Merge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerp...
[linux-2.6] / 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, 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  */
179 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb)
180 {
181         struct dccp_sock *dp = dccp_sk(sk);
182         DEFINE_WAIT(wait);
183         unsigned long delay;
184         int rc;
185
186         while (1) {
187                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
188
189                 if (sk->sk_err)
190                         goto do_error;
191                 if (signal_pending(current))
192                         goto do_interrupted;
193
194                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
195                 if (rc <= 0)
196                         break;
197                 delay = msecs_to_jiffies(rc);
198                 sk->sk_write_pending++;
199                 release_sock(sk);
200                 schedule_timeout(delay);
201                 lock_sock(sk);
202                 sk->sk_write_pending--;
203         }
204 out:
205         finish_wait(sk->sk_sleep, &wait);
206         return rc;
207
208 do_error:
209         rc = -EPIPE;
210         goto out;
211 do_interrupted:
212         rc = -EINTR;
213         goto out;
214 }
215
216 static void dccp_write_xmit_timer(unsigned long data) {
217         struct sock *sk = (struct sock *)data;
218         struct dccp_sock *dp = dccp_sk(sk);
219
220         bh_lock_sock(sk);
221         if (sock_owned_by_user(sk))
222                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
223         else
224                 dccp_write_xmit(sk, 0);
225         bh_unlock_sock(sk);
226         sock_put(sk);
227 }
228
229 void dccp_write_xmit(struct sock *sk, int block)
230 {
231         struct dccp_sock *dp = dccp_sk(sk);
232         struct sk_buff *skb;
233
234         while ((skb = skb_peek(&sk->sk_write_queue))) {
235                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
236
237                 if (err > 0) {
238                         if (!block) {
239                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
240                                                 msecs_to_jiffies(err)+jiffies);
241                                 break;
242                         } else
243                                 err = dccp_wait_for_ccid(sk, skb);
244                         if (err && err != -EINTR)
245                                 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
246                 }
247
248                 skb_dequeue(&sk->sk_write_queue);
249                 if (err == 0) {
250                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
251                         const int len = skb->len;
252
253                         if (sk->sk_state == DCCP_PARTOPEN) {
254                                 /* See 8.1.5.  Handshake Completion */
255                                 inet_csk_schedule_ack(sk);
256                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
257                                                   inet_csk(sk)->icsk_rto,
258                                                   DCCP_RTO_MAX);
259                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
260                         } else if (dccp_ack_pending(sk))
261                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
262                         else
263                                 dcb->dccpd_type = DCCP_PKT_DATA;
264
265                         err = dccp_transmit_skb(sk, skb);
266                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
267                         if (err)
268                                 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
269                                          err);
270                 } else {
271                         dccp_pr_debug("packet discarded\n");
272                         kfree(skb);
273                 }
274         }
275 }
276
277 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
278 {
279         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
280                 return -EHOSTUNREACH; /* Routing failure or similar. */
281
282         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
283                                       pskb_copy(skb, GFP_ATOMIC):
284                                       skb_clone(skb, GFP_ATOMIC)));
285 }
286
287 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
288                                    struct request_sock *req)
289 {
290         struct dccp_hdr *dh;
291         struct dccp_request_sock *dreq;
292         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
293                                      sizeof(struct dccp_hdr_ext) +
294                                      sizeof(struct dccp_hdr_response);
295         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
296                                            GFP_ATOMIC);
297         if (skb == NULL)
298                 return NULL;
299
300         /* Reserve space for headers. */
301         skb_reserve(skb, sk->sk_prot->max_header);
302
303         skb->dst = dst_clone(dst);
304
305         dreq = dccp_rsk(req);
306         if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
307                 dccp_inc_seqno(&dreq->dreq_iss);
308         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
309         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
310
311         if (dccp_insert_options(sk, skb)) {
312                 kfree_skb(skb);
313                 return NULL;
314         }
315
316         /* Build and checksum header */
317         dh = dccp_zeroed_hdr(skb, dccp_header_size);
318
319         dh->dccph_sport = inet_sk(sk)->sport;
320         dh->dccph_dport = inet_rsk(req)->rmt_port;
321         dh->dccph_doff  = (dccp_header_size +
322                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
323         dh->dccph_type  = DCCP_PKT_RESPONSE;
324         dh->dccph_x     = 1;
325         dccp_hdr_set_seq(dh, dreq->dreq_iss);
326         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
327         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
328
329         dccp_csum_outgoing(skb);
330
331         /* We use `acked' to remember that a Response was already sent. */
332         inet_rsk(req)->acked = 1;
333         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
334         return skb;
335 }
336
337 EXPORT_SYMBOL_GPL(dccp_make_response);
338
339 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
340                                        const enum dccp_reset_codes code)
341 {
342         struct dccp_hdr *dh;
343         struct dccp_sock *dp = dccp_sk(sk);
344         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
345                                      sizeof(struct dccp_hdr_ext) +
346                                      sizeof(struct dccp_hdr_reset);
347         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
348                                            GFP_ATOMIC);
349         if (skb == NULL)
350                 return NULL;
351
352         /* Reserve space for headers. */
353         skb_reserve(skb, sk->sk_prot->max_header);
354
355         skb->dst = dst_clone(dst);
356
357         dccp_inc_seqno(&dp->dccps_gss);
358
359         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
360         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
361         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
362
363         if (dccp_insert_options(sk, skb)) {
364                 kfree_skb(skb);
365                 return NULL;
366         }
367
368         dh = dccp_zeroed_hdr(skb, dccp_header_size);
369
370         dh->dccph_sport = inet_sk(sk)->sport;
371         dh->dccph_dport = inet_sk(sk)->dport;
372         dh->dccph_doff  = (dccp_header_size +
373                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
374         dh->dccph_type  = DCCP_PKT_RESET;
375         dh->dccph_x     = 1;
376         dccp_hdr_set_seq(dh, dp->dccps_gss);
377         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
378
379         dccp_hdr_reset(skb)->dccph_reset_code = code;
380         inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
381
382         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
383         return skb;
384 }
385
386 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
387 {
388         /*
389          * FIXME: what if rebuild_header fails?
390          * Should we be doing a rebuild_header here?
391          */
392         int err = inet_sk_rebuild_header(sk);
393
394         if (err == 0) {
395                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
396                                                       code);
397                 if (skb != NULL) {
398                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
399                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
400                         return net_xmit_eval(err);
401                 }
402         }
403
404         return err;
405 }
406
407 /*
408  * Do all connect socket setups that can be done AF independent.
409  */
410 static inline void dccp_connect_init(struct sock *sk)
411 {
412         struct dccp_sock *dp = dccp_sk(sk);
413         struct dst_entry *dst = __sk_dst_get(sk);
414         struct inet_connection_sock *icsk = inet_csk(sk);
415
416         sk->sk_err = 0;
417         sock_reset_flag(sk, SOCK_DONE);
418         
419         dccp_sync_mss(sk, dst_mtu(dst));
420
421         /*
422          * SWL and AWL are initially adjusted so that they are not less than
423          * the initial Sequence Numbers received and sent, respectively:
424          *      SWL := max(GSR + 1 - floor(W/4), ISR),
425          *      AWL := max(GSS - W' + 1, ISS).
426          * These adjustments MUST be applied only at the beginning of the
427          * connection.
428          */
429         dccp_update_gss(sk, dp->dccps_iss);
430         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
431
432         /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
433          *         initialized to S.ISS (sec. 8.5)                            */
434         dp->dccps_gar = dp->dccps_iss;
435
436         icsk->icsk_retransmits = 0;
437         init_timer(&dp->dccps_xmit_timer);
438         dp->dccps_xmit_timer.data = (unsigned long)sk;
439         dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
440 }
441
442 int dccp_connect(struct sock *sk)
443 {
444         struct sk_buff *skb;
445         struct inet_connection_sock *icsk = inet_csk(sk);
446
447         dccp_connect_init(sk);
448
449         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
450         if (unlikely(skb == NULL))
451                 return -ENOBUFS;
452
453         /* Reserve space for headers. */
454         skb_reserve(skb, sk->sk_prot->max_header);
455
456         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
457
458         dccp_skb_entail(sk, skb);
459         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
460         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
461
462         /* Timer for repeating the REQUEST until an answer. */
463         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
464                                   icsk->icsk_rto, DCCP_RTO_MAX);
465         return 0;
466 }
467
468 EXPORT_SYMBOL_GPL(dccp_connect);
469
470 void dccp_send_ack(struct sock *sk)
471 {
472         /* If we have been reset, we may not send again. */
473         if (sk->sk_state != DCCP_CLOSED) {
474                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
475                                                 GFP_ATOMIC);
476
477                 if (skb == NULL) {
478                         inet_csk_schedule_ack(sk);
479                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
480                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
481                                                   TCP_DELACK_MAX,
482                                                   DCCP_RTO_MAX);
483                         return;
484                 }
485
486                 /* Reserve space for headers */
487                 skb_reserve(skb, sk->sk_prot->max_header);
488                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
489                 dccp_transmit_skb(sk, skb);
490         }
491 }
492
493 EXPORT_SYMBOL_GPL(dccp_send_ack);
494
495 void dccp_send_delayed_ack(struct sock *sk)
496 {
497         struct inet_connection_sock *icsk = inet_csk(sk);
498         /*
499          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
500          * with using 2s, and active senders also piggyback the ACK into a
501          * DATAACK packet, so this is really for quiescent senders.
502          */
503         unsigned long timeout = jiffies + 2 * HZ;
504
505         /* Use new timeout only if there wasn't a older one earlier. */
506         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
507                 /* If delack timer was blocked or is about to expire,
508                  * send ACK now.
509                  *
510                  * FIXME: check the "about to expire" part
511                  */
512                 if (icsk->icsk_ack.blocked) {
513                         dccp_send_ack(sk);
514                         return;
515                 }
516
517                 if (!time_before(timeout, icsk->icsk_ack.timeout))
518                         timeout = icsk->icsk_ack.timeout;
519         }
520         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
521         icsk->icsk_ack.timeout = timeout;
522         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
523 }
524
525 void dccp_send_sync(struct sock *sk, const u64 seq,
526                     const enum dccp_pkt_type pkt_type)
527 {
528         /*
529          * We are not putting this on the write queue, so
530          * dccp_transmit_skb() will set the ownership to this
531          * sock.
532          */
533         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
534
535         if (skb == NULL)
536                 /* FIXME: how to make sure the sync is sent? */
537                 return;
538
539         /* Reserve space for headers and prepare control bits. */
540         skb_reserve(skb, sk->sk_prot->max_header);
541         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
542         DCCP_SKB_CB(skb)->dccpd_seq = seq;
543
544         dccp_transmit_skb(sk, skb);
545 }
546
547 EXPORT_SYMBOL_GPL(dccp_send_sync);
548
549 /*
550  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
551  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
552  * any circumstances.
553  */
554 void dccp_send_close(struct sock *sk, const int active)
555 {
556         struct dccp_sock *dp = dccp_sk(sk);
557         struct sk_buff *skb;
558         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
559
560         skb = alloc_skb(sk->sk_prot->max_header, prio);
561         if (skb == NULL)
562                 return;
563
564         /* Reserve space for headers and prepare control bits. */
565         skb_reserve(skb, sk->sk_prot->max_header);
566         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
567                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
568
569         if (active) {
570                 dccp_write_xmit(sk, 1);
571                 dccp_skb_entail(sk, skb);
572                 dccp_transmit_skb(sk, skb_clone(skb, prio));
573                 /* FIXME do we need a retransmit timer here? */
574         } else
575                 dccp_transmit_skb(sk, skb);
576 }