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/kernel.h>
  16 #include <linux/skbuff.h>
  17
  18 #include <net/sock.h>
  19
  20 #include "ackvec.h"
  21 #include "ccid.h"
  22 #include "dccp.h"
  23
  24 static inline void dccp_event_ack_sent(struct sock *sk)
  25 {
  26         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  27 }
  28
  29 static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  30 {
  31         skb_set_owner_w(skb, sk);
  32         WARN_ON(sk->sk_send_head);
  33         sk->sk_send_head = skb;
  34 }
  35
  36 /*
  37  * All SKB's seen here are completely headerless. It is our
  38  * job to build the DCCP header, and pass the packet down to
  39  * IP so it can do the same plus pass the packet off to the
  40  * device.
  41  */
  42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  43 {
  44         if (likely(skb != NULL)) {
  45                 const struct inet_sock *inet = inet_sk(sk);
  46                 struct dccp_sock *dp = dccp_sk(sk);
  47                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  48                 struct dccp_hdr *dh;
  49                 /* XXX For now we're using only 48 bits sequence numbers */
  50                 const int dccp_header_size = sizeof(*dh) +
  51                                              sizeof(struct dccp_hdr_ext) +
  52                                           dccp_packet_hdr_len(dcb->dccpd_type);
  53                 int err, set_ack = 1;
  54                 u64 ackno = dp->dccps_gsr;
  55
  56                 dccp_inc_seqno(&dp->dccps_gss);
  57
  58                 switch (dcb->dccpd_type) {
  59                 case DCCP_PKT_DATA:
  60                         set_ack = 0;
  61                         /* fall through */
  62                 case DCCP_PKT_DATAACK:
  63                         break;
  64
  65                 case DCCP_PKT_SYNC:
  66                 case DCCP_PKT_SYNCACK:
  67                         ackno = dcb->dccpd_seq;
  68                         /* fall through */
  69                 default:
  70                         /*
  71                          * Only data packets should come through with skb->sk
  72                          * set.
  73                          */
  74                         WARN_ON(skb->sk);
  75                         skb_set_owner_w(skb, sk);
  76                         break;
  77                 }
  78
  79                 dcb->dccpd_seq = dp->dccps_gss;
  80                 dccp_insert_options(sk, skb);
  81
  82                 skb->h.raw = skb_push(skb, dccp_header_size);
  83                 dh = dccp_hdr(skb);
  84
  85                 /* Build DCCP header and checksum it. */
  86                 memset(dh, 0, dccp_header_size);
  87                 dh->dccph_type  = dcb->dccpd_type;
  88                 dh->dccph_sport = inet->sport;
  89                 dh->dccph_dport = inet->dport;
  90                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  91                 dh->dccph_ccval = dcb->dccpd_ccval;
  92                 /* XXX For now we're using only 48 bits sequence numbers */
  93                 dh->dccph_x     = 1;
  94
  95                 dp->dccps_awh = dp->dccps_gss;
  96                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  97                 if (set_ack)
  98                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  99
 100                 switch (dcb->dccpd_type) {
 101                 case DCCP_PKT_REQUEST:
 102                         dccp_hdr_request(skb)->dccph_req_service =
 103                                                         dp->dccps_service;
 104                         break;
 105                 case DCCP_PKT_RESET:
 106                         dccp_hdr_reset(skb)->dccph_reset_code =
 107                                                         dcb->dccpd_reset_code;
 108                         break;
 109                 }
 110
 111                 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
 112                                                       inet->daddr);
 113
 114                 if (set_ack)
 115                         dccp_event_ack_sent(sk);
 116
 117                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 118
 119                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 120                 err = ip_queue_xmit(skb, 0);
 121                 if (err <= 0)
 122                         return err;
 123
 124                 /* NET_XMIT_CN is special. It does not guarantee,
 125                  * that this packet is lost. It tells that device
 126                  * is about to start to drop packets or already
 127                  * drops some packets of the same priority and
 128                  * invokes us to send less aggressively.
 129                  */
 130                 return err == NET_XMIT_CN ? 0 : err;
 131         }
 132         return -ENOBUFS;
 133 }
 134
 135 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 136 {
 137         struct dccp_sock *dp = dccp_sk(sk);
 138         int mss_now;
 139
 140         /*
 141          * FIXME: we really should be using the af_specific thing to support
 142          *        IPv6.
 143          * mss_now = pmtu - tp->af_specific->net_header_len -
 144          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 145          */
 146         mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
 147                   sizeof(struct dccp_hdr_ext);
 148
 149         /* Now subtract optional transport overhead */
 150         mss_now -= dp->dccps_ext_header_len;
 151
 152         /*
 153          * FIXME: this should come from the CCID infrastructure, where, say,
 154          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 155          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 156          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 157          * make it a multiple of 4
 158          */
 159
 160         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 161
 162         /* And store cached results */
 163         dp->dccps_pmtu_cookie = pmtu;
 164         dp->dccps_mss_cache = mss_now;
 165
 166         return mss_now;
 167 }
 168
 169 void dccp_write_space(struct sock *sk)
 170 {
 171         read_lock(&sk->sk_callback_lock);
 172
 173         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 174                 wake_up_interruptible(sk->sk_sleep);
 175         /* Should agree with poll, otherwise some programs break */
 176         if (sock_writeable(sk))
 177                 sk_wake_async(sk, 2, POLL_OUT);
 178
 179         read_unlock(&sk->sk_callback_lock);
 180 }
 181
 182 /**
 183  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 184  * @sk: socket to wait for
 185  * @timeo: for how long
 186  */
 187 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 188                               long *timeo)
 189 {
 190         struct dccp_sock *dp = dccp_sk(sk);
 191         DEFINE_WAIT(wait);
 192         long delay;
 193         int rc;
 194
 195         while (1) {
 196                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 197
 198                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 199                         goto do_error;
 200                 if (!*timeo)
 201                         goto do_nonblock;
 202                 if (signal_pending(current))
 203                         goto do_interrupted;
 204
 205                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 206                                             skb->len);
 207                 if (rc <= 0)
 208                         break;
 209                 delay = msecs_to_jiffies(rc);
 210                 if (delay > *timeo || delay < 0)
 211                         goto do_nonblock;
 212
 213                 sk->sk_write_pending++;
 214                 release_sock(sk);
 215                 *timeo -= schedule_timeout(delay);
 216                 lock_sock(sk);
 217                 sk->sk_write_pending--;
 218         }
 219 out:
 220         finish_wait(sk->sk_sleep, &wait);
 221         return rc;
 222
 223 do_error:
 224         rc = -EPIPE;
 225         goto out;
 226 do_nonblock:
 227         rc = -EAGAIN;
 228         goto out;
 229 do_interrupted:
 230         rc = sock_intr_errno(*timeo);
 231         goto out;
 232 }
 233
 234 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 235 {
 236         const struct dccp_sock *dp = dccp_sk(sk);
 237         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 238                                          skb->len);
 239
 240         if (err > 0)
 241                 err = dccp_wait_for_ccid(sk, skb, timeo);
 242
 243         if (err == 0) {
 244                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 245                 const int len = skb->len;
 246
 247                 if (sk->sk_state == DCCP_PARTOPEN) {
 248                         /* See 8.1.5.  Handshake Completion */
 249                         inet_csk_schedule_ack(sk);
 250                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 251                                                   inet_csk(sk)->icsk_rto,
 252                                                   DCCP_RTO_MAX);
 253                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 254                 } else if (dccp_ack_pending(sk))
 255                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 256                 else
 257                         dcb->dccpd_type = DCCP_PKT_DATA;
 258
 259                 err = dccp_transmit_skb(sk, skb);
 260                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 261         } else
 262                 kfree_skb(skb);
 263
 264         return err;
 265 }
 266
 267 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 268 {
 269         if (inet_sk_rebuild_header(sk) != 0)
 270                 return -EHOSTUNREACH; /* Routing failure or similar. */
 271
 272         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 273                                       pskb_copy(skb, GFP_ATOMIC):
 274                                       skb_clone(skb, GFP_ATOMIC)));
 275 }
 276
 277 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 278                                    struct request_sock *req)
 279 {
 280         struct dccp_hdr *dh;
 281         struct dccp_request_sock *dreq;
 282         const int dccp_header_size = sizeof(struct dccp_hdr) +
 283                                      sizeof(struct dccp_hdr_ext) +
 284                                      sizeof(struct dccp_hdr_response);
 285         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 286                                                dccp_header_size, 1,
 287                                            GFP_ATOMIC);
 288         if (skb == NULL)
 289                 return NULL;
 290
 291         /* Reserve space for headers. */
 292         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 293
 294         skb->dst = dst_clone(dst);
 295         skb->csum = 0;
 296
 297         dreq = dccp_rsk(req);
 298         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 299         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 300         dccp_insert_options(sk, skb);
 301
 302         skb->h.raw = skb_push(skb, dccp_header_size);
 303
 304         dh = dccp_hdr(skb);
 305         memset(dh, 0, dccp_header_size);
 306
 307         dh->dccph_sport = inet_sk(sk)->sport;
 308         dh->dccph_dport = inet_rsk(req)->rmt_port;
 309         dh->dccph_doff  = (dccp_header_size +
 310                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 311         dh->dccph_type  = DCCP_PKT_RESPONSE;
 312         dh->dccph_x     = 1;
 313         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 314         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 315         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 316
 317         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 318                                               inet_rsk(req)->rmt_addr);
 319
 320         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 321         return skb;
 322 }
 323
 324 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 325                                 const enum dccp_reset_codes code)
 326
 327 {
 328         struct dccp_hdr *dh;
 329         struct dccp_sock *dp = dccp_sk(sk);
 330         const int dccp_header_size = sizeof(struct dccp_hdr) +
 331                                      sizeof(struct dccp_hdr_ext) +
 332                                      sizeof(struct dccp_hdr_reset);
 333         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 334                                                dccp_header_size, 1,
 335                                            GFP_ATOMIC);
 336         if (skb == NULL)
 337                 return NULL;
 338
 339         /* Reserve space for headers. */
 340         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 341
 342         skb->dst = dst_clone(dst);
 343         skb->csum = 0;
 344
 345         dccp_inc_seqno(&dp->dccps_gss);
 346
 347         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 348         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 349         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 350         dccp_insert_options(sk, skb);
 351
 352         skb->h.raw = skb_push(skb, dccp_header_size);
 353
 354         dh = dccp_hdr(skb);
 355         memset(dh, 0, dccp_header_size);
 356
 357         dh->dccph_sport = inet_sk(sk)->sport;
 358         dh->dccph_dport = inet_sk(sk)->dport;
 359         dh->dccph_doff  = (dccp_header_size +
 360                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 361         dh->dccph_type  = DCCP_PKT_RESET;
 362         dh->dccph_x     = 1;
 363         dccp_hdr_set_seq(dh, dp->dccps_gss);
 364         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 365
 366         dccp_hdr_reset(skb)->dccph_reset_code = code;
 367
 368         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 369                                               inet_sk(sk)->daddr);
 370
 371         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 372         return skb;
 373 }
 374
 375 /*
 376  * Do all connect socket setups that can be done AF independent.
 377  */
 378 static inline void dccp_connect_init(struct sock *sk)
 379 {
 380         struct dst_entry *dst = __sk_dst_get(sk);
 381         struct inet_connection_sock *icsk = inet_csk(sk);
 382
 383         sk->sk_err = 0;
 384         sock_reset_flag(sk, SOCK_DONE);
 385
 386         dccp_sync_mss(sk, dst_mtu(dst));
 387
 388         /*
 389          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 390          * something like dccp_inc_seq
 391          */
 392
 393         icsk->icsk_retransmits = 0;
 394 }
 395
 396 int dccp_connect(struct sock *sk)
 397 {
 398         struct sk_buff *skb;
 399         struct inet_connection_sock *icsk = inet_csk(sk);
 400
 401         dccp_connect_init(sk);
 402
 403         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 404         if (unlikely(skb == NULL))
 405                 return -ENOBUFS;
 406
 407         /* Reserve space for headers. */
 408         skb_reserve(skb, MAX_DCCP_HEADER);
 409
 410         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 411         skb->csum = 0;
 412
 413         dccp_skb_entail(sk, skb);
 414         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 415         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 416
 417         /* Timer for repeating the REQUEST until an answer. */
 418         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 419                                   icsk->icsk_rto, DCCP_RTO_MAX);
 420         return 0;
 421 }
 422
 423 void dccp_send_ack(struct sock *sk)
 424 {
 425         /* If we have been reset, we may not send again. */
 426         if (sk->sk_state != DCCP_CLOSED) {
 427                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 428
 429                 if (skb == NULL) {
 430                         inet_csk_schedule_ack(sk);
 431                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 432                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 433                                                   TCP_DELACK_MAX,
 434                                                   DCCP_RTO_MAX);
 435                         return;
 436                 }
 437
 438                 /* Reserve space for headers */
 439                 skb_reserve(skb, MAX_DCCP_HEADER);
 440                 skb->csum = 0;
 441                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 442                 dccp_transmit_skb(sk, skb);
 443         }
 444 }
 445
 446 EXPORT_SYMBOL_GPL(dccp_send_ack);
 447
 448 void dccp_send_delayed_ack(struct sock *sk)
 449 {
 450         struct inet_connection_sock *icsk = inet_csk(sk);
 451         /*
 452          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 453          * with using 2s, and active senders also piggyback the ACK into a
 454          * DATAACK packet, so this is really for quiescent senders.
 455          */
 456         unsigned long timeout = jiffies + 2 * HZ;
 457
 458         /* Use new timeout only if there wasn't a older one earlier. */
 459         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 460                 /* If delack timer was blocked or is about to expire,
 461                  * send ACK now.
 462                  *
 463                  * FIXME: check the "about to expire" part
 464                  */
 465                 if (icsk->icsk_ack.blocked) {
 466                         dccp_send_ack(sk);
 467                         return;
 468                 }
 469
 470                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 471                         timeout = icsk->icsk_ack.timeout;
 472         }
 473         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 474         icsk->icsk_ack.timeout = timeout;
 475         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 476 }
 477
 478 void dccp_send_sync(struct sock *sk, const u64 seq,
 479                     const enum dccp_pkt_type pkt_type)
 480 {
 481         /*
 482          * We are not putting this on the write queue, so
 483          * dccp_transmit_skb() will set the ownership to this
 484          * sock.
 485          */
 486         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 487
 488         if (skb == NULL)
 489                 /* FIXME: how to make sure the sync is sent? */
 490                 return;
 491
 492         /* Reserve space for headers and prepare control bits. */
 493         skb_reserve(skb, MAX_DCCP_HEADER);
 494         skb->csum = 0;
 495         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 496         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 497
 498         dccp_transmit_skb(sk, skb);
 499 }
 500
 501 /*
 502  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 503  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 504  * any circumstances.
 505  */
 506 void dccp_send_close(struct sock *sk, const int active)
 507 {
 508         struct dccp_sock *dp = dccp_sk(sk);
 509         struct sk_buff *skb;
 510         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 511
 512         skb = alloc_skb(sk->sk_prot->max_header, prio);
 513         if (skb == NULL)
 514                 return;
 515
 516         /* Reserve space for headers and prepare control bits. */
 517         skb_reserve(skb, sk->sk_prot->max_header);
 518         skb->csum = 0;
 519         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 520                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 521
 522         if (active) {
 523                 dccp_skb_entail(sk, skb);
 524                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 525         } else
 526                 dccp_transmit_skb(sk, skb);
 527 }