[SCTP]: Switch ->cmp_addr() and sctp_cmp_addr_exact() to net-endian.
[linux-2.6/verdex.git] / net / dccp / ackvec.c
blob1b4b60d8bdec89acedbf8314d3aab1f034f08dfe
1 /*
2 * net/dccp/ackvec.c
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;
12 #include "ackvec.h"
13 #include "dccp.h"
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>
22 #include <net/sock.h>
24 static kmem_cache_t *dccp_ackvec_slab;
25 static kmem_cache_t *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);
32 if (avr != NULL)
33 INIT_LIST_HEAD(&avr->dccpavr_node);
35 return avr;
38 static void dccp_ackvec_record_delete(struct dccp_ackvec_record *avr)
40 if (unlikely(avr == NULL))
41 return;
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.
53 * -sorbo.
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,
59 dccpavr_node);
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 int len = av->dccpav_vec_len + 2;
72 struct timeval now;
73 u32 elapsed_time;
74 unsigned char *to, *from;
75 struct dccp_ackvec_record *avr;
77 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
78 return -1;
80 dccp_timestamp(sk, &now);
81 elapsed_time = timeval_delta(&now, &av->dccpav_time) / 10;
83 if (elapsed_time != 0 &&
84 dccp_insert_option_elapsed_time(sk, skb, elapsed_time))
85 return -1;
87 avr = dccp_ackvec_record_new();
88 if (avr == NULL)
89 return -1;
91 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
93 to = skb_push(skb, len);
94 *to++ = DCCPO_ACK_VECTOR_0;
95 *to++ = len;
97 len = av->dccpav_vec_len;
98 from = av->dccpav_buf + av->dccpav_buf_head;
100 /* Check if buf_head wraps */
101 if ((int)av->dccpav_buf_head + len > DCCP_MAX_ACKVEC_LEN) {
102 const u32 tailsize = DCCP_MAX_ACKVEC_LEN - av->dccpav_buf_head;
104 memcpy(to, from, tailsize);
105 to += tailsize;
106 len -= tailsize;
107 from = av->dccpav_buf;
110 memcpy(to, from, len);
112 * From RFC 4340, A.2:
114 * For each acknowledgement it sends, the HC-Receiver will add an
115 * acknowledgement record. ack_seqno will equal the HC-Receiver
116 * sequence number it used for the ack packet; ack_ptr will equal
117 * buf_head; ack_ackno will equal buf_ackno; and ack_nonce will
118 * equal buf_nonce.
120 avr->dccpavr_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
121 avr->dccpavr_ack_ptr = av->dccpav_buf_head;
122 avr->dccpavr_ack_ackno = av->dccpav_buf_ackno;
123 avr->dccpavr_ack_nonce = av->dccpav_buf_nonce;
124 avr->dccpavr_sent_len = av->dccpav_vec_len;
126 dccp_ackvec_insert_avr(av, avr);
128 dccp_pr_debug("%s ACK Vector 0, len=%d, ack_seqno=%llu, "
129 "ack_ackno=%llu\n",
130 dccp_role(sk), avr->dccpavr_sent_len,
131 (unsigned long long)avr->dccpavr_ack_seqno,
132 (unsigned long long)avr->dccpavr_ack_ackno);
133 return 0;
136 struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
138 struct dccp_ackvec *av = kmem_cache_alloc(dccp_ackvec_slab, priority);
140 if (av != NULL) {
141 av->dccpav_buf_head = DCCP_MAX_ACKVEC_LEN - 1;
142 av->dccpav_buf_ackno = DCCP_MAX_SEQNO + 1;
143 av->dccpav_buf_nonce = av->dccpav_buf_nonce = 0;
144 av->dccpav_ack_ptr = 0;
145 av->dccpav_time.tv_sec = 0;
146 av->dccpav_time.tv_usec = 0;
147 av->dccpav_vec_len = 0;
148 INIT_LIST_HEAD(&av->dccpav_records);
151 return av;
154 void dccp_ackvec_free(struct dccp_ackvec *av)
156 if (unlikely(av == NULL))
157 return;
159 if (!list_empty(&av->dccpav_records)) {
160 struct dccp_ackvec_record *avr, *next;
162 list_for_each_entry_safe(avr, next, &av->dccpav_records,
163 dccpavr_node) {
164 list_del_init(&avr->dccpavr_node);
165 dccp_ackvec_record_delete(avr);
169 kmem_cache_free(dccp_ackvec_slab, av);
172 static inline u8 dccp_ackvec_state(const struct dccp_ackvec *av,
173 const u8 index)
175 return av->dccpav_buf[index] & DCCP_ACKVEC_STATE_MASK;
178 static inline u8 dccp_ackvec_len(const struct dccp_ackvec *av,
179 const u8 index)
181 return av->dccpav_buf[index] & DCCP_ACKVEC_LEN_MASK;
185 * If several packets are missing, the HC-Receiver may prefer to enter multiple
186 * bytes with run length 0, rather than a single byte with a larger run length;
187 * this simplifies table updates if one of the missing packets arrives.
189 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec *av,
190 const unsigned int packets,
191 const unsigned char state)
193 unsigned int gap;
194 long new_head;
196 if (av->dccpav_vec_len + packets > DCCP_MAX_ACKVEC_LEN)
197 return -ENOBUFS;
199 gap = packets - 1;
200 new_head = av->dccpav_buf_head - packets;
202 if (new_head < 0) {
203 if (gap > 0) {
204 memset(av->dccpav_buf, DCCP_ACKVEC_STATE_NOT_RECEIVED,
205 gap + new_head + 1);
206 gap = -new_head;
208 new_head += DCCP_MAX_ACKVEC_LEN;
211 av->dccpav_buf_head = new_head;
213 if (gap > 0)
214 memset(av->dccpav_buf + av->dccpav_buf_head + 1,
215 DCCP_ACKVEC_STATE_NOT_RECEIVED, gap);
217 av->dccpav_buf[av->dccpav_buf_head] = state;
218 av->dccpav_vec_len += packets;
219 return 0;
223 * Implements the RFC 4340, Appendix A
225 int dccp_ackvec_add(struct dccp_ackvec *av, const struct sock *sk,
226 const u64 ackno, const u8 state)
229 * Check at the right places if the buffer is full, if it is, tell the
230 * caller to start dropping packets till the HC-Sender acks our ACK
231 * vectors, when we will free up space in dccpav_buf.
233 * We may well decide to do buffer compression, etc, but for now lets
234 * just drop.
236 * From Appendix A.1.1 (`New Packets'):
238 * Of course, the circular buffer may overflow, either when the
239 * HC-Sender is sending data at a very high rate, when the
240 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
241 * or when the HC-Sender is forgetting to acknowledge those acks
242 * (so the HC-Receiver is unable to clean up old state). In this
243 * case, the HC-Receiver should either compress the buffer (by
244 * increasing run lengths when possible), transfer its state to
245 * a larger buffer, or, as a last resort, drop all received
246 * packets, without processing them whatsoever, until its buffer
247 * shrinks again.
250 /* See if this is the first ackno being inserted */
251 if (av->dccpav_vec_len == 0) {
252 av->dccpav_buf[av->dccpav_buf_head] = state;
253 av->dccpav_vec_len = 1;
254 } else if (after48(ackno, av->dccpav_buf_ackno)) {
255 const u64 delta = dccp_delta_seqno(av->dccpav_buf_ackno,
256 ackno);
259 * Look if the state of this packet is the same as the
260 * previous ackno and if so if we can bump the head len.
262 if (delta == 1 &&
263 dccp_ackvec_state(av, av->dccpav_buf_head) == state &&
264 (dccp_ackvec_len(av, av->dccpav_buf_head) <
265 DCCP_ACKVEC_LEN_MASK))
266 av->dccpav_buf[av->dccpav_buf_head]++;
267 else if (dccp_ackvec_set_buf_head_state(av, delta, state))
268 return -ENOBUFS;
269 } else {
271 * A.1.2. Old Packets
273 * When a packet with Sequence Number S <= buf_ackno
274 * arrives, the HC-Receiver will scan the table for
275 * the byte corresponding to S. (Indexing structures
276 * could reduce the complexity of this scan.)
278 u64 delta = dccp_delta_seqno(ackno, av->dccpav_buf_ackno);
279 u8 index = av->dccpav_buf_head;
281 while (1) {
282 const u8 len = dccp_ackvec_len(av, index);
283 const u8 state = dccp_ackvec_state(av, index);
285 * valid packets not yet in dccpav_buf have a reserved
286 * entry, with a len equal to 0.
288 if (state == DCCP_ACKVEC_STATE_NOT_RECEIVED &&
289 len == 0 && delta == 0) { /* Found our
290 reserved seat! */
291 dccp_pr_debug("Found %llu reserved seat!\n",
292 (unsigned long long)ackno);
293 av->dccpav_buf[index] = state;
294 goto out;
296 /* len == 0 means one packet */
297 if (delta < len + 1)
298 goto out_duplicate;
300 delta -= len + 1;
301 if (++index == DCCP_MAX_ACKVEC_LEN)
302 index = 0;
306 av->dccpav_buf_ackno = ackno;
307 dccp_timestamp(sk, &av->dccpav_time);
308 out:
309 return 0;
311 out_duplicate:
312 /* Duplicate packet */
313 dccp_pr_debug("Received a dup or already considered lost "
314 "packet: %llu\n", (unsigned long long)ackno);
315 return -EILSEQ;
318 #ifdef CONFIG_IP_DCCP_DEBUG
319 void dccp_ackvector_print(const u64 ackno, const unsigned char *vector, int len)
321 dccp_pr_debug_cat("ACK vector len=%d, ackno=%llu |", len,
322 (unsigned long long)ackno);
324 while (len--) {
325 const u8 state = (*vector & DCCP_ACKVEC_STATE_MASK) >> 6;
326 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
328 dccp_pr_debug_cat("%d,%d|", state, rl);
329 ++vector;
332 dccp_pr_debug_cat("\n");
335 void dccp_ackvec_print(const struct dccp_ackvec *av)
337 dccp_ackvector_print(av->dccpav_buf_ackno,
338 av->dccpav_buf + av->dccpav_buf_head,
339 av->dccpav_vec_len);
341 #endif
343 static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
344 struct dccp_ackvec_record *avr)
346 struct dccp_ackvec_record *next;
348 /* sort out vector length */
349 if (av->dccpav_buf_head <= avr->dccpavr_ack_ptr)
350 av->dccpav_vec_len = avr->dccpavr_ack_ptr - av->dccpav_buf_head;
351 else
352 av->dccpav_vec_len = DCCP_MAX_ACKVEC_LEN - 1
353 - av->dccpav_buf_head
354 + avr->dccpavr_ack_ptr;
356 /* free records */
357 list_for_each_entry_safe_from(avr, next, &av->dccpav_records,
358 dccpavr_node) {
359 list_del_init(&avr->dccpavr_node);
360 dccp_ackvec_record_delete(avr);
364 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
365 const u64 ackno)
367 struct dccp_ackvec_record *avr;
370 * If we traverse backwards, it should be faster when we have large
371 * windows. We will be receiving ACKs for stuff we sent a while back
372 * -sorbo.
374 list_for_each_entry_reverse(avr, &av->dccpav_records, dccpavr_node) {
375 if (ackno == avr->dccpavr_ack_seqno) {
376 dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
377 "ack_ackno=%llu, ACKED!\n",
378 dccp_role(sk), 1,
379 (unsigned long long)avr->dccpavr_ack_seqno,
380 (unsigned long long)avr->dccpavr_ack_ackno);
381 dccp_ackvec_throw_record(av, avr);
382 break;
383 } else if (avr->dccpavr_ack_seqno > ackno)
384 break; /* old news */
388 static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec *av,
389 struct sock *sk, u64 ackno,
390 const unsigned char len,
391 const unsigned char *vector)
393 unsigned char i;
394 struct dccp_ackvec_record *avr;
396 /* Check if we actually sent an ACK vector */
397 if (list_empty(&av->dccpav_records))
398 return;
400 i = len;
402 * XXX
403 * I think it might be more efficient to work backwards. See comment on
404 * rcv_ackno. -sorbo.
406 avr = list_entry(av->dccpav_records.next, struct dccp_ackvec_record,
407 dccpavr_node);
408 while (i--) {
409 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
410 u64 ackno_end_rl;
412 dccp_set_seqno(&ackno_end_rl, ackno - rl);
415 * If our AVR sequence number is greater than the ack, go
416 * forward in the AVR list until it is not so.
418 list_for_each_entry_from(avr, &av->dccpav_records,
419 dccpavr_node) {
420 if (!after48(avr->dccpavr_ack_seqno, ackno))
421 goto found;
423 /* End of the dccpav_records list, not found, exit */
424 break;
425 found:
426 if (between48(avr->dccpavr_ack_seqno, ackno_end_rl, ackno)) {
427 const u8 state = *vector & DCCP_ACKVEC_STATE_MASK;
428 if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED) {
429 dccp_pr_debug("%s ACK vector 0, len=%d, "
430 "ack_seqno=%llu, ack_ackno=%llu, "
431 "ACKED!\n",
432 dccp_role(sk), len,
433 (unsigned long long)
434 avr->dccpavr_ack_seqno,
435 (unsigned long long)
436 avr->dccpavr_ack_ackno);
437 dccp_ackvec_throw_record(av, avr);
438 break;
441 * If it wasn't received, continue scanning... we might
442 * find another one.
446 dccp_set_seqno(&ackno, ackno_end_rl - 1);
447 ++vector;
451 int dccp_ackvec_parse(struct sock *sk, const struct sk_buff *skb,
452 const u8 opt, const u8 *value, const u8 len)
454 if (len > DCCP_MAX_ACKVEC_LEN)
455 return -1;
457 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
458 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk)->dccps_hc_rx_ackvec, sk,
459 DCCP_SKB_CB(skb)->dccpd_ack_seq,
460 len, value);
461 return 0;
464 int __init dccp_ackvec_init(void)
466 dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
467 sizeof(struct dccp_ackvec), 0,
468 SLAB_HWCACHE_ALIGN, NULL, NULL);
469 if (dccp_ackvec_slab == NULL)
470 goto out_err;
472 dccp_ackvec_record_slab =
473 kmem_cache_create("dccp_ackvec_record",
474 sizeof(struct dccp_ackvec_record),
475 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
476 if (dccp_ackvec_record_slab == NULL)
477 goto out_destroy_slab;
479 return 0;
481 out_destroy_slab:
482 kmem_cache_destroy(dccp_ackvec_slab);
483 dccp_ackvec_slab = NULL;
484 out_err:
485 DCCP_CRIT("Unable to create Ack Vector slab cache");
486 return -ENOBUFS;
489 void dccp_ackvec_exit(void)
491 if (dccp_ackvec_slab != NULL) {
492 kmem_cache_destroy(dccp_ackvec_slab);
493 dccp_ackvec_slab = NULL;
495 if (dccp_ackvec_record_slab != NULL) {
496 kmem_cache_destroy(dccp_ackvec_record_slab);
497 dccp_ackvec_record_slab = NULL;