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/dccp.h>
  14 #include <linux/kernel.h>
  15 #include <linux/skbuff.h>
  16
  17 #include <net/inet_sock.h>
  18 #include <net/sock.h>
  19
  20 #include "ackvec.h"
  21 #include "ccid.h"
  22 #include "dccp.h"
  23
  24 static inline void dccp_event_ack_sent(struct sock *sk)
  25 {
  26         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  27 }
  28
  29 static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  30 {
  31         skb_set_owner_w(skb, sk);
  32         WARN_ON(sk->sk_send_head);
  33         sk->sk_send_head = skb;
  34 }
  35
  36 /*
  37  * All SKB's seen here are completely headerless. It is our
  38  * job to build the DCCP header, and pass the packet down to
  39  * IP so it can do the same plus pass the packet off to the
  40  * device.
  41  */
  42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  43 {
  44         if (likely(skb != NULL)) {
  45                 const struct inet_sock *inet = inet_sk(sk);
  46                 const struct inet_connection_sock *icsk = inet_csk(sk);
  47                 struct dccp_sock *dp = dccp_sk(sk);
  48                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  49                 struct dccp_hdr *dh;
  50                 /* XXX For now we're using only 48 bits sequence numbers */
  51                 const u32 dccp_header_size = sizeof(*dh) +
  52                                              sizeof(struct dccp_hdr_ext) +
  53                                           dccp_packet_hdr_len(dcb->dccpd_type);
  54                 int err, set_ack = 1;
  55                 u64 ackno = dp->dccps_gsr;
  56
  57                 dccp_inc_seqno(&dp->dccps_gss);
  58
  59                 switch (dcb->dccpd_type) {
  60                 case DCCP_PKT_DATA:
  61                         set_ack = 0;
  62                         /* fall through */
  63                 case DCCP_PKT_DATAACK:
  64                         break;
  65
  66                 case DCCP_PKT_REQUEST:
  67                         set_ack = 0;
  68                         /* fall through */
  69
  70                 case DCCP_PKT_SYNC:
  71                 case DCCP_PKT_SYNCACK:
  72                         ackno = dcb->dccpd_seq;
  73                         /* fall through */
  74                 default:
  75                         /*
  76                          * Only data packets should come through with skb->sk
  77                          * set.
  78                          */
  79                         WARN_ON(skb->sk);
  80                         skb_set_owner_w(skb, sk);
  81                         break;
  82                 }
  83
  84                 dcb->dccpd_seq = dp->dccps_gss;
  85
  86                 if (dccp_insert_options(sk, skb)) {
  87                         kfree_skb(skb);
  88                         return -EPROTO;
  89                 }
  90
  91
  92                 /* Build DCCP header and checksum it. */
  93                 dh = dccp_zeroed_hdr(skb, dccp_header_size);
  94                 dh->dccph_type  = dcb->dccpd_type;
  95                 dh->dccph_sport = inet->sport;
  96                 dh->dccph_dport = inet->dport;
  97                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  98                 dh->dccph_ccval = dcb->dccpd_ccval;
  99                 dh->dccph_cscov = dp->dccps_pcslen;
 100                 /* XXX For now we're using only 48 bits sequence numbers */
 101                 dh->dccph_x     = 1;
 102
 103                 dp->dccps_awh = dp->dccps_gss;
 104                 dccp_hdr_set_seq(dh, dp->dccps_gss);
 105                 if (set_ack)
 106                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 107
 108                 switch (dcb->dccpd_type) {
 109                 case DCCP_PKT_REQUEST:
 110                         dccp_hdr_request(skb)->dccph_req_service =
 111                                                         dp->dccps_service;
 112                         break;
 113                 case DCCP_PKT_RESET:
 114                         dccp_hdr_reset(skb)->dccph_reset_code =
 115                                                         dcb->dccpd_reset_code;
 116                         break;
 117                 }
 118
 119                 icsk->icsk_af_ops->send_check(sk, 0, skb);
 120
 121                 if (set_ack)
 122                         dccp_event_ack_sent(sk);
 123
 124                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 125
 126                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 127                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 128                 return net_xmit_eval(err);
 129         }
 130         return -ENOBUFS;
 131 }
 132
 133 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 134 {
 135         struct inet_connection_sock *icsk = inet_csk(sk);
 136         struct dccp_sock *dp = dccp_sk(sk);
 137         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
 138                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 139
 140         /* Now subtract optional transport overhead */
 141         mss_now -= icsk->icsk_ext_hdr_len;
 142
 143         /*
 144          * FIXME: this should come from the CCID infrastructure, where, say,
 145          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 146          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 147          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 148          * make it a multiple of 4
 149          */
 150
 151         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 152
 153         /* And store cached results */
 154         icsk->icsk_pmtu_cookie = pmtu;
 155         dp->dccps_mss_cache = mss_now;
 156
 157         return mss_now;
 158 }
 159
 160 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 161
 162 void dccp_write_space(struct sock *sk)
 163 {
 164         read_lock(&sk->sk_callback_lock);
 165
 166         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 167                 wake_up_interruptible(sk->sk_sleep);
 168         /* Should agree with poll, otherwise some programs break */
 169         if (sock_writeable(sk))
 170                 sk_wake_async(sk, 2, POLL_OUT);
 171
 172         read_unlock(&sk->sk_callback_lock);
 173 }
 174
 175 /**
 176  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 177  * @sk: socket to wait for
 178  */
 179 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb)
 180 {
 181         struct dccp_sock *dp = dccp_sk(sk);
 182         DEFINE_WAIT(wait);
 183         unsigned long delay;
 184         int rc;
 185
 186         while (1) {
 187                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 188
 189                 if (sk->sk_err)
 190                         goto do_error;
 191                 if (signal_pending(current))
 192                         goto do_interrupted;
 193
 194                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 195                 if (rc <= 0)
 196                         break;
 197                 dccp_pr_debug("delayed send by %d msec\n", rc);
 198                 delay = msecs_to_jiffies(rc);
 199                 sk->sk_write_pending++;
 200                 release_sock(sk);
 201                 schedule_timeout(delay);
 202                 lock_sock(sk);
 203                 sk->sk_write_pending--;
 204         }
 205 out:
 206         finish_wait(sk->sk_sleep, &wait);
 207         return rc;
 208
 209 do_error:
 210         rc = -EPIPE;
 211         goto out;
 212 do_interrupted:
 213         rc = -EINTR;
 214         goto out;
 215 }
 216
 217 void dccp_write_xmit(struct sock *sk, int block)
 218 {
 219         struct dccp_sock *dp = dccp_sk(sk);
 220         struct sk_buff *skb;
 221
 222         while ((skb = skb_peek(&sk->sk_write_queue))) {
 223                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 224
 225                 if (err > 0) {
 226                         if (!block) {
 227                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
 228                                                 msecs_to_jiffies(err)+jiffies);
 229                                 break;
 230                         } else
 231                                 err = dccp_wait_for_ccid(sk, skb);
 232                         if (err && err != -EINTR)
 233                                 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
 234                 }
 235
 236                 skb_dequeue(&sk->sk_write_queue);
 237                 if (err == 0) {
 238                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 239                         const int len = skb->len;
 240
 241                         if (sk->sk_state == DCCP_PARTOPEN) {
 242                                 /* See 8.1.5.  Handshake Completion */
 243                                 inet_csk_schedule_ack(sk);
 244                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 245                                                   inet_csk(sk)->icsk_rto,
 246                                                   DCCP_RTO_MAX);
 247                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
 248                         } else if (dccp_ack_pending(sk))
 249                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
 250                         else
 251                                 dcb->dccpd_type = DCCP_PKT_DATA;
 252
 253                         err = dccp_transmit_skb(sk, skb);
 254                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 255                         if (err)
 256                                 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
 257                                          err);
 258                 } else {
 259                         dccp_pr_debug("packet discarded due to err=%d\n", err);
 260                         kfree_skb(skb);
 261                 }
 262         }
 263 }
 264
 265 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 266 {
 267         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 268                 return -EHOSTUNREACH; /* Routing failure or similar. */
 269
 270         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 271                                       pskb_copy(skb, GFP_ATOMIC):
 272                                       skb_clone(skb, GFP_ATOMIC)));
 273 }
 274
 275 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 276                                    struct request_sock *req)
 277 {
 278         struct dccp_hdr *dh;
 279         struct dccp_request_sock *dreq;
 280         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 281                                      sizeof(struct dccp_hdr_ext) +
 282                                      sizeof(struct dccp_hdr_response);
 283         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 284                                            GFP_ATOMIC);
 285         if (skb == NULL)
 286                 return NULL;
 287
 288         /* Reserve space for headers. */
 289         skb_reserve(skb, sk->sk_prot->max_header);
 290
 291         skb->dst = dst_clone(dst);
 292
 293         dreq = dccp_rsk(req);
 294         if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
 295                 dccp_inc_seqno(&dreq->dreq_iss);
 296         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 297         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 298
 299         if (dccp_insert_options(sk, skb)) {
 300                 kfree_skb(skb);
 301                 return NULL;
 302         }
 303
 304         /* Build and checksum header */
 305         dh = dccp_zeroed_hdr(skb, dccp_header_size);
 306
 307         dh->dccph_sport = inet_sk(sk)->sport;
 308         dh->dccph_dport = inet_rsk(req)->rmt_port;
 309         dh->dccph_doff  = (dccp_header_size +
 310                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 311         dh->dccph_type  = DCCP_PKT_RESPONSE;
 312         dh->dccph_x     = 1;
 313         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 314         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 315         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 316
 317         dccp_csum_outgoing(skb);
 318
 319         /* We use `acked' to remember that a Response was already sent. */
 320         inet_rsk(req)->acked = 1;
 321         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 322         return skb;
 323 }
 324
 325 EXPORT_SYMBOL_GPL(dccp_make_response);
 326
 327 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 328                                        const enum dccp_reset_codes code)
 329 {
 330         struct dccp_hdr *dh;
 331         struct dccp_sock *dp = dccp_sk(sk);
 332         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 333                                      sizeof(struct dccp_hdr_ext) +
 334                                      sizeof(struct dccp_hdr_reset);
 335         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 336                                            GFP_ATOMIC);
 337         if (skb == NULL)
 338                 return NULL;
 339
 340         /* Reserve space for headers. */
 341         skb_reserve(skb, sk->sk_prot->max_header);
 342
 343         skb->dst = dst_clone(dst);
 344
 345         dccp_inc_seqno(&dp->dccps_gss);
 346
 347         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 348         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 349         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 350
 351         if (dccp_insert_options(sk, skb)) {
 352                 kfree_skb(skb);
 353                 return NULL;
 354         }
 355
 356         dh = dccp_zeroed_hdr(skb, dccp_header_size);
 357
 358         dh->dccph_sport = inet_sk(sk)->sport;
 359         dh->dccph_dport = inet_sk(sk)->dport;
 360         dh->dccph_doff  = (dccp_header_size +
 361                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 362         dh->dccph_type  = DCCP_PKT_RESET;
 363         dh->dccph_x     = 1;
 364         dccp_hdr_set_seq(dh, dp->dccps_gss);
 365         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 366
 367         dccp_hdr_reset(skb)->dccph_reset_code = code;
 368         inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
 369
 370         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 371         return skb;
 372 }
 373
 374 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
 375 {
 376         /*
 377          * FIXME: what if rebuild_header fails?
 378          * Should we be doing a rebuild_header here?
 379          */
 380         int err = inet_sk_rebuild_header(sk);
 381
 382         if (err == 0) {
 383                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
 384                                                       code);
 385                 if (skb != NULL) {
 386                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 387                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
 388                         return net_xmit_eval(err);
 389                 }
 390         }
 391
 392         return err;
 393 }
 394
 395 /*
 396  * Do all connect socket setups that can be done AF independent.
 397  */
 398 static inline void dccp_connect_init(struct sock *sk)
 399 {
 400         struct dccp_sock *dp = dccp_sk(sk);
 401         struct dst_entry *dst = __sk_dst_get(sk);
 402         struct inet_connection_sock *icsk = inet_csk(sk);
 403
 404         sk->sk_err = 0;
 405         sock_reset_flag(sk, SOCK_DONE);
 406
 407         dccp_sync_mss(sk, dst_mtu(dst));
 408
 409         /*
 410          * SWL and AWL are initially adjusted so that they are not less than
 411          * the initial Sequence Numbers received and sent, respectively:
 412          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 413          *      AWL := max(GSS - W' + 1, ISS).
 414          * These adjustments MUST be applied only at the beginning of the
 415          * connection.
 416          */
 417         dccp_update_gss(sk, dp->dccps_iss);
 418         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 419
 420         /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
 421          *         initialized to S.ISS (sec. 8.5)                            */
 422         dp->dccps_gar = dp->dccps_iss;
 423
 424         icsk->icsk_retransmits = 0;
 425 }
 426
 427 int dccp_connect(struct sock *sk)
 428 {
 429         struct sk_buff *skb;
 430         struct inet_connection_sock *icsk = inet_csk(sk);
 431
 432         dccp_connect_init(sk);
 433
 434         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
 435         if (unlikely(skb == NULL))
 436                 return -ENOBUFS;
 437
 438         /* Reserve space for headers. */
 439         skb_reserve(skb, sk->sk_prot->max_header);
 440
 441         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 442
 443         dccp_skb_entail(sk, skb);
 444         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 445         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 446
 447         /* Timer for repeating the REQUEST until an answer. */
 448         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 449                                   icsk->icsk_rto, DCCP_RTO_MAX);
 450         return 0;
 451 }
 452
 453 EXPORT_SYMBOL_GPL(dccp_connect);
 454
 455 void dccp_send_ack(struct sock *sk)
 456 {
 457         /* If we have been reset, we may not send again. */
 458         if (sk->sk_state != DCCP_CLOSED) {
 459                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
 460                                                 GFP_ATOMIC);
 461
 462                 if (skb == NULL) {
 463                         inet_csk_schedule_ack(sk);
 464                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 465                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 466                                                   TCP_DELACK_MAX,
 467                                                   DCCP_RTO_MAX);
 468                         return;
 469                 }
 470
 471                 /* Reserve space for headers */
 472                 skb_reserve(skb, sk->sk_prot->max_header);
 473                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 474                 dccp_transmit_skb(sk, skb);
 475         }
 476 }
 477
 478 EXPORT_SYMBOL_GPL(dccp_send_ack);
 479
 480 void dccp_send_delayed_ack(struct sock *sk)
 481 {
 482         struct inet_connection_sock *icsk = inet_csk(sk);
 483         /*
 484          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 485          * with using 2s, and active senders also piggyback the ACK into a
 486          * DATAACK packet, so this is really for quiescent senders.
 487          */
 488         unsigned long timeout = jiffies + 2 * HZ;
 489
 490         /* Use new timeout only if there wasn't a older one earlier. */
 491         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 492                 /* If delack timer was blocked or is about to expire,
 493                  * send ACK now.
 494                  *
 495                  * FIXME: check the "about to expire" part
 496                  */
 497                 if (icsk->icsk_ack.blocked) {
 498                         dccp_send_ack(sk);
 499                         return;
 500                 }
 501
 502                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 503                         timeout = icsk->icsk_ack.timeout;
 504         }
 505         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 506         icsk->icsk_ack.timeout = timeout;
 507         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 508 }
 509
 510 void dccp_send_sync(struct sock *sk, const u64 seq,
 511                     const enum dccp_pkt_type pkt_type)
 512 {
 513         /*
 514          * We are not putting this on the write queue, so
 515          * dccp_transmit_skb() will set the ownership to this
 516          * sock.
 517          */
 518         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 519
 520         if (skb == NULL)
 521                 /* FIXME: how to make sure the sync is sent? */
 522                 return;
 523
 524         /* Reserve space for headers and prepare control bits. */
 525         skb_reserve(skb, sk->sk_prot->max_header);
 526         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 527         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 528
 529         dccp_transmit_skb(sk, skb);
 530 }
 531
 532 EXPORT_SYMBOL_GPL(dccp_send_sync);
 533
 534 /*
 535  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 536  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 537  * any circumstances.
 538  */
 539 void dccp_send_close(struct sock *sk, const int active)
 540 {
 541         struct dccp_sock *dp = dccp_sk(sk);
 542         struct sk_buff *skb;
 543         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 544
 545         skb = alloc_skb(sk->sk_prot->max_header, prio);
 546         if (skb == NULL)
 547                 return;
 548
 549         /* Reserve space for headers and prepare control bits. */
 550         skb_reserve(skb, sk->sk_prot->max_header);
 551         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 552                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 553
 554         if (active) {
 555                 dccp_write_xmit(sk, 1);
 556                 dccp_skb_entail(sk, skb);
 557                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 558                 /* FIXME do we need a retransmit timer here? */
 559         } else
 560                 dccp_transmit_skb(sk, skb);
 561 }