Expose basic coverage information to userspace through kcov interface.

In Linux, a kernel configuration is set that compiles the kernel with a
custom function that is called at the beginning of every basic block, which
updates the memory-mapped coverage information. The Go coverage tool does not
allow us to inject arbitrary instructions into basic blocks, but it does
provide data that we can convert to a kcov-like format and transfer them to
userspace through a memory mapping.

Note that this is not a strict implementation of kcov, which is especially
tricky to do because we do not have the same coverage tools available in Go
that that are available for the actual Linux kernel. In Linux, a kernel
configuration is set that compiles the kernel with a custom function that is
called at the beginning of every basic block to write program counters to the
kcov memory mapping. In Go, however, coverage tools only give us a count of
basic blocks as they are executed. Every time we return to userspace, we
collect the coverage information and write out PCs for each block that was
executed, providing userspace with the illusion that the kcov data is always
up to date. For convenience, we also generate a unique synthetic PC for each
block instead of using actual PCs. Finally, we do not provide thread-specific
coverage data (each kcov instance only contains PCs executed by the thread
owning it); instead, we will supply data for any file specified by --
instrumentation_filter.

Also, fix issue in nogo that was causing pkg/coverage:coverage_nogo
compilation to fail.

PiperOrigin-RevId: 328426526

6 files changed, 373 insertions, 1 deletions

diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index 5416a310d..d1ecceba3 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -91,6 +91,8 @@ go_library(
         "fd_table_unsafe.go",
         "fs_context.go",
         "ipc_namespace.go",
+        "kcov.go",
+        "kcov_unsafe.go",
         "kernel.go",
         "kernel_opts.go",
         "kernel_state.go",
@@ -157,6 +159,7 @@ go_library(
         "//pkg/bits",
         "//pkg/bpf",
         "//pkg/context",
+        "//pkg/coverage",
         "//pkg/cpuid",
         "//pkg/eventchannel",
         "//pkg/fspath",
diff --git a/pkg/sentry/kernel/kcov.go b/pkg/sentry/kernel/kcov.go
new file mode 100644
index 000000000..aad63aa99
--- /dev/null
+++ b/pkg/sentry/kernel/kcov.go
@@ -0,0 +1,321 @@
+// 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 kernel
+
+import (
+	"fmt"
+	"io"
+	"sync"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/coverage"
+	"gvisor.dev/gvisor/pkg/safemem"
+	"gvisor.dev/gvisor/pkg/sentry/memmap"
+	"gvisor.dev/gvisor/pkg/sentry/mm"
+	"gvisor.dev/gvisor/pkg/sentry/pgalloc"
+	"gvisor.dev/gvisor/pkg/sentry/usage"
+	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
+)
+
+// kcovAreaSizeMax is the maximum number of uint64 entries allowed in the kcov
+// area. On Linux, the maximum is INT_MAX / 8.
+const kcovAreaSizeMax = 10 * 1024 * 1024
+
+// Kcov provides kernel coverage data to userspace through a memory-mapped
+// region, as kcov does in Linux.
+//
+// To give the illusion that the data is always up to date, we update the shared
+// memory every time before we return to userspace.
+type Kcov struct {
+	// mfp provides application memory. It is immutable after creation.
+	mfp pgalloc.MemoryFileProvider
+
+	// mu protects all of the fields below.
+	mu sync.RWMutex
+
+	// mode is the current kcov mode.
+	mode uint8
+
+	// size is the size of the mapping through which the kernel conveys coverage
+	// information to userspace.
+	size uint64
+
+	// owningTask is the task that currently owns coverage data on the system. The
+	// interface for kcov essentially requires that coverage is only going to a
+	// single task. Note that kcov should only generate coverage data for the
+	// owning task, but we currently generate global coverage.
+	owningTask *Task
+
+	// count is a locally cached version of the first uint64 in the kcov data,
+	// which is the number of subsequent entries representing PCs.
+	//
+	// It is used with kcovInode.countBlock(), to copy in/out the first element of
+	// the actual data in an efficient manner, avoid boilerplate, and prevent
+	// accidental garbage escapes by the temporary counts.
+	count uint64
+
+	mappable *mm.SpecialMappable
+}
+
+// NewKcov creates and returns a Kcov instance.
+func (k *Kernel) NewKcov() *Kcov {
+	return &Kcov{
+		mfp: k,
+	}
+}
+
+var coveragePool = sync.Pool{
+	New: func() interface{} {
+		return make([]byte, 0)
+	},
+}
+
+// TaskWork implements TaskWorker.TaskWork.
+func (kcov *Kcov) TaskWork(t *Task) {
+	kcov.mu.Lock()
+	defer kcov.mu.Unlock()
+
+	rw := &kcovReadWriter{
+		mf: kcov.mfp.MemoryFile(),
+		fr: kcov.mappable.FileRange(),
+	}
+
+	// Read in the PC count.
+	if _, err := safemem.ReadFullToBlocks(rw, kcov.countBlock()); err != nil {
+		panic(fmt.Sprintf("Internal error reading count from kcov area: %v", err))
+	}
+
+	rw.off = 8 * (1 + kcov.count)
+	n := coverage.ConsumeCoverageData(&kcovIOWriter{rw})
+
+	// Update the pc count, based on the number of entries written. Note that if
+	// we reached the end of the kcov area, we may not have written everything in
+	// output.
+	kcov.count += uint64(n / 8)
+	rw.off = 0
+	if _, err := safemem.WriteFullFromBlocks(rw, kcov.countBlock()); err != nil {
+		panic(fmt.Sprintf("Internal error writing count to kcov area: %v", err))
+	}
+
+	// Re-register for future work.
+	t.RegisterWork(kcov)
+}
+
+// InitTrace performs the KCOV_INIT_TRACE ioctl.
+func (kcov *Kcov) InitTrace(size uint64) error {
+	kcov.mu.Lock()
+	defer kcov.mu.Unlock()
+
+	if kcov.mode != linux.KCOV_MODE_DISABLED {
+		return syserror.EBUSY
+	}
+
+	// To simplify all the logic around mapping, we require that the length of the
+	// shared region is a multiple of the system page size.
+	if (8*size)&(usermem.PageSize-1) != 0 {
+		return syserror.EINVAL
+	}
+
+	// We need space for at least two uint64s to hold current position and a
+	// single PC.
+	if size < 2 || size > kcovAreaSizeMax {
+		return syserror.EINVAL
+	}
+
+	kcov.size = size
+	kcov.mode = linux.KCOV_MODE_INIT
+	return nil
+}
+
+// EnableTrace performs the KCOV_ENABLE_TRACE ioctl.
+func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error {
+	t := TaskFromContext(ctx)
+	if t == nil {
+		panic("kcovInode.EnableTrace() cannot be used outside of a task goroutine")
+	}
+
+	kcov.mu.Lock()
+	defer kcov.mu.Unlock()
+
+	// KCOV_ENABLE must be preceded by KCOV_INIT_TRACE and an mmap call.
+	if kcov.mode != linux.KCOV_MODE_INIT || kcov.mappable == nil {
+		return syserror.EINVAL
+	}
+
+	switch traceMode {
+	case linux.KCOV_TRACE_PC:
+		kcov.mode = traceMode
+	case linux.KCOV_TRACE_CMP:
+		// We do not support KCOV_MODE_TRACE_CMP.
+		return syserror.ENOTSUP
+	default:
+		return syserror.EINVAL
+	}
+
+	if kcov.owningTask != nil && kcov.owningTask != t {
+		return syserror.EBUSY
+	}
+
+	kcov.owningTask = t
+	t.RegisterWork(kcov)
+
+	// Clear existing coverage data; the task expects to read only coverage data
+	// from the time it is activated.
+	coverage.ClearCoverageData()
+	return nil
+}
+
+// DisableTrace performs the KCOV_DISABLE_TRACE ioctl.
+func (kcov *Kcov) DisableTrace(ctx context.Context) error {
+	kcov.mu.Lock()
+	defer kcov.mu.Unlock()
+
+	t := TaskFromContext(ctx)
+	if t == nil {
+		panic("kcovInode.EnableTrace() cannot be used outside of a task goroutine")
+	}
+
+	if t != kcov.owningTask {
+		return syserror.EINVAL
+	}
+	kcov.owningTask = nil
+	kcov.mode = linux.KCOV_MODE_INIT
+	kcov.resetLocked()
+	return nil
+}
+
+// Reset is called when the owning task exits.
+func (kcov *Kcov) Reset() {
+	kcov.mu.Lock()
+	kcov.resetLocked()
+	kcov.mu.Unlock()
+}
+
+// The kcov instance is reset when the owning task exits or when tracing is
+// disabled.
+func (kcov *Kcov) resetLocked() {
+	kcov.owningTask = nil
+	if kcov.mappable != nil {
+		kcov.mappable = nil
+	}
+}
+
+// ConfigureMMap is called by the vfs.FileDescription for this kcov instance to
+// implement vfs.FileDescription.ConfigureMMap.
+func (kcov *Kcov) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error {
+	kcov.mu.Lock()
+	defer kcov.mu.Unlock()
+
+	if kcov.mode != linux.KCOV_MODE_INIT {
+		return syserror.EINVAL
+	}
+
+	if kcov.mappable == nil {
+		// Set up the kcov area.
+		fr, err := kcov.mfp.MemoryFile().Allocate(kcov.size*8, usage.Anonymous)
+		if err != nil {
+			return err
+		}
+
+		// Get the thread id for the mmap name.
+		t := TaskFromContext(ctx)
+		if t == nil {
+			panic("ThreadFromContext returned nil")
+		}
+		// For convenience, a special mappable is used here. Note that these mappings
+		// will look different under /proc/[pid]/maps than they do on Linux.
+		kcov.mappable = mm.NewSpecialMappable(fmt.Sprintf("[kcov:%d]", t.ThreadID()), kcov.mfp, fr)
+	}
+	opts.Mappable = kcov.mappable
+	opts.MappingIdentity = kcov.mappable
+	return nil
+}
+
+// kcovReadWriter implements safemem.Reader and safemem.Writer.
+type kcovReadWriter struct {
+	off uint64
+	mf  *pgalloc.MemoryFile
+	fr  memmap.FileRange
+}
+
+// ReadToBlocks implements safemem.Reader.ReadToBlocks.
+func (rw *kcovReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) {
+	if dsts.IsEmpty() {
+		return 0, nil
+	}
+
+	// Limit the read to the kcov range and check for overflow.
+	if rw.fr.Length() <= rw.off {
+		return 0, io.EOF
+	}
+	start := rw.fr.Start + rw.off
+	end := rw.fr.Start + rw.fr.Length()
+	if rend := start + dsts.NumBytes(); rend < end {
+		end = rend
+	}
+
+	// Get internal mappings.
+	bs, err := rw.mf.MapInternal(memmap.FileRange{start, end}, usermem.Read)
+	if err != nil {
+		return 0, err
+	}
+
+	// Copy from internal mappings.
+	n, err := safemem.CopySeq(dsts, bs)
+	rw.off += n
+	return n, err
+}
+
+// WriteFromBlocks implements safemem.Writer.WriteFromBlocks.
+func (rw *kcovReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) {
+	if srcs.IsEmpty() {
+		return 0, nil
+	}
+
+	// Limit the write to the kcov area and check for overflow.
+	if rw.fr.Length() <= rw.off {
+		return 0, io.EOF
+	}
+	start := rw.fr.Start + rw.off
+	end := rw.fr.Start + rw.fr.Length()
+	if wend := start + srcs.NumBytes(); wend < end {
+		end = wend
+	}
+
+	// Get internal mapping.
+	bs, err := rw.mf.MapInternal(memmap.FileRange{start, end}, usermem.Write)
+	if err != nil {
+		return 0, err
+	}
+
+	// Copy to internal mapping.
+	n, err := safemem.CopySeq(bs, srcs)
+	rw.off += n
+	return n, err
+}
+
+// kcovIOWriter implements io.Writer as a basic wrapper over kcovReadWriter.
+type kcovIOWriter struct {
+	rw *kcovReadWriter
+}
+
+// Write implements io.Writer.Write.
+func (w *kcovIOWriter) Write(p []byte) (int, error) {
+	bs := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(p))
+	n, err := safemem.WriteFullFromBlocks(w.rw, bs)
+	return int(n), err
+}
diff --git a/pkg/sentry/kernel/kcov_unsafe.go b/pkg/sentry/kernel/kcov_unsafe.go
new file mode 100644
index 000000000..6f64022eb
--- /dev/null
+++ b/pkg/sentry/kernel/kcov_unsafe.go
@@ -0,0 +1,28 @@
+// 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 kernel
+
+import (
+	"unsafe"
+
+	"gvisor.dev/gvisor/pkg/safemem"
+)
+
+// countBlock provides a safemem.BlockSeq for k.count.
+//
+// Like k.count, the block returned is protected by k.mu.
+func (k *Kcov) countBlock() safemem.BlockSeq {
+	return safemem.BlockSeqOf(safemem.BlockFromSafePointer(unsafe.Pointer(&k.count), int(unsafe.Sizeof(k.count))))
+}
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 2e0175e36..402aa1718 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -248,7 +248,7 @@ type Kernel struct {
 	// SpecialOpts contains special kernel options.
 	SpecialOpts
 
-	// VFS keeps the filesystem state used across the kernel.
+	// vfs keeps the filesystem state used across the kernel.
 	vfs vfs.VirtualFilesystem
 
 	// hostMount is the Mount used for file descriptors that were imported
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 5aee699e7..a436610c9 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -574,6 +574,11 @@ type Task struct {
 	//
 	// startTime is protected by mu.
 	startTime ktime.Time
+
+	// kcov is the kcov instance providing code coverage owned by this task.
+	//
+	// kcov is exclusive to the task goroutine.
+	kcov *Kcov
 }
 
 func (t *Task) savePtraceTracer() *Task {
@@ -903,3 +908,16 @@ func (t *Task) UID() uint32 {
 func (t *Task) GID() uint32 {
 	return uint32(t.Credentials().EffectiveKGID)
 }
+
+// SetKcov sets the kcov instance associated with t.
+func (t *Task) SetKcov(k *Kcov) {
+	t.kcov = k
+}
+
+// ResetKcov clears the kcov instance associated with t.
+func (t *Task) ResetKcov() {
+	if t.kcov != nil {
+		t.kcov.Reset()
+		t.kcov = nil
+	}
+}
diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go
index c165d6cb1..b76f7f503 100644
--- a/pkg/sentry/kernel/task_exit.go
+++ b/pkg/sentry/kernel/task_exit.go
@@ -239,6 +239,8 @@ func (*runExitMain) execute(t *Task) taskRunState {
 	t.traceExitEvent()
 	lastExiter := t.exitThreadGroup()
 
+	t.ResetKcov()
+
 	// If the task has a cleartid, and the thread group wasn't killed by a
 	// signal, handle that before releasing the MM.
 	if t.cleartid != 0 {