
cc_binary(
name = "arch_prctl_test",
testonly = 1,
- srcs = ["arch_prctl.cc"],
+ srcs = select_arch(
+ amd64 = ["arch_prctl.cc"],
+ arm64 = [],
+ ),
linkstatic = 1,
deps = [
"//test/util:file_descriptor",
@@ -1626,6 +1632,19 @@ cc_binary(
)
cc_binary(
+ name = "proc_pid_oomscore_test",
+ testonly = 1,
+ srcs = ["proc_pid_oomscore.cc"],
+ linkstatic = 1,
+ deps = [
+ "//test/util:fs_util",
+ "//test/util:test_main",
+ "//test/util:test_util",
+ "@com_google_absl//absl/strings",
+ ],
+)
+
+cc_binary(
name = "proc_pid_smaps_test",
testonly = 1,
srcs = ["proc_pid_smaps.cc"],
@@ -3317,7 +3336,10 @@ cc_binary(
cc_binary(
name = "sysret_test",
testonly = 1,
- srcs = ["sysret.cc"],
+ srcs = select_arch(
+ amd64 = ["sysret.cc"],
+ arm64 = [],
+ ),
linkstatic = 1,
deps = [
gtest,
diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc
index f91187e75..5a70f6c3b 100644
--- a/test/syscalls/linux/proc.cc
+++ b/test/syscalls/linux/proc.cc
@@ -1431,6 +1431,12 @@ TEST(ProcPidFile, SubprocessRunning) {
EXPECT_THAT(ReadWhileRunning("uid_map", buf, sizeof(buf)),
SyscallSucceedsWithValue(sizeof(buf)));
+
+ EXPECT_THAT(ReadWhileRunning("oom_score", buf, sizeof(buf)),
+ SyscallSucceedsWithValue(sizeof(buf)));
+
+ EXPECT_THAT(ReadWhileRunning("oom_score_adj", buf, sizeof(buf)),
+ SyscallSucceedsWithValue(sizeof(buf)));
}
// Test whether /proc/PID/ files can be read for a zombie process.
@@ -1466,6 +1472,12 @@ TEST(ProcPidFile, SubprocessZombie) {
EXPECT_THAT(ReadWhileZombied("uid_map", buf, sizeof(buf)),
SyscallSucceedsWithValue(sizeof(buf)));
+ EXPECT_THAT(ReadWhileZombied("oom_score", buf, sizeof(buf)),
+ SyscallSucceedsWithValue(sizeof(buf)));
+
+ EXPECT_THAT(ReadWhileZombied("oom_score_adj", buf, sizeof(buf)),
+ SyscallSucceedsWithValue(sizeof(buf)));
+
// FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux
// on proc files.
//
@@ -1527,6 +1539,15 @@ TEST(ProcPidFile, SubprocessExited) {
EXPECT_THAT(ReadWhileExited("uid_map", buf, sizeof(buf)),
SyscallSucceedsWithValue(sizeof(buf)));
+
+ if (!IsRunningOnGvisor()) {
+ // FIXME(gvisor.dev/issue/164): Succeeds on gVisor.
+ EXPECT_THAT(ReadWhileExited("oom_score", buf, sizeof(buf)),
+ SyscallFailsWithErrno(ESRCH));
+ }
+
+ EXPECT_THAT(ReadWhileExited("oom_score_adj", buf, sizeof(buf)),
+ SyscallFailsWithErrno(ESRCH));
}
PosixError DirContainsImpl(absl::string_view path,
diff --git a/test/syscalls/linux/proc_pid_oomscore.cc b/test/syscalls/linux/proc_pid_oomscore.cc
new file mode 100644
index 000000000..707821a3f
--- /dev/null
+++ b/test/syscalls/linux/proc_pid_oomscore.cc
@@ -0,0 +1,72 @@
+// 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.
+
+#include <errno.h>
+
+#include <exception>
+#include <iostream>
+#include <string>
+
+#include "test/util/fs_util.h"
+#include "test/util/test_util.h"
+
+namespace gvisor {
+namespace testing {
+
+namespace {
+
+PosixErrorOr<int> ReadProcNumber(std::string path) {
+ ASSIGN_OR_RETURN_ERRNO(std::string contents, GetContents(path));
+ EXPECT_EQ(contents[contents.length() - 1], '\n');
+
+ int num;
+ if (!absl::SimpleAtoi(contents, &num)) {
+ return PosixError(EINVAL, "invalid value: " + contents);
+ }
+
+ return num;
+}
+
+TEST(ProcPidOomscoreTest, BasicRead) {
+ auto const oom_score =
+ ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score"));
+ EXPECT_LE(oom_score, 1000);
+ EXPECT_GE(oom_score, -1000);
+}
+
+TEST(ProcPidOomscoreAdjTest, BasicRead) {
+ auto const oom_score =
+ ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score_adj"));
+
+ // oom_score_adj defaults to 0.
+ EXPECT_EQ(oom_score, 0);
+}
+
+TEST(ProcPidOomscoreAdjTest, BasicWrite) {
+ constexpr int test_value = 7;
+ FileDescriptor fd =
+ ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/oom_score_adj", O_WRONLY));
+ ASSERT_THAT(
+ RetryEINTR(write)(fd.get(), std::to_string(test_value).c_str(), 1),
+ SyscallSucceeds());
+
+ auto const oom_score =
+ ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score_adj"));
+ EXPECT_EQ(oom_score, test_value);
+}
+
+} // namespace
+
+} // namespace testing
+} // namespace gvisor
diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc
index 579463384..d9c1ac0e1 100644
--- a/test/syscalls/linux/tcp_socket.cc
+++ b/test/syscalls/linux/tcp_socket.cc
@@ -143,6 +143,20 @@ TEST_P(TcpSocketTest, ConnectOnEstablishedConnection) {
SyscallFailsWithErrno(EISCONN));
}
+TEST_P(TcpSocketTest, ShutdownWriteInTimeWait) {
+ EXPECT_THAT(shutdown(t_, SHUT_WR), SyscallSucceeds());
+ EXPECT_THAT(shutdown(s_, SHUT_RDWR), SyscallSucceeds());
+ absl::SleepFor(absl::Seconds(1)); // Wait to enter TIME_WAIT.
+ EXPECT_THAT(shutdown(t_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN));
+}
+
+TEST_P(TcpSocketTest, ShutdownWriteInFinWait1) {
+ EXPECT_THAT(shutdown(t_, SHUT_WR), SyscallSucceeds());
+ EXPECT_THAT(shutdown(t_, SHUT_WR), SyscallSucceeds());
+ absl::SleepFor(absl::Seconds(1)); // Wait to enter FIN-WAIT2.
+ EXPECT_THAT(shutdown(t_, SHUT_WR), SyscallSucceeds());
+}
+
TEST_P(TcpSocketTest, DataCoalesced) {
char buf[10];
diff --git a/test/syscalls/linux/tuntap_hostinet.cc b/test/syscalls/linux/tuntap_hostinet.cc
index 0c527419e..1513fb9d5 100644
--- a/test/syscalls/linux/tuntap_hostinet.cc
+++ b/test/syscalls/linux/tuntap_hostinet.cc
@@ -26,6 +26,7 @@ namespace {
TEST(TuntapHostInetTest, NoNetTun) {
SKIP_IF(!IsRunningOnGvisor());
+ SKIP_IF(!IsRunningWithHostinet());
struct stat statbuf;
ASSERT_THAT(stat("/dev/net/tun", &statbuf), SyscallFailsWithErrno(ENOENT));
diff --git a/vdso/vdso.cc b/vdso/vdso.cc
index 8bb80a7a4..c2585d592 100644
--- a/vdso/vdso.cc
+++ b/vdso/vdso.cc
@@ -126,6 +126,10 @@ extern "C" int __kernel_clock_getres(clockid_t clock, struct timespec* res) {
case CLOCK_REALTIME:
case CLOCK_MONOTONIC:
case CLOCK_BOOTTIME: {
+ if (res == nullptr) {
+ return 0;
+ }
+
res->tv_sec = 0;
res->tv_nsec = 1;
break;