4 * An implementation of Ack Vectors for the DCCP protocol
5 * Copyright (c) 2007 University of Aberdeen, Scotland, UK
6 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License as published by the
10 * Free Software Foundation; version 2 of the License;
16 #include <linux/init.h>
17 #include <linux/errno.h>
18 #include <linux/kernel.h>
19 #include <linux/skbuff.h>
20 #include <linux/slab.h>
24 static struct kmem_cache
*dccp_ackvec_slab
;
25 static struct kmem_cache
*dccp_ackvec_record_slab
;
27 struct dccp_ackvec
*dccp_ackvec_alloc(const gfp_t priority
)
29 struct dccp_ackvec
*av
= kmem_cache_zalloc(dccp_ackvec_slab
, priority
);
32 av
->av_buf_head
= av
->av_buf_tail
= DCCPAV_MAX_ACKVEC_LEN
- 1;
33 INIT_LIST_HEAD(&av
->av_records
);
38 static void dccp_ackvec_purge_records(struct dccp_ackvec
*av
)
40 struct dccp_ackvec_record
*cur
, *next
;
42 list_for_each_entry_safe(cur
, next
, &av
->av_records
, avr_node
)
43 kmem_cache_free(dccp_ackvec_record_slab
, cur
);
44 INIT_LIST_HEAD(&av
->av_records
);
47 void dccp_ackvec_free(struct dccp_ackvec
*av
)
49 if (likely(av
!= NULL
)) {
50 dccp_ackvec_purge_records(av
);
51 kmem_cache_free(dccp_ackvec_slab
, av
);
56 * dccp_ackvec_update_records - Record information about sent Ack Vectors
57 * @av: Ack Vector records to update
58 * @seqno: Sequence number of the packet carrying the Ack Vector just sent
59 * @nonce_sum: The sum of all buffer nonces contained in the Ack Vector
61 int dccp_ackvec_update_records(struct dccp_ackvec
*av
, u64 seqno
, u8 nonce_sum
)
63 struct dccp_ackvec_record
*avr
;
65 avr
= kmem_cache_alloc(dccp_ackvec_record_slab
, GFP_ATOMIC
);
69 avr
->avr_ack_seqno
= seqno
;
70 avr
->avr_ack_ptr
= av
->av_buf_head
;
71 avr
->avr_ack_ackno
= av
->av_buf_ackno
;
72 avr
->avr_ack_nonce
= nonce_sum
;
73 avr
->avr_ack_runlen
= dccp_ackvec_runlen(av
->av_buf
+ av
->av_buf_head
);
75 * When the buffer overflows, we keep no more than one record. This is
76 * the simplest way of disambiguating sender-Acks dating from before the
77 * overflow from sender-Acks which refer to after the overflow; a simple
78 * solution is preferable here since we are handling an exception.
81 dccp_ackvec_purge_records(av
);
83 * Since GSS is incremented for each packet, the list is automatically
84 * arranged in descending order of @ack_seqno.
86 list_add(&avr
->avr_node
, &av
->av_records
);
88 dccp_pr_debug("Added Vector, ack_seqno=%llu, ack_ackno=%llu (rl=%u)\n",
89 (unsigned long long)avr
->avr_ack_seqno
,
90 (unsigned long long)avr
->avr_ack_ackno
,
95 static struct dccp_ackvec_record
*dccp_ackvec_lookup(struct list_head
*av_list
,
98 struct dccp_ackvec_record
*avr
;
100 * Exploit that records are inserted in descending order of sequence
101 * number, start with the oldest record first. If @ackno is `before'
102 * the earliest ack_ackno, the packet is too old to be considered.
104 list_for_each_entry_reverse(avr
, av_list
, avr_node
) {
105 if (avr
->avr_ack_seqno
== ackno
)
107 if (before48(ackno
, avr
->avr_ack_seqno
))
114 * Buffer index and length computation using modulo-buffersize arithmetic.
115 * Note that, as pointers move from right to left, head is `before' tail.
117 static inline u16
__ackvec_idx_add(const u16 a
, const u16 b
)
119 return (a
+ b
) % DCCPAV_MAX_ACKVEC_LEN
;
122 static inline u16
__ackvec_idx_sub(const u16 a
, const u16 b
)
124 return __ackvec_idx_add(a
, DCCPAV_MAX_ACKVEC_LEN
- b
);
127 u16
dccp_ackvec_buflen(const struct dccp_ackvec
*av
)
129 if (unlikely(av
->av_overflow
))
130 return DCCPAV_MAX_ACKVEC_LEN
;
131 return __ackvec_idx_sub(av
->av_buf_tail
, av
->av_buf_head
);
135 * dccp_ackvec_update_old - Update previous state as per RFC 4340, 11.4.1
136 * @av: non-empty buffer to update
137 * @distance: negative or zero distance of @seqno from buf_ackno downward
138 * @seqno: the (old) sequence number whose record is to be updated
139 * @state: state in which packet carrying @seqno was received
141 static void dccp_ackvec_update_old(struct dccp_ackvec
*av
, s64 distance
,
142 u64 seqno
, enum dccp_ackvec_states state
)
144 u16 ptr
= av
->av_buf_head
;
146 BUG_ON(distance
> 0);
147 if (unlikely(dccp_ackvec_is_empty(av
)))
151 u8 runlen
= dccp_ackvec_runlen(av
->av_buf
+ ptr
);
153 if (distance
+ runlen
>= 0) {
155 * Only update the state if packet has not been received
156 * yet. This is OK as per the second table in RFC 4340,
157 * 11.4.1; i.e. here we are using the following table:
167 * The "Not Received" state was set by reserve_seats().
169 if (av
->av_buf
[ptr
] == DCCPAV_NOT_RECEIVED
)
170 av
->av_buf
[ptr
] = state
;
172 dccp_pr_debug("Not changing %llu state to %u\n",
173 (unsigned long long)seqno
, state
);
177 distance
+= runlen
+ 1;
178 ptr
= __ackvec_idx_add(ptr
, 1);
180 } while (ptr
!= av
->av_buf_tail
);
183 /* Mark @num entries after buf_head as "Not yet received". */
184 static void dccp_ackvec_reserve_seats(struct dccp_ackvec
*av
, u16 num
)
186 u16 start
= __ackvec_idx_add(av
->av_buf_head
, 1),
187 len
= DCCPAV_MAX_ACKVEC_LEN
- start
;
189 /* check for buffer wrap-around */
191 memset(av
->av_buf
+ start
, DCCPAV_NOT_RECEIVED
, len
);
196 memset(av
->av_buf
+ start
, DCCPAV_NOT_RECEIVED
, num
);
200 * dccp_ackvec_add_new - Record one or more new entries in Ack Vector buffer
201 * @av: container of buffer to update (can be empty or non-empty)
202 * @num_packets: number of packets to register (must be >= 1)
203 * @seqno: sequence number of the first packet in @num_packets
204 * @state: state in which packet carrying @seqno was received
206 static void dccp_ackvec_add_new(struct dccp_ackvec
*av
, u32 num_packets
,
207 u64 seqno
, enum dccp_ackvec_states state
)
209 u32 num_cells
= num_packets
;
211 if (num_packets
> DCCPAV_BURST_THRESH
) {
212 u32 lost_packets
= num_packets
- 1;
214 DCCP_WARN("Warning: large burst loss (%u)\n", lost_packets
);
216 * We received 1 packet and have a loss of size "num_packets-1"
217 * which we squeeze into num_cells-1 rather than reserving an
218 * entire byte for each lost packet.
219 * The reason is that the vector grows in O(burst_length); when
220 * it grows too large there will no room left for the payload.
221 * This is a trade-off: if a few packets out of the burst show
222 * up later, their state will not be changed; it is simply too
223 * costly to reshuffle/reallocate/copy the buffer each time.
224 * Should such problems persist, we will need to switch to a
225 * different underlying data structure.
227 for (num_packets
= num_cells
= 1; lost_packets
; ++num_cells
) {
228 u8 len
= min(lost_packets
, (u32
)DCCPAV_MAX_RUNLEN
);
230 av
->av_buf_head
= __ackvec_idx_sub(av
->av_buf_head
, 1);
231 av
->av_buf
[av
->av_buf_head
] = DCCPAV_NOT_RECEIVED
| len
;
237 if (num_cells
+ dccp_ackvec_buflen(av
) >= DCCPAV_MAX_ACKVEC_LEN
) {
238 DCCP_CRIT("Ack Vector buffer overflow: dropping old entries\n");
239 av
->av_overflow
= true;
242 av
->av_buf_head
= __ackvec_idx_sub(av
->av_buf_head
, num_packets
);
244 av
->av_buf_tail
= av
->av_buf_head
;
246 av
->av_buf
[av
->av_buf_head
] = state
;
247 av
->av_buf_ackno
= seqno
;
250 dccp_ackvec_reserve_seats(av
, num_packets
- 1);
254 * dccp_ackvec_input - Register incoming packet in the buffer
256 void dccp_ackvec_input(struct dccp_ackvec
*av
, struct sk_buff
*skb
)
258 u64 seqno
= DCCP_SKB_CB(skb
)->dccpd_seq
;
259 enum dccp_ackvec_states state
= DCCPAV_RECEIVED
;
261 if (dccp_ackvec_is_empty(av
)) {
262 dccp_ackvec_add_new(av
, 1, seqno
, state
);
263 av
->av_tail_ackno
= seqno
;
266 s64 num_packets
= dccp_delta_seqno(av
->av_buf_ackno
, seqno
);
267 u8
*current_head
= av
->av_buf
+ av
->av_buf_head
;
269 if (num_packets
== 1 &&
270 dccp_ackvec_state(current_head
) == state
&&
271 dccp_ackvec_runlen(current_head
) < DCCPAV_MAX_RUNLEN
) {
274 av
->av_buf_ackno
= seqno
;
276 } else if (num_packets
> 0) {
277 dccp_ackvec_add_new(av
, num_packets
, seqno
, state
);
279 dccp_ackvec_update_old(av
, num_packets
, seqno
, state
);
285 * If several packets are missing, the HC-Receiver may prefer to enter multiple
286 * bytes with run length 0, rather than a single byte with a larger run length;
287 * this simplifies table updates if one of the missing packets arrives.
289 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec
*av
,
290 const unsigned int packets
,
291 const unsigned char state
)
296 if (av
->av_vec_len
+ packets
> DCCPAV_MAX_ACKVEC_LEN
)
300 new_head
= av
->av_buf_head
- packets
;
304 memset(av
->av_buf
, DCCPAV_NOT_RECEIVED
,
308 new_head
+= DCCPAV_MAX_ACKVEC_LEN
;
311 av
->av_buf_head
= new_head
;
314 memset(av
->av_buf
+ av
->av_buf_head
+ 1,
315 DCCPAV_NOT_RECEIVED
, gap
);
317 av
->av_buf
[av
->av_buf_head
] = state
;
318 av
->av_vec_len
+= packets
;
323 * Implements the RFC 4340, Appendix A
325 int dccp_ackvec_add(struct dccp_ackvec
*av
, const struct sock
*sk
,
326 const u64 ackno
, const u8 state
)
328 u8
*cur_head
= av
->av_buf
+ av
->av_buf_head
,
329 *buf_end
= av
->av_buf
+ DCCPAV_MAX_ACKVEC_LEN
;
331 * Check at the right places if the buffer is full, if it is, tell the
332 * caller to start dropping packets till the HC-Sender acks our ACK
333 * vectors, when we will free up space in av_buf.
335 * We may well decide to do buffer compression, etc, but for now lets
338 * From Appendix A.1.1 (`New Packets'):
340 * Of course, the circular buffer may overflow, either when the
341 * HC-Sender is sending data at a very high rate, when the
342 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
343 * or when the HC-Sender is forgetting to acknowledge those acks
344 * (so the HC-Receiver is unable to clean up old state). In this
345 * case, the HC-Receiver should either compress the buffer (by
346 * increasing run lengths when possible), transfer its state to
347 * a larger buffer, or, as a last resort, drop all received
348 * packets, without processing them whatsoever, until its buffer
352 /* See if this is the first ackno being inserted */
353 if (av
->av_vec_len
== 0) {
356 } else if (after48(ackno
, av
->av_buf_ackno
)) {
357 const u64 delta
= dccp_delta_seqno(av
->av_buf_ackno
, ackno
);
360 * Look if the state of this packet is the same as the
361 * previous ackno and if so if we can bump the head len.
363 if (delta
== 1 && dccp_ackvec_state(cur_head
) == state
&&
364 dccp_ackvec_runlen(cur_head
) < DCCPAV_MAX_RUNLEN
)
366 else if (dccp_ackvec_set_buf_head_state(av
, delta
, state
))
372 * When a packet with Sequence Number S <= buf_ackno
373 * arrives, the HC-Receiver will scan the table for
374 * the byte corresponding to S. (Indexing structures
375 * could reduce the complexity of this scan.)
377 u64 delta
= dccp_delta_seqno(ackno
, av
->av_buf_ackno
);
380 const u8 len
= dccp_ackvec_runlen(cur_head
);
382 * valid packets not yet in av_buf have a reserved
383 * entry, with a len equal to 0.
385 if (*cur_head
== DCCPAV_NOT_RECEIVED
&& delta
== 0) {
386 dccp_pr_debug("Found %llu reserved seat!\n",
387 (unsigned long long)ackno
);
391 /* len == 0 means one packet */
396 if (++cur_head
== buf_end
)
397 cur_head
= av
->av_buf
;
401 av
->av_buf_ackno
= ackno
;
406 /* Duplicate packet */
407 dccp_pr_debug("Received a dup or already considered lost "
408 "packet: %llu\n", (unsigned long long)ackno
);
412 static void dccp_ackvec_throw_record(struct dccp_ackvec
*av
,
413 struct dccp_ackvec_record
*avr
)
415 struct dccp_ackvec_record
*next
;
417 /* sort out vector length */
418 if (av
->av_buf_head
<= avr
->avr_ack_ptr
)
419 av
->av_vec_len
= avr
->avr_ack_ptr
- av
->av_buf_head
;
421 av
->av_vec_len
= DCCPAV_MAX_ACKVEC_LEN
- 1 -
422 av
->av_buf_head
+ avr
->avr_ack_ptr
;
425 list_for_each_entry_safe_from(avr
, next
, &av
->av_records
, avr_node
) {
426 list_del(&avr
->avr_node
);
427 kmem_cache_free(dccp_ackvec_record_slab
, avr
);
431 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec
*av
, struct sock
*sk
,
434 struct dccp_ackvec_record
*avr
;
437 * If we traverse backwards, it should be faster when we have large
438 * windows. We will be receiving ACKs for stuff we sent a while back
441 list_for_each_entry_reverse(avr
, &av
->av_records
, avr_node
) {
442 if (ackno
== avr
->avr_ack_seqno
) {
443 dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
444 "ack_ackno=%llu, ACKED!\n",
445 dccp_role(sk
), avr
->avr_ack_runlen
,
446 (unsigned long long)avr
->avr_ack_seqno
,
447 (unsigned long long)avr
->avr_ack_ackno
);
448 dccp_ackvec_throw_record(av
, avr
);
450 } else if (avr
->avr_ack_seqno
> ackno
)
451 break; /* old news */
455 static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec
*av
,
456 struct sock
*sk
, u64
*ackno
,
457 const unsigned char len
,
458 const unsigned char *vector
)
461 struct dccp_ackvec_record
*avr
;
463 /* Check if we actually sent an ACK vector */
464 if (list_empty(&av
->av_records
))
470 * I think it might be more efficient to work backwards. See comment on
473 avr
= list_entry(av
->av_records
.next
, struct dccp_ackvec_record
, avr_node
);
475 const u8 rl
= dccp_ackvec_runlen(vector
);
478 dccp_set_seqno(&ackno_end_rl
, *ackno
- rl
);
481 * If our AVR sequence number is greater than the ack, go
482 * forward in the AVR list until it is not so.
484 list_for_each_entry_from(avr
, &av
->av_records
, avr_node
) {
485 if (!after48(avr
->avr_ack_seqno
, *ackno
))
488 /* End of the av_records list, not found, exit */
491 if (between48(avr
->avr_ack_seqno
, ackno_end_rl
, *ackno
)) {
492 if (dccp_ackvec_state(vector
) != DCCPAV_NOT_RECEIVED
) {
493 dccp_pr_debug("%s ACK vector 0, len=%d, "
494 "ack_seqno=%llu, ack_ackno=%llu, "
501 dccp_ackvec_throw_record(av
, avr
);
505 * If it wasn't received, continue scanning... we might
510 dccp_set_seqno(ackno
, ackno_end_rl
- 1);
515 int dccp_ackvec_parse(struct sock
*sk
, const struct sk_buff
*skb
,
516 u64
*ackno
, const u8 opt
, const u8
*value
, const u8 len
)
518 if (len
> DCCP_SINGLE_OPT_MAXLEN
)
521 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
522 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk
)->dccps_hc_rx_ackvec
, sk
,
528 * dccp_ackvec_clear_state - Perform house-keeping / garbage-collection
529 * This routine is called when the peer acknowledges the receipt of Ack Vectors
530 * up to and including @ackno. While based on on section A.3 of RFC 4340, here
531 * are additional precautions to prevent corrupted buffer state. In particular,
532 * we use tail_ackno to identify outdated records; it always marks the earliest
533 * packet of group (2) in 11.4.2.
535 void dccp_ackvec_clear_state(struct dccp_ackvec
*av
, const u64 ackno
)
537 struct dccp_ackvec_record
*avr
, *next
;
538 u8 runlen_now
, eff_runlen
;
541 avr
= dccp_ackvec_lookup(&av
->av_records
, ackno
);
545 * Deal with outdated acknowledgments: this arises when e.g. there are
546 * several old records and the acks from the peer come in slowly. In
547 * that case we may still have records that pre-date tail_ackno.
549 delta
= dccp_delta_seqno(av
->av_tail_ackno
, avr
->avr_ack_ackno
);
553 * Deal with overlapping Ack Vectors: don't subtract more than the
554 * number of packets between tail_ackno and ack_ackno.
556 eff_runlen
= delta
< avr
->avr_ack_runlen
? delta
: avr
->avr_ack_runlen
;
558 runlen_now
= dccp_ackvec_runlen(av
->av_buf
+ avr
->avr_ack_ptr
);
560 * The run length of Ack Vector cells does not decrease over time. If
561 * the run length is the same as at the time the Ack Vector was sent, we
562 * free the ack_ptr cell. That cell can however not be freed if the run
563 * length has increased: in this case we need to move the tail pointer
564 * backwards (towards higher indices), to its next-oldest neighbour.
566 if (runlen_now
> eff_runlen
) {
568 av
->av_buf
[avr
->avr_ack_ptr
] -= eff_runlen
+ 1;
569 av
->av_buf_tail
= __ackvec_idx_add(avr
->avr_ack_ptr
, 1);
571 /* This move may not have cleared the overflow flag. */
573 av
->av_overflow
= (av
->av_buf_head
== av
->av_buf_tail
);
575 av
->av_buf_tail
= avr
->avr_ack_ptr
;
577 * We have made sure that avr points to a valid cell within the
578 * buffer. This cell is either older than head, or equals head
579 * (empty buffer): in both cases we no longer have any overflow.
585 * The peer has acknowledged up to and including ack_ackno. Hence the
586 * first packet in group (2) of 11.4.2 is the successor of ack_ackno.
588 av
->av_tail_ackno
= ADD48(avr
->avr_ack_ackno
, 1);
591 list_for_each_entry_safe_from(avr
, next
, &av
->av_records
, avr_node
) {
592 list_del(&avr
->avr_node
);
593 kmem_cache_free(dccp_ackvec_record_slab
, avr
);
597 int __init
dccp_ackvec_init(void)
599 dccp_ackvec_slab
= kmem_cache_create("dccp_ackvec",
600 sizeof(struct dccp_ackvec
), 0,
601 SLAB_HWCACHE_ALIGN
, NULL
);
602 if (dccp_ackvec_slab
== NULL
)
605 dccp_ackvec_record_slab
= kmem_cache_create("dccp_ackvec_record",
606 sizeof(struct dccp_ackvec_record
),
607 0, SLAB_HWCACHE_ALIGN
, NULL
);
608 if (dccp_ackvec_record_slab
== NULL
)
609 goto out_destroy_slab
;
614 kmem_cache_destroy(dccp_ackvec_slab
);
615 dccp_ackvec_slab
= NULL
;
617 DCCP_CRIT("Unable to create Ack Vector slab cache");
621 void dccp_ackvec_exit(void)
623 if (dccp_ackvec_slab
!= NULL
) {
624 kmem_cache_destroy(dccp_ackvec_slab
);
625 dccp_ackvec_slab
= NULL
;
627 if (dccp_ackvec_record_slab
!= NULL
) {
628 kmem_cache_destroy(dccp_ackvec_record_slab
);
629 dccp_ackvec_record_slab
= NULL
;