diff options
-rw-r--r-- | pkg/tcpip/header/parse/parse.go | 3 | ||||
-rw-r--r-- | pkg/tcpip/network/fragmentation/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/network/fragmentation/fragmentation.go | 71 | ||||
-rw-r--r-- | pkg/tcpip/network/fragmentation/fragmentation_test.go | 207 | ||||
-rw-r--r-- | pkg/tcpip/network/fragmentation/reassembler.go | 50 | ||||
-rw-r--r-- | pkg/tcpip/network/fragmentation/reassembler_test.go | 23 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/icmp.go | 15 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 41 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/icmp.go | 13 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ipv6.go | 27 |
10 files changed, 227 insertions, 224 deletions
diff --git a/pkg/tcpip/header/parse/parse.go b/pkg/tcpip/header/parse/parse.go index 5ca75c834..2042f214a 100644 --- a/pkg/tcpip/header/parse/parse.go +++ b/pkg/tcpip/header/parse/parse.go @@ -109,6 +109,9 @@ traverseExtensions: fragOffset = extHdr.FragmentOffset() fragMore = extHdr.More() } + rawPayload := it.AsRawHeader(true /* consume */) + extensionsSize = dataClone.Size() - rawPayload.Buf.Size() + break traverseExtensions case header.IPv6RawPayloadHeader: // We've found the payload after any extensions. diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index 47fb63290..d8e4a3b54 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -47,6 +47,7 @@ go_test( "//pkg/tcpip/buffer", "//pkg/tcpip/faketime", "//pkg/tcpip/network/testutil", + "//pkg/tcpip/stack", "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 936601287..c75ca7d71 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -71,16 +71,25 @@ type FragmentID struct { // Fragmentation is the main structure that other modules // of the stack should use to implement IP Fragmentation. type Fragmentation struct { - mu sync.Mutex - highLimit int - lowLimit int - reassemblers map[FragmentID]*reassembler - rList reassemblerList - size int - timeout time.Duration - blockSize uint16 - clock tcpip.Clock - releaseJob *tcpip.Job + mu sync.Mutex + highLimit int + lowLimit int + reassemblers map[FragmentID]*reassembler + rList reassemblerList + size int + timeout time.Duration + blockSize uint16 + clock tcpip.Clock + releaseJob *tcpip.Job + timeoutHandler TimeoutHandler +} + +// TimeoutHandler is consulted if a packet reassembly has timed out. +type TimeoutHandler interface { + // OnReassemblyTimeout will be called with the first fragment (or nil, if the + // first fragment has not been received) of a packet whose reassembly has + // timed out. + OnReassemblyTimeout(pkt *stack.PacketBuffer) } // NewFragmentation creates a new Fragmentation. @@ -97,7 +106,7 @@ 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, clock tcpip.Clock) *Fragmentation { +func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration, clock tcpip.Clock, timeoutHandler TimeoutHandler) *Fragmentation { if lowMemoryLimit >= highMemoryLimit { lowMemoryLimit = highMemoryLimit } @@ -111,12 +120,13 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea } f := &Fragmentation{ - reassemblers: make(map[FragmentID]*reassembler), - highLimit: highMemoryLimit, - lowLimit: lowMemoryLimit, - timeout: reassemblingTimeout, - blockSize: blockSize, - clock: clock, + reassemblers: make(map[FragmentID]*reassembler), + highLimit: highMemoryLimit, + lowLimit: lowMemoryLimit, + timeout: reassemblingTimeout, + blockSize: blockSize, + clock: clock, + timeoutHandler: timeoutHandler, } f.releaseJob = tcpip.NewJob(f.clock, &f.mu, f.releaseReassemblersLocked) @@ -136,16 +146,8 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea // proto is the protocol number marked in the fragment being processed. It has // to be given here outside of the FragmentID struct because IPv6 should not use // the protocol to identify a fragment. -// -// releaseCB is a callback that will run when the fragment reassembly of a -// packet is complete or cancelled. releaseCB take a a boolean argument which is -// true iff the reassembly is cancelled due to timeout. releaseCB should be -// passed only with the first fragment of a packet. If more than one releaseCB -// are passed for the same packet, only the first releaseCB will be saved for -// the packet and the succeeding ones will be dropped by running them -// immediately with a false argument. func (f *Fragmentation) Process( - id FragmentID, first, last uint16, more bool, proto uint8, vv buffer.VectorisedView, releaseCB func(bool)) ( + id FragmentID, first, last uint16, more bool, proto uint8, pkt *stack.PacketBuffer) ( buffer.VectorisedView, uint8, bool, error) { if first > last { return buffer.VectorisedView{}, 0, false, fmt.Errorf("first=%d is greater than last=%d: %w", first, last, ErrInvalidArgs) @@ -160,10 +162,9 @@ func (f *Fragmentation) Process( return buffer.VectorisedView{}, 0, 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{}, 0, 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) + if l := pkt.Data.Size(); l != int(fragmentSize) { + return buffer.VectorisedView{}, 0, false, fmt.Errorf("got fragment size=%d bytes not equal to 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] @@ -179,15 +180,9 @@ func (f *Fragmentation) Process( f.releaseReassemblersLocked() } } - if releaseCB != nil { - if !r.setCallback(releaseCB) { - // We got a duplicate callback. Release it immediately. - releaseCB(false /* timedOut */) - } - } f.mu.Unlock() - res, firstFragmentProto, done, consumed, err := r.process(first, last, more, proto, vv) + res, firstFragmentProto, done, consumed, err := r.process(first, last, more, proto, pkt) if err != nil { // We probably got an invalid sequence of fragments. Just // discard the reassembler and move on. @@ -231,7 +226,9 @@ func (f *Fragmentation) release(r *reassembler, timedOut bool) { f.size = 0 } - r.release(timedOut) // releaseCB may run. + if h := f.timeoutHandler; timedOut && h != nil { + h.OnReassemblyTimeout(r.pkt) + } } // releaseReassemblersLocked releases already-expired reassemblers, then diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index 5dcd10730..3a79688a8 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/faketime" "gvisor.dev/gvisor/pkg/tcpip/network/testutil" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) // reassembleTimeout is dummy timeout used for testing, where the clock never @@ -40,13 +41,19 @@ func vv(size int, pieces ...string) buffer.VectorisedView { return buffer.NewVectorisedView(size, views) } +func pkt(size int, pieces ...string) *stack.PacketBuffer { + return stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: vv(size, pieces...), + }) +} + type processInput struct { id FragmentID first uint16 last uint16 more bool proto uint8 - vv buffer.VectorisedView + pkt *stack.PacketBuffer } type processOutput struct { @@ -63,8 +70,8 @@ var processTestCases = []struct { { comment: "One ID", in: []processInput{ - {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, vv: vv(2, "01")}, - {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, vv: vv(2, "23")}, + {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, pkt: pkt(2, "01")}, + {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, pkt: pkt(2, "23")}, }, out: []processOutput{ {vv: buffer.VectorisedView{}, done: false}, @@ -74,8 +81,8 @@ var processTestCases = []struct { { comment: "Next Header protocol mismatch", in: []processInput{ - {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, proto: 6, vv: vv(2, "01")}, - {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, proto: 17, vv: vv(2, "23")}, + {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, proto: 6, pkt: pkt(2, "01")}, + {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, proto: 17, pkt: pkt(2, "23")}, }, out: []processOutput{ {vv: buffer.VectorisedView{}, done: false}, @@ -85,10 +92,10 @@ var processTestCases = []struct { { comment: "Two IDs", in: []processInput{ - {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, vv: vv(2, "01")}, - {id: FragmentID{ID: 1}, first: 0, last: 1, more: true, vv: vv(2, "ab")}, - {id: FragmentID{ID: 1}, first: 2, last: 3, more: false, vv: vv(2, "cd")}, - {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, vv: vv(2, "23")}, + {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, pkt: pkt(2, "01")}, + {id: FragmentID{ID: 1}, first: 0, last: 1, more: true, pkt: pkt(2, "ab")}, + {id: FragmentID{ID: 1}, first: 2, last: 3, more: false, pkt: pkt(2, "cd")}, + {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, pkt: pkt(2, "23")}, }, out: []processOutput{ {vv: buffer.VectorisedView{}, done: false}, @@ -102,17 +109,17 @@ var processTestCases = []struct { func TestFragmentationProcess(t *testing.T) { for _, c := range processTestCases { t.Run(c.comment, func(t *testing.T) { - f := NewFragmentation(minBlockSize, 1024, 512, reassembleTimeout, &faketime.NullClock{}) + f := NewFragmentation(minBlockSize, 1024, 512, reassembleTimeout, &faketime.NullClock{}, nil) firstFragmentProto := c.in[0].proto for i, in := range c.in { - vv, proto, done, err := f.Process(in.id, in.first, in.last, in.more, in.proto, in.vv, nil) + vv, proto, done, err := f.Process(in.id, in.first, in.last, in.more, in.proto, in.pkt) if err != nil { - t.Fatalf("f.Process(%+v, %d, %d, %t, %d, %X) failed: %s", - in.id, in.first, in.last, in.more, in.proto, in.vv.ToView(), err) + t.Fatalf("f.Process(%+v, %d, %d, %t, %d, %#v) failed: %s", + in.id, in.first, in.last, in.more, in.proto, in.pkt, err) } if !reflect.DeepEqual(vv, c.out[i].vv) { - t.Errorf("got Process(%+v, %d, %d, %t, %d, %X) = (%X, _, _, _), want = (%X, _, _, _)", - in.id, in.first, in.last, in.more, in.proto, in.vv.ToView(), vv.ToView(), c.out[i].vv.ToView()) + t.Errorf("got Process(%+v, %d, %d, %t, %d, %#v) = (%X, _, _, _), want = (%X, _, _, _)", + in.id, in.first, in.last, in.more, in.proto, in.pkt, vv.ToView(), c.out[i].vv.ToView()) } if done != c.out[i].done { t.Errorf("got Process(%+v, %d, %d, %t, %d, _) = (_, _, %t, _), want = (_, _, %t, _)", @@ -236,11 +243,11 @@ func TestReassemblingTimeout(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { clock := faketime.NewManualClock() - f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassemblyTimeout, clock) + f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassemblyTimeout, clock, nil) for _, event := range test.events { clock.Advance(event.clockAdvance) if frag := event.fragment; frag != nil { - _, _, done, err := f.Process(FragmentID{}, frag.first, frag.last, frag.more, protocol, vv(len(frag.data), frag.data), nil) + _, _, done, err := f.Process(FragmentID{}, frag.first, frag.last, frag.more, protocol, pkt(len(frag.data), frag.data)) if err != nil { t.Fatalf("%s: f.Process failed: %s", event.name, err) } @@ -257,17 +264,17 @@ func TestReassemblingTimeout(t *testing.T) { } func TestMemoryLimits(t *testing.T) { - f := NewFragmentation(minBlockSize, 3, 1, reassembleTimeout, &faketime.NullClock{}) + f := NewFragmentation(minBlockSize, 3, 1, reassembleTimeout, &faketime.NullClock{}, nil) // Send first fragment with id = 0. - f.Process(FragmentID{ID: 0}, 0, 0, true, 0xFF, vv(1, "0"), nil) + f.Process(FragmentID{ID: 0}, 0, 0, true, 0xFF, pkt(1, "0")) // Send first fragment with id = 1. - f.Process(FragmentID{ID: 1}, 0, 0, true, 0xFF, vv(1, "1"), nil) + f.Process(FragmentID{ID: 1}, 0, 0, true, 0xFF, pkt(1, "1")) // Send first fragment with id = 2. - f.Process(FragmentID{ID: 2}, 0, 0, true, 0xFF, vv(1, "2"), nil) + f.Process(FragmentID{ID: 2}, 0, 0, true, 0xFF, pkt(1, "2")) // Send first fragment with id = 3. This should caused id = 0 and id = 1 to be // evicted. - f.Process(FragmentID{ID: 3}, 0, 0, true, 0xFF, vv(1, "3"), nil) + f.Process(FragmentID{ID: 3}, 0, 0, true, 0xFF, pkt(1, "3")) if _, ok := f.reassemblers[FragmentID{ID: 0}]; ok { t.Errorf("Memory limits are not respected: id=0 has not been evicted.") @@ -281,11 +288,11 @@ func TestMemoryLimits(t *testing.T) { } func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { - f := NewFragmentation(minBlockSize, 1, 0, reassembleTimeout, &faketime.NullClock{}) + f := NewFragmentation(minBlockSize, 1, 0, reassembleTimeout, &faketime.NullClock{}, nil) // Send first fragment with id = 0. - f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0"), nil) + f.Process(FragmentID{}, 0, 0, true, 0xFF, pkt(1, "0")) // Send the same packet again. - f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0"), nil) + f.Process(FragmentID{}, 0, 0, true, 0xFF, pkt(1, "0")) got := f.size want := 1 @@ -327,6 +334,7 @@ func TestErrors(t *testing.T) { last: 3, more: true, data: "012", + err: ErrInvalidArgs, }, { name: "exact block size with more and too little data", @@ -376,8 +384,8 @@ func TestErrors(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, reassembleTimeout, &faketime.NullClock{}) - _, _, done, err := f.Process(FragmentID{}, test.first, test.last, test.more, 0, vv(len(test.data), test.data), nil) + f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, reassembleTimeout, &faketime.NullClock{}, nil) + _, _, done, err := f.Process(FragmentID{}, test.first, test.last, test.more, 0, pkt(len(test.data), test.data)) if !errors.Is(err, test.err) { t.Errorf("got Process(_, %d, %d, %t, _, %q) = (_, _, _, %v), want = (_, _, _, %v)", test.first, test.last, test.more, test.data, err, test.err) } @@ -498,57 +506,92 @@ func TestPacketFragmenter(t *testing.T) { } } -func TestReleaseCallback(t *testing.T) { +type testTimeoutHandler struct { + pkt *stack.PacketBuffer +} + +func (h *testTimeoutHandler) OnReassemblyTimeout(pkt *stack.PacketBuffer) { + h.pkt = pkt +} + +func TestTimeoutHandler(t *testing.T) { const ( proto = 99 ) - var result int - var callbackReasonIsTimeout bool - cb1 := func(timedOut bool) { result = 1; callbackReasonIsTimeout = timedOut } - cb2 := func(timedOut bool) { result = 2; callbackReasonIsTimeout = timedOut } + pk1 := pkt(1, "1") + pk2 := pkt(1, "2") + + type processParam struct { + first uint16 + last uint16 + more bool + pkt *stack.PacketBuffer + } tests := []struct { - name string - callbacks []func(bool) - timeout bool - wantResult int - wantCallbackReasonIsTimeout bool + name string + params []processParam + wantError bool + wantPkt *stack.PacketBuffer }{ { - name: "callback runs on release", - callbacks: []func(bool){cb1}, - timeout: false, - wantResult: 1, - wantCallbackReasonIsTimeout: false, - }, - { - name: "first callback is nil", - callbacks: []func(bool){nil, cb2}, - timeout: false, - wantResult: 2, - wantCallbackReasonIsTimeout: false, + name: "onTimeout runs", + params: []processParam{ + { + first: 0, + last: 0, + more: true, + pkt: pk1, + }, + }, + wantError: false, + wantPkt: pk1, }, { - name: "two callbacks - first one is set", - callbacks: []func(bool){cb1, cb2}, - timeout: false, - wantResult: 1, - wantCallbackReasonIsTimeout: false, + name: "no first fragment", + params: []processParam{ + { + first: 1, + last: 1, + more: true, + pkt: pk1, + }, + }, + wantError: false, + wantPkt: nil, }, { - name: "callback runs on timeout", - callbacks: []func(bool){cb1}, - timeout: true, - wantResult: 1, - wantCallbackReasonIsTimeout: true, + name: "second pkt is ignored", + params: []processParam{ + { + first: 0, + last: 0, + more: true, + pkt: pk1, + }, + { + first: 0, + last: 0, + more: true, + pkt: pk2, + }, + }, + wantError: false, + wantPkt: pk1, }, { - name: "no callbacks", - callbacks: []func(bool){nil}, - timeout: false, - wantResult: 0, - wantCallbackReasonIsTimeout: false, + name: "invalid args - first is greater than last", + params: []processParam{ + { + first: 1, + last: 0, + more: true, + pkt: pk1, + }, + }, + wantError: true, + wantPkt: nil, }, } @@ -556,29 +599,31 @@ func TestReleaseCallback(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - result = 0 - callbackReasonIsTimeout = false + handler := &testTimeoutHandler{pkt: nil} - f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassembleTimeout, &faketime.NullClock{}) + f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassembleTimeout, &faketime.NullClock{}, handler) - for i, cb := range test.callbacks { - _, _, _, err := f.Process(id, uint16(i), uint16(i), true, proto, vv(1, "0"), cb) - if err != nil { + for _, p := range test.params { + if _, _, _, err := f.Process(id, p.first, p.last, p.more, proto, p.pkt); err != nil && !test.wantError { t.Errorf("f.Process error = %s", err) } } - - r, ok := f.reassemblers[id] - if !ok { - t.Fatalf("Reassemberr not found") - } - f.release(r, test.timeout) - - if result != test.wantResult { - t.Errorf("got result = %d, want = %d", result, test.wantResult) + if !test.wantError { + r, ok := f.reassemblers[id] + if !ok { + t.Fatal("Reassembler not found") + } + f.release(r, true) } - if callbackReasonIsTimeout != test.wantCallbackReasonIsTimeout { - t.Errorf("got callbackReasonIsTimeout = %t, want = %t", callbackReasonIsTimeout, test.wantCallbackReasonIsTimeout) + switch { + case handler.pkt != nil && test.wantPkt == nil: + t.Errorf("got handler.pkt = not nil (pkt.Data = %x), want = nil", handler.pkt.Data.ToView()) + case handler.pkt == nil && test.wantPkt != nil: + t.Errorf("got handler.pkt = nil, want = not nil (pkt.Data = %x)", test.wantPkt.Data.ToView()) + case handler.pkt != nil && test.wantPkt != nil: + if diff := cmp.Diff(test.wantPkt.Data.ToView(), handler.pkt.Data.ToView()); diff != "" { + t.Errorf("pkt.Data mismatch (-want, +got):\n%s", diff) + } } }) } diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index c0cc0bde0..19f4920b3 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) type hole struct { @@ -41,7 +42,7 @@ type reassembler struct { heap fragHeap done bool creationTime int64 - callback func(bool) + pkt *stack.PacketBuffer } func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler { @@ -79,7 +80,7 @@ func (r *reassembler) updateHoles(first, last uint16, more bool) bool { return used } -func (r *reassembler) process(first, last uint16, more bool, proto uint8, vv buffer.VectorisedView) (buffer.VectorisedView, uint8, bool, int, error) { +func (r *reassembler) process(first, last uint16, more bool, proto uint8, pkt *stack.PacketBuffer) (buffer.VectorisedView, uint8, bool, int, error) { r.mu.Lock() defer r.mu.Unlock() consumed := 0 @@ -89,18 +90,20 @@ func (r *reassembler) process(first, last uint16, more bool, proto uint8, vv buf // was waiting on the mutex. We don't have to do anything in this case. return buffer.VectorisedView{}, 0, false, consumed, nil } - // For IPv6, it is possible to have different Protocol values between - // fragments of a packet (because, unlike IPv4, the Protocol is not used to - // identify a fragment). In this case, only the Protocol of the first - // fragment must be used as per RFC 8200 Section 4.5. - // - // TODO(gvisor.dev/issue/3648): The entire first IP header should be recorded - // here (instead of just the protocol) because most IP options should be - // derived from the first fragment. - if first == 0 { - r.proto = proto - } if r.updateHoles(first, last, more) { + // For IPv6, it is possible to have different Protocol values between + // fragments of a packet (because, unlike IPv4, the Protocol is not used to + // identify a fragment). In this case, only the Protocol of the first + // fragment must be used as per RFC 8200 Section 4.5. + // + // TODO(gvisor.dev/issue/3648): During reassembly of an IPv6 packet, IP + // options received in the first fragment should be used - and they should + // override options from following fragments. + if first == 0 { + r.pkt = pkt + r.proto = proto + } + vv := pkt.Data // We store the incoming packet only if it filled some holes. heap.Push(&r.heap, fragment{offset: first, vv: vv.Clone(nil)}) consumed = vv.Size() @@ -124,24 +127,3 @@ func (r *reassembler) checkDoneOrMark() bool { r.mu.Unlock() return prev } - -func (r *reassembler) setCallback(c func(bool)) bool { - r.mu.Lock() - defer r.mu.Unlock() - if r.callback != nil { - return false - } - r.callback = c - return true -} - -func (r *reassembler) release(timedOut bool) { - r.mu.Lock() - callback := r.callback - r.callback = nil - r.mu.Unlock() - - if callback != nil { - callback(timedOut) - } -} diff --git a/pkg/tcpip/network/fragmentation/reassembler_test.go b/pkg/tcpip/network/fragmentation/reassembler_test.go index fa2a70dc8..a0a04a027 100644 --- a/pkg/tcpip/network/fragmentation/reassembler_test.go +++ b/pkg/tcpip/network/fragmentation/reassembler_test.go @@ -105,26 +105,3 @@ func TestUpdateHoles(t *testing.T) { } } } - -func TestSetCallback(t *testing.T) { - result := 0 - reasonTimeout := false - - cb1 := func(timedOut bool) { result = 1; reasonTimeout = timedOut } - cb2 := func(timedOut bool) { result = 2; reasonTimeout = timedOut } - - r := newReassembler(FragmentID{}, &faketime.NullClock{}) - if !r.setCallback(cb1) { - t.Errorf("setCallback failed") - } - if r.setCallback(cb2) { - t.Errorf("setCallback should fail if one is already set") - } - r.release(true) - if result != 1 { - t.Errorf("got result = %d, want = 1", result) - } - if !reasonTimeout { - t.Errorf("got reasonTimeout = %t, want = true", reasonTimeout) - } -} diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 204b182e6..488945226 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -514,3 +514,18 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi counter.Increment() return nil } + +// OnReassemblyTimeout implements fragmentation.TimeoutHandler. +func (p *protocol) OnReassemblyTimeout(pkt *stack.PacketBuffer) { + // OnReassemblyTimeout sends a Time Exceeded Message, as per RFC 792: + // + // If a host reassembling a fragmented datagram cannot complete the + // reassembly due to missing fragments within its time limit it discards the + // datagram, and it may send a time exceeded message. + // + // If fragment zero is not available then no time exceeded need be sent at + // all. + if pkt != nil { + p.returnError(&icmpReasonReassemblyTimeout{}, pkt) + } +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index a9a38b851..1efe6297a 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -650,29 +650,8 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { return } - // Set up a callback in case we need to send a Time Exceeded Message, as per - // RFC 792: - // - // If a host reassembling a fragmented datagram cannot complete the - // reassembly due to missing fragments within its time limit it discards - // the datagram, and it may send a time exceeded message. - // - // If fragment zero is not available then no time exceeded need be sent at - // all. - var releaseCB func(bool) - if start == 0 { - pkt := pkt.Clone() - releaseCB = func(timedOut bool) { - if timedOut { - _ = e.protocol.returnError(&icmpReasonReassemblyTimeout{}, pkt) - } - } - } - - var ready bool - var err error proto := h.Protocol() - pkt.Data, _, ready, err = e.protocol.fragmentation.Process( + data, _, ready, err := e.protocol.fragmentation.Process( // As per RFC 791 section 2.3, the identification value is unique // for a source-destination pair and protocol. fragmentation.FragmentID{ @@ -685,8 +664,7 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { start+uint16(pkt.Data.Size())-1, h.More(), proto, - pkt.Data, - releaseCB, + pkt, ) if err != nil { stats.IP.MalformedPacketsReceived.Increment() @@ -696,6 +674,7 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { if !ready { return } + pkt.Data = data // The reassembler doesn't take care of fixing up the header, so we need // to do it here. @@ -863,6 +842,7 @@ func (e *endpoint) IsInGroup(addr tcpip.Address) bool { var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) var _ stack.NetworkProtocol = (*protocol)(nil) +var _ fragmentation.TimeoutHandler = (*protocol)(nil) type protocol struct { stack *stack.Stack @@ -1027,13 +1007,14 @@ func NewProtocol(s *stack.Stack) stack.NetworkProtocol { } hashIV := r[buckets] - return &protocol{ - stack: s, - ids: ids, - hashIV: hashIV, - defaultTTL: DefaultTTL, - fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock()), + p := &protocol{ + stack: s, + ids: ids, + hashIV: hashIV, + defaultTTL: DefaultTTL, } + p.fragmentation = fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock(), p) + return p } func buildNextFragment(pf *fragmentation.PacketFragmenter, originalIPHeader header.IPv4) (*stack.PacketBuffer, bool) { diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 8d788af80..beb8f562e 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -922,3 +922,16 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi counter.Increment() return nil } + +// OnReassemblyTimeout implements fragmentation.TimeoutHandler. +func (p *protocol) OnReassemblyTimeout(pkt *stack.PacketBuffer) { + // OnReassemblyTimeout sends a Time Exceeded Message as per RFC 2460 Section + // 4.5: + // + // If the first fragment (i.e., the one with a Fragment Offset of zero) has + // been received, an ICMP Time Exceeded -- Fragment Reassembly Time Exceeded + // message should be sent to the source of that fragment. + if pkt != nil { + p.returnError(&icmpReasonReassemblyTimeout{}, pkt) + } +} diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 7697ff987..7a00f6314 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -967,18 +967,6 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { return } - // Set up a callback in case we need to send a Time Exceeded Message as - // per RFC 2460 Section 4.5. - var releaseCB func(bool) - if start == 0 { - pkt := pkt.Clone() - releaseCB = func(timedOut bool) { - if timedOut { - _ = e.protocol.returnError(&icmpReasonReassemblyTimeout{}, pkt) - } - } - } - // Note that pkt doesn't have its transport header set after reassembly, // and won't until DeliverNetworkPacket sets it. data, proto, ready, err := e.protocol.fragmentation.Process( @@ -993,17 +981,17 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { start+uint16(fragmentPayloadLen)-1, extHdr.More(), uint8(rawPayload.Identifier), - rawPayload.Buf, - releaseCB, + pkt, ) if err != nil { stats.IP.MalformedPacketsReceived.Increment() stats.IP.MalformedFragmentsReceived.Increment() return } - pkt.Data = data if ready { + pkt.Data = data + // We create a new iterator with the reassembled packet because we could // have more extension headers in the reassembled payload, as per RFC // 8200 section 4.5. We also use the NextHeader value from the first @@ -1414,6 +1402,7 @@ func (e *endpoint) IsInGroup(addr tcpip.Address) bool { var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) var _ stack.NetworkProtocol = (*protocol)(nil) +var _ fragmentation.TimeoutHandler = (*protocol)(nil) type protocol struct { stack *stack.Stack @@ -1669,10 +1658,9 @@ 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, ReassembleTimeout, s.Clock()), - ids: ids, - hashIV: hashIV, + stack: s, + ids: ids, + hashIV: hashIV, ndpDisp: opts.NDPDisp, ndpConfigs: opts.NDPConfigs, @@ -1680,6 +1668,7 @@ func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory { tempIIDSeed: opts.TempIIDSeed, autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, } + p.fragmentation = fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock(), p) p.mu.eps = make(map[*endpoint]struct{}) p.SetDefaultTTL(DefaultTTL) return p |