net/dccp/ccids/lib/packet_history.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
   4  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
   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  *  or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
  11  *
  12  *  This code also uses code from Lulea University, rereleased as GPL by its
  13  *  authors:
  14  *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
  15  *
  16  *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
  17  *  and to make it work as a loadable module in the DCCP stack written by
  18  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
  19  *
  20  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  21  */
  22
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include "packet_history.h"
  26 #include "../../dccp.h"
  27
  28 /*
  29  * Transmitter History Routines
  30  */
  31 static struct kmem_cache *tfrc_tx_hist_slab;
  32
  33 int __init tfrc_tx_packet_history_init(void)
  34 {
  35         tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
  36                                               sizeof(struct tfrc_tx_hist_entry),
  37                                               0, SLAB_HWCACHE_ALIGN, NULL);
  38         return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
  39 }
  40
  41 void tfrc_tx_packet_history_exit(void)
  42 {
  43         if (tfrc_tx_hist_slab != NULL) {
  44                 kmem_cache_destroy(tfrc_tx_hist_slab);
  45                 tfrc_tx_hist_slab = NULL;
  46         }
  47 }
  48
  49 int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
  50 {
  51         struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
  52
  53         if (entry == NULL)
  54                 return -ENOBUFS;
  55         entry->seqno = seqno;
  56         entry->stamp = ktime_get_real();
  57         entry->next  = *headp;
  58         *headp       = entry;
  59         return 0;
  60 }
  61
  62 void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
  63 {
  64         struct tfrc_tx_hist_entry *head = *headp;
  65
  66         while (head != NULL) {
  67                 struct tfrc_tx_hist_entry *next = head->next;
  68
  69                 kmem_cache_free(tfrc_tx_hist_slab, head);
  70                 head = next;
  71         }
  72
  73         *headp = NULL;
  74 }
  75
  76 /*
  77  *      Receiver History Routines
  78  */
  79 static struct kmem_cache *tfrc_rx_hist_slab;
  80
  81 int __init tfrc_rx_packet_history_init(void)
  82 {
  83         tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
  84                                               sizeof(struct tfrc_rx_hist_entry),
  85                                               0, SLAB_HWCACHE_ALIGN, NULL);
  86         return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
  87 }
  88
  89 void tfrc_rx_packet_history_exit(void)
  90 {
  91         if (tfrc_rx_hist_slab != NULL) {
  92                 kmem_cache_destroy(tfrc_rx_hist_slab);
  93                 tfrc_rx_hist_slab = NULL;
  94         }
  95 }
  96
  97 static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
  98                                                const struct sk_buff *skb,
  99                                                const u64 ndp)
 100 {
 101         const struct dccp_hdr *dh = dccp_hdr(skb);
 102
 103         entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
 104         entry->tfrchrx_ccval = dh->dccph_ccval;
 105         entry->tfrchrx_type  = dh->dccph_type;
 106         entry->tfrchrx_ndp   = ndp;
 107         entry->tfrchrx_tstamp = ktime_get_real();
 108 }
 109
 110 void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
 111                              const struct sk_buff *skb,
 112                              const u64 ndp)
 113 {
 114         struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
 115
 116         tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
 117 }
 118
 119 /* has the packet contained in skb been seen before? */
 120 int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
 121 {
 122         const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
 123         int i;
 124
 125         if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
 126                 return 1;
 127
 128         for (i = 1; i <= h->loss_count; i++)
 129                 if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
 130                         return 1;
 131
 132         return 0;
 133 }
 134
 135 static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
 136 {
 137         const u8 idx_a = tfrc_rx_hist_index(h, a),
 138                  idx_b = tfrc_rx_hist_index(h, b);
 139
 140         swap(h->ring[idx_a], h->ring[idx_b]);
 141 }
 142
 143 /*
 144  * Private helper functions for loss detection.
 145  *
 146  * In the descriptions, `Si' refers to the sequence number of entry number i,
 147  * whose NDP count is `Ni' (lower case is used for variables).
 148  * Note: All __xxx_loss functions expect that a test against duplicates has been
 149  *       performed already: the seqno of the skb must not be less than the seqno
 150  *       of loss_prev; and it must not equal that of any valid history entry.
 151  */
 152 static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
 153 {
 154         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 155             s1 = DCCP_SKB_CB(skb)->dccpd_seq;
 156
 157         if (!dccp_loss_free(s0, s1, n1)) {      /* gap between S0 and S1 */
 158                 h->loss_count = 1;
 159                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n1);
 160         }
 161 }
 162
 163 static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
 164 {
 165         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 166             s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 167             s2 = DCCP_SKB_CB(skb)->dccpd_seq;
 168
 169         if (likely(dccp_delta_seqno(s1, s2) > 0)) {     /* S1  <  S2 */
 170                 h->loss_count = 2;
 171                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
 172                 return;
 173         }
 174
 175         /* S0  <  S2  <  S1 */
 176
 177         if (dccp_loss_free(s0, s2, n2)) {
 178                 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
 179
 180                 if (dccp_loss_free(s2, s1, n1)) {
 181                         /* hole is filled: S0, S2, and S1 are consecutive */
 182                         h->loss_count = 0;
 183                         h->loss_start = tfrc_rx_hist_index(h, 1);
 184                 } else
 185                         /* gap between S2 and S1: just update loss_prev */
 186                         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
 187
 188         } else {        /* gap between S0 and S2 */
 189                 /*
 190                  * Reorder history to insert S2 between S0 and S1
 191                  */
 192                 tfrc_rx_hist_swap(h, 0, 3);
 193                 h->loss_start = tfrc_rx_hist_index(h, 3);
 194                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
 195                 h->loss_count = 2;
 196         }
 197 }
 198
 199 /* return 1 if a new loss event has been identified */
 200 static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
 201 {
 202         u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
 203             s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 204             s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
 205             s3 = DCCP_SKB_CB(skb)->dccpd_seq;
 206
 207         if (likely(dccp_delta_seqno(s2, s3) > 0)) {     /* S2  <  S3 */
 208                 h->loss_count = 3;
 209                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
 210                 return 1;
 211         }
 212
 213         /* S3  <  S2 */
 214
 215         if (dccp_delta_seqno(s1, s3) > 0) {             /* S1  <  S3  <  S2 */
 216                 /*
 217                  * Reorder history to insert S3 between S1 and S2
 218                  */
 219                 tfrc_rx_hist_swap(h, 2, 3);
 220                 tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
 221                 h->loss_count = 3;
 222                 return 1;
 223         }
 224
 225         /* S0  <  S3  <  S1 */
 226
 227         if (dccp_loss_free(s0, s3, n3)) {
 228                 u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
 229
 230                 if (dccp_loss_free(s3, s1, n1)) {
 231                         /* hole between S0 and S1 filled by S3 */
 232                         u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
 233
 234                         if (dccp_loss_free(s1, s2, n2)) {
 235                                 /* entire hole filled by S0, S3, S1, S2 */
 236                                 h->loss_start = tfrc_rx_hist_index(h, 2);
 237                                 h->loss_count = 0;
 238                         } else {
 239                                 /* gap remains between S1 and S2 */
 240                                 h->loss_start = tfrc_rx_hist_index(h, 1);
 241                                 h->loss_count = 1;
 242                         }
 243
 244                 } else /* gap exists between S3 and S1, loss_count stays at 2 */
 245                         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
 246
 247                 return 0;
 248         }
 249
 250         /*
 251          * The remaining case:  S0  <  S3  <  S1  <  S2;  gap between S0 and S3
 252          * Reorder history to insert S3 between S0 and S1.
 253          */
 254         tfrc_rx_hist_swap(h, 0, 3);
 255         h->loss_start = tfrc_rx_hist_index(h, 3);
 256         tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
 257         h->loss_count = 3;
 258
 259         return 1;
 260 }
 261
 262 /* recycle RX history records to continue loss detection if necessary */
 263 static void __three_after_loss(struct tfrc_rx_hist *h)
 264 {
 265         /*
 266          * At this stage we know already that there is a gap between S0 and S1
 267          * (since S0 was the highest sequence number received before detecting
 268          * the loss). To recycle the loss record, it is thus only necessary to
 269          * check for other possible gaps between S1/S2 and between S2/S3.
 270          */
 271         u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
 272             s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
 273             s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
 274         u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
 275             n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
 276
 277         if (dccp_loss_free(s1, s2, n2)) {
 278
 279                 if (dccp_loss_free(s2, s3, n3)) {
 280                         /* no gap between S2 and S3: entire hole is filled */
 281                         h->loss_start = tfrc_rx_hist_index(h, 3);
 282                         h->loss_count = 0;
 283                 } else {
 284                         /* gap between S2 and S3 */
 285                         h->loss_start = tfrc_rx_hist_index(h, 2);
 286                         h->loss_count = 1;
 287                 }
 288
 289         } else {        /* gap between S1 and S2 */
 290                 h->loss_start = tfrc_rx_hist_index(h, 1);
 291                 h->loss_count = 2;
 292         }
 293 }
 294
 295 /**
 296  *  tfrc_rx_handle_loss  -  Loss detection and further processing
 297  *  @h:             The non-empty RX history object
 298  *  @lh:            Loss Intervals database to update
 299  *  @skb:           Currently received packet
 300  *  @ndp:           The NDP count belonging to @skb
 301  *  @calc_first_li: Caller-dependent computation of first loss interval in @lh
 302  *  @sk:            Used by @calc_first_li (see tfrc_lh_interval_add)
 303  *
 304  *  Chooses action according to pending loss, updates LI database when a new
 305  *  loss was detected, and does required post-processing. Returns 1 when caller
 306  *  should send feedback, 0 otherwise.
 307  *  Since it also takes care of reordering during loss detection and updates the
 308  *  records accordingly, the caller should not perform any more RX history
 309  *  operations when loss_count is greater than 0 after calling this function.
 310  */
 311 int tfrc_rx_handle_loss(struct tfrc_rx_hist *h,
 312                         struct tfrc_loss_hist *lh,
 313                         struct sk_buff *skb, const u64 ndp,
 314                         u32 (*calc_first_li)(struct sock *), struct sock *sk)
 315 {
 316         int is_new_loss = 0;
 317
 318         if (h->loss_count == 0) {
 319                 __do_track_loss(h, skb, ndp);
 320         } else if (h->loss_count == 1) {
 321                 __one_after_loss(h, skb, ndp);
 322         } else if (h->loss_count != 2) {
 323                 DCCP_BUG("invalid loss_count %d", h->loss_count);
 324         } else if (__two_after_loss(h, skb, ndp)) {
 325                 /*
 326                  * Update Loss Interval database and recycle RX records
 327                  */
 328                 is_new_loss = tfrc_lh_interval_add(lh, h, calc_first_li, sk);
 329                 __three_after_loss(h);
 330         }
 331         return is_new_loss;
 332 }
 333
 334 int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
 335 {
 336         int i;
 337
 338         for (i = 0; i <= TFRC_NDUPACK; i++) {
 339                 h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
 340                 if (h->ring[i] == NULL)
 341                         goto out_free;
 342         }
 343
 344         h->loss_count = h->loss_start = 0;
 345         return 0;
 346
 347 out_free:
 348         while (i-- != 0) {
 349                 kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
 350                 h->ring[i] = NULL;
 351         }
 352         return -ENOBUFS;
 353 }
 354
 355 void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
 356 {
 357         int i;
 358
 359         for (i = 0; i <= TFRC_NDUPACK; ++i)
 360                 if (h->ring[i] != NULL) {
 361                         kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
 362                         h->ring[i] = NULL;
 363                 }
 364 }
 365
 366 /**
 367  * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
 368  */
 369 static inline struct tfrc_rx_hist_entry *
 370                         tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
 371 {
 372         return h->ring[0];
 373 }
 374
 375 /**
 376  * tfrc_rx_hist_rtt_prev_s - previously suitable (wrt rtt_last_s) RTT-sampling entry
 377  */
 378 static inline struct tfrc_rx_hist_entry *
 379                         tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
 380 {
 381         return h->ring[h->rtt_sample_prev];
 382 }
 383
 384 /**
 385  * tfrc_rx_hist_sample_rtt  -  Sample RTT from timestamp / CCVal
 386  * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
 387  * to compute a sample with given data - calling function should check this.
 388  */
 389 u32 tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
 390 {
 391         u32 sample = 0,
 392             delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
 393                             tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 394
 395         if (delta_v < 1 || delta_v > 4) {       /* unsuitable CCVal delta */
 396                 if (h->rtt_sample_prev == 2) {  /* previous candidate stored */
 397                         sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
 398                                        tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 399                         if (sample)
 400                                 sample = 4 / sample *
 401                                          ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
 402                                                         tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
 403                         else    /*
 404                                  * FIXME: This condition is in principle not
 405                                  * possible but occurs when CCID is used for
 406                                  * two-way data traffic. I have tried to trace
 407                                  * it, but the cause does not seem to be here.
 408                                  */
 409                                 DCCP_BUG("please report to dccp@vger.kernel.org"
 410                                          " => prev = %u, last = %u",
 411                                          tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
 412                                          tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
 413                 } else if (delta_v < 1) {
 414                         h->rtt_sample_prev = 1;
 415                         goto keep_ref_for_next_time;
 416                 }
 417
 418         } else if (delta_v == 4) /* optimal match */
 419                 sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
 420         else {                   /* suboptimal match */
 421                 h->rtt_sample_prev = 2;
 422                 goto keep_ref_for_next_time;
 423         }
 424
 425         if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
 426                 DCCP_WARN("RTT sample %u too large, using max\n", sample);
 427                 sample = DCCP_SANE_RTT_MAX;
 428         }
 429
 430         h->rtt_sample_prev = 0;        /* use current entry as next reference */
 431 keep_ref_for_next_time:
 432
 433         return sample;
 434 }