err.no 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/skbuff.h>
  16
  17 #include <net/sock.h>
  18
  19 #include "ccid.h"
  20 #include "dccp.h"
  21
  22 static inline void dccp_event_ack_sent(struct sock *sk)
  23 {
  24         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  25 }
  26
  27 /*
  28  * All SKB's seen here are completely headerless. It is our
  29  * job to build the DCCP header, and pass the packet down to
  30  * IP so it can do the same plus pass the packet off to the
  31  * device.
  32  */
  33 int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  34 {
  35         if (likely(skb != NULL)) {
  36                 const struct inet_sock *inet = inet_sk(sk);
  37                 struct dccp_sock *dp = dccp_sk(sk);
  38                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  39                 struct dccp_hdr *dh;
  40                 /* XXX For now we're using only 48 bits sequence numbers */
  41                 const int dccp_header_size = sizeof(*dh) +
  42                                              sizeof(struct dccp_hdr_ext) +
  43                                           dccp_packet_hdr_len(dcb->dccpd_type);
  44                 int err, set_ack = 1;
  45                 u64 ackno = dp->dccps_gsr;
  46
  47                 dccp_inc_seqno(&dp->dccps_gss);
  48
  49                 switch (dcb->dccpd_type) {
  50                 case DCCP_PKT_DATA:
  51                         set_ack = 0;
  52                         break;
  53                 case DCCP_PKT_SYNC:
  54                 case DCCP_PKT_SYNCACK:
  55                         ackno = dcb->dccpd_seq;
  56                         break;
  57                 }
  58
  59                 dcb->dccpd_seq = dp->dccps_gss;
  60                 dccp_insert_options(sk, skb);
  61
  62                 skb->h.raw = skb_push(skb, dccp_header_size);
  63                 dh = dccp_hdr(skb);
  64                 /*
  65                  * Data packets are not cloned as they are never retransmitted
  66                  */
  67                 if (skb_cloned(skb))
  68                         skb_set_owner_w(skb, sk);
  69
  70                 /* Build DCCP header and checksum it. */
  71                 memset(dh, 0, dccp_header_size);
  72                 dh->dccph_type  = dcb->dccpd_type;
  73                 dh->dccph_sport = inet->sport;
  74                 dh->dccph_dport = inet->dport;
  75                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  76                 dh->dccph_ccval = dcb->dccpd_ccval;
  77                 /* XXX For now we're using only 48 bits sequence numbers */
  78                 dh->dccph_x     = 1;
  79
  80                 dp->dccps_awh = dp->dccps_gss;
  81                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  82                 if (set_ack)
  83                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  84
  85                 switch (dcb->dccpd_type) {
  86                 case DCCP_PKT_REQUEST:
  87                         dccp_hdr_request(skb)->dccph_req_service =
  88                                                         dcb->dccpd_service;
  89                         break;
  90                 case DCCP_PKT_RESET:
  91                         dccp_hdr_reset(skb)->dccph_reset_code =
  92                                                         dcb->dccpd_reset_code;
  93                         break;
  94                 }
  95
  96                 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
  97                                                       inet->daddr);
  98
  99                 if (set_ack)
 100                         dccp_event_ack_sent(sk);
 101
 102                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 103
 104                 err = ip_queue_xmit(skb, 0);
 105                 if (err <= 0)
 106                         return err;
 107
 108                 /* NET_XMIT_CN is special. It does not guarantee,
 109                  * that this packet is lost. It tells that device
 110                  * is about to start to drop packets or already
 111                  * drops some packets of the same priority and
 112                  * invokes us to send less aggressively.
 113                  */
 114                 return err == NET_XMIT_CN ? 0 : err;
 115         }
 116         return -ENOBUFS;
 117 }
 118
 119 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 120 {
 121         struct dccp_sock *dp = dccp_sk(sk);
 122         int mss_now;
 123
 124         /*
 125          * FIXME: we really should be using the af_specific thing to support
 126          *        IPv6.
 127          * mss_now = pmtu - tp->af_specific->net_header_len -
 128          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 129          */
 130         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
 131                   sizeof(struct dccp_hdr_ext);
 132
 133         /* Now subtract optional transport overhead */
 134         mss_now -= dp->dccps_ext_header_len;
 135
 136         /*
 137          * FIXME: this should come from the CCID infrastructure, where, say,
 138          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 139          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 140          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 141          * make it a multiple of 4
 142          */
 143
 144         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 145
 146         /* And store cached results */
 147         dp->dccps_pmtu_cookie = pmtu;
 148         dp->dccps_mss_cache = mss_now;
 149
 150         return mss_now;
 151 }
 152
 153 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, const int len)
 154 {
 155         const struct dccp_sock *dp = dccp_sk(sk);
 156         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb, len);
 157
 158         if (err == 0) {
 159                 const struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts;
 160                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 161
 162                 if (sk->sk_state == DCCP_PARTOPEN) {
 163                         /* See 8.1.5.  Handshake Completion */
 164                         inet_csk_schedule_ack(sk);
 165                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 166                                                   inet_csk(sk)->icsk_rto,
 167                                                   DCCP_RTO_MAX);
 168                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 169                         /*
 170                          * FIXME: we really should have a
 171                          * dccps_ack_pending or use icsk.
 172                          */
 173                 } else if (inet_csk_ack_scheduled(sk) ||
 174                            dp->dccps_timestamp_echo != 0 ||
 175                            (dp->dccps_options.dccpo_send_ack_vector &&
 176                             ap->dccpap_buf_ackno != DCCP_MAX_SEQNO + 1 &&
 177                             ap->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1))
 178                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 179                 else
 180                         dcb->dccpd_type = DCCP_PKT_DATA;
 181
 182                 err = dccp_transmit_skb(sk, skb);
 183                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 184         }
 185
 186         return err;
 187 }
 188
 189 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 190 {
 191         if (inet_sk_rebuild_header(sk) != 0)
 192                 return -EHOSTUNREACH; /* Routing failure or similar. */
 193
 194         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 195                                       pskb_copy(skb, GFP_ATOMIC):
 196                                       skb_clone(skb, GFP_ATOMIC)));
 197 }
 198
 199 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 200                                    struct request_sock *req)
 201 {
 202         struct dccp_hdr *dh;
 203         const int dccp_header_size = sizeof(struct dccp_hdr) +
 204                                      sizeof(struct dccp_hdr_ext) +
 205                                      sizeof(struct dccp_hdr_response);
 206         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 207                                                dccp_header_size, 1,
 208                                            GFP_ATOMIC);
 209         if (skb == NULL)
 210                 return NULL;
 211
 212         /* Reserve space for headers. */
 213         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 214
 215         skb->dst = dst_clone(dst);
 216         skb->csum = 0;
 217
 218         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 219         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_rsk(req)->dreq_iss;
 220         dccp_insert_options(sk, skb);
 221
 222         skb->h.raw = skb_push(skb, dccp_header_size);
 223
 224         dh = dccp_hdr(skb);
 225         memset(dh, 0, dccp_header_size);
 226
 227         dh->dccph_sport = inet_sk(sk)->sport;
 228         dh->dccph_dport = inet_rsk(req)->rmt_port;
 229         dh->dccph_doff  = (dccp_header_size +
 230                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 231         dh->dccph_type  = DCCP_PKT_RESPONSE;
 232         dh->dccph_x     = 1;
 233         dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss);
 234         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr);
 235
 236         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 237                                               inet_rsk(req)->rmt_addr);
 238
 239         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 240         return skb;
 241 }
 242
 243 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 244                                 const enum dccp_reset_codes code)
 245
 246 {
 247         struct dccp_hdr *dh;
 248         struct dccp_sock *dp = dccp_sk(sk);
 249         const int dccp_header_size = sizeof(struct dccp_hdr) +
 250                                      sizeof(struct dccp_hdr_ext) +
 251                                      sizeof(struct dccp_hdr_reset);
 252         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 253                                                dccp_header_size, 1,
 254                                            GFP_ATOMIC);
 255         if (skb == NULL)
 256                 return NULL;
 257
 258         /* Reserve space for headers. */
 259         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 260
 261         skb->dst = dst_clone(dst);
 262         skb->csum = 0;
 263
 264         dccp_inc_seqno(&dp->dccps_gss);
 265
 266         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 267         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 268         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 269         dccp_insert_options(sk, skb);
 270
 271         skb->h.raw = skb_push(skb, dccp_header_size);
 272
 273         dh = dccp_hdr(skb);
 274         memset(dh, 0, dccp_header_size);
 275
 276         dh->dccph_sport = inet_sk(sk)->sport;
 277         dh->dccph_dport = inet_sk(sk)->dport;
 278         dh->dccph_doff  = (dccp_header_size +
 279                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 280         dh->dccph_type  = DCCP_PKT_RESET;
 281         dh->dccph_x     = 1;
 282         dccp_hdr_set_seq(dh, dp->dccps_gss);
 283         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 284
 285         dccp_hdr_reset(skb)->dccph_reset_code = code;
 286
 287         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 288                                               inet_sk(sk)->daddr);
 289
 290         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 291         return skb;
 292 }
 293
 294 /*
 295  * Do all connect socket setups that can be done AF independent.
 296  */
 297 static inline void dccp_connect_init(struct sock *sk)
 298 {
 299         struct dst_entry *dst = __sk_dst_get(sk);
 300         struct inet_connection_sock *icsk = inet_csk(sk);
 301
 302         sk->sk_err = 0;
 303         sock_reset_flag(sk, SOCK_DONE);
 304
 305         dccp_sync_mss(sk, dst_mtu(dst));
 306
 307         /*
 308          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 309          * something like dccp_inc_seq
 310          */
 311
 312         icsk->icsk_retransmits = 0;
 313 }
 314
 315 int dccp_connect(struct sock *sk)
 316 {
 317         struct sk_buff *skb;
 318         struct inet_connection_sock *icsk = inet_csk(sk);
 319
 320         dccp_connect_init(sk);
 321
 322         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 323         if (unlikely(skb == NULL))
 324                 return -ENOBUFS;
 325
 326         /* Reserve space for headers. */
 327         skb_reserve(skb, MAX_DCCP_HEADER);
 328
 329         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 330         /* FIXME: set service to something meaningful, coming
 331          * from userspace*/
 332         DCCP_SKB_CB(skb)->dccpd_service = 0;
 333         skb->csum = 0;
 334         skb_set_owner_w(skb, sk);
 335
 336         BUG_TRAP(sk->sk_send_head == NULL);
 337         sk->sk_send_head = skb;
 338         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 339         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 340
 341         /* Timer for repeating the REQUEST until an answer. */
 342         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 343                                   icsk->icsk_rto, DCCP_RTO_MAX);
 344         return 0;
 345 }
 346
 347 void dccp_send_ack(struct sock *sk)
 348 {
 349         /* If we have been reset, we may not send again. */
 350         if (sk->sk_state != DCCP_CLOSED) {
 351                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 352
 353                 if (skb == NULL) {
 354                         inet_csk_schedule_ack(sk);
 355                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 356                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 357                                                   TCP_DELACK_MAX,
 358                                                   DCCP_RTO_MAX);
 359                         return;
 360                 }
 361
 362                 /* Reserve space for headers */
 363                 skb_reserve(skb, MAX_DCCP_HEADER);
 364                 skb->csum = 0;
 365                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 366                 skb_set_owner_w(skb, sk);
 367                 dccp_transmit_skb(sk, skb);
 368         }
 369 }
 370
 371 EXPORT_SYMBOL_GPL(dccp_send_ack);
 372
 373 void dccp_send_delayed_ack(struct sock *sk)
 374 {
 375         struct inet_connection_sock *icsk = inet_csk(sk);
 376         /*
 377          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 378          * with using 2s, and active senders also piggyback the ACK into a
 379          * DATAACK packet, so this is really for quiescent senders.
 380          */
 381         unsigned long timeout = jiffies + 2 * HZ;
 382
 383         /* Use new timeout only if there wasn't a older one earlier. */
 384         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 385                 /* If delack timer was blocked or is about to expire,
 386                  * send ACK now.
 387                  *
 388                  * FIXME: check the "about to expire" part
 389                  */
 390                 if (icsk->icsk_ack.blocked) {
 391                         dccp_send_ack(sk);
 392                         return;
 393                 }
 394
 395                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 396                         timeout = icsk->icsk_ack.timeout;
 397         }
 398         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 399         icsk->icsk_ack.timeout = timeout;
 400         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 401 }
 402
 403 void dccp_send_sync(struct sock *sk, const u64 seq,
 404                     const enum dccp_pkt_type pkt_type)
 405 {
 406         /*
 407          * We are not putting this on the write queue, so
 408          * dccp_transmit_skb() will set the ownership to this
 409          * sock.
 410          */
 411         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 412
 413         if (skb == NULL)
 414                 /* FIXME: how to make sure the sync is sent? */
 415                 return;
 416
 417         /* Reserve space for headers and prepare control bits. */
 418         skb_reserve(skb, MAX_DCCP_HEADER);
 419         skb->csum = 0;
 420         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 421         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 422
 423         skb_set_owner_w(skb, sk);
 424         dccp_transmit_skb(sk, skb);
 425 }
 426
 427 /*
 428  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 429  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 430  * any circumstances.
 431  */
 432 void dccp_send_close(struct sock *sk, const int active)
 433 {
 434         struct dccp_sock *dp = dccp_sk(sk);
 435         struct sk_buff *skb;
 436         const unsigned int prio = active ? GFP_KERNEL : GFP_ATOMIC;
 437
 438         skb = alloc_skb(sk->sk_prot->max_header, prio);
 439         if (skb == NULL)
 440                 return;
 441
 442         /* Reserve space for headers and prepare control bits. */
 443         skb_reserve(skb, sk->sk_prot->max_header);
 444         skb->csum = 0;
 445         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 446                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 447
 448         skb_set_owner_w(skb, sk);
 449         if (active) {
 450                 BUG_TRAP(sk->sk_send_head == NULL);
 451                 sk->sk_send_head = skb;
 452                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 453         } else
 454                 dccp_transmit_skb(sk, skb);
 455
 456         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 457         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 458 }