2 * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
4 * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
6 * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 * This implementation should follow RFC 4341
26 #include <linux/slab.h>
31 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
32 static bool ccid2_debug;
33 #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
35 #define ccid2_pr_debug(format, a...)
38 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
40 struct ccid2_seq *seqp;
43 /* check if we have space to preserve the pointer to the buffer */
44 if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
45 sizeof(struct ccid2_seq *)))
48 /* allocate buffer and initialize linked list */
49 seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
53 for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
54 seqp[i].ccid2s_next = &seqp[i + 1];
55 seqp[i + 1].ccid2s_prev = &seqp[i];
57 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
58 seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
60 /* This is the first allocation. Initiate the head and tail. */
61 if (hc->tx_seqbufc == 0)
62 hc->tx_seqh = hc->tx_seqt = seqp;
64 /* link the existing list with the one we just created */
65 hc->tx_seqh->ccid2s_next = seqp;
66 seqp->ccid2s_prev = hc->tx_seqh;
68 hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
69 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
72 /* store the original pointer to the buffer so we can free it */
73 hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
79 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
81 if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
82 return CCID_PACKET_WILL_DEQUEUE_LATER;
83 return CCID_PACKET_SEND_AT_ONCE;
86 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
88 u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
91 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
92 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
93 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
94 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
96 if (val == 0 || val > max_ratio) {
97 DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
100 dccp_feat_signal_nn_change(sk, DCCPF_ACK_RATIO,
101 min_t(u32, val, DCCPF_ACK_RATIO_MAX));
104 static void ccid2_check_l_ack_ratio(struct sock *sk)
106 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
109 * After a loss, idle period, application limited period, or RTO we
110 * need to check that the ack ratio is still less than the congestion
111 * window. Otherwise, we will send an entire congestion window of
112 * packets and got no response because we haven't sent ack ratio
114 * If the ack ratio does need to be reduced, we reduce it to half of
115 * the congestion window (or 1 if that's zero) instead of to the
116 * congestion window. This prevents problems if one ack is lost.
118 if (dccp_feat_nn_get(sk, DCCPF_ACK_RATIO) > hc->tx_cwnd)
119 ccid2_change_l_ack_ratio(sk, hc->tx_cwnd/2 ? : 1U);
122 static void ccid2_change_l_seq_window(struct sock *sk, u64 val)
124 dccp_feat_signal_nn_change(sk, DCCPF_SEQUENCE_WINDOW,
125 clamp_val(val, DCCPF_SEQ_WMIN,
129 static void dccp_tasklet_schedule(struct sock *sk)
131 struct tasklet_struct *t = &dccp_sk(sk)->dccps_xmitlet;
133 if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
135 __tasklet_schedule(t);
139 static void ccid2_hc_tx_rto_expire(struct timer_list *t)
141 struct ccid2_hc_tx_sock *hc = from_timer(hc, t, tx_rtotimer);
142 struct sock *sk = hc->sk;
143 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
146 if (sock_owned_by_user(sk)) {
147 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
151 ccid2_pr_debug("RTO_EXPIRE\n");
153 if (sk->sk_state == DCCP_CLOSED)
158 if (hc->tx_rto > DCCP_RTO_MAX)
159 hc->tx_rto = DCCP_RTO_MAX;
161 /* adjust pipe, cwnd etc */
162 hc->tx_ssthresh = hc->tx_cwnd / 2;
163 if (hc->tx_ssthresh < 2)
168 /* clear state about stuff we sent */
169 hc->tx_seqt = hc->tx_seqh;
170 hc->tx_packets_acked = 0;
172 /* clear ack ratio state. */
174 hc->tx_rpdupack = -1;
175 ccid2_change_l_ack_ratio(sk, 1);
177 /* if we were blocked before, we may now send cwnd=1 packet */
178 if (sender_was_blocked)
179 dccp_tasklet_schedule(sk);
180 /* restart backed-off timer */
181 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
188 * Congestion window validation (RFC 2861).
190 static bool ccid2_do_cwv = true;
191 module_param(ccid2_do_cwv, bool, 0644);
192 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation");
195 * ccid2_update_used_window - Track how much of cwnd is actually used
196 * This is done in addition to CWV. The sender needs to have an idea of how many
197 * packets may be in flight, to set the local Sequence Window value accordingly
198 * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the
199 * maximum-used window. We use an EWMA low-pass filter to filter out noise.
201 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd)
203 hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4;
206 /* This borrows the code of tcp_cwnd_application_limited() */
207 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now)
209 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
210 /* don't reduce cwnd below the initial window (IW) */
211 u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache),
212 win_used = max(hc->tx_cwnd_used, init_win);
214 if (win_used < hc->tx_cwnd) {
215 hc->tx_ssthresh = max(hc->tx_ssthresh,
216 (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2));
217 hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1;
219 hc->tx_cwnd_used = 0;
220 hc->tx_cwnd_stamp = now;
222 ccid2_check_l_ack_ratio(sk);
225 /* This borrows the code of tcp_cwnd_restart() */
226 static void ccid2_cwnd_restart(struct sock *sk, const u32 now)
228 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
229 u32 cwnd = hc->tx_cwnd, restart_cwnd,
230 iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache);
232 hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2));
234 /* don't reduce cwnd below the initial window (IW) */
235 restart_cwnd = min(cwnd, iwnd);
236 cwnd >>= (now - hc->tx_lsndtime) / hc->tx_rto;
237 hc->tx_cwnd = max(cwnd, restart_cwnd);
239 hc->tx_cwnd_stamp = now;
240 hc->tx_cwnd_used = 0;
242 ccid2_check_l_ack_ratio(sk);
245 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
247 struct dccp_sock *dp = dccp_sk(sk);
248 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
249 const u32 now = ccid2_jiffies32;
250 struct ccid2_seq *next;
252 /* slow-start after idle periods (RFC 2581, RFC 2861) */
253 if (ccid2_do_cwv && !hc->tx_pipe &&
254 (s32)(now - hc->tx_lsndtime) >= hc->tx_rto)
255 ccid2_cwnd_restart(sk, now);
257 hc->tx_lsndtime = now;
260 /* see whether cwnd was fully used (RFC 2861), update expected window */
261 if (ccid2_cwnd_network_limited(hc)) {
262 ccid2_update_used_window(hc, hc->tx_cwnd);
263 hc->tx_cwnd_used = 0;
264 hc->tx_cwnd_stamp = now;
266 if (hc->tx_pipe > hc->tx_cwnd_used)
267 hc->tx_cwnd_used = hc->tx_pipe;
269 ccid2_update_used_window(hc, hc->tx_cwnd_used);
271 if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto)
272 ccid2_cwnd_application_limited(sk, now);
275 hc->tx_seqh->ccid2s_seq = dp->dccps_gss;
276 hc->tx_seqh->ccid2s_acked = 0;
277 hc->tx_seqh->ccid2s_sent = now;
279 next = hc->tx_seqh->ccid2s_next;
280 /* check if we need to alloc more space */
281 if (next == hc->tx_seqt) {
282 if (ccid2_hc_tx_alloc_seq(hc)) {
283 DCCP_CRIT("packet history - out of memory!");
284 /* FIXME: find a more graceful way to bail out */
287 next = hc->tx_seqh->ccid2s_next;
288 BUG_ON(next == hc->tx_seqt);
292 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
295 * FIXME: The code below is broken and the variables have been removed
296 * from the socket struct. The `ackloss' variable was always set to 0,
297 * and with arsent there are several problems:
298 * (i) it doesn't just count the number of Acks, but all sent packets;
299 * (ii) it is expressed in # of packets, not # of windows, so the
300 * comparison below uses the wrong formula: Appendix A of RFC 4341
301 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
302 * of data with no lost or marked Ack packets. If arsent were the # of
303 * consecutive Acks received without loss, then Ack Ratio needs to be
304 * decreased by 1 when
305 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
306 * where cwnd / R is the number of Acks received per window of data
307 * (cf. RFC 4341, App. A). The problems are that
308 * - arsent counts other packets as well;
309 * - the comparison uses a formula different from RFC 4341;
310 * - computing a cubic/quadratic equation each time is too complicated.
311 * Hence a different algorithm is needed.
314 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
316 /* We had an ack loss in this window... */
317 if (hc->tx_ackloss) {
318 if (hc->tx_arsent >= hc->tx_cwnd) {
323 /* No acks lost up to now... */
324 /* decrease ack ratio if enough packets were sent */
325 if (dp->dccps_l_ack_ratio > 1) {
326 /* XXX don't calculate denominator each time */
327 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
328 dp->dccps_l_ack_ratio;
330 denom = hc->tx_cwnd * hc->tx_cwnd / denom;
332 if (hc->tx_arsent >= denom) {
333 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
337 /* we can't increase ack ratio further [1] */
338 hc->tx_arsent = 0; /* or maybe set it to cwnd*/
343 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
345 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
347 struct ccid2_seq *seqp = hc->tx_seqt;
349 while (seqp != hc->tx_seqh) {
350 ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
351 (unsigned long long)seqp->ccid2s_seq,
352 seqp->ccid2s_acked, seqp->ccid2s_sent);
353 seqp = seqp->ccid2s_next;
356 ccid2_pr_debug("=========\n");
361 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
362 * This code is almost identical with TCP's tcp_rtt_estimator(), since
363 * - it has a higher sampling frequency (recommended by RFC 1323),
364 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
365 * - it is simple (cf. more complex proposals such as Eifel timer or research
366 * which suggests that the gain should be set according to window size),
367 * - in tests it was found to work well with CCID2 [gerrit].
369 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
371 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
374 if (hc->tx_srtt == 0) {
375 /* First measurement m */
376 hc->tx_srtt = m << 3;
377 hc->tx_mdev = m << 1;
379 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
380 hc->tx_rttvar = hc->tx_mdev_max;
382 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
384 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
385 m -= (hc->tx_srtt >> 3);
388 /* Similarly, update scaled mdev with regard to |m| */
391 m -= (hc->tx_mdev >> 2);
393 * This neutralises RTO increase when RTT < SRTT - mdev
394 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
395 * in Linux TCP", USENIX 2002, pp. 49-62).
400 m -= (hc->tx_mdev >> 2);
404 if (hc->tx_mdev > hc->tx_mdev_max) {
405 hc->tx_mdev_max = hc->tx_mdev;
406 if (hc->tx_mdev_max > hc->tx_rttvar)
407 hc->tx_rttvar = hc->tx_mdev_max;
411 * Decay RTTVAR at most once per flight, exploiting that
412 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
413 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
414 * GAR is a useful bound for FlightSize = pipe.
415 * AWL is probably too low here, as it over-estimates pipe.
417 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
418 if (hc->tx_mdev_max < hc->tx_rttvar)
419 hc->tx_rttvar -= (hc->tx_rttvar -
420 hc->tx_mdev_max) >> 2;
421 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
422 hc->tx_mdev_max = tcp_rto_min(sk);
427 * Set RTO from SRTT and RTTVAR
428 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
429 * This agrees with RFC 4341, 5:
430 * "Because DCCP does not retransmit data, DCCP does not require
431 * TCP's recommended minimum timeout of one second".
433 hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
435 if (hc->tx_rto > DCCP_RTO_MAX)
436 hc->tx_rto = DCCP_RTO_MAX;
439 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
440 unsigned int *maxincr)
442 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
443 struct dccp_sock *dp = dccp_sk(sk);
444 int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio;
446 if (hc->tx_cwnd < dp->dccps_l_seq_win &&
447 r_seq_used < dp->dccps_r_seq_win) {
448 if (hc->tx_cwnd < hc->tx_ssthresh) {
449 if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) {
452 hc->tx_packets_acked = 0;
454 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
456 hc->tx_packets_acked = 0;
461 * Adjust the local sequence window and the ack ratio to allow about
462 * 5 times the number of packets in the network (RFC 4340 7.5.2)
464 if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win)
465 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2);
466 else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2)
467 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U);
469 if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win)
470 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2);
471 else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2)
472 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2);
475 * FIXME: RTT is sampled several times per acknowledgment (for each
476 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
477 * This causes the RTT to be over-estimated, since the older entries
478 * in the Ack Vector have earlier sending times.
479 * The cleanest solution is to not use the ccid2s_sent field at all
480 * and instead use DCCP timestamps: requires changes in other places.
482 ccid2_rtt_estimator(sk, ccid2_jiffies32 - seqp->ccid2s_sent);
485 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
487 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
489 if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
490 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
494 hc->tx_last_cong = ccid2_jiffies32;
496 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U;
497 hc->tx_ssthresh = max(hc->tx_cwnd, 2U);
499 ccid2_check_l_ack_ratio(sk);
502 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
503 u8 option, u8 *optval, u8 optlen)
505 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
508 case DCCPO_ACK_VECTOR_0:
509 case DCCPO_ACK_VECTOR_1:
510 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
511 option - DCCPO_ACK_VECTOR_0);
516 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
518 struct dccp_sock *dp = dccp_sk(sk);
519 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
520 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
521 struct dccp_ackvec_parsed *avp;
523 struct ccid2_seq *seqp;
525 unsigned int maxincr = 0;
527 /* check reverse path congestion */
528 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
530 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
531 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
534 /* need to bootstrap */
535 if (hc->tx_rpdupack == -1) {
537 hc->tx_rpseq = seqno;
539 /* check if packet is consecutive */
540 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
541 hc->tx_rpseq = seqno;
542 /* it's a later packet */
543 else if (after48(seqno, hc->tx_rpseq)) {
546 /* check if we got enough dupacks */
547 if (hc->tx_rpdupack >= NUMDUPACK) {
548 hc->tx_rpdupack = -1; /* XXX lame */
550 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__
552 * FIXME: Ack Congestion Control is broken; in
553 * the current state instabilities occurred with
554 * Ack Ratios greater than 1; causing hang-ups
555 * and long RTO timeouts. This needs to be fixed
556 * before opening up dynamic changes. -- gerrit
558 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
564 /* check forward path congestion */
565 if (dccp_packet_without_ack(skb))
568 /* still didn't send out new data packets */
569 if (hc->tx_seqh == hc->tx_seqt)
572 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
573 if (after48(ackno, hc->tx_high_ack))
574 hc->tx_high_ack = ackno;
577 while (before48(seqp->ccid2s_seq, ackno)) {
578 seqp = seqp->ccid2s_next;
579 if (seqp == hc->tx_seqh) {
580 seqp = hc->tx_seqh->ccid2s_prev;
586 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
587 * packets per acknowledgement. Rounding up avoids that cwnd is not
588 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
590 if (hc->tx_cwnd < hc->tx_ssthresh)
591 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
593 /* go through all ack vectors */
594 list_for_each_entry(avp, &hc->tx_av_chunks, node) {
595 /* go through this ack vector */
596 for (; avp->len--; avp->vec++) {
597 u64 ackno_end_rl = SUB48(ackno,
598 dccp_ackvec_runlen(avp->vec));
600 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
601 (unsigned long long)ackno,
602 dccp_ackvec_state(avp->vec) >> 6,
603 dccp_ackvec_runlen(avp->vec));
604 /* if the seqno we are analyzing is larger than the
605 * current ackno, then move towards the tail of our
608 while (after48(seqp->ccid2s_seq, ackno)) {
609 if (seqp == hc->tx_seqt) {
613 seqp = seqp->ccid2s_prev;
618 /* check all seqnos in the range of the vector
621 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
622 const u8 state = dccp_ackvec_state(avp->vec);
624 /* new packet received or marked */
625 if (state != DCCPAV_NOT_RECEIVED &&
626 !seqp->ccid2s_acked) {
627 if (state == DCCPAV_ECN_MARKED)
628 ccid2_congestion_event(sk,
631 ccid2_new_ack(sk, seqp,
634 seqp->ccid2s_acked = 1;
635 ccid2_pr_debug("Got ack for %llu\n",
636 (unsigned long long)seqp->ccid2s_seq);
639 if (seqp == hc->tx_seqt) {
643 seqp = seqp->ccid2s_prev;
648 ackno = SUB48(ackno_end_rl, 1);
654 /* The state about what is acked should be correct now
655 * Check for NUMDUPACK
658 while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
659 seqp = seqp->ccid2s_next;
660 if (seqp == hc->tx_seqh) {
661 seqp = hc->tx_seqh->ccid2s_prev;
667 if (seqp->ccid2s_acked) {
669 if (done == NUMDUPACK)
672 if (seqp == hc->tx_seqt)
674 seqp = seqp->ccid2s_prev;
677 /* If there are at least 3 acknowledgements, anything unacknowledged
678 * below the last sequence number is considered lost
680 if (done == NUMDUPACK) {
681 struct ccid2_seq *last_acked = seqp;
683 /* check for lost packets */
685 if (!seqp->ccid2s_acked) {
686 ccid2_pr_debug("Packet lost: %llu\n",
687 (unsigned long long)seqp->ccid2s_seq);
688 /* XXX need to traverse from tail -> head in
689 * order to detect multiple congestion events in
692 ccid2_congestion_event(sk, seqp);
695 if (seqp == hc->tx_seqt)
697 seqp = seqp->ccid2s_prev;
700 hc->tx_seqt = last_acked;
703 /* trim acked packets in tail */
704 while (hc->tx_seqt != hc->tx_seqh) {
705 if (!hc->tx_seqt->ccid2s_acked)
708 hc->tx_seqt = hc->tx_seqt->ccid2s_next;
711 /* restart RTO timer if not all outstanding data has been acked */
712 if (hc->tx_pipe == 0)
713 sk_stop_timer(sk, &hc->tx_rtotimer);
715 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
717 /* check if incoming Acks allow pending packets to be sent */
718 if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
719 dccp_tasklet_schedule(sk);
720 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
723 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
725 struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
726 struct dccp_sock *dp = dccp_sk(sk);
729 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
730 hc->tx_ssthresh = ~0U;
732 /* Use larger initial windows (RFC 4341, section 5). */
733 hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
734 hc->tx_expected_wnd = hc->tx_cwnd;
736 /* Make sure that Ack Ratio is enabled and within bounds. */
737 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
738 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
739 dp->dccps_l_ack_ratio = max_ratio;
741 /* XXX init ~ to window size... */
742 if (ccid2_hc_tx_alloc_seq(hc))
745 hc->tx_rto = DCCP_TIMEOUT_INIT;
746 hc->tx_rpdupack = -1;
747 hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_jiffies32;
748 hc->tx_cwnd_used = 0;
750 timer_setup(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, 0);
751 INIT_LIST_HEAD(&hc->tx_av_chunks);
755 static void ccid2_hc_tx_exit(struct sock *sk)
757 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
760 sk_stop_timer(sk, &hc->tx_rtotimer);
762 for (i = 0; i < hc->tx_seqbufc; i++)
763 kfree(hc->tx_seqbuf[i]);
765 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
768 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
770 struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
772 if (!dccp_data_packet(skb))
775 if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) {
777 hc->rx_num_data_pkts = 0;
781 struct ccid_operations ccid2_ops = {
782 .ccid_id = DCCPC_CCID2,
783 .ccid_name = "TCP-like",
784 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
785 .ccid_hc_tx_init = ccid2_hc_tx_init,
786 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
787 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
788 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
789 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
790 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
791 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
792 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
795 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
796 module_param(ccid2_debug, bool, 0644);
797 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");