ccid3.c 36 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271
  1. /*
  2. * net/dccp/ccids/ccid3.c
  3. *
  4. * Copyright (c) 2005 The University of Waikato, Hamilton, New Zealand.
  5. * Copyright (c) 2005-6 Ian McDonald <ian.mcdonald@jandi.co.nz>
  6. *
  7. * An implementation of the DCCP protocol
  8. *
  9. * This code has been developed by the University of Waikato WAND
  10. * research group. For further information please see http://www.wand.net.nz/
  11. *
  12. * This code also uses code from Lulea University, rereleased as GPL by its
  13. * authors:
  14. * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
  15. *
  16. * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
  17. * and to make it work as a loadable module in the DCCP stack written by
  18. * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
  19. *
  20. * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  21. *
  22. * This program is free software; you can redistribute it and/or modify
  23. * it under the terms of the GNU General Public License as published by
  24. * the Free Software Foundation; either version 2 of the License, or
  25. * (at your option) any later version.
  26. *
  27. * This program is distributed in the hope that it will be useful,
  28. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  29. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  30. * GNU General Public License for more details.
  31. *
  32. * You should have received a copy of the GNU General Public License
  33. * along with this program; if not, write to the Free Software
  34. * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  35. */
  36. #include "../ccid.h"
  37. #include "../dccp.h"
  38. #include "lib/packet_history.h"
  39. #include "lib/loss_interval.h"
  40. #include "lib/tfrc.h"
  41. #include "ccid3.h"
  42. /*
  43. * Reason for maths here is to avoid 32 bit overflow when a is big.
  44. * With this we get close to the limit.
  45. */
  46. static u32 usecs_div(const u32 a, const u32 b)
  47. {
  48. const u32 div = a < (UINT_MAX / (USEC_PER_SEC / 10)) ? 10 :
  49. a < (UINT_MAX / (USEC_PER_SEC / 50)) ? 50 :
  50. a < (UINT_MAX / (USEC_PER_SEC / 100)) ? 100 :
  51. a < (UINT_MAX / (USEC_PER_SEC / 500)) ? 500 :
  52. a < (UINT_MAX / (USEC_PER_SEC / 1000)) ? 1000 :
  53. a < (UINT_MAX / (USEC_PER_SEC / 5000)) ? 5000 :
  54. a < (UINT_MAX / (USEC_PER_SEC / 10000)) ? 10000 :
  55. a < (UINT_MAX / (USEC_PER_SEC / 50000)) ? 50000 :
  56. 100000;
  57. const u32 tmp = a * (USEC_PER_SEC / div);
  58. return (b >= 2 * div) ? tmp / (b / div) : tmp;
  59. }
  60. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  61. static int ccid3_debug;
  62. #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
  63. #else
  64. #define ccid3_pr_debug(format, a...)
  65. #endif
  66. static struct dccp_tx_hist *ccid3_tx_hist;
  67. static struct dccp_rx_hist *ccid3_rx_hist;
  68. static struct dccp_li_hist *ccid3_li_hist;
  69. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  70. static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
  71. {
  72. static char *ccid3_state_names[] = {
  73. [TFRC_SSTATE_NO_SENT] = "NO_SENT",
  74. [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
  75. [TFRC_SSTATE_FBACK] = "FBACK",
  76. [TFRC_SSTATE_TERM] = "TERM",
  77. };
  78. return ccid3_state_names[state];
  79. }
  80. #endif
  81. static void ccid3_hc_tx_set_state(struct sock *sk,
  82. enum ccid3_hc_tx_states state)
  83. {
  84. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  85. enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
  86. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  87. dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
  88. ccid3_tx_state_name(state));
  89. WARN_ON(state == oldstate);
  90. hctx->ccid3hctx_state = state;
  91. }
  92. /*
  93. * Recalculate scheduled nominal send time t_nom, inter-packet interval
  94. * t_ipi, and delta value. Should be called after each change to X.
  95. */
  96. static inline void ccid3_update_send_time(struct ccid3_hc_tx_sock *hctx)
  97. {
  98. timeval_sub_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
  99. /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
  100. hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_x);
  101. /* Update nominal send time with regard to the new t_ipi */
  102. timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
  103. /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
  104. hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
  105. TFRC_OPSYS_HALF_TIME_GRAN);
  106. }
  107. /*
  108. * Update X by
  109. * If (p > 0)
  110. * x_calc = calcX(s, R, p);
  111. * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
  112. * Else
  113. * If (now - tld >= R)
  114. * X = max(min(2 * X, 2 * X_recv), s / R);
  115. * tld = now;
  116. */
  117. static void ccid3_hc_tx_update_x(struct sock *sk, struct timeval *now)
  118. {
  119. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  120. const __u32 old_x = hctx->ccid3hctx_x;
  121. /* To avoid large error in calcX */
  122. if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
  123. hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
  124. hctx->ccid3hctx_rtt,
  125. hctx->ccid3hctx_p);
  126. hctx->ccid3hctx_x = max_t(u32, min(hctx->ccid3hctx_x_calc,
  127. hctx->ccid3hctx_x_recv * 2),
  128. hctx->ccid3hctx_s / TFRC_T_MBI);
  129. } else if (timeval_delta(now, &hctx->ccid3hctx_t_ld) >=
  130. hctx->ccid3hctx_rtt) {
  131. hctx->ccid3hctx_x = max(min(hctx->ccid3hctx_x_recv,
  132. hctx->ccid3hctx_x ) * 2,
  133. usecs_div(hctx->ccid3hctx_s,
  134. hctx->ccid3hctx_rtt) );
  135. hctx->ccid3hctx_t_ld = *now;
  136. } else
  137. ccid3_pr_debug("Not changing X\n");
  138. if (hctx->ccid3hctx_x != old_x)
  139. ccid3_update_send_time(hctx);
  140. }
  141. /*
  142. * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
  143. * @len: DCCP packet payload size in bytes
  144. */
  145. static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
  146. {
  147. if (unlikely(len == 0))
  148. ccid3_pr_debug("Packet payload length is 0 - not updating\n");
  149. else
  150. hctx->ccid3hctx_s = hctx->ccid3hctx_s == 0 ? len :
  151. (9 * hctx->ccid3hctx_s + len) / 10;
  152. /*
  153. * Note: We could do a potential optimisation here - when `s' changes,
  154. * recalculate sending rate and consequently t_ipi, t_delta, and
  155. * t_now. This is however non-standard, and the benefits are not
  156. * clear, so it is currently left out.
  157. */
  158. }
  159. static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
  160. {
  161. struct sock *sk = (struct sock *)data;
  162. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  163. unsigned long t_nfb = USEC_PER_SEC / 5;
  164. bh_lock_sock(sk);
  165. if (sock_owned_by_user(sk)) {
  166. /* Try again later. */
  167. /* XXX: set some sensible MIB */
  168. goto restart_timer;
  169. }
  170. ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
  171. ccid3_tx_state_name(hctx->ccid3hctx_state));
  172. switch (hctx->ccid3hctx_state) {
  173. case TFRC_SSTATE_NO_FBACK:
  174. /* RFC 3448, 4.4: Halve send rate directly */
  175. hctx->ccid3hctx_x = min_t(u32, hctx->ccid3hctx_x / 2,
  176. hctx->ccid3hctx_s / TFRC_T_MBI);
  177. ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
  178. "bytes/s\n",
  179. dccp_role(sk), sk,
  180. ccid3_tx_state_name(hctx->ccid3hctx_state),
  181. hctx->ccid3hctx_x);
  182. /* The value of R is still undefined and so we can not recompute
  183. * the timout value. Keep initial value as per [RFC 4342, 5]. */
  184. t_nfb = TFRC_INITIAL_TIMEOUT;
  185. ccid3_update_send_time(hctx);
  186. break;
  187. case TFRC_SSTATE_FBACK:
  188. /*
  189. * Check if IDLE since last timeout and recv rate is less than
  190. * 4 packets per RTT
  191. */
  192. if (!hctx->ccid3hctx_idle ||
  193. (hctx->ccid3hctx_x_recv >=
  194. 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
  195. struct timeval now;
  196. ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
  197. dccp_role(sk), sk,
  198. ccid3_tx_state_name(hctx->ccid3hctx_state));
  199. /* Halve sending rate */
  200. /* If (X_calc > 2 * X_recv)
  201. * X_recv = max(X_recv / 2, s / (2 * t_mbi));
  202. * Else
  203. * X_recv = X_calc / 4;
  204. */
  205. BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
  206. hctx->ccid3hctx_x_calc == 0);
  207. /* check also if p is zero -> x_calc is infinity? */
  208. if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
  209. hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
  210. hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
  211. hctx->ccid3hctx_s / (2 * TFRC_T_MBI));
  212. else
  213. hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
  214. /* Update sending rate */
  215. dccp_timestamp(sk, &now);
  216. ccid3_hc_tx_update_x(sk, &now);
  217. }
  218. /*
  219. * Schedule no feedback timer to expire in
  220. * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
  221. * See comments in packet_recv() regarding the value of t_RTO.
  222. */
  223. t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
  224. break;
  225. case TFRC_SSTATE_NO_SENT:
  226. DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
  227. /* fall through */
  228. case TFRC_SSTATE_TERM:
  229. goto out;
  230. }
  231. hctx->ccid3hctx_idle = 1;
  232. restart_timer:
  233. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  234. jiffies + usecs_to_jiffies(t_nfb));
  235. out:
  236. bh_unlock_sock(sk);
  237. sock_put(sk);
  238. }
  239. /*
  240. * returns
  241. * > 0: delay (in msecs) that should pass before actually sending
  242. * = 0: can send immediately
  243. * < 0: error condition; do not send packet
  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 *hctx = ccid3_hc_tx_sk(sk);
  249. struct dccp_tx_hist_entry *new_packet;
  250. struct timeval now;
  251. long delay;
  252. BUG_ON(hctx == NULL);
  253. /*
  254. * This function is called only for Data and DataAck packets. Sending
  255. * zero-sized Data(Ack)s is theoretically possible, but for congestion
  256. * control this case is pathological - ignore it.
  257. */
  258. if (unlikely(skb->len == 0))
  259. return -EBADMSG;
  260. /* See if last packet allocated was not sent */
  261. new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
  262. if (new_packet == NULL || new_packet->dccphtx_sent) {
  263. new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
  264. SLAB_ATOMIC);
  265. if (unlikely(new_packet == NULL)) {
  266. DCCP_WARN("%s, sk=%p, not enough mem to add to history,"
  267. "send refused\n", dccp_role(sk), sk);
  268. return -ENOBUFS;
  269. }
  270. dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
  271. }
  272. dccp_timestamp(sk, &now);
  273. switch (hctx->ccid3hctx_state) {
  274. case TFRC_SSTATE_NO_SENT:
  275. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  276. jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
  277. hctx->ccid3hctx_last_win_count = 0;
  278. hctx->ccid3hctx_t_last_win_count = now;
  279. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
  280. /* Set initial sending rate to 1 packet per second */
  281. ccid3_hc_tx_update_s(hctx, skb->len);
  282. hctx->ccid3hctx_x = hctx->ccid3hctx_s;
  283. /* First timeout, according to [RFC 3448, 4.2], is 1 second */
  284. hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
  285. /* Initial delta: minimum of 0.5 sec and t_gran/2 */
  286. hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
  287. /* Set t_0 for initial packet */
  288. hctx->ccid3hctx_t_nom = now;
  289. break;
  290. case TFRC_SSTATE_NO_FBACK:
  291. case TFRC_SSTATE_FBACK:
  292. delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
  293. /*
  294. * Scheduling of packet transmissions [RFC 3448, 4.6]
  295. *
  296. * if (t_now > t_nom - delta)
  297. * // send the packet now
  298. * else
  299. * // send the packet in (t_nom - t_now) milliseconds.
  300. */
  301. if (delay >= hctx->ccid3hctx_delta)
  302. return delay / 1000L;
  303. break;
  304. case TFRC_SSTATE_TERM:
  305. DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
  306. return -EINVAL;
  307. }
  308. /* prepare to send now (add options etc.) */
  309. dp->dccps_hc_tx_insert_options = 1;
  310. new_packet->dccphtx_ccval = DCCP_SKB_CB(skb)->dccpd_ccval =
  311. hctx->ccid3hctx_last_win_count;
  312. timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
  313. return 0;
  314. }
  315. static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len)
  316. {
  317. const struct dccp_sock *dp = dccp_sk(sk);
  318. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  319. struct timeval now;
  320. unsigned long quarter_rtt;
  321. struct dccp_tx_hist_entry *packet;
  322. BUG_ON(hctx == NULL);
  323. dccp_timestamp(sk, &now);
  324. ccid3_hc_tx_update_s(hctx, len);
  325. packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
  326. if (unlikely(packet == NULL)) {
  327. DCCP_WARN("packet doesn't exist in history!\n");
  328. return;
  329. }
  330. if (unlikely(packet->dccphtx_sent)) {
  331. DCCP_WARN("no unsent packet in history!\n");
  332. return;
  333. }
  334. packet->dccphtx_tstamp = now;
  335. packet->dccphtx_seqno = dp->dccps_gss;
  336. /*
  337. * Check if win_count have changed
  338. * Algorithm in "8.1. Window Counter Value" in RFC 4342.
  339. */
  340. quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
  341. if (likely(hctx->ccid3hctx_rtt > 8))
  342. quarter_rtt /= hctx->ccid3hctx_rtt / 4;
  343. if (quarter_rtt > 0) {
  344. hctx->ccid3hctx_t_last_win_count = now;
  345. hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
  346. min_t(unsigned long, quarter_rtt, 5)) % 16;
  347. ccid3_pr_debug("%s, sk=%p, window changed from "
  348. "%u to %u!\n",
  349. dccp_role(sk), sk,
  350. packet->dccphtx_ccval,
  351. hctx->ccid3hctx_last_win_count);
  352. }
  353. hctx->ccid3hctx_idle = 0;
  354. packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
  355. packet->dccphtx_sent = 1;
  356. }
  357. static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
  358. {
  359. const struct dccp_sock *dp = dccp_sk(sk);
  360. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  361. struct ccid3_options_received *opt_recv;
  362. struct dccp_tx_hist_entry *packet;
  363. struct timeval now;
  364. unsigned long t_nfb;
  365. u32 t_elapsed;
  366. u32 pinv;
  367. u32 x_recv;
  368. u32 r_sample;
  369. BUG_ON(hctx == NULL);
  370. /* we are only interested in ACKs */
  371. if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
  372. DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
  373. return;
  374. opt_recv = &hctx->ccid3hctx_options_received;
  375. t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
  376. x_recv = opt_recv->ccid3or_receive_rate;
  377. pinv = opt_recv->ccid3or_loss_event_rate;
  378. switch (hctx->ccid3hctx_state) {
  379. case TFRC_SSTATE_NO_FBACK:
  380. case TFRC_SSTATE_FBACK:
  381. /* Calculate new round trip sample by
  382. * R_sample = (now - t_recvdata) - t_delay */
  383. /* get t_recvdata from history */
  384. packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
  385. DCCP_SKB_CB(skb)->dccpd_ack_seq);
  386. if (unlikely(packet == NULL)) {
  387. DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
  388. "in history!\n", dccp_role(sk), sk,
  389. (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
  390. dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
  391. return;
  392. }
  393. /* Update RTT */
  394. dccp_timestamp(sk, &now);
  395. r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
  396. if (unlikely(r_sample <= t_elapsed))
  397. DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
  398. r_sample, t_elapsed);
  399. else
  400. r_sample -= t_elapsed;
  401. /* Update RTT estimate by
  402. * If (No feedback recv)
  403. * R = R_sample;
  404. * Else
  405. * R = q * R + (1 - q) * R_sample;
  406. *
  407. * q is a constant, RFC 3448 recomments 0.9
  408. */
  409. if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
  410. /* Use Larger Initial Windows [RFC 4342, sec. 5]
  411. * We deviate in that we use `s' instead of `MSS'. */
  412. u16 w_init = max( 4 * hctx->ccid3hctx_s,
  413. max(2 * hctx->ccid3hctx_s, 4380));
  414. hctx->ccid3hctx_rtt = r_sample;
  415. hctx->ccid3hctx_x = usecs_div(w_init, r_sample);
  416. hctx->ccid3hctx_t_ld = now;
  417. ccid3_update_send_time(hctx);
  418. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
  419. } else {
  420. hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
  421. r_sample / 10;
  422. ccid3_hc_tx_update_x(sk, &now);
  423. }
  424. ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
  425. "r_sample=%us\n", dccp_role(sk), sk,
  426. hctx->ccid3hctx_rtt, r_sample);
  427. /* Update receive rate */
  428. hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
  429. /* Update loss event rate */
  430. if (pinv == ~0 || pinv == 0)
  431. hctx->ccid3hctx_p = 0;
  432. else {
  433. hctx->ccid3hctx_p = 1000000 / pinv;
  434. if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
  435. hctx->ccid3hctx_p = TFRC_SMALLEST_P;
  436. ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
  437. dccp_role(sk), sk);
  438. }
  439. }
  440. /* unschedule no feedback timer */
  441. sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
  442. /* remove all packets older than the one acked from history */
  443. dccp_tx_hist_purge_older(ccid3_tx_hist,
  444. &hctx->ccid3hctx_hist, packet);
  445. /*
  446. * As we have calculated new ipi, delta, t_nom it is possible that
  447. * we now can send a packet, so wake up dccp_wait_for_ccid
  448. */
  449. sk->sk_write_space(sk);
  450. /*
  451. * Update timeout interval for the nofeedback timer.
  452. * We use a configuration option to increase the lower bound.
  453. * This can help avoid triggering the nofeedback timer too often
  454. * ('spinning') on LANs with small RTTs.
  455. */
  456. hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
  457. CONFIG_IP_DCCP_CCID3_RTO *
  458. (USEC_PER_SEC/1000) );
  459. /*
  460. * Schedule no feedback timer to expire in
  461. * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
  462. */
  463. t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
  464. ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
  465. "expire in %lu jiffies (%luus)\n",
  466. dccp_role(sk), sk,
  467. usecs_to_jiffies(t_nfb), t_nfb);
  468. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  469. jiffies + usecs_to_jiffies(t_nfb));
  470. /* set idle flag */
  471. hctx->ccid3hctx_idle = 1;
  472. break;
  473. case TFRC_SSTATE_NO_SENT:
  474. DCCP_WARN("Illegal ACK received - no packet has been sent\n");
  475. /* fall through */
  476. case TFRC_SSTATE_TERM: /* ignore feedback when closing */
  477. break;
  478. }
  479. }
  480. static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
  481. {
  482. const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  483. BUG_ON(hctx == NULL);
  484. if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
  485. DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
  486. return 0;
  487. }
  488. static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
  489. unsigned char len, u16 idx,
  490. unsigned char *value)
  491. {
  492. int rc = 0;
  493. const struct dccp_sock *dp = dccp_sk(sk);
  494. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  495. struct ccid3_options_received *opt_recv;
  496. BUG_ON(hctx == NULL);
  497. opt_recv = &hctx->ccid3hctx_options_received;
  498. if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
  499. opt_recv->ccid3or_seqno = dp->dccps_gsr;
  500. opt_recv->ccid3or_loss_event_rate = ~0;
  501. opt_recv->ccid3or_loss_intervals_idx = 0;
  502. opt_recv->ccid3or_loss_intervals_len = 0;
  503. opt_recv->ccid3or_receive_rate = 0;
  504. }
  505. switch (option) {
  506. case TFRC_OPT_LOSS_EVENT_RATE:
  507. if (unlikely(len != 4)) {
  508. DCCP_WARN("%s, sk=%p, invalid len %d "
  509. "for TFRC_OPT_LOSS_EVENT_RATE\n",
  510. dccp_role(sk), sk, len);
  511. rc = -EINVAL;
  512. } else {
  513. opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
  514. ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
  515. dccp_role(sk), sk,
  516. opt_recv->ccid3or_loss_event_rate);
  517. }
  518. break;
  519. case TFRC_OPT_LOSS_INTERVALS:
  520. opt_recv->ccid3or_loss_intervals_idx = idx;
  521. opt_recv->ccid3or_loss_intervals_len = len;
  522. ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
  523. dccp_role(sk), sk,
  524. opt_recv->ccid3or_loss_intervals_idx,
  525. opt_recv->ccid3or_loss_intervals_len);
  526. break;
  527. case TFRC_OPT_RECEIVE_RATE:
  528. if (unlikely(len != 4)) {
  529. DCCP_WARN("%s, sk=%p, invalid len %d "
  530. "for TFRC_OPT_RECEIVE_RATE\n",
  531. dccp_role(sk), sk, len);
  532. rc = -EINVAL;
  533. } else {
  534. opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
  535. ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
  536. dccp_role(sk), sk,
  537. opt_recv->ccid3or_receive_rate);
  538. }
  539. break;
  540. }
  541. return rc;
  542. }
  543. static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
  544. {
  545. struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
  546. hctx->ccid3hctx_s = 0;
  547. hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
  548. INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
  549. hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
  550. hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
  551. init_timer(&hctx->ccid3hctx_no_feedback_timer);
  552. return 0;
  553. }
  554. static void ccid3_hc_tx_exit(struct sock *sk)
  555. {
  556. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  557. BUG_ON(hctx == NULL);
  558. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
  559. sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
  560. /* Empty packet history */
  561. dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
  562. }
  563. /*
  564. * RX Half Connection methods
  565. */
  566. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  567. static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
  568. {
  569. static char *ccid3_rx_state_names[] = {
  570. [TFRC_RSTATE_NO_DATA] = "NO_DATA",
  571. [TFRC_RSTATE_DATA] = "DATA",
  572. [TFRC_RSTATE_TERM] = "TERM",
  573. };
  574. return ccid3_rx_state_names[state];
  575. }
  576. #endif
  577. static void ccid3_hc_rx_set_state(struct sock *sk,
  578. enum ccid3_hc_rx_states state)
  579. {
  580. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  581. enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
  582. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  583. dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
  584. ccid3_rx_state_name(state));
  585. WARN_ON(state == oldstate);
  586. hcrx->ccid3hcrx_state = state;
  587. }
  588. static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
  589. {
  590. if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
  591. ccid3_pr_debug("Packet payload length is 0 - not updating\n");
  592. else
  593. hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
  594. (9 * hcrx->ccid3hcrx_s + len) / 10;
  595. }
  596. static void ccid3_hc_rx_send_feedback(struct sock *sk)
  597. {
  598. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  599. struct dccp_sock *dp = dccp_sk(sk);
  600. struct dccp_rx_hist_entry *packet;
  601. struct timeval now;
  602. ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
  603. dccp_timestamp(sk, &now);
  604. switch (hcrx->ccid3hcrx_state) {
  605. case TFRC_RSTATE_NO_DATA:
  606. hcrx->ccid3hcrx_x_recv = 0;
  607. break;
  608. case TFRC_RSTATE_DATA: {
  609. const u32 delta = timeval_delta(&now,
  610. &hcrx->ccid3hcrx_tstamp_last_feedback);
  611. hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
  612. delta);
  613. }
  614. break;
  615. case TFRC_RSTATE_TERM:
  616. DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
  617. return;
  618. }
  619. packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
  620. if (unlikely(packet == NULL)) {
  621. DCCP_WARN("%s, sk=%p, no data packet in history!\n",
  622. dccp_role(sk), sk);
  623. return;
  624. }
  625. hcrx->ccid3hcrx_tstamp_last_feedback = now;
  626. hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
  627. hcrx->ccid3hcrx_bytes_recv = 0;
  628. /* Convert to multiples of 10us */
  629. hcrx->ccid3hcrx_elapsed_time =
  630. timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
  631. if (hcrx->ccid3hcrx_p == 0)
  632. hcrx->ccid3hcrx_pinv = ~0;
  633. else
  634. hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
  635. dp->dccps_hc_rx_insert_options = 1;
  636. dccp_send_ack(sk);
  637. }
  638. static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
  639. {
  640. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  641. __be32 x_recv, pinv;
  642. BUG_ON(hcrx == NULL);
  643. if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
  644. return 0;
  645. DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
  646. if (dccp_packet_without_ack(skb))
  647. return 0;
  648. x_recv = htonl(hcrx->ccid3hcrx_x_recv);
  649. pinv = htonl(hcrx->ccid3hcrx_pinv);
  650. if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
  651. dccp_insert_option_elapsed_time(sk, skb,
  652. hcrx->ccid3hcrx_elapsed_time)) ||
  653. dccp_insert_option_timestamp(sk, skb) ||
  654. dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
  655. &pinv, sizeof(pinv)) ||
  656. dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
  657. &x_recv, sizeof(x_recv)))
  658. return -1;
  659. return 0;
  660. }
  661. /* calculate first loss interval
  662. *
  663. * returns estimated loss interval in usecs */
  664. static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
  665. {
  666. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  667. struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
  668. u32 rtt, delta, x_recv, fval, p, tmp2;
  669. struct timeval tstamp = { 0, };
  670. int interval = 0;
  671. int win_count = 0;
  672. int step = 0;
  673. u64 tmp1;
  674. list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
  675. dccphrx_node) {
  676. if (dccp_rx_hist_entry_data_packet(entry)) {
  677. tail = entry;
  678. switch (step) {
  679. case 0:
  680. tstamp = entry->dccphrx_tstamp;
  681. win_count = entry->dccphrx_ccval;
  682. step = 1;
  683. break;
  684. case 1:
  685. interval = win_count - entry->dccphrx_ccval;
  686. if (interval < 0)
  687. interval += TFRC_WIN_COUNT_LIMIT;
  688. if (interval > 4)
  689. goto found;
  690. break;
  691. }
  692. }
  693. }
  694. if (unlikely(step == 0)) {
  695. DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
  696. dccp_role(sk), sk);
  697. return ~0;
  698. }
  699. if (unlikely(interval == 0)) {
  700. DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
  701. "Defaulting to 1\n", dccp_role(sk), sk);
  702. interval = 1;
  703. }
  704. found:
  705. if (!tail) {
  706. DCCP_CRIT("tail is null\n");
  707. return ~0;
  708. }
  709. rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
  710. ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
  711. dccp_role(sk), sk, rtt);
  712. if (rtt == 0) {
  713. DCCP_WARN("RTT==0, setting to 1\n");
  714. rtt = 1;
  715. }
  716. dccp_timestamp(sk, &tstamp);
  717. delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
  718. x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
  719. if (x_recv == 0)
  720. x_recv = hcrx->ccid3hcrx_x_recv;
  721. tmp1 = (u64)x_recv * (u64)rtt;
  722. do_div(tmp1,10000000);
  723. tmp2 = (u32)tmp1;
  724. if (!tmp2) {
  725. DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
  726. return ~0;
  727. }
  728. fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
  729. /* do not alter order above or you will get overflow on 32 bit */
  730. p = tfrc_calc_x_reverse_lookup(fval);
  731. ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
  732. "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
  733. if (p == 0)
  734. return ~0;
  735. else
  736. return 1000000 / p;
  737. }
  738. static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
  739. {
  740. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  741. struct dccp_li_hist_entry *head;
  742. u64 seq_temp;
  743. if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
  744. if (!dccp_li_hist_interval_new(ccid3_li_hist,
  745. &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
  746. return;
  747. head = list_entry(hcrx->ccid3hcrx_li_hist.next,
  748. struct dccp_li_hist_entry, dccplih_node);
  749. head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
  750. } else {
  751. struct dccp_li_hist_entry *entry;
  752. struct list_head *tail;
  753. head = list_entry(hcrx->ccid3hcrx_li_hist.next,
  754. struct dccp_li_hist_entry, dccplih_node);
  755. /* FIXME win count check removed as was wrong */
  756. /* should make this check with receive history */
  757. /* and compare there as per section 10.2 of RFC4342 */
  758. /* new loss event detected */
  759. /* calculate last interval length */
  760. seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
  761. entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
  762. if (entry == NULL) {
  763. DCCP_BUG("out of memory - can not allocate entry");
  764. return;
  765. }
  766. list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
  767. tail = hcrx->ccid3hcrx_li_hist.prev;
  768. list_del(tail);
  769. kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
  770. /* Create the newest interval */
  771. entry->dccplih_seqno = seq_loss;
  772. entry->dccplih_interval = seq_temp;
  773. entry->dccplih_win_count = win_loss;
  774. }
  775. }
  776. static int ccid3_hc_rx_detect_loss(struct sock *sk,
  777. struct dccp_rx_hist_entry *packet)
  778. {
  779. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  780. struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
  781. u64 seqno = packet->dccphrx_seqno;
  782. u64 tmp_seqno;
  783. int loss = 0;
  784. u8 ccval;
  785. tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
  786. if (!rx_hist ||
  787. follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
  788. hcrx->ccid3hcrx_seqno_nonloss = seqno;
  789. hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
  790. goto detect_out;
  791. }
  792. while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
  793. > TFRC_RECV_NUM_LATE_LOSS) {
  794. loss = 1;
  795. ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
  796. hcrx->ccid3hcrx_ccval_nonloss);
  797. tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
  798. dccp_inc_seqno(&tmp_seqno);
  799. hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
  800. dccp_inc_seqno(&tmp_seqno);
  801. while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
  802. tmp_seqno, &ccval)) {
  803. hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
  804. hcrx->ccid3hcrx_ccval_nonloss = ccval;
  805. dccp_inc_seqno(&tmp_seqno);
  806. }
  807. }
  808. /* FIXME - this code could be simplified with above while */
  809. /* but works at moment */
  810. if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
  811. hcrx->ccid3hcrx_seqno_nonloss = seqno;
  812. hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
  813. }
  814. detect_out:
  815. dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
  816. &hcrx->ccid3hcrx_li_hist, packet,
  817. hcrx->ccid3hcrx_seqno_nonloss);
  818. return loss;
  819. }
  820. static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
  821. {
  822. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  823. const struct dccp_options_received *opt_recv;
  824. struct dccp_rx_hist_entry *packet;
  825. struct timeval now;
  826. u32 p_prev, rtt_prev, r_sample, t_elapsed;
  827. int loss, payload_size;
  828. BUG_ON(hcrx == NULL);
  829. opt_recv = &dccp_sk(sk)->dccps_options_received;
  830. switch (DCCP_SKB_CB(skb)->dccpd_type) {
  831. case DCCP_PKT_ACK:
  832. if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
  833. return;
  834. case DCCP_PKT_DATAACK:
  835. if (opt_recv->dccpor_timestamp_echo == 0)
  836. break;
  837. rtt_prev = hcrx->ccid3hcrx_rtt;
  838. dccp_timestamp(sk, &now);
  839. timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
  840. r_sample = timeval_usecs(&now);
  841. t_elapsed = opt_recv->dccpor_elapsed_time * 10;
  842. if (unlikely(r_sample <= t_elapsed))
  843. DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
  844. r_sample, t_elapsed);
  845. else
  846. r_sample -= t_elapsed;
  847. if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
  848. hcrx->ccid3hcrx_rtt = r_sample;
  849. else
  850. hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
  851. r_sample / 10;
  852. if (rtt_prev != hcrx->ccid3hcrx_rtt)
  853. ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
  854. dccp_role(sk), hcrx->ccid3hcrx_rtt,
  855. opt_recv->dccpor_elapsed_time);
  856. break;
  857. case DCCP_PKT_DATA:
  858. break;
  859. default: /* We're not interested in other packet types, move along */
  860. return;
  861. }
  862. packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
  863. skb, SLAB_ATOMIC);
  864. if (unlikely(packet == NULL)) {
  865. DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
  866. "to history, consider it lost!\n", dccp_role(sk), sk);
  867. return;
  868. }
  869. loss = ccid3_hc_rx_detect_loss(sk, packet);
  870. if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
  871. return;
  872. payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
  873. ccid3_hc_rx_update_s(hcrx, payload_size);
  874. switch (hcrx->ccid3hcrx_state) {
  875. case TFRC_RSTATE_NO_DATA:
  876. ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
  877. "feedback\n",
  878. dccp_role(sk), sk,
  879. dccp_state_name(sk->sk_state), skb);
  880. ccid3_hc_rx_send_feedback(sk);
  881. ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
  882. return;
  883. case TFRC_RSTATE_DATA:
  884. hcrx->ccid3hcrx_bytes_recv += payload_size;
  885. if (loss)
  886. break;
  887. dccp_timestamp(sk, &now);
  888. if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
  889. hcrx->ccid3hcrx_rtt) {
  890. hcrx->ccid3hcrx_tstamp_last_ack = now;
  891. ccid3_hc_rx_send_feedback(sk);
  892. }
  893. return;
  894. case TFRC_RSTATE_TERM:
  895. DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
  896. return;
  897. }
  898. /* Dealing with packet loss */
  899. ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
  900. dccp_role(sk), sk, dccp_state_name(sk->sk_state));
  901. p_prev = hcrx->ccid3hcrx_p;
  902. /* Calculate loss event rate */
  903. if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
  904. u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
  905. /* Scaling up by 1000000 as fixed decimal */
  906. if (i_mean != 0)
  907. hcrx->ccid3hcrx_p = 1000000 / i_mean;
  908. } else
  909. DCCP_BUG("empty loss history");
  910. if (hcrx->ccid3hcrx_p > p_prev) {
  911. ccid3_hc_rx_send_feedback(sk);
  912. return;
  913. }
  914. }
  915. static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
  916. {
  917. struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
  918. ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
  919. hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
  920. INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
  921. INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
  922. dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
  923. hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
  924. hcrx->ccid3hcrx_s = 0;
  925. hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
  926. return 0;
  927. }
  928. static void ccid3_hc_rx_exit(struct sock *sk)
  929. {
  930. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  931. BUG_ON(hcrx == NULL);
  932. ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
  933. /* Empty packet history */
  934. dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
  935. /* Empty loss interval history */
  936. dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
  937. }
  938. static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
  939. {
  940. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  941. /* Listen socks doesn't have a private CCID block */
  942. if (sk->sk_state == DCCP_LISTEN)
  943. return;
  944. BUG_ON(hcrx == NULL);
  945. info->tcpi_ca_state = hcrx->ccid3hcrx_state;
  946. info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
  947. info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
  948. }
  949. static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
  950. {
  951. const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  952. /* Listen socks doesn't have a private CCID block */
  953. if (sk->sk_state == DCCP_LISTEN)
  954. return;
  955. BUG_ON(hctx == NULL);
  956. info->tcpi_rto = hctx->ccid3hctx_t_rto;
  957. info->tcpi_rtt = hctx->ccid3hctx_rtt;
  958. }
  959. static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
  960. u32 __user *optval, int __user *optlen)
  961. {
  962. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  963. const void *val;
  964. /* Listen socks doesn't have a private CCID block */
  965. if (sk->sk_state == DCCP_LISTEN)
  966. return -EINVAL;
  967. switch (optname) {
  968. case DCCP_SOCKOPT_CCID_RX_INFO:
  969. if (len < sizeof(hcrx->ccid3hcrx_tfrc))
  970. return -EINVAL;
  971. len = sizeof(hcrx->ccid3hcrx_tfrc);
  972. val = &hcrx->ccid3hcrx_tfrc;
  973. break;
  974. default:
  975. return -ENOPROTOOPT;
  976. }
  977. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  978. return -EFAULT;
  979. return 0;
  980. }
  981. static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
  982. u32 __user *optval, int __user *optlen)
  983. {
  984. const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  985. const void *val;
  986. /* Listen socks doesn't have a private CCID block */
  987. if (sk->sk_state == DCCP_LISTEN)
  988. return -EINVAL;
  989. switch (optname) {
  990. case DCCP_SOCKOPT_CCID_TX_INFO:
  991. if (len < sizeof(hctx->ccid3hctx_tfrc))
  992. return -EINVAL;
  993. len = sizeof(hctx->ccid3hctx_tfrc);
  994. val = &hctx->ccid3hctx_tfrc;
  995. break;
  996. default:
  997. return -ENOPROTOOPT;
  998. }
  999. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  1000. return -EFAULT;
  1001. return 0;
  1002. }
  1003. static struct ccid_operations ccid3 = {
  1004. .ccid_id = DCCPC_CCID3,
  1005. .ccid_name = "ccid3",
  1006. .ccid_owner = THIS_MODULE,
  1007. .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
  1008. .ccid_hc_tx_init = ccid3_hc_tx_init,
  1009. .ccid_hc_tx_exit = ccid3_hc_tx_exit,
  1010. .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
  1011. .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
  1012. .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
  1013. .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
  1014. .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
  1015. .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
  1016. .ccid_hc_rx_init = ccid3_hc_rx_init,
  1017. .ccid_hc_rx_exit = ccid3_hc_rx_exit,
  1018. .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
  1019. .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
  1020. .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
  1021. .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
  1022. .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
  1023. .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
  1024. };
  1025. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  1026. module_param(ccid3_debug, int, 0444);
  1027. MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
  1028. #endif
  1029. static __init int ccid3_module_init(void)
  1030. {
  1031. int rc = -ENOBUFS;
  1032. ccid3_rx_hist = dccp_rx_hist_new("ccid3");
  1033. if (ccid3_rx_hist == NULL)
  1034. goto out;
  1035. ccid3_tx_hist = dccp_tx_hist_new("ccid3");
  1036. if (ccid3_tx_hist == NULL)
  1037. goto out_free_rx;
  1038. ccid3_li_hist = dccp_li_hist_new("ccid3");
  1039. if (ccid3_li_hist == NULL)
  1040. goto out_free_tx;
  1041. rc = ccid_register(&ccid3);
  1042. if (rc != 0)
  1043. goto out_free_loss_interval_history;
  1044. out:
  1045. return rc;
  1046. out_free_loss_interval_history:
  1047. dccp_li_hist_delete(ccid3_li_hist);
  1048. ccid3_li_hist = NULL;
  1049. out_free_tx:
  1050. dccp_tx_hist_delete(ccid3_tx_hist);
  1051. ccid3_tx_hist = NULL;
  1052. out_free_rx:
  1053. dccp_rx_hist_delete(ccid3_rx_hist);
  1054. ccid3_rx_hist = NULL;
  1055. goto out;
  1056. }
  1057. module_init(ccid3_module_init);
  1058. static __exit void ccid3_module_exit(void)
  1059. {
  1060. ccid_unregister(&ccid3);
  1061. if (ccid3_tx_hist != NULL) {
  1062. dccp_tx_hist_delete(ccid3_tx_hist);
  1063. ccid3_tx_hist = NULL;
  1064. }
  1065. if (ccid3_rx_hist != NULL) {
  1066. dccp_rx_hist_delete(ccid3_rx_hist);
  1067. ccid3_rx_hist = NULL;
  1068. }
  1069. if (ccid3_li_hist != NULL) {
  1070. dccp_li_hist_delete(ccid3_li_hist);
  1071. ccid3_li_hist = NULL;
  1072. }
  1073. }
  1074. module_exit(ccid3_module_exit);
  1075. MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
  1076. "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
  1077. MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
  1078. MODULE_LICENSE("GPL");
  1079. MODULE_ALIAS("net-dccp-ccid-3");