SAFE public projects git trees. - safe/jmp/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/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                 dccp_insert_options(sk, skb);
  87
  88                 skb->h.raw = skb_push(skb, dccp_header_size);
  89                 dh = dccp_hdr(skb);
  90
  91                 /* Build DCCP header and checksum it. */
  92                 memset(dh, 0, dccp_header_size);
  93                 dh->dccph_type  = dcb->dccpd_type;
  94                 dh->dccph_sport = inet->sport;
  95                 dh->dccph_dport = inet->dport;
  96                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  97                 dh->dccph_ccval = dcb->dccpd_ccval;
  98                 /* XXX For now we're using only 48 bits sequence numbers */
  99                 dh->dccph_x     = 1;
 100
 101                 dp->dccps_awh = dp->dccps_gss;
 102                 dccp_hdr_set_seq(dh, dp->dccps_gss);
 103                 if (set_ack)
 104                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 105
 106                 switch (dcb->dccpd_type) {
 107                 case DCCP_PKT_REQUEST:
 108                         dccp_hdr_request(skb)->dccph_req_service =
 109                                                         dp->dccps_service;
 110                         break;
 111                 case DCCP_PKT_RESET:
 112                         dccp_hdr_reset(skb)->dccph_reset_code =
 113                                                         dcb->dccpd_reset_code;
 114                         break;
 115                 }
 116
 117                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 118
 119                 if (set_ack)
 120                         dccp_event_ack_sent(sk);
 121
 122                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 123
 124                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 125                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 126                 if (err <= 0)
 127                         return err;
 128
 129                 /* NET_XMIT_CN is special. It does not guarantee,
 130                  * that this packet is lost. It tells that device
 131                  * is about to start to drop packets or already
 132                  * drops some packets of the same priority and
 133                  * invokes us to send less aggressively.
 134                  */
 135                 return err == NET_XMIT_CN ? 0 : err;
 136         }
 137         return -ENOBUFS;
 138 }
 139
 140 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 141 {
 142         struct inet_connection_sock *icsk = inet_csk(sk);
 143         struct dccp_sock *dp = dccp_sk(sk);
 144         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
 145                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 146
 147         /* Now subtract optional transport overhead */
 148         mss_now -= icsk->icsk_ext_hdr_len;
 149
 150         /*
 151          * FIXME: this should come from the CCID infrastructure, where, say,
 152          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 153          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 154          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 155          * make it a multiple of 4
 156          */
 157
 158         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 159
 160         /* And store cached results */
 161         icsk->icsk_pmtu_cookie = pmtu;
 162         dp->dccps_mss_cache = mss_now;
 163
 164         return mss_now;
 165 }
 166
 167 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 168
 169 void dccp_write_space(struct sock *sk)
 170 {
 171         read_lock(&sk->sk_callback_lock);
 172
 173         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 174                 wake_up_interruptible(sk->sk_sleep);
 175         /* Should agree with poll, otherwise some programs break */
 176         if (sock_writeable(sk))
 177                 sk_wake_async(sk, 2, POLL_OUT);
 178
 179         read_unlock(&sk->sk_callback_lock);
 180 }
 181
 182 /**
 183  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 184  * @sk: socket to wait for
 185  * @timeo: for how long
 186  */
 187 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 188                               long *timeo)
 189 {
 190         struct dccp_sock *dp = dccp_sk(sk);
 191         DEFINE_WAIT(wait);
 192         long delay;
 193         int rc;
 194
 195         while (1) {
 196                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 197
 198                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 199                         goto do_error;
 200                 if (!*timeo)
 201                         goto do_nonblock;
 202                 if (signal_pending(current))
 203                         goto do_interrupted;
 204
 205                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 206                                             skb->len);
 207                 if (rc <= 0)
 208                         break;
 209                 delay = msecs_to_jiffies(rc);
 210                 if (delay > *timeo || delay < 0)
 211                         goto do_nonblock;
 212
 213                 sk->sk_write_pending++;
 214                 release_sock(sk);
 215                 *timeo -= schedule_timeout(delay);
 216                 lock_sock(sk);
 217                 sk->sk_write_pending--;
 218         }
 219 out:
 220         finish_wait(sk->sk_sleep, &wait);
 221         return rc;
 222
 223 do_error:
 224         rc = -EPIPE;
 225         goto out;
 226 do_nonblock:
 227         rc = -EAGAIN;
 228         goto out;
 229 do_interrupted:
 230         rc = sock_intr_errno(*timeo);
 231         goto out;
 232 }
 233
 234 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 235 {
 236         const struct dccp_sock *dp = dccp_sk(sk);
 237         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 238                                          skb->len);
 239
 240         if (err > 0)
 241                 err = dccp_wait_for_ccid(sk, skb, timeo);
 242
 243         if (err == 0) {
 244                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 245                 const int len = skb->len;
 246
 247                 if (sk->sk_state == DCCP_PARTOPEN) {
 248                         /* See 8.1.5.  Handshake Completion */
 249                         inet_csk_schedule_ack(sk);
 250                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 251                                                   inet_csk(sk)->icsk_rto,
 252                                                   DCCP_RTO_MAX);
 253                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 254                 } else if (dccp_ack_pending(sk))
 255                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 256                 else
 257                         dcb->dccpd_type = DCCP_PKT_DATA;
 258
 259                 err = dccp_transmit_skb(sk, skb);
 260                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 261         } else
 262                 kfree_skb(skb);
 263
 264         return err;
 265 }
 266
 267 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 268 {
 269         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 270                 return -EHOSTUNREACH; /* Routing failure or similar. */
 271
 272         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 273                                       pskb_copy(skb, GFP_ATOMIC):
 274                                       skb_clone(skb, GFP_ATOMIC)));
 275 }
 276
 277 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 278                                    struct request_sock *req)
 279 {
 280         struct dccp_hdr *dh;
 281         struct dccp_request_sock *dreq;
 282         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 283                                      sizeof(struct dccp_hdr_ext) +
 284                                      sizeof(struct dccp_hdr_response);
 285         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 286                                            GFP_ATOMIC);
 287         if (skb == NULL)
 288                 return NULL;
 289
 290         /* Reserve space for headers. */
 291         skb_reserve(skb, sk->sk_prot->max_header);
 292
 293         skb->dst = dst_clone(dst);
 294         skb->csum = 0;
 295
 296         dreq = dccp_rsk(req);
 297         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 298         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 299         dccp_insert_options(sk, skb);
 300
 301         skb->h.raw = skb_push(skb, dccp_header_size);
 302
 303         dh = dccp_hdr(skb);
 304         memset(dh, 0, dccp_header_size);
 305
 306         dh->dccph_sport = inet_sk(sk)->sport;
 307         dh->dccph_dport = inet_rsk(req)->rmt_port;
 308         dh->dccph_doff  = (dccp_header_size +
 309                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 310         dh->dccph_type  = DCCP_PKT_RESPONSE;
 311         dh->dccph_x     = 1;
 312         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 313         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 314         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 315
 316         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 317         return skb;
 318 }
 319
 320 EXPORT_SYMBOL_GPL(dccp_make_response);
 321
 322 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 323                                        const enum dccp_reset_codes code)
 324
 325 {
 326         struct dccp_hdr *dh;
 327         struct dccp_sock *dp = dccp_sk(sk);
 328         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
 329                                      sizeof(struct dccp_hdr_ext) +
 330                                      sizeof(struct dccp_hdr_reset);
 331         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
 332                                            GFP_ATOMIC);
 333         if (skb == NULL)
 334                 return NULL;
 335
 336         /* Reserve space for headers. */
 337         skb_reserve(skb, sk->sk_prot->max_header);
 338
 339         skb->dst = dst_clone(dst);
 340         skb->csum = 0;
 341
 342         dccp_inc_seqno(&dp->dccps_gss);
 343
 344         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 345         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 346         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 347         dccp_insert_options(sk, skb);
 348
 349         skb->h.raw = skb_push(skb, dccp_header_size);
 350
 351         dh = dccp_hdr(skb);
 352         memset(dh, 0, dccp_header_size);
 353
 354         dh->dccph_sport = inet_sk(sk)->sport;
 355         dh->dccph_dport = inet_sk(sk)->dport;
 356         dh->dccph_doff  = (dccp_header_size +
 357                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 358         dh->dccph_type  = DCCP_PKT_RESET;
 359         dh->dccph_x     = 1;
 360         dccp_hdr_set_seq(dh, dp->dccps_gss);
 361         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 362
 363         dccp_hdr_reset(skb)->dccph_reset_code = code;
 364         inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
 365
 366         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 367         return skb;
 368 }
 369
 370 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
 371 {
 372         /*
 373          * FIXME: what if rebuild_header fails?
 374          * Should we be doing a rebuild_header here?
 375          */
 376         int err = inet_sk_rebuild_header(sk);
 377
 378         if (err == 0) {
 379                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
 380                                                       code);
 381                 if (skb != NULL) {
 382                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 383                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
 384                         if (err == NET_XMIT_CN)
 385                                 err = 0;
 386                 }
 387         }
 388
 389         return err;
 390 }
 391
 392 /*
 393  * Do all connect socket setups that can be done AF independent.
 394  */
 395 static inline void dccp_connect_init(struct sock *sk)
 396 {
 397         struct dccp_sock *dp = dccp_sk(sk);
 398         struct dst_entry *dst = __sk_dst_get(sk);
 399         struct inet_connection_sock *icsk = inet_csk(sk);
 400
 401         sk->sk_err = 0;
 402         sock_reset_flag(sk, SOCK_DONE);
 403
 404         dccp_sync_mss(sk, dst_mtu(dst));
 405
 406         dccp_update_gss(sk, dp->dccps_iss);
 407         /*
 408          * SWL and AWL are initially adjusted so that they are not less than
 409          * the initial Sequence Numbers received and sent, respectively:
 410          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 411          *      AWL := max(GSS - W' + 1, ISS).
 412          * These adjustments MUST be applied only at the beginning of the
 413          * connection.
 414          */
 415         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 416
 417         icsk->icsk_retransmits = 0;
 418 }
 419
 420 int dccp_connect(struct sock *sk)
 421 {
 422         struct sk_buff *skb;
 423         struct inet_connection_sock *icsk = inet_csk(sk);
 424
 425         dccp_connect_init(sk);
 426
 427         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
 428         if (unlikely(skb == NULL))
 429                 return -ENOBUFS;
 430
 431         /* Reserve space for headers. */
 432         skb_reserve(skb, sk->sk_prot->max_header);
 433
 434         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 435         skb->csum = 0;
 436
 437         dccp_skb_entail(sk, skb);
 438         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 439         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 440
 441         /* Timer for repeating the REQUEST until an answer. */
 442         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 443                                   icsk->icsk_rto, DCCP_RTO_MAX);
 444         return 0;
 445 }
 446
 447 EXPORT_SYMBOL_GPL(dccp_connect);
 448
 449 void dccp_send_ack(struct sock *sk)
 450 {
 451         /* If we have been reset, we may not send again. */
 452         if (sk->sk_state != DCCP_CLOSED) {
 453                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
 454                                                 GFP_ATOMIC);
 455
 456                 if (skb == NULL) {
 457                         inet_csk_schedule_ack(sk);
 458                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 459                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 460                                                   TCP_DELACK_MAX,
 461                                                   DCCP_RTO_MAX);
 462                         return;
 463                 }
 464
 465                 /* Reserve space for headers */
 466                 skb_reserve(skb, sk->sk_prot->max_header);
 467                 skb->csum = 0;
 468                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 469                 dccp_transmit_skb(sk, skb);
 470         }
 471 }
 472
 473 EXPORT_SYMBOL_GPL(dccp_send_ack);
 474
 475 void dccp_send_delayed_ack(struct sock *sk)
 476 {
 477         struct inet_connection_sock *icsk = inet_csk(sk);
 478         /*
 479          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 480          * with using 2s, and active senders also piggyback the ACK into a
 481          * DATAACK packet, so this is really for quiescent senders.
 482          */
 483         unsigned long timeout = jiffies + 2 * HZ;
 484
 485         /* Use new timeout only if there wasn't a older one earlier. */
 486         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 487                 /* If delack timer was blocked or is about to expire,
 488                  * send ACK now.
 489                  *
 490                  * FIXME: check the "about to expire" part
 491                  */
 492                 if (icsk->icsk_ack.blocked) {
 493                         dccp_send_ack(sk);
 494                         return;
 495                 }
 496
 497                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 498                         timeout = icsk->icsk_ack.timeout;
 499         }
 500         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 501         icsk->icsk_ack.timeout = timeout;
 502         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 503 }
 504
 505 void dccp_send_sync(struct sock *sk, const u64 seq,
 506                     const enum dccp_pkt_type pkt_type)
 507 {
 508         /*
 509          * We are not putting this on the write queue, so
 510          * dccp_transmit_skb() will set the ownership to this
 511          * sock.
 512          */
 513         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 514
 515         if (skb == NULL)
 516                 /* FIXME: how to make sure the sync is sent? */
 517                 return;
 518
 519         /* Reserve space for headers and prepare control bits. */
 520         skb_reserve(skb, sk->sk_prot->max_header);
 521         skb->csum = 0;
 522         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 523         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 524
 525         dccp_transmit_skb(sk, skb);
 526 }
 527
 528 EXPORT_SYMBOL_GPL(dccp_send_sync);
 529
 530 /*
 531  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 532  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 533  * any circumstances.
 534  */
 535 void dccp_send_close(struct sock *sk, const int active)
 536 {
 537         struct dccp_sock *dp = dccp_sk(sk);
 538         struct sk_buff *skb;
 539         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 540
 541         skb = alloc_skb(sk->sk_prot->max_header, prio);
 542         if (skb == NULL)
 543                 return;
 544
 545         /* Reserve space for headers and prepare control bits. */
 546         skb_reserve(skb, sk->sk_prot->max_header);
 547         skb->csum = 0;
 548         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 549                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 550
 551         if (active) {
 552                 dccp_skb_entail(sk, skb);
 553                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 554         } else
 555                 dccp_transmit_skb(sk, skb);
 556 }