blob: 25b7a8d1ad58eac67d2306e0c86ff8da7cea0b3b [file] [log] [blame]
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -07001/*
2 * net/dccp/ackvec.c
3 *
Gerrit Renkerf17a37c2010-11-10 21:20:07 +01004 * An implementation of Ack Vectors for the DCCP protocol
5 * Copyright (c) 2007 University of Aberdeen, Scotland, UK
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -07006 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
7 *
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;
11 */
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070012#include "dccp.h"
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -080013#include <linux/kernel.h>
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -080014#include <linux/slab.h>
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070015
Christoph Lametere18b8902006-12-06 20:33:20 -080016static struct kmem_cache *dccp_ackvec_slab;
17static struct kmem_cache *dccp_ackvec_record_slab;
Andrea Bittau02bcf282006-03-20 17:19:55 -080018
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010019struct dccp_ackvec *dccp_ackvec_alloc(const gfp_t priority)
Andrea Bittau02bcf282006-03-20 17:19:55 -080020{
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010021 struct dccp_ackvec *av = kmem_cache_zalloc(dccp_ackvec_slab, priority);
Andrea Bittau02bcf282006-03-20 17:19:55 -080022
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010023 if (av != NULL) {
Gerrit Renkerb3d14bf2010-11-10 21:21:35 +010024 av->av_buf_head = av->av_buf_tail = DCCPAV_MAX_ACKVEC_LEN - 1;
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010025 INIT_LIST_HEAD(&av->av_records);
26 }
27 return av;
Andrea Bittau02bcf282006-03-20 17:19:55 -080028}
29
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010030static void dccp_ackvec_purge_records(struct dccp_ackvec *av)
Andrea Bittau02bcf282006-03-20 17:19:55 -080031{
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010032 struct dccp_ackvec_record *cur, *next;
33
34 list_for_each_entry_safe(cur, next, &av->av_records, avr_node)
35 kmem_cache_free(dccp_ackvec_record_slab, cur);
36 INIT_LIST_HEAD(&av->av_records);
37}
38
39void dccp_ackvec_free(struct dccp_ackvec *av)
40{
41 if (likely(av != NULL)) {
42 dccp_ackvec_purge_records(av);
43 kmem_cache_free(dccp_ackvec_slab, av);
44 }
Andrea Bittau02bcf282006-03-20 17:19:55 -080045}
46
Gerrit Renker7d870932010-11-10 21:21:02 +010047/**
48 * dccp_ackvec_update_records - Record information about sent Ack Vectors
49 * @av: Ack Vector records to update
50 * @seqno: Sequence number of the packet carrying the Ack Vector just sent
51 * @nonce_sum: The sum of all buffer nonces contained in the Ack Vector
52 */
53int dccp_ackvec_update_records(struct dccp_ackvec *av, u64 seqno, u8 nonce_sum)
Andrea Bittau02bcf282006-03-20 17:19:55 -080054{
Andrea Bittau02bcf282006-03-20 17:19:55 -080055 struct dccp_ackvec_record *avr;
56
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010057 avr = kmem_cache_alloc(dccp_ackvec_record_slab, GFP_ATOMIC);
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -080058 if (avr == NULL)
Gerrit Renker7d870932010-11-10 21:21:02 +010059 return -ENOBUFS;
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070060
Gerrit Renker7d870932010-11-10 21:21:02 +010061 avr->avr_ack_seqno = seqno;
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010062 avr->avr_ack_ptr = av->av_buf_head;
63 avr->avr_ack_ackno = av->av_buf_ackno;
Gerrit Renker7d870932010-11-10 21:21:02 +010064 avr->avr_ack_nonce = nonce_sum;
Gerrit Renkerf17a37c2010-11-10 21:20:07 +010065 avr->avr_ack_runlen = dccp_ackvec_runlen(av->av_buf + av->av_buf_head);
Gerrit Renker7d870932010-11-10 21:21:02 +010066 /*
Gerrit Renkerb3d14bf2010-11-10 21:21:35 +010067 * When the buffer overflows, we keep no more than one record. This is
68 * the simplest way of disambiguating sender-Acks dating from before the
69 * overflow from sender-Acks which refer to after the overflow; a simple
70 * solution is preferable here since we are handling an exception.
71 */
72 if (av->av_overflow)
73 dccp_ackvec_purge_records(av);
74 /*
Gerrit Renker7d870932010-11-10 21:21:02 +010075 * Since GSS is incremented for each packet, the list is automatically
76 * arranged in descending order of @ack_seqno.
77 */
78 list_add(&avr->avr_node, &av->av_records);
Andrea Bittau02bcf282006-03-20 17:19:55 -080079
Gerrit Renker7d870932010-11-10 21:21:02 +010080 dccp_pr_debug("Added Vector, ack_seqno=%llu, ack_ackno=%llu (rl=%u)\n",
Gerrit Renkera47c5102007-12-30 04:19:31 -080081 (unsigned long long)avr->avr_ack_seqno,
Gerrit Renker7d870932010-11-10 21:21:02 +010082 (unsigned long long)avr->avr_ack_ackno,
83 avr->avr_ack_runlen);
Andrea Bittau02bcf282006-03-20 17:19:55 -080084 return 0;
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070085}
86
Gerrit Renker5753fdf2010-11-14 17:25:11 +010087static struct dccp_ackvec_record *dccp_ackvec_lookup(struct list_head *av_list,
88 const u64 ackno)
89{
90 struct dccp_ackvec_record *avr;
91 /*
92 * Exploit that records are inserted in descending order of sequence
93 * number, start with the oldest record first. If @ackno is `before'
94 * the earliest ack_ackno, the packet is too old to be considered.
95 */
96 list_for_each_entry_reverse(avr, av_list, avr_node) {
97 if (avr->avr_ack_seqno == ackno)
98 return avr;
99 if (before48(ackno, avr->avr_ack_seqno))
100 break;
101 }
102 return NULL;
103}
104
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -0700105/*
Gerrit Renkerb3d14bf2010-11-10 21:21:35 +0100106 * Buffer index and length computation using modulo-buffersize arithmetic.
107 * Note that, as pointers move from right to left, head is `before' tail.
108 */
109static inline u16 __ackvec_idx_add(const u16 a, const u16 b)
110{
111 return (a + b) % DCCPAV_MAX_ACKVEC_LEN;
112}
113
114static inline u16 __ackvec_idx_sub(const u16 a, const u16 b)
115{
116 return __ackvec_idx_add(a, DCCPAV_MAX_ACKVEC_LEN - b);
117}
118
119u16 dccp_ackvec_buflen(const struct dccp_ackvec *av)
120{
121 if (unlikely(av->av_overflow))
122 return DCCPAV_MAX_ACKVEC_LEN;
123 return __ackvec_idx_sub(av->av_buf_tail, av->av_buf_head);
124}
125
Gerrit Renker38024082010-11-14 17:25:23 +0100126/**
127 * dccp_ackvec_update_old - Update previous state as per RFC 4340, 11.4.1
128 * @av: non-empty buffer to update
129 * @distance: negative or zero distance of @seqno from buf_ackno downward
130 * @seqno: the (old) sequence number whose record is to be updated
131 * @state: state in which packet carrying @seqno was received
132 */
133static void dccp_ackvec_update_old(struct dccp_ackvec *av, s64 distance,
134 u64 seqno, enum dccp_ackvec_states state)
135{
136 u16 ptr = av->av_buf_head;
137
138 BUG_ON(distance > 0);
139 if (unlikely(dccp_ackvec_is_empty(av)))
140 return;
141
142 do {
143 u8 runlen = dccp_ackvec_runlen(av->av_buf + ptr);
144
145 if (distance + runlen >= 0) {
146 /*
147 * Only update the state if packet has not been received
148 * yet. This is OK as per the second table in RFC 4340,
149 * 11.4.1; i.e. here we are using the following table:
150 * RECEIVED
151 * 0 1 3
152 * S +---+---+---+
153 * T 0 | 0 | 0 | 0 |
154 * O +---+---+---+
155 * R 1 | 1 | 1 | 1 |
156 * E +---+---+---+
157 * D 3 | 0 | 1 | 3 |
158 * +---+---+---+
159 * The "Not Received" state was set by reserve_seats().
160 */
161 if (av->av_buf[ptr] == DCCPAV_NOT_RECEIVED)
162 av->av_buf[ptr] = state;
163 else
164 dccp_pr_debug("Not changing %llu state to %u\n",
165 (unsigned long long)seqno, state);
166 break;
167 }
168
169 distance += runlen + 1;
170 ptr = __ackvec_idx_add(ptr, 1);
171
172 } while (ptr != av->av_buf_tail);
173}
174
175/* Mark @num entries after buf_head as "Not yet received". */
176static void dccp_ackvec_reserve_seats(struct dccp_ackvec *av, u16 num)
177{
178 u16 start = __ackvec_idx_add(av->av_buf_head, 1),
179 len = DCCPAV_MAX_ACKVEC_LEN - start;
180
181 /* check for buffer wrap-around */
182 if (num > len) {
183 memset(av->av_buf + start, DCCPAV_NOT_RECEIVED, len);
184 start = 0;
185 num -= len;
186 }
187 if (num)
188 memset(av->av_buf + start, DCCPAV_NOT_RECEIVED, num);
189}
190
191/**
192 * dccp_ackvec_add_new - Record one or more new entries in Ack Vector buffer
193 * @av: container of buffer to update (can be empty or non-empty)
194 * @num_packets: number of packets to register (must be >= 1)
195 * @seqno: sequence number of the first packet in @num_packets
196 * @state: state in which packet carrying @seqno was received
197 */
198static void dccp_ackvec_add_new(struct dccp_ackvec *av, u32 num_packets,
199 u64 seqno, enum dccp_ackvec_states state)
200{
201 u32 num_cells = num_packets;
202
203 if (num_packets > DCCPAV_BURST_THRESH) {
204 u32 lost_packets = num_packets - 1;
205
206 DCCP_WARN("Warning: large burst loss (%u)\n", lost_packets);
207 /*
208 * We received 1 packet and have a loss of size "num_packets-1"
209 * which we squeeze into num_cells-1 rather than reserving an
210 * entire byte for each lost packet.
211 * The reason is that the vector grows in O(burst_length); when
212 * it grows too large there will no room left for the payload.
213 * This is a trade-off: if a few packets out of the burst show
214 * up later, their state will not be changed; it is simply too
215 * costly to reshuffle/reallocate/copy the buffer each time.
216 * Should such problems persist, we will need to switch to a
217 * different underlying data structure.
218 */
219 for (num_packets = num_cells = 1; lost_packets; ++num_cells) {
220 u8 len = min(lost_packets, (u32)DCCPAV_MAX_RUNLEN);
221
222 av->av_buf_head = __ackvec_idx_sub(av->av_buf_head, 1);
223 av->av_buf[av->av_buf_head] = DCCPAV_NOT_RECEIVED | len;
224
225 lost_packets -= len;
226 }
227 }
228
229 if (num_cells + dccp_ackvec_buflen(av) >= DCCPAV_MAX_ACKVEC_LEN) {
230 DCCP_CRIT("Ack Vector buffer overflow: dropping old entries\n");
231 av->av_overflow = true;
232 }
233
234 av->av_buf_head = __ackvec_idx_sub(av->av_buf_head, num_packets);
235 if (av->av_overflow)
236 av->av_buf_tail = av->av_buf_head;
237
238 av->av_buf[av->av_buf_head] = state;
239 av->av_buf_ackno = seqno;
240
241 if (num_packets > 1)
242 dccp_ackvec_reserve_seats(av, num_packets - 1);
243}
244
245/**
246 * dccp_ackvec_input - Register incoming packet in the buffer
247 */
248void dccp_ackvec_input(struct dccp_ackvec *av, struct sk_buff *skb)
249{
250 u64 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
251 enum dccp_ackvec_states state = DCCPAV_RECEIVED;
252
253 if (dccp_ackvec_is_empty(av)) {
254 dccp_ackvec_add_new(av, 1, seqno, state);
255 av->av_tail_ackno = seqno;
256
257 } else {
258 s64 num_packets = dccp_delta_seqno(av->av_buf_ackno, seqno);
259 u8 *current_head = av->av_buf + av->av_buf_head;
260
261 if (num_packets == 1 &&
262 dccp_ackvec_state(current_head) == state &&
263 dccp_ackvec_runlen(current_head) < DCCPAV_MAX_RUNLEN) {
264
265 *current_head += 1;
266 av->av_buf_ackno = seqno;
267
268 } else if (num_packets > 0) {
269 dccp_ackvec_add_new(av, num_packets, seqno, state);
270 } else {
271 dccp_ackvec_update_old(av, num_packets, seqno, state);
272 }
273 }
274}
275
Gerrit Renker5753fdf2010-11-14 17:25:11 +0100276/**
277 * dccp_ackvec_clear_state - Perform house-keeping / garbage-collection
278 * This routine is called when the peer acknowledges the receipt of Ack Vectors
279 * up to and including @ackno. While based on on section A.3 of RFC 4340, here
280 * are additional precautions to prevent corrupted buffer state. In particular,
281 * we use tail_ackno to identify outdated records; it always marks the earliest
282 * packet of group (2) in 11.4.2.
283 */
284void dccp_ackvec_clear_state(struct dccp_ackvec *av, const u64 ackno)
Gerrit Renkerf72f2f42010-11-18 09:37:07 -0800285{
Gerrit Renker5753fdf2010-11-14 17:25:11 +0100286 struct dccp_ackvec_record *avr, *next;
287 u8 runlen_now, eff_runlen;
288 s64 delta;
289
290 avr = dccp_ackvec_lookup(&av->av_records, ackno);
291 if (avr == NULL)
292 return;
293 /*
294 * Deal with outdated acknowledgments: this arises when e.g. there are
295 * several old records and the acks from the peer come in slowly. In
296 * that case we may still have records that pre-date tail_ackno.
297 */
298 delta = dccp_delta_seqno(av->av_tail_ackno, avr->avr_ack_ackno);
299 if (delta < 0)
300 goto free_records;
301 /*
302 * Deal with overlapping Ack Vectors: don't subtract more than the
303 * number of packets between tail_ackno and ack_ackno.
304 */
305 eff_runlen = delta < avr->avr_ack_runlen ? delta : avr->avr_ack_runlen;
306
307 runlen_now = dccp_ackvec_runlen(av->av_buf + avr->avr_ack_ptr);
308 /*
309 * The run length of Ack Vector cells does not decrease over time. If
310 * the run length is the same as at the time the Ack Vector was sent, we
311 * free the ack_ptr cell. That cell can however not be freed if the run
312 * length has increased: in this case we need to move the tail pointer
313 * backwards (towards higher indices), to its next-oldest neighbour.
314 */
315 if (runlen_now > eff_runlen) {
316
317 av->av_buf[avr->avr_ack_ptr] -= eff_runlen + 1;
318 av->av_buf_tail = __ackvec_idx_add(avr->avr_ack_ptr, 1);
319
320 /* This move may not have cleared the overflow flag. */
321 if (av->av_overflow)
322 av->av_overflow = (av->av_buf_head == av->av_buf_tail);
323 } else {
324 av->av_buf_tail = avr->avr_ack_ptr;
325 /*
326 * We have made sure that avr points to a valid cell within the
327 * buffer. This cell is either older than head, or equals head
328 * (empty buffer): in both cases we no longer have any overflow.
329 */
330 av->av_overflow = 0;
331 }
332
333 /*
334 * The peer has acknowledged up to and including ack_ackno. Hence the
335 * first packet in group (2) of 11.4.2 is the successor of ack_ackno.
336 */
337 av->av_tail_ackno = ADD48(avr->avr_ack_ackno, 1);
338
339free_records:
340 list_for_each_entry_safe_from(avr, next, &av->av_records, avr_node) {
341 list_del(&avr->avr_node);
342 kmem_cache_free(dccp_ackvec_record_slab, avr);
343 }
344}
345
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100346/*
347 * Routines to keep track of Ack Vectors received in an skb
348 */
349int dccp_ackvec_parsed_add(struct list_head *head, u8 *vec, u8 len, u8 nonce)
350{
351 struct dccp_ackvec_parsed *new = kmalloc(sizeof(*new), GFP_ATOMIC);
352
353 if (new == NULL)
354 return -ENOBUFS;
355 new->vec = vec;
356 new->len = len;
357 new->nonce = nonce;
358
359 list_add_tail(&new->node, head);
360 return 0;
361}
362EXPORT_SYMBOL_GPL(dccp_ackvec_parsed_add);
363
364void dccp_ackvec_parsed_cleanup(struct list_head *parsed_chunks)
365{
366 struct dccp_ackvec_parsed *cur, *next;
367
368 list_for_each_entry_safe(cur, next, parsed_chunks, node)
369 kfree(cur);
370 INIT_LIST_HEAD(parsed_chunks);
371}
372EXPORT_SYMBOL_GPL(dccp_ackvec_parsed_cleanup);
373
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -0800374int __init dccp_ackvec_init(void)
375{
376 dccp_ackvec_slab = kmem_cache_create("dccp_ackvec",
377 sizeof(struct dccp_ackvec), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +0900378 SLAB_HWCACHE_ALIGN, NULL);
Andrea Bittau02bcf282006-03-20 17:19:55 -0800379 if (dccp_ackvec_slab == NULL)
380 goto out_err;
381
Gerrit Renkerf17a37c2010-11-10 21:20:07 +0100382 dccp_ackvec_record_slab = kmem_cache_create("dccp_ackvec_record",
383 sizeof(struct dccp_ackvec_record),
384 0, SLAB_HWCACHE_ALIGN, NULL);
Andrea Bittau02bcf282006-03-20 17:19:55 -0800385 if (dccp_ackvec_record_slab == NULL)
386 goto out_destroy_slab;
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -0800387
388 return 0;
Andrea Bittau02bcf282006-03-20 17:19:55 -0800389
390out_destroy_slab:
391 kmem_cache_destroy(dccp_ackvec_slab);
392 dccp_ackvec_slab = NULL;
393out_err:
Gerrit Renker59348b12006-11-20 18:39:23 -0200394 DCCP_CRIT("Unable to create Ack Vector slab cache");
Andrea Bittau02bcf282006-03-20 17:19:55 -0800395 return -ENOBUFS;
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -0800396}
397
398void dccp_ackvec_exit(void)
399{
400 if (dccp_ackvec_slab != NULL) {
401 kmem_cache_destroy(dccp_ackvec_slab);
402 dccp_ackvec_slab = NULL;
403 }
Andrea Bittau02bcf282006-03-20 17:19:55 -0800404 if (dccp_ackvec_record_slab != NULL) {
405 kmem_cache_destroy(dccp_ackvec_record_slab);
406 dccp_ackvec_record_slab = NULL;
407 }
Arnaldo Carvalho de Melo9b07ef52006-03-20 17:16:17 -0800408}