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 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);
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 #ifdef CONFIG_IP_DCCP_DEBUG
71 const char *debug_prefix = dp->dccps_role == DCCP_ROLE_CLIENT ?
72 "CLIENT tx: " : "server tx: ";
74 struct dccp_ackvec *av = dp->dccps_hc_rx_ackvec;
75 int len = av->dccpav_vec_len + 2;
78 unsigned char *to, *from;
79 struct dccp_ackvec_record *avr;
81 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
84 avr = dccp_ackvec_record_new();
88 dccp_timestamp(sk, &now);
89 elapsed_time = timeval_delta(&now, &av->dccpav_time) / 10;
91 if (elapsed_time != 0)
92 dccp_insert_option_elapsed_time(sk, skb, elapsed_time);
94 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
96 to = skb_push(skb, len);
97 *to++ = DCCPO_ACK_VECTOR_0;
100 len = av->dccpav_vec_len;
101 from = av->dccpav_buf + av->dccpav_buf_head;
103 /* Check if buf_head wraps */
104 if ((int)av->dccpav_buf_head + len > DCCP_MAX_ACKVEC_LEN) {
105 const u32 tailsize = DCCP_MAX_ACKVEC_LEN - av->dccpav_buf_head;
107 memcpy(to, from, tailsize);
110 from = av->dccpav_buf;
113 memcpy(to, from, len);
115 * From draft-ietf-dccp-spec-11.txt:
117 * For each acknowledgement it sends, the HC-Receiver will add an
118 * acknowledgement record. ack_seqno will equal the HC-Receiver
119 * sequence number it used for the ack packet; ack_ptr will equal
120 * buf_head; ack_ackno will equal buf_ackno; and ack_nonce will
123 avr->dccpavr_ack_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
124 avr->dccpavr_ack_ptr = av->dccpav_buf_head;
125 avr->dccpavr_ack_ackno = av->dccpav_buf_ackno;
126 avr->dccpavr_ack_nonce = av->dccpav_buf_nonce;
127 avr->dccpavr_sent_len = av->dccpav_vec_len;
129 dccp_ackvec_insert_avr(av, avr);
131 dccp_pr_debug("%sACK Vector 0, len=%d, ack_seqno=%llu, "
133 debug_prefix, avr->dccpavr_sent_len,
134 (unsigned long long)avr->dccpavr_ack_seqno,
135 (unsigned long long)avr->dccpavr_ack_ackno);
139 struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
141 struct dccp_ackvec *av = kmem_cache_alloc(dccp_ackvec_slab, priority);
144 av->dccpav_buf_head =
145 av->dccpav_buf_tail = DCCP_MAX_ACKVEC_LEN - 1;
146 av->dccpav_buf_ackno = DCCP_MAX_SEQNO + 1;
147 av->dccpav_buf_nonce = av->dccpav_buf_nonce = 0;
148 av->dccpav_ack_ptr = 0;
149 av->dccpav_time.tv_sec = 0;
150 av->dccpav_time.tv_usec = 0;
151 av->dccpav_sent_len = av->dccpav_vec_len = 0;
152 INIT_LIST_HEAD(&av->dccpav_records);
158 void dccp_ackvec_free(struct dccp_ackvec *av)
160 if (unlikely(av == NULL))
162 WARN_ON(!list_empty(&av->dccpav_records));
163 kmem_cache_free(dccp_ackvec_slab, av);
166 static inline u8 dccp_ackvec_state(const struct dccp_ackvec *av,
169 return av->dccpav_buf[index] & DCCP_ACKVEC_STATE_MASK;
172 static inline u8 dccp_ackvec_len(const struct dccp_ackvec *av,
175 return av->dccpav_buf[index] & DCCP_ACKVEC_LEN_MASK;
179 * If several packets are missing, the HC-Receiver may prefer to enter multiple
180 * bytes with run length 0, rather than a single byte with a larger run length;
181 * this simplifies table updates if one of the missing packets arrives.
183 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec *av,
184 const unsigned int packets,
185 const unsigned char state)
190 if (av->dccpav_vec_len + packets > DCCP_MAX_ACKVEC_LEN)
194 new_head = av->dccpav_buf_head - packets;
198 memset(av->dccpav_buf, DCCP_ACKVEC_STATE_NOT_RECEIVED,
202 new_head += DCCP_MAX_ACKVEC_LEN;
205 av->dccpav_buf_head = new_head;
208 memset(av->dccpav_buf + av->dccpav_buf_head + 1,
209 DCCP_ACKVEC_STATE_NOT_RECEIVED, gap);
211 av->dccpav_buf[av->dccpav_buf_head] = state;
212 av->dccpav_vec_len += packets;
217 * Implements the draft-ietf-dccp-spec-11.txt Appendix A
219 int dccp_ackvec_add(struct dccp_ackvec *av, const struct sock *sk,
220 const u64 ackno, const u8 state)
223 * Check at the right places if the buffer is full, if it is, tell the
224 * caller to start dropping packets till the HC-Sender acks our ACK
225 * vectors, when we will free up space in dccpav_buf.
227 * We may well decide to do buffer compression, etc, but for now lets
232 * Of course, the circular buffer may overflow, either when the
233 * HC-Sender is sending data at a very high rate, when the
234 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
235 * or when the HC-Sender is forgetting to acknowledge those acks
236 * (so the HC-Receiver is unable to clean up old state). In this
237 * case, the HC-Receiver should either compress the buffer (by
238 * increasing run lengths when possible), transfer its state to
239 * a larger buffer, or, as a last resort, drop all received
240 * packets, without processing them whatsoever, until its buffer
244 /* See if this is the first ackno being inserted */
245 if (av->dccpav_vec_len == 0) {
246 av->dccpav_buf[av->dccpav_buf_head] = state;
247 av->dccpav_vec_len = 1;
248 } else if (after48(ackno, av->dccpav_buf_ackno)) {
249 const u64 delta = dccp_delta_seqno(av->dccpav_buf_ackno,
253 * Look if the state of this packet is the same as the
254 * previous ackno and if so if we can bump the head len.
257 dccp_ackvec_state(av, av->dccpav_buf_head) == state &&
258 (dccp_ackvec_len(av, av->dccpav_buf_head) <
259 DCCP_ACKVEC_LEN_MASK))
260 av->dccpav_buf[av->dccpav_buf_head]++;
261 else if (dccp_ackvec_set_buf_head_state(av, delta, state))
267 * When a packet with Sequence Number S arrives, and
268 * S <= buf_ackno, the HC-Receiver will scan the table
269 * for the byte corresponding to S. (Indexing structures
270 * could reduce the complexity of this scan.)
272 u64 delta = dccp_delta_seqno(ackno, av->dccpav_buf_ackno);
273 u8 index = av->dccpav_buf_head;
276 const u8 len = dccp_ackvec_len(av, index);
277 const u8 state = dccp_ackvec_state(av, index);
279 * valid packets not yet in dccpav_buf have a reserved
280 * entry, with a len equal to 0.
282 if (state == DCCP_ACKVEC_STATE_NOT_RECEIVED &&
283 len == 0 && delta == 0) { /* Found our
285 dccp_pr_debug("Found %llu reserved seat!\n",
286 (unsigned long long)ackno);
287 av->dccpav_buf[index] = state;
290 /* len == 0 means one packet */
295 if (++index == DCCP_MAX_ACKVEC_LEN)
300 av->dccpav_buf_ackno = ackno;
301 dccp_timestamp(sk, &av->dccpav_time);
307 /* Duplicate packet */
308 dccp_pr_debug("Received a dup or already considered lost "
309 "packet: %llu\n", (unsigned long long)ackno);
313 #ifdef CONFIG_IP_DCCP_DEBUG
314 void dccp_ackvector_print(const u64 ackno, const unsigned char *vector, int len)
319 printk("ACK vector len=%d, ackno=%llu |", len,
320 (unsigned long long)ackno);
323 const u8 state = (*vector & DCCP_ACKVEC_STATE_MASK) >> 6;
324 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
326 printk("%d,%d|", state, rl);
333 void dccp_ackvec_print(const struct dccp_ackvec *av)
335 dccp_ackvector_print(av->dccpav_buf_ackno,
336 av->dccpav_buf + av->dccpav_buf_head,
341 static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
342 struct dccp_ackvec_record *avr)
344 struct dccp_ackvec_record *next;
346 av->dccpav_buf_tail = avr->dccpavr_ack_ptr - 1;
347 if (av->dccpav_buf_tail == 0)
348 av->dccpav_buf_tail = DCCP_MAX_ACKVEC_LEN - 1;
350 av->dccpav_vec_len -= avr->dccpavr_sent_len;
353 list_for_each_entry_safe_from(avr, next, &av->dccpav_records,
355 list_del_init(&avr->dccpavr_node);
356 dccp_ackvec_record_delete(avr);
360 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
363 struct dccp_ackvec_record *avr;
366 * If we traverse backwards, it should be faster when we have large
367 * windows. We will be receiving ACKs for stuff we sent a while back
370 list_for_each_entry_reverse(avr, &av->dccpav_records, dccpavr_node) {
371 if (ackno == avr->dccpavr_ack_seqno) {
372 #ifdef CONFIG_IP_DCCP_DEBUG
373 struct dccp_sock *dp = dccp_sk(sk);
374 const char *debug_prefix = dp->dccps_role == DCCP_ROLE_CLIENT ?
375 "CLIENT rx ack: " : "server rx ack: ";
377 dccp_pr_debug("%sACK packet 0, len=%d, ack_seqno=%llu, "
378 "ack_ackno=%llu, ACKED!\n",
380 (unsigned long long)avr->dccpavr_ack_seqno,
381 (unsigned long long)avr->dccpavr_ack_ackno);
382 dccp_ackvec_throw_record(av, avr);
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)
394 struct dccp_ackvec_record *avr;
396 /* Check if we actually sent an ACK vector */
397 if (list_empty(&av->dccpav_records))
403 * I think it might be more efficient to work backwards. See comment on
406 avr = list_entry(av->dccpav_records.next, struct dccp_ackvec_record,
409 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
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,
420 if (!after48(avr->dccpavr_ack_seqno, ackno))
423 /* End of the dccpav_records list, not found, exit */
426 if (between48(avr->dccpavr_ack_seqno, ackno_end_rl, ackno)) {
427 const u8 state = (*vector &
428 DCCP_ACKVEC_STATE_MASK) >> 6;
429 if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED) {
430 #ifdef CONFIG_IP_DCCP_DEBUG
431 struct dccp_sock *dp = dccp_sk(sk);
432 const char *debug_prefix =
433 dp->dccps_role == DCCP_ROLE_CLIENT ?
434 "CLIENT rx ack: " : "server rx ack: ";
436 dccp_pr_debug("%sACK vector 0, len=%d, "
437 "ack_seqno=%llu, ack_ackno=%llu, "
441 avr->dccpavr_ack_seqno,
443 avr->dccpavr_ack_ackno);
444 dccp_ackvec_throw_record(av, avr);
447 * If it wasn't received, continue scanning... we might
452 dccp_set_seqno(&ackno, ackno_end_rl - 1);
457 int dccp_ackvec_parse(struct sock *sk, const struct sk_buff *skb,
458 const u8 opt, const u8 *value, const u8 len)
460 if (len > DCCP_MAX_ACKVEC_LEN)
463 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
464 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk)->dccps_hc_rx_ackvec, sk,
465 DCCP_SKB_CB(skb)->dccpd_ack_seq,
470 static char dccp_ackvec_slab_msg[] __initdata =
471 KERN_CRIT "DCCP: Unable to create ack vectors slab caches\n";
473 int __init dccp_ackvec_init(void)
475 dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
476 sizeof(struct dccp_ackvec), 0,
477 SLAB_HWCACHE_ALIGN, NULL, NULL);
478 if (dccp_ackvec_slab == NULL)
481 dccp_ackvec_record_slab =
482 kmem_cache_create("dccp_ackvec_record",
483 sizeof(struct dccp_ackvec_record),
484 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
485 if (dccp_ackvec_record_slab == NULL)
486 goto out_destroy_slab;
491 kmem_cache_destroy(dccp_ackvec_slab);
492 dccp_ackvec_slab = NULL;
494 printk(dccp_ackvec_slab_msg);
498 void dccp_ackvec_exit(void)
500 if (dccp_ackvec_slab != NULL) {
501 kmem_cache_destroy(dccp_ackvec_slab);
502 dccp_ackvec_slab = NULL;
504 if (dccp_ackvec_record_slab != NULL) {
505 kmem_cache_destroy(dccp_ackvec_record_slab);
506 dccp_ackvec_record_slab = NULL;