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