ccid3.c 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872
  1. /*
  2. * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
  3. * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
  4. * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
  5. *
  6. * An implementation of the DCCP protocol
  7. *
  8. * This code has been developed by the University of Waikato WAND
  9. * research group. For further information please see http://www.wand.net.nz/
  10. *
  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
  14. *
  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>.
  18. *
  19. * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  20. *
  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.
  25. *
  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.
  30. *
  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.
  34. */
  35. #include "../dccp.h"
  36. #include "ccid3.h"
  37. #include <asm/unaligned.h>
  38. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  39. static bool ccid3_debug;
  40. #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
  41. #else
  42. #define ccid3_pr_debug(format, a...)
  43. #endif
  44. /*
  45. * Transmitter Half-Connection Routines
  46. */
  47. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  48. static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
  49. {
  50. static const char *const ccid3_state_names[] = {
  51. [TFRC_SSTATE_NO_SENT] = "NO_SENT",
  52. [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
  53. [TFRC_SSTATE_FBACK] = "FBACK",
  54. };
  55. return ccid3_state_names[state];
  56. }
  57. #endif
  58. static void ccid3_hc_tx_set_state(struct sock *sk,
  59. enum ccid3_hc_tx_states state)
  60. {
  61. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  62. enum ccid3_hc_tx_states oldstate = hc->tx_state;
  63. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  64. dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
  65. ccid3_tx_state_name(state));
  66. WARN_ON(state == oldstate);
  67. hc->tx_state = state;
  68. }
  69. /*
  70. * Compute the initial sending rate X_init in the manner of RFC 3390:
  71. *
  72. * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
  73. *
  74. * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
  75. * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
  76. * For consistency with other parts of the code, X_init is scaled by 2^6.
  77. */
  78. static inline u64 rfc3390_initial_rate(struct sock *sk)
  79. {
  80. const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  81. const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
  82. return scaled_div(w_init << 6, hc->tx_rtt);
  83. }
  84. /**
  85. * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
  86. * This respects the granularity of X_inst (64 * bytes/second).
  87. */
  88. static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
  89. {
  90. hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
  91. DCCP_BUG_ON(hc->tx_t_ipi == 0);
  92. ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
  93. hc->tx_s, (unsigned int)(hc->tx_x >> 6));
  94. }
  95. static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
  96. {
  97. u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
  98. return delta / hc->tx_rtt;
  99. }
  100. /**
  101. * ccid3_hc_tx_update_x - Update allowed sending rate X
  102. * @stamp: most recent time if available - can be left NULL.
  103. *
  104. * This function tracks draft rfc3448bis, check there for latest details.
  105. *
  106. * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
  107. * fine-grained resolution of sending rates. This requires scaling by 2^6
  108. * throughout the code. Only X_calc is unscaled (in bytes/second).
  109. *
  110. */
  111. static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
  112. {
  113. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  114. __u64 min_rate = 2 * hc->tx_x_recv;
  115. const __u64 old_x = hc->tx_x;
  116. ktime_t now = stamp ? *stamp : ktime_get_real();
  117. /*
  118. * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
  119. * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
  120. * a sender is idle if it has not sent anything over a 2-RTT-period.
  121. * For consistency with X and X_recv, min_rate is also scaled by 2^6.
  122. */
  123. if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
  124. min_rate = rfc3390_initial_rate(sk);
  125. min_rate = max(min_rate, 2 * hc->tx_x_recv);
  126. }
  127. if (hc->tx_p > 0) {
  128. hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
  129. hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
  130. } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
  131. hc->tx_x = min(2 * hc->tx_x, min_rate);
  132. hc->tx_x = max(hc->tx_x,
  133. scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
  134. hc->tx_t_ld = now;
  135. }
  136. if (hc->tx_x != old_x) {
  137. ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
  138. "X_recv=%u\n", (unsigned int)(old_x >> 6),
  139. (unsigned int)(hc->tx_x >> 6), hc->tx_x_calc,
  140. (unsigned int)(hc->tx_x_recv >> 6));
  141. ccid3_update_send_interval(hc);
  142. }
  143. }
  144. /**
  145. * ccid3_hc_tx_update_s - Track the mean packet size `s'
  146. * @len: DCCP packet payload size in bytes
  147. *
  148. * cf. RFC 4342, 5.3 and RFC 3448, 4.1
  149. */
  150. static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
  151. {
  152. const u16 old_s = hc->tx_s;
  153. hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
  154. if (hc->tx_s != old_s)
  155. ccid3_update_send_interval(hc);
  156. }
  157. /*
  158. * Update Window Counter using the algorithm from [RFC 4342, 8.1].
  159. * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
  160. */
  161. static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
  162. ktime_t now)
  163. {
  164. u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
  165. quarter_rtts = (4 * delta) / hc->tx_rtt;
  166. if (quarter_rtts > 0) {
  167. hc->tx_t_last_win_count = now;
  168. hc->tx_last_win_count += min(quarter_rtts, 5U);
  169. hc->tx_last_win_count &= 0xF; /* mod 16 */
  170. }
  171. }
  172. static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
  173. {
  174. struct sock *sk = (struct sock *)data;
  175. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  176. unsigned long t_nfb = USEC_PER_SEC / 5;
  177. bh_lock_sock(sk);
  178. if (sock_owned_by_user(sk)) {
  179. /* Try again later. */
  180. /* XXX: set some sensible MIB */
  181. goto restart_timer;
  182. }
  183. ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
  184. ccid3_tx_state_name(hc->tx_state));
  185. /* Ignore and do not restart after leaving the established state */
  186. if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
  187. goto out;
  188. /* Reset feedback state to "no feedback received" */
  189. if (hc->tx_state == TFRC_SSTATE_FBACK)
  190. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
  191. /*
  192. * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
  193. * RTO is 0 if and only if no feedback has been received yet.
  194. */
  195. if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
  196. /* halve send rate directly */
  197. hc->tx_x = max(hc->tx_x / 2,
  198. (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
  199. ccid3_update_send_interval(hc);
  200. } else {
  201. /*
  202. * Modify the cached value of X_recv
  203. *
  204. * If (X_calc > 2 * X_recv)
  205. * X_recv = max(X_recv / 2, s / (2 * t_mbi));
  206. * Else
  207. * X_recv = X_calc / 4;
  208. *
  209. * Note that X_recv is scaled by 2^6 while X_calc is not
  210. */
  211. if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
  212. hc->tx_x_recv =
  213. max(hc->tx_x_recv / 2,
  214. (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
  215. else {
  216. hc->tx_x_recv = hc->tx_x_calc;
  217. hc->tx_x_recv <<= 4;
  218. }
  219. ccid3_hc_tx_update_x(sk, NULL);
  220. }
  221. ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
  222. (unsigned long long)hc->tx_x);
  223. /*
  224. * Set new timeout for the nofeedback timer.
  225. * See comments in packet_recv() regarding the value of t_RTO.
  226. */
  227. if (unlikely(hc->tx_t_rto == 0)) /* no feedback received yet */
  228. t_nfb = TFRC_INITIAL_TIMEOUT;
  229. else
  230. t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
  231. restart_timer:
  232. sk_reset_timer(sk, &hc->tx_no_feedback_timer,
  233. jiffies + usecs_to_jiffies(t_nfb));
  234. out:
  235. bh_unlock_sock(sk);
  236. sock_put(sk);
  237. }
  238. /**
  239. * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
  240. * @skb: next packet candidate to send on @sk
  241. *
  242. * This function uses the convention of ccid_packet_dequeue_eval() and
  243. * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
  244. */
  245. static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
  246. {
  247. struct dccp_sock *dp = dccp_sk(sk);
  248. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  249. ktime_t now = ktime_get_real();
  250. s64 delay;
  251. /*
  252. * This function is called only for Data and DataAck packets. Sending
  253. * zero-sized Data(Ack)s is theoretically possible, but for congestion
  254. * control this case is pathological - ignore it.
  255. */
  256. if (unlikely(skb->len == 0))
  257. return -EBADMSG;
  258. if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
  259. sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
  260. usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
  261. hc->tx_last_win_count = 0;
  262. hc->tx_t_last_win_count = now;
  263. /* Set t_0 for initial packet */
  264. hc->tx_t_nom = now;
  265. hc->tx_s = skb->len;
  266. /*
  267. * Use initial RTT sample when available: recommended by erratum
  268. * to RFC 4342. This implements the initialisation procedure of
  269. * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
  270. */
  271. if (dp->dccps_syn_rtt) {
  272. ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
  273. hc->tx_rtt = dp->dccps_syn_rtt;
  274. hc->tx_x = rfc3390_initial_rate(sk);
  275. hc->tx_t_ld = now;
  276. } else {
  277. /*
  278. * Sender does not have RTT sample:
  279. * - set fallback RTT (RFC 4340, 3.4) since a RTT value
  280. * is needed in several parts (e.g. window counter);
  281. * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
  282. */
  283. hc->tx_rtt = DCCP_FALLBACK_RTT;
  284. hc->tx_x = hc->tx_s;
  285. hc->tx_x <<= 6;
  286. }
  287. ccid3_update_send_interval(hc);
  288. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
  289. } else {
  290. delay = ktime_us_delta(hc->tx_t_nom, now);
  291. ccid3_pr_debug("delay=%ld\n", (long)delay);
  292. /*
  293. * Scheduling of packet transmissions (RFC 5348, 8.3)
  294. *
  295. * if (t_now > t_nom - delta)
  296. * // send the packet now
  297. * else
  298. * // send the packet in (t_nom - t_now) milliseconds.
  299. */
  300. if (delay >= TFRC_T_DELTA)
  301. return (u32)delay / USEC_PER_MSEC;
  302. ccid3_hc_tx_update_win_count(hc, now);
  303. }
  304. /* prepare to send now (add options etc.) */
  305. dp->dccps_hc_tx_insert_options = 1;
  306. DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count;
  307. /* set the nominal send time for the next following packet */
  308. hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
  309. return CCID_PACKET_SEND_AT_ONCE;
  310. }
  311. static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
  312. {
  313. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  314. ccid3_hc_tx_update_s(hc, len);
  315. if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
  316. DCCP_CRIT("packet history - out of memory!");
  317. }
  318. static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
  319. {
  320. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  321. struct tfrc_tx_hist_entry *acked;
  322. ktime_t now;
  323. unsigned long t_nfb;
  324. u32 r_sample;
  325. /* we are only interested in ACKs */
  326. if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
  327. DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
  328. return;
  329. /*
  330. * Locate the acknowledged packet in the TX history.
  331. *
  332. * Returning "entry not found" here can for instance happen when
  333. * - the host has not sent out anything (e.g. a passive server),
  334. * - the Ack is outdated (packet with higher Ack number was received),
  335. * - it is a bogus Ack (for a packet not sent on this connection).
  336. */
  337. acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
  338. if (acked == NULL)
  339. return;
  340. /* For the sake of RTT sampling, ignore/remove all older entries */
  341. tfrc_tx_hist_purge(&acked->next);
  342. /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
  343. now = ktime_get_real();
  344. r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
  345. hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
  346. /*
  347. * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
  348. */
  349. if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
  350. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
  351. if (hc->tx_t_rto == 0) {
  352. /*
  353. * Initial feedback packet: Larger Initial Windows (4.2)
  354. */
  355. hc->tx_x = rfc3390_initial_rate(sk);
  356. hc->tx_t_ld = now;
  357. ccid3_update_send_interval(hc);
  358. goto done_computing_x;
  359. } else if (hc->tx_p == 0) {
  360. /*
  361. * First feedback after nofeedback timer expiry (4.3)
  362. */
  363. goto done_computing_x;
  364. }
  365. }
  366. /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
  367. if (hc->tx_p > 0)
  368. hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
  369. ccid3_hc_tx_update_x(sk, &now);
  370. done_computing_x:
  371. ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
  372. "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
  373. dccp_role(sk), sk, hc->tx_rtt, r_sample,
  374. hc->tx_s, hc->tx_p, hc->tx_x_calc,
  375. (unsigned int)(hc->tx_x_recv >> 6),
  376. (unsigned int)(hc->tx_x >> 6));
  377. /* unschedule no feedback timer */
  378. sk_stop_timer(sk, &hc->tx_no_feedback_timer);
  379. /*
  380. * As we have calculated new ipi, delta, t_nom it is possible
  381. * that we now can send a packet, so wake up dccp_wait_for_ccid
  382. */
  383. sk->sk_write_space(sk);
  384. /*
  385. * Update timeout interval for the nofeedback timer. In order to control
  386. * rate halving on networks with very low RTTs (<= 1 ms), use per-route
  387. * tunable RTAX_RTO_MIN value as the lower bound.
  388. */
  389. hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
  390. USEC_PER_SEC/HZ * tcp_rto_min(sk));
  391. /*
  392. * Schedule no feedback timer to expire in
  393. * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
  394. */
  395. t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
  396. ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
  397. "expire in %lu jiffies (%luus)\n",
  398. dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
  399. sk_reset_timer(sk, &hc->tx_no_feedback_timer,
  400. jiffies + usecs_to_jiffies(t_nfb));
  401. }
  402. static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
  403. u8 option, u8 *optval, u8 optlen)
  404. {
  405. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  406. __be32 opt_val;
  407. switch (option) {
  408. case TFRC_OPT_RECEIVE_RATE:
  409. case TFRC_OPT_LOSS_EVENT_RATE:
  410. /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
  411. if (packet_type == DCCP_PKT_DATA)
  412. break;
  413. if (unlikely(optlen != 4)) {
  414. DCCP_WARN("%s(%p), invalid len %d for %u\n",
  415. dccp_role(sk), sk, optlen, option);
  416. return -EINVAL;
  417. }
  418. opt_val = ntohl(get_unaligned((__be32 *)optval));
  419. if (option == TFRC_OPT_RECEIVE_RATE) {
  420. /* Receive Rate is kept in units of 64 bytes/second */
  421. hc->tx_x_recv = opt_val;
  422. hc->tx_x_recv <<= 6;
  423. ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
  424. dccp_role(sk), sk, opt_val);
  425. } else {
  426. /* Update the fixpoint Loss Event Rate fraction */
  427. hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
  428. ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
  429. dccp_role(sk), sk, opt_val);
  430. }
  431. }
  432. return 0;
  433. }
  434. static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
  435. {
  436. struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
  437. hc->tx_state = TFRC_SSTATE_NO_SENT;
  438. hc->tx_hist = NULL;
  439. setup_timer(&hc->tx_no_feedback_timer,
  440. ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
  441. return 0;
  442. }
  443. static void ccid3_hc_tx_exit(struct sock *sk)
  444. {
  445. struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  446. sk_stop_timer(sk, &hc->tx_no_feedback_timer);
  447. tfrc_tx_hist_purge(&hc->tx_hist);
  448. }
  449. static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
  450. {
  451. info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
  452. info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
  453. }
  454. static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
  455. u32 __user *optval, int __user *optlen)
  456. {
  457. const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
  458. struct tfrc_tx_info tfrc;
  459. const void *val;
  460. switch (optname) {
  461. case DCCP_SOCKOPT_CCID_TX_INFO:
  462. if (len < sizeof(tfrc))
  463. return -EINVAL;
  464. memset(&tfrc, 0, sizeof(tfrc));
  465. tfrc.tfrctx_x = hc->tx_x;
  466. tfrc.tfrctx_x_recv = hc->tx_x_recv;
  467. tfrc.tfrctx_x_calc = hc->tx_x_calc;
  468. tfrc.tfrctx_rtt = hc->tx_rtt;
  469. tfrc.tfrctx_p = hc->tx_p;
  470. tfrc.tfrctx_rto = hc->tx_t_rto;
  471. tfrc.tfrctx_ipi = hc->tx_t_ipi;
  472. len = sizeof(tfrc);
  473. val = &tfrc;
  474. break;
  475. default:
  476. return -ENOPROTOOPT;
  477. }
  478. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  479. return -EFAULT;
  480. return 0;
  481. }
  482. /*
  483. * Receiver Half-Connection Routines
  484. */
  485. /* CCID3 feedback types */
  486. enum ccid3_fback_type {
  487. CCID3_FBACK_NONE = 0,
  488. CCID3_FBACK_INITIAL,
  489. CCID3_FBACK_PERIODIC,
  490. CCID3_FBACK_PARAM_CHANGE
  491. };
  492. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  493. static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
  494. {
  495. static const char *const ccid3_rx_state_names[] = {
  496. [TFRC_RSTATE_NO_DATA] = "NO_DATA",
  497. [TFRC_RSTATE_DATA] = "DATA",
  498. };
  499. return ccid3_rx_state_names[state];
  500. }
  501. #endif
  502. static void ccid3_hc_rx_set_state(struct sock *sk,
  503. enum ccid3_hc_rx_states state)
  504. {
  505. struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  506. enum ccid3_hc_rx_states oldstate = hc->rx_state;
  507. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  508. dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
  509. ccid3_rx_state_name(state));
  510. WARN_ON(state == oldstate);
  511. hc->rx_state = state;
  512. }
  513. static void ccid3_hc_rx_send_feedback(struct sock *sk,
  514. const struct sk_buff *skb,
  515. enum ccid3_fback_type fbtype)
  516. {
  517. struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  518. struct dccp_sock *dp = dccp_sk(sk);
  519. ktime_t now = ktime_get();
  520. s64 delta = 0;
  521. switch (fbtype) {
  522. case CCID3_FBACK_INITIAL:
  523. hc->rx_x_recv = 0;
  524. hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */
  525. break;
  526. case CCID3_FBACK_PARAM_CHANGE:
  527. /*
  528. * When parameters change (new loss or p > p_prev), we do not
  529. * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
  530. * need to reuse the previous value of X_recv. However, when
  531. * X_recv was 0 (due to early loss), this would kill X down to
  532. * s/t_mbi (i.e. one packet in 64 seconds).
  533. * To avoid such drastic reduction, we approximate X_recv as
  534. * the number of bytes since last feedback.
  535. * This is a safe fallback, since X is bounded above by X_calc.
  536. */
  537. if (hc->rx_x_recv > 0)
  538. break;
  539. /* fall through */
  540. case CCID3_FBACK_PERIODIC:
  541. delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
  542. if (delta <= 0)
  543. delta = 1;
  544. hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
  545. break;
  546. default:
  547. return;
  548. }
  549. ccid3_pr_debug("Interval %lldusec, X_recv=%u, 1/p=%u\n", delta,
  550. hc->rx_x_recv, hc->rx_pinv);
  551. hc->rx_tstamp_last_feedback = now;
  552. hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval;
  553. hc->rx_bytes_recv = 0;
  554. dp->dccps_hc_rx_insert_options = 1;
  555. dccp_send_ack(sk);
  556. }
  557. static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
  558. {
  559. const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  560. __be32 x_recv, pinv;
  561. if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
  562. return 0;
  563. if (dccp_packet_without_ack(skb))
  564. return 0;
  565. x_recv = htonl(hc->rx_x_recv);
  566. pinv = htonl(hc->rx_pinv);
  567. if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
  568. &pinv, sizeof(pinv)) ||
  569. dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
  570. &x_recv, sizeof(x_recv)))
  571. return -1;
  572. return 0;
  573. }
  574. /**
  575. * ccid3_first_li - Implements [RFC 5348, 6.3.1]
  576. *
  577. * Determine the length of the first loss interval via inverse lookup.
  578. * Assume that X_recv can be computed by the throughput equation
  579. * s
  580. * X_recv = --------
  581. * R * fval
  582. * Find some p such that f(p) = fval; return 1/p (scaled).
  583. */
  584. static u32 ccid3_first_li(struct sock *sk)
  585. {
  586. struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  587. u32 x_recv, p;
  588. s64 delta;
  589. u64 fval;
  590. if (hc->rx_rtt == 0) {
  591. DCCP_WARN("No RTT estimate available, using fallback RTT\n");
  592. hc->rx_rtt = DCCP_FALLBACK_RTT;
  593. }
  594. delta = ktime_us_delta(ktime_get(), hc->rx_tstamp_last_feedback);
  595. if (delta <= 0)
  596. delta = 1;
  597. x_recv = scaled_div32(hc->rx_bytes_recv, delta);
  598. if (x_recv == 0) { /* would also trigger divide-by-zero */
  599. DCCP_WARN("X_recv==0\n");
  600. if (hc->rx_x_recv == 0) {
  601. DCCP_BUG("stored value of X_recv is zero");
  602. return ~0U;
  603. }
  604. x_recv = hc->rx_x_recv;
  605. }
  606. fval = scaled_div(hc->rx_s, hc->rx_rtt);
  607. fval = scaled_div32(fval, x_recv);
  608. p = tfrc_calc_x_reverse_lookup(fval);
  609. ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
  610. "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
  611. return p == 0 ? ~0U : scaled_div(1, p);
  612. }
  613. static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
  614. {
  615. struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  616. enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
  617. const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
  618. const bool is_data_packet = dccp_data_packet(skb);
  619. if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
  620. if (is_data_packet) {
  621. const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
  622. do_feedback = CCID3_FBACK_INITIAL;
  623. ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
  624. hc->rx_s = payload;
  625. /*
  626. * Not necessary to update rx_bytes_recv here,
  627. * since X_recv = 0 for the first feedback packet (cf.
  628. * RFC 3448, 6.3) -- gerrit
  629. */
  630. }
  631. goto update_records;
  632. }
  633. if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
  634. return; /* done receiving */
  635. if (is_data_packet) {
  636. const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
  637. /*
  638. * Update moving-average of s and the sum of received payload bytes
  639. */
  640. hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
  641. hc->rx_bytes_recv += payload;
  642. }
  643. /*
  644. * Perform loss detection and handle pending losses
  645. */
  646. if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
  647. skb, ndp, ccid3_first_li, sk)) {
  648. do_feedback = CCID3_FBACK_PARAM_CHANGE;
  649. goto done_receiving;
  650. }
  651. if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
  652. return; /* done receiving */
  653. /*
  654. * Handle data packets: RTT sampling and monitoring p
  655. */
  656. if (unlikely(!is_data_packet))
  657. goto update_records;
  658. if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
  659. const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
  660. /*
  661. * Empty loss history: no loss so far, hence p stays 0.
  662. * Sample RTT values, since an RTT estimate is required for the
  663. * computation of p when the first loss occurs; RFC 3448, 6.3.1.
  664. */
  665. if (sample != 0)
  666. hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
  667. } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
  668. /*
  669. * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
  670. * has decreased (resp. p has increased), send feedback now.
  671. */
  672. do_feedback = CCID3_FBACK_PARAM_CHANGE;
  673. }
  674. /*
  675. * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
  676. */
  677. if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
  678. do_feedback = CCID3_FBACK_PERIODIC;
  679. update_records:
  680. tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
  681. done_receiving:
  682. if (do_feedback)
  683. ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
  684. }
  685. static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
  686. {
  687. struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
  688. hc->rx_state = TFRC_RSTATE_NO_DATA;
  689. tfrc_lh_init(&hc->rx_li_hist);
  690. return tfrc_rx_hist_alloc(&hc->rx_hist);
  691. }
  692. static void ccid3_hc_rx_exit(struct sock *sk)
  693. {
  694. struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  695. tfrc_rx_hist_purge(&hc->rx_hist);
  696. tfrc_lh_cleanup(&hc->rx_li_hist);
  697. }
  698. static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
  699. {
  700. info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
  701. info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
  702. info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rx_rtt;
  703. }
  704. static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
  705. u32 __user *optval, int __user *optlen)
  706. {
  707. const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
  708. struct tfrc_rx_info rx_info;
  709. const void *val;
  710. switch (optname) {
  711. case DCCP_SOCKOPT_CCID_RX_INFO:
  712. if (len < sizeof(rx_info))
  713. return -EINVAL;
  714. rx_info.tfrcrx_x_recv = hc->rx_x_recv;
  715. rx_info.tfrcrx_rtt = hc->rx_rtt;
  716. rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hc->rx_pinv);
  717. len = sizeof(rx_info);
  718. val = &rx_info;
  719. break;
  720. default:
  721. return -ENOPROTOOPT;
  722. }
  723. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  724. return -EFAULT;
  725. return 0;
  726. }
  727. struct ccid_operations ccid3_ops = {
  728. .ccid_id = DCCPC_CCID3,
  729. .ccid_name = "TCP-Friendly Rate Control",
  730. .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
  731. .ccid_hc_tx_init = ccid3_hc_tx_init,
  732. .ccid_hc_tx_exit = ccid3_hc_tx_exit,
  733. .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
  734. .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
  735. .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
  736. .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
  737. .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
  738. .ccid_hc_rx_init = ccid3_hc_rx_init,
  739. .ccid_hc_rx_exit = ccid3_hc_rx_exit,
  740. .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
  741. .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
  742. .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
  743. .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
  744. .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
  745. .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
  746. };
  747. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  748. module_param(ccid3_debug, bool, 0644);
  749. MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
  750. #endif