4 * An implementation of the DCCP protocol
5 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License as published by the
9 * Free Software Foundation; version 2 of the License;
15 #include <linux/dccp.h>
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 static struct dccp_ackvec_record *dccp_ackvec_record_new(void)
29 struct dccp_ackvec_record *avr =
30 kmem_cache_alloc(dccp_ackvec_record_slab, GFP_ATOMIC);
33 INIT_LIST_HEAD(&avr->dccpavr_node);
38 static void dccp_ackvec_record_delete(struct dccp_ackvec_record *avr)
40 if (unlikely(avr == NULL))
42 /* Check if deleting a linked record */
43 WARN_ON(!list_empty(&avr->dccpavr_node));
44 kmem_cache_free(dccp_ackvec_record_slab, avr);
47 static void dccp_ackvec_insert_avr(struct dccp_ackvec *av,
48 struct dccp_ackvec_record *avr)
51 * AVRs are sorted by seqno. Since we are sending them in order, we
52 * just add the AVR at the head of the list.
55 if (!list_empty(&av->dccpav_records)) {
56 const struct dccp_ackvec_record *head =
57 list_entry(av->dccpav_records.next,
58 struct dccp_ackvec_record,
60 BUG_ON(before48(avr->dccpavr_ack_seqno,
61 head->dccpavr_ack_seqno));
64 list_add(&avr->dccpavr_node, &av->dccpav_records);
67 int dccp_insert_option_ackvec(struct sock *sk, struct sk_buff *skb)
69 struct dccp_sock *dp = dccp_sk(sk);
70 struct dccp_ackvec *av = dp->dccps_hc_rx_ackvec;
71 /* Figure out how many options do we need to represent the ackvec */
72 const u16 nr_opts = (av->dccpav_vec_len +
73 DCCP_MAX_ACKVEC_OPT_LEN - 1) /
74 DCCP_MAX_ACKVEC_OPT_LEN;
75 u16 len = av->dccpav_vec_len + 2 * nr_opts, i;
77 const unsigned char *tail, *from;
79 struct dccp_ackvec_record *avr;
82 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
85 delta = ktime_us_delta(ktime_get_real(), av->dccpav_time);
86 elapsed_time = delta / 10;
88 if (elapsed_time != 0 &&
89 dccp_insert_option_elapsed_time(sk, skb, elapsed_time))
92 avr = dccp_ackvec_record_new();
96 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
98 to = skb_push(skb, len);
99 len = av->dccpav_vec_len;
100 from = av->dccpav_buf + av->dccpav_buf_head;
101 tail = av->dccpav_buf + DCCP_MAX_ACKVEC_LEN;
103 for (i = 0; i < nr_opts; ++i) {
106 if (len > DCCP_MAX_ACKVEC_OPT_LEN)
107 copylen = DCCP_MAX_ACKVEC_OPT_LEN;
109 *to++ = DCCPO_ACK_VECTOR_0;
112 /* Check if buf_head wraps */
113 if (from + copylen > tail) {
114 const u16 tailsize = tail - from;
116 memcpy(to, from, tailsize);
120 from = av->dccpav_buf;
123 memcpy(to, from, copylen);
130 * From RFC 4340, A.2:
132 * For each acknowledgement it sends, the HC-Receiver will add an
133 * acknowledgement record. ack_seqno will equal the HC-Receiver
134 * sequence number it used for the ack packet; ack_ptr will equal
135 * buf_head; ack_ackno will equal buf_ackno; and ack_nonce will
138 avr->dccpavr_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
139 avr->dccpavr_ack_ptr = av->dccpav_buf_head;
140 avr->dccpavr_ack_ackno = av->dccpav_buf_ackno;
141 avr->dccpavr_ack_nonce = av->dccpav_buf_nonce;
142 avr->dccpavr_sent_len = av->dccpav_vec_len;
144 dccp_ackvec_insert_avr(av, avr);
146 dccp_pr_debug("%s ACK Vector 0, len=%d, ack_seqno=%llu, "
148 dccp_role(sk), avr->dccpavr_sent_len,
149 (unsigned long long)avr->dccpavr_ack_seqno,
150 (unsigned long long)avr->dccpavr_ack_ackno);
154 struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
156 struct dccp_ackvec *av = kmem_cache_alloc(dccp_ackvec_slab, priority);
159 av->dccpav_buf_head = DCCP_MAX_ACKVEC_LEN - 1;
160 av->dccpav_buf_ackno = UINT48_MAX + 1;
161 av->dccpav_buf_nonce = av->dccpav_buf_nonce = 0;
162 av->dccpav_time = ktime_set(0, 0);
163 av->dccpav_vec_len = 0;
164 INIT_LIST_HEAD(&av->dccpav_records);
170 void dccp_ackvec_free(struct dccp_ackvec *av)
172 if (unlikely(av == NULL))
175 if (!list_empty(&av->dccpav_records)) {
176 struct dccp_ackvec_record *avr, *next;
178 list_for_each_entry_safe(avr, next, &av->dccpav_records,
180 list_del_init(&avr->dccpavr_node);
181 dccp_ackvec_record_delete(avr);
185 kmem_cache_free(dccp_ackvec_slab, av);
188 static inline u8 dccp_ackvec_state(const struct dccp_ackvec *av,
191 return av->dccpav_buf[index] & DCCP_ACKVEC_STATE_MASK;
194 static inline u8 dccp_ackvec_len(const struct dccp_ackvec *av,
197 return av->dccpav_buf[index] & DCCP_ACKVEC_LEN_MASK;
201 * If several packets are missing, the HC-Receiver may prefer to enter multiple
202 * bytes with run length 0, rather than a single byte with a larger run length;
203 * this simplifies table updates if one of the missing packets arrives.
205 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec *av,
206 const unsigned int packets,
207 const unsigned char state)
212 if (av->dccpav_vec_len + packets > DCCP_MAX_ACKVEC_LEN)
216 new_head = av->dccpav_buf_head - packets;
220 memset(av->dccpav_buf, DCCP_ACKVEC_STATE_NOT_RECEIVED,
224 new_head += DCCP_MAX_ACKVEC_LEN;
227 av->dccpav_buf_head = new_head;
230 memset(av->dccpav_buf + av->dccpav_buf_head + 1,
231 DCCP_ACKVEC_STATE_NOT_RECEIVED, gap);
233 av->dccpav_buf[av->dccpav_buf_head] = state;
234 av->dccpav_vec_len += packets;
239 * Implements the RFC 4340, Appendix A
241 int dccp_ackvec_add(struct dccp_ackvec *av, const struct sock *sk,
242 const u64 ackno, const u8 state)
245 * Check at the right places if the buffer is full, if it is, tell the
246 * caller to start dropping packets till the HC-Sender acks our ACK
247 * vectors, when we will free up space in dccpav_buf.
249 * We may well decide to do buffer compression, etc, but for now lets
252 * From Appendix A.1.1 (`New Packets'):
254 * Of course, the circular buffer may overflow, either when the
255 * HC-Sender is sending data at a very high rate, when the
256 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
257 * or when the HC-Sender is forgetting to acknowledge those acks
258 * (so the HC-Receiver is unable to clean up old state). In this
259 * case, the HC-Receiver should either compress the buffer (by
260 * increasing run lengths when possible), transfer its state to
261 * a larger buffer, or, as a last resort, drop all received
262 * packets, without processing them whatsoever, until its buffer
266 /* See if this is the first ackno being inserted */
267 if (av->dccpav_vec_len == 0) {
268 av->dccpav_buf[av->dccpav_buf_head] = state;
269 av->dccpav_vec_len = 1;
270 } else if (after48(ackno, av->dccpav_buf_ackno)) {
271 const u64 delta = dccp_delta_seqno(av->dccpav_buf_ackno,
275 * Look if the state of this packet is the same as the
276 * previous ackno and if so if we can bump the head len.
279 dccp_ackvec_state(av, av->dccpav_buf_head) == state &&
280 (dccp_ackvec_len(av, av->dccpav_buf_head) <
281 DCCP_ACKVEC_LEN_MASK))
282 av->dccpav_buf[av->dccpav_buf_head]++;
283 else if (dccp_ackvec_set_buf_head_state(av, delta, state))
289 * When a packet with Sequence Number S <= buf_ackno
290 * arrives, the HC-Receiver will scan the table for
291 * the byte corresponding to S. (Indexing structures
292 * could reduce the complexity of this scan.)
294 u64 delta = dccp_delta_seqno(ackno, av->dccpav_buf_ackno);
295 u32 index = av->dccpav_buf_head;
298 const u8 len = dccp_ackvec_len(av, index);
299 const u8 state = dccp_ackvec_state(av, index);
301 * valid packets not yet in dccpav_buf have a reserved
302 * entry, with a len equal to 0.
304 if (state == DCCP_ACKVEC_STATE_NOT_RECEIVED &&
305 len == 0 && delta == 0) { /* Found our
307 dccp_pr_debug("Found %llu reserved seat!\n",
308 (unsigned long long)ackno);
309 av->dccpav_buf[index] = state;
312 /* len == 0 means one packet */
317 if (++index == DCCP_MAX_ACKVEC_LEN)
322 av->dccpav_buf_ackno = ackno;
323 av->dccpav_time = ktime_get_real();
328 /* Duplicate packet */
329 dccp_pr_debug("Received a dup or already considered lost "
330 "packet: %llu\n", (unsigned long long)ackno);
334 #ifdef CONFIG_IP_DCCP_DEBUG
335 void dccp_ackvector_print(const u64 ackno, const unsigned char *vector, int len)
337 dccp_pr_debug_cat("ACK vector len=%d, ackno=%llu |", len,
338 (unsigned long long)ackno);
341 const u8 state = (*vector & DCCP_ACKVEC_STATE_MASK) >> 6;
342 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
344 dccp_pr_debug_cat("%d,%d|", state, rl);
348 dccp_pr_debug_cat("\n");
351 void dccp_ackvec_print(const struct dccp_ackvec *av)
353 dccp_ackvector_print(av->dccpav_buf_ackno,
354 av->dccpav_buf + av->dccpav_buf_head,
359 static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
360 struct dccp_ackvec_record *avr)
362 struct dccp_ackvec_record *next;
364 /* sort out vector length */
365 if (av->dccpav_buf_head <= avr->dccpavr_ack_ptr)
366 av->dccpav_vec_len = avr->dccpavr_ack_ptr - av->dccpav_buf_head;
368 av->dccpav_vec_len = DCCP_MAX_ACKVEC_LEN - 1
369 - av->dccpav_buf_head
370 + avr->dccpavr_ack_ptr;
373 list_for_each_entry_safe_from(avr, next, &av->dccpav_records,
375 list_del_init(&avr->dccpavr_node);
376 dccp_ackvec_record_delete(avr);
380 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
383 struct dccp_ackvec_record *avr;
386 * If we traverse backwards, it should be faster when we have large
387 * windows. We will be receiving ACKs for stuff we sent a while back
390 list_for_each_entry_reverse(avr, &av->dccpav_records, dccpavr_node) {
391 if (ackno == avr->dccpavr_ack_seqno) {
392 dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
393 "ack_ackno=%llu, ACKED!\n",
395 (unsigned long long)avr->dccpavr_ack_seqno,
396 (unsigned long long)avr->dccpavr_ack_ackno);
397 dccp_ackvec_throw_record(av, avr);
399 } else if (avr->dccpavr_ack_seqno > ackno)
400 break; /* old news */
404 static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec *av,
405 struct sock *sk, u64 *ackno,
406 const unsigned char len,
407 const unsigned char *vector)
410 struct dccp_ackvec_record *avr;
412 /* Check if we actually sent an ACK vector */
413 if (list_empty(&av->dccpav_records))
419 * I think it might be more efficient to work backwards. See comment on
422 avr = list_entry(av->dccpav_records.next, struct dccp_ackvec_record,
425 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
428 dccp_set_seqno(&ackno_end_rl, *ackno - rl);
431 * If our AVR sequence number is greater than the ack, go
432 * forward in the AVR list until it is not so.
434 list_for_each_entry_from(avr, &av->dccpav_records,
436 if (!after48(avr->dccpavr_ack_seqno, *ackno))
439 /* End of the dccpav_records list, not found, exit */
442 if (between48(avr->dccpavr_ack_seqno, ackno_end_rl, *ackno)) {
443 const u8 state = *vector & DCCP_ACKVEC_STATE_MASK;
444 if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED) {
445 dccp_pr_debug("%s ACK vector 0, len=%d, "
446 "ack_seqno=%llu, ack_ackno=%llu, "
450 avr->dccpavr_ack_seqno,
452 avr->dccpavr_ack_ackno);
453 dccp_ackvec_throw_record(av, avr);
457 * If it wasn't received, continue scanning... we might
462 dccp_set_seqno(ackno, ackno_end_rl - 1);
467 int dccp_ackvec_parse(struct sock *sk, const struct sk_buff *skb,
468 u64 *ackno, const u8 opt, const u8 *value, const u8 len)
470 if (len > DCCP_MAX_ACKVEC_OPT_LEN)
473 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
474 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk)->dccps_hc_rx_ackvec, sk,
479 int __init dccp_ackvec_init(void)
481 dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
482 sizeof(struct dccp_ackvec), 0,
483 SLAB_HWCACHE_ALIGN, NULL);
484 if (dccp_ackvec_slab == NULL)
487 dccp_ackvec_record_slab =
488 kmem_cache_create("dccp_ackvec_record",
489 sizeof(struct dccp_ackvec_record),
490 0, SLAB_HWCACHE_ALIGN, NULL);
491 if (dccp_ackvec_record_slab == NULL)
492 goto out_destroy_slab;
497 kmem_cache_destroy(dccp_ackvec_slab);
498 dccp_ackvec_slab = NULL;
500 DCCP_CRIT("Unable to create Ack Vector slab cache");
504 void dccp_ackvec_exit(void)
506 if (dccp_ackvec_slab != NULL) {
507 kmem_cache_destroy(dccp_ackvec_slab);
508 dccp_ackvec_slab = NULL;
510 if (dccp_ackvec_record_slab != NULL) {
511 kmem_cache_destroy(dccp_ackvec_record_slab);
512 dccp_ackvec_record_slab = NULL;