400c30b6fcae4d1e0357188d2081e60d06072e55
[pandora-kernel.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
92                 /* Build DCCP header and checksum it. */
93                 dh = dccp_zeroed_hdr(skb, dccp_header_size);
94                 dh->dccph_type  = dcb->dccpd_type;
95                 dh->dccph_sport = inet->sport;
96                 dh->dccph_dport = inet->dport;
97                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
98                 dh->dccph_ccval = dcb->dccpd_ccval;
99                 dh->dccph_cscov = dp->dccps_pcslen;
100                 /* XXX For now we're using only 48 bits sequence numbers */
101                 dh->dccph_x     = 1;
102
103                 dp->dccps_awh = dp->dccps_gss;
104                 dccp_hdr_set_seq(dh, dp->dccps_gss);
105                 if (set_ack)
106                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
107
108                 switch (dcb->dccpd_type) {
109                 case DCCP_PKT_REQUEST:
110                         dccp_hdr_request(skb)->dccph_req_service =
111                                                         dp->dccps_service;
112                         break;
113                 case DCCP_PKT_RESET:
114                         dccp_hdr_reset(skb)->dccph_reset_code =
115                                                         dcb->dccpd_reset_code;
116                         break;
117                 }
118
119                 icsk->icsk_af_ops->send_check(sk, 0, skb);
120
121                 if (set_ack)
122                         dccp_event_ack_sent(sk);
123
124                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
125
126                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
127                 err = icsk->icsk_af_ops->queue_xmit(skb, sk, 0);
128                 return net_xmit_eval(err);
129         }
130         return -ENOBUFS;
131 }
132
133 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
134 {
135         struct inet_connection_sock *icsk = inet_csk(sk);
136         struct dccp_sock *dp = dccp_sk(sk);
137         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
138                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
139
140         /* Now subtract optional transport overhead */
141         mss_now -= icsk->icsk_ext_hdr_len;
142
143         /*
144          * FIXME: this should come from the CCID infrastructure, where, say,
145          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
146          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
147          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
148          * make it a multiple of 4
149          */
150
151         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
152
153         /* And store cached results */
154         icsk->icsk_pmtu_cookie = pmtu;
155         dp->dccps_mss_cache = mss_now;
156
157         return mss_now;
158 }
159
160 EXPORT_SYMBOL_GPL(dccp_sync_mss);
161
162 void dccp_write_space(struct sock *sk)
163 {
164         read_lock(&sk->sk_callback_lock);
165
166         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
167                 wake_up_interruptible(sk->sk_sleep);
168         /* Should agree with poll, otherwise some programs break */
169         if (sock_writeable(sk))
170                 sk_wake_async(sk, 2, POLL_OUT);
171
172         read_unlock(&sk->sk_callback_lock);
173 }
174
175 /**
176  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
177  * @sk: socket to wait for
178  * @timeo: for how long
179  */
180 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
181                               long *timeo)
182 {
183         struct dccp_sock *dp = dccp_sk(sk);
184         DEFINE_WAIT(wait);
185         long delay;
186         int rc;
187
188         while (1) {
189                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
190
191                 if (sk->sk_err)
192                         goto do_error;
193                 if (!*timeo)
194                         goto do_nonblock;
195                 if (signal_pending(current))
196                         goto do_interrupted;
197
198                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
199                 if (rc <= 0)
200                         break;
201                 delay = msecs_to_jiffies(rc);
202                 if (delay > *timeo || delay < 0)
203                         goto do_nonblock;
204
205                 sk->sk_write_pending++;
206                 release_sock(sk);
207                 *timeo -= schedule_timeout(delay);
208                 lock_sock(sk);
209                 sk->sk_write_pending--;
210         }
211 out:
212         finish_wait(sk->sk_sleep, &wait);
213         return rc;
214
215 do_error:
216         rc = -EPIPE;
217         goto out;
218 do_nonblock:
219         rc = -EAGAIN;
220         goto out;
221 do_interrupted:
222         rc = sock_intr_errno(*timeo);
223         goto out;
224 }
225
226 static void dccp_write_xmit_timer(unsigned long data) {
227         struct sock *sk = (struct sock *)data;
228         struct dccp_sock *dp = dccp_sk(sk);
229
230         bh_lock_sock(sk);
231         if (sock_owned_by_user(sk))
232                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
233         else
234                 dccp_write_xmit(sk, 0);
235         bh_unlock_sock(sk);
236         sock_put(sk);
237 }
238
239 void dccp_write_xmit(struct sock *sk, int block)
240 {
241         struct dccp_sock *dp = dccp_sk(sk);
242         struct sk_buff *skb;
243         long timeo = DCCP_XMIT_TIMEO;   /* If a packet is taking longer than
244                                            this we have other issues */
245
246         while ((skb = skb_peek(&sk->sk_write_queue))) {
247                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
248
249                 if (err > 0) {
250                         if (!block) {
251                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
252                                                 msecs_to_jiffies(err)+jiffies);
253                                 break;
254                         } else {
255                                 err = dccp_wait_for_ccid(sk, skb, &timeo);
256                                 timeo = DCCP_XMIT_TIMEO;
257                         }
258                         if (err)
259                                 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
260                 }
261
262                 skb_dequeue(&sk->sk_write_queue);
263                 if (err == 0) {
264                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
265                         const int len = skb->len;
266
267                         if (sk->sk_state == DCCP_PARTOPEN) {
268                                 /* See 8.1.5.  Handshake Completion */
269                                 inet_csk_schedule_ack(sk);
270                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
271                                                   inet_csk(sk)->icsk_rto,
272                                                   DCCP_RTO_MAX);
273                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
274                         } else if (dccp_ack_pending(sk))
275                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
276                         else
277                                 dcb->dccpd_type = DCCP_PKT_DATA;
278
279                         err = dccp_transmit_skb(sk, skb);
280                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
281                         if (err)
282                                 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
283                                          err);
284                 } else
285                         kfree(skb);
286         }
287 }
288
289 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
290 {
291         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
292                 return -EHOSTUNREACH; /* Routing failure or similar. */
293
294         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
295                                       pskb_copy(skb, GFP_ATOMIC):
296                                       skb_clone(skb, GFP_ATOMIC)));
297 }
298
299 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
300                                    struct request_sock *req)
301 {
302         struct dccp_hdr *dh;
303         struct dccp_request_sock *dreq;
304         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
305                                      sizeof(struct dccp_hdr_ext) +
306                                      sizeof(struct dccp_hdr_response);
307         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
308                                            GFP_ATOMIC);
309         if (skb == NULL)
310                 return NULL;
311
312         /* Reserve space for headers. */
313         skb_reserve(skb, sk->sk_prot->max_header);
314
315         skb->dst = dst_clone(dst);
316
317         dreq = dccp_rsk(req);
318         if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
319                 dccp_inc_seqno(&dreq->dreq_iss);
320         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
321         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
322
323         if (dccp_insert_options(sk, skb)) {
324                 kfree_skb(skb);
325                 return NULL;
326         }
327
328         /* Build and checksum header */
329         dh = dccp_zeroed_hdr(skb, dccp_header_size);
330
331         dh->dccph_sport = inet_sk(sk)->sport;
332         dh->dccph_dport = inet_rsk(req)->rmt_port;
333         dh->dccph_doff  = (dccp_header_size +
334                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
335         dh->dccph_type  = DCCP_PKT_RESPONSE;
336         dh->dccph_x     = 1;
337         dccp_hdr_set_seq(dh, dreq->dreq_iss);
338         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
339         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
340
341         dccp_csum_outgoing(skb);
342
343         /* We use `acked' to remember that a Response was already sent. */
344         inet_rsk(req)->acked = 1;
345         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
346         return skb;
347 }
348
349 EXPORT_SYMBOL_GPL(dccp_make_response);
350
351 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
352                                        const enum dccp_reset_codes code)
353                                    
354 {
355         struct dccp_hdr *dh;
356         struct dccp_sock *dp = dccp_sk(sk);
357         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
358                                      sizeof(struct dccp_hdr_ext) +
359                                      sizeof(struct dccp_hdr_reset);
360         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
361                                            GFP_ATOMIC);
362         if (skb == NULL)
363                 return NULL;
364
365         /* Reserve space for headers. */
366         skb_reserve(skb, sk->sk_prot->max_header);
367
368         skb->dst = dst_clone(dst);
369
370         dccp_inc_seqno(&dp->dccps_gss);
371
372         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
373         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
374         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
375
376         if (dccp_insert_options(sk, skb)) {
377                 kfree_skb(skb);
378                 return NULL;
379         }
380
381         dh = dccp_zeroed_hdr(skb, dccp_header_size);
382
383         dh->dccph_sport = inet_sk(sk)->sport;
384         dh->dccph_dport = inet_sk(sk)->dport;
385         dh->dccph_doff  = (dccp_header_size +
386                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
387         dh->dccph_type  = DCCP_PKT_RESET;
388         dh->dccph_x     = 1;
389         dccp_hdr_set_seq(dh, dp->dccps_gss);
390         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
391
392         dccp_hdr_reset(skb)->dccph_reset_code = code;
393         inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
394
395         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
396         return skb;
397 }
398
399 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
400 {
401         /*
402          * FIXME: what if rebuild_header fails?
403          * Should we be doing a rebuild_header here?
404          */
405         int err = inet_sk_rebuild_header(sk);
406
407         if (err == 0) {
408                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
409                                                       code);
410                 if (skb != NULL) {
411                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
412                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, sk, 0);
413                         return net_xmit_eval(err);
414                 }
415         }
416
417         return err;
418 }
419
420 /*
421  * Do all connect socket setups that can be done AF independent.
422  */
423 static inline void dccp_connect_init(struct sock *sk)
424 {
425         struct dccp_sock *dp = dccp_sk(sk);
426         struct dst_entry *dst = __sk_dst_get(sk);
427         struct inet_connection_sock *icsk = inet_csk(sk);
428
429         sk->sk_err = 0;
430         sock_reset_flag(sk, SOCK_DONE);
431         
432         dccp_sync_mss(sk, dst_mtu(dst));
433
434         /*
435          * SWL and AWL are initially adjusted so that they are not less than
436          * the initial Sequence Numbers received and sent, respectively:
437          *      SWL := max(GSR + 1 - floor(W/4), ISR),
438          *      AWL := max(GSS - W' + 1, ISS).
439          * These adjustments MUST be applied only at the beginning of the
440          * connection.
441          */
442         dccp_update_gss(sk, dp->dccps_iss);
443         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
444
445         /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
446          *         initialized to S.ISS (sec. 8.5)                            */
447         dp->dccps_gar = dp->dccps_iss;
448
449         icsk->icsk_retransmits = 0;
450         init_timer(&dp->dccps_xmit_timer);
451         dp->dccps_xmit_timer.data = (unsigned long)sk;
452         dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
453 }
454
455 int dccp_connect(struct sock *sk)
456 {
457         struct sk_buff *skb;
458         struct inet_connection_sock *icsk = inet_csk(sk);
459
460         dccp_connect_init(sk);
461
462         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
463         if (unlikely(skb == NULL))
464                 return -ENOBUFS;
465
466         /* Reserve space for headers. */
467         skb_reserve(skb, sk->sk_prot->max_header);
468
469         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
470
471         dccp_skb_entail(sk, skb);
472         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
473         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
474
475         /* Timer for repeating the REQUEST until an answer. */
476         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
477                                   icsk->icsk_rto, DCCP_RTO_MAX);
478         return 0;
479 }
480
481 EXPORT_SYMBOL_GPL(dccp_connect);
482
483 void dccp_send_ack(struct sock *sk)
484 {
485         /* If we have been reset, we may not send again. */
486         if (sk->sk_state != DCCP_CLOSED) {
487                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
488                                                 GFP_ATOMIC);
489
490                 if (skb == NULL) {
491                         inet_csk_schedule_ack(sk);
492                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
493                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
494                                                   TCP_DELACK_MAX,
495                                                   DCCP_RTO_MAX);
496                         return;
497                 }
498
499                 /* Reserve space for headers */
500                 skb_reserve(skb, sk->sk_prot->max_header);
501                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
502                 dccp_transmit_skb(sk, skb);
503         }
504 }
505
506 EXPORT_SYMBOL_GPL(dccp_send_ack);
507
508 void dccp_send_delayed_ack(struct sock *sk)
509 {
510         struct inet_connection_sock *icsk = inet_csk(sk);
511         /*
512          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
513          * with using 2s, and active senders also piggyback the ACK into a
514          * DATAACK packet, so this is really for quiescent senders.
515          */
516         unsigned long timeout = jiffies + 2 * HZ;
517
518         /* Use new timeout only if there wasn't a older one earlier. */
519         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
520                 /* If delack timer was blocked or is about to expire,
521                  * send ACK now.
522                  *
523                  * FIXME: check the "about to expire" part
524                  */
525                 if (icsk->icsk_ack.blocked) {
526                         dccp_send_ack(sk);
527                         return;
528                 }
529
530                 if (!time_before(timeout, icsk->icsk_ack.timeout))
531                         timeout = icsk->icsk_ack.timeout;
532         }
533         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
534         icsk->icsk_ack.timeout = timeout;
535         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
536 }
537
538 void dccp_send_sync(struct sock *sk, const u64 seq,
539                     const enum dccp_pkt_type pkt_type)
540 {
541         /*
542          * We are not putting this on the write queue, so
543          * dccp_transmit_skb() will set the ownership to this
544          * sock.
545          */
546         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
547
548         if (skb == NULL)
549                 /* FIXME: how to make sure the sync is sent? */
550                 return;
551
552         /* Reserve space for headers and prepare control bits. */
553         skb_reserve(skb, sk->sk_prot->max_header);
554         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
555         DCCP_SKB_CB(skb)->dccpd_seq = seq;
556
557         dccp_transmit_skb(sk, skb);
558 }
559
560 EXPORT_SYMBOL_GPL(dccp_send_sync);
561
562 /*
563  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
564  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
565  * any circumstances.
566  */
567 void dccp_send_close(struct sock *sk, const int active)
568 {
569         struct dccp_sock *dp = dccp_sk(sk);
570         struct sk_buff *skb;
571         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
572
573         skb = alloc_skb(sk->sk_prot->max_header, prio);
574         if (skb == NULL)
575                 return;
576
577         /* Reserve space for headers and prepare control bits. */
578         skb_reserve(skb, sk->sk_prot->max_header);
579         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
580                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
581
582         if (active) {
583                 dccp_write_xmit(sk, 1);
584                 dccp_skb_entail(sk, skb);
585                 dccp_transmit_skb(sk, skb_clone(skb, prio));
586                 /* FIXME do we need a retransmit timer here? */
587         } else
588                 dccp_transmit_skb(sk, skb);
589 }