19 files changed, 612 insertions, 27 deletions

diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index e61d39c82..aba2414d4 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -1,9 +1,11 @@
 load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library")
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("@rules_cc//cc:defs.bzl", "cc_proto_library")
 
 package(licenses = ["notice"])
 
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_template_instance(
     name = "pending_signals_list",
@@ -83,6 +85,12 @@ proto_library(
     deps = ["//pkg/sentry/arch:registers_proto"],
 )
 
+cc_proto_library(
+    name = "uncaught_signal_cc_proto",
+    visibility = ["//visibility:public"],
+    deps = [":uncaught_signal_proto"],
+)
+
 go_proto_library(
     name = "uncaught_signal_go_proto",
     importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto",
@@ -144,6 +152,7 @@ go_library(
         "threads.go",
         "timekeeper.go",
         "timekeeper_state.go",
+        "tty.go",
         "uts_namespace.go",
         "vdso.go",
         "version.go",
diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD
index f46c43128..65427b112 100644
--- a/pkg/sentry/kernel/epoll/BUILD
+++ b/pkg/sentry/kernel/epoll/BUILD
@@ -1,7 +1,9 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+
 package(licenses = ["notice"])
 
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_template_instance(
     name = "epoll_list",
diff --git a/pkg/sentry/kernel/eventfd/BUILD b/pkg/sentry/kernel/eventfd/BUILD
index 1c5f979d4..983ca67ed 100644
--- a/pkg/sentry/kernel/eventfd/BUILD
+++ b/pkg/sentry/kernel/eventfd/BUILD
@@ -1,6 +1,8 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+
 package(licenses = ["notice"])
 
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_library(
     name = "eventfd",
diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD
index 6a31dc044..41f44999c 100644
--- a/pkg/sentry/kernel/futex/BUILD
+++ b/pkg/sentry/kernel/futex/BUILD
@@ -1,7 +1,9 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+
 package(licenses = ["notice"])
 
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_template_instance(
     name = "atomicptr_bucket",
diff --git a/pkg/sentry/kernel/memevent/BUILD b/pkg/sentry/kernel/memevent/BUILD
index ebcfaa619..d7a7d1169 100644
--- a/pkg/sentry/kernel/memevent/BUILD
+++ b/pkg/sentry/kernel/memevent/BUILD
@@ -1,5 +1,6 @@
 load("//tools/go_stateify:defs.bzl", "go_library")
 load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library")
+load("@rules_cc//cc:defs.bzl", "cc_proto_library")
 
 package(licenses = ["notice"])
 
@@ -24,6 +25,12 @@ proto_library(
     visibility = ["//visibility:public"],
 )
 
+cc_proto_library(
+    name = "memory_events_cc_proto",
+    visibility = ["//visibility:public"],
+    deps = [":memory_events_proto"],
+)
+
 go_proto_library(
     name = "memory_events_go_proto",
     importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/memevent/memory_events_go_proto",
diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD
index 4d15cca85..2ce8952e2 100644
--- a/pkg/sentry/kernel/pipe/BUILD
+++ b/pkg/sentry/kernel/pipe/BUILD
@@ -1,7 +1,9 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+
 package(licenses = ["notice"])
 
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_template_instance(
     name = "buffer_list",
diff --git a/pkg/sentry/kernel/pipe/buffer.go b/pkg/sentry/kernel/pipe/buffer.go
index 69ef2a720..95bee2d37 100644
--- a/pkg/sentry/kernel/pipe/buffer.go
+++ b/pkg/sentry/kernel/pipe/buffer.go
@@ -15,6 +15,7 @@
 package pipe
 
 import (
+	"io"
 	"sync"
 
 	"gvisor.dev/gvisor/pkg/sentry/safemem"
@@ -67,6 +68,17 @@ func (b *buffer) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) {
 	return n, err
 }
 
+// WriteFromReader writes to the buffer from an io.Reader.
+func (b *buffer) WriteFromReader(r io.Reader, count int64) (int64, error) {
+	dst := b.data[b.write:]
+	if count < int64(len(dst)) {
+		dst = b.data[b.write:][:count]
+	}
+	n, err := r.Read(dst)
+	b.write += n
+	return int64(n), err
+}
+
 // ReadToBlocks implements safemem.Reader.ReadToBlocks.
 func (b *buffer) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) {
 	src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(b.data[b.read:b.write]))
@@ -75,6 +87,19 @@ func (b *buffer) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) {
 	return n, err
 }
 
+// ReadToWriter reads from the buffer into an io.Writer.
+func (b *buffer) ReadToWriter(w io.Writer, count int64, dup bool) (int64, error) {
+	src := b.data[b.read:b.write]
+	if count < int64(len(src)) {
+		src = b.data[b.read:][:count]
+	}
+	n, err := w.Write(src)
+	if !dup {
+		b.read += n
+	}
+	return int64(n), err
+}
+
 // bufferPool is a pool for buffers.
 var bufferPool = sync.Pool{
 	New: func() interface{} {
diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go
index 247e2928e..93b50669f 100644
--- a/pkg/sentry/kernel/pipe/pipe.go
+++ b/pkg/sentry/kernel/pipe/pipe.go
@@ -23,7 +23,6 @@ import (
 
 	"gvisor.dev/gvisor/pkg/sentry/context"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
-	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
@@ -173,13 +172,24 @@ func (p *Pipe) Open(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) *fs.F
 	}
 }
 
+type readOps struct {
+	// left returns the bytes remaining.
+	left func() int64
+
+	// limit limits subsequence reads.
+	limit func(int64)
+
+	// read performs the actual read operation.
+	read func(*buffer) (int64, error)
+}
+
 // read reads data from the pipe into dst and returns the number of bytes
 // read, or returns ErrWouldBlock if the pipe is empty.
 //
 // Precondition: this pipe must have readers.
-func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error) {
+func (p *Pipe) read(ctx context.Context, ops readOps) (int64, error) {
 	// Don't block for a zero-length read even if the pipe is empty.
-	if dst.NumBytes() == 0 {
+	if ops.left() == 0 {
 		return 0, nil
 	}
 
@@ -196,12 +206,12 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error)
 	}
 
 	// Limit how much we consume.
-	if dst.NumBytes() > p.size {
-		dst = dst.TakeFirst64(p.size)
+	if ops.left() > p.size {
+		ops.limit(p.size)
 	}
 
 	done := int64(0)
-	for dst.NumBytes() > 0 {
+	for ops.left() > 0 {
 		// Pop the first buffer.
 		first := p.data.Front()
 		if first == nil {
@@ -209,10 +219,9 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error)
 		}
 
 		// Copy user data.
-		n, err := dst.CopyOutFrom(ctx, first)
+		n, err := ops.read(first)
 		done += int64(n)
 		p.size -= n
-		dst = dst.DropFirst64(n)
 
 		// Empty buffer?
 		if first.Empty() {
@@ -230,12 +239,57 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error)
 	return done, nil
 }
 
+// dup duplicates all data from this pipe into the given writer.
+//
+// There is no blocking behavior implemented here. The writer may propagate
+// some blocking error. All the writes must be complete writes.
+func (p *Pipe) dup(ctx context.Context, ops readOps) (int64, error) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	// Is the pipe empty?
+	if p.size == 0 {
+		if !p.HasWriters() {
+			// See above.
+			return 0, nil
+		}
+		return 0, syserror.ErrWouldBlock
+	}
+
+	// Limit how much we consume.
+	if ops.left() > p.size {
+		ops.limit(p.size)
+	}
+
+	done := int64(0)
+	for buf := p.data.Front(); buf != nil; buf = buf.Next() {
+		n, err := ops.read(buf)
+		done += n
+		if err != nil {
+			return done, err
+		}
+	}
+
+	return done, nil
+}
+
+type writeOps struct {
+	// left returns the bytes remaining.
+	left func() int64
+
+	// limit should limit subsequent writes.
+	limit func(int64)
+
+	// write should write to the provided buffer.
+	write func(*buffer) (int64, error)
+}
+
 // write writes data from sv into the pipe and returns the number of bytes
 // written. If no bytes are written because the pipe is full (or has less than
 // atomicIOBytes free capacity), write returns ErrWouldBlock.
 //
 // Precondition: this pipe must have writers.
-func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error) {
+func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) {
 	p.mu.Lock()
 	defer p.mu.Unlock()
 
@@ -246,17 +300,16 @@ func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error)
 
 	// POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be
 	// atomic, but requires no atomicity for writes larger than this.
-	wanted := src.NumBytes()
+	wanted := ops.left()
 	if avail := p.max - p.size; wanted > avail {
 		if wanted <= p.atomicIOBytes {
 			return 0, syserror.ErrWouldBlock
 		}
-		// Limit to the available capacity.
-		src = src.TakeFirst64(avail)
+		ops.limit(avail)
 	}
 
 	done := int64(0)
-	for src.NumBytes() > 0 {
+	for ops.left() > 0 {
 		// Need a new buffer?
 		last := p.data.Back()
 		if last == nil || last.Full() {
@@ -266,10 +319,9 @@ func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error)
 		}
 
 		// Copy user data.
-		n, err := src.CopyInTo(ctx, last)
+		n, err := ops.write(last)
 		done += int64(n)
 		p.size += n
-		src = src.DropFirst64(n)
 
 		// Handle errors.
 		if err != nil {
diff --git a/pkg/sentry/kernel/pipe/reader_writer.go b/pkg/sentry/kernel/pipe/reader_writer.go
index f69dbf27b..7c307f013 100644
--- a/pkg/sentry/kernel/pipe/reader_writer.go
+++ b/pkg/sentry/kernel/pipe/reader_writer.go
@@ -15,6 +15,7 @@
 package pipe
 
 import (
+	"io"
 	"math"
 	"syscall"
 
@@ -55,7 +56,45 @@ func (rw *ReaderWriter) Release() {
 
 // Read implements fs.FileOperations.Read.
 func (rw *ReaderWriter) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) {
-	n, err := rw.Pipe.read(ctx, dst)
+	n, err := rw.Pipe.read(ctx, readOps{
+		left: func() int64 {
+			return dst.NumBytes()
+		},
+		limit: func(l int64) {
+			dst = dst.TakeFirst64(l)
+		},
+		read: func(buf *buffer) (int64, error) {
+			n, err := dst.CopyOutFrom(ctx, buf)
+			dst = dst.DropFirst64(n)
+			return n, err
+		},
+	})
+	if n > 0 {
+		rw.Pipe.Notify(waiter.EventOut)
+	}
+	return n, err
+}
+
+// WriteTo implements fs.FileOperations.WriteTo.
+func (rw *ReaderWriter) WriteTo(ctx context.Context, _ *fs.File, w io.Writer, count int64, dup bool) (int64, error) {
+	ops := readOps{
+		left: func() int64 {
+			return count
+		},
+		limit: func(l int64) {
+			count = l
+		},
+		read: func(buf *buffer) (int64, error) {
+			n, err := buf.ReadToWriter(w, count, dup)
+			count -= n
+			return n, err
+		},
+	}
+	if dup {
+		// There is no notification for dup operations.
+		return rw.Pipe.dup(ctx, ops)
+	}
+	n, err := rw.Pipe.read(ctx, ops)
 	if n > 0 {
 		rw.Pipe.Notify(waiter.EventOut)
 	}
@@ -64,7 +103,40 @@ func (rw *ReaderWriter) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequ
 
 // Write implements fs.FileOperations.Write.
 func (rw *ReaderWriter) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) {
-	n, err := rw.Pipe.write(ctx, src)
+	n, err := rw.Pipe.write(ctx, writeOps{
+		left: func() int64 {
+			return src.NumBytes()
+		},
+		limit: func(l int64) {
+			src = src.TakeFirst64(l)
+		},
+		write: func(buf *buffer) (int64, error) {
+			n, err := src.CopyInTo(ctx, buf)
+			src = src.DropFirst64(n)
+			return n, err
+		},
+	})
+	if n > 0 {
+		rw.Pipe.Notify(waiter.EventIn)
+	}
+	return n, err
+}
+
+// ReadFrom implements fs.FileOperations.WriteTo.
+func (rw *ReaderWriter) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader, count int64) (int64, error) {
+	n, err := rw.Pipe.write(ctx, writeOps{
+		left: func() int64 {
+			return count
+		},
+		limit: func(l int64) {
+			count = l
+		},
+		write: func(buf *buffer) (int64, error) {
+			n, err := buf.WriteFromReader(r, count)
+			count -= n
+			return n, err
+		},
+	})
 	if n > 0 {
 		rw.Pipe.Notify(waiter.EventIn)
 	}
diff --git a/pkg/sentry/kernel/sched/BUILD b/pkg/sentry/kernel/sched/BUILD
index 1725b8562..98ea7a0d8 100644
--- a/pkg/sentry/kernel/sched/BUILD
+++ b/pkg/sentry/kernel/sched/BUILD
@@ -1,4 +1,5 @@
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
 
 package(licenses = ["notice"])
 
diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD
index 36edf10f3..80e5e5da3 100644
--- a/pkg/sentry/kernel/semaphore/BUILD
+++ b/pkg/sentry/kernel/semaphore/BUILD
@@ -1,7 +1,9 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+
 package(licenses = ["notice"])
 
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library", "go_test")
+load("//tools/go_stateify:defs.bzl", "go_library")
 
 go_template_instance(
     name = "waiter_list",
diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go
index 81fcd8258..047b5214d 100644
--- a/pkg/sentry/kernel/sessions.go
+++ b/pkg/sentry/kernel/sessions.go
@@ -47,6 +47,11 @@ type Session struct {
 	// The id is immutable.
 	id SessionID
 
+	// foreground is the foreground process group.
+	//
+	// This is protected by TaskSet.mu.
+	foreground *ProcessGroup
+
 	// ProcessGroups is a list of process groups in this Session. This is
 	// protected by TaskSet.mu.
 	processGroups processGroupList
@@ -260,12 +265,14 @@ func (pg *ProcessGroup) SendSignal(info *arch.SignalInfo) error {
 func (tg *ThreadGroup) CreateSession() error {
 	tg.pidns.owner.mu.Lock()
 	defer tg.pidns.owner.mu.Unlock()
+	tg.signalHandlers.mu.Lock()
+	defer tg.signalHandlers.mu.Unlock()
 	return tg.createSession()
 }
 
 // createSession creates a new session for a threadgroup.
 //
-// Precondition: callers must hold TaskSet.mu for writing.
+// Precondition: callers must hold TaskSet.mu and the signal mutex for writing.
 func (tg *ThreadGroup) createSession() error {
 	// Get the ID for this thread in the current namespace.
 	id := tg.pidns.tgids[tg]
@@ -321,8 +328,14 @@ func (tg *ThreadGroup) createSession() error {
 			childTG.processGroup.incRefWithParent(pg)
 			childTG.processGroup.decRefWithParent(oldParentPG)
 		})
-		tg.processGroup.decRefWithParent(oldParentPG)
+		// If tg.processGroup is an orphan, decRefWithParent will lock
+		// the signal mutex of each thread group in tg.processGroup.
+		// However, tg's signal mutex may already be locked at this
+		// point. We change tg's process group before calling
+		// decRefWithParent to avoid locking tg's signal mutex twice.
+		oldPG := tg.processGroup
 		tg.processGroup = pg
+		oldPG.decRefWithParent(oldParentPG)
 	} else {
 		// The current process group may be nil only in the case of an
 		// unparented thread group (i.e. the init process). This would
@@ -346,6 +359,9 @@ func (tg *ThreadGroup) createSession() error {
 		ns.processGroups[ProcessGroupID(local)] = pg
 	}
 
+	// Disconnect from the controlling terminal.
+	tg.tty = nil
+
 	return nil
 }
 
diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD
new file mode 100644
index 000000000..50b69d154
--- /dev/null
+++ b/pkg/sentry/kernel/signalfd/BUILD
@@ -0,0 +1,22 @@
+package(licenses = ["notice"])
+
+load("//tools/go_stateify:defs.bzl", "go_library")
+
+go_library(
+    name = "signalfd",
+    srcs = ["signalfd.go"],
+    importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/signalfd",
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/binary",
+        "//pkg/sentry/context",
+        "//pkg/sentry/fs",
+        "//pkg/sentry/fs/anon",
+        "//pkg/sentry/fs/fsutil",
+        "//pkg/sentry/kernel",
+        "//pkg/sentry/usermem",
+        "//pkg/syserror",
+        "//pkg/waiter",
+    ],
+)
diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go
new file mode 100644
index 000000000..06fd5ec88
--- /dev/null
+++ b/pkg/sentry/kernel/signalfd/signalfd.go
@@ -0,0 +1,137 @@
+// Copyright 2019 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 signalfd provides an implementation of signal file descriptors.
+package signalfd
+
+import (
+	"sync"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/binary"
+	"gvisor.dev/gvisor/pkg/sentry/context"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/fs/anon"
+	"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+// SignalOperations represent a file with signalfd semantics.
+//
+// +stateify savable
+type SignalOperations struct {
+	fsutil.FileNoopRelease          `state:"nosave"`
+	fsutil.FilePipeSeek             `state:"nosave"`
+	fsutil.FileNotDirReaddir        `state:"nosave"`
+	fsutil.FileNoIoctl              `state:"nosave"`
+	fsutil.FileNoFsync              `state:"nosave"`
+	fsutil.FileNoMMap               `state:"nosave"`
+	fsutil.FileNoSplice             `state:"nosave"`
+	fsutil.FileNoWrite              `state:"nosave"`
+	fsutil.FileNoopFlush            `state:"nosave"`
+	fsutil.FileUseInodeUnstableAttr `state:"nosave"`
+
+	// target is the original task target.
+	//
+	// The semantics here are a bit broken. Linux will always use current
+	// for all reads, regardless of where the signalfd originated. We can't
+	// do exactly that because we need to plumb the context through
+	// EventRegister in order to support proper blocking behavior. This
+	// will undoubtedly become very complicated quickly.
+	target *kernel.Task
+
+	// mu protects below.
+	mu sync.Mutex `state:"nosave"`
+
+	// mask is the signal mask. Protected by mu.
+	mask linux.SignalSet
+}
+
+// New creates a new signalfd object with the supplied mask.
+func New(ctx context.Context, mask linux.SignalSet) (*fs.File, error) {
+	t := kernel.TaskFromContext(ctx)
+	if t == nil {
+		// No task context? Not valid.
+		return nil, syserror.EINVAL
+	}
+	// name matches fs/signalfd.c:signalfd4.
+	dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[signalfd]")
+	return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &SignalOperations{
+		target: t,
+		mask:   mask,
+	}), nil
+}
+
+// Release implements fs.FileOperations.Release.
+func (s *SignalOperations) Release() {}
+
+// Mask returns the signal mask.
+func (s *SignalOperations) Mask() linux.SignalSet {
+	s.mu.Lock()
+	mask := s.mask
+	s.mu.Unlock()
+	return mask
+}
+
+// SetMask sets the signal mask.
+func (s *SignalOperations) SetMask(mask linux.SignalSet) {
+	s.mu.Lock()
+	s.mask = mask
+	s.mu.Unlock()
+}
+
+// Read implements fs.FileOperations.Read.
+func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) {
+	// Attempt to dequeue relevant signals.
+	info, err := s.target.Sigtimedwait(s.Mask(), 0)
+	if err != nil {
+		// There must be no signal available.
+		return 0, syserror.ErrWouldBlock
+	}
+
+	// Copy out the signal info using the specified format.
+	var buf [128]byte
+	binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{
+		Signo:   uint32(info.Signo),
+		Errno:   info.Errno,
+		Code:    info.Code,
+		PID:     uint32(info.Pid()),
+		UID:     uint32(info.Uid()),
+		Status:  info.Status(),
+		Overrun: uint32(info.Overrun()),
+		Addr:    info.Addr(),
+	})
+	n, err := dst.CopyOut(ctx, buf[:])
+	return int64(n), err
+}
+
+// Readiness implements waiter.Waitable.Readiness.
+func (s *SignalOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
+	return mask & waiter.EventIn
+}
+
+// EventRegister implements waiter.Waitable.EventRegister.
+func (s *SignalOperations) EventRegister(entry *waiter.Entry, _ waiter.EventMask) {
+	// Register for the signal set; ignore the passed events.
+	s.target.SignalRegister(entry, waiter.EventMask(s.Mask()))
+}
+
+// EventUnregister implements waiter.Waitable.EventUnregister.
+func (s *SignalOperations) EventUnregister(entry *waiter.Entry) {
+	// Unregister the original entry.
+	s.target.SignalUnregister(entry)
+}
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index e91f82bb3..c82ef5486 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -35,6 +35,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/uniqueid"
 	"gvisor.dev/gvisor/pkg/sentry/usage"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/waiter"
 	"gvisor.dev/gvisor/third_party/gvsync"
 )
 
@@ -133,6 +134,13 @@ type Task struct {
 	// signalStack is exclusive to the task goroutine.
 	signalStack arch.SignalStack
 
+	// signalQueue is a set of registered waiters for signal-related events.
+	//
+	// signalQueue is protected by the signalMutex. Note that the task does
+	// not implement all queue methods, specifically the readiness checks.
+	// The task only broadcast a notification on signal delivery.
+	signalQueue waiter.Queue `state:"zerovalue"`
+
 	// If groupStopPending is true, the task should participate in a group
 	// stop in the interrupt path.
 	//
diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go
index 266959a07..39cd1340d 100644
--- a/pkg/sentry/kernel/task_signals.go
+++ b/pkg/sentry/kernel/task_signals.go
@@ -28,6 +28,7 @@ import (
 	ucspb "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/waiter"
 )
 
 // SignalAction is an internal signal action.
@@ -497,6 +498,9 @@ func (tg *ThreadGroup) applySignalSideEffectsLocked(sig linux.Signal) {
 //
 // Preconditions: The signal mutex must be locked.
 func (t *Task) canReceiveSignalLocked(sig linux.Signal) bool {
+	// Notify that the signal is queued.
+	t.signalQueue.Notify(waiter.EventMask(linux.MakeSignalSet(sig)))
+
 	// - Do not choose tasks that are blocking the signal.
 	if linux.SignalSetOf(sig)&t.signalMask != 0 {
 		return false
@@ -1108,3 +1112,17 @@ func (*runInterruptAfterSignalDeliveryStop) execute(t *Task) taskRunState {
 	t.tg.signalHandlers.mu.Unlock()
 	return t.deliverSignal(info, act)
 }
+
+// SignalRegister registers a waiter for pending signals.
+func (t *Task) SignalRegister(e *waiter.Entry, mask waiter.EventMask) {
+	t.tg.signalHandlers.mu.Lock()
+	t.signalQueue.EventRegister(e, mask)
+	t.tg.signalHandlers.mu.Unlock()
+}
+
+// SignalUnregister unregisters a waiter for pending signals.
+func (t *Task) SignalUnregister(e *waiter.Entry) {
+	t.tg.signalHandlers.mu.Lock()
+	t.signalQueue.EventUnregister(e)
+	t.tg.signalHandlers.mu.Unlock()
+}
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index d60cd62c7..ae6fc4025 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -172,9 +172,10 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) {
 		if parentPG := tg.parentPG(); parentPG == nil {
 			tg.createSession()
 		} else {
-			// Inherit the process group.
+			// Inherit the process group and terminal.
 			parentPG.incRefWithParent(parentPG)
 			tg.processGroup = parentPG
+			tg.tty = t.parent.tg.tty
 		}
 	}
 	tg.tasks.PushBack(t)
diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go
index 2a97e3e8e..0eef24bfb 100644
--- a/pkg/sentry/kernel/thread_group.go
+++ b/pkg/sentry/kernel/thread_group.go
@@ -19,10 +19,13 @@ import (
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"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/limits"
 	"gvisor.dev/gvisor/pkg/sentry/usage"
+	"gvisor.dev/gvisor/pkg/syserror"
 )
 
 // A ThreadGroup is a logical grouping of tasks that has widespread
@@ -245,6 +248,12 @@ type ThreadGroup struct {
 	//
 	// mounts is immutable.
 	mounts *fs.MountNamespace
+
+	// tty is the thread group's controlling terminal. If nil, there is no
+	// controlling terminal.
+	//
+	// tty is protected by the signal mutex.
+	tty *TTY
 }
 
 // newThreadGroup returns a new, empty thread group in PID namespace ns. The
@@ -324,6 +333,176 @@ func (tg *ThreadGroup) forEachChildThreadGroupLocked(fn func(*ThreadGroup)) {
 	}
 }
 
+// SetControllingTTY sets tty as the controlling terminal of tg.
+func (tg *ThreadGroup) SetControllingTTY(tty *TTY, arg int32) error {
+	tty.mu.Lock()
+	defer tty.mu.Unlock()
+
+	// We might be asked to set the controlling terminal of multiple
+	// processes, so we lock both the TaskSet and SignalHandlers.
+	tg.pidns.owner.mu.Lock()
+	defer tg.pidns.owner.mu.Unlock()
+	tg.signalHandlers.mu.Lock()
+	defer tg.signalHandlers.mu.Unlock()
+
+	// "The calling process must be a session leader and not have a
+	// controlling terminal already." - tty_ioctl(4)
+	if tg.processGroup.session.leader != tg || tg.tty != nil {
+		return syserror.EINVAL
+	}
+
+	// "If this terminal is already the controlling terminal of a different
+	// session group, then the ioctl fails with EPERM, unless the caller
+	// has the CAP_SYS_ADMIN capability and arg equals 1, in which case the
+	// terminal is stolen, and all processes that had it as controlling
+	// terminal lose it." - tty_ioctl(4)
+	if tty.tg != nil && tg.processGroup.session != tty.tg.processGroup.session {
+		if !auth.CredentialsFromContext(tg.leader).HasCapability(linux.CAP_SYS_ADMIN) || arg != 1 {
+			return syserror.EPERM
+		}
+		// Steal the TTY away. Unlike TIOCNOTTY, don't send signals.
+		for othertg := range tg.pidns.owner.Root.tgids {
+			// This won't deadlock by locking tg.signalHandlers
+			// because at this point:
+			// - We only lock signalHandlers if it's in the same
+			//   session as the tty's controlling thread group.
+			// - We know that the calling thread group is not in
+			//   the same session as the tty's controlling thread
+			//   group.
+			if othertg.processGroup.session == tty.tg.processGroup.session {
+				othertg.signalHandlers.mu.Lock()
+				othertg.tty = nil
+				othertg.signalHandlers.mu.Unlock()
+			}
+		}
+	}
+
+	// Set the controlling terminal and foreground process group.
+	tg.tty = tty
+	tg.processGroup.session.foreground = tg.processGroup
+	// Set this as the controlling process of the terminal.
+	tty.tg = tg
+
+	return nil
+}
+
+// ReleaseControllingTTY gives up tty as the controlling tty of tg.
+func (tg *ThreadGroup) ReleaseControllingTTY(tty *TTY) error {
+	tty.mu.Lock()
+	defer tty.mu.Unlock()
+
+	// We might be asked to set the controlling terminal of multiple
+	// processes, so we lock both the TaskSet and SignalHandlers.
+	tg.pidns.owner.mu.RLock()
+	defer tg.pidns.owner.mu.RUnlock()
+
+	// Just below, we may re-lock signalHandlers in order to send signals.
+	// Thus we can't defer Unlock here.
+	tg.signalHandlers.mu.Lock()
+
+	if tg.tty == nil || tg.tty != tty {
+		tg.signalHandlers.mu.Unlock()
+		return syserror.ENOTTY
+	}
+
+	// "If the process was session leader, then send SIGHUP and SIGCONT to
+	// the foreground process group and all processes in the current
+	// session lose their controlling terminal." - tty_ioctl(4)
+	// Remove tty as the controlling tty for each process in the session,
+	// then send them SIGHUP and SIGCONT.
+
+	// If we're not the session leader, we don't have to do much.
+	if tty.tg != tg {
+		tg.tty = nil
+		tg.signalHandlers.mu.Unlock()
+		return nil
+	}
+
+	tg.signalHandlers.mu.Unlock()
+
+	// We're the session leader. SIGHUP and SIGCONT the foreground process
+	// group and remove all controlling terminals in the session.
+	var lastErr error
+	for othertg := range tg.pidns.owner.Root.tgids {
+		if othertg.processGroup.session == tg.processGroup.session {
+			othertg.signalHandlers.mu.Lock()
+			othertg.tty = nil
+			if othertg.processGroup == tg.processGroup.session.foreground {
+				if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGHUP)}, true /* group */); err != nil {
+					lastErr = err
+				}
+				if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGCONT)}, true /* group */); err != nil {
+					lastErr = err
+				}
+			}
+			othertg.signalHandlers.mu.Unlock()
+		}
+	}
+
+	return lastErr
+}
+
+// ForegroundProcessGroup returns the process group ID of the foreground
+// process group.
+func (tg *ThreadGroup) ForegroundProcessGroup(tty *TTY) (int32, error) {
+	tty.mu.Lock()
+	defer tty.mu.Unlock()
+
+	tg.pidns.owner.mu.Lock()
+	defer tg.pidns.owner.mu.Unlock()
+	tg.signalHandlers.mu.Lock()
+	defer tg.signalHandlers.mu.Unlock()
+
+	// "When fd does not refer to the controlling terminal of the calling
+	// process, -1 is returned" - tcgetpgrp(3)
+	if tg.tty != tty {
+		return -1, syserror.ENOTTY
+	}
+
+	return int32(tg.processGroup.session.foreground.id), nil
+}
+
+// SetForegroundProcessGroup sets the foreground process group of tty to pgid.
+func (tg *ThreadGroup) SetForegroundProcessGroup(tty *TTY, pgid ProcessGroupID) (int32, error) {
+	tty.mu.Lock()
+	defer tty.mu.Unlock()
+
+	tg.pidns.owner.mu.Lock()
+	defer tg.pidns.owner.mu.Unlock()
+	tg.signalHandlers.mu.Lock()
+	defer tg.signalHandlers.mu.Unlock()
+
+	// TODO(b/129283598): "If tcsetpgrp() is called by a member of a
+	// background process group in its session, and the calling process is
+	// not blocking or ignoring SIGTTOU, a SIGTTOU signal is sent to all
+	// members of this background process group."
+
+	// tty must be the controlling terminal.
+	if tg.tty != tty {
+		return -1, syserror.ENOTTY
+	}
+
+	// pgid must be positive.
+	if pgid < 0 {
+		return -1, syserror.EINVAL
+	}
+
+	// pg must not be empty. Empty process groups are removed from their
+	// pid namespaces.
+	pg, ok := tg.pidns.processGroups[pgid]
+	if !ok {
+		return -1, syserror.ESRCH
+	}
+
+	// pg must be part of this process's session.
+	if tg.processGroup.session != pg.session {
+		return -1, syserror.EPERM
+	}
+
+	tg.processGroup.session.foreground.id = pgid
+	return 0, nil
+}
+
 // itimerRealListener implements ktime.Listener for ITIMER_REAL expirations.
 //
 // +stateify savable
diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go
new file mode 100644
index 000000000..34f84487a
--- /dev/null
+++ b/pkg/sentry/kernel/tty.go
@@ -0,0 +1,28 @@
+// 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 kernel
+
+import "sync"
+
+// TTY defines the relationship between a thread group and its controlling
+// terminal.
+//
+// +stateify savable
+type TTY struct {
+	mu sync.Mutex `state:"nosave"`
+
+	// tg is protected by mu.
+	tg *ThreadGroup
+}