2 * net/dccp/packet_history.c
4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
7 * An implementation of the DCCP protocol
9 * This code has been developed by the University of Waikato WAND
10 * research group. For further information please see http://www.wand.net.nz/
11 * or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
13 * This code also uses code from Lulea University, rereleased as GPL by its
15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18 * and to make it work as a loadable module in the DCCP stack written by
19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
33 * You should have received a copy of the GNU General Public License
34 * along with this program; if not, write to the Free Software
35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include "packet_history.h"
41 #include "../../dccp.h"
44 * Transmitter History Routines
46 static struct kmem_cache
*tfrc_tx_hist_slab
;
48 int __init
tfrc_tx_packet_history_init(void)
50 tfrc_tx_hist_slab
= kmem_cache_create("tfrc_tx_hist",
51 sizeof(struct tfrc_tx_hist_entry
),
52 0, SLAB_HWCACHE_ALIGN
, NULL
);
53 return tfrc_tx_hist_slab
== NULL
? -ENOBUFS
: 0;
56 void tfrc_tx_packet_history_exit(void)
58 if (tfrc_tx_hist_slab
!= NULL
) {
59 kmem_cache_destroy(tfrc_tx_hist_slab
);
60 tfrc_tx_hist_slab
= NULL
;
64 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry
**headp
, u64 seqno
)
66 struct tfrc_tx_hist_entry
*entry
= kmem_cache_alloc(tfrc_tx_hist_slab
, gfp_any());
71 entry
->stamp
= ktime_get_real();
76 EXPORT_SYMBOL_GPL(tfrc_tx_hist_add
);
78 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry
**headp
)
80 struct tfrc_tx_hist_entry
*head
= *headp
;
82 while (head
!= NULL
) {
83 struct tfrc_tx_hist_entry
*next
= head
->next
;
85 kmem_cache_free(tfrc_tx_hist_slab
, head
);
91 EXPORT_SYMBOL_GPL(tfrc_tx_hist_purge
);
94 * Receiver History Routines
96 static struct kmem_cache
*tfrc_rx_hist_slab
;
98 int __init
tfrc_rx_packet_history_init(void)
100 tfrc_rx_hist_slab
= kmem_cache_create("tfrc_rxh_cache",
101 sizeof(struct tfrc_rx_hist_entry
),
102 0, SLAB_HWCACHE_ALIGN
, NULL
);
103 return tfrc_rx_hist_slab
== NULL
? -ENOBUFS
: 0;
106 void tfrc_rx_packet_history_exit(void)
108 if (tfrc_rx_hist_slab
!= NULL
) {
109 kmem_cache_destroy(tfrc_rx_hist_slab
);
110 tfrc_rx_hist_slab
= NULL
;
114 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry
*entry
,
115 const struct sk_buff
*skb
,
118 const struct dccp_hdr
*dh
= dccp_hdr(skb
);
120 entry
->tfrchrx_seqno
= DCCP_SKB_CB(skb
)->dccpd_seq
;
121 entry
->tfrchrx_ccval
= dh
->dccph_ccval
;
122 entry
->tfrchrx_type
= dh
->dccph_type
;
123 entry
->tfrchrx_ndp
= ndp
;
124 entry
->tfrchrx_tstamp
= ktime_get_real();
127 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist
*h
,
128 const struct sk_buff
*skb
,
131 struct tfrc_rx_hist_entry
*entry
= tfrc_rx_hist_last_rcv(h
);
133 tfrc_rx_hist_entry_from_skb(entry
, skb
, ndp
);
135 EXPORT_SYMBOL_GPL(tfrc_rx_hist_add_packet
);
137 /* has the packet contained in skb been seen before? */
138 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist
*h
, struct sk_buff
*skb
)
140 const u64 seq
= DCCP_SKB_CB(skb
)->dccpd_seq
;
143 if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h
)->tfrchrx_seqno
, seq
) <= 0)
146 for (i
= 1; i
<= h
->loss_count
; i
++)
147 if (tfrc_rx_hist_entry(h
, i
)->tfrchrx_seqno
== seq
)
152 EXPORT_SYMBOL_GPL(tfrc_rx_hist_duplicate
);
155 static void __tfrc_rx_hist_swap(struct tfrc_rx_hist
*h
, const u8 a
, const u8 b
)
157 struct tfrc_rx_hist_entry
*tmp
= h
->ring
[a
];
159 h
->ring
[a
] = h
->ring
[b
];
163 static void tfrc_rx_hist_swap(struct tfrc_rx_hist
*h
, const u8 a
, const u8 b
)
165 __tfrc_rx_hist_swap(h
, tfrc_rx_hist_index(h
, a
),
166 tfrc_rx_hist_index(h
, b
));
170 * tfrc_rx_hist_resume_rtt_sampling - Prepare RX history for RTT sampling
171 * This is called after loss detection has finished, when the history entry
172 * with the index of `loss_count' holds the highest-received sequence number.
173 * RTT sampling requires this information at ring[0] (tfrc_rx_hist_sample_rtt).
175 static inline void tfrc_rx_hist_resume_rtt_sampling(struct tfrc_rx_hist
*h
)
177 __tfrc_rx_hist_swap(h
, 0, tfrc_rx_hist_index(h
, h
->loss_count
));
178 h
->loss_count
= h
->loss_start
= 0;
182 * Private helper functions for loss detection.
184 * In the descriptions, `Si' refers to the sequence number of entry number i,
185 * whose NDP count is `Ni' (lower case is used for variables).
186 * Note: All __xxx_loss functions expect that a test against duplicates has been
187 * performed already: the seqno of the skb must not be less than the seqno
188 * of loss_prev; and it must not equal that of any valid history entry.
190 static void __do_track_loss(struct tfrc_rx_hist
*h
, struct sk_buff
*skb
, u64 n1
)
192 u64 s0
= tfrc_rx_hist_loss_prev(h
)->tfrchrx_seqno
,
193 s1
= DCCP_SKB_CB(skb
)->dccpd_seq
;
195 if (!dccp_loss_free(s0
, s1
, n1
)) { /* gap between S0 and S1 */
197 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 1), skb
, n1
);
201 static void __one_after_loss(struct tfrc_rx_hist
*h
, struct sk_buff
*skb
, u32 n2
)
203 u64 s0
= tfrc_rx_hist_loss_prev(h
)->tfrchrx_seqno
,
204 s1
= tfrc_rx_hist_entry(h
, 1)->tfrchrx_seqno
,
205 s2
= DCCP_SKB_CB(skb
)->dccpd_seq
;
207 if (likely(dccp_delta_seqno(s1
, s2
) > 0)) { /* S1 < S2 */
209 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 2), skb
, n2
);
215 if (dccp_loss_free(s0
, s2
, n2
)) {
216 u64 n1
= tfrc_rx_hist_entry(h
, 1)->tfrchrx_ndp
;
218 if (dccp_loss_free(s2
, s1
, n1
)) {
219 /* hole is filled: S0, S2, and S1 are consecutive */
220 tfrc_rx_hist_resume_rtt_sampling(h
);
222 /* gap between S2 and S1: just update loss_prev */
223 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h
), skb
, n2
);
225 } else { /* gap between S0 and S2 */
227 * Reorder history to insert S2 between S0 and S1
229 tfrc_rx_hist_swap(h
, 0, 3);
230 h
->loss_start
= tfrc_rx_hist_index(h
, 3);
231 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 1), skb
, n2
);
236 /* return 1 if a new loss event has been identified */
237 static int __two_after_loss(struct tfrc_rx_hist
*h
, struct sk_buff
*skb
, u32 n3
)
239 u64 s0
= tfrc_rx_hist_loss_prev(h
)->tfrchrx_seqno
,
240 s1
= tfrc_rx_hist_entry(h
, 1)->tfrchrx_seqno
,
241 s2
= tfrc_rx_hist_entry(h
, 2)->tfrchrx_seqno
,
242 s3
= DCCP_SKB_CB(skb
)->dccpd_seq
;
244 if (likely(dccp_delta_seqno(s2
, s3
) > 0)) { /* S2 < S3 */
246 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 3), skb
, n3
);
252 if (dccp_delta_seqno(s1
, s3
) > 0) { /* S1 < S3 < S2 */
254 * Reorder history to insert S3 between S1 and S2
256 tfrc_rx_hist_swap(h
, 2, 3);
257 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 2), skb
, n3
);
264 if (dccp_loss_free(s0
, s3
, n3
)) {
265 u64 n1
= tfrc_rx_hist_entry(h
, 1)->tfrchrx_ndp
;
267 if (dccp_loss_free(s3
, s1
, n1
)) {
268 /* hole between S0 and S1 filled by S3 */
269 u64 n2
= tfrc_rx_hist_entry(h
, 2)->tfrchrx_ndp
;
271 if (dccp_loss_free(s1
, s2
, n2
)) {
272 /* entire hole filled by S0, S3, S1, S2 */
273 tfrc_rx_hist_resume_rtt_sampling(h
);
275 /* gap remains between S1 and S2 */
276 h
->loss_start
= tfrc_rx_hist_index(h
, 1);
280 } else /* gap exists between S3 and S1, loss_count stays at 2 */
281 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h
), skb
, n3
);
287 * The remaining case: S0 < S3 < S1 < S2; gap between S0 and S3
288 * Reorder history to insert S3 between S0 and S1.
290 tfrc_rx_hist_swap(h
, 0, 3);
291 h
->loss_start
= tfrc_rx_hist_index(h
, 3);
292 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h
, 1), skb
, n3
);
298 /* recycle RX history records to continue loss detection if necessary */
299 static void __three_after_loss(struct tfrc_rx_hist
*h
)
302 * At this stage we know already that there is a gap between S0 and S1
303 * (since S0 was the highest sequence number received before detecting
304 * the loss). To recycle the loss record, it is thus only necessary to
305 * check for other possible gaps between S1/S2 and between S2/S3.
307 u64 s1
= tfrc_rx_hist_entry(h
, 1)->tfrchrx_seqno
,
308 s2
= tfrc_rx_hist_entry(h
, 2)->tfrchrx_seqno
,
309 s3
= tfrc_rx_hist_entry(h
, 3)->tfrchrx_seqno
;
310 u64 n2
= tfrc_rx_hist_entry(h
, 2)->tfrchrx_ndp
,
311 n3
= tfrc_rx_hist_entry(h
, 3)->tfrchrx_ndp
;
313 if (dccp_loss_free(s1
, s2
, n2
)) {
315 if (dccp_loss_free(s2
, s3
, n3
)) {
316 /* no gap between S2 and S3: entire hole is filled */
317 tfrc_rx_hist_resume_rtt_sampling(h
);
319 /* gap between S2 and S3 */
320 h
->loss_start
= tfrc_rx_hist_index(h
, 2);
324 } else { /* gap between S1 and S2 */
325 h
->loss_start
= tfrc_rx_hist_index(h
, 1);
331 * tfrc_rx_handle_loss - Loss detection and further processing
332 * @h: The non-empty RX history object
333 * @lh: Loss Intervals database to update
334 * @skb: Currently received packet
335 * @ndp: The NDP count belonging to @skb
336 * @calc_first_li: Caller-dependent computation of first loss interval in @lh
337 * @sk: Used by @calc_first_li (see tfrc_lh_interval_add)
338 * Chooses action according to pending loss, updates LI database when a new
339 * loss was detected, and does required post-processing. Returns 1 when caller
340 * should send feedback, 0 otherwise.
341 * Since it also takes care of reordering during loss detection and updates the
342 * records accordingly, the caller should not perform any more RX history
343 * operations when loss_count is greater than 0 after calling this function.
345 int tfrc_rx_handle_loss(struct tfrc_rx_hist
*h
,
346 struct tfrc_loss_hist
*lh
,
347 struct sk_buff
*skb
, const u64 ndp
,
348 u32 (*calc_first_li
)(struct sock
*), struct sock
*sk
)
352 if (tfrc_rx_hist_duplicate(h
, skb
))
355 if (h
->loss_count
== 0) {
356 __do_track_loss(h
, skb
, ndp
);
357 tfrc_rx_hist_sample_rtt(h
, skb
);
358 } else if (h
->loss_count
== 1) {
359 __one_after_loss(h
, skb
, ndp
);
360 } else if (h
->loss_count
!= 2) {
361 DCCP_BUG("invalid loss_count %d", h
->loss_count
);
362 } else if (__two_after_loss(h
, skb
, ndp
)) {
364 * Update Loss Interval database and recycle RX records
366 is_new_loss
= tfrc_lh_interval_add(lh
, h
, calc_first_li
, sk
);
367 __three_after_loss(h
);
371 * Update moving-average of `s' and the sum of received payload bytes.
373 if (dccp_data_packet(skb
)) {
374 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
376 h
->packet_size
= tfrc_ewma(h
->packet_size
, payload
, 9);
377 h
->bytes_recvd
+= payload
;
380 /* RFC 3448, 6.1: update I_0, whose growth implies p <= p_prev */
382 tfrc_lh_update_i_mean(lh
, skb
);
386 EXPORT_SYMBOL_GPL(tfrc_rx_handle_loss
);
388 void tfrc_rx_hist_purge(struct tfrc_rx_hist
*h
)
392 for (i
= 0; i
<= TFRC_NDUPACK
; ++i
)
393 if (h
->ring
[i
] != NULL
) {
394 kmem_cache_free(tfrc_rx_hist_slab
, h
->ring
[i
]);
398 EXPORT_SYMBOL_GPL(tfrc_rx_hist_purge
);
400 static int tfrc_rx_hist_alloc(struct tfrc_rx_hist
*h
)
404 memset(h
, 0, sizeof(*h
));
406 for (i
= 0; i
<= TFRC_NDUPACK
; i
++) {
407 h
->ring
[i
] = kmem_cache_alloc(tfrc_rx_hist_slab
, GFP_ATOMIC
);
408 if (h
->ring
[i
] == NULL
) {
409 tfrc_rx_hist_purge(h
);
416 int tfrc_rx_hist_init(struct tfrc_rx_hist
*h
, struct sock
*sk
)
418 if (tfrc_rx_hist_alloc(h
))
421 * Initialise first entry with GSR to start loss detection as early as
422 * possible. Code using this must not use any other fields. The entry
423 * will be overwritten once the CCID updates its received packets.
425 tfrc_rx_hist_loss_prev(h
)->tfrchrx_seqno
= dccp_sk(sk
)->dccps_gsr
;
428 EXPORT_SYMBOL_GPL(tfrc_rx_hist_init
);
431 * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
433 static inline struct tfrc_rx_hist_entry
*
434 tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist
*h
)
440 * tfrc_rx_hist_rtt_prev_s: previously suitable (wrt rtt_last_s) RTT-sampling entry
442 static inline struct tfrc_rx_hist_entry
*
443 tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist
*h
)
445 return h
->ring
[h
->rtt_sample_prev
];
449 * tfrc_rx_hist_sample_rtt - Sample RTT from timestamp / CCVal
450 * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
451 * to compute a sample with given data - calling function should check this.
453 void tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist
*h
, const struct sk_buff
*skb
)
455 u32 sample
= 0, delta_v
;
458 * When not to sample:
459 * - on non-data packets
460 * (RFC 4342, 8.1: CCVal only fully defined for data packets);
461 * - when no data packets have been received yet
462 * (FIXME: using sampled packet size as indicator here);
463 * - as long as there are gaps in the sequence space (pending loss).
465 if (!dccp_data_packet(skb
) || h
->packet_size
== 0 ||
466 tfrc_rx_hist_loss_pending(h
))
469 delta_v
= SUB16(dccp_hdr(skb
)->dccph_ccval
,
470 tfrc_rx_hist_rtt_last_s(h
)->tfrchrx_ccval
);
472 if (delta_v
< 1 || delta_v
> 4) { /* unsuitable CCVal delta */
473 if (h
->rtt_sample_prev
== 2) { /* previous candidate stored */
474 sample
= SUB16(tfrc_rx_hist_rtt_prev_s(h
)->tfrchrx_ccval
,
475 tfrc_rx_hist_rtt_last_s(h
)->tfrchrx_ccval
);
477 sample
= 4 / sample
*
478 ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h
)->tfrchrx_tstamp
,
479 tfrc_rx_hist_rtt_last_s(h
)->tfrchrx_tstamp
);
481 * FIXME: This condition is in principle not
482 * possible but occurs when CCID is used for
483 * two-way data traffic. I have tried to trace
484 * it, but the cause does not seem to be here.
486 DCCP_BUG("please report to dccp@vger.kernel.org"
487 " => prev = %u, last = %u",
488 tfrc_rx_hist_rtt_prev_s(h
)->tfrchrx_ccval
,
489 tfrc_rx_hist_rtt_last_s(h
)->tfrchrx_ccval
);
490 } else if (delta_v
< 1) {
491 h
->rtt_sample_prev
= 1;
492 goto keep_ref_for_next_time
;
495 } else if (delta_v
== 4) /* optimal match */
496 sample
= ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h
)->tfrchrx_tstamp
));
497 else { /* suboptimal match */
498 h
->rtt_sample_prev
= 2;
499 goto keep_ref_for_next_time
;
502 if (unlikely(sample
> DCCP_SANE_RTT_MAX
)) {
503 DCCP_WARN("RTT sample %u too large, using max\n", sample
);
504 sample
= DCCP_SANE_RTT_MAX
;
507 h
->rtt_sample_prev
= 0; /* use current entry as next reference */
508 keep_ref_for_next_time
:
510 h
->rtt_estimate
= tfrc_ewma(h
->rtt_estimate
, sample
, 9);
512 EXPORT_SYMBOL_GPL(tfrc_rx_hist_sample_rtt
);
This page took 0.041398 seconds and 6 git commands to generate.