kernel/net/dccp/ccids/lib/packet_history.c

   1 /*
   2  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
   3  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
   4  *
   5  *  An implementation of the DCCP protocol
   6  *
   7  *  This code has been developed by the University of Waikato WAND
   8  *  research group. For further information please see http://www.wand.net.nz/
   9  *  or e-mail Ian McDonald - ian.mcdonald@jandi.co.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
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include "packet_history.h"
  39 #include "../../dccp.h"
  40
  41 /*
  42  * Transmitter History Routines
  43  */
  44 static struct kmem_cache *tfrc_tx_hist_slab;
  45
  46 int __init tfrc_tx_packet_history_init(void)
  47 {
  48         tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
  49                                               sizeof(struct tfrc_tx_hist_entry),
  50                                               0, SLAB_HWCACHE_ALIGN, NULL);
  51         return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
  52 }
  53
  54 void tfrc_tx_packet_history_exit(void)
  55 {
  56         if (tfrc_tx_hist_slab != NULL) {
  57                 kmem_cache_destroy(tfrc_tx_hist_slab);
  58                 tfrc_tx_hist_slab = NULL;
  59         }
  60 }
  61
  62 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
  63 {
  64         struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
  65
  66         if (entry == NULL)
  67                 return -ENOBUFS;
  68         entry->seqno = seqno;
  69         entry->stamp = ktime_get_real();
  70         entry->next  = *headp;
  71         *headp       = entry;
  72         return 0;
  73 }
  74
  75 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
  76 {
  77         struct tfrc_tx_hist_entry *head = *headp;
  78
  79         while (head != NULL) {
  80                 struct tfrc_tx_hist_entry *next = head->next;
  81
  82                 kmem_cache_free(tfrc_tx_hist_slab, head);
  83                 head = next;
  84         }
  85
  86         *headp = NULL;
  87 }
  88
  89 /*
  90  *      Receiver History Routines
  91  */
  92 static struct kmem_cache *tfrc_rx_hist_slab;
  93
  94 int __init tfrc_rx_packet_history_init(void)
  95 {
  96         tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
  97                                               sizeof(struct tfrc_rx_hist_entry),
  98                                               0, SLAB_HWCACHE_ALIGN, NULL);
  99         return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
 100 }
 101
 102 void tfrc_rx_packet_history_exit(void)
 103 {
 104         if (tfrc_rx_hist_slab != NULL) {
 105                 kmem_cache_destroy(tfrc_rx_hist_slab);
 106                 tfrc_rx_hist_slab = NULL;
 107         }
 108 }
 109
 110 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
 111                                                const struct sk_buff *skb,
 112                                                const u64 ndp)
 113 {
 114         const struct dccp_hdr *dh = dccp_hdr(skb);
 115
 116         entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
 117         entry->tfrchrx_ccval = dh->dccph_ccval;
 118         entry->tfrchrx_type  = dh->dccph_type;
 119         entry->tfrchrx_ndp   = ndp;
 120         entry->tfrchrx_tstamp = ktime_get_real();
 121 }
 122
 123 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
 124                              const struct sk_buff *skb,
 125                              const u64 ndp)
 126 {
 127         struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
 128
 129         tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
 130 }
 131
 132 /* has the packet contained in skb been seen before? */
 133 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
 134 {
 135         const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
 136         int i;
 137
 138         if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
 139                 return 1;
 140
 141         for (i = 1; i <= h->loss_count; i++)
 142                 if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
 143                         return 1;
 144
 145         return 0;
 146 }
 147
 148 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
 149 {
 150         const u8 idx_a = tfrc_rx_hist_index(h, a),
 151                  idx_b = tfrc_rx_hist_index(h, b);
 152         struct tfrc_rx_hist_entry *tmp = h->ring[idx_a];
 153
 154         h->ring[idx_a] = h->ring[idx_b];
 155         h->ring[idx_b] = tmp;
 156 }
 157
 158 /*
 159  * Private helper functions for loss detection.
 160  *
 161  * In the descriptions, `Si' refers to the sequence number of entry number i,
 162  * whose NDP count is `Ni' (lower case is used for variables).
 163  * Note: All __xxx_loss functions expect that a test against duplicates has been
 164  *       performed already: the seqno of the skb must not be less than the seqno
 165  *       of loss_prev; and it must not equal that of any valid history entry.
 166  */
 167 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
 168 {
 169         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 170             s1 = DCCP_SKB_CB(skb)->dccpd_seq;
 171
 172         if (!dccp_loss_free(s0, s1, n1)) {      /* gap between S0 and S1 */
 173                 h->loss_count = 1;
 174                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
 175         }
 176 }
 177
 178 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
 179 {
 180         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 181             s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 182             s2 = DCCP_SKB_CB(skb)->dccpd_seq;
 183
 184         if (likely(dccp_delta_seqno(s1, s2) > 0)) {     /* S1  <  S2 */
 185                 h->loss_count = 2;
 186                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
 187                 return;
 188         }
 189
 190         /* S0  <  S2  <  S1 */
 191
 192         if (dccp_loss_free(s0, s2, n2)) {
 193                 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
 194
 195                 if (dccp_loss_free(s2, s1, n1)) {
 196                         /* hole is filled: S0, S2, and S1 are consecutive */
 197                         h->loss_count = 0;
 198                         h->loss_start = tfrc_rx_hist_index(h, 1);
 199                 } else
 200                         /* gap between S2 and S1: just update loss_prev */
 201                         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
 202
 203         } else {        /* gap between S0 and S2 */
 204                 /*
 205                  * Reorder history to insert S2 between S0 and S1
 206                  */
 207                 tfrc_rx_hist_swap(h, 0, 3);
 208                 h->loss_start = tfrc_rx_hist_index(h, 3);
 209                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
 210                 h->loss_count = 2;
 211         }
 212 }
 213
 214 /* return 1 if a new loss event has been identified */
 215 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
 216 {
 217         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 218             s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 219             s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
 220             s3 = DCCP_SKB_CB(skb)->dccpd_seq;
 221
 222         if (likely(dccp_delta_seqno(s2, s3) > 0)) {     /* S2  <  S3 */
 223                 h->loss_count = 3;
 224                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
 225                 return 1;
 226         }
 227
 228         /* S3  <  S2 */
 229
 230         if (dccp_delta_seqno(s1, s3) > 0) {             /* S1  <  S3  <  S2 */
 231                 /*
 232                  * Reorder history to insert S3 between S1 and S2
 233                  */
 234                 tfrc_rx_hist_swap(h, 2, 3);
 235                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
 236                 h->loss_count = 3;
 237                 return 1;
 238         }
 239
 240         /* S0  <  S3  <  S1 */
 241
 242         if (dccp_loss_free(s0, s3, n3)) {
 243                 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
 244
 245                 if (dccp_loss_free(s3, s1, n1)) {
 246                         /* hole between S0 and S1 filled by S3 */
 247                         u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
 248
 249                         if (dccp_loss_free(s1, s2, n2)) {
 250                                 /* entire hole filled by S0, S3, S1, S2 */
 251                                 h->loss_start = tfrc_rx_hist_index(h, 2);
 252                                 h->loss_count = 0;
 253                         } else {
 254                                 /* gap remains between S1 and S2 */
 255                                 h->loss_start = tfrc_rx_hist_index(h, 1);
 256                                 h->loss_count = 1;
 257                         }
 258
 259                 } else /* gap exists between S3 and S1, loss_count stays at 2 */
 260                         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
 261
 262                 return 0;
 263         }
 264
 265         /*
 266          * The remaining case:  S0  <  S3  <  S1  <  S2;  gap between S0 and S3
 267          * Reorder history to insert S3 between S0 and S1.
 268          */
 269         tfrc_rx_hist_swap(h, 0, 3);
 270         h->loss_start = tfrc_rx_hist_index(h, 3);
 271         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
 272         h->loss_count = 3;
 273
 274         return 1;
 275 }
 276
 277 /* recycle RX history records to continue loss detection if necessary */
 278 static void __three_after_loss(struct tfrc_rx_hist *h)
 279 {
 280         /*
 281          * At this stage we know already that there is a gap between S0 and S1
 282          * (since S0 was the highest sequence number received before detecting
 283          * the loss). To recycle the loss record, it is thus only necessary to
 284          * check for other possible gaps between S1/S2 and between S2/S3.
 285          */
 286         u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 287             s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
 288             s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
 289         u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
 290             n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
 291
 292         if (dccp_loss_free(s1, s2, n2)) {
 293
 294                 if (dccp_loss_free(s2, s3, n3)) {
 295                         /* no gap between S2 and S3: entire hole is filled */
 296                         h->loss_start = tfrc_rx_hist_index(h, 3);
 297                         h->loss_count = 0;
 298                 } else {
 299                         /* gap between S2 and S3 */
 300                         h->loss_start = tfrc_rx_hist_index(h, 2);
 301                         h->loss_count = 1;
 302                 }
 303
 304         } else {        /* gap between S1 and S2 */
 305                 h->loss_start = tfrc_rx_hist_index(h, 1);
 306                 h->loss_count = 2;
 307         }
 308 }
 309
 310 /**
 311  *  tfrc_rx_handle_loss  -  Loss detection and further processing
 312  *  @h:             The non-empty RX history object
 313  *  @lh:            Loss Intervals database to update
 314  *  @skb:           Currently received packet
 315  *  @ndp:           The NDP count belonging to @skb
 316  *  @calc_first_li: Caller-dependent computation of first loss interval in @lh
 317  *  @sk:            Used by @calc_first_li (see tfrc_lh_interval_add)
 318  *
 319  *  Chooses action according to pending loss, updates LI database when a new
 320  *  loss was detected, and does required post-processing. Returns 1 when caller
 321  *  should send feedback, 0 otherwise.
 322  *  Since it also takes care of reordering during loss detection and updates the
 323  *  records accordingly, the caller should not perform any more RX history
 324  *  operations when loss_count is greater than 0 after calling this function.
 325  */
 326 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
 327                         struct tfrc_loss_hist *lh,
 328                         struct sk_buff *skb, const u64 ndp,
 329                         u32 (*calc_first_li)(struct sock *), struct sock *sk)
 330 {
 331         int is_new_loss = 0;
 332
 333         if (h->loss_count == 0) {
 334                 __do_track_loss(h, skb, ndp);
 335         } else if (h->loss_count == 1) {
 336                 __one_after_loss(h, skb, ndp);
 337         } else if (h->loss_count != 2) {
 338                 DCCP_BUG("invalid loss_count %d", h->loss_count);
 339         } else if (__two_after_loss(h, skb, ndp)) {
 340                 /*
 341                  * Update Loss Interval database and recycle RX records
 342                  */
 343                 is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
 344                 __three_after_loss(h);
 345         }
 346         return is_new_loss;
 347 }
 348
 349 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
 350 {
 351         int i;
 352
 353         for (i = 0; i <= TFRC_NDUPACK; i++) {
 354                 h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
 355                 if (h->ring[i] == NULL)
 356                         goto out_free;
 357         }
 358
 359         h->loss_count = h->loss_start = 0;
 360         return 0;
 361
 362 out_free:
 363         while (i-- != 0) {
 364                 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
 365                 h->ring[i] = NULL;
 366         }
 367         return -ENOBUFS;
 368 }
 369
 370 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
 371 {
 372         int i;
 373
 374         for (i = 0; i <= TFRC_NDUPACK; ++i)
 375                 if (h->ring[i] != NULL) {
 376                         kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
 377                         h->ring[i] = NULL;
 378                 }
 379 }
 380
 381 /**
 382  * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
 383  */
 384 static inline struct tfrc_rx_hist_entry *
 385                         tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
 386 {
 387         return h->ring[0];
 388 }
 389
 390 /**
 391  * tfrc_rx_hist_rtt_prev_s - previously suitable (wrt rtt_last_s) RTT-sampling entry
 392  */
 393 static inline struct tfrc_rx_hist_entry *
 394                         tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
 395 {
 396         return h->ring[h->rtt_sample_prev];
 397 }
 398
 399 /**
 400  * tfrc_rx_hist_sample_rtt  -  Sample RTT from timestamp / CCVal
 401  * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
 402  * to compute a sample with given data - calling function should check this.
 403  */
 404 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
 405 {
 406         u32 sample = 0,
 407             delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
 408                             tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 409
 410         if (delta_v < 1 || delta_v > 4) {       /* unsuitable CCVal delta */
 411                 if (h->rtt_sample_prev == 2) {  /* previous candidate stored */
 412                         sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
 413                                        tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 414                         if (sample)
 415                                 sample = 4 / sample *
 416                                          ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
 417                                                         tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
 418                         else    /*
 419                                  * FIXME: This condition is in principle not
 420                                  * possible but occurs when CCID is used for
 421                                  * two-way data traffic. I have tried to trace
 422                                  * it, but the cause does not seem to be here.
 423                                  */
 424                                 DCCP_BUG("please report to dccp@vger.kernel.org"
 425                                          " => prev = %u, last = %u",
 426                                          tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
 427                                          tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 428                 } else if (delta_v < 1) {
 429                         h->rtt_sample_prev = 1;
 430                         goto keep_ref_for_next_time;
 431                 }
 432
 433         } else if (delta_v == 4) /* optimal match */
 434                 sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
 435         else {                   /* suboptimal match */
 436                 h->rtt_sample_prev = 2;
 437                 goto keep_ref_for_next_time;
 438         }
 439
 440         if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
 441                 DCCP_WARN("RTT sample %u too large, using max\n", sample);
 442                 sample = DCCP_SANE_RTT_MAX;
 443         }
 444
 445         h->rtt_sample_prev = 0;        /* use current entry as next reference */
 446 keep_ref_for_next_time:
 447
 448         return sample;
 449 }