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