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 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 u32 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_REQUEST:
  68                         set_ack = 0;
  69                         /* fall through */
  70
  71                 case DCCP_PKT_SYNC:
  72                 case DCCP_PKT_SYNCACK:
  73                         ackno = dcb->dccpd_seq;
  74                         /* fall through */
  75                 default:
  76                         /*
  77                          * Only data packets should come through with skb->sk
  78                          * set.
  79                          */
  80                         WARN_ON(skb->sk);
  81                         skb_set_owner_w(skb, sk);
  82                         break;
  83                 }
  84
  85                 dcb->dccpd_seq = dp->dccps_gss;
  86
  87                 if (dccp_insert_options(sk, skb)) {
  88                         kfree_skb(skb);
  89                         return -EPROTO;
  90                 }
  91
  92                 skb->h.raw = skb_push(skb, dccp_header_size);
  93                 dh = dccp_hdr(skb);
  94
  95                 /* Build DCCP header and checksum it. */
  96                 memset(dh, 0, 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                 /* XXX For now we're using only 48 bits sequence numbers */
 103                 dh->dccph_x     = 1;
 104
 105                 dp->dccps_awh = dp->dccps_gss;
 106                 dccp_hdr_set_seq(dh, dp->dccps_gss);
 107                 if (set_ack)
 108                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 109
 110                 switch (dcb->dccpd_type) {
 111                 case DCCP_PKT_REQUEST:
 112                         dccp_hdr_request(skb)->dccph_req_service =
 113                                                         dp->dccps_service;
 114                         break;
 115                 case DCCP_PKT_RESET:
 116                         dccp_hdr_reset(skb)->dccph_reset_code =
 117                                                         dcb->dccpd_reset_code;
 118                         break;
 119                 }
 120
 121                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 122
 123                 if (set_ack)
 124                         dccp_event_ack_sent(sk);
 125
 126                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 127
 128                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 129                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 130                 if (err <= 0)
 131                         return err;
 132
 133                 /* NET_XMIT_CN is special. It does not guarantee,
 134                  * that this packet is lost. It tells that device
 135                  * is about to start to drop packets or already
 136                  * drops some packets of the same priority and
 137                  * invokes us to send less aggressively.
 138                  */
 139                 return err == NET_XMIT_CN ? 0 : err;
 140         }
 141         return -ENOBUFS;
 142 }
 143
 144 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 145 {
 146         struct inet_connection_sock *icsk = inet_csk(sk);
 147         struct dccp_sock *dp = dccp_sk(sk);
 148         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
 149                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 150
 151         /* Now subtract optional transport overhead */
 152         mss_now -= icsk->icsk_ext_hdr_len;
 153
 154         /*
 155          * FIXME: this should come from the CCID infrastructure, where, say,
 156          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 157          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 158          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 159          * make it a multiple of 4
 160          */
 161
 162         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 163
 164         /* And store cached results */
 165         icsk->icsk_pmtu_cookie = pmtu;
 166         dp->dccps_mss_cache = mss_now;
 167
 168         return mss_now;
 169 }
 170
 171 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 172
 173 void dccp_write_space(struct sock *sk)
 174 {
 175         read_lock(&sk->sk_callback_lock);
 176
 177         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 178                 wake_up_interruptible(sk->sk_sleep);
 179         /* Should agree with poll, otherwise some programs break */
 180         if (sock_writeable(sk))
 181                 sk_wake_async(sk, 2, POLL_OUT);
 182
 183         read_unlock(&sk->sk_callback_lock);
 184 }
 185
 186 /**
 187  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 188  * @sk: socket to wait for
 189  * @timeo: for how long
 190  */
 191 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 192                               long *timeo)
 193 {
 194         struct dccp_sock *dp = dccp_sk(sk);
 195         DEFINE_WAIT(wait);
 196         long delay;
 197         int rc;
 198
 199         while (1) {
 200                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 201
 202                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 203                         goto do_error;
 204                 if (!*timeo)
 205                         goto do_nonblock;
 206                 if (signal_pending(current))
 207                         goto do_interrupted;
 208
 209                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 210                                             skb->len);
 211                 if (rc <= 0)
 212                         break;
 213                 delay = msecs_to_jiffies(rc);
 214                 if (delay > *timeo || delay < 0)
 215                         goto do_nonblock;
 216
 217                 sk->sk_write_pending++;
 218                 release_sock(sk);
 219                 *timeo -= schedule_timeout(delay);
 220                 lock_sock(sk);
 221                 sk->sk_write_pending--;
 222         }
 223 out:
 224         finish_wait(sk->sk_sleep, &wait);
 225         return rc;
 226
 227 do_error:
 228         rc = -EPIPE;
 229         goto out;
 230 do_nonblock:
 231         rc = -EAGAIN;
 232         goto out;
 233 do_interrupted:
 234         rc = sock_intr_errno(*timeo);
 235         goto out;
 236 }
 237
 238 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 239 {
 240         const struct dccp_sock *dp = dccp_sk(sk);
 241         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 242                                          skb->len);
 243
 244         if (err > 0)
 245                 err = dccp_wait_for_ccid(sk, skb, timeo);
 246
 247         if (err == 0) {
 248                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 249                 const int len = skb->len;
 250
 251                 if (sk->sk_state == DCCP_PARTOPEN) {
 252                         /* See 8.1.5.  Handshake Completion */
 253                         inet_csk_schedule_ack(sk);
 254                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 255                                                   inet_csk(sk)->icsk_rto,
 256                                                   DCCP_RTO_MAX);
 257                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 258                 } else if (dccp_ack_pending(sk))
 259                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 260                 else
 261                         dcb->dccpd_type = DCCP_PKT_DATA;
 262
 263                 err = dccp_transmit_skb(sk, skb);
 264                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 265         } else
 266                 kfree_skb(skb);
 267
 268         return err;
 269 }
 270
 271 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 272 {
 273         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 274                 return -EHOSTUNREACH; /* Routing failure or similar. */
 275
 276         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 277                                       pskb_copy(skb, GFP_ATOMIC):
 278                                       skb_clone(skb, GFP_ATOMIC)));
 279 }
 280
 281 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 282                                    struct request_sock *req)
 283 {
 284         struct dccp_hdr *dh;
 285         struct dccp_request_sock *dreq;
 286         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 287                                      sizeof(struct dccp_hdr_ext) +
 288                                      sizeof(struct dccp_hdr_response);
 289         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 290                                            GFP_ATOMIC);
 291         if (skb == NULL)
 292                 return NULL;
 293
 294         /* Reserve space for headers. */
 295         skb_reserve(skb, sk->sk_prot->max_header);
 296
 297         skb->dst = dst_clone(dst);
 298         skb->csum = 0;
 299
 300         dreq = dccp_rsk(req);
 301         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 302         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 303
 304         if (dccp_insert_options(sk, skb)) {
 305                 kfree_skb(skb);
 306                 return NULL;
 307         }
 308
 309         skb->h.raw = skb_push(skb, dccp_header_size);
 310
 311         dh = dccp_hdr(skb);
 312         memset(dh, 0, dccp_header_size);
 313
 314         dh->dccph_sport = inet_sk(sk)->sport;
 315         dh->dccph_dport = inet_rsk(req)->rmt_port;
 316         dh->dccph_doff  = (dccp_header_size +
 317                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 318         dh->dccph_type  = DCCP_PKT_RESPONSE;
 319         dh->dccph_x     = 1;
 320         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 321         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 322         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 323
 324         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 325         return skb;
 326 }
 327
 328 EXPORT_SYMBOL_GPL(dccp_make_response);
 329
 330 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 331                                        const enum dccp_reset_codes code)
 332
 333 {
 334         struct dccp_hdr *dh;
 335         struct dccp_sock *dp = dccp_sk(sk);
 336         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 337                                      sizeof(struct dccp_hdr_ext) +
 338                                      sizeof(struct dccp_hdr_reset);
 339         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 340                                            GFP_ATOMIC);
 341         if (skb == NULL)
 342                 return NULL;
 343
 344         /* Reserve space for headers. */
 345         skb_reserve(skb, sk->sk_prot->max_header);
 346
 347         skb->dst = dst_clone(dst);
 348         skb->csum = 0;
 349
 350         dccp_inc_seqno(&dp->dccps_gss);
 351
 352         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 353         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 354         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 355
 356         if (dccp_insert_options(sk, skb)) {
 357                 kfree_skb(skb);
 358                 return NULL;
 359         }
 360
 361         skb->h.raw = skb_push(skb, dccp_header_size);
 362
 363         dh = dccp_hdr(skb);
 364         memset(dh, 0, dccp_header_size);
 365
 366         dh->dccph_sport = inet_sk(sk)->sport;
 367         dh->dccph_dport = inet_sk(sk)->dport;
 368         dh->dccph_doff  = (dccp_header_size +
 369                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 370         dh->dccph_type  = DCCP_PKT_RESET;
 371         dh->dccph_x     = 1;
 372         dccp_hdr_set_seq(dh, dp->dccps_gss);
 373         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 374
 375         dccp_hdr_reset(skb)->dccph_reset_code = code;
 376         inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
 377
 378         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 379         return skb;
 380 }
 381
 382 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
 383 {
 384         /*
 385          * FIXME: what if rebuild_header fails?
 386          * Should we be doing a rebuild_header here?
 387          */
 388         int err = inet_sk_rebuild_header(sk);
 389
 390         if (err == 0) {
 391                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
 392                                                       code);
 393                 if (skb != NULL) {
 394                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 395                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
 396                         if (err == NET_XMIT_CN)
 397                                 err = 0;
 398                 }
 399         }
 400
 401         return err;
 402 }
 403
 404 /*
 405  * Do all connect socket setups that can be done AF independent.
 406  */
 407 static inline void dccp_connect_init(struct sock *sk)
 408 {
 409         struct dccp_sock *dp = dccp_sk(sk);
 410         struct dst_entry *dst = __sk_dst_get(sk);
 411         struct inet_connection_sock *icsk = inet_csk(sk);
 412
 413         sk->sk_err = 0;
 414         sock_reset_flag(sk, SOCK_DONE);
 415
 416         dccp_sync_mss(sk, dst_mtu(dst));
 417
 418         dccp_update_gss(sk, dp->dccps_iss);
 419         /*
 420          * SWL and AWL are initially adjusted so that they are not less than
 421          * the initial Sequence Numbers received and sent, respectively:
 422          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 423          *      AWL := max(GSS - W' + 1, ISS).
 424          * These adjustments MUST be applied only at the beginning of the
 425          * connection.
 426          */
 427         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 428
 429         icsk->icsk_retransmits = 0;
 430 }
 431
 432 int dccp_connect(struct sock *sk)
 433 {
 434         struct sk_buff *skb;
 435         struct inet_connection_sock *icsk = inet_csk(sk);
 436
 437         dccp_connect_init(sk);
 438
 439         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
 440         if (unlikely(skb == NULL))
 441                 return -ENOBUFS;
 442
 443         /* Reserve space for headers. */
 444         skb_reserve(skb, sk->sk_prot->max_header);
 445
 446         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 447         skb->csum = 0;
 448
 449         dccp_skb_entail(sk, skb);
 450         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 451         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 452
 453         /* Timer for repeating the REQUEST until an answer. */
 454         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 455                                   icsk->icsk_rto, DCCP_RTO_MAX);
 456         return 0;
 457 }
 458
 459 EXPORT_SYMBOL_GPL(dccp_connect);
 460
 461 void dccp_send_ack(struct sock *sk)
 462 {
 463         /* If we have been reset, we may not send again. */
 464         if (sk->sk_state != DCCP_CLOSED) {
 465                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
 466                                                 GFP_ATOMIC);
 467
 468                 if (skb == NULL) {
 469                         inet_csk_schedule_ack(sk);
 470                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 471                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 472                                                   TCP_DELACK_MAX,
 473                                                   DCCP_RTO_MAX);
 474                         return;
 475                 }
 476
 477                 /* Reserve space for headers */
 478                 skb_reserve(skb, sk->sk_prot->max_header);
 479                 skb->csum = 0;
 480                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 481                 dccp_transmit_skb(sk, skb);
 482         }
 483 }
 484
 485 EXPORT_SYMBOL_GPL(dccp_send_ack);
 486
 487 void dccp_send_delayed_ack(struct sock *sk)
 488 {
 489         struct inet_connection_sock *icsk = inet_csk(sk);
 490         /*
 491          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 492          * with using 2s, and active senders also piggyback the ACK into a
 493          * DATAACK packet, so this is really for quiescent senders.
 494          */
 495         unsigned long timeout = jiffies + 2 * HZ;
 496
 497         /* Use new timeout only if there wasn't a older one earlier. */
 498         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 499                 /* If delack timer was blocked or is about to expire,
 500                  * send ACK now.
 501                  *
 502                  * FIXME: check the "about to expire" part
 503                  */
 504                 if (icsk->icsk_ack.blocked) {
 505                         dccp_send_ack(sk);
 506                         return;
 507                 }
 508
 509                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 510                         timeout = icsk->icsk_ack.timeout;
 511         }
 512         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 513         icsk->icsk_ack.timeout = timeout;
 514         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 515 }
 516
 517 void dccp_send_sync(struct sock *sk, const u64 seq,
 518                     const enum dccp_pkt_type pkt_type)
 519 {
 520         /*
 521          * We are not putting this on the write queue, so
 522          * dccp_transmit_skb() will set the ownership to this
 523          * sock.
 524          */
 525         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 526
 527         if (skb == NULL)
 528                 /* FIXME: how to make sure the sync is sent? */
 529                 return;
 530
 531         /* Reserve space for headers and prepare control bits. */
 532         skb_reserve(skb, sk->sk_prot->max_header);
 533         skb->csum = 0;
 534         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 535         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 536
 537         dccp_transmit_skb(sk, skb);
 538 }
 539
 540 EXPORT_SYMBOL_GPL(dccp_send_sync);
 541
 542 /*
 543  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 544  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 545  * any circumstances.
 546  */
 547 void dccp_send_close(struct sock *sk, const int active)
 548 {
 549         struct dccp_sock *dp = dccp_sk(sk);
 550         struct sk_buff *skb;
 551         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 552
 553         skb = alloc_skb(sk->sk_prot->max_header, prio);
 554         if (skb == NULL)
 555                 return;
 556
 557         /* Reserve space for headers and prepare control bits. */
 558         skb_reserve(skb, sk->sk_prot->max_header);
 559         skb->csum = 0;
 560         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 561                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 562
 563         if (active) {
 564                 dccp_skb_entail(sk, skb);
 565                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 566         } else
 567                 dccp_transmit_skb(sk, skb);
 568 }