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/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                 struct dccp_sock *dp = dccp_sk(sk);
  47                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  48                 struct dccp_hdr *dh;
  49                 /* XXX For now we're using only 48 bits sequence numbers */
  50                 const int dccp_header_size = sizeof(*dh) +
  51                                              sizeof(struct dccp_hdr_ext) +
  52                                           dccp_packet_hdr_len(dcb->dccpd_type);
  53                 int err, set_ack = 1;
  54                 u64 ackno = dp->dccps_gsr;
  55
  56                 dccp_inc_seqno(&dp->dccps_gss);
  57
  58                 switch (dcb->dccpd_type) {
  59                 case DCCP_PKT_DATA:
  60                         set_ack = 0;
  61                         break;
  62                 case DCCP_PKT_SYNC:
  63                 case DCCP_PKT_SYNCACK:
  64                         ackno = dcb->dccpd_seq;
  65                         break;
  66                 }
  67
  68                 dcb->dccpd_seq = dp->dccps_gss;
  69                 dccp_insert_options(sk, skb);
  70
  71                 skb->h.raw = skb_push(skb, dccp_header_size);
  72                 dh = dccp_hdr(skb);
  73
  74                 /*
  75                  * Only data packets should come through with skb->sk set.
  76                  */
  77                 if (!skb->sk)
  78                         skb_set_owner_w(skb, sk);
  79
  80                 /* Build DCCP header and checksum it. */
  81                 memset(dh, 0, dccp_header_size);
  82                 dh->dccph_type  = dcb->dccpd_type;
  83                 dh->dccph_sport = inet->sport;
  84                 dh->dccph_dport = inet->dport;
  85                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  86                 dh->dccph_ccval = dcb->dccpd_ccval;
  87                 /* XXX For now we're using only 48 bits sequence numbers */
  88                 dh->dccph_x     = 1;
  89
  90                 dp->dccps_awh = dp->dccps_gss;
  91                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  92                 if (set_ack)
  93                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  94
  95                 switch (dcb->dccpd_type) {
  96                 case DCCP_PKT_REQUEST:
  97                         dccp_hdr_request(skb)->dccph_req_service =
  98                                                         dp->dccps_service;
  99                         break;
 100                 case DCCP_PKT_RESET:
 101                         dccp_hdr_reset(skb)->dccph_reset_code =
 102                                                         dcb->dccpd_reset_code;
 103                         break;
 104                 }
 105
 106                 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
 107                                                       inet->daddr);
 108
 109                 if (set_ack)
 110                         dccp_event_ack_sent(sk);
 111
 112                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 113
 114                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 115                 err = ip_queue_xmit(skb, 0);
 116                 if (err <= 0)
 117                         return err;
 118
 119                 /* NET_XMIT_CN is special. It does not guarantee,
 120                  * that this packet is lost. It tells that device
 121                  * is about to start to drop packets or already
 122                  * drops some packets of the same priority and
 123                  * invokes us to send less aggressively.
 124                  */
 125                 return err == NET_XMIT_CN ? 0 : err;
 126         }
 127         return -ENOBUFS;
 128 }
 129
 130 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 131 {
 132         struct dccp_sock *dp = dccp_sk(sk);
 133         int mss_now;
 134
 135         /*
 136          * FIXME: we really should be using the af_specific thing to support
 137          *        IPv6.
 138          * mss_now = pmtu - tp->af_specific->net_header_len -
 139          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 140          */
 141         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
 142                   sizeof(struct dccp_hdr_ext);
 143
 144         /* Now subtract optional transport overhead */
 145         mss_now -= dp->dccps_ext_header_len;
 146
 147         /*
 148          * FIXME: this should come from the CCID infrastructure, where, say,
 149          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 150          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 151          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 152          * make it a multiple of 4
 153          */
 154
 155         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 156
 157         /* And store cached results */
 158         dp->dccps_pmtu_cookie = pmtu;
 159         dp->dccps_mss_cache = mss_now;
 160
 161         return mss_now;
 162 }
 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_sk_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 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 320                                 const enum dccp_reset_codes code)
 321
 322 {
 323         struct dccp_hdr *dh;
 324         struct dccp_sock *dp = dccp_sk(sk);
 325         const int dccp_header_size = sizeof(struct dccp_hdr) +
 326                                      sizeof(struct dccp_hdr_ext) +
 327                                      sizeof(struct dccp_hdr_reset);
 328         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 329                                                dccp_header_size, 1,
 330                                            GFP_ATOMIC);
 331         if (skb == NULL)
 332                 return NULL;
 333
 334         /* Reserve space for headers. */
 335         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 336
 337         skb->dst = dst_clone(dst);
 338         skb->csum = 0;
 339
 340         dccp_inc_seqno(&dp->dccps_gss);
 341
 342         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 343         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 344         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 345         dccp_insert_options(sk, skb);
 346
 347         skb->h.raw = skb_push(skb, dccp_header_size);
 348
 349         dh = dccp_hdr(skb);
 350         memset(dh, 0, dccp_header_size);
 351
 352         dh->dccph_sport = inet_sk(sk)->sport;
 353         dh->dccph_dport = inet_sk(sk)->dport;
 354         dh->dccph_doff  = (dccp_header_size +
 355                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 356         dh->dccph_type  = DCCP_PKT_RESET;
 357         dh->dccph_x     = 1;
 358         dccp_hdr_set_seq(dh, dp->dccps_gss);
 359         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 360
 361         dccp_hdr_reset(skb)->dccph_reset_code = code;
 362
 363         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 364                                               inet_sk(sk)->daddr);
 365
 366         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 367         return skb;
 368 }
 369
 370 /*
 371  * Do all connect socket setups that can be done AF independent.
 372  */
 373 static inline void dccp_connect_init(struct sock *sk)
 374 {
 375         struct dst_entry *dst = __sk_dst_get(sk);
 376         struct inet_connection_sock *icsk = inet_csk(sk);
 377
 378         sk->sk_err = 0;
 379         sock_reset_flag(sk, SOCK_DONE);
 380
 381         dccp_sync_mss(sk, dst_mtu(dst));
 382
 383         /*
 384          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 385          * something like dccp_inc_seq
 386          */
 387
 388         icsk->icsk_retransmits = 0;
 389 }
 390
 391 int dccp_connect(struct sock *sk)
 392 {
 393         struct sk_buff *skb;
 394         struct inet_connection_sock *icsk = inet_csk(sk);
 395
 396         dccp_connect_init(sk);
 397
 398         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 399         if (unlikely(skb == NULL))
 400                 return -ENOBUFS;
 401
 402         /* Reserve space for headers. */
 403         skb_reserve(skb, MAX_DCCP_HEADER);
 404
 405         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 406         skb->csum = 0;
 407
 408         dccp_skb_entail(sk, skb);
 409         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 410         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 411
 412         /* Timer for repeating the REQUEST until an answer. */
 413         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 414                                   icsk->icsk_rto, DCCP_RTO_MAX);
 415         return 0;
 416 }
 417
 418 void dccp_send_ack(struct sock *sk)
 419 {
 420         /* If we have been reset, we may not send again. */
 421         if (sk->sk_state != DCCP_CLOSED) {
 422                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 423
 424                 if (skb == NULL) {
 425                         inet_csk_schedule_ack(sk);
 426                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 427                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 428                                                   TCP_DELACK_MAX,
 429                                                   DCCP_RTO_MAX);
 430                         return;
 431                 }
 432
 433                 /* Reserve space for headers */
 434                 skb_reserve(skb, MAX_DCCP_HEADER);
 435                 skb->csum = 0;
 436                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 437                 dccp_transmit_skb(sk, skb);
 438         }
 439 }
 440
 441 EXPORT_SYMBOL_GPL(dccp_send_ack);
 442
 443 void dccp_send_delayed_ack(struct sock *sk)
 444 {
 445         struct inet_connection_sock *icsk = inet_csk(sk);
 446         /*
 447          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 448          * with using 2s, and active senders also piggyback the ACK into a
 449          * DATAACK packet, so this is really for quiescent senders.
 450          */
 451         unsigned long timeout = jiffies + 2 * HZ;
 452
 453         /* Use new timeout only if there wasn't a older one earlier. */
 454         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 455                 /* If delack timer was blocked or is about to expire,
 456                  * send ACK now.
 457                  *
 458                  * FIXME: check the "about to expire" part
 459                  */
 460                 if (icsk->icsk_ack.blocked) {
 461                         dccp_send_ack(sk);
 462                         return;
 463                 }
 464
 465                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 466                         timeout = icsk->icsk_ack.timeout;
 467         }
 468         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 469         icsk->icsk_ack.timeout = timeout;
 470         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 471 }
 472
 473 void dccp_send_sync(struct sock *sk, const u64 seq,
 474                     const enum dccp_pkt_type pkt_type)
 475 {
 476         /*
 477          * We are not putting this on the write queue, so
 478          * dccp_transmit_skb() will set the ownership to this
 479          * sock.
 480          */
 481         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 482
 483         if (skb == NULL)
 484                 /* FIXME: how to make sure the sync is sent? */
 485                 return;
 486
 487         /* Reserve space for headers and prepare control bits. */
 488         skb_reserve(skb, MAX_DCCP_HEADER);
 489         skb->csum = 0;
 490         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 491         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 492
 493         dccp_transmit_skb(sk, skb);
 494 }
 495
 496 /*
 497  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 498  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 499  * any circumstances.
 500  */
 501 void dccp_send_close(struct sock *sk, const int active)
 502 {
 503         struct dccp_sock *dp = dccp_sk(sk);
 504         struct sk_buff *skb;
 505         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 506
 507         skb = alloc_skb(sk->sk_prot->max_header, prio);
 508         if (skb == NULL)
 509                 return;
 510
 511         /* Reserve space for headers and prepare control bits. */
 512         skb_reserve(skb, sk->sk_prot->max_header);
 513         skb->csum = 0;
 514         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 515                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 516
 517         if (active) {
 518                 dccp_skb_entail(sk, skb);
 519                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 520         } else
 521                 dccp_transmit_skb(sk, skb);
 522 }