git.oblomov.eu Git - linux-2.6/blob - net/dccp/output.c

   1 /*
   2  *  net/dccp/output.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/kernel.h>
  16 #include <linux/skbuff.h>
  17
  18 #include <net/sock.h>
  19
  20 #include "ackvec.h"
  21 #include "ccid.h"
  22 #include "dccp.h"
  23
  24 static inline void dccp_event_ack_sent(struct sock *sk)
  25 {
  26         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  27 }
  28
  29 static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  30 {
  31         skb_set_owner_w(skb, sk);
  32         WARN_ON(sk->sk_send_head);
  33         sk->sk_send_head = skb;
  34 }
  35
  36 /*
  37  * All SKB's seen here are completely headerless. It is our
  38  * job to build the DCCP header, and pass the packet down to
  39  * IP so it can do the same plus pass the packet off to the
  40  * device.
  41  */
  42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  43 {
  44         if (likely(skb != NULL)) {
  45                 const struct inet_sock *inet = inet_sk(sk);
  46                 const struct inet_connection_sock *icsk = inet_csk(sk);
  47                 struct dccp_sock *dp = dccp_sk(sk);
  48                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  49                 struct dccp_hdr *dh;
  50                 /* XXX For now we're using only 48 bits sequence numbers */
  51                 const int dccp_header_size = sizeof(*dh) +
  52                                              sizeof(struct dccp_hdr_ext) +
  53                                           dccp_packet_hdr_len(dcb->dccpd_type);
  54                 int err, set_ack = 1;
  55                 u64 ackno = dp->dccps_gsr;
  56
  57                 dccp_inc_seqno(&dp->dccps_gss);
  58
  59                 switch (dcb->dccpd_type) {
  60                 case DCCP_PKT_DATA:
  61                         set_ack = 0;
  62                         /* fall through */
  63                 case DCCP_PKT_DATAACK:
  64                         break;
  65
  66                 case DCCP_PKT_SYNC:
  67                 case DCCP_PKT_SYNCACK:
  68                         ackno = dcb->dccpd_seq;
  69                         /* fall through */
  70                 default:
  71                         /*
  72                          * Only data packets should come through with skb->sk
  73                          * set.
  74                          */
  75                         WARN_ON(skb->sk);
  76                         skb_set_owner_w(skb, sk);
  77                         break;
  78                 }
  79
  80                 dcb->dccpd_seq = dp->dccps_gss;
  81                 dccp_insert_options(sk, skb);
  82
  83                 skb->h.raw = skb_push(skb, dccp_header_size);
  84                 dh = dccp_hdr(skb);
  85
  86                 /* Build DCCP header and checksum it. */
  87                 memset(dh, 0, dccp_header_size);
  88                 dh->dccph_type  = dcb->dccpd_type;
  89                 dh->dccph_sport = inet->sport;
  90                 dh->dccph_dport = inet->dport;
  91                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  92                 dh->dccph_ccval = dcb->dccpd_ccval;
  93                 /* XXX For now we're using only 48 bits sequence numbers */
  94                 dh->dccph_x     = 1;
  95
  96                 dp->dccps_awh = dp->dccps_gss;
  97                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  98                 if (set_ack)
  99                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 100
 101                 switch (dcb->dccpd_type) {
 102                 case DCCP_PKT_REQUEST:
 103                         dccp_hdr_request(skb)->dccph_req_service =
 104                                                         dp->dccps_service;
 105                         break;
 106                 case DCCP_PKT_RESET:
 107                         dccp_hdr_reset(skb)->dccph_reset_code =
 108                                                         dcb->dccpd_reset_code;
 109                         break;
 110                 }
 111
 112                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 113
 114                 if (set_ack)
 115                         dccp_event_ack_sent(sk);
 116
 117                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 118
 119                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 120                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 121                 if (err <= 0)
 122                         return err;
 123
 124                 /* NET_XMIT_CN is special. It does not guarantee,
 125                  * that this packet is lost. It tells that device
 126                  * is about to start to drop packets or already
 127                  * drops some packets of the same priority and
 128                  * invokes us to send less aggressively.
 129                  */
 130                 return err == NET_XMIT_CN ? 0 : err;
 131         }
 132         return -ENOBUFS;
 133 }
 134
 135 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 136 {
 137         struct inet_connection_sock *icsk = inet_csk(sk);
 138         struct dccp_sock *dp = dccp_sk(sk);
 139         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
 140                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 141
 142         /* Now subtract optional transport overhead */
 143         mss_now -= icsk->icsk_ext_hdr_len;
 144
 145         /*
 146          * FIXME: this should come from the CCID infrastructure, where, say,
 147          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 148          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 149          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 150          * make it a multiple of 4
 151          */
 152
 153         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 154
 155         /* And store cached results */
 156         icsk->icsk_pmtu_cookie = pmtu;
 157         dp->dccps_mss_cache = mss_now;
 158
 159         return mss_now;
 160 }
 161
 162 EXPORT_SYMBOL_GPL(dccp_sync_mss);
 163
 164 void dccp_write_space(struct sock *sk)
 165 {
 166         read_lock(&sk->sk_callback_lock);
 167
 168         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 169                 wake_up_interruptible(sk->sk_sleep);
 170         /* Should agree with poll, otherwise some programs break */
 171         if (sock_writeable(sk))
 172                 sk_wake_async(sk, 2, POLL_OUT);
 173
 174         read_unlock(&sk->sk_callback_lock);
 175 }
 176
 177 /**
 178  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 179  * @sk: socket to wait for
 180  * @timeo: for how long
 181  */
 182 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 183                               long *timeo)
 184 {
 185         struct dccp_sock *dp = dccp_sk(sk);
 186         DEFINE_WAIT(wait);
 187         long delay;
 188         int rc;
 189
 190         while (1) {
 191                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 192
 193                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 194                         goto do_error;
 195                 if (!*timeo)
 196                         goto do_nonblock;
 197                 if (signal_pending(current))
 198                         goto do_interrupted;
 199
 200                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 201                                             skb->len);
 202                 if (rc <= 0)
 203                         break;
 204                 delay = msecs_to_jiffies(rc);
 205                 if (delay > *timeo || delay < 0)
 206                         goto do_nonblock;
 207
 208                 sk->sk_write_pending++;
 209                 release_sock(sk);
 210                 *timeo -= schedule_timeout(delay);
 211                 lock_sock(sk);
 212                 sk->sk_write_pending--;
 213         }
 214 out:
 215         finish_wait(sk->sk_sleep, &wait);
 216         return rc;
 217
 218 do_error:
 219         rc = -EPIPE;
 220         goto out;
 221 do_nonblock:
 222         rc = -EAGAIN;
 223         goto out;
 224 do_interrupted:
 225         rc = sock_intr_errno(*timeo);
 226         goto out;
 227 }
 228
 229 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 230 {
 231         const struct dccp_sock *dp = dccp_sk(sk);
 232         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 233                                          skb->len);
 234
 235         if (err > 0)
 236                 err = dccp_wait_for_ccid(sk, skb, timeo);
 237
 238         if (err == 0) {
 239                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 240                 const int len = skb->len;
 241
 242                 if (sk->sk_state == DCCP_PARTOPEN) {
 243                         /* See 8.1.5.  Handshake Completion */
 244                         inet_csk_schedule_ack(sk);
 245                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 246                                                   inet_csk(sk)->icsk_rto,
 247                                                   DCCP_RTO_MAX);
 248                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 249                 } else if (dccp_ack_pending(sk))
 250                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 251                 else
 252                         dcb->dccpd_type = DCCP_PKT_DATA;
 253
 254                 err = dccp_transmit_skb(sk, skb);
 255                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 256         } else
 257                 kfree_skb(skb);
 258
 259         return err;
 260 }
 261
 262 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 263 {
 264         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 265                 return -EHOSTUNREACH; /* Routing failure or similar. */
 266
 267         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 268                                       pskb_copy(skb, GFP_ATOMIC):
 269                                       skb_clone(skb, GFP_ATOMIC)));
 270 }
 271
 272 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 273                                    struct request_sock *req)
 274 {
 275         struct dccp_hdr *dh;
 276         struct dccp_request_sock *dreq;
 277         const int dccp_header_size = sizeof(struct dccp_hdr) +
 278                                      sizeof(struct dccp_hdr_ext) +
 279                                      sizeof(struct dccp_hdr_response);
 280         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 281                                                dccp_header_size, 1,
 282                                            GFP_ATOMIC);
 283         if (skb == NULL)
 284                 return NULL;
 285
 286         /* Reserve space for headers. */
 287         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 288
 289         skb->dst = dst_clone(dst);
 290         skb->csum = 0;
 291
 292         dreq = dccp_rsk(req);
 293         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 294         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 295         dccp_insert_options(sk, skb);
 296
 297         skb->h.raw = skb_push(skb, dccp_header_size);
 298
 299         dh = dccp_hdr(skb);
 300         memset(dh, 0, dccp_header_size);
 301
 302         dh->dccph_sport = inet_sk(sk)->sport;
 303         dh->dccph_dport = inet_rsk(req)->rmt_port;
 304         dh->dccph_doff  = (dccp_header_size +
 305                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 306         dh->dccph_type  = DCCP_PKT_RESPONSE;
 307         dh->dccph_x     = 1;
 308         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 309         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 310         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 311
 312         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 313                                               inet_rsk(req)->rmt_addr);
 314
 315         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 316         return skb;
 317 }
 318
 319 EXPORT_SYMBOL_GPL(dccp_make_response);
 320
 321 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 322                                 const enum dccp_reset_codes code)
 323
 324 {
 325         struct dccp_hdr *dh;
 326         struct dccp_sock *dp = dccp_sk(sk);
 327         const int dccp_header_size = sizeof(struct dccp_hdr) +
 328                                      sizeof(struct dccp_hdr_ext) +
 329                                      sizeof(struct dccp_hdr_reset);
 330         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 331                                                dccp_header_size, 1,
 332                                            GFP_ATOMIC);
 333         if (skb == NULL)
 334                 return NULL;
 335
 336         /* Reserve space for headers. */
 337         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 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
 365         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 366                                               inet_sk(sk)->daddr);
 367
 368         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 369         return skb;
 370 }
 371
 372 /*
 373  * Do all connect socket setups that can be done AF independent.
 374  */
 375 static inline void dccp_connect_init(struct sock *sk)
 376 {
 377         struct dccp_sock *dp = dccp_sk(sk);
 378         struct dst_entry *dst = __sk_dst_get(sk);
 379         struct inet_connection_sock *icsk = inet_csk(sk);
 380
 381         sk->sk_err = 0;
 382         sock_reset_flag(sk, SOCK_DONE);
 383
 384         dccp_sync_mss(sk, dst_mtu(dst));
 385
 386         dccp_update_gss(sk, dp->dccps_iss);
 387         /*
 388          * SWL and AWL are initially adjusted so that they are not less than
 389          * the initial Sequence Numbers received and sent, respectively:
 390          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 391          *      AWL := max(GSS - W' + 1, ISS).
 392          * These adjustments MUST be applied only at the beginning of the
 393          * connection.
 394          */
 395         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 396
 397         icsk->icsk_retransmits = 0;
 398 }
 399
 400 int dccp_connect(struct sock *sk)
 401 {
 402         struct sk_buff *skb;
 403         struct inet_connection_sock *icsk = inet_csk(sk);
 404
 405         dccp_connect_init(sk);
 406
 407         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 408         if (unlikely(skb == NULL))
 409                 return -ENOBUFS;
 410
 411         /* Reserve space for headers. */
 412         skb_reserve(skb, MAX_DCCP_HEADER);
 413
 414         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 415         skb->csum = 0;
 416
 417         dccp_skb_entail(sk, skb);
 418         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 419         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 420
 421         /* Timer for repeating the REQUEST until an answer. */
 422         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 423                                   icsk->icsk_rto, DCCP_RTO_MAX);
 424         return 0;
 425 }
 426
 427 EXPORT_SYMBOL_GPL(dccp_connect);
 428
 429 void dccp_send_ack(struct sock *sk)
 430 {
 431         /* If we have been reset, we may not send again. */
 432         if (sk->sk_state != DCCP_CLOSED) {
 433                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 434
 435                 if (skb == NULL) {
 436                         inet_csk_schedule_ack(sk);
 437                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 438                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 439                                                   TCP_DELACK_MAX,
 440                                                   DCCP_RTO_MAX);
 441                         return;
 442                 }
 443
 444                 /* Reserve space for headers */
 445                 skb_reserve(skb, MAX_DCCP_HEADER);
 446                 skb->csum = 0;
 447                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 448                 dccp_transmit_skb(sk, skb);
 449         }
 450 }
 451
 452 EXPORT_SYMBOL_GPL(dccp_send_ack);
 453
 454 void dccp_send_delayed_ack(struct sock *sk)
 455 {
 456         struct inet_connection_sock *icsk = inet_csk(sk);
 457         /*
 458          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 459          * with using 2s, and active senders also piggyback the ACK into a
 460          * DATAACK packet, so this is really for quiescent senders.
 461          */
 462         unsigned long timeout = jiffies + 2 * HZ;
 463
 464         /* Use new timeout only if there wasn't a older one earlier. */
 465         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 466                 /* If delack timer was blocked or is about to expire,
 467                  * send ACK now.
 468                  *
 469                  * FIXME: check the "about to expire" part
 470                  */
 471                 if (icsk->icsk_ack.blocked) {
 472                         dccp_send_ack(sk);
 473                         return;
 474                 }
 475
 476                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 477                         timeout = icsk->icsk_ack.timeout;
 478         }
 479         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 480         icsk->icsk_ack.timeout = timeout;
 481         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 482 }
 483
 484 void dccp_send_sync(struct sock *sk, const u64 seq,
 485                     const enum dccp_pkt_type pkt_type)
 486 {
 487         /*
 488          * We are not putting this on the write queue, so
 489          * dccp_transmit_skb() will set the ownership to this
 490          * sock.
 491          */
 492         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 493
 494         if (skb == NULL)
 495                 /* FIXME: how to make sure the sync is sent? */
 496                 return;
 497
 498         /* Reserve space for headers and prepare control bits. */
 499         skb_reserve(skb, MAX_DCCP_HEADER);
 500         skb->csum = 0;
 501         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 502         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 503
 504         dccp_transmit_skb(sk, skb);
 505 }
 506
 507 /*
 508  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 509  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 510  * any circumstances.
 511  */
 512 void dccp_send_close(struct sock *sk, const int active)
 513 {
 514         struct dccp_sock *dp = dccp_sk(sk);
 515         struct sk_buff *skb;
 516         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 517
 518         skb = alloc_skb(sk->sk_prot->max_header, prio);
 519         if (skb == NULL)
 520                 return;
 521
 522         /* Reserve space for headers and prepare control bits. */
 523         skb_reserve(skb, sk->sk_prot->max_header);
 524         skb->csum = 0;
 525         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 526                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 527
 528         if (active) {
 529                 dccp_skb_entail(sk, skb);
 530                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 531         } else
 532                 dccp_transmit_skb(sk, skb);
 533 }