1 /* 2 * net/dccp/packet_history.c 3 * 4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK 5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand. 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 * or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz 12 * 13 * This code also uses code from Lulea University, rereleased as GPL by its 14 * authors: 15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon 16 * 17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft 18 * and to make it work as a loadable module in the DCCP stack written by 19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>. 20 * 21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br> 22 * 23 * This program is free software; you can redistribute it and/or modify 24 * it under the terms of the GNU General Public License as published by 25 * the Free Software Foundation; either version 2 of the License, or 26 * (at your option) any later version. 27 * 28 * This program is distributed in the hope that it will be useful, 29 * but WITHOUT ANY WARRANTY; without even the implied warranty of 30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 31 * GNU General Public License for more details. 32 * 33 * You should have received a copy of the GNU General Public License 34 * along with this program; if not, write to the Free Software 35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 36 */ 37 38 #include <linux/string.h> 39 #include <linux/slab.h> 40 #include "packet_history.h" 41 #include "../../dccp.h" 42 43 /** 44 * tfrc_tx_hist_entry - Simple singly-linked TX history list 45 * @next: next oldest entry (LIFO order) 46 * @seqno: sequence number of this entry 47 * @stamp: send time of packet with sequence number @seqno 48 */ 49 struct tfrc_tx_hist_entry { 50 struct tfrc_tx_hist_entry *next; 51 u64 seqno; 52 ktime_t stamp; 53 }; 54 55 /* 56 * Transmitter History Routines 57 */ 58 static struct kmem_cache *tfrc_tx_hist_slab; 59 60 int __init tfrc_tx_packet_history_init(void) 61 { 62 tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist", 63 sizeof(struct tfrc_tx_hist_entry), 64 0, SLAB_HWCACHE_ALIGN, NULL); 65 return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0; 66 } 67 68 void tfrc_tx_packet_history_exit(void) 69 { 70 if (tfrc_tx_hist_slab != NULL) { 71 kmem_cache_destroy(tfrc_tx_hist_slab); 72 tfrc_tx_hist_slab = NULL; 73 } 74 } 75 76 static struct tfrc_tx_hist_entry * 77 tfrc_tx_hist_find_entry(struct tfrc_tx_hist_entry *head, u64 seqno) 78 { 79 while (head != NULL && head->seqno != seqno) 80 head = head->next; 81 82 return head; 83 } 84 85 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno) 86 { 87 struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any()); 88 89 if (entry == NULL) 90 return -ENOBUFS; 91 entry->seqno = seqno; 92 entry->stamp = ktime_get_real(); 93 entry->next = *headp; 94 *headp = entry; 95 return 0; 96 } 97 EXPORT_SYMBOL_GPL(tfrc_tx_hist_add); 98 99 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp) 100 { 101 struct tfrc_tx_hist_entry *head = *headp; 102 103 while (head != NULL) { 104 struct tfrc_tx_hist_entry *next = head->next; 105 106 kmem_cache_free(tfrc_tx_hist_slab, head); 107 head = next; 108 } 109 110 *headp = NULL; 111 } 112 EXPORT_SYMBOL_GPL(tfrc_tx_hist_purge); 113 114 u32 tfrc_tx_hist_rtt(struct tfrc_tx_hist_entry *head, const u64 seqno, 115 const ktime_t now) 116 { 117 u32 rtt = 0; 118 struct tfrc_tx_hist_entry *packet = tfrc_tx_hist_find_entry(head, seqno); 119 120 if (packet != NULL) { 121 rtt = ktime_us_delta(now, packet->stamp); 122 /* 123 * Garbage-collect older (irrelevant) entries: 124 */ 125 tfrc_tx_hist_purge(&packet->next); 126 } 127 128 return rtt; 129 } 130 EXPORT_SYMBOL_GPL(tfrc_tx_hist_rtt); 131 132 133 /* 134 * Receiver History Routines 135 */ 136 static struct kmem_cache *tfrc_rx_hist_slab; 137 138 int __init tfrc_rx_packet_history_init(void) 139 { 140 tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache", 141 sizeof(struct tfrc_rx_hist_entry), 142 0, SLAB_HWCACHE_ALIGN, NULL); 143 return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0; 144 } 145 146 void tfrc_rx_packet_history_exit(void) 147 { 148 if (tfrc_rx_hist_slab != NULL) { 149 kmem_cache_destroy(tfrc_rx_hist_slab); 150 tfrc_rx_hist_slab = NULL; 151 } 152 } 153 154 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry, 155 const struct sk_buff *skb, 156 const u32 ndp) 157 { 158 const struct dccp_hdr *dh = dccp_hdr(skb); 159 160 entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq; 161 entry->tfrchrx_ccval = dh->dccph_ccval; 162 entry->tfrchrx_type = dh->dccph_type; 163 entry->tfrchrx_ndp = ndp; 164 entry->tfrchrx_tstamp = ktime_get_real(); 165 } 166 167 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h, 168 const struct sk_buff *skb, 169 const u32 ndp) 170 { 171 struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h); 172 173 tfrc_rx_hist_entry_from_skb(entry, skb, ndp); 174 } 175 EXPORT_SYMBOL_GPL(tfrc_rx_hist_add_packet); 176 177 /* has the packet contained in skb been seen before? */ 178 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb) 179 { 180 const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq; 181 int i; 182 183 if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0) 184 return 1; 185 186 for (i = 1; i <= h->loss_count; i++) 187 if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq) 188 return 1; 189 190 return 0; 191 } 192 EXPORT_SYMBOL_GPL(tfrc_rx_hist_duplicate); 193 194 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b) 195 { 196 const u8 idx_a = tfrc_rx_hist_index(h, a), 197 idx_b = tfrc_rx_hist_index(h, b); 198 struct tfrc_rx_hist_entry *tmp = h->ring[idx_a]; 199 200 h->ring[idx_a] = h->ring[idx_b]; 201 h->ring[idx_b] = tmp; 202 } 203 204 /* 205 * Private helper functions for loss detection. 206 * 207 * In the descriptions, `Si' refers to the sequence number of entry number i, 208 * whose NDP count is `Ni' (lower case is used for variables). 209 * Note: All __after_loss functions expect that a test against duplicates has 210 * been performed already: the seqno of the skb must not be less than the 211 * seqno of loss_prev; and it must not equal that of any valid hist_entry. 212 */ 213 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2) 214 { 215 u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, 216 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno, 217 s2 = DCCP_SKB_CB(skb)->dccpd_seq; 218 int n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp, 219 d12 = dccp_delta_seqno(s1, s2), d2; 220 221 if (d12 > 0) { /* S1 < S2 */ 222 h->loss_count = 2; 223 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2); 224 return; 225 } 226 227 /* S0 < S2 < S1 */ 228 d2 = dccp_delta_seqno(s0, s2); 229 230 if (d2 == 1 || n2 >= d2) { /* S2 is direct successor of S0 */ 231 int d21 = -d12; 232 233 if (d21 == 1 || n1 >= d21) { 234 /* hole is filled: S0, S2, and S1 are consecutive */ 235 h->loss_count = 0; 236 h->loss_start = tfrc_rx_hist_index(h, 1); 237 } else 238 /* gap between S2 and S1: just update loss_prev */ 239 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2); 240 241 } else { /* hole between S0 and S2 */ 242 /* 243 * Reorder history to insert S2 between S0 and s1 244 */ 245 tfrc_rx_hist_swap(h, 0, 3); 246 h->loss_start = tfrc_rx_hist_index(h, 3); 247 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2); 248 h->loss_count = 2; 249 } 250 } 251 252 /* return 1 if a new loss event has been identified */ 253 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3) 254 { 255 u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, 256 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno, 257 s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno, 258 s3 = DCCP_SKB_CB(skb)->dccpd_seq; 259 int n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp, 260 d23 = dccp_delta_seqno(s2, s3), d13, d3, d31; 261 262 if (d23 > 0) { /* S2 < S3 */ 263 h->loss_count = 3; 264 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3); 265 return 1; 266 } 267 268 /* S3 < S2 */ 269 d13 = dccp_delta_seqno(s1, s3); 270 271 if (d13 > 0) { 272 /* 273 * The sequence number order is S1, S3, S2 274 * Reorder history to insert entry between S1 and S2 275 */ 276 tfrc_rx_hist_swap(h, 2, 3); 277 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3); 278 h->loss_count = 3; 279 return 1; 280 } 281 282 /* S0 < S3 < S1 */ 283 d31 = -d13; 284 d3 = dccp_delta_seqno(s0, s3); 285 286 if (d3 == 1 || n3 >= d3) { /* S3 is a successor of S0 */ 287 288 if (d31 == 1 || n1 >= d31) { 289 /* hole between S0 and S1 filled by S3 */ 290 int d2 = dccp_delta_seqno(s1, s2), 291 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp; 292 293 if (d2 == 1 || n2 >= d2) { 294 /* entire hole filled by S0, S3, S1, S2 */ 295 h->loss_start = tfrc_rx_hist_index(h, 2); 296 h->loss_count = 0; 297 } else { 298 /* gap remains between S1 and S2 */ 299 h->loss_start = tfrc_rx_hist_index(h, 1); 300 h->loss_count = 1; 301 } 302 303 } else /* gap exists between S3 and S1, loss_count stays at 2 */ 304 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3); 305 306 return 0; 307 } 308 309 /* 310 * The remaining case: S3 is not a successor of S0. 311 * Sequence order is S0, S3, S1, S2; reorder to insert between S0 and S1 312 */ 313 tfrc_rx_hist_swap(h, 0, 3); 314 h->loss_start = tfrc_rx_hist_index(h, 3); 315 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3); 316 h->loss_count = 3; 317 318 return 1; 319 } 320 321 /* return the signed modulo-2^48 sequence number distance from entry e1 to e2 */ 322 static s64 tfrc_rx_hist_delta_seqno(struct tfrc_rx_hist *h, u8 e1, u8 e2) 323 { 324 DCCP_BUG_ON(e1 > h->loss_count || e2 > h->loss_count); 325 326 return dccp_delta_seqno(tfrc_rx_hist_entry(h, e1)->tfrchrx_seqno, 327 tfrc_rx_hist_entry(h, e2)->tfrchrx_seqno); 328 } 329 330 /* recycle RX history records to continue loss detection if necessary */ 331 static void __three_after_loss(struct tfrc_rx_hist *h) 332 { 333 /* 334 * The distance between S0 and S1 is always greater than 1 and the NDP 335 * count of S1 is smaller than this distance. Otherwise there would 336 * have been no loss. Hence it is only necessary to see whether there 337 * are further missing data packets between S1/S2 and S2/S3. 338 */ 339 int d2 = tfrc_rx_hist_delta_seqno(h, 1, 2), 340 d3 = tfrc_rx_hist_delta_seqno(h, 2, 3), 341 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp, 342 n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp; 343 344 if (d2 == 1 || n2 >= d2) { /* S2 is successor to S1 */ 345 346 if (d3 == 1 || n3 >= d3) { 347 /* S3 is successor of S2: entire hole is filled */ 348 h->loss_start = tfrc_rx_hist_index(h, 3); 349 h->loss_count = 0; 350 } else { 351 /* gap between S2 and S3 */ 352 h->loss_start = tfrc_rx_hist_index(h, 2); 353 h->loss_count = 1; 354 } 355 356 } else { /* gap between S1 and S2 */ 357 h->loss_start = tfrc_rx_hist_index(h, 1); 358 h->loss_count = 2; 359 } 360 } 361 362 /** 363 * tfrc_rx_handle_loss - Loss detection and further processing 364 * @h: The non-empty RX history object 365 * @lh: Loss Intervals database to update 366 * @skb: Currently received packet 367 * @ndp: The NDP count belonging to @skb 368 * @calc_first_li: Caller-dependent computation of first loss interval in @lh 369 * @sk: Used by @calc_first_li (see tfrc_lh_interval_add) 370 * Chooses action according to pending loss, updates LI database when a new 371 * loss was detected, and does required post-processing. Returns 1 when caller 372 * should send feedback, 0 otherwise. 373 */ 374 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h, 375 struct tfrc_loss_hist *lh, 376 struct sk_buff *skb, u32 ndp, 377 u32 (*calc_first_li)(struct sock *), struct sock *sk) 378 { 379 int is_new_loss = 0; 380 381 if (h->loss_count == 1) { 382 __one_after_loss(h, skb, ndp); 383 } else if (h->loss_count != 2) { 384 DCCP_BUG("invalid loss_count %d", h->loss_count); 385 } else if (__two_after_loss(h, skb, ndp)) { 386 /* 387 * Update Loss Interval database and recycle RX records 388 */ 389 is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk); 390 __three_after_loss(h); 391 } 392 return is_new_loss; 393 } 394 EXPORT_SYMBOL_GPL(tfrc_rx_handle_loss); 395 396 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h) 397 { 398 int i; 399 400 for (i = 0; i <= TFRC_NDUPACK; i++) { 401 h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC); 402 if (h->ring[i] == NULL) 403 goto out_free; 404 } 405 406 h->loss_count = h->loss_start = 0; 407 return 0; 408 409 out_free: 410 while (i-- != 0) { 411 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]); 412 h->ring[i] = NULL; 413 } 414 return -ENOBUFS; 415 } 416 EXPORT_SYMBOL_GPL(tfrc_rx_hist_alloc); 417 418 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h) 419 { 420 int i; 421 422 for (i = 0; i <= TFRC_NDUPACK; ++i) 423 if (h->ring[i] != NULL) { 424 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]); 425 h->ring[i] = NULL; 426 } 427 } 428 EXPORT_SYMBOL_GPL(tfrc_rx_hist_purge); 429 430 /** 431 * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against 432 */ 433 static inline struct tfrc_rx_hist_entry * 434 tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h) 435 { 436 return h->ring[0]; 437 } 438 439 /** 440 * tfrc_rx_hist_rtt_prev_s: previously suitable (wrt rtt_last_s) RTT-sampling entry 441 */ 442 static inline struct tfrc_rx_hist_entry * 443 tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h) 444 { 445 return h->ring[h->rtt_sample_prev]; 446 } 447 448 /** 449 * tfrc_rx_hist_sample_rtt - Sample RTT from timestamp / CCVal 450 * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able 451 * to compute a sample with given data - calling function should check this. 452 */ 453 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb) 454 { 455 u32 sample = 0, 456 delta_v = SUB16(dccp_hdr(skb)->dccph_ccval, 457 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval); 458 459 if (delta_v < 1 || delta_v > 4) { /* unsuitable CCVal delta */ 460 if (h->rtt_sample_prev == 2) { /* previous candidate stored */ 461 sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval, 462 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval); 463 if (sample) 464 sample = 4 / sample * 465 ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp, 466 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp); 467 else /* 468 * FIXME: This condition is in principle not 469 * possible but occurs when CCID is used for 470 * two-way data traffic. I have tried to trace 471 * it, but the cause does not seem to be here. 472 */ 473 DCCP_BUG("please report to dccp@vger.kernel.org" 474 " => prev = %u, last = %u", 475 tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval, 476 tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval); 477 } else if (delta_v < 1) { 478 h->rtt_sample_prev = 1; 479 goto keep_ref_for_next_time; 480 } 481 482 } else if (delta_v == 4) /* optimal match */ 483 sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp)); 484 else { /* suboptimal match */ 485 h->rtt_sample_prev = 2; 486 goto keep_ref_for_next_time; 487 } 488 489 if (unlikely(sample > DCCP_SANE_RTT_MAX)) { 490 DCCP_WARN("RTT sample %u too large, using max\n", sample); 491 sample = DCCP_SANE_RTT_MAX; 492 } 493 494 h->rtt_sample_prev = 0; /* use current entry as next reference */ 495 keep_ref_for_next_time: 496 497 return sample; 498 } 499 EXPORT_SYMBOL_GPL(tfrc_rx_hist_sample_rtt); 500