Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-drm-fsl-dcu.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                         break;
65
66                 case DCCP_PKT_REQUEST:
67                         set_ack = 0;
68                         /* fall through */
69
70                 case DCCP_PKT_SYNC:
71                 case DCCP_PKT_SYNCACK:
72                         ackno = dcb->dccpd_seq;
73                         /* fall through */
74                 default:
75                         /*
76                          * Only data packets should come through with skb->sk
77                          * set.
78                          */
79                         WARN_ON(skb->sk);
80                         skb_set_owner_w(skb, sk);
81                         break;
82                 }
83
84                 dcb->dccpd_seq = dp->dccps_gss;
85
86                 if (dccp_insert_options(sk, skb)) {
87                         kfree_skb(skb);
88                         return -EPROTO;
89                 }
90                 
91                 skb->h.raw = skb_push(skb, dccp_header_size);
92                 dh = dccp_hdr(skb);
93
94                 /* Build DCCP header and checksum it. */
95                 memset(dh, 0, dccp_header_size);
96                 dh->dccph_type  = dcb->dccpd_type;
97                 dh->dccph_sport = inet->sport;
98                 dh->dccph_dport = inet->dport;
99                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
100                 dh->dccph_ccval = dcb->dccpd_ccval;
101                 /* XXX For now we're using only 48 bits sequence numbers */
102                 dh->dccph_x     = 1;
103
104                 dp->dccps_awh = dp->dccps_gss;
105                 dccp_hdr_set_seq(dh, dp->dccps_gss);
106                 if (set_ack)
107                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
108
109                 switch (dcb->dccpd_type) {
110                 case DCCP_PKT_REQUEST:
111                         dccp_hdr_request(skb)->dccph_req_service =
112                                                         dp->dccps_service;
113                         break;
114                 case DCCP_PKT_RESET:
115                         dccp_hdr_reset(skb)->dccph_reset_code =
116                                                         dcb->dccpd_reset_code;
117                         break;
118                 }
119
120                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
121
122                 if (set_ack)
123                         dccp_event_ack_sent(sk);
124
125                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
126
127                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
128                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
129                 if (err <= 0)
130                         return err;
131
132                 /* NET_XMIT_CN is special. It does not guarantee,
133                  * that this packet is lost. It tells that device
134                  * is about to start to drop packets or already
135                  * drops some packets of the same priority and
136                  * invokes us to send less aggressively.
137                  */
138                 return err == NET_XMIT_CN ? 0 : err;
139         }
140         return -ENOBUFS;
141 }
142
143 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
144 {
145         struct inet_connection_sock *icsk = inet_csk(sk);
146         struct dccp_sock *dp = dccp_sk(sk);
147         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
148                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
149
150         /* Now subtract optional transport overhead */
151         mss_now -= icsk->icsk_ext_hdr_len;
152
153         /*
154          * FIXME: this should come from the CCID infrastructure, where, say,
155          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
156          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
157          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
158          * make it a multiple of 4
159          */
160
161         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
162
163         /* And store cached results */
164         icsk->icsk_pmtu_cookie = pmtu;
165         dp->dccps_mss_cache = mss_now;
166
167         return mss_now;
168 }
169
170 EXPORT_SYMBOL_GPL(dccp_sync_mss);
171
172 void dccp_write_space(struct sock *sk)
173 {
174         read_lock(&sk->sk_callback_lock);
175
176         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
177                 wake_up_interruptible(sk->sk_sleep);
178         /* Should agree with poll, otherwise some programs break */
179         if (sock_writeable(sk))
180                 sk_wake_async(sk, 2, POLL_OUT);
181
182         read_unlock(&sk->sk_callback_lock);
183 }
184
185 /**
186  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
187  * @sk: socket to wait for
188  * @timeo: for how long
189  */
190 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
191                               long *timeo)
192 {
193         struct dccp_sock *dp = dccp_sk(sk);
194         DEFINE_WAIT(wait);
195         long delay;
196         int rc;
197
198         while (1) {
199                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
200
201                 if (sk->sk_err)
202                         goto do_error;
203                 if (!*timeo)
204                         goto do_nonblock;
205                 if (signal_pending(current))
206                         goto do_interrupted;
207
208                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
209                                             skb->len);
210                 if (rc <= 0)
211                         break;
212                 delay = msecs_to_jiffies(rc);
213                 if (delay > *timeo || delay < 0)
214                         goto do_nonblock;
215
216                 sk->sk_write_pending++;
217                 release_sock(sk);
218                 *timeo -= schedule_timeout(delay);
219                 lock_sock(sk);
220                 sk->sk_write_pending--;
221         }
222 out:
223         finish_wait(sk->sk_sleep, &wait);
224         return rc;
225
226 do_error:
227         rc = -EPIPE;
228         goto out;
229 do_nonblock:
230         rc = -EAGAIN;
231         goto out;
232 do_interrupted:
233         rc = sock_intr_errno(*timeo);
234         goto out;
235 }
236
237 static void dccp_write_xmit_timer(unsigned long data) {
238         struct sock *sk = (struct sock *)data;
239         struct dccp_sock *dp = dccp_sk(sk);
240
241         bh_lock_sock(sk);
242         if (sock_owned_by_user(sk))
243                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
244         else
245                 dccp_write_xmit(sk, 0);
246         bh_unlock_sock(sk);
247         sock_put(sk);
248 }
249
250 void dccp_write_xmit(struct sock *sk, int block)
251 {
252         struct dccp_sock *dp = dccp_sk(sk);
253         struct sk_buff *skb;
254         long timeo = 30000;     /* If a packet is taking longer than 2 secs
255                                    we have other issues */
256
257         while ((skb = skb_peek(&sk->sk_write_queue))) {
258                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
259                                          skb->len);
260
261                 if (err > 0) {
262                         if (!block) {
263                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
264                                                 msecs_to_jiffies(err)+jiffies);
265                                 break;
266                         } else
267                                 err = dccp_wait_for_ccid(sk, skb, &timeo);
268                         if (err) {
269                                 printk(KERN_CRIT "%s:err at dccp_wait_for_ccid"
270                                                  " %d\n", __FUNCTION__, err);
271                                 dump_stack();
272                         }
273                 }
274
275                 skb_dequeue(&sk->sk_write_queue);
276                 if (err == 0) {
277                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
278                         const int len = skb->len;
279
280                         if (sk->sk_state == DCCP_PARTOPEN) {
281                                 /* See 8.1.5.  Handshake Completion */
282                                 inet_csk_schedule_ack(sk);
283                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
284                                                   inet_csk(sk)->icsk_rto,
285                                                   DCCP_RTO_MAX);
286                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
287                         } else if (dccp_ack_pending(sk))
288                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
289                         else
290                                 dcb->dccpd_type = DCCP_PKT_DATA;
291
292                         err = dccp_transmit_skb(sk, skb);
293                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
294                         if (err) {
295                                 printk(KERN_CRIT "%s:err from "
296                                                  "ccid_hc_tx_packet_sent %d\n",
297                                                  __FUNCTION__, err);
298                                 dump_stack();
299                         }
300                 } else
301                         kfree(skb);
302         }
303 }
304
305 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
306 {
307         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
308                 return -EHOSTUNREACH; /* Routing failure or similar. */
309
310         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
311                                       pskb_copy(skb, GFP_ATOMIC):
312                                       skb_clone(skb, GFP_ATOMIC)));
313 }
314
315 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
316                                    struct request_sock *req)
317 {
318         struct dccp_hdr *dh;
319         struct dccp_request_sock *dreq;
320         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
321                                      sizeof(struct dccp_hdr_ext) +
322                                      sizeof(struct dccp_hdr_response);
323         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
324                                            GFP_ATOMIC);
325         if (skb == NULL)
326                 return NULL;
327
328         /* Reserve space for headers. */
329         skb_reserve(skb, sk->sk_prot->max_header);
330
331         skb->dst = dst_clone(dst);
332         skb->csum = 0;
333
334         dreq = dccp_rsk(req);
335         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
336         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
337
338         if (dccp_insert_options(sk, skb)) {
339                 kfree_skb(skb);
340                 return NULL;
341         }
342
343         skb->h.raw = skb_push(skb, dccp_header_size);
344
345         dh = dccp_hdr(skb);
346         memset(dh, 0, dccp_header_size);
347
348         dh->dccph_sport = inet_sk(sk)->sport;
349         dh->dccph_dport = inet_rsk(req)->rmt_port;
350         dh->dccph_doff  = (dccp_header_size +
351                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
352         dh->dccph_type  = DCCP_PKT_RESPONSE;
353         dh->dccph_x     = 1;
354         dccp_hdr_set_seq(dh, dreq->dreq_iss);
355         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
356         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
357
358         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
359         return skb;
360 }
361
362 EXPORT_SYMBOL_GPL(dccp_make_response);
363
364 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
365                                        const enum dccp_reset_codes code)
366                                    
367 {
368         struct dccp_hdr *dh;
369         struct dccp_sock *dp = dccp_sk(sk);
370         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
371                                      sizeof(struct dccp_hdr_ext) +
372                                      sizeof(struct dccp_hdr_reset);
373         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
374                                            GFP_ATOMIC);
375         if (skb == NULL)
376                 return NULL;
377
378         /* Reserve space for headers. */
379         skb_reserve(skb, sk->sk_prot->max_header);
380
381         skb->dst = dst_clone(dst);
382         skb->csum = 0;
383
384         dccp_inc_seqno(&dp->dccps_gss);
385
386         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
387         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
388         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
389
390         if (dccp_insert_options(sk, skb)) {
391                 kfree_skb(skb);
392                 return NULL;
393         }
394
395         skb->h.raw = skb_push(skb, dccp_header_size);
396
397         dh = dccp_hdr(skb);
398         memset(dh, 0, dccp_header_size);
399
400         dh->dccph_sport = inet_sk(sk)->sport;
401         dh->dccph_dport = inet_sk(sk)->dport;
402         dh->dccph_doff  = (dccp_header_size +
403                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
404         dh->dccph_type  = DCCP_PKT_RESET;
405         dh->dccph_x     = 1;
406         dccp_hdr_set_seq(dh, dp->dccps_gss);
407         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
408
409         dccp_hdr_reset(skb)->dccph_reset_code = code;
410         inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
411
412         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
413         return skb;
414 }
415
416 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
417 {
418         /*
419          * FIXME: what if rebuild_header fails?
420          * Should we be doing a rebuild_header here?
421          */
422         int err = inet_sk_rebuild_header(sk);
423
424         if (err == 0) {
425                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
426                                                       code);
427                 if (skb != NULL) {
428                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
429                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
430                         if (err == NET_XMIT_CN)
431                                 err = 0;
432                 }
433         }
434
435         return err;
436 }
437
438 /*
439  * Do all connect socket setups that can be done AF independent.
440  */
441 static inline void dccp_connect_init(struct sock *sk)
442 {
443         struct dccp_sock *dp = dccp_sk(sk);
444         struct dst_entry *dst = __sk_dst_get(sk);
445         struct inet_connection_sock *icsk = inet_csk(sk);
446
447         sk->sk_err = 0;
448         sock_reset_flag(sk, SOCK_DONE);
449         
450         dccp_sync_mss(sk, dst_mtu(dst));
451
452         dccp_update_gss(sk, dp->dccps_iss);
453         /*
454          * SWL and AWL are initially adjusted so that they are not less than
455          * the initial Sequence Numbers received and sent, respectively:
456          *      SWL := max(GSR + 1 - floor(W/4), ISR),
457          *      AWL := max(GSS - W' + 1, ISS).
458          * These adjustments MUST be applied only at the beginning of the
459          * connection.
460          */
461         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
462
463         icsk->icsk_retransmits = 0;
464         init_timer(&dp->dccps_xmit_timer);
465         dp->dccps_xmit_timer.data = (unsigned long)sk;
466         dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
467 }
468
469 int dccp_connect(struct sock *sk)
470 {
471         struct sk_buff *skb;
472         struct inet_connection_sock *icsk = inet_csk(sk);
473
474         dccp_connect_init(sk);
475
476         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
477         if (unlikely(skb == NULL))
478                 return -ENOBUFS;
479
480         /* Reserve space for headers. */
481         skb_reserve(skb, sk->sk_prot->max_header);
482
483         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
484         skb->csum = 0;
485
486         dccp_skb_entail(sk, skb);
487         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
488         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
489
490         /* Timer for repeating the REQUEST until an answer. */
491         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
492                                   icsk->icsk_rto, DCCP_RTO_MAX);
493         return 0;
494 }
495
496 EXPORT_SYMBOL_GPL(dccp_connect);
497
498 void dccp_send_ack(struct sock *sk)
499 {
500         /* If we have been reset, we may not send again. */
501         if (sk->sk_state != DCCP_CLOSED) {
502                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
503                                                 GFP_ATOMIC);
504
505                 if (skb == NULL) {
506                         inet_csk_schedule_ack(sk);
507                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
508                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
509                                                   TCP_DELACK_MAX,
510                                                   DCCP_RTO_MAX);
511                         return;
512                 }
513
514                 /* Reserve space for headers */
515                 skb_reserve(skb, sk->sk_prot->max_header);
516                 skb->csum = 0;
517                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
518                 dccp_transmit_skb(sk, skb);
519         }
520 }
521
522 EXPORT_SYMBOL_GPL(dccp_send_ack);
523
524 void dccp_send_delayed_ack(struct sock *sk)
525 {
526         struct inet_connection_sock *icsk = inet_csk(sk);
527         /*
528          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
529          * with using 2s, and active senders also piggyback the ACK into a
530          * DATAACK packet, so this is really for quiescent senders.
531          */
532         unsigned long timeout = jiffies + 2 * HZ;
533
534         /* Use new timeout only if there wasn't a older one earlier. */
535         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
536                 /* If delack timer was blocked or is about to expire,
537                  * send ACK now.
538                  *
539                  * FIXME: check the "about to expire" part
540                  */
541                 if (icsk->icsk_ack.blocked) {
542                         dccp_send_ack(sk);
543                         return;
544                 }
545
546                 if (!time_before(timeout, icsk->icsk_ack.timeout))
547                         timeout = icsk->icsk_ack.timeout;
548         }
549         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
550         icsk->icsk_ack.timeout = timeout;
551         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
552 }
553
554 void dccp_send_sync(struct sock *sk, const u64 seq,
555                     const enum dccp_pkt_type pkt_type)
556 {
557         /*
558          * We are not putting this on the write queue, so
559          * dccp_transmit_skb() will set the ownership to this
560          * sock.
561          */
562         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
563
564         if (skb == NULL)
565                 /* FIXME: how to make sure the sync is sent? */
566                 return;
567
568         /* Reserve space for headers and prepare control bits. */
569         skb_reserve(skb, sk->sk_prot->max_header);
570         skb->csum = 0;
571         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
572         DCCP_SKB_CB(skb)->dccpd_seq = seq;
573
574         dccp_transmit_skb(sk, skb);
575 }
576
577 EXPORT_SYMBOL_GPL(dccp_send_sync);
578
579 /*
580  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
581  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
582  * any circumstances.
583  */
584 void dccp_send_close(struct sock *sk, const int active)
585 {
586         struct dccp_sock *dp = dccp_sk(sk);
587         struct sk_buff *skb;
588         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
589
590         skb = alloc_skb(sk->sk_prot->max_header, prio);
591         if (skb == NULL)
592                 return;
593
594         /* Reserve space for headers and prepare control bits. */
595         skb_reserve(skb, sk->sk_prot->max_header);
596         skb->csum = 0;
597         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
598                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
599
600         if (active) {
601                 dccp_write_xmit(sk, 1);
602                 dccp_skb_entail(sk, skb);
603                 dccp_transmit_skb(sk, skb_clone(skb, prio));
604                 /* FIXME do we need a retransmit timer here? */
605         } else
606                 dccp_transmit_skb(sk, skb);
607 }