1 files changed, 78 insertions, 16 deletions
diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go
index 2982450f8..1827666c5 100644
--- a/pkg/tcpip/network/fragmentation/fragmentation.go
+++ b/pkg/tcpip/network/fragmentation/fragmentation.go
@@ -17,28 +17,58 @@
 package fragmentation
 
 import (
+	"errors"
 	"fmt"
 	"log"
 	"time"
 
 	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 )
 
-// DefaultReassembleTimeout is based on the linux stack: net.ipv4.ipfrag_time.
-const DefaultReassembleTimeout = 30 * time.Second
+const (
+	// DefaultReassembleTimeout is based on the linux stack: net.ipv4.ipfrag_time.
+	DefaultReassembleTimeout = 30 * time.Second
 
-// HighFragThreshold is the threshold at which we start trimming old
-// fragmented packets. Linux uses a default value of 4 MB. See
-// net.ipv4.ipfrag_high_thresh for more information.
-const HighFragThreshold = 4 << 20 // 4MB
+	// HighFragThreshold is the threshold at which we start trimming old
+	// fragmented packets. Linux uses a default value of 4 MB. See
+	// net.ipv4.ipfrag_high_thresh for more information.
+	HighFragThreshold = 4 << 20 // 4MB
 
-// LowFragThreshold is the threshold we reach to when we start dropping
-// older fragmented packets. It's important that we keep enough room for newer
-// packets to be re-assembled. Hence, this needs to be lower than
-// HighFragThreshold enough. Linux uses a default value of 3 MB. See
-// net.ipv4.ipfrag_low_thresh for more information.
-const LowFragThreshold = 3 << 20 // 3MB
+	// LowFragThreshold is the threshold we reach to when we start dropping
+	// older fragmented packets. It's important that we keep enough room for newer
+	// packets to be re-assembled. Hence, this needs to be lower than
+	// HighFragThreshold enough. Linux uses a default value of 3 MB. See
+	// net.ipv4.ipfrag_low_thresh for more information.
+	LowFragThreshold = 3 << 20 // 3MB
+
+	// minBlockSize is the minimum block size for fragments.
+	minBlockSize = 1
+)
+
+var (
+	// ErrInvalidArgs indicates to the caller that that an invalid argument was
+	// provided.
+	ErrInvalidArgs = errors.New("invalid args")
+)
+
+// FragmentID is the identifier for a fragment.
+type FragmentID struct {
+	// Source is the source address of the fragment.
+	Source tcpip.Address
+
+	// Destination is the destination address of the fragment.
+	Destination tcpip.Address
+
+	// ID is the identification value of the fragment.
+	//
+	// This is a uint32 because IPv6 uses a 32-bit identification value.
+	ID uint32
+
+	// The protocol for the packet.
+	Protocol uint8
+}
 
 // Fragmentation is the main structure that other modules
 // of the stack should use to implement IP Fragmentation.
@@ -46,14 +76,17 @@ type Fragmentation struct {
 	mu           sync.Mutex
 	highLimit    int
 	lowLimit     int
-	reassemblers map[uint32]*reassembler
+	reassemblers map[FragmentID]*reassembler
 	rList        reassemblerList
 	size         int
 	timeout      time.Duration
+	blockSize    uint16
 }
 
 // NewFragmentation creates a new Fragmentation.
 //
+// blockSize specifies the fragment block size, in bytes.
+//
 // highMemoryLimit specifies the limit on the memory consumed
 // by the fragments stored by Fragmentation (overhead of internal data-structures
 // is not accounted). Fragments are dropped when the limit is reached.
@@ -64,7 +97,7 @@ 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(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation {
+func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation {
 	if lowMemoryLimit >= highMemoryLimit {
 		lowMemoryLimit = highMemoryLimit
 	}
@@ -73,17 +106,46 @@ func NewFragmentation(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout t
 		lowMemoryLimit = 0
 	}
 
+	if blockSize < minBlockSize {
+		blockSize = minBlockSize
+	}
+
 	return &Fragmentation{
-		reassemblers: make(map[uint32]*reassembler),
+		reassemblers: make(map[FragmentID]*reassembler),
 		highLimit:    highMemoryLimit,
 		lowLimit:     lowMemoryLimit,
 		timeout:      reassemblingTimeout,
+		blockSize:    blockSize,
 	}
 }
 
 // Process processes an incoming fragment belonging to an ID and returns a
 // complete packet when all the packets belonging to that ID have been received.
-func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) {
+//
+// [first, last] is the range of the fragment bytes.
+//
+// first must be a multiple of the block size f is configured with. The size
+// of the fragment data must be a multiple of the block size, unless there are
+// no fragments following this fragment (more set to false).
+func (f *Fragmentation) Process(id FragmentID, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) {
+	if first > last {
+		return buffer.VectorisedView{}, false, fmt.Errorf("first=%d is greater than last=%d: %w", first, last, ErrInvalidArgs)
+	}
+
+	if first%f.blockSize != 0 {
+		return buffer.VectorisedView{}, false, fmt.Errorf("first=%d is not a multiple of block size=%d: %w", first, f.blockSize, ErrInvalidArgs)
+	}
+
+	fragmentSize := last - first + 1
+	if more && fragmentSize%f.blockSize != 0 {
+		return buffer.VectorisedView{}, false, fmt.Errorf("fragment size=%d bytes is not a multiple of block size=%d on non-final fragment: %w", fragmentSize, f.blockSize, ErrInvalidArgs)
+	}
+
+	if l := vv.Size(); l < int(fragmentSize) {
+		return buffer.VectorisedView{}, false, fmt.Errorf("got fragment size=%d bytes less than the expected fragment size=%d bytes (first=%d last=%d): %w", l, fragmentSize, first, last, ErrInvalidArgs)
+	}
+	vv.CapLength(int(fragmentSize))
+
 	f.mu.Lock()
 	r, ok := f.reassemblers[id]
 	if ok && r.tooOld(f.timeout) {