Merge release-20200921.0-63-gf15182243 (automated)

4 files changed, 46 insertions, 18 deletions

diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go
index 6a4843f92..e1909fab0 100644
--- a/pkg/tcpip/network/fragmentation/fragmentation.go
+++ b/pkg/tcpip/network/fragmentation/fragmentation.go
@@ -81,6 +81,8 @@ type Fragmentation struct {
 	size         int
 	timeout      time.Duration
 	blockSize    uint16
+	clock        tcpip.Clock
+	releaseJob   *tcpip.Job
 }
 
 // NewFragmentation creates a new Fragmentation.
@@ -97,7 +99,7 @@ type Fragmentation struct {
 // reassemblingTimeout specifies the maximum time allowed to reassemble a packet.
 // Fragments are lazily evicted only when a new a packet with an
 // already existing fragmentation-id arrives after the timeout.
-func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation {
+func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration, clock tcpip.Clock) *Fragmentation {
 	if lowMemoryLimit >= highMemoryLimit {
 		lowMemoryLimit = highMemoryLimit
 	}
@@ -110,13 +112,17 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea
 		blockSize = minBlockSize
 	}
 
-	return &Fragmentation{
+	f := &Fragmentation{
 		reassemblers: make(map[FragmentID]*reassembler),
 		highLimit:    highMemoryLimit,
 		lowLimit:     lowMemoryLimit,
 		timeout:      reassemblingTimeout,
 		blockSize:    blockSize,
+		clock:        clock,
 	}
+	f.releaseJob = tcpip.NewJob(f.clock, &f.mu, f.releaseReassemblersLocked)
+
+	return f
 }
 
 // Process processes an incoming fragment belonging to an ID and returns a
@@ -155,15 +161,17 @@ func (f *Fragmentation) Process(
 
 	f.mu.Lock()
 	r, ok := f.reassemblers[id]
-	if ok && r.tooOld(f.timeout) {
-		// This is very likely to be an id-collision or someone performing a slow-rate attack.
-		f.release(r)
-		ok = false
-	}
 	if !ok {
-		r = newReassembler(id)
+		r = newReassembler(id, f.clock)
 		f.reassemblers[id] = r
+		wasEmpty := f.rList.Empty()
 		f.rList.PushFront(r)
+		if wasEmpty {
+			// If we have just pushed a first reassembler into an empty list, we
+			// should kickstart the release job. The release job will keep
+			// rescheduling itself until the list becomes empty.
+			f.releaseReassemblersLocked()
+		}
 	}
 	f.mu.Unlock()
 
@@ -211,3 +219,27 @@ func (f *Fragmentation) release(r *reassembler) {
 		f.size = 0
 	}
 }
+
+// releaseReassemblersLocked releases already-expired reassemblers, then
+// schedules the job to call back itself for the remaining reassemblers if
+// any. This function must be called with f.mu locked.
+func (f *Fragmentation) releaseReassemblersLocked() {
+	now := f.clock.NowMonotonic()
+	for {
+		// The reassembler at the end of the list is the oldest.
+		r := f.rList.Back()
+		if r == nil {
+			// The list is empty.
+			break
+		}
+		elapsed := time.Duration(now-r.creationTime) * time.Nanosecond
+		if f.timeout > elapsed {
+			// If the oldest reassembler has not expired, schedule the release
+			// job so that this function is called back when it has expired.
+			f.releaseJob.Schedule(f.timeout - elapsed)
+			break
+		}
+		// If the oldest reassembler has already expired, release it.
+		f.release(r)
+	}
+}
diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go
index f044867dc..9bb051a30 100644
--- a/pkg/tcpip/network/fragmentation/reassembler.go
+++ b/pkg/tcpip/network/fragmentation/reassembler.go
@@ -18,9 +18,9 @@ import (
 	"container/heap"
 	"fmt"
 	"math"
-	"time"
 
 	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 )
 
@@ -40,15 +40,15 @@ type reassembler struct {
 	deleted      int
 	heap         fragHeap
 	done         bool
-	creationTime time.Time
+	creationTime int64
 }
 
-func newReassembler(id FragmentID) *reassembler {
+func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler {
 	r := &reassembler{
 		id:           id,
 		holes:        make([]hole, 0, 16),
 		heap:         make(fragHeap, 0, 8),
-		creationTime: time.Now(),
+		creationTime: clock.NowMonotonic(),
 	}
 	r.holes = append(r.holes, hole{
 		first:   0,
@@ -116,10 +116,6 @@ func (r *reassembler) process(first, last uint16, more bool, proto uint8, vv buf
 	return res, r.proto, true, consumed, nil
 }
 
-func (r *reassembler) tooOld(timeout time.Duration) bool {
-	return time.Now().Sub(r.creationTime) > timeout
-}
-
 func (r *reassembler) checkDoneOrMark() bool {
 	r.mu.Lock()
 	prev := r.done
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 7adf0fac3..cf0a76274 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -804,6 +804,6 @@ func NewProtocol(s *stack.Stack) stack.NetworkProtocol {
 		ids:           ids,
 		hashIV:        hashIV,
 		defaultTTL:    DefaultTTL,
-		fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout),
+		fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout, s.Clock()),
 	}
 }
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index d1ad7acb7..990e67210 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -1306,7 +1306,7 @@ func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory {
 	return func(s *stack.Stack) stack.NetworkProtocol {
 		p := &protocol{
 			stack:         s,
-			fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout),
+			fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout, s.Clock()),
 
 			ndpDisp:              opts.NDPDisp,
 			ndpConfigs:           opts.NDPConfigs,