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/skbuff.h>
  16
  17 #include <net/sock.h>
  18
  19 #include "ccid.h"
  20 #include "dccp.h"
  21
  22 static inline void dccp_event_ack_sent(struct sock *sk)
  23 {
  24         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  25 }
  26
  27 /*
  28  * All SKB's seen here are completely headerless. It is our
  29  * job to build the DCCP header, and pass the packet down to
  30  * IP so it can do the same plus pass the packet off to the
  31  * device.
  32  */
  33 int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  34 {
  35         if (likely(skb != NULL)) {
  36                 const struct inet_sock *inet = inet_sk(sk);
  37                 struct dccp_sock *dp = dccp_sk(sk);
  38                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  39                 struct dccp_hdr *dh;
  40                 /* XXX For now we're using only 48 bits sequence numbers */
  41                 const int dccp_header_size = sizeof(*dh) +
  42                                              sizeof(struct dccp_hdr_ext) +
  43                                           dccp_packet_hdr_len(dcb->dccpd_type);
  44                 int err, set_ack = 1;
  45                 u64 ackno = dp->dccps_gsr;
  46
  47                 dccp_inc_seqno(&dp->dccps_gss);
  48
  49                 switch (dcb->dccpd_type) {
  50                 case DCCP_PKT_DATA:
  51                         set_ack = 0;
  52                         break;
  53                 case DCCP_PKT_SYNC:
  54                 case DCCP_PKT_SYNCACK:
  55                         ackno = dcb->dccpd_seq;
  56                         break;
  57                 }
  58
  59                 dcb->dccpd_seq = dp->dccps_gss;
  60                 dccp_insert_options(sk, skb);
  61
  62                 skb->h.raw = skb_push(skb, dccp_header_size);
  63                 dh = dccp_hdr(skb);
  64                 /*
  65                  * Data packets are not cloned as they are never retransmitted
  66                  */
  67                 if (skb_cloned(skb))
  68                         skb_set_owner_w(skb, sk);
  69
  70                 /* Build DCCP header and checksum it. */
  71                 memset(dh, 0, dccp_header_size);
  72                 dh->dccph_type  = dcb->dccpd_type;
  73                 dh->dccph_sport = inet->sport;
  74                 dh->dccph_dport = inet->dport;
  75                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  76                 dh->dccph_ccval = dcb->dccpd_ccval;
  77                 /* XXX For now we're using only 48 bits sequence numbers */
  78                 dh->dccph_x     = 1;
  79
  80                 dp->dccps_awh = dp->dccps_gss;
  81                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  82                 if (set_ack)
  83                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  84
  85                 switch (dcb->dccpd_type) {
  86                 case DCCP_PKT_REQUEST:
  87                         dccp_hdr_request(skb)->dccph_req_service =
  88                                                         dcb->dccpd_service;
  89                         break;
  90                 case DCCP_PKT_RESET:
  91                         dccp_hdr_reset(skb)->dccph_reset_code =
  92                                                         dcb->dccpd_reset_code;
  93                         break;
  94                 }
  95
  96                 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
  97                                                       inet->daddr);
  98
  99                 if (set_ack)
 100                         dccp_event_ack_sent(sk);
 101
 102                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 103
 104                 err = ip_queue_xmit(skb, 0);
 105                 if (err <= 0)
 106                         return err;
 107
 108                 /* NET_XMIT_CN is special. It does not guarantee,
 109                  * that this packet is lost. It tells that device
 110                  * is about to start to drop packets or already
 111                  * drops some packets of the same priority and
 112                  * invokes us to send less aggressively.
 113                  */
 114                 return err == NET_XMIT_CN ? 0 : err;
 115         }
 116         return -ENOBUFS;
 117 }
 118
 119 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 120 {
 121         struct dccp_sock *dp = dccp_sk(sk);
 122         int mss_now;
 123
 124         /*
 125          * FIXME: we really should be using the af_specific thing to support
 126          *        IPv6.
 127          * mss_now = pmtu - tp->af_specific->net_header_len -
 128          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 129          */
 130         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
 131                   sizeof(struct dccp_hdr_ext);
 132
 133         /* Now subtract optional transport overhead */
 134         mss_now -= dp->dccps_ext_header_len;
 135
 136         /*
 137          * FIXME: this should come from the CCID infrastructure, where, say,
 138          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 139          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 140          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 141          * make it a multiple of 4
 142          */
 143
 144         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 145
 146         /* And store cached results */
 147         dp->dccps_pmtu_cookie = pmtu;
 148         dp->dccps_mss_cache = mss_now;
 149
 150         return mss_now;
 151 }
 152
 153 /**
 154  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 155  * @sk: socket to wait for
 156  * @timeo: for how long
 157  */
 158 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 159                               long *timeo)
 160 {
 161         struct dccp_sock *dp = dccp_sk(sk);
 162         DEFINE_WAIT(wait);
 163         long delay;
 164         int rc;
 165
 166         while (1) {
 167                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 168
 169                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 170                         goto do_error;
 171                 if (!*timeo)
 172                         goto do_nonblock;
 173                 if (signal_pending(current))
 174                         goto do_interrupted;
 175
 176                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 177                                             skb->len);
 178                 if (rc <= 0)
 179                         break;
 180                 delay = msecs_to_jiffies(rc);
 181                 if (delay > *timeo || delay < 0)
 182                         goto do_nonblock;
 183
 184                 sk->sk_write_pending++;
 185                 release_sock(sk);
 186                 *timeo -= schedule_timeout(delay);
 187                 lock_sock(sk);
 188                 sk->sk_write_pending--;
 189         }
 190 out:
 191         finish_wait(sk->sk_sleep, &wait);
 192         return rc;
 193
 194 do_error:
 195         rc = -EPIPE;
 196         goto out;
 197 do_nonblock:
 198         rc = -EAGAIN;
 199         goto out;
 200 do_interrupted:
 201         rc = sock_intr_errno(*timeo);
 202         goto out;
 203 }
 204
 205 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 206 {
 207         const struct dccp_sock *dp = dccp_sk(sk);
 208         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 209                                          skb->len);
 210
 211         if (err > 0)
 212                 err = dccp_wait_for_ccid(sk, skb, timeo);
 213
 214         if (err == 0) {
 215                 const struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts;
 216                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 217                 const int len = skb->len;
 218
 219                 if (sk->sk_state == DCCP_PARTOPEN) {
 220                         /* See 8.1.5.  Handshake Completion */
 221                         inet_csk_schedule_ack(sk);
 222                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 223                                                   inet_csk(sk)->icsk_rto,
 224                                                   DCCP_RTO_MAX);
 225                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 226                         /*
 227                          * FIXME: we really should have a
 228                          * dccps_ack_pending or use icsk.
 229                          */
 230                 } else if (inet_csk_ack_scheduled(sk) ||
 231                            dp->dccps_timestamp_echo != 0 ||
 232                            (dp->dccps_options.dccpo_send_ack_vector &&
 233                             ap->dccpap_buf_ackno != DCCP_MAX_SEQNO + 1 &&
 234                             ap->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1))
 235                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 236                 else
 237                         dcb->dccpd_type = DCCP_PKT_DATA;
 238
 239                 err = dccp_transmit_skb(sk, skb);
 240                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 241         }
 242
 243         return err;
 244 }
 245
 246 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 247 {
 248         if (inet_sk_rebuild_header(sk) != 0)
 249                 return -EHOSTUNREACH; /* Routing failure or similar. */
 250
 251         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 252                                       pskb_copy(skb, GFP_ATOMIC):
 253                                       skb_clone(skb, GFP_ATOMIC)));
 254 }
 255
 256 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 257                                    struct request_sock *req)
 258 {
 259         struct dccp_hdr *dh;
 260         const int dccp_header_size = sizeof(struct dccp_hdr) +
 261                                      sizeof(struct dccp_hdr_ext) +
 262                                      sizeof(struct dccp_hdr_response);
 263         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 264                                                dccp_header_size, 1,
 265                                            GFP_ATOMIC);
 266         if (skb == NULL)
 267                 return NULL;
 268
 269         /* Reserve space for headers. */
 270         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 271
 272         skb->dst = dst_clone(dst);
 273         skb->csum = 0;
 274
 275         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 276         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_rsk(req)->dreq_iss;
 277         dccp_insert_options(sk, skb);
 278
 279         skb->h.raw = skb_push(skb, dccp_header_size);
 280
 281         dh = dccp_hdr(skb);
 282         memset(dh, 0, dccp_header_size);
 283
 284         dh->dccph_sport = inet_sk(sk)->sport;
 285         dh->dccph_dport = inet_rsk(req)->rmt_port;
 286         dh->dccph_doff  = (dccp_header_size +
 287                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 288         dh->dccph_type  = DCCP_PKT_RESPONSE;
 289         dh->dccph_x     = 1;
 290         dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss);
 291         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr);
 292
 293         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 294                                               inet_rsk(req)->rmt_addr);
 295
 296         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 297         return skb;
 298 }
 299
 300 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 301                                 const enum dccp_reset_codes code)
 302
 303 {
 304         struct dccp_hdr *dh;
 305         struct dccp_sock *dp = dccp_sk(sk);
 306         const int dccp_header_size = sizeof(struct dccp_hdr) +
 307                                      sizeof(struct dccp_hdr_ext) +
 308                                      sizeof(struct dccp_hdr_reset);
 309         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 310                                                dccp_header_size, 1,
 311                                            GFP_ATOMIC);
 312         if (skb == NULL)
 313                 return NULL;
 314
 315         /* Reserve space for headers. */
 316         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 317
 318         skb->dst = dst_clone(dst);
 319         skb->csum = 0;
 320
 321         dccp_inc_seqno(&dp->dccps_gss);
 322
 323         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 324         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 325         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 326         dccp_insert_options(sk, skb);
 327
 328         skb->h.raw = skb_push(skb, dccp_header_size);
 329
 330         dh = dccp_hdr(skb);
 331         memset(dh, 0, dccp_header_size);
 332
 333         dh->dccph_sport = inet_sk(sk)->sport;
 334         dh->dccph_dport = inet_sk(sk)->dport;
 335         dh->dccph_doff  = (dccp_header_size +
 336                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 337         dh->dccph_type  = DCCP_PKT_RESET;
 338         dh->dccph_x     = 1;
 339         dccp_hdr_set_seq(dh, dp->dccps_gss);
 340         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 341
 342         dccp_hdr_reset(skb)->dccph_reset_code = code;
 343
 344         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 345                                               inet_sk(sk)->daddr);
 346
 347         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 348         return skb;
 349 }
 350
 351 /*
 352  * Do all connect socket setups that can be done AF independent.
 353  */
 354 static inline void dccp_connect_init(struct sock *sk)
 355 {
 356         struct dst_entry *dst = __sk_dst_get(sk);
 357         struct inet_connection_sock *icsk = inet_csk(sk);
 358
 359         sk->sk_err = 0;
 360         sock_reset_flag(sk, SOCK_DONE);
 361
 362         dccp_sync_mss(sk, dst_mtu(dst));
 363
 364         /*
 365          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 366          * something like dccp_inc_seq
 367          */
 368
 369         icsk->icsk_retransmits = 0;
 370 }
 371
 372 int dccp_connect(struct sock *sk)
 373 {
 374         struct sk_buff *skb;
 375         struct inet_connection_sock *icsk = inet_csk(sk);
 376
 377         dccp_connect_init(sk);
 378
 379         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 380         if (unlikely(skb == NULL))
 381                 return -ENOBUFS;
 382
 383         /* Reserve space for headers. */
 384         skb_reserve(skb, MAX_DCCP_HEADER);
 385
 386         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 387         /* FIXME: set service to something meaningful, coming
 388          * from userspace*/
 389         DCCP_SKB_CB(skb)->dccpd_service = 0;
 390         skb->csum = 0;
 391         skb_set_owner_w(skb, sk);
 392
 393         BUG_TRAP(sk->sk_send_head == NULL);
 394         sk->sk_send_head = skb;
 395         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 396         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 397
 398         /* Timer for repeating the REQUEST until an answer. */
 399         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 400                                   icsk->icsk_rto, DCCP_RTO_MAX);
 401         return 0;
 402 }
 403
 404 void dccp_send_ack(struct sock *sk)
 405 {
 406         /* If we have been reset, we may not send again. */
 407         if (sk->sk_state != DCCP_CLOSED) {
 408                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 409
 410                 if (skb == NULL) {
 411                         inet_csk_schedule_ack(sk);
 412                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 413                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 414                                                   TCP_DELACK_MAX,
 415                                                   DCCP_RTO_MAX);
 416                         return;
 417                 }
 418
 419                 /* Reserve space for headers */
 420                 skb_reserve(skb, MAX_DCCP_HEADER);
 421                 skb->csum = 0;
 422                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 423                 skb_set_owner_w(skb, sk);
 424                 dccp_transmit_skb(sk, skb);
 425         }
 426 }
 427
 428 EXPORT_SYMBOL_GPL(dccp_send_ack);
 429
 430 void dccp_send_delayed_ack(struct sock *sk)
 431 {
 432         struct inet_connection_sock *icsk = inet_csk(sk);
 433         /*
 434          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 435          * with using 2s, and active senders also piggyback the ACK into a
 436          * DATAACK packet, so this is really for quiescent senders.
 437          */
 438         unsigned long timeout = jiffies + 2 * HZ;
 439
 440         /* Use new timeout only if there wasn't a older one earlier. */
 441         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 442                 /* If delack timer was blocked or is about to expire,
 443                  * send ACK now.
 444                  *
 445                  * FIXME: check the "about to expire" part
 446                  */
 447                 if (icsk->icsk_ack.blocked) {
 448                         dccp_send_ack(sk);
 449                         return;
 450                 }
 451
 452                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 453                         timeout = icsk->icsk_ack.timeout;
 454         }
 455         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 456         icsk->icsk_ack.timeout = timeout;
 457         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 458 }
 459
 460 void dccp_send_sync(struct sock *sk, const u64 seq,
 461                     const enum dccp_pkt_type pkt_type)
 462 {
 463         /*
 464          * We are not putting this on the write queue, so
 465          * dccp_transmit_skb() will set the ownership to this
 466          * sock.
 467          */
 468         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 469
 470         if (skb == NULL)
 471                 /* FIXME: how to make sure the sync is sent? */
 472                 return;
 473
 474         /* Reserve space for headers and prepare control bits. */
 475         skb_reserve(skb, MAX_DCCP_HEADER);
 476         skb->csum = 0;
 477         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 478         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 479
 480         skb_set_owner_w(skb, sk);
 481         dccp_transmit_skb(sk, skb);
 482 }
 483
 484 /*
 485  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 486  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 487  * any circumstances.
 488  */
 489 void dccp_send_close(struct sock *sk, const int active)
 490 {
 491         struct dccp_sock *dp = dccp_sk(sk);
 492         struct sk_buff *skb;
 493         const unsigned int prio = active ? GFP_KERNEL : GFP_ATOMIC;
 494
 495         skb = alloc_skb(sk->sk_prot->max_header, prio);
 496         if (skb == NULL)
 497                 return;
 498
 499         /* Reserve space for headers and prepare control bits. */
 500         skb_reserve(skb, sk->sk_prot->max_header);
 501         skb->csum = 0;
 502         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 503                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 504
 505         skb_set_owner_w(skb, sk);
 506         if (active) {
 507                 BUG_TRAP(sk->sk_send_head == NULL);
 508                 sk->sk_send_head = skb;
 509                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 510         } else
 511                 dccp_transmit_skb(sk, skb);
 512
 513         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 514         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 515 }