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 int 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 [TFRC_SSTATE_TERM] = "TERM", 58 }; 59 60 return ccid3_state_names[state]; 61} 62#endif 63 64static void ccid3_hc_tx_set_state(struct sock *sk, 65 enum ccid3_hc_tx_states state) 66{ 67 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 68 enum ccid3_hc_tx_states oldstate = hc->tx_state; 69 70 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n", 71 dccp_role(sk), sk, ccid3_tx_state_name(oldstate), 72 ccid3_tx_state_name(state)); 73 WARN_ON(state == oldstate); 74 hc->tx_state = state; 75} 76 77/* 78 * Compute the initial sending rate X_init in the manner of RFC 3390: 79 * 80 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT 81 * 82 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis 83 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula. 84 * For consistency with other parts of the code, X_init is scaled by 2^6. 85 */ 86static inline u64 rfc3390_initial_rate(struct sock *sk) 87{ 88 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 89 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s); 90 91 return scaled_div(w_init << 6, hc->tx_rtt); 92} 93 94/* 95 * Recalculate t_ipi and delta (should be called whenever X changes) 96 */ 97static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc) 98{ 99 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */ 100 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x); 101 102 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */ 103 hc->tx_delta = min_t(u32, hc->tx_t_ipi / 2, TFRC_OPSYS_HALF_TIME_GRAN); 104 105 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n", hc->tx_t_ipi, 106 hc->tx_delta, hc->tx_s, (unsigned)(hc->tx_x >> 6)); 107} 108 109static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now) 110{ 111 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count); 112 113 return delta / hc->tx_rtt; 114} 115 116/** 117 * ccid3_hc_tx_update_x - Update allowed sending rate X 118 * @stamp: most recent time if available - can be left NULL. 119 * This function tracks draft rfc3448bis, check there for latest details. 120 * 121 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support 122 * fine-grained resolution of sending rates. This requires scaling by 2^6 123 * throughout the code. Only X_calc is unscaled (in bytes/second). 124 * 125 */ 126static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp) 127{ 128 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 129 __u64 min_rate = 2 * hc->tx_x_recv; 130 const __u64 old_x = hc->tx_x; 131 ktime_t now = stamp ? *stamp : ktime_get_real(); 132 133 /* 134 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate 135 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis: 136 * a sender is idle if it has not sent anything over a 2-RTT-period. 137 * For consistency with X and X_recv, min_rate is also scaled by 2^6. 138 */ 139 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) { 140 min_rate = rfc3390_initial_rate(sk); 141 min_rate = max(min_rate, 2 * hc->tx_x_recv); 142 } 143 144 if (hc->tx_p > 0) { 145 146 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate); 147 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI); 148 149 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) { 150 151 hc->tx_x = min(2 * hc->tx_x, min_rate); 152 hc->tx_x = max(hc->tx_x, 153 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt)); 154 hc->tx_t_ld = now; 155 } 156 157 if (hc->tx_x != old_x) { 158 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, " 159 "X_recv=%u\n", (unsigned)(old_x >> 6), 160 (unsigned)(hc->tx_x >> 6), hc->tx_x_calc, 161 (unsigned)(hc->tx_x_recv >> 6)); 162 163 ccid3_update_send_interval(hc); 164 } 165} 166 167/* 168 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1) 169 * @len: DCCP packet payload size in bytes 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 goto restart_timer; 208 } 209 210 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk, 211 ccid3_tx_state_name(hc->tx_state)); 212 213 if (hc->tx_state == TFRC_SSTATE_FBACK) 214 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK); 215 else if (hc->tx_state != TFRC_SSTATE_NO_FBACK) 216 goto out; 217 218 /* 219 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4 220 */ 221 if (hc->tx_t_rto == 0 || /* no feedback received yet */ 222 hc->tx_p == 0) { 223 224 /* halve send rate directly */ 225 hc->tx_x = max(hc->tx_x / 2, 226 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI); 227 ccid3_update_send_interval(hc); 228 } else { 229 /* 230 * Modify the cached value of X_recv 231 * 232 * If (X_calc > 2 * X_recv) 233 * X_recv = max(X_recv / 2, s / (2 * t_mbi)); 234 * Else 235 * X_recv = X_calc / 4; 236 * 237 * Note that X_recv is scaled by 2^6 while X_calc is not 238 */ 239 BUG_ON(hc->tx_p && !hc->tx_x_calc); 240 241 if (hc->tx_x_calc > (hc->tx_x_recv >> 5)) 242 hc->tx_x_recv = 243 max(hc->tx_x_recv / 2, 244 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI)); 245 else { 246 hc->tx_x_recv = hc->tx_x_calc; 247 hc->tx_x_recv <<= 4; 248 } 249 ccid3_hc_tx_update_x(sk, NULL); 250 } 251 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n", 252 (unsigned long long)hc->tx_x); 253 254 /* 255 * Set new timeout for the nofeedback timer. 256 * See comments in packet_recv() regarding the value of t_RTO. 257 */ 258 if (unlikely(hc->tx_t_rto == 0)) /* no feedback yet */ 259 t_nfb = TFRC_INITIAL_TIMEOUT; 260 else 261 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi); 262 263restart_timer: 264 sk_reset_timer(sk, &hc->tx_no_feedback_timer, 265 jiffies + usecs_to_jiffies(t_nfb)); 266out: 267 bh_unlock_sock(sk); 268 sock_put(sk); 269} 270 271/* 272 * returns 273 * > 0: delay (in msecs) that should pass before actually sending 274 * = 0: can send immediately 275 * < 0: error condition; do not send packet 276 */ 277static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb) 278{ 279 struct dccp_sock *dp = dccp_sk(sk); 280 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 281 ktime_t now = ktime_get_real(); 282 s64 delay; 283 284 /* 285 * This function is called only for Data and DataAck packets. Sending 286 * zero-sized Data(Ack)s is theoretically possible, but for congestion 287 * control this case is pathological - ignore it. 288 */ 289 if (unlikely(skb->len == 0)) 290 return -EBADMSG; 291 292 switch (hc->tx_state) { 293 case TFRC_SSTATE_NO_SENT: 294 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies + 295 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT))); 296 hc->tx_last_win_count = 0; 297 hc->tx_t_last_win_count = now; 298 299 /* Set t_0 for initial packet */ 300 hc->tx_t_nom = now; 301 302 hc->tx_s = skb->len; 303 304 /* 305 * Use initial RTT sample when available: recommended by erratum 306 * to RFC 4342. This implements the initialisation procedure of 307 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6. 308 */ 309 if (dp->dccps_syn_rtt) { 310 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt); 311 hc->tx_rtt = dp->dccps_syn_rtt; 312 hc->tx_x = rfc3390_initial_rate(sk); 313 hc->tx_t_ld = now; 314 } else { 315 /* 316 * Sender does not have RTT sample: 317 * - set fallback RTT (RFC 4340, 3.4) since a RTT value 318 * is needed in several parts (e.g. window counter); 319 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2. 320 */ 321 hc->tx_rtt = DCCP_FALLBACK_RTT; 322 hc->tx_x = hc->tx_s; 323 hc->tx_x <<= 6; 324 } 325 ccid3_update_send_interval(hc); 326 327 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK); 328 break; 329 case TFRC_SSTATE_NO_FBACK: 330 case TFRC_SSTATE_FBACK: 331 delay = ktime_us_delta(hc->tx_t_nom, now); 332 ccid3_pr_debug("delay=%ld\n", (long)delay); 333 /* 334 * Scheduling of packet transmissions [RFC 3448, 4.6] 335 * 336 * if (t_now > t_nom - delta) 337 * // send the packet now 338 * else 339 * // send the packet in (t_nom - t_now) milliseconds. 340 */ 341 if (delay - (s64)hc->tx_delta >= 1000) 342 return (u32)delay / 1000L; 343 344 ccid3_hc_tx_update_win_count(hc, now); 345 break; 346 case TFRC_SSTATE_TERM: 347 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk); 348 return -EINVAL; 349 } 350 351 /* prepare to send now (add options etc.) */ 352 dp->dccps_hc_tx_insert_options = 1; 353 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count; 354 355 /* set the nominal send time for the next following packet */ 356 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi); 357 return 0; 358} 359 360static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, 361 unsigned int len) 362{ 363 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 364 365 ccid3_hc_tx_update_s(hc, len); 366 367 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss)) 368 DCCP_CRIT("packet history - out of memory!"); 369} 370 371static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) 372{ 373 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 374 struct ccid3_options_received *opt_recv; 375 ktime_t now; 376 unsigned long t_nfb; 377 u32 pinv, r_sample; 378 379 /* we are only interested in ACKs */ 380 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK || 381 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK)) 382 return; 383 /* ... and only in the established state */ 384 if (hc->tx_state != TFRC_SSTATE_FBACK && 385 hc->tx_state != TFRC_SSTATE_NO_FBACK) 386 return; 387 388 opt_recv = &hc->tx_options_received; 389 now = ktime_get_real(); 390 391 /* Estimate RTT from history if ACK number is valid */ 392 r_sample = tfrc_tx_hist_rtt(hc->tx_hist, 393 DCCP_SKB_CB(skb)->dccpd_ack_seq, now); 394 if (r_sample == 0) { 395 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk, 396 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type), 397 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq); 398 return; 399 } 400 401 /* Update receive rate in units of 64 * bytes/second */ 402 hc->tx_x_recv = opt_recv->ccid3or_receive_rate; 403 hc->tx_x_recv <<= 6; 404 405 /* Update loss event rate (which is scaled by 1e6) */ 406 pinv = opt_recv->ccid3or_loss_event_rate; 407 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */ 408 hc->tx_p = 0; 409 else /* can not exceed 100% */ 410 hc->tx_p = scaled_div(1, pinv); 411 /* 412 * Validate new RTT sample and update moving average 413 */ 414 r_sample = dccp_sample_rtt(sk, r_sample); 415 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9); 416 /* 417 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3 418 */ 419 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) { 420 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK); 421 422 if (hc->tx_t_rto == 0) { 423 /* 424 * Initial feedback packet: Larger Initial Windows (4.2) 425 */ 426 hc->tx_x = rfc3390_initial_rate(sk); 427 hc->tx_t_ld = now; 428 429 ccid3_update_send_interval(hc); 430 431 goto done_computing_x; 432 } else if (hc->tx_p == 0) { 433 /* 434 * First feedback after nofeedback timer expiry (4.3) 435 */ 436 goto done_computing_x; 437 } 438 } 439 440 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */ 441 if (hc->tx_p > 0) 442 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p); 443 ccid3_hc_tx_update_x(sk, &now); 444 445done_computing_x: 446 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, " 447 "p=%u, X_calc=%u, X_recv=%u, X=%u\n", 448 dccp_role(sk), sk, hc->tx_rtt, r_sample, 449 hc->tx_s, hc->tx_p, hc->tx_x_calc, 450 (unsigned)(hc->tx_x_recv >> 6), 451 (unsigned)(hc->tx_x >> 6)); 452 453 /* unschedule no feedback timer */ 454 sk_stop_timer(sk, &hc->tx_no_feedback_timer); 455 456 /* 457 * As we have calculated new ipi, delta, t_nom it is possible 458 * that we now can send a packet, so wake up dccp_wait_for_ccid 459 */ 460 sk->sk_write_space(sk); 461 462 /* 463 * Update timeout interval for the nofeedback timer. 464 * We use a configuration option to increase the lower bound. 465 * This can help avoid triggering the nofeedback timer too 466 * often ('spinning') on LANs with small RTTs. 467 */ 468 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt, (CONFIG_IP_DCCP_CCID3_RTO * 469 (USEC_PER_SEC / 1000))); 470 /* 471 * Schedule no feedback timer to expire in 472 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi) 473 */ 474 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi); 475 476 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to " 477 "expire in %lu jiffies (%luus)\n", 478 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb); 479 480 sk_reset_timer(sk, &hc->tx_no_feedback_timer, 481 jiffies + usecs_to_jiffies(t_nfb)); 482} 483 484static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option, 485 unsigned char len, u16 idx, 486 unsigned char *value) 487{ 488 int rc = 0; 489 const struct dccp_sock *dp = dccp_sk(sk); 490 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 491 struct ccid3_options_received *opt_recv; 492 __be32 opt_val; 493 494 opt_recv = &hc->tx_options_received; 495 496 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) { 497 opt_recv->ccid3or_seqno = dp->dccps_gsr; 498 opt_recv->ccid3or_loss_event_rate = ~0; 499 opt_recv->ccid3or_loss_intervals_idx = 0; 500 opt_recv->ccid3or_loss_intervals_len = 0; 501 opt_recv->ccid3or_receive_rate = 0; 502 } 503 504 switch (option) { 505 case TFRC_OPT_LOSS_EVENT_RATE: 506 if (unlikely(len != 4)) { 507 DCCP_WARN("%s(%p), invalid len %d " 508 "for TFRC_OPT_LOSS_EVENT_RATE\n", 509 dccp_role(sk), sk, len); 510 rc = -EINVAL; 511 } else { 512 opt_val = get_unaligned((__be32 *)value); 513 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val); 514 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n", 515 dccp_role(sk), sk, 516 opt_recv->ccid3or_loss_event_rate); 517 } 518 break; 519 case TFRC_OPT_LOSS_INTERVALS: 520 opt_recv->ccid3or_loss_intervals_idx = idx; 521 opt_recv->ccid3or_loss_intervals_len = len; 522 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n", 523 dccp_role(sk), sk, 524 opt_recv->ccid3or_loss_intervals_idx, 525 opt_recv->ccid3or_loss_intervals_len); 526 break; 527 case TFRC_OPT_RECEIVE_RATE: 528 if (unlikely(len != 4)) { 529 DCCP_WARN("%s(%p), invalid len %d " 530 "for TFRC_OPT_RECEIVE_RATE\n", 531 dccp_role(sk), sk, len); 532 rc = -EINVAL; 533 } else { 534 opt_val = get_unaligned((__be32 *)value); 535 opt_recv->ccid3or_receive_rate = ntohl(opt_val); 536 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n", 537 dccp_role(sk), sk, 538 opt_recv->ccid3or_receive_rate); 539 } 540 break; 541 } 542 543 return rc; 544} 545 546static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk) 547{ 548 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid); 549 550 hc->tx_state = TFRC_SSTATE_NO_SENT; 551 hc->tx_hist = NULL; 552 setup_timer(&hc->tx_no_feedback_timer, 553 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk); 554 return 0; 555} 556 557static void ccid3_hc_tx_exit(struct sock *sk) 558{ 559 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk); 560 561 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM); 562 sk_stop_timer(sk, &hc->tx_no_feedback_timer); 563 564 tfrc_tx_hist_purge(&hc->tx_hist); 565} 566 567static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info) 568{ 569 struct ccid3_hc_tx_sock *hc; 570 571 /* Listen socks doesn't have a private CCID block */ 572 if (sk->sk_state == DCCP_LISTEN) 573 return; 574 575 hc = ccid3_hc_tx_sk(sk); 576 info->tcpi_rto = hc->tx_t_rto; 577 info->tcpi_rtt = hc->tx_rtt; 578} 579 580static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len, 581 u32 __user *optval, int __user *optlen) 582{ 583 const struct ccid3_hc_tx_sock *hc; 584 const void *val; 585 586 /* Listen socks doesn't have a private CCID block */ 587 if (sk->sk_state == DCCP_LISTEN) 588 return -EINVAL; 589 590 hc = ccid3_hc_tx_sk(sk); 591 switch (optname) { 592 case DCCP_SOCKOPT_CCID_TX_INFO: 593 if (len < sizeof(hc->tx_tfrc)) 594 return -EINVAL; 595 len = sizeof(hc->tx_tfrc); 596 val = &hc->tx_tfrc; 597 break; 598 default: 599 return -ENOPROTOOPT; 600 } 601 602 if (put_user(len, optlen) || copy_to_user(optval, val, len)) 603 return -EFAULT; 604 605 return 0; 606} 607 608/* 609 * Receiver Half-Connection Routines 610 */ 611 612/* CCID3 feedback types */ 613enum ccid3_fback_type { 614 CCID3_FBACK_NONE = 0, 615 CCID3_FBACK_INITIAL, 616 CCID3_FBACK_PERIODIC, 617 CCID3_FBACK_PARAM_CHANGE 618}; 619 620#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 621static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state) 622{ 623 static const char *const ccid3_rx_state_names[] = { 624 [TFRC_RSTATE_NO_DATA] = "NO_DATA", 625 [TFRC_RSTATE_DATA] = "DATA", 626 [TFRC_RSTATE_TERM] = "TERM", 627 }; 628 629 return ccid3_rx_state_names[state]; 630} 631#endif 632 633static void ccid3_hc_rx_set_state(struct sock *sk, 634 enum ccid3_hc_rx_states state) 635{ 636 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 637 enum ccid3_hc_rx_states oldstate = hc->rx_state; 638 639 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n", 640 dccp_role(sk), sk, ccid3_rx_state_name(oldstate), 641 ccid3_rx_state_name(state)); 642 WARN_ON(state == oldstate); 643 hc->rx_state = state; 644} 645 646static void ccid3_hc_rx_send_feedback(struct sock *sk, 647 const struct sk_buff *skb, 648 enum ccid3_fback_type fbtype) 649{ 650 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 651 struct dccp_sock *dp = dccp_sk(sk); 652 ktime_t now; 653 s64 delta = 0; 654 655 if (unlikely(hc->rx_state == TFRC_RSTATE_TERM)) 656 return; 657 658 now = ktime_get_real(); 659 660 switch (fbtype) { 661 case CCID3_FBACK_INITIAL: 662 hc->rx_x_recv = 0; 663 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */ 664 break; 665 case CCID3_FBACK_PARAM_CHANGE: 666 /* 667 * When parameters change (new loss or p > p_prev), we do not 668 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so 669 * need to reuse the previous value of X_recv. However, when 670 * X_recv was 0 (due to early loss), this would kill X down to 671 * s/t_mbi (i.e. one packet in 64 seconds). 672 * To avoid such drastic reduction, we approximate X_recv as 673 * the number of bytes since last feedback. 674 * This is a safe fallback, since X is bounded above by X_calc. 675 */ 676 if (hc->rx_x_recv > 0) 677 break; 678 /* fall through */ 679 case CCID3_FBACK_PERIODIC: 680 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback); 681 if (delta <= 0) 682 DCCP_BUG("delta (%ld) <= 0", (long)delta); 683 else 684 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta); 685 break; 686 default: 687 return; 688 } 689 690 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta, 691 hc->rx_x_recv, hc->rx_pinv); 692 693 hc->rx_tstamp_last_feedback = now; 694 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval; 695 hc->rx_bytes_recv = 0; 696 697 dp->dccps_hc_rx_insert_options = 1; 698 dccp_send_ack(sk); 699} 700 701static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb) 702{ 703 const struct ccid3_hc_rx_sock *hc; 704 __be32 x_recv, pinv; 705 706 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)) 707 return 0; 708 709 hc = ccid3_hc_rx_sk(sk); 710 711 if (dccp_packet_without_ack(skb)) 712 return 0; 713 714 x_recv = htonl(hc->rx_x_recv); 715 pinv = htonl(hc->rx_pinv); 716 717 if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE, 718 &pinv, sizeof(pinv)) || 719 dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE, 720 &x_recv, sizeof(x_recv))) 721 return -1; 722 723 return 0; 724} 725 726/** 727 * ccid3_first_li - Implements [RFC 5348, 6.3.1] 728 * 729 * Determine the length of the first loss interval via inverse lookup. 730 * Assume that X_recv can be computed by the throughput equation 731 * s 732 * X_recv = -------- 733 * R * fval 734 * Find some p such that f(p) = fval; return 1/p (scaled). 735 */ 736static u32 ccid3_first_li(struct sock *sk) 737{ 738 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 739 u32 x_recv, p, delta; 740 u64 fval; 741 742 if (hc->rx_rtt == 0) { 743 DCCP_WARN("No RTT estimate available, using fallback RTT\n"); 744 hc->rx_rtt = DCCP_FALLBACK_RTT; 745 } 746 747 delta = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback)); 748 x_recv = scaled_div32(hc->rx_bytes_recv, delta); 749 if (x_recv == 0) { /* would also trigger divide-by-zero */ 750 DCCP_WARN("X_recv==0\n"); 751 if ((x_recv = hc->rx_x_recv) == 0) { 752 DCCP_BUG("stored value of X_recv is zero"); 753 return ~0U; 754 } 755 } 756 757 fval = scaled_div(hc->rx_s, hc->rx_rtt); 758 fval = scaled_div32(fval, x_recv); 759 p = tfrc_calc_x_reverse_lookup(fval); 760 761 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied " 762 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p); 763 764 return p == 0 ? ~0U : scaled_div(1, p); 765} 766 767static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) 768{ 769 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 770 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE; 771 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp; 772 const bool is_data_packet = dccp_data_packet(skb); 773 774 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) { 775 if (is_data_packet) { 776 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4; 777 do_feedback = CCID3_FBACK_INITIAL; 778 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA); 779 hc->rx_s = payload; 780 /* 781 * Not necessary to update rx_bytes_recv here, 782 * since X_recv = 0 for the first feedback packet (cf. 783 * RFC 3448, 6.3) -- gerrit 784 */ 785 } 786 goto update_records; 787 } 788 789 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb)) 790 return; /* done receiving */ 791 792 if (is_data_packet) { 793 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4; 794 /* 795 * Update moving-average of s and the sum of received payload bytes 796 */ 797 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9); 798 hc->rx_bytes_recv += payload; 799 } 800 801 /* 802 * Perform loss detection and handle pending losses 803 */ 804 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist, 805 skb, ndp, ccid3_first_li, sk)) { 806 do_feedback = CCID3_FBACK_PARAM_CHANGE; 807 goto done_receiving; 808 } 809 810 if (tfrc_rx_hist_loss_pending(&hc->rx_hist)) 811 return; /* done receiving */ 812 813 /* 814 * Handle data packets: RTT sampling and monitoring p 815 */ 816 if (unlikely(!is_data_packet)) 817 goto update_records; 818 819 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) { 820 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb); 821 /* 822 * Empty loss history: no loss so far, hence p stays 0. 823 * Sample RTT values, since an RTT estimate is required for the 824 * computation of p when the first loss occurs; RFC 3448, 6.3.1. 825 */ 826 if (sample != 0) 827 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9); 828 829 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) { 830 /* 831 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean 832 * has decreased (resp. p has increased), send feedback now. 833 */ 834 do_feedback = CCID3_FBACK_PARAM_CHANGE; 835 } 836 837 /* 838 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3 839 */ 840 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3) 841 do_feedback = CCID3_FBACK_PERIODIC; 842 843update_records: 844 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp); 845 846done_receiving: 847 if (do_feedback) 848 ccid3_hc_rx_send_feedback(sk, skb, do_feedback); 849} 850 851static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk) 852{ 853 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid); 854 855 hc->rx_state = TFRC_RSTATE_NO_DATA; 856 tfrc_lh_init(&hc->rx_li_hist); 857 return tfrc_rx_hist_alloc(&hc->rx_hist); 858} 859 860static void ccid3_hc_rx_exit(struct sock *sk) 861{ 862 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk); 863 864 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM); 865 866 tfrc_rx_hist_purge(&hc->rx_hist); 867 tfrc_lh_cleanup(&hc->rx_li_hist); 868} 869 870static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info) 871{ 872 const struct ccid3_hc_rx_sock *hc; 873 874 /* Listen socks doesn't have a private CCID block */ 875 if (sk->sk_state == DCCP_LISTEN) 876 return; 877 878 hc = ccid3_hc_rx_sk(sk); 879 info->tcpi_ca_state = hc->rx_state; 880 info->tcpi_options |= TCPI_OPT_TIMESTAMPS; 881 info->tcpi_rcv_rtt = hc->rx_rtt; 882} 883 884static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len, 885 u32 __user *optval, int __user *optlen) 886{ 887 const struct ccid3_hc_rx_sock *hc; 888 struct tfrc_rx_info rx_info; 889 const void *val; 890 891 /* Listen socks doesn't have a private CCID block */ 892 if (sk->sk_state == DCCP_LISTEN) 893 return -EINVAL; 894 895 hc = ccid3_hc_rx_sk(sk); 896 switch (optname) { 897 case DCCP_SOCKOPT_CCID_RX_INFO: 898 if (len < sizeof(rx_info)) 899 return -EINVAL; 900 rx_info.tfrcrx_x_recv = hc->rx_x_recv; 901 rx_info.tfrcrx_rtt = hc->rx_rtt; 902 rx_info.tfrcrx_p = hc->rx_pinv == 0 ? ~0U : 903 scaled_div(1, hc->rx_pinv); 904 len = sizeof(rx_info); 905 val = &rx_info; 906 break; 907 default: 908 return -ENOPROTOOPT; 909 } 910 911 if (put_user(len, optlen) || copy_to_user(optval, val, len)) 912 return -EFAULT; 913 914 return 0; 915} 916 917struct ccid_operations ccid3_ops = { 918 .ccid_id = DCCPC_CCID3, 919 .ccid_name = "TCP-Friendly Rate Control", 920 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock), 921 .ccid_hc_tx_init = ccid3_hc_tx_init, 922 .ccid_hc_tx_exit = ccid3_hc_tx_exit, 923 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet, 924 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent, 925 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv, 926 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options, 927 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock), 928 .ccid_hc_rx_init = ccid3_hc_rx_init, 929 .ccid_hc_rx_exit = ccid3_hc_rx_exit, 930 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options, 931 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv, 932 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info, 933 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info, 934 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt, 935 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt, 936}; 937 938#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 939module_param(ccid3_debug, bool, 0644); 940MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages"); 941#endif 942