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 38#include <asm/unaligned.h> 39 40#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 41static bool ccid3_debug; 42#define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a) 43#else 44#define ccid3_pr_debug(format, a...) 45#endif 46 47/* 48 * Transmitter Half-Connection Routines 49 */ 50#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 51static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state) 52{ 53 static const char *const ccid3_state_names[] = { 54 [TFRC_SSTATE_NO_SENT] = "NO_SENT", 55 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK", 56 [TFRC_SSTATE_FBACK] = "FBACK", 57 }; 58 59 return ccid3_state_names[state]; 60} 61#endif 62 63static void ccid3_hc_tx_set_state(struct sock *sk, 64 enum ccid3_hc_tx_states state) 65{ 66 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 67 enum ccid3_hc_tx_states oldstate = hc->tx_state; 68 69 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n", 70 dccp_role(sk), sk, ccid3_tx_state_name(oldstate), 71 ccid3_tx_state_name(state)); 72 WARN_ON(state == oldstate); 73 hc->tx_state = state; 74} 75 76/* 77 * Compute the initial sending rate X_init in the manner of RFC 3390: 78 * 79 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT 80 * 81 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis 82 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula. 83 * For consistency with other parts of the code, X_init is scaled by 2^6. 84 */ 85static inline u64 rfc3390_initial_rate(struct sock *sk) 86{ 87 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 88 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s); 89 90 return scaled_div(w_init << 6, hc->tx_rtt); 91} 92 93/** 94 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst 95 * This respects the granularity of X_inst (64 * bytes/second). 96 */ 97static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc) 98{ 99 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x); 100 101 DCCP_BUG_ON(hc->tx_t_ipi == 0); 102 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi, 103 hc->tx_s, (unsigned int)(hc->tx_x >> 6)); 104} 105 106static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now) 107{ 108 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count); 109 110 return delta / hc->tx_rtt; 111} 112 113/** 114 * ccid3_hc_tx_update_x - Update allowed sending rate X 115 * @stamp: most recent time if available - can be left NULL. 116 * 117 * This function tracks draft rfc3448bis, check there for latest details. 118 * 119 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support 120 * fine-grained resolution of sending rates. This requires scaling by 2^6 121 * throughout the code. Only X_calc is unscaled (in bytes/second). 122 * 123 */ 124static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp) 125{ 126 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 127 __u64 min_rate = 2 * hc->tx_x_recv; 128 const __u64 old_x = hc->tx_x; 129 ktime_t now = stamp ? *stamp : ktime_get_real(); 130 131 /* 132 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate 133 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis: 134 * a sender is idle if it has not sent anything over a 2-RTT-period. 135 * For consistency with X and X_recv, min_rate is also scaled by 2^6. 136 */ 137 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) { 138 min_rate = rfc3390_initial_rate(sk); 139 min_rate = max(min_rate, 2 * hc->tx_x_recv); 140 } 141 142 if (hc->tx_p > 0) { 143 144 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate); 145 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI); 146 147 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) { 148 149 hc->tx_x = min(2 * hc->tx_x, min_rate); 150 hc->tx_x = max(hc->tx_x, 151 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt)); 152 hc->tx_t_ld = now; 153 } 154 155 if (hc->tx_x != old_x) { 156 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, " 157 "X_recv=%u\n", (unsigned int)(old_x >> 6), 158 (unsigned int)(hc->tx_x >> 6), hc->tx_x_calc, 159 (unsigned int)(hc->tx_x_recv >> 6)); 160 161 ccid3_update_send_interval(hc); 162 } 163} 164 165/** 166 * ccid3_hc_tx_update_s - Track the mean packet size `s' 167 * @len: DCCP packet payload size in bytes 168 * 169 * cf. RFC 4342, 5.3 and RFC 3448, 4.1 170 */ 171static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len) 172{ 173 const u16 old_s = hc->tx_s; 174 175 hc->tx_s = tfrc_ewma(hc->tx_s, len, 9); 176 177 if (hc->tx_s != old_s) 178 ccid3_update_send_interval(hc); 179} 180 181/* 182 * Update Window Counter using the algorithm from [RFC 4342, 8.1]. 183 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt(). 184 */ 185static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc, 186 ktime_t now) 187{ 188 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count), 189 quarter_rtts = (4 * delta) / hc->tx_rtt; 190 191 if (quarter_rtts > 0) { 192 hc->tx_t_last_win_count = now; 193 hc->tx_last_win_count += min(quarter_rtts, 5U); 194 hc->tx_last_win_count &= 0xF; /* mod 16 */ 195 } 196} 197 198static void ccid3_hc_tx_no_feedback_timer(unsigned long data) 199{ 200 struct sock *sk = (struct sock *)data; 201 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 202 unsigned long t_nfb = USEC_PER_SEC / 5; 203 204 bh_lock_sock(sk); 205 if (sock_owned_by_user(sk)) { 206 /* Try again later. */ 207 /* XXX: set some sensible MIB */ 208 goto restart_timer; 209 } 210 211 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk, 212 ccid3_tx_state_name(hc->tx_state)); 213 214 /* Ignore and do not restart after leaving the established state */ 215 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN)) 216 goto out; 217 218 /* Reset feedback state to "no feedback received" */ 219 if (hc->tx_state == TFRC_SSTATE_FBACK) 220 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK); 221 222 /* 223 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4 224 * RTO is 0 if and only if no feedback has been received yet. 225 */ 226 if (hc->tx_t_rto == 0 || hc->tx_p == 0) { 227 228 /* halve send rate directly */ 229 hc->tx_x = max(hc->tx_x / 2, 230 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI); 231 ccid3_update_send_interval(hc); 232 } else { 233 /* 234 * Modify the cached value of X_recv 235 * 236 * If (X_calc > 2 * X_recv) 237 * X_recv = max(X_recv / 2, s / (2 * t_mbi)); 238 * Else 239 * X_recv = X_calc / 4; 240 * 241 * Note that X_recv is scaled by 2^6 while X_calc is not 242 */ 243 if (hc->tx_x_calc > (hc->tx_x_recv >> 5)) 244 hc->tx_x_recv = 245 max(hc->tx_x_recv / 2, 246 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI)); 247 else { 248 hc->tx_x_recv = hc->tx_x_calc; 249 hc->tx_x_recv <<= 4; 250 } 251 ccid3_hc_tx_update_x(sk, NULL); 252 } 253 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n", 254 (unsigned long long)hc->tx_x); 255 256 /* 257 * Set new timeout for the nofeedback timer. 258 * See comments in packet_recv() regarding the value of t_RTO. 259 */ 260 if (unlikely(hc->tx_t_rto == 0)) /* no feedback received yet */ 261 t_nfb = TFRC_INITIAL_TIMEOUT; 262 else 263 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi); 264 265restart_timer: 266 sk_reset_timer(sk, &hc->tx_no_feedback_timer, 267 jiffies + usecs_to_jiffies(t_nfb)); 268out: 269 bh_unlock_sock(sk); 270 sock_put(sk); 271} 272 273/** 274 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets 275 * @skb: next packet candidate to send on @sk 276 * 277 * This function uses the convention of ccid_packet_dequeue_eval() and 278 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec. 279 */ 280static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb) 281{ 282 struct dccp_sock *dp = dccp_sk(sk); 283 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 284 ktime_t now = ktime_get_real(); 285 s64 delay; 286 287 /* 288 * This function is called only for Data and DataAck packets. Sending 289 * zero-sized Data(Ack)s is theoretically possible, but for congestion 290 * control this case is pathological - ignore it. 291 */ 292 if (unlikely(skb->len == 0)) 293 return -EBADMSG; 294 295 if (hc->tx_state == TFRC_SSTATE_NO_SENT) { 296 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies + 297 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT))); 298 hc->tx_last_win_count = 0; 299 hc->tx_t_last_win_count = now; 300 301 /* Set t_0 for initial packet */ 302 hc->tx_t_nom = now; 303 304 hc->tx_s = skb->len; 305 306 /* 307 * Use initial RTT sample when available: recommended by erratum 308 * to RFC 4342. This implements the initialisation procedure of 309 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6. 310 */ 311 if (dp->dccps_syn_rtt) { 312 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt); 313 hc->tx_rtt = dp->dccps_syn_rtt; 314 hc->tx_x = rfc3390_initial_rate(sk); 315 hc->tx_t_ld = now; 316 } else { 317 /* 318 * Sender does not have RTT sample: 319 * - set fallback RTT (RFC 4340, 3.4) since a RTT value 320 * is needed in several parts (e.g. window counter); 321 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2. 322 */ 323 hc->tx_rtt = DCCP_FALLBACK_RTT; 324 hc->tx_x = hc->tx_s; 325 hc->tx_x <<= 6; 326 } 327 ccid3_update_send_interval(hc); 328 329 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK); 330 331 } else { 332 delay = ktime_us_delta(hc->tx_t_nom, now); 333 ccid3_pr_debug("delay=%ld\n", (long)delay); 334 /* 335 * Scheduling of packet transmissions (RFC 5348, 8.3) 336 * 337 * if (t_now > t_nom - delta) 338 * // send the packet now 339 * else 340 * // send the packet in (t_nom - t_now) milliseconds. 341 */ 342 if (delay >= TFRC_T_DELTA) 343 return (u32)delay / USEC_PER_MSEC; 344 345 ccid3_hc_tx_update_win_count(hc, now); 346 } 347 348 /* prepare to send now (add options etc.) */ 349 dp->dccps_hc_tx_insert_options = 1; 350 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count; 351 352 /* set the nominal send time for the next following packet */ 353 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi); 354 return CCID_PACKET_SEND_AT_ONCE; 355} 356 357static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len) 358{ 359 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 360 361 ccid3_hc_tx_update_s(hc, len); 362 363 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss)) 364 DCCP_CRIT("packet history - out of memory!"); 365} 366 367static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) 368{ 369 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 370 struct tfrc_tx_hist_entry *acked; 371 ktime_t now; 372 unsigned long t_nfb; 373 u32 r_sample; 374 375 /* we are only interested in ACKs */ 376 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK || 377 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK)) 378 return; 379 /* 380 * Locate the acknowledged packet in the TX history. 381 * 382 * Returning "entry not found" here can for instance happen when 383 * - the host has not sent out anything (e.g. a passive server), 384 * - the Ack is outdated (packet with higher Ack number was received), 385 * - it is a bogus Ack (for a packet not sent on this connection). 386 */ 387 acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb)); 388 if (acked == NULL) 389 return; 390 /* For the sake of RTT sampling, ignore/remove all older entries */ 391 tfrc_tx_hist_purge(&acked->next); 392 393 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */ 394 now = ktime_get_real(); 395 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp)); 396 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9); 397 398 /* 399 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3 400 */ 401 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) { 402 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK); 403 404 if (hc->tx_t_rto == 0) { 405 /* 406 * Initial feedback packet: Larger Initial Windows (4.2) 407 */ 408 hc->tx_x = rfc3390_initial_rate(sk); 409 hc->tx_t_ld = now; 410 411 ccid3_update_send_interval(hc); 412 413 goto done_computing_x; 414 } else if (hc->tx_p == 0) { 415 /* 416 * First feedback after nofeedback timer expiry (4.3) 417 */ 418 goto done_computing_x; 419 } 420 } 421 422 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */ 423 if (hc->tx_p > 0) 424 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p); 425 ccid3_hc_tx_update_x(sk, &now); 426 427done_computing_x: 428 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, " 429 "p=%u, X_calc=%u, X_recv=%u, X=%u\n", 430 dccp_role(sk), sk, hc->tx_rtt, r_sample, 431 hc->tx_s, hc->tx_p, hc->tx_x_calc, 432 (unsigned int)(hc->tx_x_recv >> 6), 433 (unsigned int)(hc->tx_x >> 6)); 434 435 /* unschedule no feedback timer */ 436 sk_stop_timer(sk, &hc->tx_no_feedback_timer); 437 438 /* 439 * As we have calculated new ipi, delta, t_nom it is possible 440 * that we now can send a packet, so wake up dccp_wait_for_ccid 441 */ 442 sk->sk_write_space(sk); 443 444 /* 445 * Update timeout interval for the nofeedback timer. In order to control 446 * rate halving on networks with very low RTTs (<= 1 ms), use per-route 447 * tunable RTAX_RTO_MIN value as the lower bound. 448 */ 449 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt, 450 USEC_PER_SEC/HZ * tcp_rto_min(sk)); 451 /* 452 * Schedule no feedback timer to expire in 453 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi) 454 */ 455 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi); 456 457 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to " 458 "expire in %lu jiffies (%luus)\n", 459 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb); 460 461 sk_reset_timer(sk, &hc->tx_no_feedback_timer, 462 jiffies + usecs_to_jiffies(t_nfb)); 463} 464 465static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type, 466 u8 option, u8 *optval, u8 optlen) 467{ 468 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 469 __be32 opt_val; 470 471 switch (option) { 472 case TFRC_OPT_RECEIVE_RATE: 473 case TFRC_OPT_LOSS_EVENT_RATE: 474 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */ 475 if (packet_type == DCCP_PKT_DATA) 476 break; 477 if (unlikely(optlen != 4)) { 478 DCCP_WARN("%s(%p), invalid len %d for %u\n", 479 dccp_role(sk), sk, optlen, option); 480 return -EINVAL; 481 } 482 opt_val = ntohl(get_unaligned((__be32 *)optval)); 483 484 if (option == TFRC_OPT_RECEIVE_RATE) { 485 /* Receive Rate is kept in units of 64 bytes/second */ 486 hc->tx_x_recv = opt_val; 487 hc->tx_x_recv <<= 6; 488 489 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n", 490 dccp_role(sk), sk, opt_val); 491 } else { 492 /* Update the fixpoint Loss Event Rate fraction */ 493 hc->tx_p = tfrc_invert_loss_event_rate(opt_val); 494 495 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n", 496 dccp_role(sk), sk, opt_val); 497 } 498 } 499 return 0; 500} 501 502static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk) 503{ 504 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid); 505 506 hc->tx_state = TFRC_SSTATE_NO_SENT; 507 hc->tx_hist = NULL; 508 setup_timer(&hc->tx_no_feedback_timer, 509 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk); 510 return 0; 511} 512 513static void ccid3_hc_tx_exit(struct sock *sk) 514{ 515 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 516 517 sk_stop_timer(sk, &hc->tx_no_feedback_timer); 518 tfrc_tx_hist_purge(&hc->tx_hist); 519} 520 521static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info) 522{ 523 info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto; 524 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt; 525} 526 527static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len, 528 u32 __user *optval, int __user *optlen) 529{ 530 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 531 struct tfrc_tx_info tfrc; 532 const void *val; 533 534 switch (optname) { 535 case DCCP_SOCKOPT_CCID_TX_INFO: 536 if (len < sizeof(tfrc)) 537 return -EINVAL; 538 memset(&tfrc, 0, sizeof(tfrc)); 539 tfrc.tfrctx_x = hc->tx_x; 540 tfrc.tfrctx_x_recv = hc->tx_x_recv; 541 tfrc.tfrctx_x_calc = hc->tx_x_calc; 542 tfrc.tfrctx_rtt = hc->tx_rtt; 543 tfrc.tfrctx_p = hc->tx_p; 544 tfrc.tfrctx_rto = hc->tx_t_rto; 545 tfrc.tfrctx_ipi = hc->tx_t_ipi; 546 len = sizeof(tfrc); 547 val = &tfrc; 548 break; 549 default: 550 return -ENOPROTOOPT; 551 } 552 553 if (put_user(len, optlen) || copy_to_user(optval, val, len)) 554 return -EFAULT; 555 556 return 0; 557} 558 559/* 560 * Receiver Half-Connection Routines 561 */ 562 563/* CCID3 feedback types */ 564enum ccid3_fback_type { 565 CCID3_FBACK_NONE = 0, 566 CCID3_FBACK_INITIAL, 567 CCID3_FBACK_PERIODIC, 568 CCID3_FBACK_PARAM_CHANGE 569}; 570 571#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 572static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state) 573{ 574 static const char *const ccid3_rx_state_names[] = { 575 [TFRC_RSTATE_NO_DATA] = "NO_DATA", 576 [TFRC_RSTATE_DATA] = "DATA", 577 }; 578 579 return ccid3_rx_state_names[state]; 580} 581#endif 582 583static void ccid3_hc_rx_set_state(struct sock *sk, 584 enum ccid3_hc_rx_states state) 585{ 586 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 587 enum ccid3_hc_rx_states oldstate = hc->rx_state; 588 589 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n", 590 dccp_role(sk), sk, ccid3_rx_state_name(oldstate), 591 ccid3_rx_state_name(state)); 592 WARN_ON(state == oldstate); 593 hc->rx_state = state; 594} 595 596static void ccid3_hc_rx_send_feedback(struct sock *sk, 597 const struct sk_buff *skb, 598 enum ccid3_fback_type fbtype) 599{ 600 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 601 struct dccp_sock *dp = dccp_sk(sk); 602 ktime_t now = ktime_get_real(); 603 s64 delta = 0; 604 605 switch (fbtype) { 606 case CCID3_FBACK_INITIAL: 607 hc->rx_x_recv = 0; 608 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */ 609 break; 610 case CCID3_FBACK_PARAM_CHANGE: 611 /* 612 * When parameters change (new loss or p > p_prev), we do not 613 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so 614 * need to reuse the previous value of X_recv. However, when 615 * X_recv was 0 (due to early loss), this would kill X down to 616 * s/t_mbi (i.e. one packet in 64 seconds). 617 * To avoid such drastic reduction, we approximate X_recv as 618 * the number of bytes since last feedback. 619 * This is a safe fallback, since X is bounded above by X_calc. 620 */ 621 if (hc->rx_x_recv > 0) 622 break; 623 /* fall through */ 624 case CCID3_FBACK_PERIODIC: 625 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback); 626 if (delta <= 0) 627 DCCP_BUG("delta (%ld) <= 0", (long)delta); 628 else 629 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta); 630 break; 631 default: 632 return; 633 } 634 635 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta, 636 hc->rx_x_recv, hc->rx_pinv); 637 638 hc->rx_tstamp_last_feedback = now; 639 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval; 640 hc->rx_bytes_recv = 0; 641 642 dp->dccps_hc_rx_insert_options = 1; 643 dccp_send_ack(sk); 644} 645 646static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb) 647{ 648 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 649 __be32 x_recv, pinv; 650 651 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)) 652 return 0; 653 654 if (dccp_packet_without_ack(skb)) 655 return 0; 656 657 x_recv = htonl(hc->rx_x_recv); 658 pinv = htonl(hc->rx_pinv); 659 660 if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE, 661 &pinv, sizeof(pinv)) || 662 dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE, 663 &x_recv, sizeof(x_recv))) 664 return -1; 665 666 return 0; 667} 668 669/** 670 * ccid3_first_li - Implements [RFC 5348, 6.3.1] 671 * 672 * Determine the length of the first loss interval via inverse lookup. 673 * Assume that X_recv can be computed by the throughput equation 674 * s 675 * X_recv = -------- 676 * R * fval 677 * Find some p such that f(p) = fval; return 1/p (scaled). 678 */ 679static u32 ccid3_first_li(struct sock *sk) 680{ 681 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 682 u32 x_recv, p, delta; 683 u64 fval; 684 685 if (hc->rx_rtt == 0) { 686 DCCP_WARN("No RTT estimate available, using fallback RTT\n"); 687 hc->rx_rtt = DCCP_FALLBACK_RTT; 688 } 689 690 delta = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback)); 691 x_recv = scaled_div32(hc->rx_bytes_recv, delta); 692 if (x_recv == 0) { /* would also trigger divide-by-zero */ 693 DCCP_WARN("X_recv==0\n"); 694 if (hc->rx_x_recv == 0) { 695 DCCP_BUG("stored value of X_recv is zero"); 696 return ~0U; 697 } 698 x_recv = hc->rx_x_recv; 699 } 700 701 fval = scaled_div(hc->rx_s, hc->rx_rtt); 702 fval = scaled_div32(fval, x_recv); 703 p = tfrc_calc_x_reverse_lookup(fval); 704 705 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied " 706 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p); 707 708 return p == 0 ? ~0U : scaled_div(1, p); 709} 710 711static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) 712{ 713 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 714 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE; 715 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp; 716 const bool is_data_packet = dccp_data_packet(skb); 717 718 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) { 719 if (is_data_packet) { 720 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4; 721 do_feedback = CCID3_FBACK_INITIAL; 722 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA); 723 hc->rx_s = payload; 724 /* 725 * Not necessary to update rx_bytes_recv here, 726 * since X_recv = 0 for the first feedback packet (cf. 727 * RFC 3448, 6.3) -- gerrit 728 */ 729 } 730 goto update_records; 731 } 732 733 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb)) 734 return; /* done receiving */ 735 736 if (is_data_packet) { 737 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4; 738 /* 739 * Update moving-average of s and the sum of received payload bytes 740 */ 741 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9); 742 hc->rx_bytes_recv += payload; 743 } 744 745 /* 746 * Perform loss detection and handle pending losses 747 */ 748 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist, 749 skb, ndp, ccid3_first_li, sk)) { 750 do_feedback = CCID3_FBACK_PARAM_CHANGE; 751 goto done_receiving; 752 } 753 754 if (tfrc_rx_hist_loss_pending(&hc->rx_hist)) 755 return; /* done receiving */ 756 757 /* 758 * Handle data packets: RTT sampling and monitoring p 759 */ 760 if (unlikely(!is_data_packet)) 761 goto update_records; 762 763 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) { 764 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb); 765 /* 766 * Empty loss history: no loss so far, hence p stays 0. 767 * Sample RTT values, since an RTT estimate is required for the 768 * computation of p when the first loss occurs; RFC 3448, 6.3.1. 769 */ 770 if (sample != 0) 771 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9); 772 773 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) { 774 /* 775 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean 776 * has decreased (resp. p has increased), send feedback now. 777 */ 778 do_feedback = CCID3_FBACK_PARAM_CHANGE; 779 } 780 781 /* 782 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3 783 */ 784 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3) 785 do_feedback = CCID3_FBACK_PERIODIC; 786 787update_records: 788 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp); 789 790done_receiving: 791 if (do_feedback) 792 ccid3_hc_rx_send_feedback(sk, skb, do_feedback); 793} 794 795static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk) 796{ 797 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid); 798 799 hc->rx_state = TFRC_RSTATE_NO_DATA; 800 tfrc_lh_init(&hc->rx_li_hist); 801 return tfrc_rx_hist_alloc(&hc->rx_hist); 802} 803 804static void ccid3_hc_rx_exit(struct sock *sk) 805{ 806 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 807 808 tfrc_rx_hist_purge(&hc->rx_hist); 809 tfrc_lh_cleanup(&hc->rx_li_hist); 810} 811 812static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info) 813{ 814 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state; 815 info->tcpi_options |= TCPI_OPT_TIMESTAMPS; 816 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rx_rtt; 817} 818 819static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len, 820 u32 __user *optval, int __user *optlen) 821{ 822 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 823 struct tfrc_rx_info rx_info; 824 const void *val; 825 826 switch (optname) { 827 case DCCP_SOCKOPT_CCID_RX_INFO: 828 if (len < sizeof(rx_info)) 829 return -EINVAL; 830 rx_info.tfrcrx_x_recv = hc->rx_x_recv; 831 rx_info.tfrcrx_rtt = hc->rx_rtt; 832 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hc->rx_pinv); 833 len = sizeof(rx_info); 834 val = &rx_info; 835 break; 836 default: 837 return -ENOPROTOOPT; 838 } 839 840 if (put_user(len, optlen) || copy_to_user(optval, val, len)) 841 return -EFAULT; 842 843 return 0; 844} 845 846struct ccid_operations ccid3_ops = { 847 .ccid_id = DCCPC_CCID3, 848 .ccid_name = "TCP-Friendly Rate Control", 849 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock), 850 .ccid_hc_tx_init = ccid3_hc_tx_init, 851 .ccid_hc_tx_exit = ccid3_hc_tx_exit, 852 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet, 853 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent, 854 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv, 855 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options, 856 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock), 857 .ccid_hc_rx_init = ccid3_hc_rx_init, 858 .ccid_hc_rx_exit = ccid3_hc_rx_exit, 859 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options, 860 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv, 861 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info, 862 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info, 863 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt, 864 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt, 865}; 866 867#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 868module_param(ccid3_debug, bool, 0644); 869MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages"); 870#endif 871