Add linux-next specific files for 20110831
[linux-2.6/next.git] / net / dccp / ccids / lib / packet_history.c
blobde8fe294bf0bccfce4d865ea15590162e1225e06
1 /*
2 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
3 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
5 * An implementation of the DCCP protocol
7 * This code has been developed by the University of Waikato WAND
8 * research group. For further information please see http://www.wand.net.nz/
9 * or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
11 * This code also uses code from Lulea University, rereleased as GPL by its
12 * authors:
13 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
16 * and to make it work as a loadable module in the DCCP stack written by
17 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21 * This program is free software; you can redistribute it and/or modify
22 * it under the terms of the GNU General Public License as published by
23 * the Free Software Foundation; either version 2 of the License, or
24 * (at your option) any later version.
26 * This program is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * GNU General Public License for more details.
31 * You should have received a copy of the GNU General Public License
32 * along with this program; if not, write to the Free Software
33 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
36 #include <linux/string.h>
37 #include <linux/slab.h>
38 #include "packet_history.h"
39 #include "../../dccp.h"
42 * Transmitter History Routines
44 static struct kmem_cache *tfrc_tx_hist_slab;
46 int __init tfrc_tx_packet_history_init(void)
48 tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
49 sizeof(struct tfrc_tx_hist_entry),
50 0, SLAB_HWCACHE_ALIGN, NULL);
51 return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
54 void tfrc_tx_packet_history_exit(void)
56 if (tfrc_tx_hist_slab != NULL) {
57 kmem_cache_destroy(tfrc_tx_hist_slab);
58 tfrc_tx_hist_slab = NULL;
62 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
64 struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
66 if (entry == NULL)
67 return -ENOBUFS;
68 entry->seqno = seqno;
69 entry->stamp = ktime_get_real();
70 entry->next = *headp;
71 *headp = entry;
72 return 0;
75 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
77 struct tfrc_tx_hist_entry *head = *headp;
79 while (head != NULL) {
80 struct tfrc_tx_hist_entry *next = head->next;
82 kmem_cache_free(tfrc_tx_hist_slab, head);
83 head = next;
86 *headp = NULL;
90 * Receiver History Routines
92 static struct kmem_cache *tfrc_rx_hist_slab;
94 int __init tfrc_rx_packet_history_init(void)
96 tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
97 sizeof(struct tfrc_rx_hist_entry),
98 0, SLAB_HWCACHE_ALIGN, NULL);
99 return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
102 void tfrc_rx_packet_history_exit(void)
104 if (tfrc_rx_hist_slab != NULL) {
105 kmem_cache_destroy(tfrc_rx_hist_slab);
106 tfrc_rx_hist_slab = NULL;
110 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
111 const struct sk_buff *skb,
112 const u64 ndp)
114 const struct dccp_hdr *dh = dccp_hdr(skb);
116 entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
117 entry->tfrchrx_ccval = dh->dccph_ccval;
118 entry->tfrchrx_type = dh->dccph_type;
119 entry->tfrchrx_ndp = ndp;
120 entry->tfrchrx_tstamp = ktime_get_real();
123 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
124 const struct sk_buff *skb,
125 const u64 ndp)
127 struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
129 tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
132 /* has the packet contained in skb been seen before? */
133 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
135 const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
136 int i;
138 if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
139 return 1;
141 for (i = 1; i <= h->loss_count; i++)
142 if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
143 return 1;
145 return 0;
148 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
150 const u8 idx_a = tfrc_rx_hist_index(h, a),
151 idx_b = tfrc_rx_hist_index(h, b);
152 struct tfrc_rx_hist_entry *tmp = h->ring[idx_a];
154 h->ring[idx_a] = h->ring[idx_b];
155 h->ring[idx_b] = tmp;
159 * Private helper functions for loss detection.
161 * In the descriptions, `Si' refers to the sequence number of entry number i,
162 * whose NDP count is `Ni' (lower case is used for variables).
163 * Note: All __xxx_loss functions expect that a test against duplicates has been
164 * performed already: the seqno of the skb must not be less than the seqno
165 * of loss_prev; and it must not equal that of any valid history entry.
167 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
169 u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
170 s1 = DCCP_SKB_CB(skb)->dccpd_seq;
172 if (!dccp_loss_free(s0, s1, n1)) { /* gap between S0 and S1 */
173 h->loss_count = 1;
174 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
178 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
180 u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
181 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
182 s2 = DCCP_SKB_CB(skb)->dccpd_seq;
184 if (likely(dccp_delta_seqno(s1, s2) > 0)) { /* S1 < S2 */
185 h->loss_count = 2;
186 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
187 return;
190 /* S0 < S2 < S1 */
192 if (dccp_loss_free(s0, s2, n2)) {
193 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
195 if (dccp_loss_free(s2, s1, n1)) {
196 /* hole is filled: S0, S2, and S1 are consecutive */
197 h->loss_count = 0;
198 h->loss_start = tfrc_rx_hist_index(h, 1);
199 } else
200 /* gap between S2 and S1: just update loss_prev */
201 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
203 } else { /* gap between S0 and S2 */
205 * Reorder history to insert S2 between S0 and S1
207 tfrc_rx_hist_swap(h, 0, 3);
208 h->loss_start = tfrc_rx_hist_index(h, 3);
209 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
210 h->loss_count = 2;
214 /* return 1 if a new loss event has been identified */
215 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
217 u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
218 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
219 s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
220 s3 = DCCP_SKB_CB(skb)->dccpd_seq;
222 if (likely(dccp_delta_seqno(s2, s3) > 0)) { /* S2 < S3 */
223 h->loss_count = 3;
224 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
225 return 1;
228 /* S3 < S2 */
230 if (dccp_delta_seqno(s1, s3) > 0) { /* S1 < S3 < S2 */
232 * Reorder history to insert S3 between S1 and S2
234 tfrc_rx_hist_swap(h, 2, 3);
235 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
236 h->loss_count = 3;
237 return 1;
240 /* S0 < S3 < S1 */
242 if (dccp_loss_free(s0, s3, n3)) {
243 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
245 if (dccp_loss_free(s3, s1, n1)) {
246 /* hole between S0 and S1 filled by S3 */
247 u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
249 if (dccp_loss_free(s1, s2, n2)) {
250 /* entire hole filled by S0, S3, S1, S2 */
251 h->loss_start = tfrc_rx_hist_index(h, 2);
252 h->loss_count = 0;
253 } else {
254 /* gap remains between S1 and S2 */
255 h->loss_start = tfrc_rx_hist_index(h, 1);
256 h->loss_count = 1;
259 } else /* gap exists between S3 and S1, loss_count stays at 2 */
260 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
262 return 0;
266 * The remaining case: S0 < S3 < S1 < S2; gap between S0 and S3
267 * Reorder history to insert S3 between S0 and S1.
269 tfrc_rx_hist_swap(h, 0, 3);
270 h->loss_start = tfrc_rx_hist_index(h, 3);
271 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
272 h->loss_count = 3;
274 return 1;
277 /* recycle RX history records to continue loss detection if necessary */
278 static void __three_after_loss(struct tfrc_rx_hist *h)
281 * At this stage we know already that there is a gap between S0 and S1
282 * (since S0 was the highest sequence number received before detecting
283 * the loss). To recycle the loss record, it is thus only necessary to
284 * check for other possible gaps between S1/S2 and between S2/S3.
286 u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
287 s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
288 s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
289 u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
290 n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
292 if (dccp_loss_free(s1, s2, n2)) {
294 if (dccp_loss_free(s2, s3, n3)) {
295 /* no gap between S2 and S3: entire hole is filled */
296 h->loss_start = tfrc_rx_hist_index(h, 3);
297 h->loss_count = 0;
298 } else {
299 /* gap between S2 and S3 */
300 h->loss_start = tfrc_rx_hist_index(h, 2);
301 h->loss_count = 1;
304 } else { /* gap between S1 and S2 */
305 h->loss_start = tfrc_rx_hist_index(h, 1);
306 h->loss_count = 2;
311 * tfrc_rx_handle_loss - Loss detection and further processing
312 * @h: The non-empty RX history object
313 * @lh: Loss Intervals database to update
314 * @skb: Currently received packet
315 * @ndp: The NDP count belonging to @skb
316 * @calc_first_li: Caller-dependent computation of first loss interval in @lh
317 * @sk: Used by @calc_first_li (see tfrc_lh_interval_add)
318 * Chooses action according to pending loss, updates LI database when a new
319 * loss was detected, and does required post-processing. Returns 1 when caller
320 * should send feedback, 0 otherwise.
321 * Since it also takes care of reordering during loss detection and updates the
322 * records accordingly, the caller should not perform any more RX history
323 * operations when loss_count is greater than 0 after calling this function.
325 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
326 struct tfrc_loss_hist *lh,
327 struct sk_buff *skb, const u64 ndp,
328 u32 (*calc_first_li)(struct sock *), struct sock *sk)
330 int is_new_loss = 0;
332 if (h->loss_count == 0) {
333 __do_track_loss(h, skb, ndp);
334 } else if (h->loss_count == 1) {
335 __one_after_loss(h, skb, ndp);
336 } else if (h->loss_count != 2) {
337 DCCP_BUG("invalid loss_count %d", h->loss_count);
338 } else if (__two_after_loss(h, skb, ndp)) {
340 * Update Loss Interval database and recycle RX records
342 is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
343 __three_after_loss(h);
345 return is_new_loss;
348 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
350 int i;
352 for (i = 0; i <= TFRC_NDUPACK; i++) {
353 h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
354 if (h->ring[i] == NULL)
355 goto out_free;
358 h->loss_count = h->loss_start = 0;
359 return 0;
361 out_free:
362 while (i-- != 0) {
363 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
364 h->ring[i] = NULL;
366 return -ENOBUFS;
369 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
371 int i;
373 for (i = 0; i <= TFRC_NDUPACK; ++i)
374 if (h->ring[i] != NULL) {
375 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
376 h->ring[i] = NULL;
381 * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
383 static inline struct tfrc_rx_hist_entry *
384 tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
386 return h->ring[0];
390 * tfrc_rx_hist_rtt_prev_s: previously suitable (wrt rtt_last_s) RTT-sampling entry
392 static inline struct tfrc_rx_hist_entry *
393 tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
395 return h->ring[h->rtt_sample_prev];
399 * tfrc_rx_hist_sample_rtt - Sample RTT from timestamp / CCVal
400 * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
401 * to compute a sample with given data - calling function should check this.
403 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
405 u32 sample = 0,
406 delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
407 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
409 if (delta_v < 1 || delta_v > 4) { /* unsuitable CCVal delta */
410 if (h->rtt_sample_prev == 2) { /* previous candidate stored */
411 sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
412 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
413 if (sample)
414 sample = 4 / sample *
415 ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
416 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
417 else /*
418 * FIXME: This condition is in principle not
419 * possible but occurs when CCID is used for
420 * two-way data traffic. I have tried to trace
421 * it, but the cause does not seem to be here.
423 DCCP_BUG("please report to dccp@vger.kernel.org"
424 " => prev = %u, last = %u",
425 tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
426 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
427 } else if (delta_v < 1) {
428 h->rtt_sample_prev = 1;
429 goto keep_ref_for_next_time;
432 } else if (delta_v == 4) /* optimal match */
433 sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
434 else { /* suboptimal match */
435 h->rtt_sample_prev = 2;
436 goto keep_ref_for_next_time;
439 if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
440 DCCP_WARN("RTT sample %u too large, using max\n", sample);
441 sample = DCCP_SANE_RTT_MAX;
444 h->rtt_sample_prev = 0; /* use current entry as next reference */
445 keep_ref_for_next_time:
447 return sample;