blob: dede3edb88495150c2bfe9e8c7af7575abf4c498 [file] [log] [blame]
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07001/*
2 * net/dccp/output.c
Arnaldo Carvalho de Melo8109b022006-12-10 16:01:18 -02003 *
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07004 * 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
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070013#include <linux/dccp.h>
Herbert Xu48918a42005-10-30 11:20:59 +110014#include <linux/kernel.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070015#include <linux/skbuff.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070017
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020018#include <net/inet_sock.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070019#include <net/sock.h>
20
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070021#include "ackvec.h"
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070022#include "ccid.h"
23#include "dccp.h"
24
25static inline void dccp_event_ack_sent(struct sock *sk)
26{
27 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
28}
29
Gerrit Renker8695e802011-07-03 09:51:29 -060030/* enqueue @skb on sk_send_head for retransmission, return clone to send now */
31static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
Herbert Xu48918a42005-10-30 11:20:59 +110032{
33 skb_set_owner_w(skb, sk);
34 WARN_ON(sk->sk_send_head);
35 sk->sk_send_head = skb;
Gerrit Renker8695e802011-07-03 09:51:29 -060036 return skb_clone(sk->sk_send_head, gfp_any());
Herbert Xu48918a42005-10-30 11:20:59 +110037}
38
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070039/*
40 * All SKB's seen here are completely headerless. It is our
41 * job to build the DCCP header, and pass the packet down to
42 * IP so it can do the same plus pass the packet off to the
43 * device.
44 */
Herbert Xu48918a42005-10-30 11:20:59 +110045static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070046{
47 if (likely(skb != NULL)) {
David S. Millerd9d8da82011-05-06 22:23:20 -070048 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -080049 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070050 struct dccp_sock *dp = dccp_sk(sk);
51 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
52 struct dccp_hdr *dh;
53 /* XXX For now we're using only 48 bits sequence numbers */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -080054 const u32 dccp_header_size = sizeof(*dh) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070055 sizeof(struct dccp_hdr_ext) +
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -030056 dccp_packet_hdr_len(dcb->dccpd_type);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070057 int err, set_ack = 1;
58 u64 ackno = dp->dccps_gsr;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010059 /*
60 * Increment GSS here already in case the option code needs it.
61 * Update GSS for real only if option processing below succeeds.
62 */
63 dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070064
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070065 switch (dcb->dccpd_type) {
66 case DCCP_PKT_DATA:
67 set_ack = 0;
Herbert Xuedc9e812005-10-30 11:20:59 +110068 /* fall through */
69 case DCCP_PKT_DATAACK:
Gerrit Renkeree1a1592007-09-26 11:30:02 -030070 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070071 break;
Herbert Xuedc9e812005-10-30 11:20:59 +110072
Andrea Bittauafe00252006-03-20 17:43:56 -080073 case DCCP_PKT_REQUEST:
74 set_ack = 0;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010075 /* Use ISS on the first (non-retransmitted) Request. */
76 if (icsk->icsk_retransmits == 0)
77 dcb->dccpd_seq = dp->dccps_iss;
Andrea Bittauafe00252006-03-20 17:43:56 -080078 /* fall through */
79
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070080 case DCCP_PKT_SYNC:
81 case DCCP_PKT_SYNCACK:
Gerrit Renkerb0d045c2007-09-25 22:42:27 -070082 ackno = dcb->dccpd_ack_seq;
Herbert Xuedc9e812005-10-30 11:20:59 +110083 /* fall through */
84 default:
85 /*
Gerrit Renkeree1a1592007-09-26 11:30:02 -030086 * Set owner/destructor: some skbs are allocated via
87 * alloc_skb (e.g. when retransmission may happen).
88 * Only Data, DataAck, and Reset packets should come
89 * through here with skb->sk set.
Herbert Xuedc9e812005-10-30 11:20:59 +110090 */
91 WARN_ON(skb->sk);
92 skb_set_owner_w(skb, sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070093 break;
94 }
Arnaldo Carvalho de Melo24117722005-08-21 05:40:16 -030095
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -080096 if (dccp_insert_options(sk, skb)) {
97 kfree_skb(skb);
98 return -EPROTO;
99 }
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900100
Herbert Xufda0fd62005-10-14 16:38:49 +1000101
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700102 /* Build DCCP header and checksum it. */
Gerrit Renker9b420782006-11-10 11:22:32 -0200103 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700104 dh->dccph_type = dcb->dccpd_type;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000105 dh->dccph_sport = inet->inet_sport;
106 dh->dccph_dport = inet->inet_dport;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700107 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
108 dh->dccph_ccval = dcb->dccpd_ccval;
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200109 dh->dccph_cscov = dp->dccps_pcslen;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700110 /* XXX For now we're using only 48 bits sequence numbers */
111 dh->dccph_x = 1;
112
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100113 dccp_update_gss(sk, dcb->dccpd_seq);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700114 dccp_hdr_set_seq(dh, dp->dccps_gss);
115 if (set_ack)
116 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
117
118 switch (dcb->dccpd_type) {
119 case DCCP_PKT_REQUEST:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300120 dccp_hdr_request(skb)->dccph_req_service =
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700121 dp->dccps_service;
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100122 /*
123 * Limit Ack window to ISS <= P.ackno <= GSS, so that
124 * only Responses to Requests we sent are considered.
125 */
126 dp->dccps_awl = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700127 break;
128 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300129 dccp_hdr_reset(skb)->dccph_reset_code =
130 dcb->dccpd_reset_code;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700131 break;
132 }
133
Herbert Xubb296242010-04-11 02:15:55 +0000134 icsk->icsk_af_ops->send_check(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700135
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700136 if (set_ack)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700137 dccp_event_ack_sent(sk);
138
139 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
140
David S. Millerd9d8da82011-05-06 22:23:20 -0700141 err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200142 return net_xmit_eval(err);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700143 }
144 return -ENOBUFS;
145}
146
Gerrit Renker61799832007-12-13 23:37:55 -0200147/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300148 * dccp_determine_ccmps - Find out about CCID-specific packet-size limits
Gerrit Renker61799832007-12-13 23:37:55 -0200149 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
150 * since the RX CCID is restricted to feedback packets (Acks), which are small
151 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
152 */
153static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
154{
155 const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
156
157 if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
158 return 0;
159 return tx_ccid->ccid_ops->ccid_ccmps;
160}
161
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700162unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
163{
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800164 struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700165 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker61799832007-12-13 23:37:55 -0200166 u32 ccmps = dccp_determine_ccmps(dp);
Gerrit Renker361a5c12009-02-27 22:38:28 +0000167 u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700168
Gerrit Renker61799832007-12-13 23:37:55 -0200169 /* Account for header lengths and IPv4/v6 option overhead */
170 cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
171 sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700172
173 /*
Gerrit Renker361a5c12009-02-27 22:38:28 +0000174 * Leave enough headroom for common DCCP header options.
175 * This only considers options which may appear on DCCP-Data packets, as
176 * per table 3 in RFC 4340, 5.8. When running out of space for other
177 * options (eg. Ack Vector which can take up to 255 bytes), it is better
178 * to schedule a separate Ack. Thus we leave headroom for the following:
179 * - 1 byte for Slow Receiver (11.6)
180 * - 6 bytes for Timestamp (13.1)
181 * - 10 bytes for Timestamp Echo (13.3)
182 * - 8 bytes for NDP count (7.7, when activated)
183 * - 6 bytes for Data Checksum (9.3)
184 * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700185 */
Gerrit Renker361a5c12009-02-27 22:38:28 +0000186 cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
187 (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700188
189 /* And store cached results */
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800190 icsk->icsk_pmtu_cookie = pmtu;
Gerrit Renker61799832007-12-13 23:37:55 -0200191 dp->dccps_mss_cache = cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700192
Gerrit Renker61799832007-12-13 23:37:55 -0200193 return cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700194}
195
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800196EXPORT_SYMBOL_GPL(dccp_sync_mss);
197
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300198void dccp_write_space(struct sock *sk)
199{
Eric Dumazet43815482010-04-29 11:01:49 +0000200 struct socket_wq *wq;
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300201
Eric Dumazet43815482010-04-29 11:01:49 +0000202 rcu_read_lock();
203 wq = rcu_dereference(sk->sk_wq);
204 if (wq_has_sleeper(wq))
205 wake_up_interruptible(&wq->wait);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300206 /* Should agree with poll, otherwise some programs break */
207 if (sock_writeable(sk))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800208 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300209
Eric Dumazet43815482010-04-29 11:01:49 +0000210 rcu_read_unlock();
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300211}
212
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300213/**
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000214 * dccp_wait_for_ccid - Await CCID send permission
Gerrit Renkerbc849872007-10-04 14:38:49 -0700215 * @sk: socket to wait for
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000216 * @delay: timeout in jiffies
217 * This is used by CCIDs which need to delay the send time in process context.
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300218 */
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000219static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300220{
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300221 DEFINE_WAIT(wait);
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000222 long remaining;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300223
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000224 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
225 sk->sk_write_pending++;
226 release_sock(sk);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700227
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000228 remaining = schedule_timeout(delay);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300229
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000230 lock_sock(sk);
231 sk->sk_write_pending--;
Eric Dumazetaa395142010-04-20 13:03:51 +0000232 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300233
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000234 if (signal_pending(current) || sk->sk_err)
235 return -1;
236 return remaining;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300237}
238
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000239/**
240 * dccp_xmit_packet - Send data packet under control of CCID
241 * Transmits next-queued payload and informs CCID to account for the packet.
242 */
243static void dccp_xmit_packet(struct sock *sk)
244{
245 int err, len;
246 struct dccp_sock *dp = dccp_sk(sk);
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100247 struct sk_buff *skb = dccp_qpolicy_pop(sk);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000248
249 if (unlikely(skb == NULL))
250 return;
251 len = skb->len;
252
253 if (sk->sk_state == DCCP_PARTOPEN) {
254 const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
255 /*
256 * See 8.1.5 - Handshake Completion.
257 *
258 * For robustness we resend Confirm options until the client has
259 * entered OPEN. During the initial feature negotiation, the MPS
260 * is smaller than usual, reduced by the Change/Confirm options.
261 */
262 if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
263 DCCP_WARN("Payload too large (%d) for featneg.\n", len);
264 dccp_send_ack(sk);
265 dccp_feat_list_purge(&dp->dccps_featneg);
266 }
267
268 inet_csk_schedule_ack(sk);
269 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
270 inet_csk(sk)->icsk_rto,
271 DCCP_RTO_MAX);
272 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
273 } else if (dccp_ack_pending(sk)) {
274 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
275 } else {
276 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
277 }
278
279 err = dccp_transmit_skb(sk, skb);
280 if (err)
281 dccp_pr_debug("transmit_skb() returned err=%d\n", err);
282 /*
283 * Register this one as sent even if an error occurred. To the remote
284 * end a local packet drop is indistinguishable from network loss, i.e.
285 * any local drop will eventually be reported via receiver feedback.
286 */
287 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
Gerrit Renkerd83447f2010-11-14 17:25:46 +0100288
289 /*
290 * If the CCID needs to transfer additional header options out-of-band
291 * (e.g. Ack Vectors or feature-negotiation options), it activates this
292 * flag to schedule a Sync. The Sync will automatically incorporate all
293 * currently pending header options, thus clearing the backlog.
294 */
295 if (dp->dccps_sync_scheduled)
296 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000297}
298
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000299/**
300 * dccp_flush_write_queue - Drain queue at end of connection
301 * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
302 * happen that the TX queue is not empty at the end of a connection. We give the
303 * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
304 * returns with a non-empty write queue, it will be purged later.
305 */
306void dccp_flush_write_queue(struct sock *sk, long *time_budget)
307{
308 struct dccp_sock *dp = dccp_sk(sk);
309 struct sk_buff *skb;
310 long delay, rc;
311
312 while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
313 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
314
315 switch (ccid_packet_dequeue_eval(rc)) {
316 case CCID_PACKET_WILL_DEQUEUE_LATER:
317 /*
318 * If the CCID determines when to send, the next sending
319 * time is unknown or the CCID may not even send again
320 * (e.g. remote host crashes or lost Ack packets).
321 */
322 DCCP_WARN("CCID did not manage to send all packets\n");
323 return;
324 case CCID_PACKET_DELAY:
325 delay = msecs_to_jiffies(rc);
326 if (delay > *time_budget)
327 return;
328 rc = dccp_wait_for_ccid(sk, delay);
329 if (rc < 0)
330 return;
331 *time_budget -= (delay - rc);
332 /* check again if we can send now */
333 break;
334 case CCID_PACKET_SEND_AT_ONCE:
335 dccp_xmit_packet(sk);
336 break;
337 case CCID_PACKET_ERR:
338 skb_dequeue(&sk->sk_write_queue);
339 kfree_skb(skb);
340 dccp_pr_debug("packet discarded due to err=%ld\n", rc);
341 }
342 }
343}
344
345void dccp_write_xmit(struct sock *sk)
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700346{
Ian McDonald97e58482006-08-26 19:16:45 -0700347 struct dccp_sock *dp = dccp_sk(sk);
348 struct sk_buff *skb;
Ian McDonald97e58482006-08-26 19:16:45 -0700349
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100350 while ((skb = dccp_qpolicy_top(sk))) {
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000351 int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300352
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000353 switch (ccid_packet_dequeue_eval(rc)) {
354 case CCID_PACKET_WILL_DEQUEUE_LATER:
355 return;
356 case CCID_PACKET_DELAY:
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000357 sk_reset_timer(sk, &dp->dccps_xmit_timer,
358 jiffies + msecs_to_jiffies(rc));
359 return;
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000360 case CCID_PACKET_SEND_AT_ONCE:
361 dccp_xmit_packet(sk);
362 break;
363 case CCID_PACKET_ERR:
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100364 dccp_qpolicy_drop(sk, skb);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000365 dccp_pr_debug("packet discarded due to err=%d\n", rc);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200366 }
Ian McDonald97e58482006-08-26 19:16:45 -0700367 }
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700368}
369
Gerrit Renker59435442008-07-26 11:59:09 +0100370/**
371 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
372 * There are only four retransmittable packet types in DCCP:
373 * - Request in client-REQUEST state (sec. 8.1.1),
374 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
375 * - Close in node-CLOSING state (sec. 8.3),
376 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
377 * This function expects sk->sk_send_head to contain the original skb.
378 */
379int dccp_retransmit_skb(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700380{
Gerrit Renker59435442008-07-26 11:59:09 +0100381 WARN_ON(sk->sk_send_head == NULL);
382
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -0800383 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700384 return -EHOSTUNREACH; /* Routing failure or similar. */
385
Gerrit Renker59435442008-07-26 11:59:09 +0100386 /* this count is used to distinguish original and retransmitted skb */
387 inet_csk(sk)->icsk_retransmits++;
388
389 return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700390}
391
392struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
393 struct request_sock *req)
394{
395 struct dccp_hdr *dh;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700396 struct dccp_request_sock *dreq;
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800397 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700398 sizeof(struct dccp_hdr_ext) +
399 sizeof(struct dccp_hdr_response);
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800400 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700401 GFP_ATOMIC);
402 if (skb == NULL)
403 return NULL;
404
405 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800406 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700407
Eric Dumazetadf30902009-06-02 05:19:30 +0000408 skb_dst_set(skb, dst_clone(dst));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700409
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700410 dreq = dccp_rsk(req);
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200411 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
412 dccp_inc_seqno(&dreq->dreq_iss);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700413 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700414 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -0800415
Gerrit Renker0c116832008-11-16 22:49:52 -0800416 /* Resolve feature dependencies resulting from choice of CCID */
417 if (dccp_feat_server_ccid_dependencies(dreq))
418 goto response_failed;
419
420 if (dccp_insert_options_rsk(dreq, skb))
421 goto response_failed;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700422
Gerrit Renker09dbc382006-11-14 12:57:34 -0200423 /* Build and checksum header */
Gerrit Renker9b420782006-11-10 11:22:32 -0200424 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700425
Gerrit Renker944f7502008-10-19 23:36:47 -0700426 dh->dccph_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700427 dh->dccph_dport = inet_rsk(req)->rmt_port;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300428 dh->dccph_doff = (dccp_header_size +
429 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700430 dh->dccph_type = DCCP_PKT_RESPONSE;
431 dh->dccph_x = 1;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700432 dccp_hdr_set_seq(dh, dreq->dreq_iss);
433 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
434 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700435
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200436 dccp_csum_outgoing(skb);
437
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200438 /* We use `acked' to remember that a Response was already sent. */
439 inet_rsk(req)->acked = 1;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700440 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
441 return skb;
Gerrit Renker0c116832008-11-16 22:49:52 -0800442response_failed:
443 kfree_skb(skb);
444 return NULL;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700445}
446
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800447EXPORT_SYMBOL_GPL(dccp_make_response);
448
Gerrit Renkere356d372007-09-26 14:35:19 -0300449/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
Denis V. Lunev7630f022008-04-03 14:20:52 -0700450struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
Gerrit Renkere356d372007-09-26 14:35:19 -0300451{
452 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
453 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
454 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
455 sizeof(struct dccp_hdr_ext) +
456 sizeof(struct dccp_hdr_reset);
457 struct dccp_hdr_reset *dhr;
458 struct sk_buff *skb;
459
Denis V. Lunev7630f022008-04-03 14:20:52 -0700460 skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Gerrit Renkere356d372007-09-26 14:35:19 -0300461 if (skb == NULL)
462 return NULL;
463
Denis V. Lunev7630f022008-04-03 14:20:52 -0700464 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkere356d372007-09-26 14:35:19 -0300465
466 /* Swap the send and the receive. */
467 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
468 dh->dccph_type = DCCP_PKT_RESET;
469 dh->dccph_sport = rxdh->dccph_dport;
470 dh->dccph_dport = rxdh->dccph_sport;
471 dh->dccph_doff = dccp_hdr_reset_len / 4;
472 dh->dccph_x = 1;
473
474 dhr = dccp_hdr_reset(skb);
475 dhr->dccph_reset_code = dcb->dccpd_reset_code;
476
477 switch (dcb->dccpd_reset_code) {
478 case DCCP_RESET_CODE_PACKET_ERROR:
479 dhr->dccph_reset_data[0] = rxdh->dccph_type;
480 break;
481 case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
482 case DCCP_RESET_CODE_MANDATORY_ERROR:
483 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
484 break;
485 }
486 /*
487 * From RFC 4340, 8.3.1:
488 * If P.ackno exists, set R.seqno := P.ackno + 1.
489 * Else set R.seqno := 0.
490 */
491 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
492 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
493 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
494
495 dccp_csum_outgoing(skb);
496 return skb;
497}
498
499EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
500
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300501/* send Reset on established socket, to close or abort the connection */
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800502int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
503{
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300504 struct sk_buff *skb;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800505 /*
506 * FIXME: what if rebuild_header fails?
507 * Should we be doing a rebuild_header here?
508 */
Gerrit Renkerf53dc672007-11-28 08:35:08 +0000509 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800510
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300511 if (err != 0)
512 return err;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800513
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300514 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
515 if (skb == NULL)
516 return -ENOBUFS;
517
518 /* Reserve space for headers and prepare control bits. */
519 skb_reserve(skb, sk->sk_prot->max_header);
520 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
521 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
522
523 return dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800524}
525
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700526/*
527 * Do all connect socket setups that can be done AF independent.
528 */
Gerrit Renker93344af2010-10-11 20:36:33 +0200529int dccp_connect(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700530{
Gerrit Renker93344af2010-10-11 20:36:33 +0200531 struct sk_buff *skb;
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800532 struct dccp_sock *dp = dccp_sk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700533 struct dst_entry *dst = __sk_dst_get(sk);
534 struct inet_connection_sock *icsk = inet_csk(sk);
535
536 sk->sk_err = 0;
537 sock_reset_flag(sk, SOCK_DONE);
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900538
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700539 dccp_sync_mss(sk, dst_mtu(dst));
540
Gerrit Renker9eca0a42008-11-12 00:48:44 -0800541 /* do not connect if feature negotiation setup fails */
542 if (dccp_feat_finalise_settings(dccp_sk(sk)))
543 return -EPROTO;
544
Gerrit Renker93344af2010-10-11 20:36:33 +0200545 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
546 dp->dccps_gar = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700547
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800548 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700549 if (unlikely(skb == NULL))
550 return -ENOBUFS;
551
552 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800553 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700554
555 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700556
Gerrit Renker8695e802011-07-03 09:51:29 -0600557 dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700558 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
559
560 /* Timer for repeating the REQUEST until an answer. */
Gerrit Renker93344af2010-10-11 20:36:33 +0200561 icsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700562 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
563 icsk->icsk_rto, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700564 return 0;
565}
566
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800567EXPORT_SYMBOL_GPL(dccp_connect);
568
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700569void dccp_send_ack(struct sock *sk)
570{
571 /* If we have been reset, we may not send again. */
572 if (sk->sk_state != DCCP_CLOSED) {
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800573 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
574 GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700575
576 if (skb == NULL) {
577 inet_csk_schedule_ack(sk);
578 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300579 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
580 TCP_DELACK_MAX,
581 DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700582 return;
583 }
584
585 /* Reserve space for headers */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800586 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700587 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700588 dccp_transmit_skb(sk, skb);
589 }
590}
591
592EXPORT_SYMBOL_GPL(dccp_send_ack);
593
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100594#if 0
Gerrit Renker727ecc52007-09-26 11:26:04 -0300595/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700596void dccp_send_delayed_ack(struct sock *sk)
597{
598 struct inet_connection_sock *icsk = inet_csk(sk);
599 /*
600 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
601 * with using 2s, and active senders also piggyback the ACK into a
602 * DATAACK packet, so this is really for quiescent senders.
603 */
604 unsigned long timeout = jiffies + 2 * HZ;
605
606 /* Use new timeout only if there wasn't a older one earlier. */
607 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
608 /* If delack timer was blocked or is about to expire,
609 * send ACK now.
610 *
611 * FIXME: check the "about to expire" part
612 */
613 if (icsk->icsk_ack.blocked) {
614 dccp_send_ack(sk);
615 return;
616 }
617
618 if (!time_before(timeout, icsk->icsk_ack.timeout))
619 timeout = icsk->icsk_ack.timeout;
620 }
621 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
622 icsk->icsk_ack.timeout = timeout;
623 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
624}
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100625#endif
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700626
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700627void dccp_send_sync(struct sock *sk, const u64 ackno,
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300628 const enum dccp_pkt_type pkt_type)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700629{
630 /*
631 * We are not putting this on the write queue, so
632 * dccp_transmit_skb() will set the ownership to this
633 * sock.
634 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800635 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700636
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700637 if (skb == NULL) {
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700638 /* FIXME: how to make sure the sync is sent? */
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700639 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700640 return;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700641 }
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700642
643 /* Reserve space for headers and prepare control bits. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800644 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300645 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700646 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700647
Gerrit Renkerd83447f2010-11-14 17:25:46 +0100648 /*
649 * Clear the flag in case the Sync was scheduled for out-of-band data,
650 * such as carrying a long Ack Vector.
651 */
652 dccp_sk(sk)->dccps_sync_scheduled = 0;
653
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700654 dccp_transmit_skb(sk, skb);
655}
656
Arnaldo Carvalho de Melob61fafc2006-03-20 21:25:11 -0800657EXPORT_SYMBOL_GPL(dccp_send_sync);
658
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300659/*
660 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
661 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
662 * any circumstances.
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700663 */
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700664void dccp_send_close(struct sock *sk, const int active)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700665{
666 struct dccp_sock *dp = dccp_sk(sk);
667 struct sk_buff *skb;
Al Viro7d877f32005-10-21 03:20:43 -0400668 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700669
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700670 skb = alloc_skb(sk->sk_prot->max_header, prio);
671 if (skb == NULL)
672 return;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700673
674 /* Reserve space for headers and prepare control bits. */
675 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkerb8599d22007-12-13 12:25:01 -0200676 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
677 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
678 else
679 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700680
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700681 if (active) {
Gerrit Renker8695e802011-07-03 09:51:29 -0600682 skb = dccp_skb_entail(sk, skb);
Gerrit Renker92d31922007-12-13 12:02:43 -0200683 /*
684 * Retransmission timer for active-close: RFC 4340, 8.3 requires
685 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
686 * state can be left. The initial timeout is 2 RTTs.
687 * Since RTT measurement is done by the CCIDs, there is no easy
688 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
689 * is too low (200ms); we use a high value to avoid unnecessary
690 * retransmissions when the link RTT is > 0.2 seconds.
691 * FIXME: Let main module sample RTTs and use that instead.
692 */
693 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
694 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
Gerrit Renker8695e802011-07-03 09:51:29 -0600695 }
696 dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700697}