[DCCP]: Move the IPv4 specific bits from proto.c to ipv4.c
[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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/kernel.h>
16 #include <linux/skbuff.h>
17
18 #include <net/inet_sock.h>
19 #include <net/sock.h>
20
21 #include "ackvec.h"
22 #include "ccid.h"
23 #include "dccp.h"
24
25 static inline void dccp_event_ack_sent(struct sock *sk)
26 {
27         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
28 }
29
30 static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
31 {
32         skb_set_owner_w(skb, sk);
33         WARN_ON(sk->sk_send_head);
34         sk->sk_send_head = skb;
35 }
36
37 /*
38  * All SKB's seen here are completely headerless. It is our
39  * job to build the DCCP header, and pass the packet down to
40  * IP so it can do the same plus pass the packet off to the
41  * device.
42  */
43 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
44 {
45         if (likely(skb != NULL)) {
46                 const struct inet_sock *inet = inet_sk(sk);
47                 const struct inet_connection_sock *icsk = inet_csk(sk);
48                 struct dccp_sock *dp = dccp_sk(sk);
49                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
50                 struct dccp_hdr *dh;
51                 /* XXX For now we're using only 48 bits sequence numbers */
52                 const int dccp_header_size = sizeof(*dh) +
53                                              sizeof(struct dccp_hdr_ext) +
54                                           dccp_packet_hdr_len(dcb->dccpd_type);
55                 int err, set_ack = 1;
56                 u64 ackno = dp->dccps_gsr;
57
58                 dccp_inc_seqno(&dp->dccps_gss);
59
60                 switch (dcb->dccpd_type) {
61                 case DCCP_PKT_DATA:
62                         set_ack = 0;
63                         /* fall through */
64                 case DCCP_PKT_DATAACK:
65                         break;
66
67                 case DCCP_PKT_REQUEST:
68                         set_ack = 0;
69                         /* fall through */
70
71                 case DCCP_PKT_SYNC:
72                 case DCCP_PKT_SYNCACK:
73                         ackno = dcb->dccpd_seq;
74                         /* fall through */
75                 default:
76                         /*
77                          * Only data packets should come through with skb->sk
78                          * set.
79                          */
80                         WARN_ON(skb->sk);
81                         skb_set_owner_w(skb, sk);
82                         break;
83                 }
84
85                 dcb->dccpd_seq = dp->dccps_gss;
86                 dccp_insert_options(sk, skb);
87                 
88                 skb->h.raw = skb_push(skb, dccp_header_size);
89                 dh = dccp_hdr(skb);
90
91                 /* Build DCCP header and checksum it. */
92                 memset(dh, 0, dccp_header_size);
93                 dh->dccph_type  = dcb->dccpd_type;
94                 dh->dccph_sport = inet->sport;
95                 dh->dccph_dport = inet->dport;
96                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
97                 dh->dccph_ccval = dcb->dccpd_ccval;
98                 /* XXX For now we're using only 48 bits sequence numbers */
99                 dh->dccph_x     = 1;
100
101                 dp->dccps_awh = dp->dccps_gss;
102                 dccp_hdr_set_seq(dh, dp->dccps_gss);
103                 if (set_ack)
104                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
105
106                 switch (dcb->dccpd_type) {
107                 case DCCP_PKT_REQUEST:
108                         dccp_hdr_request(skb)->dccph_req_service =
109                                                         dp->dccps_service;
110                         break;
111                 case DCCP_PKT_RESET:
112                         dccp_hdr_reset(skb)->dccph_reset_code =
113                                                         dcb->dccpd_reset_code;
114                         break;
115                 }
116
117                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
118
119                 if (set_ack)
120                         dccp_event_ack_sent(sk);
121
122                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
123
124                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
125                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
126                 if (err <= 0)
127                         return err;
128
129                 /* NET_XMIT_CN is special. It does not guarantee,
130                  * that this packet is lost. It tells that device
131                  * is about to start to drop packets or already
132                  * drops some packets of the same priority and
133                  * invokes us to send less aggressively.
134                  */
135                 return err == NET_XMIT_CN ? 0 : err;
136         }
137         return -ENOBUFS;
138 }
139
140 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
141 {
142         struct inet_connection_sock *icsk = inet_csk(sk);
143         struct dccp_sock *dp = dccp_sk(sk);
144         int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
145                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
146
147         /* Now subtract optional transport overhead */
148         mss_now -= icsk->icsk_ext_hdr_len;
149
150         /*
151          * FIXME: this should come from the CCID infrastructure, where, say,
152          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
153          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
154          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
155          * make it a multiple of 4
156          */
157
158         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
159
160         /* And store cached results */
161         icsk->icsk_pmtu_cookie = pmtu;
162         dp->dccps_mss_cache = mss_now;
163
164         return mss_now;
165 }
166
167 EXPORT_SYMBOL_GPL(dccp_sync_mss);
168
169 void dccp_write_space(struct sock *sk)
170 {
171         read_lock(&sk->sk_callback_lock);
172
173         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
174                 wake_up_interruptible(sk->sk_sleep);
175         /* Should agree with poll, otherwise some programs break */
176         if (sock_writeable(sk))
177                 sk_wake_async(sk, 2, POLL_OUT);
178
179         read_unlock(&sk->sk_callback_lock);
180 }
181
182 /**
183  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
184  * @sk: socket to wait for
185  * @timeo: for how long
186  */
187 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
188                               long *timeo)
189 {
190         struct dccp_sock *dp = dccp_sk(sk);
191         DEFINE_WAIT(wait);
192         long delay;
193         int rc;
194
195         while (1) {
196                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
197
198                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
199                         goto do_error;
200                 if (!*timeo)
201                         goto do_nonblock;
202                 if (signal_pending(current))
203                         goto do_interrupted;
204
205                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
206                                             skb->len);
207                 if (rc <= 0)
208                         break;
209                 delay = msecs_to_jiffies(rc);
210                 if (delay > *timeo || delay < 0)
211                         goto do_nonblock;
212
213                 sk->sk_write_pending++;
214                 release_sock(sk);
215                 *timeo -= schedule_timeout(delay);
216                 lock_sock(sk);
217                 sk->sk_write_pending--;
218         }
219 out:
220         finish_wait(sk->sk_sleep, &wait);
221         return rc;
222
223 do_error:
224         rc = -EPIPE;
225         goto out;
226 do_nonblock:
227         rc = -EAGAIN;
228         goto out;
229 do_interrupted:
230         rc = sock_intr_errno(*timeo);
231         goto out;
232 }
233
234 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
235 {
236         const struct dccp_sock *dp = dccp_sk(sk);
237         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
238                                          skb->len);
239
240         if (err > 0)
241                 err = dccp_wait_for_ccid(sk, skb, timeo);
242
243         if (err == 0) {
244                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
245                 const int len = skb->len;
246
247                 if (sk->sk_state == DCCP_PARTOPEN) {
248                         /* See 8.1.5.  Handshake Completion */
249                         inet_csk_schedule_ack(sk);
250                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
251                                                   inet_csk(sk)->icsk_rto,
252                                                   DCCP_RTO_MAX);
253                         dcb->dccpd_type = DCCP_PKT_DATAACK;
254                 } else if (dccp_ack_pending(sk))
255                         dcb->dccpd_type = DCCP_PKT_DATAACK;
256                 else
257                         dcb->dccpd_type = DCCP_PKT_DATA;
258
259                 err = dccp_transmit_skb(sk, skb);
260                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
261         } else
262                 kfree_skb(skb);
263
264         return err;
265 }
266
267 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
268 {
269         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
270                 return -EHOSTUNREACH; /* Routing failure or similar. */
271
272         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
273                                       pskb_copy(skb, GFP_ATOMIC):
274                                       skb_clone(skb, GFP_ATOMIC)));
275 }
276
277 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
278                                    struct request_sock *req)
279 {
280         struct dccp_hdr *dh;
281         struct dccp_request_sock *dreq;
282         const int dccp_header_size = sizeof(struct dccp_hdr) +
283                                      sizeof(struct dccp_hdr_ext) +
284                                      sizeof(struct dccp_hdr_response);
285         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
286                                                dccp_header_size, 1,
287                                            GFP_ATOMIC);
288         if (skb == NULL)
289                 return NULL;
290
291         /* Reserve space for headers. */
292         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
293
294         skb->dst = dst_clone(dst);
295         skb->csum = 0;
296
297         dreq = dccp_rsk(req);
298         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
299         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
300         dccp_insert_options(sk, skb);
301
302         skb->h.raw = skb_push(skb, dccp_header_size);
303
304         dh = dccp_hdr(skb);
305         memset(dh, 0, dccp_header_size);
306
307         dh->dccph_sport = inet_sk(sk)->sport;
308         dh->dccph_dport = inet_rsk(req)->rmt_port;
309         dh->dccph_doff  = (dccp_header_size +
310                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
311         dh->dccph_type  = DCCP_PKT_RESPONSE;
312         dh->dccph_x     = 1;
313         dccp_hdr_set_seq(dh, dreq->dreq_iss);
314         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
315         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
316
317         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
318         return skb;
319 }
320
321 EXPORT_SYMBOL_GPL(dccp_make_response);
322
323 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
324                                        const enum dccp_reset_codes code)
325                                    
326 {
327         struct dccp_hdr *dh;
328         struct dccp_sock *dp = dccp_sk(sk);
329         const int dccp_header_size = sizeof(struct dccp_hdr) +
330                                      sizeof(struct dccp_hdr_ext) +
331                                      sizeof(struct dccp_hdr_reset);
332         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
333                                                dccp_header_size, 1,
334                                            GFP_ATOMIC);
335         if (skb == NULL)
336                 return NULL;
337
338         /* Reserve space for headers. */
339         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
340
341         skb->dst = dst_clone(dst);
342         skb->csum = 0;
343
344         dccp_inc_seqno(&dp->dccps_gss);
345
346         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
347         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
348         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
349         dccp_insert_options(sk, skb);
350
351         skb->h.raw = skb_push(skb, dccp_header_size);
352
353         dh = dccp_hdr(skb);
354         memset(dh, 0, dccp_header_size);
355
356         dh->dccph_sport = inet_sk(sk)->sport;
357         dh->dccph_dport = inet_sk(sk)->dport;
358         dh->dccph_doff  = (dccp_header_size +
359                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
360         dh->dccph_type  = DCCP_PKT_RESET;
361         dh->dccph_x     = 1;
362         dccp_hdr_set_seq(dh, dp->dccps_gss);
363         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
364
365         dccp_hdr_reset(skb)->dccph_reset_code = code;
366         inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
367
368         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
369         return skb;
370 }
371
372 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
373 {
374         /*
375          * FIXME: what if rebuild_header fails?
376          * Should we be doing a rebuild_header here?
377          */
378         int err = inet_sk_rebuild_header(sk);
379
380         if (err == 0) {
381                 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
382                                                       code);
383                 if (skb != NULL) {
384                         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
385                         err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
386                         if (err == NET_XMIT_CN)
387                                 err = 0;
388                 }
389         }
390
391         return err;
392 }
393
394 /*
395  * Do all connect socket setups that can be done AF independent.
396  */
397 static inline void dccp_connect_init(struct sock *sk)
398 {
399         struct dccp_sock *dp = dccp_sk(sk);
400         struct dst_entry *dst = __sk_dst_get(sk);
401         struct inet_connection_sock *icsk = inet_csk(sk);
402
403         sk->sk_err = 0;
404         sock_reset_flag(sk, SOCK_DONE);
405         
406         dccp_sync_mss(sk, dst_mtu(dst));
407
408         dccp_update_gss(sk, dp->dccps_iss);
409         /*
410          * SWL and AWL are initially adjusted so that they are not less than
411          * the initial Sequence Numbers received and sent, respectively:
412          *      SWL := max(GSR + 1 - floor(W/4), ISR),
413          *      AWL := max(GSS - W' + 1, ISS).
414          * These adjustments MUST be applied only at the beginning of the
415          * connection.
416          */
417         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
418
419         icsk->icsk_retransmits = 0;
420 }
421
422 int dccp_connect(struct sock *sk)
423 {
424         struct sk_buff *skb;
425         struct inet_connection_sock *icsk = inet_csk(sk);
426
427         dccp_connect_init(sk);
428
429         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
430         if (unlikely(skb == NULL))
431                 return -ENOBUFS;
432
433         /* Reserve space for headers. */
434         skb_reserve(skb, MAX_DCCP_HEADER);
435
436         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
437         skb->csum = 0;
438
439         dccp_skb_entail(sk, skb);
440         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
441         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
442
443         /* Timer for repeating the REQUEST until an answer. */
444         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
445                                   icsk->icsk_rto, DCCP_RTO_MAX);
446         return 0;
447 }
448
449 EXPORT_SYMBOL_GPL(dccp_connect);
450
451 void dccp_send_ack(struct sock *sk)
452 {
453         /* If we have been reset, we may not send again. */
454         if (sk->sk_state != DCCP_CLOSED) {
455                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
456
457                 if (skb == NULL) {
458                         inet_csk_schedule_ack(sk);
459                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
460                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
461                                                   TCP_DELACK_MAX,
462                                                   DCCP_RTO_MAX);
463                         return;
464                 }
465
466                 /* Reserve space for headers */
467                 skb_reserve(skb, MAX_DCCP_HEADER);
468                 skb->csum = 0;
469                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
470                 dccp_transmit_skb(sk, skb);
471         }
472 }
473
474 EXPORT_SYMBOL_GPL(dccp_send_ack);
475
476 void dccp_send_delayed_ack(struct sock *sk)
477 {
478         struct inet_connection_sock *icsk = inet_csk(sk);
479         /*
480          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
481          * with using 2s, and active senders also piggyback the ACK into a
482          * DATAACK packet, so this is really for quiescent senders.
483          */
484         unsigned long timeout = jiffies + 2 * HZ;
485
486         /* Use new timeout only if there wasn't a older one earlier. */
487         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
488                 /* If delack timer was blocked or is about to expire,
489                  * send ACK now.
490                  *
491                  * FIXME: check the "about to expire" part
492                  */
493                 if (icsk->icsk_ack.blocked) {
494                         dccp_send_ack(sk);
495                         return;
496                 }
497
498                 if (!time_before(timeout, icsk->icsk_ack.timeout))
499                         timeout = icsk->icsk_ack.timeout;
500         }
501         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
502         icsk->icsk_ack.timeout = timeout;
503         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
504 }
505
506 void dccp_send_sync(struct sock *sk, const u64 seq,
507                     const enum dccp_pkt_type pkt_type)
508 {
509         /*
510          * We are not putting this on the write queue, so
511          * dccp_transmit_skb() will set the ownership to this
512          * sock.
513          */
514         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
515
516         if (skb == NULL)
517                 /* FIXME: how to make sure the sync is sent? */
518                 return;
519
520         /* Reserve space for headers and prepare control bits. */
521         skb_reserve(skb, MAX_DCCP_HEADER);
522         skb->csum = 0;
523         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
524         DCCP_SKB_CB(skb)->dccpd_seq = seq;
525
526         dccp_transmit_skb(sk, skb);
527 }
528
529 EXPORT_SYMBOL_GPL(dccp_send_sync);
530
531 /*
532  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
533  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
534  * any circumstances.
535  */
536 void dccp_send_close(struct sock *sk, const int active)
537 {
538         struct dccp_sock *dp = dccp_sk(sk);
539         struct sk_buff *skb;
540         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
541
542         skb = alloc_skb(sk->sk_prot->max_header, prio);
543         if (skb == NULL)
544                 return;
545
546         /* Reserve space for headers and prepare control bits. */
547         skb_reserve(skb, sk->sk_prot->max_header);
548         skb->csum = 0;
549         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
550                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
551
552         if (active) {
553                 dccp_skb_entail(sk, skb);
554                 dccp_transmit_skb(sk, skb_clone(skb, prio));
555         } else
556                 dccp_transmit_skb(sk, skb);
557 }