ccid2.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687
  1. /*
  2. * net/dccp/ccids/ccid2.c
  3. *
  4. * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
  5. *
  6. * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
  7. *
  8. * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  9. *
  10. * This program is free software; you can redistribute it and/or modify
  11. * it under the terms of the GNU General Public License as published by
  12. * the Free Software Foundation; either version 2 of the License, or
  13. * (at your option) any later version.
  14. *
  15. * This program is distributed in the hope that it will be useful,
  16. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  18. * GNU General Public License for more details.
  19. *
  20. * You should have received a copy of the GNU General Public License
  21. * along with this program; if not, write to the Free Software
  22. * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23. */
  24. /*
  25. * This implementation should follow RFC 4341
  26. */
  27. #include "../feat.h"
  28. #include "../ccid.h"
  29. #include "../dccp.h"
  30. #include "ccid2.h"
  31. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  32. static int ccid2_debug;
  33. #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
  34. #else
  35. #define ccid2_pr_debug(format, a...)
  36. #endif
  37. static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hctx)
  38. {
  39. struct ccid2_seq *seqp;
  40. int i;
  41. /* check if we have space to preserve the pointer to the buffer */
  42. if (hctx->seqbufc >= sizeof(hctx->seqbuf) / sizeof(struct ccid2_seq *))
  43. return -ENOMEM;
  44. /* allocate buffer and initialize linked list */
  45. seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
  46. if (seqp == NULL)
  47. return -ENOMEM;
  48. for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
  49. seqp[i].ccid2s_next = &seqp[i + 1];
  50. seqp[i + 1].ccid2s_prev = &seqp[i];
  51. }
  52. seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
  53. seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
  54. /* This is the first allocation. Initiate the head and tail. */
  55. if (hctx->seqbufc == 0)
  56. hctx->seqh = hctx->seqt = seqp;
  57. else {
  58. /* link the existing list with the one we just created */
  59. hctx->seqh->ccid2s_next = seqp;
  60. seqp->ccid2s_prev = hctx->seqh;
  61. hctx->seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
  62. seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hctx->seqt;
  63. }
  64. /* store the original pointer to the buffer so we can free it */
  65. hctx->seqbuf[hctx->seqbufc] = seqp;
  66. hctx->seqbufc++;
  67. return 0;
  68. }
  69. static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
  70. {
  71. if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
  72. return CCID_PACKET_WILL_DEQUEUE_LATER;
  73. return CCID_PACKET_SEND_AT_ONCE;
  74. }
  75. static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
  76. {
  77. struct dccp_sock *dp = dccp_sk(sk);
  78. u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->cwnd, 2);
  79. /*
  80. * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
  81. * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
  82. * acceptable since this causes starvation/deadlock whenever cwnd < 2.
  83. * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
  84. */
  85. if (val == 0 || val > max_ratio) {
  86. DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
  87. val = max_ratio;
  88. }
  89. if (val > DCCPF_ACK_RATIO_MAX)
  90. val = DCCPF_ACK_RATIO_MAX;
  91. if (val == dp->dccps_l_ack_ratio)
  92. return;
  93. ccid2_pr_debug("changing local ack ratio to %u\n", val);
  94. dp->dccps_l_ack_ratio = val;
  95. }
  96. static void ccid2_hc_tx_rto_expire(unsigned long data)
  97. {
  98. struct sock *sk = (struct sock *)data;
  99. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  100. const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
  101. bh_lock_sock(sk);
  102. if (sock_owned_by_user(sk)) {
  103. sk_reset_timer(sk, &hctx->rtotimer, jiffies + HZ / 5);
  104. goto out;
  105. }
  106. ccid2_pr_debug("RTO_EXPIRE\n");
  107. /* back-off timer */
  108. hctx->rto <<= 1;
  109. if (hctx->rto > DCCP_RTO_MAX)
  110. hctx->rto = DCCP_RTO_MAX;
  111. /* adjust pipe, cwnd etc */
  112. hctx->ssthresh = hctx->cwnd / 2;
  113. if (hctx->ssthresh < 2)
  114. hctx->ssthresh = 2;
  115. hctx->cwnd = 1;
  116. hctx->pipe = 0;
  117. /* clear state about stuff we sent */
  118. hctx->seqt = hctx->seqh;
  119. hctx->packets_acked = 0;
  120. /* clear ack ratio state. */
  121. hctx->rpseq = 0;
  122. hctx->rpdupack = -1;
  123. ccid2_change_l_ack_ratio(sk, 1);
  124. /* if we were blocked before, we may now send cwnd=1 packet */
  125. if (sender_was_blocked)
  126. tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
  127. /* restart backed-off timer */
  128. sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
  129. out:
  130. bh_unlock_sock(sk);
  131. sock_put(sk);
  132. }
  133. static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
  134. {
  135. struct dccp_sock *dp = dccp_sk(sk);
  136. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  137. struct ccid2_seq *next;
  138. hctx->pipe++;
  139. hctx->seqh->ccid2s_seq = dp->dccps_gss;
  140. hctx->seqh->ccid2s_acked = 0;
  141. hctx->seqh->ccid2s_sent = jiffies;
  142. next = hctx->seqh->ccid2s_next;
  143. /* check if we need to alloc more space */
  144. if (next == hctx->seqt) {
  145. if (ccid2_hc_tx_alloc_seq(hctx)) {
  146. DCCP_CRIT("packet history - out of memory!");
  147. /* FIXME: find a more graceful way to bail out */
  148. return;
  149. }
  150. next = hctx->seqh->ccid2s_next;
  151. BUG_ON(next == hctx->seqt);
  152. }
  153. hctx->seqh = next;
  154. ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->cwnd, hctx->pipe);
  155. /*
  156. * FIXME: The code below is broken and the variables have been removed
  157. * from the socket struct. The `ackloss' variable was always set to 0,
  158. * and with arsent there are several problems:
  159. * (i) it doesn't just count the number of Acks, but all sent packets;
  160. * (ii) it is expressed in # of packets, not # of windows, so the
  161. * comparison below uses the wrong formula: Appendix A of RFC 4341
  162. * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
  163. * of data with no lost or marked Ack packets. If arsent were the # of
  164. * consecutive Acks received without loss, then Ack Ratio needs to be
  165. * decreased by 1 when
  166. * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
  167. * where cwnd / R is the number of Acks received per window of data
  168. * (cf. RFC 4341, App. A). The problems are that
  169. * - arsent counts other packets as well;
  170. * - the comparison uses a formula different from RFC 4341;
  171. * - computing a cubic/quadratic equation each time is too complicated.
  172. * Hence a different algorithm is needed.
  173. */
  174. #if 0
  175. /* Ack Ratio. Need to maintain a concept of how many windows we sent */
  176. hctx->arsent++;
  177. /* We had an ack loss in this window... */
  178. if (hctx->ackloss) {
  179. if (hctx->arsent >= hctx->cwnd) {
  180. hctx->arsent = 0;
  181. hctx->ackloss = 0;
  182. }
  183. } else {
  184. /* No acks lost up to now... */
  185. /* decrease ack ratio if enough packets were sent */
  186. if (dp->dccps_l_ack_ratio > 1) {
  187. /* XXX don't calculate denominator each time */
  188. int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
  189. dp->dccps_l_ack_ratio;
  190. denom = hctx->cwnd * hctx->cwnd / denom;
  191. if (hctx->arsent >= denom) {
  192. ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
  193. hctx->arsent = 0;
  194. }
  195. } else {
  196. /* we can't increase ack ratio further [1] */
  197. hctx->arsent = 0; /* or maybe set it to cwnd*/
  198. }
  199. }
  200. #endif
  201. /* setup RTO timer */
  202. if (!timer_pending(&hctx->rtotimer))
  203. sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
  204. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  205. do {
  206. struct ccid2_seq *seqp = hctx->seqt;
  207. while (seqp != hctx->seqh) {
  208. ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n",
  209. (unsigned long long)seqp->ccid2s_seq,
  210. seqp->ccid2s_acked, seqp->ccid2s_sent);
  211. seqp = seqp->ccid2s_next;
  212. }
  213. } while (0);
  214. ccid2_pr_debug("=========\n");
  215. #endif
  216. }
  217. /**
  218. * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
  219. * This code is almost identical with TCP's tcp_rtt_estimator(), since
  220. * - it has a higher sampling frequency (recommended by RFC 1323),
  221. * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
  222. * - it is simple (cf. more complex proposals such as Eifel timer or research
  223. * which suggests that the gain should be set according to window size),
  224. * - in tests it was found to work well with CCID2 [gerrit].
  225. */
  226. static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
  227. {
  228. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  229. long m = mrtt ? : 1;
  230. if (hctx->srtt == 0) {
  231. /* First measurement m */
  232. hctx->srtt = m << 3;
  233. hctx->mdev = m << 1;
  234. hctx->mdev_max = max(TCP_RTO_MIN, hctx->mdev);
  235. hctx->rttvar = hctx->mdev_max;
  236. hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
  237. } else {
  238. /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
  239. m -= (hctx->srtt >> 3);
  240. hctx->srtt += m;
  241. /* Similarly, update scaled mdev with regard to |m| */
  242. if (m < 0) {
  243. m = -m;
  244. m -= (hctx->mdev >> 2);
  245. /*
  246. * This neutralises RTO increase when RTT < SRTT - mdev
  247. * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
  248. * in Linux TCP", USENIX 2002, pp. 49-62).
  249. */
  250. if (m > 0)
  251. m >>= 3;
  252. } else {
  253. m -= (hctx->mdev >> 2);
  254. }
  255. hctx->mdev += m;
  256. if (hctx->mdev > hctx->mdev_max) {
  257. hctx->mdev_max = hctx->mdev;
  258. if (hctx->mdev_max > hctx->rttvar)
  259. hctx->rttvar = hctx->mdev_max;
  260. }
  261. /*
  262. * Decay RTTVAR at most once per flight, exploiting that
  263. * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
  264. * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
  265. * GAR is a useful bound for FlightSize = pipe, AWL is probably
  266. * too low as it over-estimates pipe.
  267. */
  268. if (after48(dccp_sk(sk)->dccps_gar, hctx->rtt_seq)) {
  269. if (hctx->mdev_max < hctx->rttvar)
  270. hctx->rttvar -= (hctx->rttvar -
  271. hctx->mdev_max) >> 2;
  272. hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
  273. hctx->mdev_max = TCP_RTO_MIN;
  274. }
  275. }
  276. /*
  277. * Set RTO from SRTT and RTTVAR
  278. * Clock granularity is ignored since the minimum error for RTTVAR is
  279. * clamped to 50msec (corresponding to HZ=20). This leads to a minimum
  280. * RTO of 200msec. This agrees with TCP and RFC 4341, 5.: "Because DCCP
  281. * does not retransmit data, DCCP does not require TCP's recommended
  282. * minimum timeout of one second".
  283. */
  284. hctx->rto = (hctx->srtt >> 3) + hctx->rttvar;
  285. if (hctx->rto > DCCP_RTO_MAX)
  286. hctx->rto = DCCP_RTO_MAX;
  287. }
  288. static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
  289. unsigned int *maxincr)
  290. {
  291. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  292. if (hctx->cwnd < hctx->ssthresh) {
  293. if (*maxincr > 0 && ++hctx->packets_acked == 2) {
  294. hctx->cwnd += 1;
  295. *maxincr -= 1;
  296. hctx->packets_acked = 0;
  297. }
  298. } else if (++hctx->packets_acked >= hctx->cwnd) {
  299. hctx->cwnd += 1;
  300. hctx->packets_acked = 0;
  301. }
  302. /*
  303. * FIXME: RTT is sampled several times per acknowledgment (for each
  304. * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
  305. * This causes the RTT to be over-estimated, since the older entries
  306. * in the Ack Vector have earlier sending times.
  307. * The cleanest solution is to not use the ccid2s_sent field at all
  308. * and instead use DCCP timestamps - need to be resolved at some time.
  309. */
  310. ccid2_rtt_estimator(sk, jiffies - seqp->ccid2s_sent);
  311. }
  312. static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
  313. {
  314. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  315. if (time_before(seqp->ccid2s_sent, hctx->last_cong)) {
  316. ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
  317. return;
  318. }
  319. hctx->last_cong = jiffies;
  320. hctx->cwnd = hctx->cwnd / 2 ? : 1U;
  321. hctx->ssthresh = max(hctx->cwnd, 2U);
  322. /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
  323. if (dccp_sk(sk)->dccps_l_ack_ratio > hctx->cwnd)
  324. ccid2_change_l_ack_ratio(sk, hctx->cwnd);
  325. }
  326. static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
  327. u8 option, u8 *optval, u8 optlen)
  328. {
  329. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  330. switch (option) {
  331. case DCCPO_ACK_VECTOR_0:
  332. case DCCPO_ACK_VECTOR_1:
  333. return dccp_ackvec_parsed_add(&hctx->av_chunks, optval, optlen,
  334. option - DCCPO_ACK_VECTOR_0);
  335. }
  336. return 0;
  337. }
  338. static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
  339. {
  340. struct dccp_sock *dp = dccp_sk(sk);
  341. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  342. const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
  343. struct dccp_ackvec_parsed *avp;
  344. u64 ackno, seqno;
  345. struct ccid2_seq *seqp;
  346. int done = 0;
  347. unsigned int maxincr = 0;
  348. /* check reverse path congestion */
  349. seqno = DCCP_SKB_CB(skb)->dccpd_seq;
  350. /* XXX this whole "algorithm" is broken. Need to fix it to keep track
  351. * of the seqnos of the dupacks so that rpseq and rpdupack are correct
  352. * -sorbo.
  353. */
  354. /* need to bootstrap */
  355. if (hctx->rpdupack == -1) {
  356. hctx->rpdupack = 0;
  357. hctx->rpseq = seqno;
  358. } else {
  359. /* check if packet is consecutive */
  360. if (dccp_delta_seqno(hctx->rpseq, seqno) == 1)
  361. hctx->rpseq = seqno;
  362. /* it's a later packet */
  363. else if (after48(seqno, hctx->rpseq)) {
  364. hctx->rpdupack++;
  365. /* check if we got enough dupacks */
  366. if (hctx->rpdupack >= NUMDUPACK) {
  367. hctx->rpdupack = -1; /* XXX lame */
  368. hctx->rpseq = 0;
  369. ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
  370. }
  371. }
  372. }
  373. /* check forward path congestion */
  374. if (dccp_packet_without_ack(skb))
  375. return;
  376. /* still didn't send out new data packets */
  377. if (hctx->seqh == hctx->seqt)
  378. goto done;
  379. ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
  380. if (after48(ackno, hctx->high_ack))
  381. hctx->high_ack = ackno;
  382. seqp = hctx->seqt;
  383. while (before48(seqp->ccid2s_seq, ackno)) {
  384. seqp = seqp->ccid2s_next;
  385. if (seqp == hctx->seqh) {
  386. seqp = hctx->seqh->ccid2s_prev;
  387. break;
  388. }
  389. }
  390. /*
  391. * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
  392. * packets per acknowledgement. Rounding up avoids that cwnd is not
  393. * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
  394. */
  395. if (hctx->cwnd < hctx->ssthresh)
  396. maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
  397. /* go through all ack vectors */
  398. list_for_each_entry(avp, &hctx->av_chunks, node) {
  399. /* go through this ack vector */
  400. for (; avp->len--; avp->vec++) {
  401. u64 ackno_end_rl = SUB48(ackno,
  402. dccp_ackvec_runlen(avp->vec));
  403. ccid2_pr_debug("ackvec %llu |%u,%u|\n",
  404. (unsigned long long)ackno,
  405. dccp_ackvec_state(avp->vec) >> 6,
  406. dccp_ackvec_runlen(avp->vec));
  407. /* if the seqno we are analyzing is larger than the
  408. * current ackno, then move towards the tail of our
  409. * seqnos.
  410. */
  411. while (after48(seqp->ccid2s_seq, ackno)) {
  412. if (seqp == hctx->seqt) {
  413. done = 1;
  414. break;
  415. }
  416. seqp = seqp->ccid2s_prev;
  417. }
  418. if (done)
  419. break;
  420. /* check all seqnos in the range of the vector
  421. * run length
  422. */
  423. while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
  424. const u8 state = dccp_ackvec_state(avp->vec);
  425. /* new packet received or marked */
  426. if (state != DCCPAV_NOT_RECEIVED &&
  427. !seqp->ccid2s_acked) {
  428. if (state == DCCPAV_ECN_MARKED)
  429. ccid2_congestion_event(sk,
  430. seqp);
  431. else
  432. ccid2_new_ack(sk, seqp,
  433. &maxincr);
  434. seqp->ccid2s_acked = 1;
  435. ccid2_pr_debug("Got ack for %llu\n",
  436. (unsigned long long)seqp->ccid2s_seq);
  437. hctx->pipe--;
  438. }
  439. if (seqp == hctx->seqt) {
  440. done = 1;
  441. break;
  442. }
  443. seqp = seqp->ccid2s_prev;
  444. }
  445. if (done)
  446. break;
  447. ackno = SUB48(ackno_end_rl, 1);
  448. }
  449. if (done)
  450. break;
  451. }
  452. /* The state about what is acked should be correct now
  453. * Check for NUMDUPACK
  454. */
  455. seqp = hctx->seqt;
  456. while (before48(seqp->ccid2s_seq, hctx->high_ack)) {
  457. seqp = seqp->ccid2s_next;
  458. if (seqp == hctx->seqh) {
  459. seqp = hctx->seqh->ccid2s_prev;
  460. break;
  461. }
  462. }
  463. done = 0;
  464. while (1) {
  465. if (seqp->ccid2s_acked) {
  466. done++;
  467. if (done == NUMDUPACK)
  468. break;
  469. }
  470. if (seqp == hctx->seqt)
  471. break;
  472. seqp = seqp->ccid2s_prev;
  473. }
  474. /* If there are at least 3 acknowledgements, anything unacknowledged
  475. * below the last sequence number is considered lost
  476. */
  477. if (done == NUMDUPACK) {
  478. struct ccid2_seq *last_acked = seqp;
  479. /* check for lost packets */
  480. while (1) {
  481. if (!seqp->ccid2s_acked) {
  482. ccid2_pr_debug("Packet lost: %llu\n",
  483. (unsigned long long)seqp->ccid2s_seq);
  484. /* XXX need to traverse from tail -> head in
  485. * order to detect multiple congestion events in
  486. * one ack vector.
  487. */
  488. ccid2_congestion_event(sk, seqp);
  489. hctx->pipe--;
  490. }
  491. if (seqp == hctx->seqt)
  492. break;
  493. seqp = seqp->ccid2s_prev;
  494. }
  495. hctx->seqt = last_acked;
  496. }
  497. /* trim acked packets in tail */
  498. while (hctx->seqt != hctx->seqh) {
  499. if (!hctx->seqt->ccid2s_acked)
  500. break;
  501. hctx->seqt = hctx->seqt->ccid2s_next;
  502. }
  503. /* restart RTO timer if not all outstanding data has been acked */
  504. if (hctx->pipe == 0)
  505. sk_stop_timer(sk, &hctx->rtotimer);
  506. else
  507. sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
  508. done:
  509. /* check if incoming Acks allow pending packets to be sent */
  510. if (sender_was_blocked && !ccid2_cwnd_network_limited(hctx))
  511. tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
  512. dccp_ackvec_parsed_cleanup(&hctx->av_chunks);
  513. }
  514. static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
  515. {
  516. struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid);
  517. struct dccp_sock *dp = dccp_sk(sk);
  518. u32 max_ratio;
  519. /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
  520. hctx->ssthresh = ~0U;
  521. /*
  522. * RFC 4341, 5: "The cwnd parameter is initialized to at most four
  523. * packets for new connections, following the rules from [RFC3390]".
  524. * We need to convert the bytes of RFC3390 into the packets of RFC 4341.
  525. */
  526. hctx->cwnd = clamp(4380U / dp->dccps_mss_cache, 2U, 4U);
  527. /* Make sure that Ack Ratio is enabled and within bounds. */
  528. max_ratio = DIV_ROUND_UP(hctx->cwnd, 2);
  529. if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
  530. dp->dccps_l_ack_ratio = max_ratio;
  531. /* XXX init ~ to window size... */
  532. if (ccid2_hc_tx_alloc_seq(hctx))
  533. return -ENOMEM;
  534. hctx->rto = DCCP_TIMEOUT_INIT;
  535. hctx->rpdupack = -1;
  536. hctx->last_cong = jiffies;
  537. setup_timer(&hctx->rtotimer, ccid2_hc_tx_rto_expire, (unsigned long)sk);
  538. INIT_LIST_HEAD(&hctx->av_chunks);
  539. return 0;
  540. }
  541. static void ccid2_hc_tx_exit(struct sock *sk)
  542. {
  543. struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
  544. int i;
  545. sk_stop_timer(sk, &hctx->rtotimer);
  546. for (i = 0; i < hctx->seqbufc; i++)
  547. kfree(hctx->seqbuf[i]);
  548. hctx->seqbufc = 0;
  549. }
  550. static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
  551. {
  552. const struct dccp_sock *dp = dccp_sk(sk);
  553. struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk);
  554. switch (DCCP_SKB_CB(skb)->dccpd_type) {
  555. case DCCP_PKT_DATA:
  556. case DCCP_PKT_DATAACK:
  557. hcrx->data++;
  558. if (hcrx->data >= dp->dccps_r_ack_ratio) {
  559. dccp_send_ack(sk);
  560. hcrx->data = 0;
  561. }
  562. break;
  563. }
  564. }
  565. static struct ccid_operations ccid2 = {
  566. .ccid_id = DCCPC_CCID2,
  567. .ccid_name = "TCP-like",
  568. .ccid_owner = THIS_MODULE,
  569. .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
  570. .ccid_hc_tx_init = ccid2_hc_tx_init,
  571. .ccid_hc_tx_exit = ccid2_hc_tx_exit,
  572. .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
  573. .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
  574. .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
  575. .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
  576. .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
  577. .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
  578. };
  579. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  580. module_param(ccid2_debug, bool, 0644);
  581. MODULE_PARM_DESC(ccid2_debug, "Enable debug messages");
  582. #endif
  583. static __init int ccid2_module_init(void)
  584. {
  585. return ccid_register(&ccid2);
  586. }
  587. module_init(ccid2_module_init);
  588. static __exit void ccid2_module_exit(void)
  589. {
  590. ccid_unregister(&ccid2);
  591. }
  592. module_exit(ccid2_module_exit);
  593. MODULE_AUTHOR("Andrea Bittau <a.bittau@cs.ucl.ac.uk>");
  594. MODULE_DESCRIPTION("DCCP TCP-Like (CCID2) CCID");
  595. MODULE_LICENSE("GPL");
  596. MODULE_ALIAS("net-dccp-ccid-2");