5 files changed, 574 insertions, 0 deletions
diff --git a/pkg/sentry/fsimpl/eventfd/BUILD b/pkg/sentry/fsimpl/eventfd/BUILD
new file mode 100644
index 000000000..ea167d38c
--- /dev/null
+++ b/pkg/sentry/fsimpl/eventfd/BUILD
@@ -0,0 +1,33 @@
+load("//tools:defs.bzl", "go_library", "go_test")
+
+licenses(["notice"])
+
+go_library(
+    name = "eventfd",
+    srcs = ["eventfd.go"],
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/context",
+        "//pkg/fdnotifier",
+        "//pkg/log",
+        "//pkg/sentry/vfs",
+        "//pkg/syserror",
+        "//pkg/usermem",
+        "//pkg/waiter",
+    ],
+)
+
+go_test(
+    name = "eventfd_test",
+    size = "small",
+    srcs = ["eventfd_test.go"],
+    library = ":eventfd",
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/sentry/contexttest",
+        "//pkg/sentry/vfs",
+        "//pkg/usermem",
+        "//pkg/waiter",
+    ],
+)
diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go
new file mode 100644
index 000000000..c573d7935
--- /dev/null
+++ b/pkg/sentry/fsimpl/eventfd/eventfd.go
@@ -0,0 +1,284 @@
+// 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 eventfd implements event fds.
+package eventfd
+
+import (
+	"math"
+	"sync"
+	"syscall"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/fdnotifier"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+// EventFileDescription implements FileDescriptionImpl for file-based event
+// notification (eventfd). Eventfds are usually internal to the Sentry but in
+// certain situations they may be converted into a host-backed eventfd.
+type EventFileDescription struct {
+	vfsfd vfs.FileDescription
+	vfs.FileDescriptionDefaultImpl
+	vfs.DentryMetadataFileDescriptionImpl
+
+	// queue is used to notify interested parties when the event object
+	// becomes readable or writable.
+	queue waiter.Queue `state:"zerovalue"`
+
+	// mu protects the fields below.
+	mu sync.Mutex `state:"nosave"`
+
+	// val is the current value of the event counter.
+	val uint64
+
+	// semMode specifies whether the event is in "semaphore" mode.
+	semMode bool
+
+	// hostfd indicates whether this eventfd is passed through to the host.
+	hostfd int
+}
+
+var _ vfs.FileDescriptionImpl = (*EventFileDescription)(nil)
+
+// New creates a new event fd.
+func New(vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) {
+	vd := vfsObj.NewAnonVirtualDentry("[eventfd]")
+	defer vd.DecRef()
+	efd := &EventFileDescription{
+		val:     initVal,
+		semMode: semMode,
+		hostfd:  -1,
+	}
+	if err := efd.vfsfd.Init(efd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{
+		UseDentryMetadata: true,
+		DenyPRead:         true,
+		DenyPWrite:        true,
+	}); err != nil {
+		return nil, err
+	}
+	return &efd.vfsfd, nil
+}
+
+// HostFD returns the host eventfd associated with this event.
+func (efd *EventFileDescription) HostFD() (int, error) {
+	efd.mu.Lock()
+	defer efd.mu.Unlock()
+	if efd.hostfd >= 0 {
+		return efd.hostfd, nil
+	}
+
+	flags := linux.EFD_NONBLOCK
+	if efd.semMode {
+		flags |= linux.EFD_SEMAPHORE
+	}
+
+	fd, _, errno := syscall.Syscall(syscall.SYS_EVENTFD2, uintptr(efd.val), uintptr(flags), 0)
+	if errno != 0 {
+		return -1, errno
+	}
+
+	if err := fdnotifier.AddFD(int32(fd), &efd.queue); err != nil {
+		if closeErr := syscall.Close(int(fd)); closeErr != nil {
+			log.Warningf("close(%d) eventfd failed: %v", fd, closeErr)
+		}
+		return -1, err
+	}
+
+	efd.hostfd = int(fd)
+	return efd.hostfd, nil
+}
+
+// Release implements FileDescriptionImpl.Release()
+func (efd *EventFileDescription) Release() {
+	efd.mu.Lock()
+	defer efd.mu.Unlock()
+	if efd.hostfd >= 0 {
+		fdnotifier.RemoveFD(int32(efd.hostfd))
+		if closeErr := syscall.Close(int(efd.hostfd)); closeErr != nil {
+			log.Warningf("close(%d) eventfd failed: %v", efd.hostfd, closeErr)
+		}
+		efd.hostfd = -1
+	}
+}
+
+// Read implements FileDescriptionImpl.Read.
+func (efd *EventFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) {
+	if dst.NumBytes() < 8 {
+		return 0, syscall.EINVAL
+	}
+	if err := efd.read(ctx, dst); err != nil {
+		return 0, err
+	}
+	return 8, nil
+}
+
+// Write implements FileDescriptionImpl.Write.
+func (efd *EventFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) {
+	if src.NumBytes() < 8 {
+		return 0, syscall.EINVAL
+	}
+	if err := efd.write(ctx, src); err != nil {
+		return 0, err
+	}
+	return 8, nil
+}
+
+// Preconditions: Must be called with efd.mu locked.
+func (efd *EventFileDescription) hostReadLocked(ctx context.Context, dst usermem.IOSequence) error {
+	var buf [8]byte
+	if _, err := syscall.Read(efd.hostfd, buf[:]); err != nil {
+		if err == syscall.EWOULDBLOCK {
+			return syserror.ErrWouldBlock
+		}
+		return err
+	}
+	_, err := dst.CopyOut(ctx, buf[:])
+	return err
+}
+
+func (efd *EventFileDescription) read(ctx context.Context, dst usermem.IOSequence) error {
+	efd.mu.Lock()
+	if efd.hostfd >= 0 {
+		defer efd.mu.Unlock()
+		return efd.hostReadLocked(ctx, dst)
+	}
+
+	// We can't complete the read if the value is currently zero.
+	if efd.val == 0 {
+		efd.mu.Unlock()
+		return syserror.ErrWouldBlock
+	}
+
+	// Update the value based on the mode the event is operating in.
+	var val uint64
+	if efd.semMode {
+		val = 1
+		// Consistent with Linux, this is done even if writing to memory fails.
+		efd.val--
+	} else {
+		val = efd.val
+		efd.val = 0
+	}
+
+	efd.mu.Unlock()
+
+	// Notify writers. We do this even if we were already writable because
+	// it is possible that a writer is waiting to write the maximum value
+	// to the event.
+	efd.queue.Notify(waiter.EventOut)
+
+	var buf [8]byte
+	usermem.ByteOrder.PutUint64(buf[:], val)
+	_, err := dst.CopyOut(ctx, buf[:])
+	return err
+}
+
+// Preconditions: Must be called with efd.mu locked.
+func (efd *EventFileDescription) hostWriteLocked(val uint64) error {
+	var buf [8]byte
+	usermem.ByteOrder.PutUint64(buf[:], val)
+	_, err := syscall.Write(efd.hostfd, buf[:])
+	if err == syscall.EWOULDBLOCK {
+		return syserror.ErrWouldBlock
+	}
+	return err
+}
+
+func (efd *EventFileDescription) write(ctx context.Context, src usermem.IOSequence) error {
+	var buf [8]byte
+	if _, err := src.CopyIn(ctx, buf[:]); err != nil {
+		return err
+	}
+	val := usermem.ByteOrder.Uint64(buf[:])
+
+	return efd.Signal(val)
+}
+
+// Signal is an internal function to signal the event fd.
+func (efd *EventFileDescription) Signal(val uint64) error {
+	if val == math.MaxUint64 {
+		return syscall.EINVAL
+	}
+
+	efd.mu.Lock()
+
+	if efd.hostfd >= 0 {
+		defer efd.mu.Unlock()
+		return efd.hostWriteLocked(val)
+	}
+
+	// We only allow writes that won't cause the value to go over the max
+	// uint64 minus 1.
+	if val > math.MaxUint64-1-efd.val {
+		efd.mu.Unlock()
+		return syserror.ErrWouldBlock
+	}
+
+	efd.val += val
+	efd.mu.Unlock()
+
+	// Always trigger a notification.
+	efd.queue.Notify(waiter.EventIn)
+
+	return nil
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (efd *EventFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask {
+	efd.mu.Lock()
+	defer efd.mu.Unlock()
+
+	if efd.hostfd >= 0 {
+		return fdnotifier.NonBlockingPoll(int32(efd.hostfd), mask)
+	}
+
+	ready := waiter.EventMask(0)
+	if efd.val > 0 {
+		ready |= waiter.EventIn
+	}
+
+	if efd.val < math.MaxUint64-1 {
+		ready |= waiter.EventOut
+	}
+
+	return mask & ready
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (efd *EventFileDescription) EventRegister(entry *waiter.Entry, mask waiter.EventMask) {
+	efd.queue.EventRegister(entry, mask)
+
+	efd.mu.Lock()
+	defer efd.mu.Unlock()
+	if efd.hostfd >= 0 {
+		fdnotifier.UpdateFD(int32(efd.hostfd))
+	}
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (efd *EventFileDescription) EventUnregister(entry *waiter.Entry) {
+	efd.queue.EventUnregister(entry)
+
+	efd.mu.Lock()
+	defer efd.mu.Unlock()
+	if efd.hostfd >= 0 {
+		fdnotifier.UpdateFD(int32(efd.hostfd))
+	}
+}
diff --git a/pkg/sentry/fsimpl/eventfd/eventfd_test.go b/pkg/sentry/fsimpl/eventfd/eventfd_test.go
new file mode 100644
index 000000000..20e3adffc
--- /dev/null
+++ b/pkg/sentry/fsimpl/eventfd/eventfd_test.go
@@ -0,0 +1,97 @@
+// 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 eventfd
+
+import (
+	"testing"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/sentry/contexttest"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/usermem"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+func TestEventFD(t *testing.T) {
+	initVals := []uint64{
+		0,
+		// Using a non-zero initial value verifies that writing to an
+		// eventfd signals when the eventfd's counter was already
+		// non-zero.
+		343,
+	}
+
+	for _, initVal := range initVals {
+		ctx := contexttest.Context(t)
+		vfsObj := &vfs.VirtualFilesystem{}
+		if err := vfsObj.Init(); err != nil {
+			t.Fatalf("VFS init: %v", err)
+		}
+
+		// Make a new eventfd that is writable.
+		eventfd, err := New(vfsObj, initVal, false, linux.O_RDWR)
+		if err != nil {
+			t.Fatalf("New() failed: %v", err)
+		}
+		defer eventfd.DecRef()
+
+		// Register a callback for a write event.
+		w, ch := waiter.NewChannelEntry(nil)
+		eventfd.EventRegister(&w, waiter.EventIn)
+		defer eventfd.EventUnregister(&w)
+
+		data := []byte("00000124")
+		// Create and submit a write request.
+		n, err := eventfd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if n != 8 {
+			t.Errorf("eventfd.write wrote %d bytes, not full int64", n)
+		}
+
+		// Check if the callback fired due to the write event.
+		select {
+		case <-ch:
+		default:
+			t.Errorf("Didn't get notified of EventIn after write")
+		}
+	}
+}
+
+func TestEventFDStat(t *testing.T) {
+	ctx := contexttest.Context(t)
+	vfsObj := &vfs.VirtualFilesystem{}
+	if err := vfsObj.Init(); err != nil {
+		t.Fatalf("VFS init: %v", err)
+	}
+
+	// Make a new eventfd that is writable.
+	eventfd, err := New(vfsObj, 0, false, linux.O_RDWR)
+	if err != nil {
+		t.Fatalf("New() failed: %v", err)
+	}
+	defer eventfd.DecRef()
+
+	statx, err := eventfd.Stat(ctx, vfs.StatOptions{
+		Mask: linux.STATX_BASIC_STATS,
+	})
+	if err != nil {
+		t.Fatalf("eventfd.Stat failed: %v", err)
+	}
+	if statx.Size != 0 {
+		t.Errorf("eventfd size should be 0")
+	}
+}
diff --git a/pkg/sentry/fsimpl/timerfd/BUILD b/pkg/sentry/fsimpl/timerfd/BUILD
new file mode 100644
index 000000000..fbb02a271
--- /dev/null
+++ b/pkg/sentry/fsimpl/timerfd/BUILD
@@ -0,0 +1,17 @@
+load("//tools:defs.bzl", "go_library")
+
+licenses(["notice"])
+
+go_library(
+    name = "timerfd",
+    srcs = ["timerfd.go"],
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/context",
+        "//pkg/sentry/kernel/time",
+        "//pkg/sentry/vfs",
+        "//pkg/syserror",
+        "//pkg/usermem",
+        "//pkg/waiter",
+    ],
+)
diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go
new file mode 100644
index 000000000..60c92d626
--- /dev/null
+++ b/pkg/sentry/fsimpl/timerfd/timerfd.go
@@ -0,0 +1,143 @@
+// 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 timerfd implements timer fds.
+package timerfd
+
+import (
+	"sync/atomic"
+
+	"gvisor.dev/gvisor/pkg/context"
+	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+// TimerFileDescription implements FileDescriptionImpl for timer fds. It also
+// implements ktime.TimerListener.
+type TimerFileDescription struct {
+	vfsfd vfs.FileDescription
+	vfs.FileDescriptionDefaultImpl
+	vfs.DentryMetadataFileDescriptionImpl
+
+	events waiter.Queue
+	timer  *ktime.Timer
+
+	// val is the number of timer expirations since the last successful
+	// call to PRead, or SetTime. val must be accessed using atomic memory
+	// operations.
+	val uint64
+}
+
+var _ vfs.FileDescriptionImpl = (*TimerFileDescription)(nil)
+var _ ktime.TimerListener = (*TimerFileDescription)(nil)
+
+// New returns a new timer fd.
+func New(vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) {
+	vd := vfsObj.NewAnonVirtualDentry("[timerfd]")
+	defer vd.DecRef()
+	tfd := &TimerFileDescription{}
+	tfd.timer = ktime.NewTimer(clock, tfd)
+	if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{
+		UseDentryMetadata: true,
+		DenyPRead:         true,
+		DenyPWrite:        true,
+	}); err != nil {
+		return nil, err
+	}
+	return &tfd.vfsfd, nil
+}
+
+// Read implements FileDescriptionImpl.Read.
+func (tfd *TimerFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+	const sizeofUint64 = 8
+	if dst.NumBytes() < sizeofUint64 {
+		return 0, syserror.EINVAL
+	}
+	if val := atomic.SwapUint64(&tfd.val, 0); val != 0 {
+		var buf [sizeofUint64]byte
+		usermem.ByteOrder.PutUint64(buf[:], val)
+		if _, err := dst.CopyOut(ctx, buf[:]); err != nil {
+			// Linux does not undo consuming the number of
+			// expirations even if writing to userspace fails.
+			return 0, err
+		}
+		return sizeofUint64, nil
+	}
+	return 0, syserror.ErrWouldBlock
+}
+
+// Clock returns the timer fd's Clock.
+func (tfd *TimerFileDescription) Clock() ktime.Clock {
+	return tfd.timer.Clock()
+}
+
+// GetTime returns the associated Timer's setting and the time at which it was
+// observed.
+func (tfd *TimerFileDescription) GetTime() (ktime.Time, ktime.Setting) {
+	return tfd.timer.Get()
+}
+
+// SetTime atomically changes the associated Timer's setting, resets the number
+// of expirations to 0, and returns the previous setting and the time at which
+// it was observed.
+func (tfd *TimerFileDescription) SetTime(s ktime.Setting) (ktime.Time, ktime.Setting) {
+	return tfd.timer.SwapAnd(s, func() { atomic.StoreUint64(&tfd.val, 0) })
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (tfd *TimerFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask {
+	var ready waiter.EventMask
+	if atomic.LoadUint64(&tfd.val) != 0 {
+		ready |= waiter.EventIn
+	}
+	return ready
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (tfd *TimerFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
+	tfd.events.EventRegister(e, mask)
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (tfd *TimerFileDescription) EventUnregister(e *waiter.Entry) {
+	tfd.events.EventUnregister(e)
+}
+
+// PauseTimer pauses the associated Timer.
+func (tfd *TimerFileDescription) PauseTimer() {
+	tfd.timer.Pause()
+}
+
+// ResumeTimer resumes the associated Timer.
+func (tfd *TimerFileDescription) ResumeTimer() {
+	tfd.timer.Resume()
+}
+
+// Release implements FileDescriptionImpl.Release()
+func (tfd *TimerFileDescription) Release() {
+	tfd.timer.Destroy()
+}
+
+// Notify implements ktime.TimerListener.Notify.
+func (tfd *TimerFileDescription) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) {
+	atomic.AddUint64(&tfd.val, exp)
+	tfd.events.Notify(waiter.EventIn)
+	return ktime.Setting{}, false
+}
+
+// Destroy implements ktime.TimerListener.Destroy.
+func (tfd *TimerFileDescription) Destroy() {}