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