]> nv-tegra.nvidia Code Review - linux-2.6.git/blob - net/dccp/output.c
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / 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/dccp.h>
14 #include <linux/kernel.h>
15 #include <linux/skbuff.h>
16
17 #include <net/inet_sock.h>
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 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                 const struct inet_connection_sock *icsk = inet_csk(sk);
47                 struct dccp_sock *dp = dccp_sk(sk);
48                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
49                 struct dccp_hdr *dh;
50                 /* XXX For now we're using only 48 bits sequence numbers */
51                 const u32 dccp_header_size = sizeof(*dh) +
52                                              sizeof(struct dccp_hdr_ext) +
53                                           dccp_packet_hdr_len(dcb->dccpd_type);
54                 int err, set_ack = 1;
55                 u64 ackno = dp->dccps_gsr;
56
57                 dccp_inc_seqno(&dp->dccps_gss);
58
59                 switch (dcb->dccpd_type) {
60                 case DCCP_PKT_DATA:
61                         set_ack = 0;
62                         /* fall through */
63                 case DCCP_PKT_DATAACK:
64                 case DCCP_PKT_RESET:
65                         break;
66
67                 case DCCP_PKT_REQUEST:
68                         set_ack = 0;
69                         /* fall through */
70
71                 case DCCP_PKT_SYNC:
72                 case DCCP_PKT_SYNCACK:
73                         ackno = dcb->dccpd_ack_seq;
74                         /* fall through */
75                 default:
76                         /*
77                          * Set owner/destructor: some skbs are allocated via
78                          * alloc_skb (e.g. when retransmission may happen).
79                          * Only Data, DataAck, and Reset packets should come
80                          * through here with skb->sk set.
81                          */
82                         WARN_ON(skb->sk);
83                         skb_set_owner_w(skb, sk);
84                         break;
85                 }
86
87                 dcb->dccpd_seq = dp->dccps_gss;
88
89                 if (dccp_insert_options(sk, skb)) {
90                         kfree_skb(skb);
91                         return -EPROTO;
92                 }
93
94
95                 /* Build DCCP header and checksum it. */
96                 dh = dccp_zeroed_hdr(skb, dccp_header_size);
97                 dh->dccph_type  = dcb->dccpd_type;
98                 dh->dccph_sport = inet->sport;
99                 dh->dccph_dport = inet->dport;
100                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
101                 dh->dccph_ccval = dcb->dccpd_ccval;
102                 dh->dccph_cscov = dp->dccps_pcslen;
103                 /* XXX For now we're using only 48 bits sequence numbers */
104                 dh->dccph_x     = 1;
105
106                 dp->dccps_awh = dp->dccps_gss;
107                 dccp_hdr_set_seq(dh, dp->dccps_gss);
108                 if (set_ack)
109                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
110
111                 switch (dcb->dccpd_type) {
112                 case DCCP_PKT_REQUEST:
113                         dccp_hdr_request(skb)->dccph_req_service =
114                                                         dp->dccps_service;
115                         break;
116                 case DCCP_PKT_RESET:
117                         dccp_hdr_reset(skb)->dccph_reset_code =
118                                                         dcb->dccpd_reset_code;
119                         break;
120                 }
121
122                 icsk->icsk_af_ops->send_check(sk, 0, skb);
123
124                 if (set_ack)
125                         dccp_event_ack_sent(sk);
126
127                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
128
129                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
130                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
131                 return net_xmit_eval(err);
132         }
133         return -ENOBUFS;
134 }
135
136 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
137 {
138         struct inet_connection_sock *icsk = inet_csk(sk);
139         struct dccp_sock *dp = dccp_sk(sk);
140         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
141                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
142
143         /* Now subtract optional transport overhead */
144         mss_now -= icsk->icsk_ext_hdr_len;
145
146         /*
147          * FIXME: this should come from the CCID infrastructure, where, say,
148          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
149          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
150          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
151          * make it a multiple of 4
152          */
153
154         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
155
156         /* And store cached results */
157         icsk->icsk_pmtu_cookie = pmtu;
158         dp->dccps_mss_cache = mss_now;
159
160         return mss_now;
161 }
162
163 EXPORT_SYMBOL_GPL(dccp_sync_mss);
164
165 void dccp_write_space(struct sock *sk)
166 {
167         read_lock(&sk->sk_callback_lock);
168
169         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
170                 wake_up_interruptible(sk->sk_sleep);
171         /* Should agree with poll, otherwise some programs break */
172         if (sock_writeable(sk))
173                 sk_wake_async(sk, 2, POLL_OUT);
174
175         read_unlock(&sk->sk_callback_lock);
176 }
177
178 /**
179  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
180  * @sk:    socket to wait for
181  * @skb:   current skb to pass on for waiting
182  * @delay: sleep timeout in milliseconds (> 0)
183  * This function is called by default when the socket is closed, and
184  * when a non-zero linger time is set on the socket. For consistency
185  */
186 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb, int delay)
187 {
188         struct dccp_sock *dp = dccp_sk(sk);
189         DEFINE_WAIT(wait);
190         unsigned long jiffdelay;
191         int rc;
192
193         do {
194                 dccp_pr_debug("delayed send by %d msec\n", delay);
195                 jiffdelay = msecs_to_jiffies(delay);
196
197                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
198
199                 sk->sk_write_pending++;
200                 release_sock(sk);
201                 schedule_timeout(jiffdelay);
202                 lock_sock(sk);
203                 sk->sk_write_pending--;
204
205                 if (sk->sk_err)
206                         goto do_error;
207                 if (signal_pending(current))
208                         goto do_interrupted;
209
210                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
211         } while ((delay = rc) > 0);
212 out:
213         finish_wait(sk->sk_sleep, &wait);
214         return rc;
215
216 do_error:
217         rc = -EPIPE;
218         goto out;
219 do_interrupted:
220         rc = -EINTR;
221         goto out;
222 }
223
224 void dccp_write_xmit(struct sock *sk, int block)
225 {
226         struct dccp_sock *dp = dccp_sk(sk);
227         struct sk_buff *skb;
228
229         while ((skb = skb_peek(&sk->sk_write_queue))) {
230                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
231
232                 if (err > 0) {
233                         if (!block) {
234                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
235                                                 msecs_to_jiffies(err)+jiffies);
236                                 break;
237                         } else
238                                 err = dccp_wait_for_ccid(sk, skb, err);
239                         if (err && err != -EINTR)
240                                 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
241                 }
242
243                 skb_dequeue(&sk->sk_write_queue);
244                 if (err == 0) {
245                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
246                         const int len = skb->len;
247
248                         if (sk->sk_state == DCCP_PARTOPEN) {
249                                 /* See 8.1.5.  Handshake Completion */
250                                 inet_csk_schedule_ack(sk);
251                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
252                                                   inet_csk(sk)->icsk_rto,
253                                                   DCCP_RTO_MAX);
254                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
255                         } else if (dccp_ack_pending(sk))
256                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
257                         else
258                                 dcb->dccpd_type = DCCP_PKT_DATA;
259
260                         err = dccp_transmit_skb(sk, skb);
261                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
262                         if (err)
263                                 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
264                                          err);
265                 } else {
266                         dccp_pr_debug("packet discarded due to err=%d\n", err);
267                         kfree_skb(skb);
268                 }
269         }
270 }
271
272 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
273 {
274         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
275                 return -EHOSTUNREACH; /* Routing failure or similar. */
276
277         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
278                                       pskb_copy(skb, GFP_ATOMIC):
279                                       skb_clone(skb, GFP_ATOMIC)));
280 }
281
282 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
283                                    struct request_sock *req)
284 {
285         struct dccp_hdr *dh;
286         struct dccp_request_sock *dreq;
287         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
288                                      sizeof(struct dccp_hdr_ext) +
289                                      sizeof(struct dccp_hdr_response);
290         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
291                                            GFP_ATOMIC);
292         if (skb == NULL)
293                 return NULL;
294
295         /* Reserve space for headers. */
296         skb_reserve(skb, sk->sk_prot->max_header);
297
298         skb->dst = dst_clone(dst);
299
300         dreq = dccp_rsk(req);
301         if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
302                 dccp_inc_seqno(&dreq->dreq_iss);
303         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
304         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
305
306         if (dccp_insert_options(sk, skb)) {
307                 kfree_skb(skb);
308                 return NULL;
309         }
310
311         /* Build and checksum header */
312         dh = dccp_zeroed_hdr(skb, dccp_header_size);
313
314         dh->dccph_sport = inet_sk(sk)->sport;
315         dh->dccph_dport = inet_rsk(req)->rmt_port;
316         dh->dccph_doff  = (dccp_header_size +
317                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
318         dh->dccph_type  = DCCP_PKT_RESPONSE;
319         dh->dccph_x     = 1;
320         dccp_hdr_set_seq(dh, dreq->dreq_iss);
321         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
322         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
323
324         dccp_csum_outgoing(skb);
325
326         /* We use `acked' to remember that a Response was already sent. */
327         inet_rsk(req)->acked = 1;
328         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
329         return skb;
330 }
331
332 EXPORT_SYMBOL_GPL(dccp_make_response);
333
334 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
335 struct sk_buff *dccp_ctl_make_reset(struct socket *ctl, struct sk_buff *rcv_skb)
336 {
337         struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
338         struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
339         const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
340                                        sizeof(struct dccp_hdr_ext) +
341                                        sizeof(struct dccp_hdr_reset);
342         struct dccp_hdr_reset *dhr;
343         struct sk_buff *skb;
344
345         skb = alloc_skb(ctl->sk->sk_prot->max_header, GFP_ATOMIC);
346         if (skb == NULL)
347                 return NULL;
348
349         skb_reserve(skb, ctl->sk->sk_prot->max_header);
350
351         /* Swap the send and the receive. */
352         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
353         dh->dccph_type  = DCCP_PKT_RESET;
354         dh->dccph_sport = rxdh->dccph_dport;
355         dh->dccph_dport = rxdh->dccph_sport;
356         dh->dccph_doff  = dccp_hdr_reset_len / 4;
357         dh->dccph_x     = 1;
358
359         dhr = dccp_hdr_reset(skb);
360         dhr->dccph_reset_code = dcb->dccpd_reset_code;
361
362         switch (dcb->dccpd_reset_code) {
363         case DCCP_RESET_CODE_PACKET_ERROR:
364                 dhr->dccph_reset_data[0] = rxdh->dccph_type;
365                 break;
366         case DCCP_RESET_CODE_OPTION_ERROR:      /* fall through */
367         case DCCP_RESET_CODE_MANDATORY_ERROR:
368                 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
369                 break;
370         }
371         /*
372          * From RFC 4340, 8.3.1:
373          *   If P.ackno exists, set R.seqno := P.ackno + 1.
374          *   Else set R.seqno := 0.
375          */
376         if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
377                 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
378         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
379
380         dccp_csum_outgoing(skb);
381         return skb;
382 }
383
384 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
385
386 /* send Reset on established socket, to close or abort the connection */
387 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
388 {
389         struct sk_buff *skb;
390         /*
391          * FIXME: what if rebuild_header fails?
392          * Should we be doing a rebuild_header here?
393          */
394         int err = inet_sk_rebuild_header(sk);
395
396         if (err != 0)
397                 return err;
398
399         skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
400         if (skb == NULL)
401                 return -ENOBUFS;
402
403         /* Reserve space for headers and prepare control bits. */
404         skb_reserve(skb, sk->sk_prot->max_header);
405         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
406         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
407
408         return dccp_transmit_skb(sk, skb);
409 }
410
411 /*
412  * Do all connect socket setups that can be done AF independent.
413  */
414 static inline void dccp_connect_init(struct sock *sk)
415 {
416         struct dccp_sock *dp = dccp_sk(sk);
417         struct dst_entry *dst = __sk_dst_get(sk);
418         struct inet_connection_sock *icsk = inet_csk(sk);
419
420         sk->sk_err = 0;
421         sock_reset_flag(sk, SOCK_DONE);
422
423         dccp_sync_mss(sk, dst_mtu(dst));
424
425         /*
426          * SWL and AWL are initially adjusted so that they are not less than
427          * the initial Sequence Numbers received and sent, respectively:
428          *      SWL := max(GSR + 1 - floor(W/4), ISR),
429          *      AWL := max(GSS - W' + 1, ISS).
430          * These adjustments MUST be applied only at the beginning of the
431          * connection.
432          */
433         dccp_update_gss(sk, dp->dccps_iss);
434         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
435
436         /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
437          *         initialized to S.ISS (sec. 8.5)                            */
438         dp->dccps_gar = dp->dccps_iss;
439
440         icsk->icsk_retransmits = 0;
441 }
442
443 int dccp_connect(struct sock *sk)
444 {
445         struct sk_buff *skb;
446         struct inet_connection_sock *icsk = inet_csk(sk);
447
448         dccp_connect_init(sk);
449
450         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
451         if (unlikely(skb == NULL))
452                 return -ENOBUFS;
453
454         /* Reserve space for headers. */
455         skb_reserve(skb, sk->sk_prot->max_header);
456
457         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
458
459         dccp_skb_entail(sk, skb);
460         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
461         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
462
463         /* Timer for repeating the REQUEST until an answer. */
464         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
465                                   icsk->icsk_rto, DCCP_RTO_MAX);
466         return 0;
467 }
468
469 EXPORT_SYMBOL_GPL(dccp_connect);
470
471 void dccp_send_ack(struct sock *sk)
472 {
473         /* If we have been reset, we may not send again. */
474         if (sk->sk_state != DCCP_CLOSED) {
475                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
476                                                 GFP_ATOMIC);
477
478                 if (skb == NULL) {
479                         inet_csk_schedule_ack(sk);
480                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
481                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
482                                                   TCP_DELACK_MAX,
483                                                   DCCP_RTO_MAX);
484                         return;
485                 }
486
487                 /* Reserve space for headers */
488                 skb_reserve(skb, sk->sk_prot->max_header);
489                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
490                 dccp_transmit_skb(sk, skb);
491         }
492 }
493
494 EXPORT_SYMBOL_GPL(dccp_send_ack);
495
496 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
497 void dccp_send_delayed_ack(struct sock *sk)
498 {
499         struct inet_connection_sock *icsk = inet_csk(sk);
500         /*
501          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
502          * with using 2s, and active senders also piggyback the ACK into a
503          * DATAACK packet, so this is really for quiescent senders.
504          */
505         unsigned long timeout = jiffies + 2 * HZ;
506
507         /* Use new timeout only if there wasn't a older one earlier. */
508         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
509                 /* If delack timer was blocked or is about to expire,
510                  * send ACK now.
511                  *
512                  * FIXME: check the "about to expire" part
513                  */
514                 if (icsk->icsk_ack.blocked) {
515                         dccp_send_ack(sk);
516                         return;
517                 }
518
519                 if (!time_before(timeout, icsk->icsk_ack.timeout))
520                         timeout = icsk->icsk_ack.timeout;
521         }
522         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
523         icsk->icsk_ack.timeout = timeout;
524         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
525 }
526
527 void dccp_send_sync(struct sock *sk, const u64 ackno,
528                     const enum dccp_pkt_type pkt_type)
529 {
530         /*
531          * We are not putting this on the write queue, so
532          * dccp_transmit_skb() will set the ownership to this
533          * sock.
534          */
535         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
536
537         if (skb == NULL) {
538                 /* FIXME: how to make sure the sync is sent? */
539                 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
540                 return;
541         }
542
543         /* Reserve space for headers and prepare control bits. */
544         skb_reserve(skb, sk->sk_prot->max_header);
545         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
546         DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
547
548         dccp_transmit_skb(sk, skb);
549 }
550
551 EXPORT_SYMBOL_GPL(dccp_send_sync);
552
553 /*
554  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
555  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
556  * any circumstances.
557  */
558 void dccp_send_close(struct sock *sk, const int active)
559 {
560         struct dccp_sock *dp = dccp_sk(sk);
561         struct sk_buff *skb;
562         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
563
564         skb = alloc_skb(sk->sk_prot->max_header, prio);
565         if (skb == NULL)
566                 return;
567
568         /* Reserve space for headers and prepare control bits. */
569         skb_reserve(skb, sk->sk_prot->max_header);
570         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
571                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
572
573         if (active) {
574                 dccp_write_xmit(sk, 1);
575                 dccp_skb_entail(sk, skb);
576                 dccp_transmit_skb(sk, skb_clone(skb, prio));
577                 /* FIXME do we need a retransmit timer here? */
578         } else
579                 dccp_transmit_skb(sk, skb);
580 }