err.no 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                 case DCCP_PKT_RESET:
  65                         break;
  66
  67                 case DCCP_PKT_REQUEST:
  68                         set_ack = 0;
  69                         /* fall through */
  70
  71                 case DCCP_PKT_SYNC:
  72                 case DCCP_PKT_SYNCACK:
  73                         ackno = dcb->dccpd_ack_seq;
  74                         /* fall through */
  75                 default:
  76                         /*
  77                          * Set owner/destructor: some skbs are allocated via
  78                          * alloc_skb (e.g. when retransmission may happen).
  79                          * Only Data, DataAck, and Reset packets should come
  80                          * through here with skb->sk set.
  81                          */
  82                         WARN_ON(skb->sk);
  83                         skb_set_owner_w(skb, sk);
  84                         break;
  85                 }
  86
  87                 dcb->dccpd_seq = dp->dccps_gss;
  88
  89                 if (dccp_insert_options(sk, skb)) {
  90                         kfree_skb(skb);
  91                         return -EPROTO;
  92                 }
  93
  94
  95                 /* Build DCCP header and checksum it. */
  96                 dh = dccp_zeroed_hdr(skb, dccp_header_size);
  97                 dh->dccph_type  = dcb->dccpd_type;
  98                 dh->dccph_sport = inet->sport;
  99                 dh->dccph_dport = inet->dport;
 100                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
 101                 dh->dccph_ccval = dcb->dccpd_ccval;
 102                 dh->dccph_cscov = dp->dccps_pcslen;
 103                 /* XXX For now we're using only 48 bits sequence numbers */
 104                 dh->dccph_x     = 1;
 105
 106                 dp->dccps_awh = dp->dccps_gss;
 107                 dccp_hdr_set_seq(dh, dp->dccps_gss);
 108                 if (set_ack)
 109                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 110
 111                 switch (dcb->dccpd_type) {
 112                 case DCCP_PKT_REQUEST:
 113                         dccp_hdr_request(skb)->dccph_req_service =
 114                                                         dp->dccps_service;
 115                         break;
 116                 case DCCP_PKT_RESET:
 117                         dccp_hdr_reset(skb)->dccph_reset_code =
 118                                                         dcb->dccpd_reset_code;
 119                         break;
 120                 }
 121
 122                 icsk->icsk_af_ops->send_check(sk, 0, skb);
 123
 124                 if (set_ack)
 125                         dccp_event_ack_sent(sk);
 126
 127                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 128
 129                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 130                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 131                 return net_xmit_eval(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  */
 182 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb)
 183 {
 184         struct dccp_sock *dp = dccp_sk(sk);
 185         DEFINE_WAIT(wait);
 186         unsigned long delay;
 187         int rc;
 188
 189         while (1) {
 190                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 191
 192                 if (sk->sk_err)
 193                         goto do_error;
 194                 if (signal_pending(current))
 195                         goto do_interrupted;
 196
 197                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 198                 if (rc <= 0)
 199                         break;
 200                 dccp_pr_debug("delayed send by %d msec\n", rc);
 201                 delay = msecs_to_jiffies(rc);
 202                 sk->sk_write_pending++;
 203                 release_sock(sk);
 204                 schedule_timeout(delay);
 205                 lock_sock(sk);
 206                 sk->sk_write_pending--;
 207         }
 208 out:
 209         finish_wait(sk->sk_sleep, &wait);
 210         return rc;
 211
 212 do_error:
 213         rc = -EPIPE;
 214         goto out;
 215 do_interrupted:
 216         rc = -EINTR;
 217         goto out;
 218 }
 219
 220 void dccp_write_xmit(struct sock *sk, int block)
 221 {
 222         struct dccp_sock *dp = dccp_sk(sk);
 223         struct sk_buff *skb;
 224
 225         while ((skb = skb_peek(&sk->sk_write_queue))) {
 226                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 227
 228                 if (err > 0) {
 229                         if (!block) {
 230                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
 231                                                 msecs_to_jiffies(err)+jiffies);
 232                                 break;
 233                         } else
 234                                 err = dccp_wait_for_ccid(sk, skb);
 235                         if (err && err != -EINTR)
 236                                 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
 237                 }
 238
 239                 skb_dequeue(&sk->sk_write_queue);
 240                 if (err == 0) {
 241                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 242                         const int len = skb->len;
 243
 244                         if (sk->sk_state == DCCP_PARTOPEN) {
 245                                 /* See 8.1.5.  Handshake Completion */
 246                                 inet_csk_schedule_ack(sk);
 247                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 248                                                   inet_csk(sk)->icsk_rto,
 249                                                   DCCP_RTO_MAX);
 250                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
 251                         } else if (dccp_ack_pending(sk))
 252                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
 253                         else
 254                                 dcb->dccpd_type = DCCP_PKT_DATA;
 255
 256                         err = dccp_transmit_skb(sk, skb);
 257                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 258                         if (err)
 259                                 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
 260                                          err);
 261                 } else {
 262                         dccp_pr_debug("packet discarded due to err=%d\n", err);
 263                         kfree_skb(skb);
 264                 }
 265         }
 266 }
 267
 268 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 269 {
 270         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 271                 return -EHOSTUNREACH; /* Routing failure or similar. */
 272
 273         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 274                                       pskb_copy(skb, GFP_ATOMIC):
 275                                       skb_clone(skb, GFP_ATOMIC)));
 276 }
 277
 278 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 279                                    struct request_sock *req)
 280 {
 281         struct dccp_hdr *dh;
 282         struct dccp_request_sock *dreq;
 283         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 284                                      sizeof(struct dccp_hdr_ext) +
 285                                      sizeof(struct dccp_hdr_response);
 286         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 287                                            GFP_ATOMIC);
 288         if (skb == NULL)
 289                 return NULL;
 290
 291         /* Reserve space for headers. */
 292         skb_reserve(skb, sk->sk_prot->max_header);
 293
 294         skb->dst = dst_clone(dst);
 295
 296         dreq = dccp_rsk(req);
 297         if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
 298                 dccp_inc_seqno(&dreq->dreq_iss);
 299         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 300         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 301
 302         if (dccp_insert_options(sk, skb)) {
 303                 kfree_skb(skb);
 304                 return NULL;
 305         }
 306
 307         /* Build and checksum header */
 308         dh = dccp_zeroed_hdr(skb, dccp_header_size);
 309
 310         dh->dccph_sport = inet_sk(sk)->sport;
 311         dh->dccph_dport = inet_rsk(req)->rmt_port;
 312         dh->dccph_doff  = (dccp_header_size +
 313                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 314         dh->dccph_type  = DCCP_PKT_RESPONSE;
 315         dh->dccph_x     = 1;
 316         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 317         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 318         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 319
 320         dccp_csum_outgoing(skb);
 321
 322         /* We use `acked' to remember that a Response was already sent. */
 323         inet_rsk(req)->acked = 1;
 324         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 325         return skb;
 326 }
 327
 328 EXPORT_SYMBOL_GPL(dccp_make_response);
 329
 330 /* send Reset on established socket, to close or abort the connection */
 331 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
 332 {
 333         struct sk_buff *skb;
 334         /*
 335          * FIXME: what if rebuild_header fails?
 336          * Should we be doing a rebuild_header here?
 337          */
 338         int err = inet_sk_rebuild_header(sk);
 339
 340         if (err != 0)
 341                 return err;
 342
 343         skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
 344         if (skb == NULL)
 345                 return -ENOBUFS;
 346
 347         /* Reserve space for headers and prepare control bits. */
 348         skb_reserve(skb, sk->sk_prot->max_header);
 349         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 350         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 351
 352         return dccp_transmit_skb(sk, skb);
 353 }
 354
 355 /*
 356  * Do all connect socket setups that can be done AF independent.
 357  */
 358 static inline void dccp_connect_init(struct sock *sk)
 359 {
 360         struct dccp_sock *dp = dccp_sk(sk);
 361         struct dst_entry *dst = __sk_dst_get(sk);
 362         struct inet_connection_sock *icsk = inet_csk(sk);
 363
 364         sk->sk_err = 0;
 365         sock_reset_flag(sk, SOCK_DONE);
 366
 367         dccp_sync_mss(sk, dst_mtu(dst));
 368
 369         /*
 370          * SWL and AWL are initially adjusted so that they are not less than
 371          * the initial Sequence Numbers received and sent, respectively:
 372          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 373          *      AWL := max(GSS - W' + 1, ISS).
 374          * These adjustments MUST be applied only at the beginning of the
 375          * connection.
 376          */
 377         dccp_update_gss(sk, dp->dccps_iss);
 378         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 379
 380         /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
 381          *         initialized to S.ISS (sec. 8.5)                            */
 382         dp->dccps_gar = dp->dccps_iss;
 383
 384         icsk->icsk_retransmits = 0;
 385 }
 386
 387 int dccp_connect(struct sock *sk)
 388 {
 389         struct sk_buff *skb;
 390         struct inet_connection_sock *icsk = inet_csk(sk);
 391
 392         dccp_connect_init(sk);
 393
 394         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
 395         if (unlikely(skb == NULL))
 396                 return -ENOBUFS;
 397
 398         /* Reserve space for headers. */
 399         skb_reserve(skb, sk->sk_prot->max_header);
 400
 401         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 402
 403         dccp_skb_entail(sk, skb);
 404         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 405         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 406
 407         /* Timer for repeating the REQUEST until an answer. */
 408         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 409                                   icsk->icsk_rto, DCCP_RTO_MAX);
 410         return 0;
 411 }
 412
 413 EXPORT_SYMBOL_GPL(dccp_connect);
 414
 415 void dccp_send_ack(struct sock *sk)
 416 {
 417         /* If we have been reset, we may not send again. */
 418         if (sk->sk_state != DCCP_CLOSED) {
 419                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
 420                                                 GFP_ATOMIC);
 421
 422                 if (skb == NULL) {
 423                         inet_csk_schedule_ack(sk);
 424                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 425                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 426                                                   TCP_DELACK_MAX,
 427                                                   DCCP_RTO_MAX);
 428                         return;
 429                 }
 430
 431                 /* Reserve space for headers */
 432                 skb_reserve(skb, sk->sk_prot->max_header);
 433                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 434                 dccp_transmit_skb(sk, skb);
 435         }
 436 }
 437
 438 EXPORT_SYMBOL_GPL(dccp_send_ack);
 439
 440 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
 441 void dccp_send_delayed_ack(struct sock *sk)
 442 {
 443         struct inet_connection_sock *icsk = inet_csk(sk);
 444         /*
 445          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 446          * with using 2s, and active senders also piggyback the ACK into a
 447          * DATAACK packet, so this is really for quiescent senders.
 448          */
 449         unsigned long timeout = jiffies + 2 * HZ;
 450
 451         /* Use new timeout only if there wasn't a older one earlier. */
 452         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 453                 /* If delack timer was blocked or is about to expire,
 454                  * send ACK now.
 455                  *
 456                  * FIXME: check the "about to expire" part
 457                  */
 458                 if (icsk->icsk_ack.blocked) {
 459                         dccp_send_ack(sk);
 460                         return;
 461                 }
 462
 463                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 464                         timeout = icsk->icsk_ack.timeout;
 465         }
 466         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 467         icsk->icsk_ack.timeout = timeout;
 468         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 469 }
 470
 471 void dccp_send_sync(struct sock *sk, const u64 ackno,
 472                     const enum dccp_pkt_type pkt_type)
 473 {
 474         /*
 475          * We are not putting this on the write queue, so
 476          * dccp_transmit_skb() will set the ownership to this
 477          * sock.
 478          */
 479         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 480
 481         if (skb == NULL) {
 482                 /* FIXME: how to make sure the sync is sent? */
 483                 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
 484                 return;
 485         }
 486
 487         /* Reserve space for headers and prepare control bits. */
 488         skb_reserve(skb, sk->sk_prot->max_header);
 489         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 490         DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
 491
 492         dccp_transmit_skb(sk, skb);
 493 }
 494
 495 EXPORT_SYMBOL_GPL(dccp_send_sync);
 496
 497 /*
 498  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 499  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 500  * any circumstances.
 501  */
 502 void dccp_send_close(struct sock *sk, const int active)
 503 {
 504         struct dccp_sock *dp = dccp_sk(sk);
 505         struct sk_buff *skb;
 506         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 507
 508         skb = alloc_skb(sk->sk_prot->max_header, prio);
 509         if (skb == NULL)
 510                 return;
 511
 512         /* Reserve space for headers and prepare control bits. */
 513         skb_reserve(skb, sk->sk_prot->max_header);
 514         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 515                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 516
 517         if (active) {
 518                 dccp_write_xmit(sk, 1);
 519                 dccp_skb_entail(sk, skb);
 520                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 521                 /* FIXME do we need a retransmit timer here? */
 522         } else
 523                 dccp_transmit_skb(sk, skb);
 524 }