diff options
Diffstat (limited to 'net/dccp/ccids/ccid2.c')
| -rw-r--r-- | net/dccp/ccids/ccid2.c | 779 |
1 files changed, 779 insertions, 0 deletions
diff --git a/net/dccp/ccids/ccid2.c b/net/dccp/ccids/ccid2.c new file mode 100644 index 000000000000..d4f9e2d33453 --- /dev/null +++ b/net/dccp/ccids/ccid2.c | |||
| @@ -0,0 +1,779 @@ | |||
| 1 | /* | ||
| 2 | * net/dccp/ccids/ccid2.c | ||
| 3 | * | ||
| 4 | * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk> | ||
| 5 | * | ||
| 6 | * Changes to meet Linux coding standards, and DCCP infrastructure fixes. | ||
| 7 | * | ||
| 8 | * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br> | ||
| 9 | * | ||
| 10 | * This program is free software; you can redistribute it and/or modify | ||
| 11 | * it under the terms of the GNU General Public License as published by | ||
| 12 | * the Free Software Foundation; either version 2 of the License, or | ||
| 13 | * (at your option) any later version. | ||
| 14 | * | ||
| 15 | * This program is distributed in the hope that it will be useful, | ||
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 18 | * GNU General Public License for more details. | ||
| 19 | * | ||
| 20 | * You should have received a copy of the GNU General Public License | ||
| 21 | * along with this program; if not, write to the Free Software | ||
| 22 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
| 23 | */ | ||
| 24 | |||
| 25 | /* | ||
| 26 | * This implementation should follow: draft-ietf-dccp-ccid2-10.txt | ||
| 27 | * | ||
| 28 | * BUGS: | ||
| 29 | * - sequence number wrapping | ||
| 30 | * - jiffies wrapping | ||
| 31 | */ | ||
| 32 | |||
| 33 | #include <linux/config.h> | ||
| 34 | #include "../ccid.h" | ||
| 35 | #include "../dccp.h" | ||
| 36 | #include "ccid2.h" | ||
| 37 | |||
| 38 | static int ccid2_debug; | ||
| 39 | |||
| 40 | #undef CCID2_DEBUG | ||
| 41 | #ifdef CCID2_DEBUG | ||
| 42 | #define ccid2_pr_debug(format, a...) \ | ||
| 43 | do { if (ccid2_debug) \ | ||
| 44 | printk(KERN_DEBUG "%s: " format, __FUNCTION__, ##a); \ | ||
| 45 | } while (0) | ||
| 46 | #else | ||
| 47 | #define ccid2_pr_debug(format, a...) | ||
| 48 | #endif | ||
| 49 | |||
| 50 | static const int ccid2_seq_len = 128; | ||
| 51 | |||
| 52 | #ifdef CCID2_DEBUG | ||
| 53 | static void ccid2_hc_tx_check_sanity(const struct ccid2_hc_tx_sock *hctx) | ||
| 54 | { | ||
| 55 | int len = 0; | ||
| 56 | int pipe = 0; | ||
| 57 | struct ccid2_seq *seqp = hctx->ccid2hctx_seqh; | ||
| 58 | |||
| 59 | /* there is data in the chain */ | ||
| 60 | if (seqp != hctx->ccid2hctx_seqt) { | ||
| 61 | seqp = seqp->ccid2s_prev; | ||
| 62 | len++; | ||
| 63 | if (!seqp->ccid2s_acked) | ||
| 64 | pipe++; | ||
| 65 | |||
| 66 | while (seqp != hctx->ccid2hctx_seqt) { | ||
| 67 | struct ccid2_seq *prev = seqp->ccid2s_prev; | ||
| 68 | |||
| 69 | len++; | ||
| 70 | if (!prev->ccid2s_acked) | ||
| 71 | pipe++; | ||
| 72 | |||
| 73 | /* packets are sent sequentially */ | ||
| 74 | BUG_ON(seqp->ccid2s_seq <= prev->ccid2s_seq); | ||
| 75 | BUG_ON(seqp->ccid2s_sent < prev->ccid2s_sent); | ||
| 76 | BUG_ON(len > ccid2_seq_len); | ||
| 77 | |||
| 78 | seqp = prev; | ||
| 79 | } | ||
| 80 | } | ||
| 81 | |||
| 82 | BUG_ON(pipe != hctx->ccid2hctx_pipe); | ||
| 83 | ccid2_pr_debug("len of chain=%d\n", len); | ||
| 84 | |||
| 85 | do { | ||
| 86 | seqp = seqp->ccid2s_prev; | ||
| 87 | len++; | ||
| 88 | BUG_ON(len > ccid2_seq_len); | ||
| 89 | } while (seqp != hctx->ccid2hctx_seqh); | ||
| 90 | |||
| 91 | BUG_ON(len != ccid2_seq_len); | ||
| 92 | ccid2_pr_debug("total len=%d\n", len); | ||
| 93 | } | ||
| 94 | #else | ||
| 95 | #define ccid2_hc_tx_check_sanity(hctx) do {} while (0) | ||
| 96 | #endif | ||
| 97 | |||
| 98 | static int ccid2_hc_tx_send_packet(struct sock *sk, | ||
| 99 | struct sk_buff *skb, int len) | ||
| 100 | { | ||
| 101 | struct ccid2_hc_tx_sock *hctx; | ||
| 102 | |||
| 103 | switch (DCCP_SKB_CB(skb)->dccpd_type) { | ||
| 104 | case 0: /* XXX data packets from userland come through like this */ | ||
| 105 | case DCCP_PKT_DATA: | ||
| 106 | case DCCP_PKT_DATAACK: | ||
| 107 | break; | ||
| 108 | /* No congestion control on other packets */ | ||
| 109 | default: | ||
| 110 | return 0; | ||
| 111 | } | ||
| 112 | |||
| 113 | hctx = ccid2_hc_tx_sk(sk); | ||
| 114 | |||
| 115 | ccid2_pr_debug("pipe=%d cwnd=%d\n", hctx->ccid2hctx_pipe, | ||
| 116 | hctx->ccid2hctx_cwnd); | ||
| 117 | |||
| 118 | if (hctx->ccid2hctx_pipe < hctx->ccid2hctx_cwnd) { | ||
| 119 | /* OK we can send... make sure previous packet was sent off */ | ||
| 120 | if (!hctx->ccid2hctx_sendwait) { | ||
| 121 | hctx->ccid2hctx_sendwait = 1; | ||
| 122 | return 0; | ||
| 123 | } | ||
| 124 | } | ||
| 125 | |||
| 126 | return 100; /* XXX */ | ||
| 127 | } | ||
| 128 | |||
| 129 | static void ccid2_change_l_ack_ratio(struct sock *sk, int val) | ||
| 130 | { | ||
| 131 | struct dccp_sock *dp = dccp_sk(sk); | ||
| 132 | /* | ||
| 133 | * XXX I don't really agree with val != 2. If cwnd is 1, ack ratio | ||
| 134 | * should be 1... it shouldn't be allowed to become 2. | ||
| 135 | * -sorbo. | ||
| 136 | */ | ||
| 137 | if (val != 2) { | ||
| 138 | const struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 139 | int max = hctx->ccid2hctx_cwnd / 2; | ||
| 140 | |||
| 141 | /* round up */ | ||
| 142 | if (hctx->ccid2hctx_cwnd & 1) | ||
| 143 | max++; | ||
| 144 | |||
| 145 | if (val > max) | ||
| 146 | val = max; | ||
| 147 | } | ||
| 148 | |||
| 149 | ccid2_pr_debug("changing local ack ratio to %d\n", val); | ||
| 150 | WARN_ON(val <= 0); | ||
| 151 | dp->dccps_l_ack_ratio = val; | ||
| 152 | } | ||
| 153 | |||
| 154 | static void ccid2_change_cwnd(struct sock *sk, int val) | ||
| 155 | { | ||
| 156 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 157 | |||
| 158 | if (val == 0) | ||
| 159 | val = 1; | ||
| 160 | |||
| 161 | /* XXX do we need to change ack ratio? */ | ||
| 162 | ccid2_pr_debug("change cwnd to %d\n", val); | ||
| 163 | |||
| 164 | BUG_ON(val < 1); | ||
| 165 | hctx->ccid2hctx_cwnd = val; | ||
| 166 | } | ||
| 167 | |||
| 168 | static void ccid2_start_rto_timer(struct sock *sk); | ||
| 169 | |||
| 170 | static void ccid2_hc_tx_rto_expire(unsigned long data) | ||
| 171 | { | ||
| 172 | struct sock *sk = (struct sock *)data; | ||
| 173 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 174 | long s; | ||
| 175 | |||
| 176 | bh_lock_sock(sk); | ||
| 177 | if (sock_owned_by_user(sk)) { | ||
| 178 | sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer, | ||
| 179 | jiffies + HZ / 5); | ||
| 180 | goto out; | ||
| 181 | } | ||
| 182 | |||
| 183 | ccid2_pr_debug("RTO_EXPIRE\n"); | ||
| 184 | |||
| 185 | ccid2_hc_tx_check_sanity(hctx); | ||
| 186 | |||
| 187 | /* back-off timer */ | ||
| 188 | hctx->ccid2hctx_rto <<= 1; | ||
| 189 | |||
| 190 | s = hctx->ccid2hctx_rto / HZ; | ||
| 191 | if (s > 60) | ||
| 192 | hctx->ccid2hctx_rto = 60 * HZ; | ||
| 193 | |||
| 194 | ccid2_start_rto_timer(sk); | ||
| 195 | |||
| 196 | /* adjust pipe, cwnd etc */ | ||
| 197 | hctx->ccid2hctx_pipe = 0; | ||
| 198 | hctx->ccid2hctx_ssthresh = hctx->ccid2hctx_cwnd >> 1; | ||
| 199 | if (hctx->ccid2hctx_ssthresh < 2) | ||
| 200 | hctx->ccid2hctx_ssthresh = 2; | ||
| 201 | ccid2_change_cwnd(sk, 1); | ||
| 202 | |||
| 203 | /* clear state about stuff we sent */ | ||
| 204 | hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqh; | ||
| 205 | hctx->ccid2hctx_ssacks = 0; | ||
| 206 | hctx->ccid2hctx_acks = 0; | ||
| 207 | hctx->ccid2hctx_sent = 0; | ||
| 208 | |||
| 209 | /* clear ack ratio state. */ | ||
| 210 | hctx->ccid2hctx_arsent = 0; | ||
| 211 | hctx->ccid2hctx_ackloss = 0; | ||
| 212 | hctx->ccid2hctx_rpseq = 0; | ||
| 213 | hctx->ccid2hctx_rpdupack = -1; | ||
| 214 | ccid2_change_l_ack_ratio(sk, 1); | ||
| 215 | ccid2_hc_tx_check_sanity(hctx); | ||
| 216 | out: | ||
| 217 | bh_unlock_sock(sk); | ||
| 218 | sock_put(sk); | ||
| 219 | } | ||
| 220 | |||
| 221 | static void ccid2_start_rto_timer(struct sock *sk) | ||
| 222 | { | ||
| 223 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 224 | |||
| 225 | ccid2_pr_debug("setting RTO timeout=%ld\n", hctx->ccid2hctx_rto); | ||
| 226 | |||
| 227 | BUG_ON(timer_pending(&hctx->ccid2hctx_rtotimer)); | ||
| 228 | sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer, | ||
| 229 | jiffies + hctx->ccid2hctx_rto); | ||
| 230 | } | ||
| 231 | |||
| 232 | static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, int len) | ||
| 233 | { | ||
| 234 | struct dccp_sock *dp = dccp_sk(sk); | ||
| 235 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 236 | u64 seq; | ||
| 237 | |||
| 238 | ccid2_hc_tx_check_sanity(hctx); | ||
| 239 | |||
| 240 | BUG_ON(!hctx->ccid2hctx_sendwait); | ||
| 241 | hctx->ccid2hctx_sendwait = 0; | ||
| 242 | hctx->ccid2hctx_pipe++; | ||
| 243 | BUG_ON(hctx->ccid2hctx_pipe < 0); | ||
| 244 | |||
| 245 | /* There is an issue. What if another packet is sent between | ||
| 246 | * packet_send() and packet_sent(). Then the sequence number would be | ||
| 247 | * wrong. | ||
| 248 | * -sorbo. | ||
| 249 | */ | ||
| 250 | seq = dp->dccps_gss; | ||
| 251 | |||
| 252 | hctx->ccid2hctx_seqh->ccid2s_seq = seq; | ||
| 253 | hctx->ccid2hctx_seqh->ccid2s_acked = 0; | ||
| 254 | hctx->ccid2hctx_seqh->ccid2s_sent = jiffies; | ||
| 255 | hctx->ccid2hctx_seqh = hctx->ccid2hctx_seqh->ccid2s_next; | ||
| 256 | |||
| 257 | ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->ccid2hctx_cwnd, | ||
| 258 | hctx->ccid2hctx_pipe); | ||
| 259 | |||
| 260 | if (hctx->ccid2hctx_seqh == hctx->ccid2hctx_seqt) { | ||
| 261 | /* XXX allocate more space */ | ||
| 262 | WARN_ON(1); | ||
| 263 | } | ||
| 264 | |||
| 265 | hctx->ccid2hctx_sent++; | ||
| 266 | |||
| 267 | /* Ack Ratio. Need to maintain a concept of how many windows we sent */ | ||
| 268 | hctx->ccid2hctx_arsent++; | ||
| 269 | /* We had an ack loss in this window... */ | ||
| 270 | if (hctx->ccid2hctx_ackloss) { | ||
| 271 | if (hctx->ccid2hctx_arsent >= hctx->ccid2hctx_cwnd) { | ||
| 272 | hctx->ccid2hctx_arsent = 0; | ||
| 273 | hctx->ccid2hctx_ackloss = 0; | ||
| 274 | } | ||
| 275 | } else { | ||
| 276 | /* No acks lost up to now... */ | ||
| 277 | /* decrease ack ratio if enough packets were sent */ | ||
| 278 | if (dp->dccps_l_ack_ratio > 1) { | ||
| 279 | /* XXX don't calculate denominator each time */ | ||
| 280 | int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio - | ||
| 281 | dp->dccps_l_ack_ratio; | ||
| 282 | |||
| 283 | denom = hctx->ccid2hctx_cwnd * hctx->ccid2hctx_cwnd / denom; | ||
| 284 | |||
| 285 | if (hctx->ccid2hctx_arsent >= denom) { | ||
| 286 | ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1); | ||
| 287 | hctx->ccid2hctx_arsent = 0; | ||
| 288 | } | ||
| 289 | } else { | ||
| 290 | /* we can't increase ack ratio further [1] */ | ||
| 291 | hctx->ccid2hctx_arsent = 0; /* or maybe set it to cwnd*/ | ||
| 292 | } | ||
| 293 | } | ||
| 294 | |||
| 295 | /* setup RTO timer */ | ||
| 296 | if (!timer_pending(&hctx->ccid2hctx_rtotimer)) | ||
| 297 | ccid2_start_rto_timer(sk); | ||
| 298 | |||
| 299 | #ifdef CCID2_DEBUG | ||
| 300 | ccid2_pr_debug("pipe=%d\n", hctx->ccid2hctx_pipe); | ||
| 301 | ccid2_pr_debug("Sent: seq=%llu\n", seq); | ||
| 302 | do { | ||
| 303 | struct ccid2_seq *seqp = hctx->ccid2hctx_seqt; | ||
| 304 | |||
| 305 | while (seqp != hctx->ccid2hctx_seqh) { | ||
| 306 | ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n", | ||
| 307 | seqp->ccid2s_seq, seqp->ccid2s_acked, | ||
| 308 | seqp->ccid2s_sent); | ||
| 309 | seqp = seqp->ccid2s_next; | ||
| 310 | } | ||
| 311 | } while (0); | ||
| 312 | ccid2_pr_debug("=========\n"); | ||
| 313 | ccid2_hc_tx_check_sanity(hctx); | ||
| 314 | #endif | ||
| 315 | } | ||
| 316 | |||
| 317 | /* XXX Lame code duplication! | ||
| 318 | * returns -1 if none was found. | ||
| 319 | * else returns the next offset to use in the function call. | ||
| 320 | */ | ||
| 321 | static int ccid2_ackvector(struct sock *sk, struct sk_buff *skb, int offset, | ||
| 322 | unsigned char **vec, unsigned char *veclen) | ||
| 323 | { | ||
| 324 | const struct dccp_hdr *dh = dccp_hdr(skb); | ||
| 325 | unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb); | ||
| 326 | unsigned char *opt_ptr; | ||
| 327 | const unsigned char *opt_end = (unsigned char *)dh + | ||
| 328 | (dh->dccph_doff * 4); | ||
| 329 | unsigned char opt, len; | ||
| 330 | unsigned char *value; | ||
| 331 | |||
| 332 | BUG_ON(offset < 0); | ||
| 333 | options += offset; | ||
| 334 | opt_ptr = options; | ||
| 335 | if (opt_ptr >= opt_end) | ||
| 336 | return -1; | ||
| 337 | |||
| 338 | while (opt_ptr != opt_end) { | ||
| 339 | opt = *opt_ptr++; | ||
| 340 | len = 0; | ||
| 341 | value = NULL; | ||
| 342 | |||
| 343 | /* Check if this isn't a single byte option */ | ||
| 344 | if (opt > DCCPO_MAX_RESERVED) { | ||
| 345 | if (opt_ptr == opt_end) | ||
| 346 | goto out_invalid_option; | ||
| 347 | |||
| 348 | len = *opt_ptr++; | ||
| 349 | if (len < 3) | ||
| 350 | goto out_invalid_option; | ||
| 351 | /* | ||
| 352 | * Remove the type and len fields, leaving | ||
| 353 | * just the value size | ||
| 354 | */ | ||
| 355 | len -= 2; | ||
| 356 | value = opt_ptr; | ||
| 357 | opt_ptr += len; | ||
| 358 | |||
| 359 | if (opt_ptr > opt_end) | ||
| 360 | goto out_invalid_option; | ||
| 361 | } | ||
| 362 | |||
| 363 | switch (opt) { | ||
| 364 | case DCCPO_ACK_VECTOR_0: | ||
| 365 | case DCCPO_ACK_VECTOR_1: | ||
| 366 | *vec = value; | ||
| 367 | *veclen = len; | ||
| 368 | return offset + (opt_ptr - options); | ||
| 369 | } | ||
| 370 | } | ||
| 371 | |||
| 372 | return -1; | ||
| 373 | |||
| 374 | out_invalid_option: | ||
| 375 | BUG_ON(1); /* should never happen... options were previously parsed ! */ | ||
| 376 | return -1; | ||
| 377 | } | ||
| 378 | |||
| 379 | static void ccid2_hc_tx_kill_rto_timer(struct sock *sk) | ||
| 380 | { | ||
| 381 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 382 | |||
| 383 | sk_stop_timer(sk, &hctx->ccid2hctx_rtotimer); | ||
| 384 | ccid2_pr_debug("deleted RTO timer\n"); | ||
| 385 | } | ||
| 386 | |||
| 387 | static inline void ccid2_new_ack(struct sock *sk, | ||
| 388 | struct ccid2_seq *seqp, | ||
| 389 | unsigned int *maxincr) | ||
| 390 | { | ||
| 391 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 392 | |||
| 393 | /* slow start */ | ||
| 394 | if (hctx->ccid2hctx_cwnd < hctx->ccid2hctx_ssthresh) { | ||
| 395 | hctx->ccid2hctx_acks = 0; | ||
| 396 | |||
| 397 | /* We can increase cwnd at most maxincr [ack_ratio/2] */ | ||
| 398 | if (*maxincr) { | ||
| 399 | /* increase every 2 acks */ | ||
| 400 | hctx->ccid2hctx_ssacks++; | ||
| 401 | if (hctx->ccid2hctx_ssacks == 2) { | ||
| 402 | ccid2_change_cwnd(sk, hctx->ccid2hctx_cwnd + 1); | ||
| 403 | hctx->ccid2hctx_ssacks = 0; | ||
| 404 | *maxincr = *maxincr - 1; | ||
| 405 | } | ||
| 406 | } else { | ||
| 407 | /* increased cwnd enough for this single ack */ | ||
| 408 | hctx->ccid2hctx_ssacks = 0; | ||
| 409 | } | ||
| 410 | } else { | ||
| 411 | hctx->ccid2hctx_ssacks = 0; | ||
| 412 | hctx->ccid2hctx_acks++; | ||
| 413 | |||
| 414 | if (hctx->ccid2hctx_acks >= hctx->ccid2hctx_cwnd) { | ||
| 415 | ccid2_change_cwnd(sk, hctx->ccid2hctx_cwnd + 1); | ||
| 416 | hctx->ccid2hctx_acks = 0; | ||
| 417 | } | ||
| 418 | } | ||
| 419 | |||
| 420 | /* update RTO */ | ||
| 421 | if (hctx->ccid2hctx_srtt == -1 || | ||
| 422 | (jiffies - hctx->ccid2hctx_lastrtt) >= hctx->ccid2hctx_srtt) { | ||
| 423 | unsigned long r = jiffies - seqp->ccid2s_sent; | ||
| 424 | int s; | ||
| 425 | |||
| 426 | /* first measurement */ | ||
| 427 | if (hctx->ccid2hctx_srtt == -1) { | ||
| 428 | ccid2_pr_debug("R: %lu Time=%lu seq=%llu\n", | ||
| 429 | r, jiffies, seqp->ccid2s_seq); | ||
| 430 | hctx->ccid2hctx_srtt = r; | ||
| 431 | hctx->ccid2hctx_rttvar = r >> 1; | ||
| 432 | } else { | ||
| 433 | /* RTTVAR */ | ||
| 434 | long tmp = hctx->ccid2hctx_srtt - r; | ||
| 435 | if (tmp < 0) | ||
| 436 | tmp *= -1; | ||
| 437 | |||
| 438 | tmp >>= 2; | ||
| 439 | hctx->ccid2hctx_rttvar *= 3; | ||
| 440 | hctx->ccid2hctx_rttvar >>= 2; | ||
| 441 | hctx->ccid2hctx_rttvar += tmp; | ||
| 442 | |||
| 443 | /* SRTT */ | ||
| 444 | hctx->ccid2hctx_srtt *= 7; | ||
| 445 | hctx->ccid2hctx_srtt >>= 3; | ||
| 446 | tmp = r >> 3; | ||
| 447 | hctx->ccid2hctx_srtt += tmp; | ||
| 448 | } | ||
| 449 | s = hctx->ccid2hctx_rttvar << 2; | ||
| 450 | /* clock granularity is 1 when based on jiffies */ | ||
| 451 | if (!s) | ||
| 452 | s = 1; | ||
| 453 | hctx->ccid2hctx_rto = hctx->ccid2hctx_srtt + s; | ||
| 454 | |||
| 455 | /* must be at least a second */ | ||
| 456 | s = hctx->ccid2hctx_rto / HZ; | ||
| 457 | /* DCCP doesn't require this [but I like it cuz my code sux] */ | ||
| 458 | #if 1 | ||
| 459 | if (s < 1) | ||
| 460 | hctx->ccid2hctx_rto = HZ; | ||
| 461 | #endif | ||
| 462 | /* max 60 seconds */ | ||
| 463 | if (s > 60) | ||
| 464 | hctx->ccid2hctx_rto = HZ * 60; | ||
| 465 | |||
| 466 | hctx->ccid2hctx_lastrtt = jiffies; | ||
| 467 | |||
| 468 | ccid2_pr_debug("srtt: %ld rttvar: %ld rto: %ld (HZ=%d) R=%lu\n", | ||
| 469 | hctx->ccid2hctx_srtt, hctx->ccid2hctx_rttvar, | ||
| 470 | hctx->ccid2hctx_rto, HZ, r); | ||
| 471 | hctx->ccid2hctx_sent = 0; | ||
| 472 | } | ||
| 473 | |||
| 474 | /* we got a new ack, so re-start RTO timer */ | ||
| 475 | ccid2_hc_tx_kill_rto_timer(sk); | ||
| 476 | ccid2_start_rto_timer(sk); | ||
| 477 | } | ||
| 478 | |||
| 479 | static void ccid2_hc_tx_dec_pipe(struct sock *sk) | ||
| 480 | { | ||
| 481 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 482 | |||
| 483 | hctx->ccid2hctx_pipe--; | ||
| 484 | BUG_ON(hctx->ccid2hctx_pipe < 0); | ||
| 485 | |||
| 486 | if (hctx->ccid2hctx_pipe == 0) | ||
| 487 | ccid2_hc_tx_kill_rto_timer(sk); | ||
| 488 | } | ||
| 489 | |||
| 490 | static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) | ||
| 491 | { | ||
| 492 | struct dccp_sock *dp = dccp_sk(sk); | ||
| 493 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 494 | u64 ackno, seqno; | ||
| 495 | struct ccid2_seq *seqp; | ||
| 496 | unsigned char *vector; | ||
| 497 | unsigned char veclen; | ||
| 498 | int offset = 0; | ||
| 499 | int done = 0; | ||
| 500 | int loss = 0; | ||
| 501 | unsigned int maxincr = 0; | ||
| 502 | |||
| 503 | ccid2_hc_tx_check_sanity(hctx); | ||
| 504 | /* check reverse path congestion */ | ||
| 505 | seqno = DCCP_SKB_CB(skb)->dccpd_seq; | ||
| 506 | |||
| 507 | /* XXX this whole "algorithm" is broken. Need to fix it to keep track | ||
| 508 | * of the seqnos of the dupacks so that rpseq and rpdupack are correct | ||
| 509 | * -sorbo. | ||
| 510 | */ | ||
| 511 | /* need to bootstrap */ | ||
| 512 | if (hctx->ccid2hctx_rpdupack == -1) { | ||
| 513 | hctx->ccid2hctx_rpdupack = 0; | ||
| 514 | hctx->ccid2hctx_rpseq = seqno; | ||
| 515 | } else { | ||
| 516 | /* check if packet is consecutive */ | ||
| 517 | if ((hctx->ccid2hctx_rpseq + 1) == seqno) | ||
| 518 | hctx->ccid2hctx_rpseq++; | ||
| 519 | /* it's a later packet */ | ||
| 520 | else if (after48(seqno, hctx->ccid2hctx_rpseq)) { | ||
| 521 | hctx->ccid2hctx_rpdupack++; | ||
| 522 | |||
| 523 | /* check if we got enough dupacks */ | ||
| 524 | if (hctx->ccid2hctx_rpdupack >= | ||
| 525 | hctx->ccid2hctx_numdupack) { | ||
| 526 | hctx->ccid2hctx_rpdupack = -1; /* XXX lame */ | ||
| 527 | hctx->ccid2hctx_rpseq = 0; | ||
| 528 | |||
| 529 | ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio << 1); | ||
| 530 | } | ||
| 531 | } | ||
| 532 | } | ||
| 533 | |||
| 534 | /* check forward path congestion */ | ||
| 535 | /* still didn't send out new data packets */ | ||
| 536 | if (hctx->ccid2hctx_seqh == hctx->ccid2hctx_seqt) | ||
| 537 | return; | ||
| 538 | |||
| 539 | switch (DCCP_SKB_CB(skb)->dccpd_type) { | ||
| 540 | case DCCP_PKT_ACK: | ||
| 541 | case DCCP_PKT_DATAACK: | ||
| 542 | break; | ||
| 543 | default: | ||
| 544 | return; | ||
| 545 | } | ||
| 546 | |||
| 547 | ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq; | ||
| 548 | seqp = hctx->ccid2hctx_seqh->ccid2s_prev; | ||
| 549 | |||
| 550 | /* If in slow-start, cwnd can increase at most Ack Ratio / 2 packets for | ||
| 551 | * this single ack. I round up. | ||
| 552 | * -sorbo. | ||
| 553 | */ | ||
| 554 | maxincr = dp->dccps_l_ack_ratio >> 1; | ||
| 555 | maxincr++; | ||
| 556 | |||
| 557 | /* go through all ack vectors */ | ||
| 558 | while ((offset = ccid2_ackvector(sk, skb, offset, | ||
| 559 | &vector, &veclen)) != -1) { | ||
| 560 | /* go through this ack vector */ | ||
| 561 | while (veclen--) { | ||
| 562 | const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK; | ||
| 563 | u64 ackno_end_rl; | ||
| 564 | |||
| 565 | dccp_set_seqno(&ackno_end_rl, ackno - rl); | ||
| 566 | ccid2_pr_debug("ackvec start:%llu end:%llu\n", ackno, | ||
| 567 | ackno_end_rl); | ||
| 568 | /* if the seqno we are analyzing is larger than the | ||
| 569 | * current ackno, then move towards the tail of our | ||
| 570 | * seqnos. | ||
| 571 | */ | ||
| 572 | while (after48(seqp->ccid2s_seq, ackno)) { | ||
| 573 | if (seqp == hctx->ccid2hctx_seqt) { | ||
| 574 | done = 1; | ||
| 575 | break; | ||
| 576 | } | ||
| 577 | seqp = seqp->ccid2s_prev; | ||
| 578 | } | ||
| 579 | if (done) | ||
| 580 | break; | ||
| 581 | |||
| 582 | /* check all seqnos in the range of the vector | ||
| 583 | * run length | ||
| 584 | */ | ||
| 585 | while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) { | ||
| 586 | const u8 state = (*vector & | ||
| 587 | DCCP_ACKVEC_STATE_MASK) >> 6; | ||
| 588 | |||
| 589 | /* new packet received or marked */ | ||
| 590 | if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED && | ||
| 591 | !seqp->ccid2s_acked) { | ||
| 592 | if (state == | ||
| 593 | DCCP_ACKVEC_STATE_ECN_MARKED) { | ||
| 594 | loss = 1; | ||
| 595 | } else | ||
| 596 | ccid2_new_ack(sk, seqp, | ||
| 597 | &maxincr); | ||
| 598 | |||
| 599 | seqp->ccid2s_acked = 1; | ||
| 600 | ccid2_pr_debug("Got ack for %llu\n", | ||
| 601 | seqp->ccid2s_seq); | ||
| 602 | ccid2_hc_tx_dec_pipe(sk); | ||
| 603 | } | ||
| 604 | if (seqp == hctx->ccid2hctx_seqt) { | ||
| 605 | done = 1; | ||
| 606 | break; | ||
| 607 | } | ||
| 608 | seqp = seqp->ccid2s_next; | ||
| 609 | } | ||
| 610 | if (done) | ||
| 611 | break; | ||
| 612 | |||
| 613 | |||
| 614 | dccp_set_seqno(&ackno, ackno_end_rl - 1); | ||
| 615 | vector++; | ||
| 616 | } | ||
| 617 | if (done) | ||
| 618 | break; | ||
| 619 | } | ||
| 620 | |||
| 621 | /* The state about what is acked should be correct now | ||
| 622 | * Check for NUMDUPACK | ||
| 623 | */ | ||
| 624 | seqp = hctx->ccid2hctx_seqh->ccid2s_prev; | ||
| 625 | done = 0; | ||
| 626 | while (1) { | ||
| 627 | if (seqp->ccid2s_acked) { | ||
| 628 | done++; | ||
| 629 | if (done == hctx->ccid2hctx_numdupack) | ||
| 630 | break; | ||
| 631 | } | ||
| 632 | if (seqp == hctx->ccid2hctx_seqt) | ||
| 633 | break; | ||
| 634 | seqp = seqp->ccid2s_prev; | ||
| 635 | } | ||
| 636 | |||
| 637 | /* If there are at least 3 acknowledgements, anything unacknowledged | ||
| 638 | * below the last sequence number is considered lost | ||
| 639 | */ | ||
| 640 | if (done == hctx->ccid2hctx_numdupack) { | ||
| 641 | struct ccid2_seq *last_acked = seqp; | ||
| 642 | |||
| 643 | /* check for lost packets */ | ||
| 644 | while (1) { | ||
| 645 | if (!seqp->ccid2s_acked) { | ||
| 646 | loss = 1; | ||
| 647 | ccid2_hc_tx_dec_pipe(sk); | ||
| 648 | } | ||
| 649 | if (seqp == hctx->ccid2hctx_seqt) | ||
| 650 | break; | ||
| 651 | seqp = seqp->ccid2s_prev; | ||
| 652 | } | ||
| 653 | |||
| 654 | hctx->ccid2hctx_seqt = last_acked; | ||
| 655 | } | ||
| 656 | |||
| 657 | /* trim acked packets in tail */ | ||
| 658 | while (hctx->ccid2hctx_seqt != hctx->ccid2hctx_seqh) { | ||
| 659 | if (!hctx->ccid2hctx_seqt->ccid2s_acked) | ||
| 660 | break; | ||
| 661 | |||
| 662 | hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqt->ccid2s_next; | ||
| 663 | } | ||
| 664 | |||
| 665 | if (loss) { | ||
| 666 | /* XXX do bit shifts guarantee a 0 as the new bit? */ | ||
| 667 | ccid2_change_cwnd(sk, hctx->ccid2hctx_cwnd >> 1); | ||
| 668 | hctx->ccid2hctx_ssthresh = hctx->ccid2hctx_cwnd; | ||
| 669 | if (hctx->ccid2hctx_ssthresh < 2) | ||
| 670 | hctx->ccid2hctx_ssthresh = 2; | ||
| 671 | } | ||
| 672 | |||
| 673 | ccid2_hc_tx_check_sanity(hctx); | ||
| 674 | } | ||
| 675 | |||
| 676 | static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk) | ||
| 677 | { | ||
| 678 | struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid); | ||
| 679 | int seqcount = ccid2_seq_len; | ||
| 680 | int i; | ||
| 681 | |||
| 682 | /* XXX init variables with proper values */ | ||
| 683 | hctx->ccid2hctx_cwnd = 1; | ||
| 684 | hctx->ccid2hctx_ssthresh = 10; | ||
| 685 | hctx->ccid2hctx_numdupack = 3; | ||
| 686 | |||
| 687 | /* XXX init ~ to window size... */ | ||
| 688 | hctx->ccid2hctx_seqbuf = kmalloc(sizeof(*hctx->ccid2hctx_seqbuf) * | ||
| 689 | seqcount, gfp_any()); | ||
| 690 | if (hctx->ccid2hctx_seqbuf == NULL) | ||
| 691 | return -ENOMEM; | ||
| 692 | |||
| 693 | for (i = 0; i < (seqcount - 1); i++) { | ||
| 694 | hctx->ccid2hctx_seqbuf[i].ccid2s_next = | ||
| 695 | &hctx->ccid2hctx_seqbuf[i + 1]; | ||
| 696 | hctx->ccid2hctx_seqbuf[i + 1].ccid2s_prev = | ||
| 697 | &hctx->ccid2hctx_seqbuf[i]; | ||
| 698 | } | ||
| 699 | hctx->ccid2hctx_seqbuf[seqcount - 1].ccid2s_next = | ||
| 700 | hctx->ccid2hctx_seqbuf; | ||
| 701 | hctx->ccid2hctx_seqbuf->ccid2s_prev = | ||
| 702 | &hctx->ccid2hctx_seqbuf[seqcount - 1]; | ||
| 703 | |||
| 704 | hctx->ccid2hctx_seqh = hctx->ccid2hctx_seqbuf; | ||
| 705 | hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqh; | ||
| 706 | hctx->ccid2hctx_sent = 0; | ||
| 707 | hctx->ccid2hctx_rto = 3 * HZ; | ||
| 708 | hctx->ccid2hctx_srtt = -1; | ||
| 709 | hctx->ccid2hctx_rttvar = -1; | ||
| 710 | hctx->ccid2hctx_lastrtt = 0; | ||
| 711 | hctx->ccid2hctx_rpdupack = -1; | ||
| 712 | |||
| 713 | hctx->ccid2hctx_rtotimer.function = &ccid2_hc_tx_rto_expire; | ||
| 714 | hctx->ccid2hctx_rtotimer.data = (unsigned long)sk; | ||
| 715 | init_timer(&hctx->ccid2hctx_rtotimer); | ||
| 716 | |||
| 717 | ccid2_hc_tx_check_sanity(hctx); | ||
| 718 | return 0; | ||
| 719 | } | ||
| 720 | |||
| 721 | static void ccid2_hc_tx_exit(struct sock *sk) | ||
| 722 | { | ||
| 723 | struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk); | ||
| 724 | |||
| 725 | ccid2_hc_tx_kill_rto_timer(sk); | ||
| 726 | kfree(hctx->ccid2hctx_seqbuf); | ||
| 727 | hctx->ccid2hctx_seqbuf = NULL; | ||
| 728 | } | ||
| 729 | |||
| 730 | static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) | ||
| 731 | { | ||
| 732 | const struct dccp_sock *dp = dccp_sk(sk); | ||
| 733 | struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk); | ||
| 734 | |||
| 735 | switch (DCCP_SKB_CB(skb)->dccpd_type) { | ||
| 736 | case DCCP_PKT_DATA: | ||
| 737 | case DCCP_PKT_DATAACK: | ||
| 738 | hcrx->ccid2hcrx_data++; | ||
| 739 | if (hcrx->ccid2hcrx_data >= dp->dccps_r_ack_ratio) { | ||
| 740 | dccp_send_ack(sk); | ||
| 741 | hcrx->ccid2hcrx_data = 0; | ||
| 742 | } | ||
| 743 | break; | ||
| 744 | } | ||
| 745 | } | ||
| 746 | |||
| 747 | static struct ccid_operations ccid2 = { | ||
| 748 | .ccid_id = 2, | ||
| 749 | .ccid_name = "ccid2", | ||
| 750 | .ccid_owner = THIS_MODULE, | ||
| 751 | .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock), | ||
| 752 | .ccid_hc_tx_init = ccid2_hc_tx_init, | ||
| 753 | .ccid_hc_tx_exit = ccid2_hc_tx_exit, | ||
| 754 | .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet, | ||
| 755 | .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent, | ||
| 756 | .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv, | ||
| 757 | .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock), | ||
| 758 | .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv, | ||
| 759 | }; | ||
| 760 | |||
| 761 | module_param(ccid2_debug, int, 0444); | ||
| 762 | MODULE_PARM_DESC(ccid2_debug, "Enable debug messages"); | ||
| 763 | |||
| 764 | static __init int ccid2_module_init(void) | ||
| 765 | { | ||
| 766 | return ccid_register(&ccid2); | ||
| 767 | } | ||
| 768 | module_init(ccid2_module_init); | ||
| 769 | |||
| 770 | static __exit void ccid2_module_exit(void) | ||
| 771 | { | ||
| 772 | ccid_unregister(&ccid2); | ||
| 773 | } | ||
| 774 | module_exit(ccid2_module_exit); | ||
| 775 | |||
| 776 | MODULE_AUTHOR("Andrea Bittau <a.bittau@cs.ucl.ac.uk>"); | ||
| 777 | MODULE_DESCRIPTION("DCCP TCP-Like (CCID2) CCID"); | ||
| 778 | MODULE_LICENSE("GPL"); | ||
| 779 | MODULE_ALIAS("net-dccp-ccid-2"); | ||
