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 #include <linux/slab.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 /* enqueue @skb on sk_send_head for retransmission, return clone to send now */
  31 static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  32 {
  33         skb_set_owner_w(skb, sk);
  34         WARN_ON(sk->sk_send_head);
  35         sk->sk_send_head = skb;
  36         return skb_clone(sk->sk_send_head, gfp_any());
  37 }
  38
  39 /*
  40  * All SKB's seen here are completely headerless. It is our
  41  * job to build the DCCP header, and pass the packet down to
  42  * IP so it can do the same plus pass the packet off to the
  43  * device.
  44  */
  45 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  46 {
  47         if (likely(skb != NULL)) {
  48                 struct inet_sock *inet = inet_sk(sk);
  49                 const struct inet_connection_sock *icsk = inet_csk(sk);
  50                 struct dccp_sock *dp = dccp_sk(sk);
  51                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  52                 struct dccp_hdr *dh;
  53                 /* XXX For now we're using only 48 bits sequence numbers */
  54                 const u32 dccp_header_size = sizeof(*dh) +
  55                                              sizeof(struct dccp_hdr_ext) +
  56                                           dccp_packet_hdr_len(dcb->dccpd_type);
  57                 int err, set_ack = 1;
  58                 u64 ackno = dp->dccps_gsr;
  59                 /*
  60                  * Increment GSS here already in case the option code needs it.
  61                  * Update GSS for real only if option processing below succeeds.
  62                  */
  63                 dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
  64
  65                 switch (dcb->dccpd_type) {
  66                 case DCCP_PKT_DATA:
  67                         set_ack = 0;
  68                         /* fall through */
  69                 case DCCP_PKT_DATAACK:
  70                 case DCCP_PKT_RESET:
  71                         break;
  72
  73                 case DCCP_PKT_REQUEST:
  74                         set_ack = 0;
  75                         /* Use ISS on the first (non-retransmitted) Request. */
  76                         if (icsk->icsk_retransmits == 0)
  77                                 dcb->dccpd_seq = dp->dccps_iss;
  78                         /* fall through */
  79
  80                 case DCCP_PKT_SYNC:
  81                 case DCCP_PKT_SYNCACK:
  82                         ackno = dcb->dccpd_ack_seq;
  83                         /* fall through */
  84                 default:
  85                         /*
  86                          * Set owner/destructor: some skbs are allocated via
  87                          * alloc_skb (e.g. when retransmission may happen).
  88                          * Only Data, DataAck, and Reset packets should come
  89                          * through here with skb->sk set.
  90                          */
  91                         WARN_ON(skb->sk);
  92                         skb_set_owner_w(skb, sk);
  93                         break;
  94                 }
  95
  96                 if (dccp_insert_options(sk, skb)) {
  97                         kfree_skb(skb);
  98                         return -EPROTO;
  99                 }
 100
 101
 102                 /* Build DCCP header and checksum it. */
 103                 dh = dccp_zeroed_hdr(skb, dccp_header_size);
 104                 dh->dccph_type  = dcb->dccpd_type;
 105                 dh->dccph_sport = inet->inet_sport;
 106                 dh->dccph_dport = inet->inet_dport;
 107                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
 108                 dh->dccph_ccval = dcb->dccpd_ccval;
 109                 dh->dccph_cscov = dp->dccps_pcslen;
 110                 /* XXX For now we're using only 48 bits sequence numbers */
 111                 dh->dccph_x     = 1;
 112
 113                 dccp_update_gss(sk, dcb->dccpd_seq);
 114                 dccp_hdr_set_seq(dh, dp->dccps_gss);
 115                 if (set_ack)
 116                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 117
 118                 switch (dcb->dccpd_type) {
 119                 case DCCP_PKT_REQUEST:
 120                         dccp_hdr_request(skb)->dccph_req_service =
 121                                                         dp->dccps_service;
 122                         /*
 123                          * Limit Ack window to ISS <= P.ackno <= GSS, so that
 124                          * only Responses to Requests we sent are considered.
 125                          */
 126                         dp->dccps_awl = dp->dccps_iss;
 127                         break;
 128                 case DCCP_PKT_RESET:
 129                         dccp_hdr_reset(skb)->dccph_reset_code =
 130                                                         dcb->dccpd_reset_code;
 131                         break;
 132                 }
 133
 134                 icsk->icsk_af_ops->send_check(sk, skb);
 135
 136                 if (set_ack)
 137                         dccp_event_ack_sent(sk);
 138
 139                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 140
 141                 err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
 142                 return net_xmit_eval(err);
 143         }
 144         return -ENOBUFS;
 145 }
 146
 147 /**
 148  * dccp_determine_ccmps  -  Find out about CCID-specific packet-size limits
 149  * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
 150  * since the RX CCID is restricted to feedback packets (Acks), which are small
 151  * in comparison with the data traffic. A value of 0 means "no current CCMPS".
 152  */
 153 static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
 154 {
 155         const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
 156
 157         if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
 158                 return 0;
 159         return tx_ccid->ccid_ops->ccid_ccmps;
 160 }
 161
 162 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 163 {
 164         struct inet_connection_sock *icsk = inet_csk(sk);
 165         struct dccp_sock *dp = dccp_sk(sk);
 166         u32 ccmps = dccp_determine_ccmps(dp);
 167         u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
 168
 169         /* Account for header lengths and IPv4/v6 option overhead */
 170         cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
 171                     sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
 172
 173         /*
 174          * Leave enough headroom for common DCCP header options.
 175          * This only considers options which may appear on DCCP-Data packets, as
 176          * per table 3 in RFC 4340, 5.8. When running out of space for other
 177          * options (eg. Ack Vector which can take up to 255 bytes), it is better
 178          * to schedule a separate Ack. Thus we leave headroom for the following:
 179          *  - 1 byte for Slow Receiver (11.6)
 180          *  - 6 bytes for Timestamp (13.1)
 181          *  - 10 bytes for Timestamp Echo (13.3)
 182          *  - 8 bytes for NDP count (7.7, when activated)
 183          *  - 6 bytes for Data Checksum (9.3)
 184          *  - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
 185          */
 186         cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
 187                            (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
 188
 189         /* And store cached results */
 190         icsk->icsk_pmtu_cookie = pmtu;
 191         dp->dccps_mss_cache = cur_mps;
 192
 193         return cur_mps;
 194 }
 195
 196 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 197
 198 void dccp_write_space(struct sock *sk)
 199 {
 200         struct socket_wq *wq;
 201
 202         rcu_read_lock();
 203         wq = rcu_dereference(sk->sk_wq);
 204         if (skwq_has_sleeper(wq))
 205                 wake_up_interruptible(&wq->wait);
 206         /* Should agree with poll, otherwise some programs break */
 207         if (sock_writeable(sk))
 208                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 209
 210         rcu_read_unlock();
 211 }
 212
 213 /**
 214  * dccp_wait_for_ccid  -  Await CCID send permission
 215  * @sk:    socket to wait for
 216  * @delay: timeout in jiffies
 217  *
 218  * This is used by CCIDs which need to delay the send time in process context.
 219  */
 220 static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
 221 {
 222         DEFINE_WAIT(wait);
 223         long remaining;
 224
 225         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 226         sk->sk_write_pending++;
 227         release_sock(sk);
 228
 229         remaining = schedule_timeout(delay);
 230
 231         lock_sock(sk);
 232         sk->sk_write_pending--;
 233         finish_wait(sk_sleep(sk), &wait);
 234
 235         if (signal_pending(current) || sk->sk_err)
 236                 return -1;
 237         return remaining;
 238 }
 239
 240 /**
 241  * dccp_xmit_packet  -  Send data packet under control of CCID
 242  * Transmits next-queued payload and informs CCID to account for the packet.
 243  */
 244 static void dccp_xmit_packet(struct sock *sk)
 245 {
 246         int err, len;
 247         struct dccp_sock *dp = dccp_sk(sk);
 248         struct sk_buff *skb = dccp_qpolicy_pop(sk);
 249
 250         if (unlikely(skb == NULL))
 251                 return;
 252         len = skb->len;
 253
 254         if (sk->sk_state == DCCP_PARTOPEN) {
 255                 const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
 256                 /*
 257                  * See 8.1.5 - Handshake Completion.
 258                  *
 259                  * For robustness we resend Confirm options until the client has
 260                  * entered OPEN. During the initial feature negotiation, the MPS
 261                  * is smaller than usual, reduced by the Change/Confirm options.
 262                  */
 263                 if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
 264                         DCCP_WARN("Payload too large (%d) for featneg.\n", len);
 265                         dccp_send_ack(sk);
 266                         dccp_feat_list_purge(&dp->dccps_featneg);
 267                 }
 268
 269                 inet_csk_schedule_ack(sk);
 270                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 271                                               inet_csk(sk)->icsk_rto,
 272                                               DCCP_RTO_MAX);
 273                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
 274         } else if (dccp_ack_pending(sk)) {
 275                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
 276         } else {
 277                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
 278         }
 279
 280         err = dccp_transmit_skb(sk, skb);
 281         if (err)
 282                 dccp_pr_debug("transmit_skb() returned err=%d\n", err);
 283         /*
 284          * Register this one as sent even if an error occurred. To the remote
 285          * end a local packet drop is indistinguishable from network loss, i.e.
 286          * any local drop will eventually be reported via receiver feedback.
 287          */
 288         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
 289
 290         /*
 291          * If the CCID needs to transfer additional header options out-of-band
 292          * (e.g. Ack Vectors or feature-negotiation options), it activates this
 293          * flag to schedule a Sync. The Sync will automatically incorporate all
 294          * currently pending header options, thus clearing the backlog.
 295          */
 296         if (dp->dccps_sync_scheduled)
 297                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 298 }
 299
 300 /**
 301  * dccp_flush_write_queue  -  Drain queue at end of connection
 302  * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
 303  * happen that the TX queue is not empty at the end of a connection. We give the
 304  * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
 305  * returns with a non-empty write queue, it will be purged later.
 306  */
 307 void dccp_flush_write_queue(struct sock *sk, long *time_budget)
 308 {
 309         struct dccp_sock *dp = dccp_sk(sk);
 310         struct sk_buff *skb;
 311         long delay, rc;
 312
 313         while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
 314                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 315
 316                 switch (ccid_packet_dequeue_eval(rc)) {
 317                 case CCID_PACKET_WILL_DEQUEUE_LATER:
 318                         /*
 319                          * If the CCID determines when to send, the next sending
 320                          * time is unknown or the CCID may not even send again
 321                          * (e.g. remote host crashes or lost Ack packets).
 322                          */
 323                         DCCP_WARN("CCID did not manage to send all packets\n");
 324                         return;
 325                 case CCID_PACKET_DELAY:
 326                         delay = msecs_to_jiffies(rc);
 327                         if (delay > *time_budget)
 328                                 return;
 329                         rc = dccp_wait_for_ccid(sk, delay);
 330                         if (rc < 0)
 331                                 return;
 332                         *time_budget -= (delay - rc);
 333                         /* check again if we can send now */
 334                         break;
 335                 case CCID_PACKET_SEND_AT_ONCE:
 336                         dccp_xmit_packet(sk);
 337                         break;
 338                 case CCID_PACKET_ERR:
 339                         skb_dequeue(&sk->sk_write_queue);
 340                         kfree_skb(skb);
 341                         dccp_pr_debug("packet discarded due to err=%ld\n", rc);
 342                 }
 343         }
 344 }
 345
 346 void dccp_write_xmit(struct sock *sk)
 347 {
 348         struct dccp_sock *dp = dccp_sk(sk);
 349         struct sk_buff *skb;
 350
 351         while ((skb = dccp_qpolicy_top(sk))) {
 352                 int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
 353
 354                 switch (ccid_packet_dequeue_eval(rc)) {
 355                 case CCID_PACKET_WILL_DEQUEUE_LATER:
 356                         return;
 357                 case CCID_PACKET_DELAY:
 358                         sk_reset_timer(sk, &dp->dccps_xmit_timer,
 359                                        jiffies + msecs_to_jiffies(rc));
 360                         return;
 361                 case CCID_PACKET_SEND_AT_ONCE:
 362                         dccp_xmit_packet(sk);
 363                         break;
 364                 case CCID_PACKET_ERR:
 365                         dccp_qpolicy_drop(sk, skb);
 366                         dccp_pr_debug("packet discarded due to err=%d\n", rc);
 367                 }
 368         }
 369 }
 370
 371 /**
 372  * dccp_retransmit_skb  -  Retransmit Request, Close, or CloseReq packets
 373  * There are only four retransmittable packet types in DCCP:
 374  * - Request  in client-REQUEST  state (sec. 8.1.1),
 375  * - CloseReq in server-CLOSEREQ state (sec. 8.3),
 376  * - Close    in   node-CLOSING  state (sec. 8.3),
 377  * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
 378  * This function expects sk->sk_send_head to contain the original skb.
 379  */
 380 int dccp_retransmit_skb(struct sock *sk)
 381 {
 382         WARN_ON(sk->sk_send_head == NULL);
 383
 384         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 385                 return -EHOSTUNREACH; /* Routing failure or similar. */
 386
 387         /* this count is used to distinguish original and retransmitted skb */
 388         inet_csk(sk)->icsk_retransmits++;
 389
 390         return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
 391 }
 392
 393 struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
 394                                    struct request_sock *req)
 395 {
 396         struct dccp_hdr *dh;
 397         struct dccp_request_sock *dreq;
 398         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 399                                      sizeof(struct dccp_hdr_ext) +
 400                                      sizeof(struct dccp_hdr_response);
 401         struct sk_buff *skb;
 402
 403         /* sk is marked const to clearly express we dont hold socket lock.
 404          * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
 405          * it is safe to promote sk to non const.
 406          */
 407         skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
 408                            GFP_ATOMIC);
 409         if (!skb)
 410                 return NULL;
 411
 412         skb_reserve(skb, MAX_DCCP_HEADER);
 413
 414         skb_dst_set(skb, dst_clone(dst));
 415
 416         dreq = dccp_rsk(req);
 417         if (inet_rsk(req)->acked)       /* increase GSS upon retransmission */
 418                 dccp_inc_seqno(&dreq->dreq_gss);
 419         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 420         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_gss;
 421
 422         /* Resolve feature dependencies resulting from choice of CCID */
 423         if (dccp_feat_server_ccid_dependencies(dreq))
 424                 goto response_failed;
 425
 426         if (dccp_insert_options_rsk(dreq, skb))
 427                 goto response_failed;
 428
 429         /* Build and checksum header */
 430         dh = dccp_zeroed_hdr(skb, dccp_header_size);
 431
 432         dh->dccph_sport = htons(inet_rsk(req)->ir_num);
 433         dh->dccph_dport = inet_rsk(req)->ir_rmt_port;
 434         dh->dccph_doff  = (dccp_header_size +
 435                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 436         dh->dccph_type  = DCCP_PKT_RESPONSE;
 437         dh->dccph_x     = 1;
 438         dccp_hdr_set_seq(dh, dreq->dreq_gss);
 439         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
 440         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 441
 442         dccp_csum_outgoing(skb);
 443
 444         /* We use `acked' to remember that a Response was already sent. */
 445         inet_rsk(req)->acked = 1;
 446         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 447         return skb;
 448 response_failed:
 449         kfree_skb(skb);
 450         return NULL;
 451 }
 452
 453 EXPORT_SYMBOL_GPL(dccp_make_response);
 454
 455 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
 456 struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
 457 {
 458         struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
 459         struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
 460         const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 461                                        sizeof(struct dccp_hdr_ext) +
 462                                        sizeof(struct dccp_hdr_reset);
 463         struct dccp_hdr_reset *dhr;
 464         struct sk_buff *skb;
 465
 466         skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 467         if (skb == NULL)
 468                 return NULL;
 469
 470         skb_reserve(skb, sk->sk_prot->max_header);
 471
 472         /* Swap the send and the receive. */
 473         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
 474         dh->dccph_type  = DCCP_PKT_RESET;
 475         dh->dccph_sport = rxdh->dccph_dport;
 476         dh->dccph_dport = rxdh->dccph_sport;
 477         dh->dccph_doff  = dccp_hdr_reset_len / 4;
 478         dh->dccph_x     = 1;
 479
 480         dhr = dccp_hdr_reset(skb);
 481         dhr->dccph_reset_code = dcb->dccpd_reset_code;
 482
 483         switch (dcb->dccpd_reset_code) {
 484         case DCCP_RESET_CODE_PACKET_ERROR:
 485                 dhr->dccph_reset_data[0] = rxdh->dccph_type;
 486                 break;
 487         case DCCP_RESET_CODE_OPTION_ERROR:      /* fall through */
 488         case DCCP_RESET_CODE_MANDATORY_ERROR:
 489                 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
 490                 break;
 491         }
 492         /*
 493          * From RFC 4340, 8.3.1:
 494          *   If P.ackno exists, set R.seqno := P.ackno + 1.
 495          *   Else set R.seqno := 0.
 496          */
 497         if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 498                 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
 499         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
 500
 501         dccp_csum_outgoing(skb);
 502         return skb;
 503 }
 504
 505 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
 506
 507 /* send Reset on established socket, to close or abort the connection */
 508 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
 509 {
 510         struct sk_buff *skb;
 511         /*
 512          * FIXME: what if rebuild_header fails?
 513          * Should we be doing a rebuild_header here?
 514          */
 515         int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
 516
 517         if (err != 0)
 518                 return err;
 519
 520         skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
 521         if (skb == NULL)
 522                 return -ENOBUFS;
 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       = DCCP_PKT_RESET;
 527         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 528
 529         return dccp_transmit_skb(sk, skb);
 530 }
 531
 532 /*
 533  * Do all connect socket setups that can be done AF independent.
 534  */
 535 int dccp_connect(struct sock *sk)
 536 {
 537         struct sk_buff *skb;
 538         struct dccp_sock *dp = dccp_sk(sk);
 539         struct dst_entry *dst = __sk_dst_get(sk);
 540         struct inet_connection_sock *icsk = inet_csk(sk);
 541
 542         sk->sk_err = 0;
 543         sock_reset_flag(sk, SOCK_DONE);
 544
 545         dccp_sync_mss(sk, dst_mtu(dst));
 546
 547         /* do not connect if feature negotiation setup fails */
 548         if (dccp_feat_finalise_settings(dccp_sk(sk)))
 549                 return -EPROTO;
 550
 551         /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
 552         dp->dccps_gar = dp->dccps_iss;
 553
 554         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
 555         if (unlikely(skb == NULL))
 556                 return -ENOBUFS;
 557
 558         /* Reserve space for headers. */
 559         skb_reserve(skb, sk->sk_prot->max_header);
 560
 561         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 562
 563         dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
 564         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 565
 566         /* Timer for repeating the REQUEST until an answer. */
 567         icsk->icsk_retransmits = 0;
 568         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 569                                   icsk->icsk_rto, DCCP_RTO_MAX);
 570         return 0;
 571 }
 572
 573 EXPORT_SYMBOL_GPL(dccp_connect);
 574
 575 void dccp_send_ack(struct sock *sk)
 576 {
 577         /* If we have been reset, we may not send again. */
 578         if (sk->sk_state != DCCP_CLOSED) {
 579                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
 580                                                 GFP_ATOMIC);
 581
 582                 if (skb == NULL) {
 583                         inet_csk_schedule_ack(sk);
 584                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 585                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 586                                                   TCP_DELACK_MAX,
 587                                                   DCCP_RTO_MAX);
 588                         return;
 589                 }
 590
 591                 /* Reserve space for headers */
 592                 skb_reserve(skb, sk->sk_prot->max_header);
 593                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 594                 dccp_transmit_skb(sk, skb);
 595         }
 596 }
 597
 598 EXPORT_SYMBOL_GPL(dccp_send_ack);
 599
 600 #if 0
 601 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
 602 void dccp_send_delayed_ack(struct sock *sk)
 603 {
 604         struct inet_connection_sock *icsk = inet_csk(sk);
 605         /*
 606          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 607          * with using 2s, and active senders also piggyback the ACK into a
 608          * DATAACK packet, so this is really for quiescent senders.
 609          */
 610         unsigned long timeout = jiffies + 2 * HZ;
 611
 612         /* Use new timeout only if there wasn't a older one earlier. */
 613         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 614                 /* If delack timer was blocked or is about to expire,
 615                  * send ACK now.
 616                  *
 617                  * FIXME: check the "about to expire" part
 618                  */
 619                 if (icsk->icsk_ack.blocked) {
 620                         dccp_send_ack(sk);
 621                         return;
 622                 }
 623
 624                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 625                         timeout = icsk->icsk_ack.timeout;
 626         }
 627         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 628         icsk->icsk_ack.timeout = timeout;
 629         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 630 }
 631 #endif
 632
 633 void dccp_send_sync(struct sock *sk, const u64 ackno,
 634                     const enum dccp_pkt_type pkt_type)
 635 {
 636         /*
 637          * We are not putting this on the write queue, so
 638          * dccp_transmit_skb() will set the ownership to this
 639          * sock.
 640          */
 641         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 642
 643         if (skb == NULL) {
 644                 /* FIXME: how to make sure the sync is sent? */
 645                 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
 646                 return;
 647         }
 648
 649         /* Reserve space for headers and prepare control bits. */
 650         skb_reserve(skb, sk->sk_prot->max_header);
 651         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 652         DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
 653
 654         /*
 655          * Clear the flag in case the Sync was scheduled for out-of-band data,
 656          * such as carrying a long Ack Vector.
 657          */
 658         dccp_sk(sk)->dccps_sync_scheduled = 0;
 659
 660         dccp_transmit_skb(sk, skb);
 661 }
 662
 663 EXPORT_SYMBOL_GPL(dccp_send_sync);
 664
 665 /*
 666  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 667  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 668  * any circumstances.
 669  */
 670 void dccp_send_close(struct sock *sk, const int active)
 671 {
 672         struct dccp_sock *dp = dccp_sk(sk);
 673         struct sk_buff *skb;
 674         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 675
 676         skb = alloc_skb(sk->sk_prot->max_header, prio);
 677         if (skb == NULL)
 678                 return;
 679
 680         /* Reserve space for headers and prepare control bits. */
 681         skb_reserve(skb, sk->sk_prot->max_header);
 682         if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
 683                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
 684         else
 685                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
 686
 687         if (active) {
 688                 skb = dccp_skb_entail(sk, skb);
 689                 /*
 690                  * Retransmission timer for active-close: RFC 4340, 8.3 requires
 691                  * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
 692                  * state can be left. The initial timeout is 2 RTTs.
 693                  * Since RTT measurement is done by the CCIDs, there is no easy
 694                  * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
 695                  * is too low (200ms); we use a high value to avoid unnecessary
 696                  * retransmissions when the link RTT is > 0.2 seconds.
 697                  * FIXME: Let main module sample RTTs and use that instead.
 698                  */
 699                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 700                                           DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
 701         }
 702         dccp_transmit_skb(sk, skb);
 703 }