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);
  97
  98                 if (dcb->dccpd_type == DCCP_PKT_ACK ||
  99                     dcb->dccpd_type == DCCP_PKT_DATAACK)
 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 IPv6.
 126          * mss_now = pmtu - tp->af_specific->net_header_len - sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 127          */
 128         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 129
 130         /* Now subtract optional transport overhead */
 131         mss_now -= dp->dccps_ext_header_len;
 132
 133         /*
 134          * FIXME: this should come from the CCID infrastructure, where, say,
 135          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 136          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 137          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 138          * make it a multiple of 4
 139          */
 140
 141         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 142
 143         /* And store cached results */
 144         dp->dccps_pmtu_cookie = pmtu;
 145         dp->dccps_mss_cache = mss_now;
 146
 147         return mss_now;
 148 }
 149
 150 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 151 {
 152         if (inet_sk_rebuild_header(sk) != 0)
 153                 return -EHOSTUNREACH; /* Routing failure or similar. */
 154
 155         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 156                                       pskb_copy(skb, GFP_ATOMIC):
 157                                       skb_clone(skb, GFP_ATOMIC)));
 158 }
 159
 160 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 161                                    struct request_sock *req)
 162 {
 163         struct dccp_hdr *dh;
 164         const int dccp_header_size = sizeof(struct dccp_hdr) +
 165                                      sizeof(struct dccp_hdr_ext) +
 166                                      sizeof(struct dccp_hdr_response);
 167         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 168                                                dccp_header_size, 1,
 169                                            GFP_ATOMIC);
 170         if (skb == NULL)
 171                 return NULL;
 172
 173         /* Reserve space for headers. */
 174         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 175
 176         skb->dst = dst_clone(dst);
 177         skb->csum = 0;
 178
 179         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 180         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_rsk(req)->dreq_iss;
 181         dccp_insert_options(sk, skb);
 182
 183         skb->h.raw = skb_push(skb, dccp_header_size);
 184
 185         dh = dccp_hdr(skb);
 186         memset(dh, 0, dccp_header_size);
 187
 188         dh->dccph_sport = inet_sk(sk)->sport;
 189         dh->dccph_dport = inet_rsk(req)->rmt_port;
 190         dh->dccph_doff  = (dccp_header_size + DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 191         dh->dccph_type  = DCCP_PKT_RESPONSE;
 192         dh->dccph_x     = 1;
 193         dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss);
 194         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr);
 195
 196         dh->dccph_checksum = dccp_v4_checksum(skb);
 197
 198         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 199         return skb;
 200 }
 201
 202 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 203                                 const enum dccp_reset_codes code)
 204
 205 {
 206         struct dccp_hdr *dh;
 207         struct dccp_sock *dp = dccp_sk(sk);
 208         const int dccp_header_size = sizeof(struct dccp_hdr) +
 209                                      sizeof(struct dccp_hdr_ext) +
 210                                      sizeof(struct dccp_hdr_reset);
 211         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 212                                                dccp_header_size, 1,
 213                                            GFP_ATOMIC);
 214         if (skb == NULL)
 215                 return NULL;
 216
 217         /* Reserve space for headers. */
 218         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 219
 220         skb->dst = dst_clone(dst);
 221         skb->csum = 0;
 222
 223         dccp_inc_seqno(&dp->dccps_gss);
 224
 225         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 226         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 227         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 228         dccp_insert_options(sk, skb);
 229
 230         skb->h.raw = skb_push(skb, dccp_header_size);
 231
 232         dh = dccp_hdr(skb);
 233         memset(dh, 0, dccp_header_size);
 234
 235         dh->dccph_sport = inet_sk(sk)->sport;
 236         dh->dccph_dport = inet_sk(sk)->dport;
 237         dh->dccph_doff  = (dccp_header_size + DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 238         dh->dccph_type  = DCCP_PKT_RESET;
 239         dh->dccph_x     = 1;
 240         dccp_hdr_set_seq(dh, dp->dccps_gss);
 241         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 242
 243         dccp_hdr_reset(skb)->dccph_reset_code = code;
 244
 245         dh->dccph_checksum = dccp_v4_checksum(skb);
 246
 247         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 248         return skb;
 249 }
 250
 251 /*
 252  * Do all connect socket setups that can be done AF independent.
 253  */
 254 static inline void dccp_connect_init(struct sock *sk)
 255 {
 256         struct dst_entry *dst = __sk_dst_get(sk);
 257         struct inet_connection_sock *icsk = inet_csk(sk);
 258
 259         sk->sk_err = 0;
 260         sock_reset_flag(sk, SOCK_DONE);
 261
 262         dccp_sync_mss(sk, dst_mtu(dst));
 263
 264         /*
 265          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 266          * something like dccp_inc_seq
 267          */
 268
 269         icsk->icsk_retransmits = 0;
 270 }
 271
 272 int dccp_connect(struct sock *sk)
 273 {
 274         struct sk_buff *skb;
 275         struct inet_connection_sock *icsk = inet_csk(sk);
 276
 277         dccp_connect_init(sk);
 278
 279         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 280         if (unlikely(skb == NULL))
 281                 return -ENOBUFS;
 282
 283         /* Reserve space for headers. */
 284         skb_reserve(skb, MAX_DCCP_HEADER);
 285
 286         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 287         /* FIXME: set service to something meaningful, coming
 288          * from userspace*/
 289         DCCP_SKB_CB(skb)->dccpd_service = 0;
 290         skb->csum = 0;
 291         skb_set_owner_w(skb, sk);
 292
 293         BUG_TRAP(sk->sk_send_head == NULL);
 294         sk->sk_send_head = skb;
 295         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 296         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 297
 298         /* Timer for repeating the REQUEST until an answer. */
 299         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
 300         return 0;
 301 }
 302
 303 void dccp_send_ack(struct sock *sk)
 304 {
 305         /* If we have been reset, we may not send again. */
 306         if (sk->sk_state != DCCP_CLOSED) {
 307                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 308
 309                 if (skb == NULL) {
 310                         inet_csk_schedule_ack(sk);
 311                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 312                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MAX, TCP_RTO_MAX);
 313                         return;
 314                 }
 315
 316                 /* Reserve space for headers */
 317                 skb_reserve(skb, MAX_DCCP_HEADER);
 318                 skb->csum = 0;
 319                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 320                 skb_set_owner_w(skb, sk);
 321                 dccp_transmit_skb(sk, skb);
 322         }
 323 }
 324
 325 EXPORT_SYMBOL_GPL(dccp_send_ack);
 326
 327 void dccp_send_delayed_ack(struct sock *sk)
 328 {
 329         struct inet_connection_sock *icsk = inet_csk(sk);
 330         /*
 331          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 332          * with using 2s, and active senders also piggyback the ACK into a
 333          * DATAACK packet, so this is really for quiescent senders.
 334          */
 335         unsigned long timeout = jiffies + 2 * HZ;
 336
 337         /* Use new timeout only if there wasn't a older one earlier. */
 338         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 339                 /* If delack timer was blocked or is about to expire,
 340                  * send ACK now.
 341                  *
 342                  * FIXME: check the "about to expire" part
 343                  */
 344                 if (icsk->icsk_ack.blocked) {
 345                         dccp_send_ack(sk);
 346                         return;
 347                 }
 348
 349                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 350                         timeout = icsk->icsk_ack.timeout;
 351         }
 352         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 353         icsk->icsk_ack.timeout = timeout;
 354         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 355 }
 356
 357 void dccp_send_sync(struct sock *sk, u64 seq)
 358 {
 359         /*
 360          * We are not putting this on the write queue, so
 361          * dccp_transmit_skb() will set the ownership to this
 362          * sock.
 363          */
 364         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 365
 366         if (skb == NULL)
 367                 /* FIXME: how to make sure the sync is sent? */
 368                 return;
 369
 370         /* Reserve space for headers and prepare control bits. */
 371         skb_reserve(skb, MAX_DCCP_HEADER);
 372         skb->csum = 0;
 373         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_SYNC;
 374         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 375
 376         skb_set_owner_w(skb, sk);
 377         dccp_transmit_skb(sk, skb);
 378 }
 379
 380 /* Send a DCCP_PKT_CLOSE/CLOSEREQ.  The caller locks the socket for us.  This cannot be
 381  * allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under any circumstances.
 382  */
 383 void dccp_send_close(struct sock *sk)
 384 {
 385         struct dccp_sock *dp = dccp_sk(sk);
 386         struct sk_buff *skb;
 387
 388         /* Socket is locked, keep trying until memory is available. */
 389         for (;;) {
 390                 skb = alloc_skb(sk->sk_prot->max_header, GFP_KERNEL);
 391                 if (skb != NULL)
 392                         break;
 393                 yield();
 394         }
 395
 396         /* Reserve space for headers and prepare control bits. */
 397         skb_reserve(skb, sk->sk_prot->max_header);
 398         skb->csum = 0;
 399         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ? DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 400
 401         skb_set_owner_w(skb, sk);
 402         dccp_transmit_skb(sk, skb);
 403
 404         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 405         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 406 }