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,
96 * Buffer index and length computation using modulo-buffersize arithmetic.
97 * Note that, as pointers move from right to left, head is `before' tail.
99 static inline u16 __ackvec_idx_add(const u16 a, const u16 b)
101 return (a + b) % DCCPAV_MAX_ACKVEC_LEN;
104 static inline u16 __ackvec_idx_sub(const u16 a, const u16 b)
106 return __ackvec_idx_add(a, DCCPAV_MAX_ACKVEC_LEN - b);
109 u16 dccp_ackvec_buflen(const struct dccp_ackvec *av)
111 if (unlikely(av->av_overflow))
112 return DCCPAV_MAX_ACKVEC_LEN;
113 return __ackvec_idx_sub(av->av_buf_tail, av->av_buf_head);
117 * If several packets are missing, the HC-Receiver may prefer to enter multiple
118 * bytes with run length 0, rather than a single byte with a larger run length;
119 * this simplifies table updates if one of the missing packets arrives.
121 static inline int dccp_ackvec_set_buf_head_state(struct dccp_ackvec *av,
122 const unsigned int packets,
123 const unsigned char state)
128 if (av->av_vec_len + packets > DCCPAV_MAX_ACKVEC_LEN)
132 new_head = av->av_buf_head - packets;
136 memset(av->av_buf, DCCPAV_NOT_RECEIVED,
140 new_head += DCCPAV_MAX_ACKVEC_LEN;
143 av->av_buf_head = new_head;
146 memset(av->av_buf + av->av_buf_head + 1,
147 DCCPAV_NOT_RECEIVED, gap);
149 av->av_buf[av->av_buf_head] = state;
150 av->av_vec_len += packets;
155 * Implements the RFC 4340, Appendix A
157 int dccp_ackvec_add(struct dccp_ackvec *av, const struct sock *sk,
158 const u64 ackno, const u8 state)
160 u8 *cur_head = av->av_buf + av->av_buf_head,
161 *buf_end = av->av_buf + DCCPAV_MAX_ACKVEC_LEN;
163 * Check at the right places if the buffer is full, if it is, tell the
164 * caller to start dropping packets till the HC-Sender acks our ACK
165 * vectors, when we will free up space in av_buf.
167 * We may well decide to do buffer compression, etc, but for now lets
170 * From Appendix A.1.1 (`New Packets'):
172 * Of course, the circular buffer may overflow, either when the
173 * HC-Sender is sending data at a very high rate, when the
174 * HC-Receiver's acknowledgements are not reaching the HC-Sender,
175 * or when the HC-Sender is forgetting to acknowledge those acks
176 * (so the HC-Receiver is unable to clean up old state). In this
177 * case, the HC-Receiver should either compress the buffer (by
178 * increasing run lengths when possible), transfer its state to
179 * a larger buffer, or, as a last resort, drop all received
180 * packets, without processing them whatsoever, until its buffer
184 /* See if this is the first ackno being inserted */
185 if (av->av_vec_len == 0) {
188 } else if (after48(ackno, av->av_buf_ackno)) {
189 const u64 delta = dccp_delta_seqno(av->av_buf_ackno, ackno);
192 * Look if the state of this packet is the same as the
193 * previous ackno and if so if we can bump the head len.
195 if (delta == 1 && dccp_ackvec_state(cur_head) == state &&
196 dccp_ackvec_runlen(cur_head) < DCCPAV_MAX_RUNLEN)
198 else if (dccp_ackvec_set_buf_head_state(av, delta, state))
204 * When a packet with Sequence Number S <= buf_ackno
205 * arrives, the HC-Receiver will scan the table for
206 * the byte corresponding to S. (Indexing structures
207 * could reduce the complexity of this scan.)
209 u64 delta = dccp_delta_seqno(ackno, av->av_buf_ackno);
212 const u8 len = dccp_ackvec_runlen(cur_head);
214 * valid packets not yet in av_buf have a reserved
215 * entry, with a len equal to 0.
217 if (*cur_head == DCCPAV_NOT_RECEIVED && delta == 0) {
218 dccp_pr_debug("Found %llu reserved seat!\n",
219 (unsigned long long)ackno);
223 /* len == 0 means one packet */
228 if (++cur_head == buf_end)
229 cur_head = av->av_buf;
233 av->av_buf_ackno = ackno;
238 /* Duplicate packet */
239 dccp_pr_debug("Received a dup or already considered lost "
240 "packet: %llu\n", (unsigned long long)ackno);
244 static void dccp_ackvec_throw_record(struct dccp_ackvec *av,
245 struct dccp_ackvec_record *avr)
247 struct dccp_ackvec_record *next;
249 /* sort out vector length */
250 if (av->av_buf_head <= avr->avr_ack_ptr)
251 av->av_vec_len = avr->avr_ack_ptr - av->av_buf_head;
253 av->av_vec_len = DCCPAV_MAX_ACKVEC_LEN - 1 -
254 av->av_buf_head + avr->avr_ack_ptr;
257 list_for_each_entry_safe_from(avr, next, &av->av_records, avr_node) {
258 list_del(&avr->avr_node);
259 kmem_cache_free(dccp_ackvec_record_slab, avr);
263 void dccp_ackvec_check_rcv_ackno(struct dccp_ackvec *av, struct sock *sk,
266 struct dccp_ackvec_record *avr;
269 * If we traverse backwards, it should be faster when we have large
270 * windows. We will be receiving ACKs for stuff we sent a while back
273 list_for_each_entry_reverse(avr, &av->av_records, avr_node) {
274 if (ackno == avr->avr_ack_seqno) {
275 dccp_pr_debug("%s ACK packet 0, len=%d, ack_seqno=%llu, "
276 "ack_ackno=%llu, ACKED!\n",
277 dccp_role(sk), avr->avr_ack_runlen,
278 (unsigned long long)avr->avr_ack_seqno,
279 (unsigned long long)avr->avr_ack_ackno);
280 dccp_ackvec_throw_record(av, avr);
282 } else if (avr->avr_ack_seqno > ackno)
283 break; /* old news */
287 static void dccp_ackvec_check_rcv_ackvector(struct dccp_ackvec *av,
288 struct sock *sk, u64 *ackno,
289 const unsigned char len,
290 const unsigned char *vector)
293 struct dccp_ackvec_record *avr;
295 /* Check if we actually sent an ACK vector */
296 if (list_empty(&av->av_records))
302 * I think it might be more efficient to work backwards. See comment on
305 avr = list_entry(av->av_records.next, struct dccp_ackvec_record, avr_node);
307 const u8 rl = dccp_ackvec_runlen(vector);
310 dccp_set_seqno(&ackno_end_rl, *ackno - rl);
313 * If our AVR sequence number is greater than the ack, go
314 * forward in the AVR list until it is not so.
316 list_for_each_entry_from(avr, &av->av_records, avr_node) {
317 if (!after48(avr->avr_ack_seqno, *ackno))
320 /* End of the av_records list, not found, exit */
323 if (between48(avr->avr_ack_seqno, ackno_end_rl, *ackno)) {
324 if (dccp_ackvec_state(vector) != DCCPAV_NOT_RECEIVED) {
325 dccp_pr_debug("%s ACK vector 0, len=%d, "
326 "ack_seqno=%llu, ack_ackno=%llu, "
333 dccp_ackvec_throw_record(av, avr);
337 * If it wasn't received, continue scanning... we might
342 dccp_set_seqno(ackno, ackno_end_rl - 1);
347 int dccp_ackvec_parse(struct sock *sk, const struct sk_buff *skb,
348 u64 *ackno, const u8 opt, const u8 *value, const u8 len)
350 if (len > DCCP_SINGLE_OPT_MAXLEN)
353 /* dccp_ackvector_print(DCCP_SKB_CB(skb)->dccpd_ack_seq, value, len); */
354 dccp_ackvec_check_rcv_ackvector(dccp_sk(sk)->dccps_hc_rx_ackvec, sk,
359 int __init dccp_ackvec_init(void)
361 dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
362 sizeof(struct dccp_ackvec), 0,
363 SLAB_HWCACHE_ALIGN, NULL);
364 if (dccp_ackvec_slab == NULL)
367 dccp_ackvec_record_slab = kmem_cache_create("dccp_ackvec_record",
368 sizeof(struct dccp_ackvec_record),
369 0, SLAB_HWCACHE_ALIGN, NULL);
370 if (dccp_ackvec_record_slab == NULL)
371 goto out_destroy_slab;
376 kmem_cache_destroy(dccp_ackvec_slab);
377 dccp_ackvec_slab = NULL;
379 DCCP_CRIT("Unable to create Ack Vector slab cache");
383 void dccp_ackvec_exit(void)
385 if (dccp_ackvec_slab != NULL) {
386 kmem_cache_destroy(dccp_ackvec_slab);
387 dccp_ackvec_slab = NULL;
389 if (dccp_ackvec_record_slab != NULL) {
390 kmem_cache_destroy(dccp_ackvec_record_slab);
391 dccp_ackvec_record_slab = NULL;