blob: 11418a9a389d3564654d3b5a971dafe481f14969 [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
Arnaldo Carvalho de Meloc25a18b2006-03-20 21:58:56 -080030static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
Herbert Xu48918a42005-10-30 11:20:59 +110031{
32 skb_set_owner_w(skb, sk);
33 WARN_ON(sk->sk_send_head);
34 sk->sk_send_head = skb;
35}
36
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070037/*
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 */
Herbert Xu48918a42005-10-30 11:20:59 +110043static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070044{
45 if (likely(skb != NULL)) {
46 const struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -080047 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070048 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 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -080052 const u32 dccp_header_size = sizeof(*dh) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070053 sizeof(struct dccp_hdr_ext) +
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -030054 dccp_packet_hdr_len(dcb->dccpd_type);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070055 int err, set_ack = 1;
56 u64 ackno = dp->dccps_gsr;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010057 /*
58 * Increment GSS here already in case the option code needs it.
59 * Update GSS for real only if option processing below succeeds.
60 */
61 dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070062
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070063 switch (dcb->dccpd_type) {
64 case DCCP_PKT_DATA:
65 set_ack = 0;
Herbert Xuedc9e812005-10-30 11:20:59 +110066 /* fall through */
67 case DCCP_PKT_DATAACK:
Gerrit Renkeree1a1592007-09-26 11:30:02 -030068 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070069 break;
Herbert Xuedc9e812005-10-30 11:20:59 +110070
Andrea Bittauafe00252006-03-20 17:43:56 -080071 case DCCP_PKT_REQUEST:
72 set_ack = 0;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010073 /* Use ISS on the first (non-retransmitted) Request. */
74 if (icsk->icsk_retransmits == 0)
75 dcb->dccpd_seq = dp->dccps_iss;
Andrea Bittauafe00252006-03-20 17:43:56 -080076 /* fall through */
77
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070078 case DCCP_PKT_SYNC:
79 case DCCP_PKT_SYNCACK:
Gerrit Renkerb0d045c2007-09-25 22:42:27 -070080 ackno = dcb->dccpd_ack_seq;
Herbert Xuedc9e812005-10-30 11:20:59 +110081 /* fall through */
82 default:
83 /*
Gerrit Renkeree1a1592007-09-26 11:30:02 -030084 * Set owner/destructor: some skbs are allocated via
85 * alloc_skb (e.g. when retransmission may happen).
86 * Only Data, DataAck, and Reset packets should come
87 * through here with skb->sk set.
Herbert Xuedc9e812005-10-30 11:20:59 +110088 */
89 WARN_ON(skb->sk);
90 skb_set_owner_w(skb, sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070091 break;
92 }
Arnaldo Carvalho de Melo24117722005-08-21 05:40:16 -030093
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -080094 if (dccp_insert_options(sk, skb)) {
95 kfree_skb(skb);
96 return -EPROTO;
97 }
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +090098
Herbert Xufda0fd62005-10-14 16:38:49 +100099
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700100 /* Build DCCP header and checksum it. */
Gerrit Renker9b420782006-11-10 11:22:32 -0200101 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700102 dh->dccph_type = dcb->dccpd_type;
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000103 dh->dccph_sport = inet->inet_sport;
104 dh->dccph_dport = inet->inet_dport;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700105 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
106 dh->dccph_ccval = dcb->dccpd_ccval;
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200107 dh->dccph_cscov = dp->dccps_pcslen;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700108 /* XXX For now we're using only 48 bits sequence numbers */
109 dh->dccph_x = 1;
110
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100111 dccp_update_gss(sk, dcb->dccpd_seq);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700112 dccp_hdr_set_seq(dh, dp->dccps_gss);
113 if (set_ack)
114 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
115
116 switch (dcb->dccpd_type) {
117 case DCCP_PKT_REQUEST:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300118 dccp_hdr_request(skb)->dccph_req_service =
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700119 dp->dccps_service;
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100120 /*
121 * Limit Ack window to ISS <= P.ackno <= GSS, so that
122 * only Responses to Requests we sent are considered.
123 */
124 dp->dccps_awl = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700125 break;
126 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300127 dccp_hdr_reset(skb)->dccph_reset_code =
128 dcb->dccpd_reset_code;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700129 break;
130 }
131
Herbert Xubb296242010-04-11 02:15:55 +0000132 icsk->icsk_af_ops->send_check(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700133
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700134 if (set_ack)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700135 dccp_event_ack_sent(sk);
136
137 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
138
Shan Wei4e15ed42010-04-15 16:43:08 +0000139 err = icsk->icsk_af_ops->queue_xmit(skb);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200140 return net_xmit_eval(err);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700141 }
142 return -ENOBUFS;
143}
144
Gerrit Renker61799832007-12-13 23:37:55 -0200145/**
146 * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
147 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
148 * since the RX CCID is restricted to feedback packets (Acks), which are small
149 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
150 */
151static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
152{
153 const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
154
155 if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
156 return 0;
157 return tx_ccid->ccid_ops->ccid_ccmps;
158}
159
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700160unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
161{
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800162 struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700163 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker61799832007-12-13 23:37:55 -0200164 u32 ccmps = dccp_determine_ccmps(dp);
Gerrit Renker361a5c12009-02-27 22:38:28 +0000165 u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700166
Gerrit Renker61799832007-12-13 23:37:55 -0200167 /* Account for header lengths and IPv4/v6 option overhead */
168 cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
169 sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700170
171 /*
Gerrit Renker361a5c12009-02-27 22:38:28 +0000172 * Leave enough headroom for common DCCP header options.
173 * This only considers options which may appear on DCCP-Data packets, as
174 * per table 3 in RFC 4340, 5.8. When running out of space for other
175 * options (eg. Ack Vector which can take up to 255 bytes), it is better
176 * to schedule a separate Ack. Thus we leave headroom for the following:
177 * - 1 byte for Slow Receiver (11.6)
178 * - 6 bytes for Timestamp (13.1)
179 * - 10 bytes for Timestamp Echo (13.3)
180 * - 8 bytes for NDP count (7.7, when activated)
181 * - 6 bytes for Data Checksum (9.3)
182 * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700183 */
Gerrit Renker361a5c12009-02-27 22:38:28 +0000184 cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
185 (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700186
187 /* And store cached results */
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800188 icsk->icsk_pmtu_cookie = pmtu;
Gerrit Renker61799832007-12-13 23:37:55 -0200189 dp->dccps_mss_cache = cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700190
Gerrit Renker61799832007-12-13 23:37:55 -0200191 return cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700192}
193
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800194EXPORT_SYMBOL_GPL(dccp_sync_mss);
195
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300196void dccp_write_space(struct sock *sk)
197{
Eric Dumazet43815482010-04-29 11:01:49 +0000198 struct socket_wq *wq;
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300199
Eric Dumazet43815482010-04-29 11:01:49 +0000200 rcu_read_lock();
201 wq = rcu_dereference(sk->sk_wq);
202 if (wq_has_sleeper(wq))
203 wake_up_interruptible(&wq->wait);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300204 /* Should agree with poll, otherwise some programs break */
205 if (sock_writeable(sk))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800206 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300207
Eric Dumazet43815482010-04-29 11:01:49 +0000208 rcu_read_unlock();
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300209}
210
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300211/**
212 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
Gerrit Renkerbc849872007-10-04 14:38:49 -0700213 * @sk: socket to wait for
214 * @skb: current skb to pass on for waiting
215 * @delay: sleep timeout in milliseconds (> 0)
216 * This function is called by default when the socket is closed, and
217 * when a non-zero linger time is set on the socket. For consistency
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300218 */
Gerrit Renkerbc849872007-10-04 14:38:49 -0700219static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb, int delay)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300220{
221 struct dccp_sock *dp = dccp_sk(sk);
222 DEFINE_WAIT(wait);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700223 unsigned long jiffdelay;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300224 int rc;
225
Gerrit Renkerbc849872007-10-04 14:38:49 -0700226 do {
227 dccp_pr_debug("delayed send by %d msec\n", delay);
228 jiffdelay = msecs_to_jiffies(delay);
229
Eric Dumazetaa395142010-04-20 13:03:51 +0000230 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300231
Gerrit Renkerbc849872007-10-04 14:38:49 -0700232 sk->sk_write_pending++;
233 release_sock(sk);
234 schedule_timeout(jiffdelay);
235 lock_sock(sk);
236 sk->sk_write_pending--;
237
Ian McDonald97e58482006-08-26 19:16:45 -0700238 if (sk->sk_err)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300239 goto do_error;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300240 if (signal_pending(current))
241 goto do_interrupted;
242
Gerrit Renker6b57c932006-11-28 19:55:06 -0200243 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700244 } while ((delay = rc) > 0);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300245out:
Eric Dumazetaa395142010-04-20 13:03:51 +0000246 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300247 return rc;
248
249do_error:
250 rc = -EPIPE;
251 goto out;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300252do_interrupted:
Ian McDonald5cc3741d2006-12-09 23:56:09 -0200253 rc = -EINTR;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300254 goto out;
255}
256
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000257/**
258 * dccp_xmit_packet - Send data packet under control of CCID
259 * Transmits next-queued payload and informs CCID to account for the packet.
260 */
261static void dccp_xmit_packet(struct sock *sk)
262{
263 int err, len;
264 struct dccp_sock *dp = dccp_sk(sk);
265 struct sk_buff *skb = skb_dequeue(&sk->sk_write_queue);
266
267 if (unlikely(skb == NULL))
268 return;
269 len = skb->len;
270
271 if (sk->sk_state == DCCP_PARTOPEN) {
272 const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
273 /*
274 * See 8.1.5 - Handshake Completion.
275 *
276 * For robustness we resend Confirm options until the client has
277 * entered OPEN. During the initial feature negotiation, the MPS
278 * is smaller than usual, reduced by the Change/Confirm options.
279 */
280 if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
281 DCCP_WARN("Payload too large (%d) for featneg.\n", len);
282 dccp_send_ack(sk);
283 dccp_feat_list_purge(&dp->dccps_featneg);
284 }
285
286 inet_csk_schedule_ack(sk);
287 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
288 inet_csk(sk)->icsk_rto,
289 DCCP_RTO_MAX);
290 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
291 } else if (dccp_ack_pending(sk)) {
292 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
293 } else {
294 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
295 }
296
297 err = dccp_transmit_skb(sk, skb);
298 if (err)
299 dccp_pr_debug("transmit_skb() returned err=%d\n", err);
300 /*
301 * Register this one as sent even if an error occurred. To the remote
302 * end a local packet drop is indistinguishable from network loss, i.e.
303 * any local drop will eventually be reported via receiver feedback.
304 */
305 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
306}
307
Ian McDonald97e58482006-08-26 19:16:45 -0700308void dccp_write_xmit(struct sock *sk, int block)
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700309{
Ian McDonald97e58482006-08-26 19:16:45 -0700310 struct dccp_sock *dp = dccp_sk(sk);
311 struct sk_buff *skb;
Ian McDonald97e58482006-08-26 19:16:45 -0700312
313 while ((skb = skb_peek(&sk->sk_write_queue))) {
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000314 int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300315
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000316 switch (ccid_packet_dequeue_eval(rc)) {
317 case CCID_PACKET_WILL_DEQUEUE_LATER:
318 return;
319 case CCID_PACKET_DELAY:
Ian McDonald97e58482006-08-26 19:16:45 -0700320 if (!block) {
321 sk_reset_timer(sk, &dp->dccps_xmit_timer,
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000322 msecs_to_jiffies(rc)+jiffies);
323 return;
324 }
325 rc = dccp_wait_for_ccid(sk, skb, rc);
326 if (rc && rc != -EINTR) {
327 DCCP_BUG("err=%d after dccp_wait_for_ccid", rc);
328 skb_dequeue(&sk->sk_write_queue);
329 kfree_skb(skb);
Ian McDonald97e58482006-08-26 19:16:45 -0700330 break;
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000331 }
332 /* fall through */
333 case CCID_PACKET_SEND_AT_ONCE:
334 dccp_xmit_packet(sk);
335 break;
336 case CCID_PACKET_ERR:
337 skb_dequeue(&sk->sk_write_queue);
Patrick McHardyb08d5842007-02-27 09:57:37 -0800338 kfree_skb(skb);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000339 dccp_pr_debug("packet discarded due to err=%d\n", rc);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200340 }
Ian McDonald97e58482006-08-26 19:16:45 -0700341 }
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700342}
343
Gerrit Renker59435442008-07-26 11:59:09 +0100344/**
345 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
346 * There are only four retransmittable packet types in DCCP:
347 * - Request in client-REQUEST state (sec. 8.1.1),
348 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
349 * - Close in node-CLOSING state (sec. 8.3),
350 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
351 * This function expects sk->sk_send_head to contain the original skb.
352 */
353int dccp_retransmit_skb(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700354{
Gerrit Renker59435442008-07-26 11:59:09 +0100355 WARN_ON(sk->sk_send_head == NULL);
356
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -0800357 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700358 return -EHOSTUNREACH; /* Routing failure or similar. */
359
Gerrit Renker59435442008-07-26 11:59:09 +0100360 /* this count is used to distinguish original and retransmitted skb */
361 inet_csk(sk)->icsk_retransmits++;
362
363 return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700364}
365
366struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
367 struct request_sock *req)
368{
369 struct dccp_hdr *dh;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700370 struct dccp_request_sock *dreq;
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800371 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700372 sizeof(struct dccp_hdr_ext) +
373 sizeof(struct dccp_hdr_response);
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800374 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700375 GFP_ATOMIC);
376 if (skb == NULL)
377 return NULL;
378
379 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800380 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700381
Eric Dumazetadf30902009-06-02 05:19:30 +0000382 skb_dst_set(skb, dst_clone(dst));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700383
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700384 dreq = dccp_rsk(req);
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200385 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
386 dccp_inc_seqno(&dreq->dreq_iss);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700387 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700388 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -0800389
Gerrit Renker0c116832008-11-16 22:49:52 -0800390 /* Resolve feature dependencies resulting from choice of CCID */
391 if (dccp_feat_server_ccid_dependencies(dreq))
392 goto response_failed;
393
394 if (dccp_insert_options_rsk(dreq, skb))
395 goto response_failed;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700396
Gerrit Renker09dbc382006-11-14 12:57:34 -0200397 /* Build and checksum header */
Gerrit Renker9b420782006-11-10 11:22:32 -0200398 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700399
Gerrit Renker944f7502008-10-19 23:36:47 -0700400 dh->dccph_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700401 dh->dccph_dport = inet_rsk(req)->rmt_port;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300402 dh->dccph_doff = (dccp_header_size +
403 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700404 dh->dccph_type = DCCP_PKT_RESPONSE;
405 dh->dccph_x = 1;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700406 dccp_hdr_set_seq(dh, dreq->dreq_iss);
407 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
408 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700409
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200410 dccp_csum_outgoing(skb);
411
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200412 /* We use `acked' to remember that a Response was already sent. */
413 inet_rsk(req)->acked = 1;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700414 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
415 return skb;
Gerrit Renker0c116832008-11-16 22:49:52 -0800416response_failed:
417 kfree_skb(skb);
418 return NULL;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700419}
420
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800421EXPORT_SYMBOL_GPL(dccp_make_response);
422
Gerrit Renkere356d372007-09-26 14:35:19 -0300423/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
Denis V. Lunev7630f022008-04-03 14:20:52 -0700424struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
Gerrit Renkere356d372007-09-26 14:35:19 -0300425{
426 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
427 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
428 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
429 sizeof(struct dccp_hdr_ext) +
430 sizeof(struct dccp_hdr_reset);
431 struct dccp_hdr_reset *dhr;
432 struct sk_buff *skb;
433
Denis V. Lunev7630f022008-04-03 14:20:52 -0700434 skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Gerrit Renkere356d372007-09-26 14:35:19 -0300435 if (skb == NULL)
436 return NULL;
437
Denis V. Lunev7630f022008-04-03 14:20:52 -0700438 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkere356d372007-09-26 14:35:19 -0300439
440 /* Swap the send and the receive. */
441 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
442 dh->dccph_type = DCCP_PKT_RESET;
443 dh->dccph_sport = rxdh->dccph_dport;
444 dh->dccph_dport = rxdh->dccph_sport;
445 dh->dccph_doff = dccp_hdr_reset_len / 4;
446 dh->dccph_x = 1;
447
448 dhr = dccp_hdr_reset(skb);
449 dhr->dccph_reset_code = dcb->dccpd_reset_code;
450
451 switch (dcb->dccpd_reset_code) {
452 case DCCP_RESET_CODE_PACKET_ERROR:
453 dhr->dccph_reset_data[0] = rxdh->dccph_type;
454 break;
455 case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
456 case DCCP_RESET_CODE_MANDATORY_ERROR:
457 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
458 break;
459 }
460 /*
461 * From RFC 4340, 8.3.1:
462 * If P.ackno exists, set R.seqno := P.ackno + 1.
463 * Else set R.seqno := 0.
464 */
465 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
466 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
467 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
468
469 dccp_csum_outgoing(skb);
470 return skb;
471}
472
473EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
474
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300475/* send Reset on established socket, to close or abort the connection */
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800476int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
477{
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300478 struct sk_buff *skb;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800479 /*
480 * FIXME: what if rebuild_header fails?
481 * Should we be doing a rebuild_header here?
482 */
Gerrit Renkerf53dc672007-11-28 08:35:08 +0000483 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800484
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300485 if (err != 0)
486 return err;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800487
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300488 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
489 if (skb == NULL)
490 return -ENOBUFS;
491
492 /* Reserve space for headers and prepare control bits. */
493 skb_reserve(skb, sk->sk_prot->max_header);
494 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
495 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
496
497 return dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800498}
499
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700500/*
501 * Do all connect socket setups that can be done AF independent.
502 */
Gerrit Renker93344af2010-10-11 20:36:33 +0200503int dccp_connect(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700504{
Gerrit Renker93344af2010-10-11 20:36:33 +0200505 struct sk_buff *skb;
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800506 struct dccp_sock *dp = dccp_sk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700507 struct dst_entry *dst = __sk_dst_get(sk);
508 struct inet_connection_sock *icsk = inet_csk(sk);
509
510 sk->sk_err = 0;
511 sock_reset_flag(sk, SOCK_DONE);
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900512
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700513 dccp_sync_mss(sk, dst_mtu(dst));
514
Gerrit Renker9eca0a42008-11-12 00:48:44 -0800515 /* do not connect if feature negotiation setup fails */
516 if (dccp_feat_finalise_settings(dccp_sk(sk)))
517 return -EPROTO;
518
Gerrit Renker93344af2010-10-11 20:36:33 +0200519 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
520 dp->dccps_gar = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700521
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800522 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700523 if (unlikely(skb == NULL))
524 return -ENOBUFS;
525
526 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800527 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700528
529 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700530
Herbert Xu48918a42005-10-30 11:20:59 +1100531 dccp_skb_entail(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700532 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
533 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
534
535 /* Timer for repeating the REQUEST until an answer. */
Gerrit Renker93344af2010-10-11 20:36:33 +0200536 icsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700537 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
538 icsk->icsk_rto, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700539 return 0;
540}
541
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800542EXPORT_SYMBOL_GPL(dccp_connect);
543
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700544void dccp_send_ack(struct sock *sk)
545{
546 /* If we have been reset, we may not send again. */
547 if (sk->sk_state != DCCP_CLOSED) {
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800548 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
549 GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700550
551 if (skb == NULL) {
552 inet_csk_schedule_ack(sk);
553 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300554 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
555 TCP_DELACK_MAX,
556 DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700557 return;
558 }
559
560 /* Reserve space for headers */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800561 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700562 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700563 dccp_transmit_skb(sk, skb);
564 }
565}
566
567EXPORT_SYMBOL_GPL(dccp_send_ack);
568
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100569#if 0
Gerrit Renker727ecc52007-09-26 11:26:04 -0300570/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700571void dccp_send_delayed_ack(struct sock *sk)
572{
573 struct inet_connection_sock *icsk = inet_csk(sk);
574 /*
575 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
576 * with using 2s, and active senders also piggyback the ACK into a
577 * DATAACK packet, so this is really for quiescent senders.
578 */
579 unsigned long timeout = jiffies + 2 * HZ;
580
581 /* Use new timeout only if there wasn't a older one earlier. */
582 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
583 /* If delack timer was blocked or is about to expire,
584 * send ACK now.
585 *
586 * FIXME: check the "about to expire" part
587 */
588 if (icsk->icsk_ack.blocked) {
589 dccp_send_ack(sk);
590 return;
591 }
592
593 if (!time_before(timeout, icsk->icsk_ack.timeout))
594 timeout = icsk->icsk_ack.timeout;
595 }
596 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
597 icsk->icsk_ack.timeout = timeout;
598 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
599}
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100600#endif
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700601
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700602void dccp_send_sync(struct sock *sk, const u64 ackno,
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300603 const enum dccp_pkt_type pkt_type)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700604{
605 /*
606 * We are not putting this on the write queue, so
607 * dccp_transmit_skb() will set the ownership to this
608 * sock.
609 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800610 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700611
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700612 if (skb == NULL) {
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700613 /* FIXME: how to make sure the sync is sent? */
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700614 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700615 return;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700616 }
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700617
618 /* Reserve space for headers and prepare control bits. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800619 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300620 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700621 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700622
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700623 dccp_transmit_skb(sk, skb);
624}
625
Arnaldo Carvalho de Melob61fafc2006-03-20 21:25:11 -0800626EXPORT_SYMBOL_GPL(dccp_send_sync);
627
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300628/*
629 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
630 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
631 * any circumstances.
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700632 */
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700633void dccp_send_close(struct sock *sk, const int active)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700634{
635 struct dccp_sock *dp = dccp_sk(sk);
636 struct sk_buff *skb;
Al Viro7d877f32005-10-21 03:20:43 -0400637 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700638
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700639 skb = alloc_skb(sk->sk_prot->max_header, prio);
640 if (skb == NULL)
641 return;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700642
643 /* Reserve space for headers and prepare control bits. */
644 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkerb8599d22007-12-13 12:25:01 -0200645 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
646 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
647 else
648 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700649
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700650 if (active) {
Ian McDonald97e58482006-08-26 19:16:45 -0700651 dccp_write_xmit(sk, 1);
Herbert Xu48918a42005-10-30 11:20:59 +1100652 dccp_skb_entail(sk, skb);
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700653 dccp_transmit_skb(sk, skb_clone(skb, prio));
Gerrit Renker92d31922007-12-13 12:02:43 -0200654 /*
655 * Retransmission timer for active-close: RFC 4340, 8.3 requires
656 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
657 * state can be left. The initial timeout is 2 RTTs.
658 * Since RTT measurement is done by the CCIDs, there is no easy
659 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
660 * is too low (200ms); we use a high value to avoid unnecessary
661 * retransmissions when the link RTT is > 0.2 seconds.
662 * FIXME: Let main module sample RTTs and use that instead.
663 */
664 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
665 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700666 } else
667 dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700668}