Merge branch 'master' into ip-forwarding

- Merges aleksej-paschenko's with HEAD
- Adds vfs2 support for ip_forward

1 files changed, 38 insertions, 101 deletions

diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go
index 1a1b38f83..297e8f28f 100644
--- a/pkg/sentry/kernel/pipe/pipe.go
+++ b/pkg/sentry/kernel/pipe/pipe.go
@@ -17,12 +17,13 @@ package pipe
 
 import (
 	"fmt"
-	"sync"
 	"sync/atomic"
 	"syscall"
 
-	"gvisor.dev/gvisor/pkg/sentry/context"
+	"gvisor.dev/gvisor/pkg/buffer"
+	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
@@ -70,10 +71,10 @@ type Pipe struct {
 	// mu protects all pipe internal state below.
 	mu sync.Mutex `state:"nosave"`
 
-	// data is the buffer queue of pipe contents.
+	// view is the underlying set of buffers.
 	//
 	// This is protected by mu.
-	data bufferList
+	view buffer.View
 
 	// max is the maximum size of the pipe in bytes. When this max has been
 	// reached, writers will get EWOULDBLOCK.
@@ -81,11 +82,6 @@ type Pipe struct {
 	// This is protected by mu.
 	max int64
 
-	// size is the current size of the pipe in bytes.
-	//
-	// This is protected by mu.
-	size int64
-
 	// hadWriter indicates if this pipe ever had a writer. Note that this
 	// does not necessarily indicate there is *currently* a writer, just
 	// that there has been a writer at some point since the pipe was
@@ -156,7 +152,7 @@ func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs.
 	d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, ms, sattr), fmt.Sprintf("pipe:[%d]", ino))
 	// The p.Open calls below will each take a reference on the Dirent. We
 	// must drop the one we already have.
-	defer d.DecRef()
+	defer d.DecRef(ctx)
 	return p.Open(ctx, d, fs.FileFlags{Read: true}), p.Open(ctx, d, fs.FileFlags{Write: true})
 }
 
@@ -196,7 +192,7 @@ type readOps struct {
 	limit func(int64)
 
 	// read performs the actual read operation.
-	read func(*buffer) (int64, error)
+	read func(*buffer.View) (int64, error)
 }
 
 // read reads data from the pipe into dst and returns the number of bytes
@@ -211,82 +207,27 @@ func (p *Pipe) read(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() {
-			// There are no writers, return EOF.
-			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 ops.left() > 0 {
-		// Pop the first buffer.
-		first := p.data.Front()
-		if first == nil {
-			break
-		}
-
-		// Copy user data.
-		n, err := ops.read(first)
-		done += int64(n)
-		p.size -= n
-
-		// Empty buffer?
-		if first.Empty() {
-			// Push to the free list.
-			p.data.Remove(first)
-			bufferPool.Put(first)
-		}
-
-		// Handle errors.
-		if err != nil {
-			return done, err
-		}
-	}
-
-	return done, nil
+	return p.readLocked(ctx, ops)
 }
 
-// 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()
-
+func (p *Pipe) readLocked(ctx context.Context, ops readOps) (int64, error) {
 	// Is the pipe empty?
-	if p.size == 0 {
+	if p.view.Size() == 0 {
 		if !p.HasWriters() {
-			// See above.
+			// There are no writers, return EOF.
 			return 0, nil
 		}
 		return 0, syserror.ErrWouldBlock
 	}
 
 	// Limit how much we consume.
-	if ops.left() > p.size {
-		ops.limit(p.size)
+	if ops.left() > p.view.Size() {
+		ops.limit(p.view.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
+	// Copy user data; the read op is responsible for trimming.
+	done, err := ops.read(&p.view)
+	return done, err
 }
 
 type writeOps struct {
@@ -297,7 +238,7 @@ type writeOps struct {
 	limit func(int64)
 
 	// write should write to the provided buffer.
-	write func(*buffer) (int64, error)
+	write func(*buffer.View) (int64, error)
 }
 
 // write writes data from sv into the pipe and returns the number of bytes
@@ -308,7 +249,10 @@ type writeOps struct {
 func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) {
 	p.mu.Lock()
 	defer p.mu.Unlock()
+	return p.writeLocked(ctx, ops)
+}
 
+func (p *Pipe) writeLocked(ctx context.Context, ops writeOps) (int64, error) {
 	// Can't write to a pipe with no readers.
 	if !p.HasReaders() {
 		return 0, syscall.EPIPE
@@ -317,35 +261,28 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) {
 	// POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be
 	// atomic, but requires no atomicity for writes larger than this.
 	wanted := ops.left()
-	if avail := p.max - p.size; wanted > avail {
+	avail := p.max - p.view.Size()
+	if wanted > avail {
 		if wanted <= p.atomicIOBytes {
 			return 0, syserror.ErrWouldBlock
 		}
 		ops.limit(avail)
 	}
 
-	done := int64(0)
-	for ops.left() > 0 {
-		// Need a new buffer?
-		last := p.data.Back()
-		if last == nil || last.Full() {
-			// Add a new buffer to the data list.
-			last = newBuffer()
-			p.data.PushBack(last)
-		}
-
-		// Copy user data.
-		n, err := ops.write(last)
-		done += int64(n)
-		p.size += n
+	// Copy user data.
+	done, err := ops.write(&p.view)
+	if err != nil {
+		return done, err
+	}
 
-		// Handle errors.
-		if err != nil {
-			return done, err
-		}
+	if done < avail {
+		// Non-failure, but short write.
+		return done, nil
 	}
-	if wanted > done {
-		// Partial write due to full pipe.
+	if done < wanted {
+		// Partial write due to full pipe. Note that this could also be
+		// the short write case above, we would expect a second call
+		// and the write to return zero bytes in this case.
 		return done, syserror.ErrWouldBlock
 	}
 
@@ -396,7 +333,7 @@ func (p *Pipe) HasWriters() bool {
 // Precondition: mu must be held.
 func (p *Pipe) rReadinessLocked() waiter.EventMask {
 	ready := waiter.EventMask(0)
-	if p.HasReaders() && p.data.Front() != nil {
+	if p.HasReaders() && p.view.Size() != 0 {
 		ready |= waiter.EventIn
 	}
 	if !p.HasWriters() && p.hadWriter {
@@ -422,7 +359,7 @@ func (p *Pipe) rReadiness() waiter.EventMask {
 // Precondition: mu must be held.
 func (p *Pipe) wReadinessLocked() waiter.EventMask {
 	ready := waiter.EventMask(0)
-	if p.HasWriters() && p.size < p.max {
+	if p.HasWriters() && p.view.Size() < p.max {
 		ready |= waiter.EventOut
 	}
 	if !p.HasReaders() {
@@ -451,7 +388,7 @@ func (p *Pipe) rwReadiness() waiter.EventMask {
 func (p *Pipe) queued() int64 {
 	p.mu.Lock()
 	defer p.mu.Unlock()
-	return p.size
+	return p.view.Size()
 }
 
 // FifoSize implements fs.FifoSizer.FifoSize.
@@ -474,7 +411,7 @@ func (p *Pipe) SetFifoSize(size int64) (int64, error) {
 	}
 	p.mu.Lock()
 	defer p.mu.Unlock()
-	if size < p.size {
+	if size < p.view.Size() {
 		return 0, syserror.EBUSY
 	}
 	p.max = size