blob: 6433187a5cc442bbf8e2eb4ea694343eae11ceb8 [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07002/*
3 * net/dccp/output.c
Arnaldo Carvalho de Melo8109b022006-12-10 16:01:18 -02004 *
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07005 * An implementation of the DCCP protocol
6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07007 */
8
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -07009#include <linux/dccp.h>
Herbert Xu48918a42005-10-30 11:20:59 +110010#include <linux/kernel.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070011#include <linux/skbuff.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/slab.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010013#include <linux/sched/signal.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070014
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020015#include <net/inet_sock.h>
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070016#include <net/sock.h>
17
Arnaldo Carvalho de Meloae31c332005-09-18 00:17:51 -070018#include "ackvec.h"
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -070019#include "ccid.h"
20#include "dccp.h"
21
22static inline void dccp_event_ack_sent(struct sock *sk)
23{
24 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
25}
26
Gerrit Renker8695e802011-07-03 09:51:29 -060027/* enqueue @skb on sk_send_head for retransmission, return clone to send now */
28static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
Herbert Xu48918a42005-10-30 11:20:59 +110029{
30 skb_set_owner_w(skb, sk);
31 WARN_ON(sk->sk_send_head);
32 sk->sk_send_head = skb;
Gerrit Renker8695e802011-07-03 09:51:29 -060033 return skb_clone(sk->sk_send_head, gfp_any());
Herbert Xu48918a42005-10-30 11:20:59 +110034}
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)) {
David S. Millerd9d8da82011-05-06 22:23:20 -070045 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
Herbert Xubb296242010-04-11 02:15:55 +0000131 icsk->icsk_af_ops->send_check(sk, 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
Eric Dumazetb0270e92014-04-15 12:58:34 -0400138 err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
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/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300145 * dccp_determine_ccmps - Find out about CCID-specific packet-size limits
Gerrit Renker61799832007-12-13 23:37:55 -0200146 * 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{
Eric Dumazet43815482010-04-29 11:01:49 +0000197 struct socket_wq *wq;
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300198
Eric Dumazet43815482010-04-29 11:01:49 +0000199 rcu_read_lock();
200 wq = rcu_dereference(sk->sk_wq);
Herbert Xu1ce0bf52015-11-26 13:55:39 +0800201 if (skwq_has_sleeper(wq))
Eric Dumazet43815482010-04-29 11:01:49 +0000202 wake_up_interruptible(&wq->wait);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300203 /* Should agree with poll, otherwise some programs break */
204 if (sock_writeable(sk))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800205 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300206
Eric Dumazet43815482010-04-29 11:01:49 +0000207 rcu_read_unlock();
Arnaldo Carvalho de Meloc530cfb2005-08-29 02:15:54 -0300208}
209
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300210/**
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000211 * dccp_wait_for_ccid - Await CCID send permission
Gerrit Renkerbc849872007-10-04 14:38:49 -0700212 * @sk: socket to wait for
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000213 * @delay: timeout in jiffies
Ben Hutchings2c530402012-07-10 10:55:09 +0000214 *
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000215 * This is used by CCIDs which need to delay the send time in process context.
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300216 */
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000217static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300218{
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300219 DEFINE_WAIT(wait);
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000220 long remaining;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300221
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000222 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
223 sk->sk_write_pending++;
224 release_sock(sk);
Gerrit Renkerbc849872007-10-04 14:38:49 -0700225
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000226 remaining = schedule_timeout(delay);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300227
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000228 lock_sock(sk);
229 sk->sk_write_pending--;
Eric Dumazetaa395142010-04-20 13:03:51 +0000230 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300231
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000232 if (signal_pending(current) || sk->sk_err)
233 return -1;
234 return remaining;
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300235}
236
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000237/**
238 * dccp_xmit_packet - Send data packet under control of CCID
239 * Transmits next-queued payload and informs CCID to account for the packet.
240 */
241static void dccp_xmit_packet(struct sock *sk)
242{
243 int err, len;
244 struct dccp_sock *dp = dccp_sk(sk);
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100245 struct sk_buff *skb = dccp_qpolicy_pop(sk);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000246
247 if (unlikely(skb == NULL))
248 return;
249 len = skb->len;
250
251 if (sk->sk_state == DCCP_PARTOPEN) {
252 const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
253 /*
254 * See 8.1.5 - Handshake Completion.
255 *
256 * For robustness we resend Confirm options until the client has
257 * entered OPEN. During the initial feature negotiation, the MPS
258 * is smaller than usual, reduced by the Change/Confirm options.
259 */
260 if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
261 DCCP_WARN("Payload too large (%d) for featneg.\n", len);
262 dccp_send_ack(sk);
263 dccp_feat_list_purge(&dp->dccps_featneg);
264 }
265
266 inet_csk_schedule_ack(sk);
267 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
268 inet_csk(sk)->icsk_rto,
269 DCCP_RTO_MAX);
270 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
271 } else if (dccp_ack_pending(sk)) {
272 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
273 } else {
274 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
275 }
276
277 err = dccp_transmit_skb(sk, skb);
278 if (err)
279 dccp_pr_debug("transmit_skb() returned err=%d\n", err);
280 /*
281 * Register this one as sent even if an error occurred. To the remote
282 * end a local packet drop is indistinguishable from network loss, i.e.
283 * any local drop will eventually be reported via receiver feedback.
284 */
285 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
Gerrit Renkerd83447f2010-11-14 17:25:46 +0100286
287 /*
288 * If the CCID needs to transfer additional header options out-of-band
289 * (e.g. Ack Vectors or feature-negotiation options), it activates this
290 * flag to schedule a Sync. The Sync will automatically incorporate all
291 * currently pending header options, thus clearing the backlog.
292 */
293 if (dp->dccps_sync_scheduled)
294 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000295}
296
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000297/**
298 * dccp_flush_write_queue - Drain queue at end of connection
299 * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
300 * happen that the TX queue is not empty at the end of a connection. We give the
301 * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
302 * returns with a non-empty write queue, it will be purged later.
303 */
304void dccp_flush_write_queue(struct sock *sk, long *time_budget)
305{
306 struct dccp_sock *dp = dccp_sk(sk);
307 struct sk_buff *skb;
308 long delay, rc;
309
310 while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
311 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
312
313 switch (ccid_packet_dequeue_eval(rc)) {
314 case CCID_PACKET_WILL_DEQUEUE_LATER:
315 /*
316 * If the CCID determines when to send, the next sending
317 * time is unknown or the CCID may not even send again
318 * (e.g. remote host crashes or lost Ack packets).
319 */
320 DCCP_WARN("CCID did not manage to send all packets\n");
321 return;
322 case CCID_PACKET_DELAY:
323 delay = msecs_to_jiffies(rc);
324 if (delay > *time_budget)
325 return;
326 rc = dccp_wait_for_ccid(sk, delay);
327 if (rc < 0)
328 return;
329 *time_budget -= (delay - rc);
330 /* check again if we can send now */
331 break;
332 case CCID_PACKET_SEND_AT_ONCE:
333 dccp_xmit_packet(sk);
334 break;
335 case CCID_PACKET_ERR:
336 skb_dequeue(&sk->sk_write_queue);
337 kfree_skb(skb);
338 dccp_pr_debug("packet discarded due to err=%ld\n", rc);
339 }
340 }
341}
342
343void dccp_write_xmit(struct sock *sk)
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700344{
Ian McDonald97e58482006-08-26 19:16:45 -0700345 struct dccp_sock *dp = dccp_sk(sk);
346 struct sk_buff *skb;
Ian McDonald97e58482006-08-26 19:16:45 -0700347
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100348 while ((skb = dccp_qpolicy_top(sk))) {
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000349 int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
Arnaldo Carvalho de Melod6809c12005-08-27 03:06:35 -0300350
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000351 switch (ccid_packet_dequeue_eval(rc)) {
352 case CCID_PACKET_WILL_DEQUEUE_LATER:
353 return;
354 case CCID_PACKET_DELAY:
Gerrit Renkerb1fcf552010-10-27 19:16:27 +0000355 sk_reset_timer(sk, &dp->dccps_xmit_timer,
356 jiffies + msecs_to_jiffies(rc));
357 return;
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000358 case CCID_PACKET_SEND_AT_ONCE:
359 dccp_xmit_packet(sk);
360 break;
361 case CCID_PACKET_ERR:
Tomasz Grobelny871a2c12010-12-04 13:38:01 +0100362 dccp_qpolicy_drop(sk, skb);
Gerrit Renkerdc841e32010-10-27 19:16:26 +0000363 dccp_pr_debug("packet discarded due to err=%d\n", rc);
Gerrit Renkerf6282f42006-12-10 00:05:12 -0200364 }
Ian McDonald97e58482006-08-26 19:16:45 -0700365 }
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700366}
367
Gerrit Renker59435442008-07-26 11:59:09 +0100368/**
369 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
370 * There are only four retransmittable packet types in DCCP:
371 * - Request in client-REQUEST state (sec. 8.1.1),
372 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
373 * - Close in node-CLOSING state (sec. 8.3),
374 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
375 * This function expects sk->sk_send_head to contain the original skb.
376 */
377int dccp_retransmit_skb(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700378{
Gerrit Renker59435442008-07-26 11:59:09 +0100379 WARN_ON(sk->sk_send_head == NULL);
380
Arnaldo Carvalho de Melo57cca052005-12-13 23:16:16 -0800381 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700382 return -EHOSTUNREACH; /* Routing failure or similar. */
383
Gerrit Renker59435442008-07-26 11:59:09 +0100384 /* this count is used to distinguish original and retransmitted skb */
385 inet_csk(sk)->icsk_retransmits++;
386
387 return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700388}
389
Eric Dumazet802885f2015-09-25 07:39:22 -0700390struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700391 struct request_sock *req)
392{
393 struct dccp_hdr *dh;
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700394 struct dccp_request_sock *dreq;
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800395 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700396 sizeof(struct dccp_hdr_ext) +
397 sizeof(struct dccp_hdr_response);
Eric Dumazet802885f2015-09-25 07:39:22 -0700398 struct sk_buff *skb;
399
400 /* sk is marked const to clearly express we dont hold socket lock.
401 * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
402 * it is safe to promote sk to non const.
403 */
404 skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
405 GFP_ATOMIC);
406 if (!skb)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700407 return NULL;
408
Eric Dumazet802885f2015-09-25 07:39:22 -0700409 skb_reserve(skb, MAX_DCCP_HEADER);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700410
Eric Dumazetadf30902009-06-02 05:19:30 +0000411 skb_dst_set(skb, dst_clone(dst));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700412
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700413 dreq = dccp_rsk(req);
Samuel Jerof541fb72012-02-26 18:22:02 -0700414 if (inet_rsk(req)->acked) /* increase GSS upon retransmission */
415 dccp_inc_seqno(&dreq->dreq_gss);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700416 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
Samuel Jerof541fb72012-02-26 18:22:02 -0700417 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_gss;
Arnaldo Carvalho de Melo2d0817d2006-03-20 22:32:06 -0800418
Gerrit Renker0c116832008-11-16 22:49:52 -0800419 /* Resolve feature dependencies resulting from choice of CCID */
420 if (dccp_feat_server_ccid_dependencies(dreq))
421 goto response_failed;
422
423 if (dccp_insert_options_rsk(dreq, skb))
424 goto response_failed;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700425
Gerrit Renker09dbc382006-11-14 12:57:34 -0200426 /* Build and checksum header */
Gerrit Renker9b420782006-11-10 11:22:32 -0200427 dh = dccp_zeroed_hdr(skb, dccp_header_size);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700428
Eric Dumazetb44084c2013-10-10 00:04:37 -0700429 dh->dccph_sport = htons(inet_rsk(req)->ir_num);
Eric Dumazet634fb9792013-10-09 15:21:29 -0700430 dh->dccph_dport = inet_rsk(req)->ir_rmt_port;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300431 dh->dccph_doff = (dccp_header_size +
432 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700433 dh->dccph_type = DCCP_PKT_RESPONSE;
434 dh->dccph_x = 1;
Samuel Jerof541fb72012-02-26 18:22:02 -0700435 dccp_hdr_set_seq(dh, dreq->dreq_gss);
436 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
Arnaldo Carvalho de Melo67e6b622005-09-16 16:58:40 -0700437 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700438
Gerrit Renker6f4e5ff2006-11-10 17:43:06 -0200439 dccp_csum_outgoing(skb);
440
Gerrit Renkere11d9d32006-11-13 13:12:07 -0200441 /* We use `acked' to remember that a Response was already sent. */
442 inet_rsk(req)->acked = 1;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700443 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
444 return skb;
Gerrit Renker0c116832008-11-16 22:49:52 -0800445response_failed:
446 kfree_skb(skb);
447 return NULL;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700448}
449
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800450EXPORT_SYMBOL_GPL(dccp_make_response);
451
Gerrit Renkere356d372007-09-26 14:35:19 -0300452/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
Denis V. Lunev7630f022008-04-03 14:20:52 -0700453struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
Gerrit Renkere356d372007-09-26 14:35:19 -0300454{
455 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
456 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
457 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
458 sizeof(struct dccp_hdr_ext) +
459 sizeof(struct dccp_hdr_reset);
460 struct dccp_hdr_reset *dhr;
461 struct sk_buff *skb;
462
Denis V. Lunev7630f022008-04-03 14:20:52 -0700463 skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Gerrit Renkere356d372007-09-26 14:35:19 -0300464 if (skb == NULL)
465 return NULL;
466
Denis V. Lunev7630f022008-04-03 14:20:52 -0700467 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkere356d372007-09-26 14:35:19 -0300468
469 /* Swap the send and the receive. */
470 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
471 dh->dccph_type = DCCP_PKT_RESET;
472 dh->dccph_sport = rxdh->dccph_dport;
473 dh->dccph_dport = rxdh->dccph_sport;
474 dh->dccph_doff = dccp_hdr_reset_len / 4;
475 dh->dccph_x = 1;
476
477 dhr = dccp_hdr_reset(skb);
478 dhr->dccph_reset_code = dcb->dccpd_reset_code;
479
480 switch (dcb->dccpd_reset_code) {
481 case DCCP_RESET_CODE_PACKET_ERROR:
482 dhr->dccph_reset_data[0] = rxdh->dccph_type;
483 break;
484 case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
485 case DCCP_RESET_CODE_MANDATORY_ERROR:
486 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
487 break;
488 }
489 /*
490 * From RFC 4340, 8.3.1:
491 * If P.ackno exists, set R.seqno := P.ackno + 1.
492 * Else set R.seqno := 0.
493 */
494 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
495 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
496 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
497
498 dccp_csum_outgoing(skb);
499 return skb;
500}
501
502EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
503
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300504/* send Reset on established socket, to close or abort the connection */
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800505int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
506{
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300507 struct sk_buff *skb;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800508 /*
509 * FIXME: what if rebuild_header fails?
510 * Should we be doing a rebuild_header here?
511 */
Gerrit Renkerf53dc672007-11-28 08:35:08 +0000512 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800513
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300514 if (err != 0)
515 return err;
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800516
Gerrit Renkeree1a1592007-09-26 11:30:02 -0300517 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
518 if (skb == NULL)
519 return -ENOBUFS;
520
521 /* Reserve space for headers and prepare control bits. */
522 skb_reserve(skb, sk->sk_prot->max_header);
523 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
524 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
525
526 return dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo017487d2006-03-20 19:25:24 -0800527}
528
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700529/*
530 * Do all connect socket setups that can be done AF independent.
531 */
Gerrit Renker93344af2010-10-11 20:36:33 +0200532int dccp_connect(struct sock *sk)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700533{
Gerrit Renker93344af2010-10-11 20:36:33 +0200534 struct sk_buff *skb;
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800535 struct dccp_sock *dp = dccp_sk(sk);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700536 struct dst_entry *dst = __sk_dst_get(sk);
537 struct inet_connection_sock *icsk = inet_csk(sk);
538
539 sk->sk_err = 0;
540 sock_reset_flag(sk, SOCK_DONE);
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900541
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700542 dccp_sync_mss(sk, dst_mtu(dst));
543
Gerrit Renker9eca0a42008-11-12 00:48:44 -0800544 /* do not connect if feature negotiation setup fails */
545 if (dccp_feat_finalise_settings(dccp_sk(sk)))
546 return -EPROTO;
547
Gerrit Renker93344af2010-10-11 20:36:33 +0200548 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
549 dp->dccps_gar = dp->dccps_iss;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700550
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800551 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700552 if (unlikely(skb == NULL))
553 return -ENOBUFS;
554
555 /* Reserve space for headers. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800556 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700557
558 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700559
Gerrit Renker8695e802011-07-03 09:51:29 -0600560 dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700561 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
562
563 /* Timer for repeating the REQUEST until an answer. */
Gerrit Renker93344af2010-10-11 20:36:33 +0200564 icsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo27258ee2005-08-09 20:30:56 -0700565 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
566 icsk->icsk_rto, DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700567 return 0;
568}
569
Arnaldo Carvalho de Melof21e68c2005-12-13 23:24:16 -0800570EXPORT_SYMBOL_GPL(dccp_connect);
571
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700572void dccp_send_ack(struct sock *sk)
573{
574 /* If we have been reset, we may not send again. */
575 if (sk->sk_state != DCCP_CLOSED) {
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800576 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
577 GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700578
579 if (skb == NULL) {
580 inet_csk_schedule_ack(sk);
581 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300582 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
583 TCP_DELACK_MAX,
584 DCCP_RTO_MAX);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700585 return;
586 }
587
588 /* Reserve space for headers */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800589 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700590 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700591 dccp_transmit_skb(sk, skb);
592 }
593}
594
595EXPORT_SYMBOL_GPL(dccp_send_ack);
596
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100597#if 0
Gerrit Renker727ecc52007-09-26 11:26:04 -0300598/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700599void dccp_send_delayed_ack(struct sock *sk)
600{
601 struct inet_connection_sock *icsk = inet_csk(sk);
602 /*
603 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
604 * with using 2s, and active senders also piggyback the ACK into a
605 * DATAACK packet, so this is really for quiescent senders.
606 */
607 unsigned long timeout = jiffies + 2 * HZ;
608
609 /* Use new timeout only if there wasn't a older one earlier. */
610 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
611 /* If delack timer was blocked or is about to expire,
612 * send ACK now.
613 *
614 * FIXME: check the "about to expire" part
615 */
616 if (icsk->icsk_ack.blocked) {
617 dccp_send_ack(sk);
618 return;
619 }
620
621 if (!time_before(timeout, icsk->icsk_ack.timeout))
622 timeout = icsk->icsk_ack.timeout;
623 }
624 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
625 icsk->icsk_ack.timeout = timeout;
626 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
627}
Gerrit Renker1e2f0e5e2008-06-11 11:19:09 +0100628#endif
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700629
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700630void dccp_send_sync(struct sock *sk, const u64 ackno,
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300631 const enum dccp_pkt_type pkt_type)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700632{
633 /*
634 * We are not putting this on the write queue, so
635 * dccp_transmit_skb() will set the ownership to this
636 * sock.
637 */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800638 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700639
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700640 if (skb == NULL) {
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700641 /* FIXME: how to make sure the sync is sent? */
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700642 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700643 return;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700644 }
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700645
646 /* Reserve space for headers and prepare control bits. */
Arnaldo Carvalho de Melo118b2c92006-03-20 22:31:09 -0800647 skb_reserve(skb, sk->sk_prot->max_header);
Arnaldo Carvalho de Meloe92ae932005-08-17 03:10:59 -0300648 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
Gerrit Renkerb0d045c2007-09-25 22:42:27 -0700649 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700650
Gerrit Renkerd83447f2010-11-14 17:25:46 +0100651 /*
652 * Clear the flag in case the Sync was scheduled for out-of-band data,
653 * such as carrying a long Ack Vector.
654 */
655 dccp_sk(sk)->dccps_sync_scheduled = 0;
656
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700657 dccp_transmit_skb(sk, skb);
658}
659
Arnaldo Carvalho de Melob61fafc2006-03-20 21:25:11 -0800660EXPORT_SYMBOL_GPL(dccp_send_sync);
661
Arnaldo Carvalho de Melo7690af32005-08-13 20:34:54 -0300662/*
663 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
664 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
665 * any circumstances.
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700666 */
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700667void dccp_send_close(struct sock *sk, const int active)
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700668{
669 struct dccp_sock *dp = dccp_sk(sk);
670 struct sk_buff *skb;
Al Viro7d877f32005-10-21 03:20:43 -0400671 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700672
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700673 skb = alloc_skb(sk->sk_prot->max_header, prio);
674 if (skb == NULL)
675 return;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700676
677 /* Reserve space for headers and prepare control bits. */
678 skb_reserve(skb, sk->sk_prot->max_header);
Gerrit Renkerb8599d22007-12-13 12:25:01 -0200679 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
680 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
681 else
682 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700683
Arnaldo Carvalho de Melo7ad07e72005-08-23 21:50:06 -0700684 if (active) {
Gerrit Renker8695e802011-07-03 09:51:29 -0600685 skb = dccp_skb_entail(sk, skb);
Gerrit Renker92d31922007-12-13 12:02:43 -0200686 /*
687 * Retransmission timer for active-close: RFC 4340, 8.3 requires
688 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
689 * state can be left. The initial timeout is 2 RTTs.
690 * Since RTT measurement is done by the CCIDs, there is no easy
691 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
692 * is too low (200ms); we use a high value to avoid unnecessary
693 * retransmissions when the link RTT is > 0.2 seconds.
694 * FIXME: Let main module sample RTTs and use that instead.
695 */
696 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
697 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
Gerrit Renker8695e802011-07-03 09:51:29 -0600698 }
699 dccp_transmit_skb(sk, skb);
Arnaldo Carvalho de Melo7c657872005-08-09 20:14:34 -0700700}