ccid2.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672
  1. /*
  2. * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
  3. *
  4. * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
  5. *
  6. * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  7. *
  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.
  12. *
  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.
  17. *
  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.
  21. */
  22. /*
  23. * This implementation should follow RFC 4341
  24. */
  25. #include <linux/slab.h>
  26. #include "../feat.h"
  27. #include "ccid2.h"
  28. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  29. static int ccid2_debug;
  30. #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
  31. #else
  32. #define ccid2_pr_debug(format, a...)
  33. #endif
  34. static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
  35. {
  36. struct ccid2_seq *seqp;
  37. int i;
  38. /* check if we have space to preserve the pointer to the buffer */
  39. if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
  40. sizeof(struct ccid2_seq *)))
  41. return -ENOMEM;
  42. /* allocate buffer and initialize linked list */
  43. seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
  44. if (seqp == NULL)
  45. return -ENOMEM;
  46. for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
  47. seqp[i].ccid2s_next = &seqp[i + 1];
  48. seqp[i + 1].ccid2s_prev = &seqp[i];
  49. }
  50. seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
  51. seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
  52. /* This is the first allocation. Initiate the head and tail. */
  53. if (hc->tx_seqbufc == 0)
  54. hc->tx_seqh = hc->tx_seqt = seqp;
  55. else {
  56. /* link the existing list with the one we just created */
  57. hc->tx_seqh->ccid2s_next = seqp;
  58. seqp->ccid2s_prev = hc->tx_seqh;
  59. hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
  60. seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
  61. }
  62. /* store the original pointer to the buffer so we can free it */
  63. hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
  64. hc->tx_seqbufc++;
  65. return 0;
  66. }
  67. static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
  68. {
  69. if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
  70. return CCID_PACKET_WILL_DEQUEUE_LATER;
  71. return CCID_PACKET_SEND_AT_ONCE;
  72. }
  73. static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
  74. {
  75. struct dccp_sock *dp = dccp_sk(sk);
  76. u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
  77. /*
  78. * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
  79. * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
  80. * acceptable since this causes starvation/deadlock whenever cwnd < 2.
  81. * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
  82. */
  83. if (val == 0 || val > max_ratio) {
  84. DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
  85. val = max_ratio;
  86. }
  87. if (val > DCCPF_ACK_RATIO_MAX)
  88. val = DCCPF_ACK_RATIO_MAX;
  89. if (val == dp->dccps_l_ack_ratio)
  90. return;
  91. ccid2_pr_debug("changing local ack ratio to %u\n", val);
  92. dp->dccps_l_ack_ratio = val;
  93. }
  94. static void ccid2_hc_tx_rto_expire(unsigned long data)
  95. {
  96. struct sock *sk = (struct sock *)data;
  97. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  98. const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
  99. bh_lock_sock(sk);
  100. if (sock_owned_by_user(sk)) {
  101. sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
  102. goto out;
  103. }
  104. ccid2_pr_debug("RTO_EXPIRE\n");
  105. /* back-off timer */
  106. hc->tx_rto <<= 1;
  107. if (hc->tx_rto > DCCP_RTO_MAX)
  108. hc->tx_rto = DCCP_RTO_MAX;
  109. /* adjust pipe, cwnd etc */
  110. hc->tx_ssthresh = hc->tx_cwnd / 2;
  111. if (hc->tx_ssthresh < 2)
  112. hc->tx_ssthresh = 2;
  113. hc->tx_cwnd = 1;
  114. hc->tx_pipe = 0;
  115. /* clear state about stuff we sent */
  116. hc->tx_seqt = hc->tx_seqh;
  117. hc->tx_packets_acked = 0;
  118. /* clear ack ratio state. */
  119. hc->tx_rpseq = 0;
  120. hc->tx_rpdupack = -1;
  121. ccid2_change_l_ack_ratio(sk, 1);
  122. /* if we were blocked before, we may now send cwnd=1 packet */
  123. if (sender_was_blocked)
  124. tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
  125. /* restart backed-off timer */
  126. sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
  127. out:
  128. bh_unlock_sock(sk);
  129. sock_put(sk);
  130. }
  131. static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
  132. {
  133. struct dccp_sock *dp = dccp_sk(sk);
  134. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  135. struct ccid2_seq *next;
  136. hc->tx_pipe++;
  137. hc->tx_seqh->ccid2s_seq = dp->dccps_gss;
  138. hc->tx_seqh->ccid2s_acked = 0;
  139. hc->tx_seqh->ccid2s_sent = ccid2_time_stamp;
  140. next = hc->tx_seqh->ccid2s_next;
  141. /* check if we need to alloc more space */
  142. if (next == hc->tx_seqt) {
  143. if (ccid2_hc_tx_alloc_seq(hc)) {
  144. DCCP_CRIT("packet history - out of memory!");
  145. /* FIXME: find a more graceful way to bail out */
  146. return;
  147. }
  148. next = hc->tx_seqh->ccid2s_next;
  149. BUG_ON(next == hc->tx_seqt);
  150. }
  151. hc->tx_seqh = next;
  152. ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
  153. /*
  154. * FIXME: The code below is broken and the variables have been removed
  155. * from the socket struct. The `ackloss' variable was always set to 0,
  156. * and with arsent there are several problems:
  157. * (i) it doesn't just count the number of Acks, but all sent packets;
  158. * (ii) it is expressed in # of packets, not # of windows, so the
  159. * comparison below uses the wrong formula: Appendix A of RFC 4341
  160. * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
  161. * of data with no lost or marked Ack packets. If arsent were the # of
  162. * consecutive Acks received without loss, then Ack Ratio needs to be
  163. * decreased by 1 when
  164. * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
  165. * where cwnd / R is the number of Acks received per window of data
  166. * (cf. RFC 4341, App. A). The problems are that
  167. * - arsent counts other packets as well;
  168. * - the comparison uses a formula different from RFC 4341;
  169. * - computing a cubic/quadratic equation each time is too complicated.
  170. * Hence a different algorithm is needed.
  171. */
  172. #if 0
  173. /* Ack Ratio. Need to maintain a concept of how many windows we sent */
  174. hc->tx_arsent++;
  175. /* We had an ack loss in this window... */
  176. if (hc->tx_ackloss) {
  177. if (hc->tx_arsent >= hc->tx_cwnd) {
  178. hc->tx_arsent = 0;
  179. hc->tx_ackloss = 0;
  180. }
  181. } else {
  182. /* No acks lost up to now... */
  183. /* decrease ack ratio if enough packets were sent */
  184. if (dp->dccps_l_ack_ratio > 1) {
  185. /* XXX don't calculate denominator each time */
  186. int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
  187. dp->dccps_l_ack_ratio;
  188. denom = hc->tx_cwnd * hc->tx_cwnd / denom;
  189. if (hc->tx_arsent >= denom) {
  190. ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
  191. hc->tx_arsent = 0;
  192. }
  193. } else {
  194. /* we can't increase ack ratio further [1] */
  195. hc->tx_arsent = 0; /* or maybe set it to cwnd*/
  196. }
  197. }
  198. #endif
  199. sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
  200. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  201. do {
  202. struct ccid2_seq *seqp = hc->tx_seqt;
  203. while (seqp != hc->tx_seqh) {
  204. ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
  205. (unsigned long long)seqp->ccid2s_seq,
  206. seqp->ccid2s_acked, seqp->ccid2s_sent);
  207. seqp = seqp->ccid2s_next;
  208. }
  209. } while (0);
  210. ccid2_pr_debug("=========\n");
  211. #endif
  212. }
  213. /**
  214. * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
  215. * This code is almost identical with TCP's tcp_rtt_estimator(), since
  216. * - it has a higher sampling frequency (recommended by RFC 1323),
  217. * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
  218. * - it is simple (cf. more complex proposals such as Eifel timer or research
  219. * which suggests that the gain should be set according to window size),
  220. * - in tests it was found to work well with CCID2 [gerrit].
  221. */
  222. static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
  223. {
  224. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  225. long m = mrtt ? : 1;
  226. if (hc->tx_srtt == 0) {
  227. /* First measurement m */
  228. hc->tx_srtt = m << 3;
  229. hc->tx_mdev = m << 1;
  230. hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
  231. hc->tx_rttvar = hc->tx_mdev_max;
  232. hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
  233. } else {
  234. /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
  235. m -= (hc->tx_srtt >> 3);
  236. hc->tx_srtt += m;
  237. /* Similarly, update scaled mdev with regard to |m| */
  238. if (m < 0) {
  239. m = -m;
  240. m -= (hc->tx_mdev >> 2);
  241. /*
  242. * This neutralises RTO increase when RTT < SRTT - mdev
  243. * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
  244. * in Linux TCP", USENIX 2002, pp. 49-62).
  245. */
  246. if (m > 0)
  247. m >>= 3;
  248. } else {
  249. m -= (hc->tx_mdev >> 2);
  250. }
  251. hc->tx_mdev += m;
  252. if (hc->tx_mdev > hc->tx_mdev_max) {
  253. hc->tx_mdev_max = hc->tx_mdev;
  254. if (hc->tx_mdev_max > hc->tx_rttvar)
  255. hc->tx_rttvar = hc->tx_mdev_max;
  256. }
  257. /*
  258. * Decay RTTVAR at most once per flight, exploiting that
  259. * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
  260. * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
  261. * GAR is a useful bound for FlightSize = pipe.
  262. * AWL is probably too low here, as it over-estimates pipe.
  263. */
  264. if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
  265. if (hc->tx_mdev_max < hc->tx_rttvar)
  266. hc->tx_rttvar -= (hc->tx_rttvar -
  267. hc->tx_mdev_max) >> 2;
  268. hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
  269. hc->tx_mdev_max = tcp_rto_min(sk);
  270. }
  271. }
  272. /*
  273. * Set RTO from SRTT and RTTVAR
  274. * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
  275. * This agrees with RFC 4341, 5:
  276. * "Because DCCP does not retransmit data, DCCP does not require
  277. * TCP's recommended minimum timeout of one second".
  278. */
  279. hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
  280. if (hc->tx_rto > DCCP_RTO_MAX)
  281. hc->tx_rto = DCCP_RTO_MAX;
  282. }
  283. static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
  284. unsigned int *maxincr)
  285. {
  286. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  287. if (hc->tx_cwnd < hc->tx_ssthresh) {
  288. if (*maxincr > 0 && ++hc->tx_packets_acked == 2) {
  289. hc->tx_cwnd += 1;
  290. *maxincr -= 1;
  291. hc->tx_packets_acked = 0;
  292. }
  293. } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
  294. hc->tx_cwnd += 1;
  295. hc->tx_packets_acked = 0;
  296. }
  297. /*
  298. * FIXME: RTT is sampled several times per acknowledgment (for each
  299. * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
  300. * This causes the RTT to be over-estimated, since the older entries
  301. * in the Ack Vector have earlier sending times.
  302. * The cleanest solution is to not use the ccid2s_sent field at all
  303. * and instead use DCCP timestamps: requires changes in other places.
  304. */
  305. ccid2_rtt_estimator(sk, ccid2_time_stamp - seqp->ccid2s_sent);
  306. }
  307. static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
  308. {
  309. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  310. if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
  311. ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
  312. return;
  313. }
  314. hc->tx_last_cong = ccid2_time_stamp;
  315. hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U;
  316. hc->tx_ssthresh = max(hc->tx_cwnd, 2U);
  317. /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
  318. if (dccp_sk(sk)->dccps_l_ack_ratio > hc->tx_cwnd)
  319. ccid2_change_l_ack_ratio(sk, hc->tx_cwnd);
  320. }
  321. static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
  322. u8 option, u8 *optval, u8 optlen)
  323. {
  324. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  325. switch (option) {
  326. case DCCPO_ACK_VECTOR_0:
  327. case DCCPO_ACK_VECTOR_1:
  328. return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
  329. option - DCCPO_ACK_VECTOR_0);
  330. }
  331. return 0;
  332. }
  333. static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
  334. {
  335. struct dccp_sock *dp = dccp_sk(sk);
  336. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  337. const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
  338. struct dccp_ackvec_parsed *avp;
  339. u64 ackno, seqno;
  340. struct ccid2_seq *seqp;
  341. int done = 0;
  342. unsigned int maxincr = 0;
  343. /* check reverse path congestion */
  344. seqno = DCCP_SKB_CB(skb)->dccpd_seq;
  345. /* XXX this whole "algorithm" is broken. Need to fix it to keep track
  346. * of the seqnos of the dupacks so that rpseq and rpdupack are correct
  347. * -sorbo.
  348. */
  349. /* need to bootstrap */
  350. if (hc->tx_rpdupack == -1) {
  351. hc->tx_rpdupack = 0;
  352. hc->tx_rpseq = seqno;
  353. } else {
  354. /* check if packet is consecutive */
  355. if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
  356. hc->tx_rpseq = seqno;
  357. /* it's a later packet */
  358. else if (after48(seqno, hc->tx_rpseq)) {
  359. hc->tx_rpdupack++;
  360. /* check if we got enough dupacks */
  361. if (hc->tx_rpdupack >= NUMDUPACK) {
  362. hc->tx_rpdupack = -1; /* XXX lame */
  363. hc->tx_rpseq = 0;
  364. ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
  365. }
  366. }
  367. }
  368. /* check forward path congestion */
  369. if (dccp_packet_without_ack(skb))
  370. return;
  371. /* still didn't send out new data packets */
  372. if (hc->tx_seqh == hc->tx_seqt)
  373. goto done;
  374. ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
  375. if (after48(ackno, hc->tx_high_ack))
  376. hc->tx_high_ack = ackno;
  377. seqp = hc->tx_seqt;
  378. while (before48(seqp->ccid2s_seq, ackno)) {
  379. seqp = seqp->ccid2s_next;
  380. if (seqp == hc->tx_seqh) {
  381. seqp = hc->tx_seqh->ccid2s_prev;
  382. break;
  383. }
  384. }
  385. /*
  386. * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
  387. * packets per acknowledgement. Rounding up avoids that cwnd is not
  388. * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
  389. */
  390. if (hc->tx_cwnd < hc->tx_ssthresh)
  391. maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
  392. /* go through all ack vectors */
  393. list_for_each_entry(avp, &hc->tx_av_chunks, node) {
  394. /* go through this ack vector */
  395. for (; avp->len--; avp->vec++) {
  396. u64 ackno_end_rl = SUB48(ackno,
  397. dccp_ackvec_runlen(avp->vec));
  398. ccid2_pr_debug("ackvec %llu |%u,%u|\n",
  399. (unsigned long long)ackno,
  400. dccp_ackvec_state(avp->vec) >> 6,
  401. dccp_ackvec_runlen(avp->vec));
  402. /* if the seqno we are analyzing is larger than the
  403. * current ackno, then move towards the tail of our
  404. * seqnos.
  405. */
  406. while (after48(seqp->ccid2s_seq, ackno)) {
  407. if (seqp == hc->tx_seqt) {
  408. done = 1;
  409. break;
  410. }
  411. seqp = seqp->ccid2s_prev;
  412. }
  413. if (done)
  414. break;
  415. /* check all seqnos in the range of the vector
  416. * run length
  417. */
  418. while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
  419. const u8 state = dccp_ackvec_state(avp->vec);
  420. /* new packet received or marked */
  421. if (state != DCCPAV_NOT_RECEIVED &&
  422. !seqp->ccid2s_acked) {
  423. if (state == DCCPAV_ECN_MARKED)
  424. ccid2_congestion_event(sk,
  425. seqp);
  426. else
  427. ccid2_new_ack(sk, seqp,
  428. &maxincr);
  429. seqp->ccid2s_acked = 1;
  430. ccid2_pr_debug("Got ack for %llu\n",
  431. (unsigned long long)seqp->ccid2s_seq);
  432. hc->tx_pipe--;
  433. }
  434. if (seqp == hc->tx_seqt) {
  435. done = 1;
  436. break;
  437. }
  438. seqp = seqp->ccid2s_prev;
  439. }
  440. if (done)
  441. break;
  442. ackno = SUB48(ackno_end_rl, 1);
  443. }
  444. if (done)
  445. break;
  446. }
  447. /* The state about what is acked should be correct now
  448. * Check for NUMDUPACK
  449. */
  450. seqp = hc->tx_seqt;
  451. while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
  452. seqp = seqp->ccid2s_next;
  453. if (seqp == hc->tx_seqh) {
  454. seqp = hc->tx_seqh->ccid2s_prev;
  455. break;
  456. }
  457. }
  458. done = 0;
  459. while (1) {
  460. if (seqp->ccid2s_acked) {
  461. done++;
  462. if (done == NUMDUPACK)
  463. break;
  464. }
  465. if (seqp == hc->tx_seqt)
  466. break;
  467. seqp = seqp->ccid2s_prev;
  468. }
  469. /* If there are at least 3 acknowledgements, anything unacknowledged
  470. * below the last sequence number is considered lost
  471. */
  472. if (done == NUMDUPACK) {
  473. struct ccid2_seq *last_acked = seqp;
  474. /* check for lost packets */
  475. while (1) {
  476. if (!seqp->ccid2s_acked) {
  477. ccid2_pr_debug("Packet lost: %llu\n",
  478. (unsigned long long)seqp->ccid2s_seq);
  479. /* XXX need to traverse from tail -> head in
  480. * order to detect multiple congestion events in
  481. * one ack vector.
  482. */
  483. ccid2_congestion_event(sk, seqp);
  484. hc->tx_pipe--;
  485. }
  486. if (seqp == hc->tx_seqt)
  487. break;
  488. seqp = seqp->ccid2s_prev;
  489. }
  490. hc->tx_seqt = last_acked;
  491. }
  492. /* trim acked packets in tail */
  493. while (hc->tx_seqt != hc->tx_seqh) {
  494. if (!hc->tx_seqt->ccid2s_acked)
  495. break;
  496. hc->tx_seqt = hc->tx_seqt->ccid2s_next;
  497. }
  498. /* restart RTO timer if not all outstanding data has been acked */
  499. if (hc->tx_pipe == 0)
  500. sk_stop_timer(sk, &hc->tx_rtotimer);
  501. else
  502. sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
  503. done:
  504. /* check if incoming Acks allow pending packets to be sent */
  505. if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
  506. tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
  507. dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
  508. }
  509. /*
  510. * Convert RFC 3390 larger initial window into an equivalent number of packets.
  511. * This is based on the numbers specified in RFC 5681, 3.1.
  512. */
  513. static inline u32 rfc3390_bytes_to_packets(const u32 smss)
  514. {
  515. return smss <= 1095 ? 4 : (smss > 2190 ? 2 : 3);
  516. }
  517. static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
  518. {
  519. struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
  520. struct dccp_sock *dp = dccp_sk(sk);
  521. u32 max_ratio;
  522. /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
  523. hc->tx_ssthresh = ~0U;
  524. /* Use larger initial windows (RFC 4341, section 5). */
  525. hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
  526. /* Make sure that Ack Ratio is enabled and within bounds. */
  527. max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
  528. if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
  529. dp->dccps_l_ack_ratio = max_ratio;
  530. /* XXX init ~ to window size... */
  531. if (ccid2_hc_tx_alloc_seq(hc))
  532. return -ENOMEM;
  533. hc->tx_rto = DCCP_TIMEOUT_INIT;
  534. hc->tx_rpdupack = -1;
  535. hc->tx_last_cong = ccid2_time_stamp;
  536. setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire,
  537. (unsigned long)sk);
  538. INIT_LIST_HEAD(&hc->tx_av_chunks);
  539. return 0;
  540. }
  541. static void ccid2_hc_tx_exit(struct sock *sk)
  542. {
  543. struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
  544. int i;
  545. sk_stop_timer(sk, &hc->tx_rtotimer);
  546. for (i = 0; i < hc->tx_seqbufc; i++)
  547. kfree(hc->tx_seqbuf[i]);
  548. hc->tx_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 *hc = ccid2_hc_rx_sk(sk);
  554. switch (DCCP_SKB_CB(skb)->dccpd_type) {
  555. case DCCP_PKT_DATA:
  556. case DCCP_PKT_DATAACK:
  557. hc->rx_data++;
  558. if (hc->rx_data >= dp->dccps_r_ack_ratio) {
  559. dccp_send_ack(sk);
  560. hc->rx_data = 0;
  561. }
  562. break;
  563. }
  564. }
  565. struct ccid_operations ccid2_ops = {
  566. .ccid_id = DCCPC_CCID2,
  567. .ccid_name = "TCP-like",
  568. .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
  569. .ccid_hc_tx_init = ccid2_hc_tx_init,
  570. .ccid_hc_tx_exit = ccid2_hc_tx_exit,
  571. .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
  572. .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
  573. .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
  574. .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
  575. .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
  576. .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
  577. };
  578. #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
  579. module_param(ccid2_debug, bool, 0644);
  580. MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
  581. #endif