projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Pull dock into release branch
[deliverable/linux.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, 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 void dccp_write_xmit(struct sock *sk, int block)
217 {
218 struct dccp_sock *dp = dccp_sk(sk);
219 struct sk_buff *skb;
220
221 while ((skb = skb_peek(&sk->sk_write_queue))) {
222 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
223
224 if (err > 0) {
225 if (!block) {
226 sk_reset_timer(sk, &dp->dccps_xmit_timer,
227 msecs_to_jiffies(err)+jiffies);
228 break;
229 } else
230 err = dccp_wait_for_ccid(sk, skb);
231 if (err && err != -EINTR)
232 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
233 }
234
235 skb_dequeue(&sk->sk_write_queue);
236 if (err == 0) {
237 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
238 const int len = skb->len;
239
240 if (sk->sk_state == DCCP_PARTOPEN) {
241 /* See 8.1.5. Handshake Completion */
242 inet_csk_schedule_ack(sk);
243 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
244 inet_csk(sk)->icsk_rto,
245 DCCP_RTO_MAX);
246 dcb->dccpd_type = DCCP_PKT_DATAACK;
247 } else if (dccp_ack_pending(sk))
248 dcb->dccpd_type = DCCP_PKT_DATAACK;
249 else
250 dcb->dccpd_type = DCCP_PKT_DATA;
251
252 err = dccp_transmit_skb(sk, skb);
253 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
254 if (err)
255 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
256 err);
257 } else {
258 dccp_pr_debug("packet discarded\n");
259 kfree_skb(skb);
260 }
261 }
262 }
263
264 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
265 {
266 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
267 return -EHOSTUNREACH; /* Routing failure or similar. */
268
269 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
270 pskb_copy(skb, GFP_ATOMIC):
271 skb_clone(skb, GFP_ATOMIC)));
272 }
273
274 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
275 struct request_sock *req)
276 {
277 struct dccp_hdr *dh;
278 struct dccp_request_sock *dreq;
279 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
280 sizeof(struct dccp_hdr_ext) +
281 sizeof(struct dccp_hdr_response);
282 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
283 GFP_ATOMIC);
284 if (skb == NULL)
285 return NULL;
286
287 /* Reserve space for headers. */
288 skb_reserve(skb, sk->sk_prot->max_header);
289
290 skb->dst = dst_clone(dst);
291
292 dreq = dccp_rsk(req);
293 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
294 dccp_inc_seqno(&dreq->dreq_iss);
295 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
296 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
297
298 if (dccp_insert_options(sk, skb)) {
299 kfree_skb(skb);
300 return NULL;
301 }
302
303 /* Build and checksum header */
304 dh = dccp_zeroed_hdr(skb, dccp_header_size);
305
306 dh->dccph_sport = inet_sk(sk)->sport;
307 dh->dccph_dport = inet_rsk(req)->rmt_port;
308 dh->dccph_doff = (dccp_header_size +
309 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
310 dh->dccph_type = DCCP_PKT_RESPONSE;
311 dh->dccph_x = 1;
312 dccp_hdr_set_seq(dh, dreq->dreq_iss);
313 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
314 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
315
316 dccp_csum_outgoing(skb);
317
318 /* We use `acked' to remember that a Response was already sent. */
319 inet_rsk(req)->acked = 1;
320 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
321 return skb;
322 }
323
324 EXPORT_SYMBOL_GPL(dccp_make_response);
325
326 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
327 const enum dccp_reset_codes code)
328 {
329 struct dccp_hdr *dh;
330 struct dccp_sock *dp = dccp_sk(sk);
331 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
332 sizeof(struct dccp_hdr_ext) +
333 sizeof(struct dccp_hdr_reset);
334 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
335 GFP_ATOMIC);
336 if (skb == NULL)
337 return NULL;
338
339 /* Reserve space for headers. */
340 skb_reserve(skb, sk->sk_prot->max_header);
341
342 skb->dst = dst_clone(dst);
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
350 if (dccp_insert_options(sk, skb)) {
351 kfree_skb(skb);
352 return NULL;
353 }
354
355 dh = dccp_zeroed_hdr(skb, dccp_header_size);
356
357 dh->dccph_sport = inet_sk(sk)->sport;
358 dh->dccph_dport = inet_sk(sk)->dport;
359 dh->dccph_doff = (dccp_header_size +
360 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
361 dh->dccph_type = DCCP_PKT_RESET;
362 dh->dccph_x = 1;
363 dccp_hdr_set_seq(dh, dp->dccps_gss);
364 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
365
366 dccp_hdr_reset(skb)->dccph_reset_code = code;
367 inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
368
369 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
370 return skb;
371 }
372
373 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
374 {
375 /*
376 * FIXME: what if rebuild_header fails?
377 * Should we be doing a rebuild_header here?
378 */
379 int err = inet_sk_rebuild_header(sk);
380
381 if (err == 0) {
382 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
383 code);
384 if (skb != NULL) {
385 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
386 err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
387 return net_xmit_eval(err);
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 /*
409 * SWL and AWL are initially adjusted so that they are not less than
410 * the initial Sequence Numbers received and sent, respectively:
411 * SWL := max(GSR + 1 - floor(W/4), ISR),
412 * AWL := max(GSS - W' + 1, ISS).
413 * These adjustments MUST be applied only at the beginning of the
414 * connection.
415 */
416 dccp_update_gss(sk, dp->dccps_iss);
417 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
418
419 /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
420 * initialized to S.ISS (sec. 8.5) */
421 dp->dccps_gar = dp->dccps_iss;
422
423 icsk->icsk_retransmits = 0;
424 }
425
426 int dccp_connect(struct sock *sk)
427 {
428 struct sk_buff *skb;
429 struct inet_connection_sock *icsk = inet_csk(sk);
430
431 dccp_connect_init(sk);
432
433 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
434 if (unlikely(skb == NULL))
435 return -ENOBUFS;
436
437 /* Reserve space for headers. */
438 skb_reserve(skb, sk->sk_prot->max_header);
439
440 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
441
442 dccp_skb_entail(sk, skb);
443 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
444 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
445
446 /* Timer for repeating the REQUEST until an answer. */
447 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
448 icsk->icsk_rto, DCCP_RTO_MAX);
449 return 0;
450 }
451
452 EXPORT_SYMBOL_GPL(dccp_connect);
453
454 void dccp_send_ack(struct sock *sk)
455 {
456 /* If we have been reset, we may not send again. */
457 if (sk->sk_state != DCCP_CLOSED) {
458 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
459 GFP_ATOMIC);
460
461 if (skb == NULL) {
462 inet_csk_schedule_ack(sk);
463 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
464 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
465 TCP_DELACK_MAX,
466 DCCP_RTO_MAX);
467 return;
468 }
469
470 /* Reserve space for headers */
471 skb_reserve(skb, sk->sk_prot->max_header);
472 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
473 dccp_transmit_skb(sk, skb);
474 }
475 }
476
477 EXPORT_SYMBOL_GPL(dccp_send_ack);
478
479 void dccp_send_delayed_ack(struct sock *sk)
480 {
481 struct inet_connection_sock *icsk = inet_csk(sk);
482 /*
483 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
484 * with using 2s, and active senders also piggyback the ACK into a
485 * DATAACK packet, so this is really for quiescent senders.
486 */
487 unsigned long timeout = jiffies + 2 * HZ;
488
489 /* Use new timeout only if there wasn't a older one earlier. */
490 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
491 /* If delack timer was blocked or is about to expire,
492 * send ACK now.
493 *
494 * FIXME: check the "about to expire" part
495 */
496 if (icsk->icsk_ack.blocked) {
497 dccp_send_ack(sk);
498 return;
499 }
500
501 if (!time_before(timeout, icsk->icsk_ack.timeout))
502 timeout = icsk->icsk_ack.timeout;
503 }
504 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
505 icsk->icsk_ack.timeout = timeout;
506 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
507 }
508
509 void dccp_send_sync(struct sock *sk, const u64 seq,
510 const enum dccp_pkt_type pkt_type)
511 {
512 /*
513 * We are not putting this on the write queue, so
514 * dccp_transmit_skb() will set the ownership to this
515 * sock.
516 */
517 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
518
519 if (skb == NULL)
520 /* FIXME: how to make sure the sync is sent? */
521 return;
522
523 /* Reserve space for headers and prepare control bits. */
524 skb_reserve(skb, sk->sk_prot->max_header);
525 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
526 DCCP_SKB_CB(skb)->dccpd_seq = seq;
527
528 dccp_transmit_skb(sk, skb);
529 }
530
531 EXPORT_SYMBOL_GPL(dccp_send_sync);
532
533 /*
534 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
535 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
536 * any circumstances.
537 */
538 void dccp_send_close(struct sock *sk, const int active)
539 {
540 struct dccp_sock *dp = dccp_sk(sk);
541 struct sk_buff *skb;
542 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
543
544 skb = alloc_skb(sk->sk_prot->max_header, prio);
545 if (skb == NULL)
546 return;
547
548 /* Reserve space for headers and prepare control bits. */
549 skb_reserve(skb, sk->sk_prot->max_header);
550 DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
551 DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
552
553 if (active) {
554 dccp_write_xmit(sk, 1);
555 dccp_skb_entail(sk, skb);
556 dccp_transmit_skb(sk, skb_clone(skb, prio));
557 /* FIXME do we need a retransmit timer here? */
558 } else
559 dccp_transmit_skb(sk, skb);
560 }
Linux kernel - Deliverables
This page took 0.070955 seconds and 6 git commands to generate.