ccid3.c 36 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288
  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. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  43. static int ccid3_debug;
  44. #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
  45. #else
  46. #define ccid3_pr_debug(format, a...)
  47. #endif
  48. static struct dccp_tx_hist *ccid3_tx_hist;
  49. static struct dccp_rx_hist *ccid3_rx_hist;
  50. static struct dccp_li_hist *ccid3_li_hist;
  51. /*
  52. * Transmitter Half-Connection Routines
  53. */
  54. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  55. static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
  56. {
  57. static char *ccid3_state_names[] = {
  58. [TFRC_SSTATE_NO_SENT] = "NO_SENT",
  59. [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
  60. [TFRC_SSTATE_FBACK] = "FBACK",
  61. [TFRC_SSTATE_TERM] = "TERM",
  62. };
  63. return ccid3_state_names[state];
  64. }
  65. #endif
  66. static void ccid3_hc_tx_set_state(struct sock *sk,
  67. enum ccid3_hc_tx_states state)
  68. {
  69. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  70. enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
  71. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  72. dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
  73. ccid3_tx_state_name(state));
  74. WARN_ON(state == oldstate);
  75. hctx->ccid3hctx_state = state;
  76. }
  77. /*
  78. * Recalculate t_ipi and delta (should be called whenever X changes)
  79. */
  80. static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
  81. {
  82. /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
  83. hctx->ccid3hctx_t_ipi = scaled_div(hctx->ccid3hctx_s,
  84. hctx->ccid3hctx_x >> 6);
  85. /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
  86. hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
  87. TFRC_OPSYS_HALF_TIME_GRAN);
  88. ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
  89. hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
  90. hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
  91. }
  92. /*
  93. * Update X by
  94. * If (p > 0)
  95. * X_calc = calcX(s, R, p);
  96. * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
  97. * Else
  98. * If (now - tld >= R)
  99. * X = max(min(2 * X, 2 * X_recv), s / R);
  100. * tld = now;
  101. *
  102. * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
  103. * fine-grained resolution of sending rates. This requires scaling by 2^6
  104. * throughout the code. Only X_calc is unscaled (in bytes/second).
  105. *
  106. */
  107. static void ccid3_hc_tx_update_x(struct sock *sk, struct timeval *now)
  108. {
  109. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  110. const __u64 old_x = hctx->ccid3hctx_x;
  111. if (hctx->ccid3hctx_p > 0) {
  112. hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
  113. hctx->ccid3hctx_x_recv * 2);
  114. hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
  115. (((__u64)hctx->ccid3hctx_s) << 6) /
  116. TFRC_T_MBI);
  117. } else if (timeval_delta(now, &hctx->ccid3hctx_t_ld) -
  118. (suseconds_t)hctx->ccid3hctx_rtt >= 0) {
  119. hctx->ccid3hctx_x =
  120. max(2 * min(hctx->ccid3hctx_x, hctx->ccid3hctx_x_recv),
  121. scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
  122. hctx->ccid3hctx_rtt));
  123. hctx->ccid3hctx_t_ld = *now;
  124. }
  125. if (hctx->ccid3hctx_x != old_x) {
  126. ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
  127. "X_recv=%u\n", (unsigned)(old_x >> 6),
  128. (unsigned)(hctx->ccid3hctx_x >> 6),
  129. hctx->ccid3hctx_x_calc,
  130. (unsigned)(hctx->ccid3hctx_x_recv >> 6));
  131. ccid3_update_send_interval(hctx);
  132. }
  133. }
  134. /*
  135. * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
  136. * @len: DCCP packet payload size in bytes
  137. */
  138. static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
  139. {
  140. const u16 old_s = hctx->ccid3hctx_s;
  141. hctx->ccid3hctx_s = old_s == 0 ? len : (9 * old_s + len) / 10;
  142. if (hctx->ccid3hctx_s != old_s)
  143. ccid3_update_send_interval(hctx);
  144. }
  145. /*
  146. * Update Window Counter using the algorithm from [RFC 4342, 8.1].
  147. * The algorithm is not applicable if RTT < 4 microseconds.
  148. */
  149. static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
  150. struct timeval *now)
  151. {
  152. suseconds_t delta;
  153. u32 quarter_rtts;
  154. if (unlikely(hctx->ccid3hctx_rtt < 4)) /* avoid divide-by-zero */
  155. return;
  156. delta = timeval_delta(now, &hctx->ccid3hctx_t_last_win_count);
  157. DCCP_BUG_ON(delta < 0);
  158. quarter_rtts = (u32)delta / (hctx->ccid3hctx_rtt / 4);
  159. if (quarter_rtts > 0) {
  160. hctx->ccid3hctx_t_last_win_count = *now;
  161. hctx->ccid3hctx_last_win_count += min_t(u32, quarter_rtts, 5);
  162. hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
  163. ccid3_pr_debug("now at %#X\n", hctx->ccid3hctx_last_win_count);
  164. }
  165. }
  166. static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
  167. {
  168. struct sock *sk = (struct sock *)data;
  169. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  170. unsigned long t_nfb = USEC_PER_SEC / 5;
  171. bh_lock_sock(sk);
  172. if (sock_owned_by_user(sk)) {
  173. /* Try again later. */
  174. /* XXX: set some sensible MIB */
  175. goto restart_timer;
  176. }
  177. ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
  178. ccid3_tx_state_name(hctx->ccid3hctx_state));
  179. switch (hctx->ccid3hctx_state) {
  180. case TFRC_SSTATE_NO_FBACK:
  181. /* RFC 3448, 4.4: Halve send rate directly */
  182. hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
  183. (((__u64)hctx->ccid3hctx_s) << 6) /
  184. TFRC_T_MBI);
  185. ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
  186. "bytes/s\n", dccp_role(sk), sk,
  187. ccid3_tx_state_name(hctx->ccid3hctx_state),
  188. (unsigned)(hctx->ccid3hctx_x >> 6));
  189. /* The value of R is still undefined and so we can not recompute
  190. * the timout value. Keep initial value as per [RFC 4342, 5]. */
  191. t_nfb = TFRC_INITIAL_TIMEOUT;
  192. ccid3_update_send_interval(hctx);
  193. break;
  194. case TFRC_SSTATE_FBACK:
  195. /*
  196. * Check if IDLE since last timeout and recv rate is less than
  197. * 4 packets (in units of 64*bytes/sec) per RTT
  198. */
  199. if (!hctx->ccid3hctx_idle ||
  200. (hctx->ccid3hctx_x_recv >= 4 *
  201. scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
  202. hctx->ccid3hctx_rtt))) {
  203. struct timeval now;
  204. ccid3_pr_debug("%s(%p, state=%s), not idle\n",
  205. dccp_role(sk), sk,
  206. ccid3_tx_state_name(hctx->ccid3hctx_state));
  207. /*
  208. * Modify the cached value of X_recv [RFC 3448, 4.4]
  209. *
  210. * If (p == 0 || X_calc > 2 * X_recv)
  211. * X_recv = max(X_recv / 2, s / (2 * t_mbi));
  212. * Else
  213. * X_recv = X_calc / 4;
  214. *
  215. * Note that X_recv is scaled by 2^6 while X_calc is not
  216. */
  217. BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
  218. if (hctx->ccid3hctx_p == 0 ||
  219. (hctx->ccid3hctx_x_calc >
  220. (hctx->ccid3hctx_x_recv >> 5))) {
  221. hctx->ccid3hctx_x_recv =
  222. max(hctx->ccid3hctx_x_recv / 2,
  223. (((__u64)hctx->ccid3hctx_s) << 6) /
  224. (2 * TFRC_T_MBI));
  225. if (hctx->ccid3hctx_p == 0)
  226. dccp_timestamp(sk, &now);
  227. } else {
  228. hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
  229. hctx->ccid3hctx_x_recv <<= 4;
  230. }
  231. /* Now recalculate X [RFC 3448, 4.3, step (4)] */
  232. ccid3_hc_tx_update_x(sk, &now);
  233. }
  234. /*
  235. * Schedule no feedback timer to expire in
  236. * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
  237. * See comments in packet_recv() regarding the value of t_RTO.
  238. */
  239. t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
  240. break;
  241. case TFRC_SSTATE_NO_SENT:
  242. DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk), sk);
  243. /* fall through */
  244. case TFRC_SSTATE_TERM:
  245. goto out;
  246. }
  247. hctx->ccid3hctx_idle = 1;
  248. restart_timer:
  249. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  250. jiffies + usecs_to_jiffies(t_nfb));
  251. out:
  252. bh_unlock_sock(sk);
  253. sock_put(sk);
  254. }
  255. /*
  256. * returns
  257. * > 0: delay (in msecs) that should pass before actually sending
  258. * = 0: can send immediately
  259. * < 0: error condition; do not send packet
  260. */
  261. static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
  262. {
  263. struct dccp_sock *dp = dccp_sk(sk);
  264. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  265. struct timeval now;
  266. suseconds_t delay;
  267. BUG_ON(hctx == NULL);
  268. /*
  269. * This function is called only for Data and DataAck packets. Sending
  270. * zero-sized Data(Ack)s is theoretically possible, but for congestion
  271. * control this case is pathological - ignore it.
  272. */
  273. if (unlikely(skb->len == 0))
  274. return -EBADMSG;
  275. dccp_timestamp(sk, &now);
  276. switch (hctx->ccid3hctx_state) {
  277. case TFRC_SSTATE_NO_SENT:
  278. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  279. (jiffies +
  280. usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
  281. hctx->ccid3hctx_last_win_count = 0;
  282. hctx->ccid3hctx_t_last_win_count = now;
  283. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
  284. /* Set initial sending rate X/s to 1pps (X is scaled by 2^6) */
  285. hctx->ccid3hctx_x = hctx->ccid3hctx_s = skb->len;
  286. hctx->ccid3hctx_x <<= 6;
  287. /* First timeout, according to [RFC 3448, 4.2], is 1 second */
  288. hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
  289. /* Initial delta: minimum of 0.5 sec and t_gran/2 */
  290. hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
  291. /* Set t_0 for initial packet */
  292. hctx->ccid3hctx_t_nom = now;
  293. break;
  294. case TFRC_SSTATE_NO_FBACK:
  295. case TFRC_SSTATE_FBACK:
  296. delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
  297. ccid3_pr_debug("delay=%ld\n", (long)delay);
  298. /*
  299. * Scheduling of packet transmissions [RFC 3448, 4.6]
  300. *
  301. * if (t_now > t_nom - delta)
  302. * // send the packet now
  303. * else
  304. * // send the packet in (t_nom - t_now) milliseconds.
  305. */
  306. if (delay - (suseconds_t)hctx->ccid3hctx_delta >= 0)
  307. return delay / 1000L;
  308. ccid3_hc_tx_update_win_count(hctx, &now);
  309. break;
  310. case TFRC_SSTATE_TERM:
  311. DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
  312. return -EINVAL;
  313. }
  314. /* prepare to send now (add options etc.) */
  315. dp->dccps_hc_tx_insert_options = 1;
  316. DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
  317. /* set the nominal send time for the next following packet */
  318. timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
  319. return 0;
  320. }
  321. static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
  322. unsigned int len)
  323. {
  324. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  325. struct timeval now;
  326. struct dccp_tx_hist_entry *packet;
  327. BUG_ON(hctx == NULL);
  328. ccid3_hc_tx_update_s(hctx, len);
  329. packet = dccp_tx_hist_entry_new(ccid3_tx_hist, GFP_ATOMIC);
  330. if (unlikely(packet == NULL)) {
  331. DCCP_CRIT("packet history - out of memory!");
  332. return;
  333. }
  334. dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
  335. dccp_timestamp(sk, &now);
  336. packet->dccphtx_tstamp = now;
  337. packet->dccphtx_seqno = dccp_sk(sk)->dccps_gss;
  338. packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
  339. packet->dccphtx_sent = 1;
  340. hctx->ccid3hctx_idle = 0;
  341. }
  342. static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
  343. {
  344. const struct dccp_sock *dp = dccp_sk(sk);
  345. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  346. struct ccid3_options_received *opt_recv;
  347. struct dccp_tx_hist_entry *packet;
  348. struct timeval now;
  349. unsigned long t_nfb;
  350. u32 pinv;
  351. suseconds_t r_sample, t_elapsed;
  352. BUG_ON(hctx == NULL);
  353. /* we are only interested in ACKs */
  354. if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
  355. DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
  356. return;
  357. opt_recv = &hctx->ccid3hctx_options_received;
  358. switch (hctx->ccid3hctx_state) {
  359. case TFRC_SSTATE_NO_FBACK:
  360. case TFRC_SSTATE_FBACK:
  361. /* get packet from history to look up t_recvdata */
  362. packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
  363. DCCP_SKB_CB(skb)->dccpd_ack_seq);
  364. if (unlikely(packet == NULL)) {
  365. DCCP_WARN("%s(%p), seqno %llu(%s) doesn't exist "
  366. "in history!\n", dccp_role(sk), sk,
  367. (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
  368. dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
  369. return;
  370. }
  371. /* Update receive rate in units of 64 * bytes/second */
  372. hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
  373. hctx->ccid3hctx_x_recv <<= 6;
  374. /* Update loss event rate */
  375. pinv = opt_recv->ccid3or_loss_event_rate;
  376. if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
  377. hctx->ccid3hctx_p = 0;
  378. else /* can not exceed 100% */
  379. hctx->ccid3hctx_p = 1000000 / pinv;
  380. dccp_timestamp(sk, &now);
  381. /*
  382. * Calculate new round trip sample as per [RFC 3448, 4.3] by
  383. * R_sample = (now - t_recvdata) - t_elapsed
  384. */
  385. r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
  386. t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
  387. DCCP_BUG_ON(r_sample < 0);
  388. if (unlikely(r_sample <= t_elapsed))
  389. DCCP_WARN("WARNING: r_sample=%dus <= t_elapsed=%dus\n",
  390. (int)r_sample, (int)t_elapsed);
  391. else
  392. r_sample -= t_elapsed;
  393. CCID3_RTT_SANITY_CHECK(r_sample);
  394. /* Update RTT estimate by
  395. * If (No feedback recv)
  396. * R = R_sample;
  397. * Else
  398. * R = q * R + (1 - q) * R_sample;
  399. *
  400. * q is a constant, RFC 3448 recomments 0.9
  401. */
  402. if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
  403. /*
  404. * Larger Initial Windows [RFC 4342, sec. 5]
  405. */
  406. __u32 w_init = min(4 * dp->dccps_mss_cache,
  407. max(2 * dp->dccps_mss_cache, 4380U));
  408. hctx->ccid3hctx_rtt = r_sample;
  409. hctx->ccid3hctx_x = scaled_div(w_init << 6, r_sample);
  410. hctx->ccid3hctx_t_ld = now;
  411. ccid3_update_send_interval(hctx);
  412. ccid3_pr_debug("%s(%p), s=%u, MSS=%u, w_init=%u, "
  413. "R_sample=%dus, X=%u\n", dccp_role(sk),
  414. sk, hctx->ccid3hctx_s,
  415. dp->dccps_mss_cache,
  416. w_init,
  417. (int)r_sample,
  418. (unsigned)(hctx->ccid3hctx_x >> 6));
  419. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
  420. } else {
  421. hctx->ccid3hctx_rtt = (9 * hctx->ccid3hctx_rtt +
  422. (u32)r_sample) / 10;
  423. /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
  424. if (hctx->ccid3hctx_p > 0)
  425. hctx->ccid3hctx_x_calc =
  426. tfrc_calc_x(hctx->ccid3hctx_s,
  427. hctx->ccid3hctx_rtt,
  428. hctx->ccid3hctx_p);
  429. ccid3_hc_tx_update_x(sk, &now);
  430. ccid3_pr_debug("%s(%p), RTT=%uus (sample=%dus), s=%u, "
  431. "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
  432. dccp_role(sk),
  433. sk, hctx->ccid3hctx_rtt, (int)r_sample,
  434. hctx->ccid3hctx_s, hctx->ccid3hctx_p,
  435. hctx->ccid3hctx_x_calc,
  436. (unsigned)(hctx->ccid3hctx_x_recv >> 6),
  437. (unsigned)(hctx->ccid3hctx_x >> 6));
  438. }
  439. /* unschedule no feedback timer */
  440. sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
  441. /* remove all packets older than the one acked from history */
  442. dccp_tx_hist_purge_older(ccid3_tx_hist,
  443. &hctx->ccid3hctx_hist, packet);
  444. /*
  445. * As we have calculated new ipi, delta, t_nom it is possible
  446. * that we now can send a packet, so wake up dccp_wait_for_ccid
  447. */
  448. sk->sk_write_space(sk);
  449. /*
  450. * Update timeout interval for the nofeedback timer.
  451. * We use a configuration option to increase the lower bound.
  452. * This can help avoid triggering the nofeedback timer too
  453. * often ('spinning') on LANs with small RTTs.
  454. */
  455. hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
  456. CONFIG_IP_DCCP_CCID3_RTO *
  457. (USEC_PER_SEC/1000));
  458. /*
  459. * Schedule no feedback timer to expire in
  460. * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
  461. */
  462. t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
  463. ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
  464. "expire in %lu jiffies (%luus)\n",
  465. dccp_role(sk),
  466. sk, usecs_to_jiffies(t_nfb), t_nfb);
  467. sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
  468. jiffies + usecs_to_jiffies(t_nfb));
  469. /* set idle flag */
  470. hctx->ccid3hctx_idle = 1;
  471. break;
  472. case TFRC_SSTATE_NO_SENT: /* fall through */
  473. case TFRC_SSTATE_TERM: /* ignore feedback when closing */
  474. break;
  475. }
  476. }
  477. static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
  478. unsigned char len, u16 idx,
  479. unsigned char *value)
  480. {
  481. int rc = 0;
  482. const struct dccp_sock *dp = dccp_sk(sk);
  483. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  484. struct ccid3_options_received *opt_recv;
  485. BUG_ON(hctx == NULL);
  486. opt_recv = &hctx->ccid3hctx_options_received;
  487. if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
  488. opt_recv->ccid3or_seqno = dp->dccps_gsr;
  489. opt_recv->ccid3or_loss_event_rate = ~0;
  490. opt_recv->ccid3or_loss_intervals_idx = 0;
  491. opt_recv->ccid3or_loss_intervals_len = 0;
  492. opt_recv->ccid3or_receive_rate = 0;
  493. }
  494. switch (option) {
  495. case TFRC_OPT_LOSS_EVENT_RATE:
  496. if (unlikely(len != 4)) {
  497. DCCP_WARN("%s(%p), invalid len %d "
  498. "for TFRC_OPT_LOSS_EVENT_RATE\n",
  499. dccp_role(sk), sk, len);
  500. rc = -EINVAL;
  501. } else {
  502. opt_recv->ccid3or_loss_event_rate =
  503. ntohl(*(__be32 *)value);
  504. ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
  505. dccp_role(sk), sk,
  506. opt_recv->ccid3or_loss_event_rate);
  507. }
  508. break;
  509. case TFRC_OPT_LOSS_INTERVALS:
  510. opt_recv->ccid3or_loss_intervals_idx = idx;
  511. opt_recv->ccid3or_loss_intervals_len = len;
  512. ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
  513. dccp_role(sk), sk,
  514. opt_recv->ccid3or_loss_intervals_idx,
  515. opt_recv->ccid3or_loss_intervals_len);
  516. break;
  517. case TFRC_OPT_RECEIVE_RATE:
  518. if (unlikely(len != 4)) {
  519. DCCP_WARN("%s(%p), invalid len %d "
  520. "for TFRC_OPT_RECEIVE_RATE\n",
  521. dccp_role(sk), sk, len);
  522. rc = -EINVAL;
  523. } else {
  524. opt_recv->ccid3or_receive_rate =
  525. ntohl(*(__be32 *)value);
  526. ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
  527. dccp_role(sk), sk,
  528. opt_recv->ccid3or_receive_rate);
  529. }
  530. break;
  531. }
  532. return rc;
  533. }
  534. static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
  535. {
  536. struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
  537. hctx->ccid3hctx_s = 0;
  538. hctx->ccid3hctx_rtt = 0;
  539. hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
  540. INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
  541. hctx->ccid3hctx_no_feedback_timer.function =
  542. ccid3_hc_tx_no_feedback_timer;
  543. hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
  544. init_timer(&hctx->ccid3hctx_no_feedback_timer);
  545. return 0;
  546. }
  547. static void ccid3_hc_tx_exit(struct sock *sk)
  548. {
  549. struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  550. BUG_ON(hctx == NULL);
  551. ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
  552. sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
  553. /* Empty packet history */
  554. dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
  555. }
  556. static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
  557. {
  558. const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  559. /* Listen socks doesn't have a private CCID block */
  560. if (sk->sk_state == DCCP_LISTEN)
  561. return;
  562. BUG_ON(hctx == NULL);
  563. info->tcpi_rto = hctx->ccid3hctx_t_rto;
  564. info->tcpi_rtt = hctx->ccid3hctx_rtt;
  565. }
  566. static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
  567. u32 __user *optval, int __user *optlen)
  568. {
  569. const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
  570. const void *val;
  571. /* Listen socks doesn't have a private CCID block */
  572. if (sk->sk_state == DCCP_LISTEN)
  573. return -EINVAL;
  574. switch (optname) {
  575. case DCCP_SOCKOPT_CCID_TX_INFO:
  576. if (len < sizeof(hctx->ccid3hctx_tfrc))
  577. return -EINVAL;
  578. len = sizeof(hctx->ccid3hctx_tfrc);
  579. val = &hctx->ccid3hctx_tfrc;
  580. break;
  581. default:
  582. return -ENOPROTOOPT;
  583. }
  584. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  585. return -EFAULT;
  586. return 0;
  587. }
  588. /*
  589. * Receiver Half-Connection Routines
  590. */
  591. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  592. static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
  593. {
  594. static char *ccid3_rx_state_names[] = {
  595. [TFRC_RSTATE_NO_DATA] = "NO_DATA",
  596. [TFRC_RSTATE_DATA] = "DATA",
  597. [TFRC_RSTATE_TERM] = "TERM",
  598. };
  599. return ccid3_rx_state_names[state];
  600. }
  601. #endif
  602. static void ccid3_hc_rx_set_state(struct sock *sk,
  603. enum ccid3_hc_rx_states state)
  604. {
  605. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  606. enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
  607. ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
  608. dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
  609. ccid3_rx_state_name(state));
  610. WARN_ON(state == oldstate);
  611. hcrx->ccid3hcrx_state = state;
  612. }
  613. static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
  614. {
  615. if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
  616. ccid3_pr_debug("Packet payload length is 0 - not updating\n");
  617. else
  618. hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
  619. (9 * hcrx->ccid3hcrx_s + len) / 10;
  620. }
  621. static void ccid3_hc_rx_send_feedback(struct sock *sk)
  622. {
  623. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  624. struct dccp_sock *dp = dccp_sk(sk);
  625. struct dccp_rx_hist_entry *packet;
  626. struct timeval now;
  627. suseconds_t delta;
  628. ccid3_pr_debug("%s(%p) - entry \n", dccp_role(sk), sk);
  629. dccp_timestamp(sk, &now);
  630. switch (hcrx->ccid3hcrx_state) {
  631. case TFRC_RSTATE_NO_DATA:
  632. hcrx->ccid3hcrx_x_recv = 0;
  633. break;
  634. case TFRC_RSTATE_DATA:
  635. delta = timeval_delta(&now,
  636. &hcrx->ccid3hcrx_tstamp_last_feedback);
  637. DCCP_BUG_ON(delta < 0);
  638. hcrx->ccid3hcrx_x_recv =
  639. scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
  640. break;
  641. case TFRC_RSTATE_TERM:
  642. DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
  643. return;
  644. }
  645. packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
  646. if (unlikely(packet == NULL)) {
  647. DCCP_WARN("%s(%p), no data packet in history!\n",
  648. dccp_role(sk), sk);
  649. return;
  650. }
  651. hcrx->ccid3hcrx_tstamp_last_feedback = now;
  652. hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
  653. hcrx->ccid3hcrx_bytes_recv = 0;
  654. /* Elapsed time information [RFC 4340, 13.2] in units of 10 * usecs */
  655. delta = timeval_delta(&now, &packet->dccphrx_tstamp);
  656. DCCP_BUG_ON(delta < 0);
  657. hcrx->ccid3hcrx_elapsed_time = delta / 10;
  658. if (hcrx->ccid3hcrx_p == 0)
  659. hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
  660. else if (hcrx->ccid3hcrx_p > 1000000) {
  661. DCCP_WARN("p (%u) > 100%%\n", hcrx->ccid3hcrx_p);
  662. hcrx->ccid3hcrx_pinv = 1; /* use 100% in this case */
  663. } else
  664. hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
  665. dp->dccps_hc_rx_insert_options = 1;
  666. dccp_send_ack(sk);
  667. }
  668. static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
  669. {
  670. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  671. __be32 x_recv, pinv;
  672. BUG_ON(hcrx == NULL);
  673. if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
  674. return 0;
  675. DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
  676. if (dccp_packet_without_ack(skb))
  677. return 0;
  678. x_recv = htonl(hcrx->ccid3hcrx_x_recv);
  679. pinv = htonl(hcrx->ccid3hcrx_pinv);
  680. if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
  681. dccp_insert_option_elapsed_time(sk, skb,
  682. hcrx->ccid3hcrx_elapsed_time)) ||
  683. dccp_insert_option_timestamp(sk, skb) ||
  684. dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
  685. &pinv, sizeof(pinv)) ||
  686. dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
  687. &x_recv, sizeof(x_recv)))
  688. return -1;
  689. return 0;
  690. }
  691. /* calculate first loss interval
  692. *
  693. * returns estimated loss interval in usecs */
  694. static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
  695. {
  696. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  697. struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
  698. u32 x_recv, p;
  699. suseconds_t rtt, delta;
  700. struct timeval tstamp = { 0, };
  701. int interval = 0;
  702. int win_count = 0;
  703. int step = 0;
  704. u64 fval;
  705. list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
  706. dccphrx_node) {
  707. if (dccp_rx_hist_entry_data_packet(entry)) {
  708. tail = entry;
  709. switch (step) {
  710. case 0:
  711. tstamp = entry->dccphrx_tstamp;
  712. win_count = entry->dccphrx_ccval;
  713. step = 1;
  714. break;
  715. case 1:
  716. interval = win_count - entry->dccphrx_ccval;
  717. if (interval < 0)
  718. interval += TFRC_WIN_COUNT_LIMIT;
  719. if (interval > 4)
  720. goto found;
  721. break;
  722. }
  723. }
  724. }
  725. if (unlikely(step == 0)) {
  726. DCCP_WARN("%s(%p), packet history has no data packets!\n",
  727. dccp_role(sk), sk);
  728. return ~0;
  729. }
  730. if (unlikely(interval == 0)) {
  731. DCCP_WARN("%s(%p), Could not find a win_count interval > 0."
  732. "Defaulting to 1\n", dccp_role(sk), sk);
  733. interval = 1;
  734. }
  735. found:
  736. if (!tail) {
  737. DCCP_CRIT("tail is null\n");
  738. return ~0;
  739. }
  740. delta = timeval_delta(&tstamp, &tail->dccphrx_tstamp);
  741. DCCP_BUG_ON(delta < 0);
  742. rtt = delta * 4 / interval;
  743. ccid3_pr_debug("%s(%p), approximated RTT to %dus\n",
  744. dccp_role(sk), sk, (int)rtt);
  745. /*
  746. * Determine the length of the first loss interval via inverse lookup.
  747. * Assume that X_recv can be computed by the throughput equation
  748. * s
  749. * X_recv = --------
  750. * R * fval
  751. * Find some p such that f(p) = fval; return 1/p [RFC 3448, 6.3.1].
  752. */
  753. if (rtt == 0) { /* would result in divide-by-zero */
  754. DCCP_WARN("RTT==0\n");
  755. return ~0;
  756. }
  757. dccp_timestamp(sk, &tstamp);
  758. delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
  759. DCCP_BUG_ON(delta <= 0);
  760. x_recv = scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
  761. if (x_recv == 0) { /* would also trigger divide-by-zero */
  762. DCCP_WARN("X_recv==0\n");
  763. if ((x_recv = hcrx->ccid3hcrx_x_recv) == 0) {
  764. DCCP_BUG("stored value of X_recv is zero");
  765. return ~0;
  766. }
  767. }
  768. fval = scaled_div(hcrx->ccid3hcrx_s, rtt);
  769. fval = scaled_div32(fval, x_recv);
  770. p = tfrc_calc_x_reverse_lookup(fval);
  771. ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
  772. "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
  773. if (p == 0)
  774. return ~0;
  775. else
  776. return 1000000 / p;
  777. }
  778. static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
  779. {
  780. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  781. struct dccp_li_hist_entry *head;
  782. u64 seq_temp;
  783. if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
  784. if (!dccp_li_hist_interval_new(ccid3_li_hist,
  785. &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
  786. return;
  787. head = list_entry(hcrx->ccid3hcrx_li_hist.next,
  788. struct dccp_li_hist_entry, dccplih_node);
  789. head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
  790. } else {
  791. struct dccp_li_hist_entry *entry;
  792. struct list_head *tail;
  793. head = list_entry(hcrx->ccid3hcrx_li_hist.next,
  794. struct dccp_li_hist_entry, dccplih_node);
  795. /* FIXME win count check removed as was wrong */
  796. /* should make this check with receive history */
  797. /* and compare there as per section 10.2 of RFC4342 */
  798. /* new loss event detected */
  799. /* calculate last interval length */
  800. seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
  801. entry = dccp_li_hist_entry_new(ccid3_li_hist, GFP_ATOMIC);
  802. if (entry == NULL) {
  803. DCCP_BUG("out of memory - can not allocate entry");
  804. return;
  805. }
  806. list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
  807. tail = hcrx->ccid3hcrx_li_hist.prev;
  808. list_del(tail);
  809. kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
  810. /* Create the newest interval */
  811. entry->dccplih_seqno = seq_loss;
  812. entry->dccplih_interval = seq_temp;
  813. entry->dccplih_win_count = win_loss;
  814. }
  815. }
  816. static int ccid3_hc_rx_detect_loss(struct sock *sk,
  817. struct dccp_rx_hist_entry *packet)
  818. {
  819. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  820. struct dccp_rx_hist_entry *rx_hist =
  821. dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
  822. u64 seqno = packet->dccphrx_seqno;
  823. u64 tmp_seqno;
  824. int loss = 0;
  825. u8 ccval;
  826. tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
  827. if (!rx_hist ||
  828. follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
  829. hcrx->ccid3hcrx_seqno_nonloss = seqno;
  830. hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
  831. goto detect_out;
  832. }
  833. while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
  834. > TFRC_RECV_NUM_LATE_LOSS) {
  835. loss = 1;
  836. ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
  837. hcrx->ccid3hcrx_ccval_nonloss);
  838. tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
  839. dccp_inc_seqno(&tmp_seqno);
  840. hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
  841. dccp_inc_seqno(&tmp_seqno);
  842. while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
  843. tmp_seqno, &ccval)) {
  844. hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
  845. hcrx->ccid3hcrx_ccval_nonloss = ccval;
  846. dccp_inc_seqno(&tmp_seqno);
  847. }
  848. }
  849. /* FIXME - this code could be simplified with above while */
  850. /* but works at moment */
  851. if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
  852. hcrx->ccid3hcrx_seqno_nonloss = seqno;
  853. hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
  854. }
  855. detect_out:
  856. dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
  857. &hcrx->ccid3hcrx_li_hist, packet,
  858. hcrx->ccid3hcrx_seqno_nonloss);
  859. return loss;
  860. }
  861. static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
  862. {
  863. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  864. const struct dccp_options_received *opt_recv;
  865. struct dccp_rx_hist_entry *packet;
  866. struct timeval now;
  867. u32 p_prev, rtt_prev;
  868. suseconds_t r_sample, t_elapsed;
  869. int loss, payload_size;
  870. BUG_ON(hcrx == NULL);
  871. opt_recv = &dccp_sk(sk)->dccps_options_received;
  872. switch (DCCP_SKB_CB(skb)->dccpd_type) {
  873. case DCCP_PKT_ACK:
  874. if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
  875. return;
  876. case DCCP_PKT_DATAACK:
  877. if (opt_recv->dccpor_timestamp_echo == 0)
  878. break;
  879. rtt_prev = hcrx->ccid3hcrx_rtt;
  880. dccp_timestamp(sk, &now);
  881. timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
  882. r_sample = timeval_usecs(&now);
  883. t_elapsed = opt_recv->dccpor_elapsed_time * 10;
  884. DCCP_BUG_ON(r_sample < 0);
  885. if (unlikely(r_sample <= t_elapsed))
  886. DCCP_WARN("r_sample=%ldus, t_elapsed=%ldus\n",
  887. (long)r_sample, (long)t_elapsed);
  888. else
  889. r_sample -= t_elapsed;
  890. CCID3_RTT_SANITY_CHECK(r_sample);
  891. if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
  892. hcrx->ccid3hcrx_rtt = r_sample;
  893. else
  894. hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
  895. r_sample / 10;
  896. if (rtt_prev != hcrx->ccid3hcrx_rtt)
  897. ccid3_pr_debug("%s(%p), New RTT=%uus, elapsed time=%u\n",
  898. dccp_role(sk), sk, hcrx->ccid3hcrx_rtt,
  899. opt_recv->dccpor_elapsed_time);
  900. break;
  901. case DCCP_PKT_DATA:
  902. break;
  903. default: /* We're not interested in other packet types, move along */
  904. return;
  905. }
  906. packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
  907. skb, GFP_ATOMIC);
  908. if (unlikely(packet == NULL)) {
  909. DCCP_WARN("%s(%p), Not enough mem to add rx packet "
  910. "to history, consider it lost!\n", dccp_role(sk), sk);
  911. return;
  912. }
  913. loss = ccid3_hc_rx_detect_loss(sk, packet);
  914. if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
  915. return;
  916. payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
  917. ccid3_hc_rx_update_s(hcrx, payload_size);
  918. switch (hcrx->ccid3hcrx_state) {
  919. case TFRC_RSTATE_NO_DATA:
  920. ccid3_pr_debug("%s(%p, state=%s), skb=%p, sending initial "
  921. "feedback\n", dccp_role(sk), sk,
  922. dccp_state_name(sk->sk_state), skb);
  923. ccid3_hc_rx_send_feedback(sk);
  924. ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
  925. return;
  926. case TFRC_RSTATE_DATA:
  927. hcrx->ccid3hcrx_bytes_recv += payload_size;
  928. if (loss)
  929. break;
  930. dccp_timestamp(sk, &now);
  931. if ((timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) -
  932. (suseconds_t)hcrx->ccid3hcrx_rtt) >= 0) {
  933. hcrx->ccid3hcrx_tstamp_last_ack = now;
  934. ccid3_hc_rx_send_feedback(sk);
  935. }
  936. return;
  937. case TFRC_RSTATE_TERM:
  938. DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
  939. return;
  940. }
  941. /* Dealing with packet loss */
  942. ccid3_pr_debug("%s(%p, state=%s), data loss! Reacting...\n",
  943. dccp_role(sk), sk, dccp_state_name(sk->sk_state));
  944. p_prev = hcrx->ccid3hcrx_p;
  945. /* Calculate loss event rate */
  946. if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
  947. u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
  948. /* Scaling up by 1000000 as fixed decimal */
  949. if (i_mean != 0)
  950. hcrx->ccid3hcrx_p = 1000000 / i_mean;
  951. } else
  952. DCCP_BUG("empty loss history");
  953. if (hcrx->ccid3hcrx_p > p_prev) {
  954. ccid3_hc_rx_send_feedback(sk);
  955. return;
  956. }
  957. }
  958. static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
  959. {
  960. struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
  961. ccid3_pr_debug("entry\n");
  962. hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
  963. INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
  964. INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
  965. dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
  966. hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
  967. hcrx->ccid3hcrx_s = 0;
  968. hcrx->ccid3hcrx_rtt = 0;
  969. return 0;
  970. }
  971. static void ccid3_hc_rx_exit(struct sock *sk)
  972. {
  973. struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  974. BUG_ON(hcrx == NULL);
  975. ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
  976. /* Empty packet history */
  977. dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
  978. /* Empty loss interval history */
  979. dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
  980. }
  981. static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
  982. {
  983. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  984. /* Listen socks doesn't have a private CCID block */
  985. if (sk->sk_state == DCCP_LISTEN)
  986. return;
  987. BUG_ON(hcrx == NULL);
  988. info->tcpi_ca_state = hcrx->ccid3hcrx_state;
  989. info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
  990. info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
  991. }
  992. static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
  993. u32 __user *optval, int __user *optlen)
  994. {
  995. const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
  996. const void *val;
  997. /* Listen socks doesn't have a private CCID block */
  998. if (sk->sk_state == DCCP_LISTEN)
  999. return -EINVAL;
  1000. switch (optname) {
  1001. case DCCP_SOCKOPT_CCID_RX_INFO:
  1002. if (len < sizeof(hcrx->ccid3hcrx_tfrc))
  1003. return -EINVAL;
  1004. len = sizeof(hcrx->ccid3hcrx_tfrc);
  1005. val = &hcrx->ccid3hcrx_tfrc;
  1006. break;
  1007. default:
  1008. return -ENOPROTOOPT;
  1009. }
  1010. if (put_user(len, optlen) || copy_to_user(optval, val, len))
  1011. return -EFAULT;
  1012. return 0;
  1013. }
  1014. static struct ccid_operations ccid3 = {
  1015. .ccid_id = DCCPC_CCID3,
  1016. .ccid_name = "ccid3",
  1017. .ccid_owner = THIS_MODULE,
  1018. .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
  1019. .ccid_hc_tx_init = ccid3_hc_tx_init,
  1020. .ccid_hc_tx_exit = ccid3_hc_tx_exit,
  1021. .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
  1022. .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
  1023. .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
  1024. .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
  1025. .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
  1026. .ccid_hc_rx_init = ccid3_hc_rx_init,
  1027. .ccid_hc_rx_exit = ccid3_hc_rx_exit,
  1028. .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
  1029. .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
  1030. .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
  1031. .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
  1032. .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
  1033. .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
  1034. };
  1035. #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
  1036. module_param(ccid3_debug, int, 0444);
  1037. MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
  1038. #endif
  1039. static __init int ccid3_module_init(void)
  1040. {
  1041. int rc = -ENOBUFS;
  1042. ccid3_rx_hist = dccp_rx_hist_new("ccid3");
  1043. if (ccid3_rx_hist == NULL)
  1044. goto out;
  1045. ccid3_tx_hist = dccp_tx_hist_new("ccid3");
  1046. if (ccid3_tx_hist == NULL)
  1047. goto out_free_rx;
  1048. ccid3_li_hist = dccp_li_hist_new("ccid3");
  1049. if (ccid3_li_hist == NULL)
  1050. goto out_free_tx;
  1051. rc = ccid_register(&ccid3);
  1052. if (rc != 0)
  1053. goto out_free_loss_interval_history;
  1054. out:
  1055. return rc;
  1056. out_free_loss_interval_history:
  1057. dccp_li_hist_delete(ccid3_li_hist);
  1058. ccid3_li_hist = NULL;
  1059. out_free_tx:
  1060. dccp_tx_hist_delete(ccid3_tx_hist);
  1061. ccid3_tx_hist = NULL;
  1062. out_free_rx:
  1063. dccp_rx_hist_delete(ccid3_rx_hist);
  1064. ccid3_rx_hist = NULL;
  1065. goto out;
  1066. }
  1067. module_init(ccid3_module_init);
  1068. static __exit void ccid3_module_exit(void)
  1069. {
  1070. ccid_unregister(&ccid3);
  1071. if (ccid3_tx_hist != NULL) {
  1072. dccp_tx_hist_delete(ccid3_tx_hist);
  1073. ccid3_tx_hist = NULL;
  1074. }
  1075. if (ccid3_rx_hist != NULL) {
  1076. dccp_rx_hist_delete(ccid3_rx_hist);
  1077. ccid3_rx_hist = NULL;
  1078. }
  1079. if (ccid3_li_hist != NULL) {
  1080. dccp_li_hist_delete(ccid3_li_hist);
  1081. ccid3_li_hist = NULL;
  1082. }
  1083. }
  1084. module_exit(ccid3_module_exit);
  1085. MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
  1086. "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
  1087. MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
  1088. MODULE_LICENSE("GPL");
  1089. MODULE_ALIAS("net-dccp-ccid-3");