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/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                 /*
  48                  * FIXME: study DCCP_PKT_SYNC[ACK] to see what is the right
  49                  * thing to do here...
  50                  */
  51                 dccp_inc_seqno(&dp->dccps_gss);
  52
  53                 dcb->dccpd_seq = dp->dccps_gss;
  54                 dccp_insert_options(sk, skb);
  55
  56                 switch (dcb->dccpd_type) {
  57                 case DCCP_PKT_DATA:
  58                         set_ack = 0;
  59                         break;
  60                 case DCCP_PKT_SYNC:
  61                 case DCCP_PKT_SYNCACK:
  62                         ackno = dcb->dccpd_seq;
  63                         break;
  64                 }
  65
  66                 skb->h.raw = skb_push(skb, dccp_header_size);
  67                 dh = dccp_hdr(skb);
  68                 /*
  69                  * Data packets are not cloned as they are never retransmitted
  70                  */
  71                 if (skb_cloned(skb))
  72                         skb_set_owner_w(skb, sk);
  73
  74                 /* Build DCCP header and checksum it. */
  75                 memset(dh, 0, dccp_header_size);
  76                 dh->dccph_type  = dcb->dccpd_type;
  77                 dh->dccph_sport = inet->sport;
  78                 dh->dccph_dport = inet->dport;
  79                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  80                 dh->dccph_ccval = dcb->dccpd_ccval;
  81                 /* XXX For now we're using only 48 bits sequence numbers */
  82                 dh->dccph_x     = 1;
  83
  84                 dp->dccps_awh = dp->dccps_gss;
  85                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  86                 if (set_ack)
  87                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  88
  89                 switch (dcb->dccpd_type) {
  90                 case DCCP_PKT_REQUEST:
  91                         dccp_hdr_request(skb)->dccph_req_service =
  92                                                         dcb->dccpd_service;
  93                         break;
  94                 case DCCP_PKT_RESET:
  95                         dccp_hdr_reset(skb)->dccph_reset_code =
  96                                                         dcb->dccpd_reset_code;
  97                         break;
  98                 }
  99
 100                 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
 101                                                       inet->daddr);
 102
 103                 if (dcb->dccpd_type == DCCP_PKT_ACK ||
 104                     dcb->dccpd_type == DCCP_PKT_DATAACK)
 105                         dccp_event_ack_sent(sk);
 106
 107                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 108
 109                 err = ip_queue_xmit(skb, 0);
 110                 if (err <= 0)
 111                         return err;
 112
 113                 /* NET_XMIT_CN is special. It does not guarantee,
 114                  * that this packet is lost. It tells that device
 115                  * is about to start to drop packets or already
 116                  * drops some packets of the same priority and
 117                  * invokes us to send less aggressively.
 118                  */
 119                 return err == NET_XMIT_CN ? 0 : err;
 120         }
 121         return -ENOBUFS;
 122 }
 123
 124 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 125 {
 126         struct dccp_sock *dp = dccp_sk(sk);
 127         int mss_now;
 128
 129         /*
 130          * FIXME: we really should be using the af_specific thing to support
 131          *        IPv6.
 132          * mss_now = pmtu - tp->af_specific->net_header_len -
 133          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 134          */
 135         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
 136                   sizeof(struct dccp_hdr_ext);
 137
 138         /* Now subtract optional transport overhead */
 139         mss_now -= dp->dccps_ext_header_len;
 140
 141         /*
 142          * FIXME: this should come from the CCID infrastructure, where, say,
 143          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 144          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 145          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 146          * make it a multiple of 4
 147          */
 148
 149         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 150
 151         /* And store cached results */
 152         dp->dccps_pmtu_cookie = pmtu;
 153         dp->dccps_mss_cache = mss_now;
 154
 155         return mss_now;
 156 }
 157
 158 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, const int len)
 159 {
 160         const struct dccp_sock *dp = dccp_sk(sk);
 161         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb, len);
 162
 163         if (err == 0) {
 164                 const struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts;
 165                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 166
 167                 if (sk->sk_state == DCCP_PARTOPEN) {
 168                         /* See 8.1.5.  Handshake Completion */
 169                         inet_csk_schedule_ack(sk);
 170                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 171                                                   inet_csk(sk)->icsk_rto,
 172                                                   DCCP_RTO_MAX);
 173                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 174                         /*
 175                          * FIXME: we really should have a
 176                          * dccps_ack_pending or use icsk.
 177                          */
 178                 } else if (inet_csk_ack_scheduled(sk) ||
 179                            (dp->dccps_options.dccpo_send_ack_vector &&
 180                             ap->dccpap_buf_ackno != DCCP_MAX_SEQNO + 1 &&
 181                             ap->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1))
 182                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 183                 else
 184                         dcb->dccpd_type = DCCP_PKT_DATA;
 185
 186                 err = dccp_transmit_skb(sk, skb);
 187                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 188         }
 189
 190         return err;
 191 }
 192
 193 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 194 {
 195         if (inet_sk_rebuild_header(sk) != 0)
 196                 return -EHOSTUNREACH; /* Routing failure or similar. */
 197
 198         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 199                                       pskb_copy(skb, GFP_ATOMIC):
 200                                       skb_clone(skb, GFP_ATOMIC)));
 201 }
 202
 203 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 204                                    struct request_sock *req)
 205 {
 206         struct dccp_hdr *dh;
 207         const int dccp_header_size = sizeof(struct dccp_hdr) +
 208                                      sizeof(struct dccp_hdr_ext) +
 209                                      sizeof(struct dccp_hdr_response);
 210         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 211                                                dccp_header_size, 1,
 212                                            GFP_ATOMIC);
 213         if (skb == NULL)
 214                 return NULL;
 215
 216         /* Reserve space for headers. */
 217         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 218
 219         skb->dst = dst_clone(dst);
 220         skb->csum = 0;
 221
 222         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 223         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_rsk(req)->dreq_iss;
 224         dccp_insert_options(sk, skb);
 225
 226         skb->h.raw = skb_push(skb, dccp_header_size);
 227
 228         dh = dccp_hdr(skb);
 229         memset(dh, 0, dccp_header_size);
 230
 231         dh->dccph_sport = inet_sk(sk)->sport;
 232         dh->dccph_dport = inet_rsk(req)->rmt_port;
 233         dh->dccph_doff  = (dccp_header_size +
 234                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 235         dh->dccph_type  = DCCP_PKT_RESPONSE;
 236         dh->dccph_x     = 1;
 237         dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss);
 238         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr);
 239
 240         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 241                                               inet_rsk(req)->rmt_addr);
 242
 243         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 244         return skb;
 245 }
 246
 247 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 248                                 const enum dccp_reset_codes code)
 249
 250 {
 251         struct dccp_hdr *dh;
 252         struct dccp_sock *dp = dccp_sk(sk);
 253         const int dccp_header_size = sizeof(struct dccp_hdr) +
 254                                      sizeof(struct dccp_hdr_ext) +
 255                                      sizeof(struct dccp_hdr_reset);
 256         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 257                                                dccp_header_size, 1,
 258                                            GFP_ATOMIC);
 259         if (skb == NULL)
 260                 return NULL;
 261
 262         /* Reserve space for headers. */
 263         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 264
 265         skb->dst = dst_clone(dst);
 266         skb->csum = 0;
 267
 268         dccp_inc_seqno(&dp->dccps_gss);
 269
 270         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 271         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 272         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 273         dccp_insert_options(sk, skb);
 274
 275         skb->h.raw = skb_push(skb, dccp_header_size);
 276
 277         dh = dccp_hdr(skb);
 278         memset(dh, 0, dccp_header_size);
 279
 280         dh->dccph_sport = inet_sk(sk)->sport;
 281         dh->dccph_dport = inet_sk(sk)->dport;
 282         dh->dccph_doff  = (dccp_header_size +
 283                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 284         dh->dccph_type  = DCCP_PKT_RESET;
 285         dh->dccph_x     = 1;
 286         dccp_hdr_set_seq(dh, dp->dccps_gss);
 287         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 288
 289         dccp_hdr_reset(skb)->dccph_reset_code = code;
 290
 291         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 292                                               inet_sk(sk)->daddr);
 293
 294         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 295         return skb;
 296 }
 297
 298 /*
 299  * Do all connect socket setups that can be done AF independent.
 300  */
 301 static inline void dccp_connect_init(struct sock *sk)
 302 {
 303         struct dst_entry *dst = __sk_dst_get(sk);
 304         struct inet_connection_sock *icsk = inet_csk(sk);
 305
 306         sk->sk_err = 0;
 307         sock_reset_flag(sk, SOCK_DONE);
 308
 309         dccp_sync_mss(sk, dst_mtu(dst));
 310
 311         /*
 312          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 313          * something like dccp_inc_seq
 314          */
 315
 316         icsk->icsk_retransmits = 0;
 317 }
 318
 319 int dccp_connect(struct sock *sk)
 320 {
 321         struct sk_buff *skb;
 322         struct inet_connection_sock *icsk = inet_csk(sk);
 323
 324         dccp_connect_init(sk);
 325
 326         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 327         if (unlikely(skb == NULL))
 328                 return -ENOBUFS;
 329
 330         /* Reserve space for headers. */
 331         skb_reserve(skb, MAX_DCCP_HEADER);
 332
 333         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 334         /* FIXME: set service to something meaningful, coming
 335          * from userspace*/
 336         DCCP_SKB_CB(skb)->dccpd_service = 0;
 337         skb->csum = 0;
 338         skb_set_owner_w(skb, sk);
 339
 340         BUG_TRAP(sk->sk_send_head == NULL);
 341         sk->sk_send_head = skb;
 342         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 343         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 344
 345         /* Timer for repeating the REQUEST until an answer. */
 346         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 347                                   icsk->icsk_rto, DCCP_RTO_MAX);
 348         return 0;
 349 }
 350
 351 void dccp_send_ack(struct sock *sk)
 352 {
 353         /* If we have been reset, we may not send again. */
 354         if (sk->sk_state != DCCP_CLOSED) {
 355                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 356
 357                 if (skb == NULL) {
 358                         inet_csk_schedule_ack(sk);
 359                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 360                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 361                                                   TCP_DELACK_MAX,
 362                                                   DCCP_RTO_MAX);
 363                         return;
 364                 }
 365
 366                 /* Reserve space for headers */
 367                 skb_reserve(skb, MAX_DCCP_HEADER);
 368                 skb->csum = 0;
 369                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 370                 skb_set_owner_w(skb, sk);
 371                 dccp_transmit_skb(sk, skb);
 372         }
 373 }
 374
 375 EXPORT_SYMBOL_GPL(dccp_send_ack);
 376
 377 void dccp_send_delayed_ack(struct sock *sk)
 378 {
 379         struct inet_connection_sock *icsk = inet_csk(sk);
 380         /*
 381          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 382          * with using 2s, and active senders also piggyback the ACK into a
 383          * DATAACK packet, so this is really for quiescent senders.
 384          */
 385         unsigned long timeout = jiffies + 2 * HZ;
 386
 387         /* Use new timeout only if there wasn't a older one earlier. */
 388         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 389                 /* If delack timer was blocked or is about to expire,
 390                  * send ACK now.
 391                  *
 392                  * FIXME: check the "about to expire" part
 393                  */
 394                 if (icsk->icsk_ack.blocked) {
 395                         dccp_send_ack(sk);
 396                         return;
 397                 }
 398
 399                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 400                         timeout = icsk->icsk_ack.timeout;
 401         }
 402         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 403         icsk->icsk_ack.timeout = timeout;
 404         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 405 }
 406
 407 void dccp_send_sync(struct sock *sk, const u64 seq,
 408                     const enum dccp_pkt_type pkt_type)
 409 {
 410         /*
 411          * We are not putting this on the write queue, so
 412          * dccp_transmit_skb() will set the ownership to this
 413          * sock.
 414          */
 415         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 416
 417         if (skb == NULL)
 418                 /* FIXME: how to make sure the sync is sent? */
 419                 return;
 420
 421         /* Reserve space for headers and prepare control bits. */
 422         skb_reserve(skb, MAX_DCCP_HEADER);
 423         skb->csum = 0;
 424         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 425         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 426
 427         skb_set_owner_w(skb, sk);
 428         dccp_transmit_skb(sk, skb);
 429 }
 430
 431 /*
 432  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 433  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 434  * any circumstances.
 435  */
 436 void dccp_send_close(struct sock *sk)
 437 {
 438         struct dccp_sock *dp = dccp_sk(sk);
 439         struct sk_buff *skb;
 440
 441         /* Socket is locked, keep trying until memory is available. */
 442         for (;;) {
 443                 skb = alloc_skb(sk->sk_prot->max_header, GFP_KERNEL);
 444                 if (skb != NULL)
 445                         break;
 446                 yield();
 447         }
 448
 449         /* Reserve space for headers and prepare control bits. */
 450         skb_reserve(skb, sk->sk_prot->max_header);
 451         skb->csum = 0;
 452         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 453                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 454
 455         skb_set_owner_w(skb, sk);
 456         dccp_transmit_skb(sk, skb);
 457
 458         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 459         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 460 }