git.oblomov.eu Git - linux-2.6/blob - net/dccp/output.c

   1 /*
   2  *  net/dccp/output.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/kernel.h>
  16 #include <linux/skbuff.h>
  17
  18 #include <net/inet_sock.h>
  19 #include <net/sock.h>
  20
  21 #include "ackvec.h"
  22 #include "ccid.h"
  23 #include "dccp.h"
  24
  25 static inline void dccp_event_ack_sent(struct sock *sk)
  26 {
  27         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  28 }
  29
  30 static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  31 {
  32         skb_set_owner_w(skb, sk);
  33         WARN_ON(sk->sk_send_head);
  34         sk->sk_send_head = skb;
  35 }
  36
  37 /*
  38  * All SKB's seen here are completely headerless. It is our
  39  * job to build the DCCP header, and pass the packet down to
  40  * IP so it can do the same plus pass the packet off to the
  41  * device.
  42  */
  43 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  44 {
  45         if (likely(skb != NULL)) {
  46                 const struct inet_sock *inet = inet_sk(sk);
  47                 const struct inet_connection_sock *icsk = inet_csk(sk);
  48                 struct dccp_sock *dp = dccp_sk(sk);
  49                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  50                 struct dccp_hdr *dh;
  51                 /* XXX For now we're using only 48 bits sequence numbers */
  52                 const int dccp_header_size = sizeof(*dh) +
  53                                              sizeof(struct dccp_hdr_ext) +
  54                                           dccp_packet_hdr_len(dcb->dccpd_type);
  55                 int err, set_ack = 1;
  56                 u64 ackno = dp->dccps_gsr;
  57
  58                 dccp_inc_seqno(&dp->dccps_gss);
  59
  60                 switch (dcb->dccpd_type) {
  61                 case DCCP_PKT_DATA:
  62                         set_ack = 0;
  63                         /* fall through */
  64                 case DCCP_PKT_DATAACK:
  65                         break;
  66
  67                 case DCCP_PKT_SYNC:
  68                 case DCCP_PKT_SYNCACK:
  69                         ackno = dcb->dccpd_seq;
  70                         /* fall through */
  71                 default:
  72                         /*
  73                          * Only data packets should come through with skb->sk
  74                          * set.
  75                          */
  76                         WARN_ON(skb->sk);
  77                         skb_set_owner_w(skb, sk);
  78                         break;
  79                 }
  80
  81                 dcb->dccpd_seq = dp->dccps_gss;
  82                 dccp_insert_options(sk, skb);
  83
  84                 skb->h.raw = skb_push(skb, dccp_header_size);
  85                 dh = dccp_hdr(skb);
  86
  87                 /* Build DCCP header and checksum it. */
  88                 memset(dh, 0, dccp_header_size);
  89                 dh->dccph_type  = dcb->dccpd_type;
  90                 dh->dccph_sport = inet->sport;
  91                 dh->dccph_dport = inet->dport;
  92                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  93                 dh->dccph_ccval = dcb->dccpd_ccval;
  94                 /* XXX For now we're using only 48 bits sequence numbers */
  95                 dh->dccph_x     = 1;
  96
  97                 dp->dccps_awh = dp->dccps_gss;
  98                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  99                 if (set_ack)
 100                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 101
 102                 switch (dcb->dccpd_type) {
 103                 case DCCP_PKT_REQUEST:
 104                         dccp_hdr_request(skb)->dccph_req_service =
 105                                                         dp->dccps_service;
 106                         break;
 107                 case DCCP_PKT_RESET:
 108                         dccp_hdr_reset(skb)->dccph_reset_code =
 109                                                         dcb->dccpd_reset_code;
 110                         break;
 111                 }
 112
 113                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 114
 115                 if (set_ack)
 116                         dccp_event_ack_sent(sk);
 117
 118                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 119
 120                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 121                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 122                 if (err <= 0)
 123                         return err;
 124
 125                 /* NET_XMIT_CN is special. It does not guarantee,
 126                  * that this packet is lost. It tells that device
 127                  * is about to start to drop packets or already
 128                  * drops some packets of the same priority and
 129                  * invokes us to send less aggressively.
 130                  */
 131                 return err == NET_XMIT_CN ? 0 : err;
 132         }
 133         return -ENOBUFS;
 134 }
 135
 136 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 137 {
 138         struct inet_connection_sock *icsk = inet_csk(sk);
 139         struct dccp_sock *dp = dccp_sk(sk);
 140         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
 141                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 142
 143         /* Now subtract optional transport overhead */
 144         mss_now -= icsk->icsk_ext_hdr_len;
 145
 146         /*
 147          * FIXME: this should come from the CCID infrastructure, where, say,
 148          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 149          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 150          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 151          * make it a multiple of 4
 152          */
 153
 154         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 155
 156         /* And store cached results */
 157         icsk->icsk_pmtu_cookie = pmtu;
 158         dp->dccps_mss_cache = mss_now;
 159
 160         return mss_now;
 161 }
 162
 163 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 164
 165 void dccp_write_space(struct sock *sk)
 166 {
 167         read_lock(&sk->sk_callback_lock);
 168
 169         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 170                 wake_up_interruptible(sk->sk_sleep);
 171         /* Should agree with poll, otherwise some programs break */
 172         if (sock_writeable(sk))
 173                 sk_wake_async(sk, 2, POLL_OUT);
 174
 175         read_unlock(&sk->sk_callback_lock);
 176 }
 177
 178 /**
 179  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 180  * @sk: socket to wait for
 181  * @timeo: for how long
 182  */
 183 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 184                               long *timeo)
 185 {
 186         struct dccp_sock *dp = dccp_sk(sk);
 187         DEFINE_WAIT(wait);
 188         long delay;
 189         int rc;
 190
 191         while (1) {
 192                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 193
 194                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 195                         goto do_error;
 196                 if (!*timeo)
 197                         goto do_nonblock;
 198                 if (signal_pending(current))
 199                         goto do_interrupted;
 200
 201                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 202                                             skb->len);
 203                 if (rc <= 0)
 204                         break;
 205                 delay = msecs_to_jiffies(rc);
 206                 if (delay > *timeo || delay < 0)
 207                         goto do_nonblock;
 208
 209                 sk->sk_write_pending++;
 210                 release_sock(sk);
 211                 *timeo -= schedule_timeout(delay);
 212                 lock_sock(sk);
 213                 sk->sk_write_pending--;
 214         }
 215 out:
 216         finish_wait(sk->sk_sleep, &wait);
 217         return rc;
 218
 219 do_error:
 220         rc = -EPIPE;
 221         goto out;
 222 do_nonblock:
 223         rc = -EAGAIN;
 224         goto out;
 225 do_interrupted:
 226         rc = sock_intr_errno(*timeo);
 227         goto out;
 228 }
 229
 230 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 231 {
 232         const struct dccp_sock *dp = dccp_sk(sk);
 233         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 234                                          skb->len);
 235
 236         if (err > 0)
 237                 err = dccp_wait_for_ccid(sk, skb, timeo);
 238
 239         if (err == 0) {
 240                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 241                 const int len = skb->len;
 242
 243                 if (sk->sk_state == DCCP_PARTOPEN) {
 244                         /* See 8.1.5.  Handshake Completion */
 245                         inet_csk_schedule_ack(sk);
 246                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 247                                                   inet_csk(sk)->icsk_rto,
 248                                                   DCCP_RTO_MAX);
 249                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 250                 } else if (dccp_ack_pending(sk))
 251                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 252                 else
 253                         dcb->dccpd_type = DCCP_PKT_DATA;
 254
 255                 err = dccp_transmit_skb(sk, skb);
 256                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 257         } else
 258                 kfree_skb(skb);
 259
 260         return err;
 261 }
 262
 263 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 264 {
 265         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 266                 return -EHOSTUNREACH; /* Routing failure or similar. */
 267
 268         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 269                                       pskb_copy(skb, GFP_ATOMIC):
 270                                       skb_clone(skb, GFP_ATOMIC)));
 271 }
 272
 273 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 274                                    struct request_sock *req)
 275 {
 276         struct dccp_hdr *dh;
 277         struct dccp_request_sock *dreq;
 278         const int dccp_header_size = sizeof(struct dccp_hdr) +
 279                                      sizeof(struct dccp_hdr_ext) +
 280                                      sizeof(struct dccp_hdr_response);
 281         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 282                                                dccp_header_size, 1,
 283                                            GFP_ATOMIC);
 284         if (skb == NULL)
 285                 return NULL;
 286
 287         /* Reserve space for headers. */
 288         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 289
 290         skb->dst = dst_clone(dst);
 291         skb->csum = 0;
 292
 293         dreq = dccp_rsk(req);
 294         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 295         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 296         dccp_insert_options(sk, skb);
 297
 298         skb->h.raw = skb_push(skb, dccp_header_size);
 299
 300         dh = dccp_hdr(skb);
 301         memset(dh, 0, dccp_header_size);
 302
 303         dh->dccph_sport = inet_sk(sk)->sport;
 304         dh->dccph_dport = inet_rsk(req)->rmt_port;
 305         dh->dccph_doff  = (dccp_header_size +
 306                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 307         dh->dccph_type  = DCCP_PKT_RESPONSE;
 308         dh->dccph_x     = 1;
 309         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 310         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 311         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 312
 313         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 314                                               inet_rsk(req)->rmt_addr);
 315
 316         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 317         return skb;
 318 }
 319
 320 EXPORT_SYMBOL_GPL(dccp_make_response);
 321
 322 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 323                                 const enum dccp_reset_codes code)
 324
 325 {
 326         struct dccp_hdr *dh;
 327         struct dccp_sock *dp = dccp_sk(sk);
 328         const int dccp_header_size = sizeof(struct dccp_hdr) +
 329                                      sizeof(struct dccp_hdr_ext) +
 330                                      sizeof(struct dccp_hdr_reset);
 331         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 332                                                dccp_header_size, 1,
 333                                            GFP_ATOMIC);
 334         if (skb == NULL)
 335                 return NULL;
 336
 337         /* Reserve space for headers. */
 338         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 339
 340         skb->dst = dst_clone(dst);
 341         skb->csum = 0;
 342
 343         dccp_inc_seqno(&dp->dccps_gss);
 344
 345         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 346         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 347         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 348         dccp_insert_options(sk, skb);
 349
 350         skb->h.raw = skb_push(skb, dccp_header_size);
 351
 352         dh = dccp_hdr(skb);
 353         memset(dh, 0, dccp_header_size);
 354
 355         dh->dccph_sport = inet_sk(sk)->sport;
 356         dh->dccph_dport = inet_sk(sk)->dport;
 357         dh->dccph_doff  = (dccp_header_size +
 358                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 359         dh->dccph_type  = DCCP_PKT_RESET;
 360         dh->dccph_x     = 1;
 361         dccp_hdr_set_seq(dh, dp->dccps_gss);
 362         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 363
 364         dccp_hdr_reset(skb)->dccph_reset_code = code;
 365
 366         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 367                                               inet_sk(sk)->daddr);
 368
 369         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 370         return skb;
 371 }
 372
 373 /*
 374  * Do all connect socket setups that can be done AF independent.
 375  */
 376 static inline void dccp_connect_init(struct sock *sk)
 377 {
 378         struct dccp_sock *dp = dccp_sk(sk);
 379         struct dst_entry *dst = __sk_dst_get(sk);
 380         struct inet_connection_sock *icsk = inet_csk(sk);
 381
 382         sk->sk_err = 0;
 383         sock_reset_flag(sk, SOCK_DONE);
 384
 385         dccp_sync_mss(sk, dst_mtu(dst));
 386
 387         dccp_update_gss(sk, dp->dccps_iss);
 388         /*
 389          * SWL and AWL are initially adjusted so that they are not less than
 390          * the initial Sequence Numbers received and sent, respectively:
 391          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 392          *      AWL := max(GSS - W' + 1, ISS).
 393          * These adjustments MUST be applied only at the beginning of the
 394          * connection.
 395          */
 396         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 397
 398         icsk->icsk_retransmits = 0;
 399 }
 400
 401 int dccp_connect(struct sock *sk)
 402 {
 403         struct sk_buff *skb;
 404         struct inet_connection_sock *icsk = inet_csk(sk);
 405
 406         dccp_connect_init(sk);
 407
 408         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 409         if (unlikely(skb == NULL))
 410                 return -ENOBUFS;
 411
 412         /* Reserve space for headers. */
 413         skb_reserve(skb, MAX_DCCP_HEADER);
 414
 415         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 416         skb->csum = 0;
 417
 418         dccp_skb_entail(sk, skb);
 419         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 420         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 421
 422         /* Timer for repeating the REQUEST until an answer. */
 423         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 424                                   icsk->icsk_rto, DCCP_RTO_MAX);
 425         return 0;
 426 }
 427
 428 EXPORT_SYMBOL_GPL(dccp_connect);
 429
 430 void dccp_send_ack(struct sock *sk)
 431 {
 432         /* If we have been reset, we may not send again. */
 433         if (sk->sk_state != DCCP_CLOSED) {
 434                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 435
 436                 if (skb == NULL) {
 437                         inet_csk_schedule_ack(sk);
 438                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 439                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 440                                                   TCP_DELACK_MAX,
 441                                                   DCCP_RTO_MAX);
 442                         return;
 443                 }
 444
 445                 /* Reserve space for headers */
 446                 skb_reserve(skb, MAX_DCCP_HEADER);
 447                 skb->csum = 0;
 448                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 449                 dccp_transmit_skb(sk, skb);
 450         }
 451 }
 452
 453 EXPORT_SYMBOL_GPL(dccp_send_ack);
 454
 455 void dccp_send_delayed_ack(struct sock *sk)
 456 {
 457         struct inet_connection_sock *icsk = inet_csk(sk);
 458         /*
 459          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 460          * with using 2s, and active senders also piggyback the ACK into a
 461          * DATAACK packet, so this is really for quiescent senders.
 462          */
 463         unsigned long timeout = jiffies + 2 * HZ;
 464
 465         /* Use new timeout only if there wasn't a older one earlier. */
 466         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 467                 /* If delack timer was blocked or is about to expire,
 468                  * send ACK now.
 469                  *
 470                  * FIXME: check the "about to expire" part
 471                  */
 472                 if (icsk->icsk_ack.blocked) {
 473                         dccp_send_ack(sk);
 474                         return;
 475                 }
 476
 477                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 478                         timeout = icsk->icsk_ack.timeout;
 479         }
 480         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 481         icsk->icsk_ack.timeout = timeout;
 482         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 483 }
 484
 485 void dccp_send_sync(struct sock *sk, const u64 seq,
 486                     const enum dccp_pkt_type pkt_type)
 487 {
 488         /*
 489          * We are not putting this on the write queue, so
 490          * dccp_transmit_skb() will set the ownership to this
 491          * sock.
 492          */
 493         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 494
 495         if (skb == NULL)
 496                 /* FIXME: how to make sure the sync is sent? */
 497                 return;
 498
 499         /* Reserve space for headers and prepare control bits. */
 500         skb_reserve(skb, MAX_DCCP_HEADER);
 501         skb->csum = 0;
 502         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 503         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 504
 505         dccp_transmit_skb(sk, skb);
 506 }
 507
 508 /*
 509  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 510  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 511  * any circumstances.
 512  */
 513 void dccp_send_close(struct sock *sk, const int active)
 514 {
 515         struct dccp_sock *dp = dccp_sk(sk);
 516         struct sk_buff *skb;
 517         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 518
 519         skb = alloc_skb(sk->sk_prot->max_header, prio);
 520         if (skb == NULL)
 521                 return;
 522
 523         /* Reserve space for headers and prepare control bits. */
 524         skb_reserve(skb, sk->sk_prot->max_header);
 525         skb->csum = 0;
 526         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 527                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 528
 529         if (active) {
 530                 dccp_skb_entail(sk, skb);
 531                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 532         } else
 533                 dccp_transmit_skb(sk, skb);
 534 }