[DCCP]: Support for partial checksums (RFC 4340, sec. 9.2)
[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                         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                 if (err <= 0)
129                         return err;
130
131                 /* NET_XMIT_CN is special. It does not guarantee,
132                  * that this packet is lost. It tells that device
133                  * is about to start to drop packets or already
134                  * drops some packets of the same priority and
135                  * invokes us to send less aggressively.
136                  */
137                 return err == NET_XMIT_CN ? 0 : err;
138         }
139         return -ENOBUFS;
140 }
141
142 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
143 {
144         struct inet_connection_sock *icsk = inet_csk(sk);
145         struct dccp_sock *dp = dccp_sk(sk);
146         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
147                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
148
149         /* Now subtract optional transport overhead */
150         mss_now -= icsk->icsk_ext_hdr_len;
151
152         /*
153          * FIXME: this should come from the CCID infrastructure, where, say,
154          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
155          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
156          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
157          * make it a multiple of 4
158          */
159
160         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
161
162         /* And store cached results */
163         icsk->icsk_pmtu_cookie = pmtu;
164         dp->dccps_mss_cache = mss_now;
165
166         return mss_now;
167 }
168
169 EXPORT_SYMBOL_GPL(dccp_sync_mss);
170
171 void dccp_write_space(struct sock *sk)
172 {
173         read_lock(&sk->sk_callback_lock);
174
175         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
176                 wake_up_interruptible(sk->sk_sleep);
177         /* Should agree with poll, otherwise some programs break */
178         if (sock_writeable(sk))
179                 sk_wake_async(sk, 2, POLL_OUT);
180
181         read_unlock(&sk->sk_callback_lock);
182 }
183
184 /**
185  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
186  * @sk: socket to wait for
187  * @timeo: for how long
188  */
189 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
190                               long *timeo)
191 {
192         struct dccp_sock *dp = dccp_sk(sk);
193         DEFINE_WAIT(wait);
194         long delay;
195         int rc;
196
197         while (1) {
198                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
199
200                 if (sk->sk_err)
201                         goto do_error;
202                 if (!*timeo)
203                         goto do_nonblock;
204                 if (signal_pending(current))
205                         goto do_interrupted;
206
207                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
208                                             skb->len);
209                 if (rc <= 0)
210                         break;
211                 delay = msecs_to_jiffies(rc);
212                 if (delay > *timeo || delay < 0)
213                         goto do_nonblock;
214
215                 sk->sk_write_pending++;
216                 release_sock(sk);
217                 *timeo -= schedule_timeout(delay);
218                 lock_sock(sk);
219                 sk->sk_write_pending--;
220         }
221 out:
222         finish_wait(sk->sk_sleep, &wait);
223         return rc;
224
225 do_error:
226         rc = -EPIPE;
227         goto out;
228 do_nonblock:
229         rc = -EAGAIN;
230         goto out;
231 do_interrupted:
232         rc = sock_intr_errno(*timeo);
233         goto out;
234 }
235
236 static void dccp_write_xmit_timer(unsigned long data) {
237         struct sock *sk = (struct sock *)data;
238         struct dccp_sock *dp = dccp_sk(sk);
239
240         bh_lock_sock(sk);
241         if (sock_owned_by_user(sk))
242                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
243         else
244                 dccp_write_xmit(sk, 0);
245         bh_unlock_sock(sk);
246         sock_put(sk);
247 }
248
249 void dccp_write_xmit(struct sock *sk, int block)
250 {
251         struct dccp_sock *dp = dccp_sk(sk);
252         struct sk_buff *skb;
253         long timeo = DCCP_XMIT_TIMEO;   /* If a packet is taking longer than
254                                            this we have other issues */
255
256         while ((skb = skb_peek(&sk->sk_write_queue))) {
257                 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
258                                          skb->len);
259
260                 if (err > 0) {
261                         if (!block) {
262                                 sk_reset_timer(sk, &dp->dccps_xmit_timer,
263                                                 msecs_to_jiffies(err)+jiffies);
264                                 break;
265                         } else {
266                                 err = dccp_wait_for_ccid(sk, skb, &timeo);
267                                 timeo = DCCP_XMIT_TIMEO;
268                         }
269                         if (err) {
270                                 printk(KERN_CRIT "%s:err at dccp_wait_for_ccid"
271                                                  " %d\n", __FUNCTION__, err);
272                                 dump_stack();
273                         }
274                 }
275
276                 skb_dequeue(&sk->sk_write_queue);
277                 if (err == 0) {
278                         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
279                         const int len = skb->len;
280
281                         if (sk->sk_state == DCCP_PARTOPEN) {
282                                 /* See 8.1.5.  Handshake Completion */
283                                 inet_csk_schedule_ack(sk);
284                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
285                                                   inet_csk(sk)->icsk_rto,
286                                                   DCCP_RTO_MAX);
287                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
288                         } else if (dccp_ack_pending(sk))
289                                 dcb->dccpd_type = DCCP_PKT_DATAACK;
290                         else
291                                 dcb->dccpd_type = DCCP_PKT_DATA;
292
293                         err = dccp_transmit_skb(sk, skb);
294                         ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
295                         if (err) {
296                                 printk(KERN_CRIT "%s:err from "
297                                                  "ccid_hc_tx_packet_sent %d\n",
298                                                  __FUNCTION__, err);
299                                 dump_stack();
300                         }
301                 } else
302                         kfree(skb);
303         }
304 }
305
306 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
307 {
308         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
309                 return -EHOSTUNREACH; /* Routing failure or similar. */
310
311         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
312                                       pskb_copy(skb, GFP_ATOMIC):
313                                       skb_clone(skb, GFP_ATOMIC)));
314 }
315
316 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
317                                    struct request_sock *req)
318 {
319         struct dccp_hdr *dh;
320         struct dccp_request_sock *dreq;
321         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
322                                      sizeof(struct dccp_hdr_ext) +
323                                      sizeof(struct dccp_hdr_response);
324         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
325                                            GFP_ATOMIC);
326         if (skb == NULL)
327                 return NULL;
328
329         /* Reserve space for headers. */
330         skb_reserve(skb, sk->sk_prot->max_header);
331
332         skb->dst = dst_clone(dst);
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         dh = dccp_zeroed_hdr(skb, dccp_header_size);
344
345         dh->dccph_sport = inet_sk(sk)->sport;
346         dh->dccph_dport = inet_rsk(req)->rmt_port;
347         dh->dccph_doff  = (dccp_header_size +
348                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
349         dh->dccph_type  = DCCP_PKT_RESPONSE;
350         dh->dccph_x     = 1;
351         dccp_hdr_set_seq(dh, dreq->dreq_iss);
352         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
353         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
354
355         dccp_csum_outgoing(skb);
356
357         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
358         return skb;
359 }
360
361 EXPORT_SYMBOL_GPL(dccp_make_response);
362
363 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
364                                        const enum dccp_reset_codes code)
365                                    
366 {
367         struct dccp_hdr *dh;
368         struct dccp_sock *dp = dccp_sk(sk);
369         const u32 dccp_header_size = sizeof(struct dccp_hdr) +
370                                      sizeof(struct dccp_hdr_ext) +
371                                      sizeof(struct dccp_hdr_reset);
372         struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
373                                            GFP_ATOMIC);
374         if (skb == NULL)
375                 return NULL;
376
377         /* Reserve space for headers. */
378         skb_reserve(skb, sk->sk_prot->max_header);
379
380         skb->dst = dst_clone(dst);
381
382         dccp_inc_seqno(&dp->dccps_gss);
383
384         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
385         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
386         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
387
388         if (dccp_insert_options(sk, skb)) {
389                 kfree_skb(skb);
390                 return NULL;
391         }
392
393         dh = dccp_zeroed_hdr(skb, dccp_header_size);
394
395         dh->dccph_sport = inet_sk(sk)->sport;
396         dh->dccph_dport = inet_sk(sk)->dport;
397         dh->dccph_doff  = (dccp_header_size +
398                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
399         dh->dccph_type  = DCCP_PKT_RESET;
400         dh->dccph_x     = 1;
401         dccp_hdr_set_seq(dh, dp->dccps_gss);
402         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
403
404         dccp_hdr_reset(skb)->dccph_reset_code = code;
405         inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
406
407         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
408         return skb;
409 }
410
411 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
412 {
413         /*
414          * FIXME: what if rebuild_header fails?
415          * Should we be doing a rebuild_header here?
416          */
417         int err = inet_sk_rebuild_header(sk);
418
419         if (err == 0) {
420                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
421                                                       code);
422                 if (skb != NULL) {
423                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
424                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, sk, 0);
425                         if (err == NET_XMIT_CN)
426                                 err = 0;
427                 }
428         }
429
430         return err;
431 }
432
433 /*
434  * Do all connect socket setups that can be done AF independent.
435  */
436 static inline void dccp_connect_init(struct sock *sk)
437 {
438         struct dccp_sock *dp = dccp_sk(sk);
439         struct dst_entry *dst = __sk_dst_get(sk);
440         struct inet_connection_sock *icsk = inet_csk(sk);
441
442         sk->sk_err = 0;
443         sock_reset_flag(sk, SOCK_DONE);
444         
445         dccp_sync_mss(sk, dst_mtu(dst));
446
447         dccp_update_gss(sk, dp->dccps_iss);
448         /*
449          * SWL and AWL are initially adjusted so that they are not less than
450          * the initial Sequence Numbers received and sent, respectively:
451          *      SWL := max(GSR + 1 - floor(W/4), ISR),
452          *      AWL := max(GSS - W' + 1, ISS).
453          * These adjustments MUST be applied only at the beginning of the
454          * connection.
455          */
456         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
457
458         icsk->icsk_retransmits = 0;
459         init_timer(&dp->dccps_xmit_timer);
460         dp->dccps_xmit_timer.data = (unsigned long)sk;
461         dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
462 }
463
464 int dccp_connect(struct sock *sk)
465 {
466         struct sk_buff *skb;
467         struct inet_connection_sock *icsk = inet_csk(sk);
468
469         dccp_connect_init(sk);
470
471         skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
472         if (unlikely(skb == NULL))
473                 return -ENOBUFS;
474
475         /* Reserve space for headers. */
476         skb_reserve(skb, sk->sk_prot->max_header);
477
478         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
479
480         dccp_skb_entail(sk, skb);
481         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
482         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
483
484         /* Timer for repeating the REQUEST until an answer. */
485         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
486                                   icsk->icsk_rto, DCCP_RTO_MAX);
487         return 0;
488 }
489
490 EXPORT_SYMBOL_GPL(dccp_connect);
491
492 void dccp_send_ack(struct sock *sk)
493 {
494         /* If we have been reset, we may not send again. */
495         if (sk->sk_state != DCCP_CLOSED) {
496                 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
497                                                 GFP_ATOMIC);
498
499                 if (skb == NULL) {
500                         inet_csk_schedule_ack(sk);
501                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
502                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
503                                                   TCP_DELACK_MAX,
504                                                   DCCP_RTO_MAX);
505                         return;
506                 }
507
508                 /* Reserve space for headers */
509                 skb_reserve(skb, sk->sk_prot->max_header);
510                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
511                 dccp_transmit_skb(sk, skb);
512         }
513 }
514
515 EXPORT_SYMBOL_GPL(dccp_send_ack);
516
517 void dccp_send_delayed_ack(struct sock *sk)
518 {
519         struct inet_connection_sock *icsk = inet_csk(sk);
520         /*
521          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
522          * with using 2s, and active senders also piggyback the ACK into a
523          * DATAACK packet, so this is really for quiescent senders.
524          */
525         unsigned long timeout = jiffies + 2 * HZ;
526
527         /* Use new timeout only if there wasn't a older one earlier. */
528         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
529                 /* If delack timer was blocked or is about to expire,
530                  * send ACK now.
531                  *
532                  * FIXME: check the "about to expire" part
533                  */
534                 if (icsk->icsk_ack.blocked) {
535                         dccp_send_ack(sk);
536                         return;
537                 }
538
539                 if (!time_before(timeout, icsk->icsk_ack.timeout))
540                         timeout = icsk->icsk_ack.timeout;
541         }
542         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
543         icsk->icsk_ack.timeout = timeout;
544         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
545 }
546
547 void dccp_send_sync(struct sock *sk, const u64 seq,
548                     const enum dccp_pkt_type pkt_type)
549 {
550         /*
551          * We are not putting this on the write queue, so
552          * dccp_transmit_skb() will set the ownership to this
553          * sock.
554          */
555         struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
556
557         if (skb == NULL)
558                 /* FIXME: how to make sure the sync is sent? */
559                 return;
560
561         /* Reserve space for headers and prepare control bits. */
562         skb_reserve(skb, sk->sk_prot->max_header);
563         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
564         DCCP_SKB_CB(skb)->dccpd_seq = seq;
565
566         dccp_transmit_skb(sk, skb);
567 }
568
569 EXPORT_SYMBOL_GPL(dccp_send_sync);
570
571 /*
572  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
573  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
574  * any circumstances.
575  */
576 void dccp_send_close(struct sock *sk, const int active)
577 {
578         struct dccp_sock *dp = dccp_sk(sk);
579         struct sk_buff *skb;
580         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
581
582         skb = alloc_skb(sk->sk_prot->max_header, prio);
583         if (skb == NULL)
584                 return;
585
586         /* Reserve space for headers and prepare control bits. */
587         skb_reserve(skb, sk->sk_prot->max_header);
588         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
589                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
590
591         if (active) {
592                 dccp_write_xmit(sk, 1);
593                 dccp_skb_entail(sk, skb);
594                 dccp_transmit_skb(sk, skb_clone(skb, prio));
595                 /* FIXME do we need a retransmit timer here? */
596         } else
597                 dccp_transmit_skb(sk, skb);
598 }