blob: d6bb753bf6ad8485552bea39107a59afa0eb971a [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>
16
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020017#include <net/inet_sock.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070018#include <net/sock.h>
19
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070020#include "ackvec.h"
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070021#include "ccid.h"
22#include "dccp.h"
23
24static inline void dccp_event_ack_sent(struct sock *sk)
25{
26 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
27}
28
Arnaldo Carvalho de Meloc25a18b2006-03-20 21:58:56 -080029static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
Herbert Xu48918a42005-10-30 11:20:59 +110030{
31 skb_set_owner_w(skb, sk);
32 WARN_ON(sk->sk_send_head);
33 sk->sk_send_head = skb;
34}
35
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070036/*
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 */
Herbert Xu48918a42005-10-30 11:20:59 +110042static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070043{
44 if (likely(skb != NULL)) {
45 const struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -080046 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070047 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 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -080051 const u32 dccp_header_size = sizeof(*dh) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070052 sizeof(struct dccp_hdr_ext) +
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -030053 dccp_packet_hdr_len(dcb->dccpd_type);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070054 int err, set_ack = 1;
55 u64 ackno = dp->dccps_gsr;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010056 /*
57 * Increment GSS here already in case the option code needs it.
58 * Update GSS for real only if option processing below succeeds.
59 */
60 dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070061
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070062 switch (dcb->dccpd_type) {
63 case DCCP_PKT_DATA:
64 set_ack = 0;
Herbert Xuedc9e812005-10-30 11:20:59 +110065 /* fall through */
66 case DCCP_PKT_DATAACK:
Gerrit Renkeree1a1592007-09-26 11:30:02 -030067 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070068 break;
Herbert Xuedc9e812005-10-30 11:20:59 +110069
Andrea Bittauafe00252006-03-20 17:43:56 -080070 case DCCP_PKT_REQUEST:
71 set_ack = 0;
Gerrit Renker73f18fd2008-07-26 11:59:10 +010072 /* Use ISS on the first (non-retransmitted) Request. */
73 if (icsk->icsk_retransmits == 0)
74 dcb->dccpd_seq = dp->dccps_iss;
Andrea Bittauafe00252006-03-20 17:43:56 -080075 /* fall through */
76
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070077 case DCCP_PKT_SYNC:
78 case DCCP_PKT_SYNCACK:
Gerrit Renkerb0d045c2007-09-25 22:42:27 -070079 ackno = dcb->dccpd_ack_seq;
Herbert Xuedc9e812005-10-30 11:20:59 +110080 /* fall through */
81 default:
82 /*
Gerrit Renkeree1a1592007-09-26 11:30:02 -030083 * Set owner/destructor: some skbs are allocated via
84 * alloc_skb (e.g. when retransmission may happen).
85 * Only Data, DataAck, and Reset packets should come
86 * through here with skb->sk set.
Herbert Xuedc9e812005-10-30 11:20:59 +110087 */
88 WARN_ON(skb->sk);
89 skb_set_owner_w(skb, sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070090 break;
91 }
Arnaldo Carvalho de Melo24117722005-08-21 05:40:16 -030092
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -080093 if (dccp_insert_options(sk, skb)) {
94 kfree_skb(skb);
95 return -EPROTO;
96 }
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +090097
Herbert Xufda0fd62005-10-14 16:38:49 +100098
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070099 /* Build DCCP header and checksum it. */
Gerrit Renker9b420782006-11-10 11:22:32 -0200100 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700101 dh->dccph_type = dcb->dccpd_type;
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000102 dh->dccph_sport = inet->inet_sport;
103 dh->dccph_dport = inet->inet_dport;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700104 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
105 dh->dccph_ccval = dcb->dccpd_ccval;
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200106 dh->dccph_cscov = dp->dccps_pcslen;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700107 /* XXX For now we're using only 48 bits sequence numbers */
108 dh->dccph_x = 1;
109
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100110 dccp_update_gss(sk, dcb->dccpd_seq);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700111 dccp_hdr_set_seq(dh, dp->dccps_gss);
112 if (set_ack)
113 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
114
115 switch (dcb->dccpd_type) {
116 case DCCP_PKT_REQUEST:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300117 dccp_hdr_request(skb)->dccph_req_service =
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700118 dp->dccps_service;
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100119 /*
120 * Limit Ack window to ISS <= P.ackno <= GSS, so that
121 * only Responses to Requests we sent are considered.
122 */
123 dp->dccps_awl = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700124 break;
125 case DCCP_PKT_RESET:
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300126 dccp_hdr_reset(skb)->dccph_reset_code =
127 dcb->dccpd_reset_code;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700128 break;
129 }
130
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200131 icsk->icsk_af_ops->send_check(sk, 0, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700132
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700133 if (set_ack)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700134 dccp_event_ack_sent(sk);
135
136 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
137
David S. Millere89862f2007-01-26 01:04:55 -0800138 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200139 return net_xmit_eval(err);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700140 }
141 return -ENOBUFS;
142}
143
Gerrit Renker61799832007-12-13 23:37:55 -0200144/**
145 * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
146 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
147 * since the RX CCID is restricted to feedback packets (Acks), which are small
148 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
149 */
150static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
151{
152 const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
153
154 if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
155 return 0;
156 return tx_ccid->ccid_ops->ccid_ccmps;
157}
158
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700159unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
160{
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800161 struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700162 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker61799832007-12-13 23:37:55 -0200163 u32 ccmps = dccp_determine_ccmps(dp);
Gerrit Renker361a5c12009-02-27 22:38:28 +0000164 u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700165
Gerrit Renker61799832007-12-13 23:37:55 -0200166 /* Account for header lengths and IPv4/v6 option overhead */
167 cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
168 sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700169
170 /*
Gerrit Renker361a5c12009-02-27 22:38:28 +0000171 * Leave enough headroom for common DCCP header options.
172 * This only considers options which may appear on DCCP-Data packets, as
173 * per table 3 in RFC 4340, 5.8. When running out of space for other
174 * options (eg. Ack Vector which can take up to 255 bytes), it is better
175 * to schedule a separate Ack. Thus we leave headroom for the following:
176 * - 1 byte for Slow Receiver (11.6)
177 * - 6 bytes for Timestamp (13.1)
178 * - 10 bytes for Timestamp Echo (13.3)
179 * - 8 bytes for NDP count (7.7, when activated)
180 * - 6 bytes for Data Checksum (9.3)
181 * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700182 */
Gerrit Renker361a5c12009-02-27 22:38:28 +0000183 cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
184 (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700185
186 /* And store cached results */
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800187 icsk->icsk_pmtu_cookie = pmtu;
Gerrit Renker61799832007-12-13 23:37:55 -0200188 dp->dccps_mss_cache = cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700189
Gerrit Renker61799832007-12-13 23:37:55 -0200190 return cur_mps;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700191}
192
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800193EXPORT_SYMBOL_GPL(dccp_sync_mss);
194
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300195void dccp_write_space(struct sock *sk)
196{
197 read_lock(&sk->sk_callback_lock);
198
Jiri Olsaa57de0b2009-07-08 12:09:13 +0000199 if (sk_has_sleeper(sk))
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300200 wake_up_interruptible(sk->sk_sleep);
201 /* Should agree with poll, otherwise some programs break */
202 if (sock_writeable(sk))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800203 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300204
205 read_unlock(&sk->sk_callback_lock);
206}
207
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300208/**
209 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
Gerrit Renkerbc849872007-10-04 14:38:49 -0700210 * @sk: socket to wait for
211 * @skb: current skb to pass on for waiting
212 * @delay: sleep timeout in milliseconds (> 0)
213 * This function is called by default when the socket is closed, and
214 * when a non-zero linger time is set on the socket. For consistency
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300215 */
Gerrit Renkerbc849872007-10-04 14:38:49 -0700216static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb, int delay)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300217{
218 struct dccp_sock *dp = dccp_sk(sk);
219 DEFINE_WAIT(wait);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700220 unsigned long jiffdelay;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300221 int rc;
222
Gerrit Renkerbc849872007-10-04 14:38:49 -0700223 do {
224 dccp_pr_debug("delayed send by %d msec\n", delay);
225 jiffdelay = msecs_to_jiffies(delay);
226
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300227 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
228
Gerrit Renkerbc849872007-10-04 14:38:49 -0700229 sk->sk_write_pending++;
230 release_sock(sk);
231 schedule_timeout(jiffdelay);
232 lock_sock(sk);
233 sk->sk_write_pending--;
234
Ian McDonald97e58482006-08-26 19:16:45 -0700235 if (sk->sk_err)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300236 goto do_error;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300237 if (signal_pending(current))
238 goto do_interrupted;
239
Gerrit Renker6b57c932006-11-28 19:55:06 -0200240 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700241 } while ((delay = rc) > 0);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300242out:
243 finish_wait(sk->sk_sleep, &wait);
244 return rc;
245
246do_error:
247 rc = -EPIPE;
248 goto out;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300249do_interrupted:
Ian McDonald5cc3741d2006-12-09 23:56:09 -0200250 rc = -EINTR;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300251 goto out;
252}
253
Ian McDonald97e58482006-08-26 19:16:45 -0700254void dccp_write_xmit(struct sock *sk, int block)
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700255{
Ian McDonald97e58482006-08-26 19:16:45 -0700256 struct dccp_sock *dp = dccp_sk(sk);
257 struct sk_buff *skb;
Ian McDonald97e58482006-08-26 19:16:45 -0700258
259 while ((skb = skb_peek(&sk->sk_write_queue))) {
Gerrit Renker6b57c932006-11-28 19:55:06 -0200260 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300261
Ian McDonald97e58482006-08-26 19:16:45 -0700262 if (err > 0) {
263 if (!block) {
264 sk_reset_timer(sk, &dp->dccps_xmit_timer,
265 msecs_to_jiffies(err)+jiffies);
266 break;
Ian McDonald5cc3741d2006-12-09 23:56:09 -0200267 } else
Gerrit Renkerbc849872007-10-04 14:38:49 -0700268 err = dccp_wait_for_ccid(sk, skb, err);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200269 if (err && err != -EINTR)
Gerrit Renker59348b12006-11-20 18:39:23 -0200270 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
Ian McDonald97e58482006-08-26 19:16:45 -0700271 }
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700272
Ian McDonald97e58482006-08-26 19:16:45 -0700273 skb_dequeue(&sk->sk_write_queue);
274 if (err == 0) {
275 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
276 const int len = skb->len;
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700277
Ian McDonald97e58482006-08-26 19:16:45 -0700278 if (sk->sk_state == DCCP_PARTOPEN) {
Gerrit Renker86739fb2009-02-27 22:38:29 +0000279 const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
280 /*
281 * See 8.1.5 - Handshake Completion.
282 *
283 * For robustness we resend Confirm options until the client has
284 * entered OPEN. During the initial feature negotiation, the MPS
285 * is smaller than usual, reduced by the Change/Confirm options.
286 */
287 if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
288 DCCP_WARN("Payload too large (%d) for featneg.\n", len);
289 dccp_send_ack(sk);
290 dccp_feat_list_purge(&dp->dccps_featneg);
291 }
292
Ian McDonald97e58482006-08-26 19:16:45 -0700293 inet_csk_schedule_ack(sk);
294 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700295 inet_csk(sk)->icsk_rto,
296 DCCP_RTO_MAX);
Ian McDonald97e58482006-08-26 19:16:45 -0700297 dcb->dccpd_type = DCCP_PKT_DATAACK;
298 } else if (dccp_ack_pending(sk))
299 dcb->dccpd_type = DCCP_PKT_DATAACK;
300 else
301 dcb->dccpd_type = DCCP_PKT_DATA;
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700302
Ian McDonald97e58482006-08-26 19:16:45 -0700303 err = dccp_transmit_skb(sk, skb);
304 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
Gerrit Renker59348b12006-11-20 18:39:23 -0200305 if (err)
306 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
307 err);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200308 } else {
Gerrit Renker6626e362007-03-20 15:00:28 -0300309 dccp_pr_debug("packet discarded due to err=%d\n", err);
Patrick McHardyb08d5842007-02-27 09:57:37 -0800310 kfree_skb(skb);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200311 }
Ian McDonald97e58482006-08-26 19:16:45 -0700312 }
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700313}
314
Gerrit Renker59435442008-07-26 11:59:09 +0100315/**
316 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
317 * There are only four retransmittable packet types in DCCP:
318 * - Request in client-REQUEST state (sec. 8.1.1),
319 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
320 * - Close in node-CLOSING state (sec. 8.3),
321 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
322 * This function expects sk->sk_send_head to contain the original skb.
323 */
324int dccp_retransmit_skb(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700325{
Gerrit Renker59435442008-07-26 11:59:09 +0100326 WARN_ON(sk->sk_send_head == NULL);
327
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -0800328 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700329 return -EHOSTUNREACH; /* Routing failure or similar. */
330
Gerrit Renker59435442008-07-26 11:59:09 +0100331 /* this count is used to distinguish original and retransmitted skb */
332 inet_csk(sk)->icsk_retransmits++;
333
334 return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700335}
336
337struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
338 struct request_sock *req)
339{
340 struct dccp_hdr *dh;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700341 struct dccp_request_sock *dreq;
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800342 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700343 sizeof(struct dccp_hdr_ext) +
344 sizeof(struct dccp_hdr_response);
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800345 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700346 GFP_ATOMIC);
347 if (skb == NULL)
348 return NULL;
349
350 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800351 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700352
Eric Dumazetadf30902009-06-02 05:19:30 +0000353 skb_dst_set(skb, dst_clone(dst));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700354
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700355 dreq = dccp_rsk(req);
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200356 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
357 dccp_inc_seqno(&dreq->dreq_iss);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700358 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700359 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -0800360
Gerrit Renker0c116832008-11-16 22:49:52 -0800361 /* Resolve feature dependencies resulting from choice of CCID */
362 if (dccp_feat_server_ccid_dependencies(dreq))
363 goto response_failed;
364
365 if (dccp_insert_options_rsk(dreq, skb))
366 goto response_failed;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700367
Gerrit Renker09dbc382006-11-14 12:57:34 -0200368 /* Build and checksum header */
Gerrit Renker9b420782006-11-10 11:22:32 -0200369 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700370
Gerrit Renker944f7502008-10-19 23:36:47 -0700371 dh->dccph_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700372 dh->dccph_dport = inet_rsk(req)->rmt_port;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300373 dh->dccph_doff = (dccp_header_size +
374 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700375 dh->dccph_type = DCCP_PKT_RESPONSE;
376 dh->dccph_x = 1;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700377 dccp_hdr_set_seq(dh, dreq->dreq_iss);
378 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
379 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700380
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200381 dccp_csum_outgoing(skb);
382
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200383 /* We use `acked' to remember that a Response was already sent. */
384 inet_rsk(req)->acked = 1;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700385 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
386 return skb;
Gerrit Renker0c116832008-11-16 22:49:52 -0800387response_failed:
388 kfree_skb(skb);
389 return NULL;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700390}
391
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800392EXPORT_SYMBOL_GPL(dccp_make_response);
393
Gerrit Renkere356d372007-09-26 14:35:19 -0300394/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
Denis V. Lunev7630f022008-04-03 14:20:52 -0700395struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
Gerrit Renkere356d372007-09-26 14:35:19 -0300396{
397 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
398 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
399 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
400 sizeof(struct dccp_hdr_ext) +
401 sizeof(struct dccp_hdr_reset);
402 struct dccp_hdr_reset *dhr;
403 struct sk_buff *skb;
404
Denis V. Lunev7630f022008-04-03 14:20:52 -0700405 skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Gerrit Renkere356d372007-09-26 14:35:19 -0300406 if (skb == NULL)
407 return NULL;
408
Denis V. Lunev7630f022008-04-03 14:20:52 -0700409 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkere356d372007-09-26 14:35:19 -0300410
411 /* Swap the send and the receive. */
412 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
413 dh->dccph_type = DCCP_PKT_RESET;
414 dh->dccph_sport = rxdh->dccph_dport;
415 dh->dccph_dport = rxdh->dccph_sport;
416 dh->dccph_doff = dccp_hdr_reset_len / 4;
417 dh->dccph_x = 1;
418
419 dhr = dccp_hdr_reset(skb);
420 dhr->dccph_reset_code = dcb->dccpd_reset_code;
421
422 switch (dcb->dccpd_reset_code) {
423 case DCCP_RESET_CODE_PACKET_ERROR:
424 dhr->dccph_reset_data[0] = rxdh->dccph_type;
425 break;
426 case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
427 case DCCP_RESET_CODE_MANDATORY_ERROR:
428 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
429 break;
430 }
431 /*
432 * From RFC 4340, 8.3.1:
433 * If P.ackno exists, set R.seqno := P.ackno + 1.
434 * Else set R.seqno := 0.
435 */
436 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
437 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
438 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
439
440 dccp_csum_outgoing(skb);
441 return skb;
442}
443
444EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
445
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300446/* send Reset on established socket, to close or abort the connection */
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800447int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
448{
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300449 struct sk_buff *skb;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800450 /*
451 * FIXME: what if rebuild_header fails?
452 * Should we be doing a rebuild_header here?
453 */
Gerrit Renkerf53dc672007-11-28 08:35:08 +0000454 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800455
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300456 if (err != 0)
457 return err;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800458
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300459 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
460 if (skb == NULL)
461 return -ENOBUFS;
462
463 /* Reserve space for headers and prepare control bits. */
464 skb_reserve(skb, sk->sk_prot->max_header);
465 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
466 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
467
468 return dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800469}
470
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700471/*
472 * Do all connect socket setups that can be done AF independent.
473 */
474static inline void dccp_connect_init(struct sock *sk)
475{
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800476 struct dccp_sock *dp = dccp_sk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700477 struct dst_entry *dst = __sk_dst_get(sk);
478 struct inet_connection_sock *icsk = inet_csk(sk);
479
480 sk->sk_err = 0;
481 sock_reset_flag(sk, SOCK_DONE);
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900482
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700483 dccp_sync_mss(sk, dst_mtu(dst));
484
Gerrit Renker73f18fd2008-07-26 11:59:10 +0100485 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
Gerrit Renkerd7f73652006-11-13 13:34:38 -0200486 dp->dccps_gar = dp->dccps_iss;
487
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700488 icsk->icsk_retransmits = 0;
489}
490
491int dccp_connect(struct sock *sk)
492{
493 struct sk_buff *skb;
494 struct inet_connection_sock *icsk = inet_csk(sk);
495
Gerrit Renker9eca0a42008-11-12 00:48:44 -0800496 /* do not connect if feature negotiation setup fails */
497 if (dccp_feat_finalise_settings(dccp_sk(sk)))
498 return -EPROTO;
499
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700500 dccp_connect_init(sk);
501
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800502 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700503 if (unlikely(skb == NULL))
504 return -ENOBUFS;
505
506 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800507 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700508
509 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700510
Herbert Xu48918a42005-10-30 11:20:59 +1100511 dccp_skb_entail(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700512 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
513 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
514
515 /* Timer for repeating the REQUEST until an answer. */
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700516 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
517 icsk->icsk_rto, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700518 return 0;
519}
520
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800521EXPORT_SYMBOL_GPL(dccp_connect);
522
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700523void dccp_send_ack(struct sock *sk)
524{
525 /* If we have been reset, we may not send again. */
526 if (sk->sk_state != DCCP_CLOSED) {
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800527 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
528 GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700529
530 if (skb == NULL) {
531 inet_csk_schedule_ack(sk);
532 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300533 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
534 TCP_DELACK_MAX,
535 DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700536 return;
537 }
538
539 /* Reserve space for headers */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800540 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700541 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700542 dccp_transmit_skb(sk, skb);
543 }
544}
545
546EXPORT_SYMBOL_GPL(dccp_send_ack);
547
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100548#if 0
Gerrit Renker727ecc52007-09-26 11:26:04 -0300549/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700550void dccp_send_delayed_ack(struct sock *sk)
551{
552 struct inet_connection_sock *icsk = inet_csk(sk);
553 /*
554 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
555 * with using 2s, and active senders also piggyback the ACK into a
556 * DATAACK packet, so this is really for quiescent senders.
557 */
558 unsigned long timeout = jiffies + 2 * HZ;
559
560 /* Use new timeout only if there wasn't a older one earlier. */
561 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
562 /* If delack timer was blocked or is about to expire,
563 * send ACK now.
564 *
565 * FIXME: check the "about to expire" part
566 */
567 if (icsk->icsk_ack.blocked) {
568 dccp_send_ack(sk);
569 return;
570 }
571
572 if (!time_before(timeout, icsk->icsk_ack.timeout))
573 timeout = icsk->icsk_ack.timeout;
574 }
575 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
576 icsk->icsk_ack.timeout = timeout;
577 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
578}
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100579#endif
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700580
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700581void dccp_send_sync(struct sock *sk, const u64 ackno,
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300582 const enum dccp_pkt_type pkt_type)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700583{
584 /*
585 * We are not putting this on the write queue, so
586 * dccp_transmit_skb() will set the ownership to this
587 * sock.
588 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800589 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700590
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700591 if (skb == NULL) {
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700592 /* FIXME: how to make sure the sync is sent? */
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700593 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700594 return;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700595 }
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700596
597 /* Reserve space for headers and prepare control bits. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800598 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300599 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700600 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700601
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700602 dccp_transmit_skb(sk, skb);
603}
604
Arnaldo Carvalho de Melob61fafc2006-03-20 21:25:11 -0800605EXPORT_SYMBOL_GPL(dccp_send_sync);
606
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300607/*
608 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
609 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
610 * any circumstances.
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700611 */
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700612void dccp_send_close(struct sock *sk, const int active)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700613{
614 struct dccp_sock *dp = dccp_sk(sk);
615 struct sk_buff *skb;
Al Viro7d877f32005-10-21 03:20:43 -0400616 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700617
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700618 skb = alloc_skb(sk->sk_prot->max_header, prio);
619 if (skb == NULL)
620 return;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700621
622 /* Reserve space for headers and prepare control bits. */
623 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkerb8599d22007-12-13 12:25:01 -0200624 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
625 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
626 else
627 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700628
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700629 if (active) {
Ian McDonald97e58482006-08-26 19:16:45 -0700630 dccp_write_xmit(sk, 1);
Herbert Xu48918a42005-10-30 11:20:59 +1100631 dccp_skb_entail(sk, skb);
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700632 dccp_transmit_skb(sk, skb_clone(skb, prio));
Gerrit Renker92d31922007-12-13 12:02:43 -0200633 /*
634 * Retransmission timer for active-close: RFC 4340, 8.3 requires
635 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
636 * state can be left. The initial timeout is 2 RTTs.
637 * Since RTT measurement is done by the CCIDs, there is no easy
638 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
639 * is too low (200ms); we use a high value to avoid unnecessary
640 * retransmissions when the link RTT is > 0.2 seconds.
641 * FIXME: Let main module sample RTTs and use that instead.
642 */
643 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
644 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700645 } else
646 dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700647}