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