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_options.dccpo_send_ack_vector &&
 175                             ap->dccpap_buf_ackno != DCCP_MAX_SEQNO + 1 &&
 176                             ap->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1))
 177                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 178                 else
 179                         dcb->dccpd_type = DCCP_PKT_DATA;
 180
 181                 err = dccp_transmit_skb(sk, skb);
 182                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 183         }
 184
 185         return err;
 186 }
 187
 188 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 189 {
 190         if (inet_sk_rebuild_header(sk) != 0)
 191                 return -EHOSTUNREACH; /* Routing failure or similar. */
 192
 193         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 194                                       pskb_copy(skb, GFP_ATOMIC):
 195                                       skb_clone(skb, GFP_ATOMIC)));
 196 }
 197
 198 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 199                                    struct request_sock *req)
 200 {
 201         struct dccp_hdr *dh;
 202         const int dccp_header_size = sizeof(struct dccp_hdr) +
 203                                      sizeof(struct dccp_hdr_ext) +
 204                                      sizeof(struct dccp_hdr_response);
 205         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 206                                                dccp_header_size, 1,
 207                                            GFP_ATOMIC);
 208         if (skb == NULL)
 209                 return NULL;
 210
 211         /* Reserve space for headers. */
 212         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 213
 214         skb->dst = dst_clone(dst);
 215         skb->csum = 0;
 216
 217         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 218         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_rsk(req)->dreq_iss;
 219         dccp_insert_options(sk, skb);
 220
 221         skb->h.raw = skb_push(skb, dccp_header_size);
 222
 223         dh = dccp_hdr(skb);
 224         memset(dh, 0, dccp_header_size);
 225
 226         dh->dccph_sport = inet_sk(sk)->sport;
 227         dh->dccph_dport = inet_rsk(req)->rmt_port;
 228         dh->dccph_doff  = (dccp_header_size +
 229                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 230         dh->dccph_type  = DCCP_PKT_RESPONSE;
 231         dh->dccph_x     = 1;
 232         dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss);
 233         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr);
 234
 235         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 236                                               inet_rsk(req)->rmt_addr);
 237
 238         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 239         return skb;
 240 }
 241
 242 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 243                                 const enum dccp_reset_codes code)
 244
 245 {
 246         struct dccp_hdr *dh;
 247         struct dccp_sock *dp = dccp_sk(sk);
 248         const int dccp_header_size = sizeof(struct dccp_hdr) +
 249                                      sizeof(struct dccp_hdr_ext) +
 250                                      sizeof(struct dccp_hdr_reset);
 251         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 252                                                dccp_header_size, 1,
 253                                            GFP_ATOMIC);
 254         if (skb == NULL)
 255                 return NULL;
 256
 257         /* Reserve space for headers. */
 258         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 259
 260         skb->dst = dst_clone(dst);
 261         skb->csum = 0;
 262
 263         dccp_inc_seqno(&dp->dccps_gss);
 264
 265         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 266         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 267         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 268         dccp_insert_options(sk, skb);
 269
 270         skb->h.raw = skb_push(skb, dccp_header_size);
 271
 272         dh = dccp_hdr(skb);
 273         memset(dh, 0, dccp_header_size);
 274
 275         dh->dccph_sport = inet_sk(sk)->sport;
 276         dh->dccph_dport = inet_sk(sk)->dport;
 277         dh->dccph_doff  = (dccp_header_size +
 278                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 279         dh->dccph_type  = DCCP_PKT_RESET;
 280         dh->dccph_x     = 1;
 281         dccp_hdr_set_seq(dh, dp->dccps_gss);
 282         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 283
 284         dccp_hdr_reset(skb)->dccph_reset_code = code;
 285
 286         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 287                                               inet_sk(sk)->daddr);
 288
 289         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 290         return skb;
 291 }
 292
 293 /*
 294  * Do all connect socket setups that can be done AF independent.
 295  */
 296 static inline void dccp_connect_init(struct sock *sk)
 297 {
 298         struct dst_entry *dst = __sk_dst_get(sk);
 299         struct inet_connection_sock *icsk = inet_csk(sk);
 300
 301         sk->sk_err = 0;
 302         sock_reset_flag(sk, SOCK_DONE);
 303
 304         dccp_sync_mss(sk, dst_mtu(dst));
 305
 306         /*
 307          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 308          * something like dccp_inc_seq
 309          */
 310
 311         icsk->icsk_retransmits = 0;
 312 }
 313
 314 int dccp_connect(struct sock *sk)
 315 {
 316         struct sk_buff *skb;
 317         struct inet_connection_sock *icsk = inet_csk(sk);
 318
 319         dccp_connect_init(sk);
 320
 321         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 322         if (unlikely(skb == NULL))
 323                 return -ENOBUFS;
 324
 325         /* Reserve space for headers. */
 326         skb_reserve(skb, MAX_DCCP_HEADER);
 327
 328         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 329         /* FIXME: set service to something meaningful, coming
 330          * from userspace*/
 331         DCCP_SKB_CB(skb)->dccpd_service = 0;
 332         skb->csum = 0;
 333         skb_set_owner_w(skb, sk);
 334
 335         BUG_TRAP(sk->sk_send_head == NULL);
 336         sk->sk_send_head = skb;
 337         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 338         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 339
 340         /* Timer for repeating the REQUEST until an answer. */
 341         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 342                                   icsk->icsk_rto, DCCP_RTO_MAX);
 343         return 0;
 344 }
 345
 346 void dccp_send_ack(struct sock *sk)
 347 {
 348         /* If we have been reset, we may not send again. */
 349         if (sk->sk_state != DCCP_CLOSED) {
 350                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 351
 352                 if (skb == NULL) {
 353                         inet_csk_schedule_ack(sk);
 354                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 355                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 356                                                   TCP_DELACK_MAX,
 357                                                   DCCP_RTO_MAX);
 358                         return;
 359                 }
 360
 361                 /* Reserve space for headers */
 362                 skb_reserve(skb, MAX_DCCP_HEADER);
 363                 skb->csum = 0;
 364                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 365                 skb_set_owner_w(skb, sk);
 366                 dccp_transmit_skb(sk, skb);
 367         }
 368 }
 369
 370 EXPORT_SYMBOL_GPL(dccp_send_ack);
 371
 372 void dccp_send_delayed_ack(struct sock *sk)
 373 {
 374         struct inet_connection_sock *icsk = inet_csk(sk);
 375         /*
 376          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 377          * with using 2s, and active senders also piggyback the ACK into a
 378          * DATAACK packet, so this is really for quiescent senders.
 379          */
 380         unsigned long timeout = jiffies + 2 * HZ;
 381
 382         /* Use new timeout only if there wasn't a older one earlier. */
 383         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 384                 /* If delack timer was blocked or is about to expire,
 385                  * send ACK now.
 386                  *
 387                  * FIXME: check the "about to expire" part
 388                  */
 389                 if (icsk->icsk_ack.blocked) {
 390                         dccp_send_ack(sk);
 391                         return;
 392                 }
 393
 394                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 395                         timeout = icsk->icsk_ack.timeout;
 396         }
 397         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 398         icsk->icsk_ack.timeout = timeout;
 399         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 400 }
 401
 402 void dccp_send_sync(struct sock *sk, const u64 seq,
 403                     const enum dccp_pkt_type pkt_type)
 404 {
 405         /*
 406          * We are not putting this on the write queue, so
 407          * dccp_transmit_skb() will set the ownership to this
 408          * sock.
 409          */
 410         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 411
 412         if (skb == NULL)
 413                 /* FIXME: how to make sure the sync is sent? */
 414                 return;
 415
 416         /* Reserve space for headers and prepare control bits. */
 417         skb_reserve(skb, MAX_DCCP_HEADER);
 418         skb->csum = 0;
 419         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 420         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 421
 422         skb_set_owner_w(skb, sk);
 423         dccp_transmit_skb(sk, skb);
 424 }
 425
 426 /*
 427  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 428  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 429  * any circumstances.
 430  */
 431 void dccp_send_close(struct sock *sk, const int active)
 432 {
 433         struct dccp_sock *dp = dccp_sk(sk);
 434         struct sk_buff *skb;
 435         const unsigned int prio = active ? GFP_KERNEL : GFP_ATOMIC;
 436
 437         skb = alloc_skb(sk->sk_prot->max_header, prio);
 438         if (skb == NULL)
 439                 return;
 440
 441         /* Reserve space for headers and prepare control bits. */
 442         skb_reserve(skb, sk->sk_prot->max_header);
 443         skb->csum = 0;
 444         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 445                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 446
 447         skb_set_owner_w(skb, sk);
 448         if (active) {
 449                 BUG_TRAP(sk->sk_send_head == NULL);
 450                 sk->sk_send_head = skb;
 451                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 452         } else
 453                 dccp_transmit_skb(sk, skb);
 454
 455         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 456         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 457 }