blob: 075deef5d826680c71e361d0be4bb6714d6a6ac5 [file] [log] [blame]
Willy Tarreaucff64112008-11-03 06:26:53 +01001/*
2 * Functions managing stream_interface structures
3 *
Willy Tarreauf873d752012-05-11 17:47:17 +02004 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
Willy Tarreaucff64112008-11-03 06:26:53 +01005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <errno.h>
14#include <fcntl.h>
15#include <stdio.h>
16#include <stdlib.h>
17
18#include <sys/socket.h>
19#include <sys/stat.h>
20#include <sys/types.h>
21
22#include <common/compat.h>
23#include <common/config.h>
24#include <common/debug.h>
25#include <common/standard.h>
26#include <common/ticks.h>
27#include <common/time.h>
28
Willy Tarreauc7e42382012-08-24 19:22:53 +020029#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020030#include <proto/connection.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010031#include <proto/fd.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020032#include <proto/pipe.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020033#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010034#include <proto/task.h>
35
Willy Tarreaufd31e532012-07-23 18:24:25 +020036#include <types/pipe.h>
37
Willy Tarreauf873d752012-05-11 17:47:17 +020038/* socket functions used when running a stream interface as a task */
Willy Tarreauf873d752012-05-11 17:47:17 +020039static void stream_int_update_embedded(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020040static void stream_int_shutr(struct stream_interface *si);
41static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020042static void stream_int_chk_rcv(struct stream_interface *si);
43static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020044static void stream_int_update_conn(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020045static void stream_int_shutr_conn(struct stream_interface *si);
46static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020047static void stream_int_chk_rcv_conn(struct stream_interface *si);
48static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau4aa36832012-10-02 20:07:22 +020049static void si_conn_recv_cb(struct connection *conn);
50static void si_conn_send_cb(struct connection *conn);
Willy Tarreau2396c1c2012-10-03 21:12:16 +020051static int si_conn_wake_cb(struct connection *conn);
Willy Tarreau27375622013-12-17 00:00:28 +010052static int si_idle_conn_wake_cb(struct connection *conn);
53static void si_idle_conn_null_cb(struct connection *conn);
Willy Tarreauf873d752012-05-11 17:47:17 +020054
Willy Tarreauc5788912012-08-24 18:12:41 +020055/* stream-interface operations for embedded tasks */
56struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020057 .update = stream_int_update_embedded,
Willy Tarreau5c979a92012-05-07 17:15:39 +020058 .chk_rcv = stream_int_chk_rcv,
59 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020060 .shutr = stream_int_shutr,
61 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020062};
63
Willy Tarreauc5788912012-08-24 18:12:41 +020064/* stream-interface operations for connections */
65struct si_ops si_conn_ops = {
66 .update = stream_int_update_conn,
67 .chk_rcv = stream_int_chk_rcv_conn,
68 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020069 .shutr = stream_int_shutr_conn,
70 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020071};
72
Willy Tarreau74beec32012-10-03 00:41:04 +020073struct data_cb si_conn_cb = {
Willy Tarreauc5788912012-08-24 18:12:41 +020074 .recv = si_conn_recv_cb,
75 .send = si_conn_send_cb,
Willy Tarreau4aa36832012-10-02 20:07:22 +020076 .wake = si_conn_wake_cb,
Willy Tarreauc5788912012-08-24 18:12:41 +020077};
78
Willy Tarreau27375622013-12-17 00:00:28 +010079struct data_cb si_idle_conn_cb = {
80 .recv = si_idle_conn_null_cb,
81 .send = si_idle_conn_null_cb,
82 .wake = si_idle_conn_wake_cb,
83};
84
Willy Tarreaucff64112008-11-03 06:26:53 +010085/*
86 * This function only has to be called once after a wakeup event in case of
87 * suspected timeout. It controls the stream interface timeouts and sets
88 * si->flags accordingly. It does NOT close anything, as this timeout may
89 * be used for any purpose. It returns 1 if the timeout fired, otherwise
90 * zero.
91 */
92int stream_int_check_timeouts(struct stream_interface *si)
93{
94 if (tick_is_expired(si->exp, now_ms)) {
95 si->flags |= SI_FL_EXP;
96 return 1;
97 }
98 return 0;
99}
100
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100101/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +0100102void stream_int_report_error(struct stream_interface *si)
103{
104 if (!si->err_type)
105 si->err_type = SI_ET_DATA_ERR;
106
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200107 si->ob->flags |= CF_WRITE_ERROR;
108 si->ib->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100109}
110
111/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100112 * Returns a message to the client ; the connection is shut down for read,
113 * and the request is cleared so that no server connection can be initiated.
114 * The buffer is marked for read shutdown on the other side to protect the
115 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100116 * "chunk". If it is null, then an empty message is used. The reply buffer does
117 * not need to be empty before this, and its contents will not be overwritten.
118 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100119 */
120void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
121{
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200122 channel_auto_read(si->ib);
123 channel_abort(si->ib);
124 channel_auto_close(si->ib);
125 channel_erase(si->ib);
Willy Tarreau798e1282010-12-12 13:06:00 +0100126
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200127 bi_erase(si->ob);
Willy Tarreau148d0992010-01-10 10:21:21 +0100128 if (likely(msg && msg->len))
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200129 bo_inject(si->ob, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100130
131 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200132 channel_auto_read(si->ob);
133 channel_auto_close(si->ob);
134 channel_shutr_now(si->ob);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100135}
136
Willy Tarreaufb90d942009-09-05 20:57:35 +0200137/* default update function for embedded tasks, to be used at the end of the i/o handler */
Willy Tarreauf873d752012-05-11 17:47:17 +0200138static void stream_int_update_embedded(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200139{
Willy Tarreau3488e252010-08-09 16:24:56 +0200140 int old_flags = si->flags;
141
Willy Tarreaufb90d942009-09-05 20:57:35 +0200142 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
143 __FUNCTION__,
144 si, si->state, si->ib->flags, si->ob->flags);
145
146 if (si->state != SI_ST_EST)
147 return;
148
Willy Tarreaub31c9712012-11-11 23:05:39 +0100149 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200150 channel_is_empty(si->ob))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200151 si_shutw(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200152
Willy Tarreaub31c9712012-11-11 23:05:39 +0100153 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200154 si->flags |= SI_FL_WAIT_DATA;
155
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200156 /* we're almost sure that we need some space if the buffer is not
157 * empty, even if it's not full, because the applets can't fill it.
158 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200159 if ((si->ib->flags & (CF_SHUTR|CF_DONT_READ)) == 0 && !channel_is_empty(si->ib))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200160 si->flags |= SI_FL_WAIT_ROOM;
161
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200162 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200163 if (tick_isset(si->ob->wex))
164 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
165 }
166
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200167 if (si->ib->flags & CF_READ_ACTIVITY ||
168 (si->ob->flags & CF_WRITE_ACTIVITY && !(si->flags & SI_FL_INDEP_STR))) {
Willy Tarreauf27b5ea2009-10-03 22:01:18 +0200169 if (tick_isset(si->ib->rex))
170 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
171 }
172
Willy Tarreau3488e252010-08-09 16:24:56 +0200173 /* save flags to detect changes */
174 old_flags = si->flags;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200175 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200176 !channel_full(si->ob) &&
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200177 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200178 si_chk_rcv(si->ob->prod);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200179
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200180 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200181 (si->ib->cons->flags & SI_FL_WAIT_DATA)) {
Willy Tarreau73b013b2012-05-21 16:31:45 +0200182 si_chk_snd(si->ib->cons);
Willy Tarreau3488e252010-08-09 16:24:56 +0200183 /* check if the consumer has freed some space */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200184 if (!channel_full(si->ib))
Willy Tarreau3488e252010-08-09 16:24:56 +0200185 si->flags &= ~SI_FL_WAIT_ROOM;
186 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200187
188 /* Note that we're trying to wake up in two conditions here :
189 * - special event, which needs the holder task attention
190 * - status indicating that the applet can go on working. This
191 * is rather hard because we might be blocking on output and
192 * don't want to wake up on input and vice-versa. The idea is
Willy Tarreau3488e252010-08-09 16:24:56 +0200193 * to only rely on the changes the chk_* might have performed.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200194 */
195 if (/* check stream interface changes */
Willy Tarreau3488e252010-08-09 16:24:56 +0200196 ((old_flags & ~si->flags) & (SI_FL_WAIT_ROOM|SI_FL_WAIT_DATA)) ||
197
198 /* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200199 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200200 si->state != SI_ST_EST ||
201 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200202 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200203 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
204
205 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200206 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
207 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreaue6300be2014-01-25 02:33:21 +0100208 ((si->ob->flags & CF_SHUTW) ||
209 ((si->ob->flags & CF_WAKE_WRITE) &&
210 (si->ob->prod->state != SI_ST_EST ||
211 (channel_is_empty(si->ob) && !si->ob->to_forward)))))) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200212 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
213 task_wakeup(si->owner, TASK_WOKEN_IO);
214 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200215 if (si->ib->flags & CF_READ_ACTIVITY)
216 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200217}
218
Willy Tarreau4a36b562012-08-06 19:31:45 +0200219/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200220 * This function performs a shutdown-read on a stream interface attached to an
221 * applet in a connected or init state (it does nothing for other states). It
222 * either shuts the read side or marks itself as closed. The buffer flags are
223 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
224 * we also forward the close to the write side. The owner task is woken up if
Willy Tarreau6fe15412013-09-29 15:16:03 +0200225 * it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200226 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200227static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200228{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200229 si->ib->flags &= ~CF_SHUTR_NOW;
230 if (si->ib->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200231 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200232 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200233 si->ib->rex = TICK_ETERNITY;
234 si->flags &= ~SI_FL_WAIT_ROOM;
235
236 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200237 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200238
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200239 if (si->ob->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200240 si->state = SI_ST_DIS;
241 si->exp = TICK_ETERNITY;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200242 si_applet_release(si);
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200243 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200244 else if (si->flags & SI_FL_NOHALF) {
245 /* we want to immediately forward this close to the write side */
246 return stream_int_shutw(si);
247 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200248
Willy Tarreau4a36b562012-08-06 19:31:45 +0200249 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200250 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200251 task_wakeup(si->owner, TASK_WOKEN_IO);
252}
253
Willy Tarreau4a36b562012-08-06 19:31:45 +0200254/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200255 * This function performs a shutdown-write on a stream interface attached to an
256 * applet in a connected or init state (it does nothing for other states). It
257 * either shuts the write side or marks itself as closed. The buffer flags are
258 * updated to reflect the new state. It does also close everything if the SI
259 * was marked as being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200260 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200261static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200262{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200263 si->ob->flags &= ~CF_SHUTW_NOW;
264 if (si->ob->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200265 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200266 si->ob->flags |= CF_SHUTW;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200267 si->ob->wex = TICK_ETERNITY;
268 si->flags &= ~SI_FL_WAIT_DATA;
269
270 switch (si->state) {
271 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200272 /* we have to shut before closing, otherwise some short messages
273 * may never leave the system, especially when there are remaining
274 * unread data in the socket input buffer, or when nolinger is set.
275 * However, if SI_FL_NOLINGER is explicitly set, we know there is
276 * no risk so we close both sides immediately.
277 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200278 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
279 !(si->ib->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200280 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200281
282 /* fall through */
283 case SI_ST_CON:
284 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100285 case SI_ST_QUE:
286 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200287 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200288 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200289 si_applet_release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200290 default:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200291 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
292 si->ib->flags &= ~CF_SHUTR_NOW;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200293 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200294 si->ib->rex = TICK_ETERNITY;
295 si->exp = TICK_ETERNITY;
296 }
297
Willy Tarreau4a36b562012-08-06 19:31:45 +0200298 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200299 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200300 task_wakeup(si->owner, TASK_WOKEN_IO);
301}
302
303/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200304static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200305{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200306 struct channel *ib = si->ib;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200307
308 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
309 __FUNCTION__,
310 si, si->state, si->ib->flags, si->ob->flags);
311
Willy Tarreaub31c9712012-11-11 23:05:39 +0100312 if (unlikely(si->state != SI_ST_EST || (ib->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200313 return;
314
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200315 if (channel_full(ib)) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200316 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200317 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200318 }
319 else {
320 /* (re)start reading */
321 si->flags &= ~SI_FL_WAIT_ROOM;
322 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
323 task_wakeup(si->owner, TASK_WOKEN_IO);
324 }
325}
326
327/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200328static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200329{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200330 struct channel *ob = si->ob;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200331
332 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
333 __FUNCTION__,
334 si, si->state, si->ib->flags, si->ob->flags);
335
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200336 if (unlikely(si->state != SI_ST_EST || (si->ob->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200337 return;
338
339 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200340 channel_is_empty(ob)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200341 return;
342
343 /* Otherwise there are remaining data to be sent in the buffer,
344 * so we tell the handler.
345 */
346 si->flags &= ~SI_FL_WAIT_DATA;
347 if (!tick_isset(ob->wex))
348 ob->wex = tick_add_ifset(now_ms, ob->wto);
349
350 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
351 task_wakeup(si->owner, TASK_WOKEN_IO);
352}
353
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100354/* Register an applet to handle a stream_interface as part of the
355 * stream interface's owner task. The SI will wake it up everytime it
356 * is solicited. The task's processing function must call the applet's
357 * function before returning. It must be deleted by the task handler
358 * using stream_int_unregister_handler(), possibly from within the
359 * function itself. It also pre-initializes the applet's context and
360 * returns it (or NULL in case it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200361 */
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100362struct appctx *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200363{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100364 struct appctx *appctx;
365
Simon Horman7abd00d2011-08-13 08:03:51 +0900366 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si->owner);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200367
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100368 appctx = si_alloc_appctx(si);
Willy Tarreau02291242014-12-22 19:34:00 +0100369 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100370 return NULL;
371
372 appctx_set_applet(appctx, app);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200373 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100374 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200375}
376
Willy Tarreaufb90d942009-09-05 20:57:35 +0200377/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100378 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200379 */
380void stream_int_unregister_handler(struct stream_interface *si)
381{
Willy Tarreau32e3c6a2013-10-11 19:34:20 +0200382 si_detach(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200383}
384
Willy Tarreau2c6be842012-07-06 17:12:34 +0200385/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200386 * established. It returns 0 if it fails in a fatal way or needs to poll to go
387 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200388 * flags (the bit is provided in <flag> by the caller). It is designed to be
389 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200390 * Note that it can emit a PROXY line by relying on the other end's address
391 * when the connection is attached to a stream interface, or by resolving the
392 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200393 */
394int conn_si_send_proxy(struct connection *conn, unsigned int flag)
395{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200397 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200398 goto out_error;
399
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100400 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200401 goto out_error;
402
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100403 if (!fd_send_ready(conn->t.sock.fd))
404 goto out_wait;
405
Willy Tarreau2c6be842012-07-06 17:12:34 +0200406 /* If we have a PROXY line to send, we'll use this to validate the
407 * connection, in which case the connection is validated only once
408 * we've sent the whole proxy line. Otherwise we use connect().
409 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200410 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200411 int ret;
412
413 /* The target server expects a PROXY line to be sent first.
414 * If the send_proxy_ofs is negative, it corresponds to the
415 * offset to start sending from then end of the proxy string
416 * (which is recomputed every time since it's constant). If
417 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200418 * We can only send a "normal" PROXY line when the connection
419 * is attached to a stream interface. Otherwise we can only
420 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200421 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200422 if (conn->data == &si_conn_cb) {
423 struct stream_interface *si = conn->owner;
424 struct connection *remote = objt_conn(si->ob->prod->end);
David Safb76832014-05-08 23:42:08 -0400425 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), remote);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200426 }
427 else {
428 /* The target server expects a LOCAL line to be sent first. Retrieving
429 * local or remote addresses may fail until the connection is established.
430 */
431 conn_get_from_addr(conn);
432 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
433 goto out_wait;
434
435 conn_get_to_addr(conn);
436 if (!(conn->flags & CO_FL_ADDR_TO_SET))
437 goto out_wait;
438
David Safb76832014-05-08 23:42:08 -0400439 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200440 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200441
Willy Tarreau2c6be842012-07-06 17:12:34 +0200442 if (!ret)
443 goto out_error;
444
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200445 if (conn->send_proxy_ofs > 0)
446 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200447
Willy Tarreaua1a74742012-08-24 12:14:49 +0200448 /* we have to send trash from (ret+sp for -sp bytes). If the
449 * data layer has a pending write, we'll also set MSG_MORE.
450 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200451 ret = send(conn->t.sock.fd, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
Willy Tarreaua1a74742012-08-24 12:14:49 +0200452 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200453
454 if (ret == 0)
455 goto out_wait;
456
457 if (ret < 0) {
Willy Tarreau95742a42013-09-03 09:02:11 +0200458 if (errno == EAGAIN || errno == ENOTCONN)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200459 goto out_wait;
Willy Tarreau7fe45692013-12-04 23:37:56 +0100460 if (errno == EINTR)
461 continue;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100462 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200463 goto out_error;
464 }
465
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200466 conn->send_proxy_ofs += ret; /* becomes zero once complete */
467 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200468 goto out_wait;
469
470 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100471 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200472 }
473
Willy Tarreaua1a74742012-08-24 12:14:49 +0200474 /* The connection is ready now, simply return and let the connection
475 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200476 */
477 if (conn->flags & CO_FL_WAIT_L4_CONN)
478 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200479 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200480 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200481
482 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200483 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200484 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200485 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200486
487 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200488 __conn_sock_stop_recv(conn);
Willy Tarreaue1f50c42014-01-22 20:02:06 +0100489 fd_cant_send(conn->t.sock.fd);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200490 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200491}
492
Willy Tarreau27375622013-12-17 00:00:28 +0100493
494/* Tiny I/O callback called on recv/send I/O events on idle connections.
495 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
496 * is notified and can kill the connection.
497 */
498static void si_idle_conn_null_cb(struct connection *conn)
499{
500 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
501 return;
502
Willy Tarreau46be2e52014-01-20 12:10:52 +0100503 if (fdtab[conn->t.sock.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) {
504 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreau27375622013-12-17 00:00:28 +0100505 conn->flags |= CO_FL_SOCK_RD_SH;
Willy Tarreau46be2e52014-01-20 12:10:52 +0100506 }
507 else {
508 conn_drain(conn);
509 }
Willy Tarreau27375622013-12-17 00:00:28 +0100510
511 /* disable draining if we were called and have no drain function */
512 if (!conn->ctrl->drain)
513 __conn_data_stop_recv(conn);
514}
515
516/* Callback to be used by connection I/O handlers when some activity is detected
517 * on an idle server connection. Its main purpose is to kill the connection once
518 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
519 * it killed the connection.
520 */
521static int si_idle_conn_wake_cb(struct connection *conn)
522{
523 struct stream_interface *si = conn->owner;
524
525 if (!conn_ctrl_ready(conn))
526 return 0;
527
528 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
529 /* warning, we can't do anything on <conn> after this call ! */
530 conn_force_close(conn);
531 conn_free(conn);
532 si->end = NULL;
533 return -1;
534 }
535 return 0;
536}
537
Willy Tarreau100c4672012-08-20 12:06:26 +0200538/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200539 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200540 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200541 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200542 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200543 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200544static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200545{
Willy Tarreaue603e692012-09-27 22:20:41 +0200546 struct stream_interface *si = conn->owner;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200547
548 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
549 __FUNCTION__,
550 si, si->state, si->ib->flags, si->ob->flags);
551
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200552 if (conn->flags & CO_FL_ERROR)
553 si->flags |= SI_FL_ERR;
554
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200555 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200556 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200557 si->exp = TICK_ETERNITY;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200558 si->ob->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200559 }
560
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200561 /* process consumer side */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200562 if (channel_is_empty(si->ob)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100563 if (((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200564 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200565 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200566 __conn_data_stop_send(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200567 si->ob->wex = TICK_ETERNITY;
568 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200569
Willy Tarreaub31c9712012-11-11 23:05:39 +0100570 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200571 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200572
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200573 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200574 /* update timeouts if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200575 if ((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200576 !channel_is_empty(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200577 if (tick_isset(si->ob->wex))
578 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200579
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200580 if (!(si->flags & SI_FL_INDEP_STR))
581 if (tick_isset(si->ib->rex))
582 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200583
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200584 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200585 !channel_full(si->ob) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200586 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
587 si_chk_rcv(si->ob->prod);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200588 }
589
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200590 /* process producer side.
591 * We might have some data the consumer is waiting for.
592 * We can do fast-forwarding, but we avoid doing this for partial
593 * buffers, because it is very likely that it will be done again
594 * immediately afterwards once the following data is parsed (eg:
595 * HTTP chunking).
596 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200597 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200598 (si->ib->pipe /* always try to send spliced data */ ||
Willy Tarreau9b28e032012-10-12 23:49:43 +0200599 (si->ib->buf->i == 0 && (si->ib->cons->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200600 int last_len = si->ib->pipe ? si->ib->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200601
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200602 si_chk_snd(si->ib->cons);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200603
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200604 /* check if the consumer has freed some space either in the
605 * buffer or in the pipe.
606 */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200607 if (!channel_full(si->ib) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200608 (!last_len || !si->ib->pipe || si->ib->pipe->data < last_len))
609 si->flags &= ~SI_FL_WAIT_ROOM;
610 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200611
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200612 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200613 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200614 si->ib->rex = TICK_ETERNITY;
615 }
Willy Tarreau66572762012-12-19 17:34:17 +0100616 else if ((si->ib->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200617 !channel_full(si->ib)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100618 /* we must re-enable reading if si_chk_snd() has freed some space */
619 __conn_data_want_recv(conn);
Willy Tarreau66572762012-12-19 17:34:17 +0100620 if (!(si->ib->flags & CF_READ_NOEXP) && tick_isset(si->ib->rex))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200621 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200622 }
623
624 /* wake the task up only when needed */
625 if (/* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200626 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200627 si->state != SI_ST_EST ||
628 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200629 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreaufd31e532012-07-23 18:24:25 +0200630 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
631
632 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200633 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
634 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreaue6300be2014-01-25 02:33:21 +0100635 ((si->ob->flags & CF_SHUTW) ||
636 ((si->ob->flags & CF_WAKE_WRITE) &&
637 (si->ob->prod->state != SI_ST_EST ||
638 (channel_is_empty(si->ob) && !si->ob->to_forward)))))) {
Willy Tarreaufd31e532012-07-23 18:24:25 +0200639 task_wakeup(si->owner, TASK_WOKEN_IO);
640 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200641 if (si->ib->flags & CF_READ_ACTIVITY)
642 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200643 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200644}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200645
Willy Tarreau5368d802012-08-21 18:22:06 +0200646/*
647 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800648 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800649 * caller to commit polling changes. The caller should check conn->flags
650 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200651 */
Godbach4f489902013-12-04 17:24:06 +0800652static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200653{
Willy Tarreaue603e692012-09-27 22:20:41 +0200654 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200655 struct channel *chn = si->ob;
Willy Tarreau5368d802012-08-21 18:22:06 +0200656 int ret;
657
Willy Tarreaucb76e592012-10-12 23:56:57 +0200658 if (chn->pipe && conn->xprt->snd_pipe) {
659 ret = conn->xprt->snd_pipe(conn, chn->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200660 if (ret > 0)
Willy Tarreaub4d05092014-09-01 20:35:55 +0200661 chn->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200662
Willy Tarreaucb76e592012-10-12 23:56:57 +0200663 if (!chn->pipe->data) {
664 put_pipe(chn->pipe);
665 chn->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200666 }
667
Willy Tarreau96199b12012-08-24 00:46:52 +0200668 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800669 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200670 }
671
672 /* At this point, the pipe is empty, but we may still have data pending
673 * in the normal buffer.
674 */
Willy Tarreau9b28e032012-10-12 23:49:43 +0200675 if (!chn->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800676 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200677
Godbache68e02d2013-10-11 15:48:29 +0800678 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200679 * data left, and that there are sendable buffered data.
680 */
Willy Tarreau310987a2014-01-22 19:46:33 +0100681 if (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH | CO_FL_WAIT_DATA | CO_FL_HANDSHAKE))) {
Willy Tarreau5368d802012-08-21 18:22:06 +0200682 /* check if we want to inform the kernel that we're interested in
683 * sending more data after this call. We want this if :
684 * - we're about to close after this last send and want to merge
685 * the ongoing FIN with the last segment.
686 * - we know we can't send everything at once and must get back
687 * here because of unaligned data
688 * - there is still a finite amount of data to forward
689 * The test is arranged so that the most common case does only 2
690 * tests.
691 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100692 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200693
Willy Tarreaucb76e592012-10-12 23:56:57 +0200694 if ((!(chn->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
695 ((chn->to_forward && chn->to_forward != CHN_INFINITE_FORWARD) ||
696 (chn->flags & CF_EXPECT_MORE))) ||
Willy Tarreaub31c9712012-11-11 23:05:39 +0100697 ((chn->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100698 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200699
Willy Tarreau7bed9452014-02-02 02:00:24 +0100700 if (chn->flags & CF_STREAMER)
701 send_flag |= CO_SFL_STREAMER;
702
Willy Tarreau9b28e032012-10-12 23:49:43 +0200703 ret = conn->xprt->snd_buf(conn, chn->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800704 if (ret > 0) {
Willy Tarreaub4d05092014-09-01 20:35:55 +0200705 chn->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200706
Godbache68e02d2013-10-11 15:48:29 +0800707 if (!chn->buf->o) {
708 /* Always clear both flags once everything has been sent, they're one-shot */
709 chn->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
710 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200711
Godbache68e02d2013-10-11 15:48:29 +0800712 /* if some data remain in the buffer, it's only because the
713 * system buffers are full, we will try next time.
714 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200715 }
Godbache68e02d2013-10-11 15:48:29 +0800716 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200717
Godbach4f489902013-12-04 17:24:06 +0800718 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200719}
720
721
Willy Tarreau100c4672012-08-20 12:06:26 +0200722/* Updates the timers and flags of a stream interface attached to a connection,
723 * depending on the buffers' flags. It should only be called once after the
724 * buffer flags have settled down, and before they are cleared. It doesn't
725 * harm to call it as often as desired (it just slightly hurts performance).
726 * It is only meant to be called by upper layers after buffer flags have been
727 * manipulated by analysers.
728 */
729void stream_int_update_conn(struct stream_interface *si)
730{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200731 struct channel *ib = si->ib;
732 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200733 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200734
Willy Tarreau100c4672012-08-20 12:06:26 +0200735 /* Check if we need to close the read side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200736 if (!(ib->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200737 /* Read not closed, update FD status and timeout for reads */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100738 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200739 /* stop reading */
740 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100741 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200742 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200743 conn_data_stop_recv(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200744 ib->rex = TICK_ETERNITY;
745 }
746 }
747 else {
748 /* (re)start reading and update timeout. Note: we don't recompute the timeout
749 * everytime we get here, otherwise it would risk never to expire. We only
750 * update it if is was not yet set. The stream socket handler will already
751 * have updated it if there has been a completed I/O.
752 */
753 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200754 conn_data_want_recv(conn);
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200755 if (!(ib->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ib->rex))
Willy Tarreau100c4672012-08-20 12:06:26 +0200756 ib->rex = tick_add_ifset(now_ms, ib->rto);
757 }
758 }
759
760 /* Check if we need to close the write side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200761 if (!(ob->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200762 /* Write not closed, update FD status and timeout for writes */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200763 if (channel_is_empty(ob)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200764 /* stop writing */
765 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100766 if ((ob->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200767 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200768 conn_data_stop_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200769 ob->wex = TICK_ETERNITY;
770 }
771 }
772 else {
773 /* (re)start writing and update timeout. Note: we don't recompute the timeout
774 * everytime we get here, otherwise it would risk never to expire. We only
775 * update it if is was not yet set. The stream socket handler will already
776 * have updated it if there has been a completed I/O.
777 */
778 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200779 conn_data_want_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200780 if (!tick_isset(ob->wex)) {
781 ob->wex = tick_add_ifset(now_ms, ob->wto);
782 if (tick_isset(ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
783 /* Note: depending on the protocol, we don't know if we're waiting
784 * for incoming data or not. So in order to prevent the socket from
785 * expiring read timeouts during writes, we refresh the read timeout,
786 * except if it was already infinite or if we have explicitly setup
787 * independent streams.
788 */
789 ib->rex = tick_add_ifset(now_ms, ib->rto);
790 }
791 }
792 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200793 }
794}
795
796/*
797 * This function performs a shutdown-read on a stream interface attached to
798 * a connection in a connected or init state (it does nothing for other
799 * states). It either shuts the read side or marks itself as closed. The buffer
800 * flags are updated to reflect the new state. If the stream interface has
801 * SI_FL_NOHALF, we also forward the close to the write side. If a control
802 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200803 * descriptors are then shutdown or closed accordingly. The function
804 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200805 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200806static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200807{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200808 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200809
810 si->ib->flags &= ~CF_SHUTR_NOW;
811 if (si->ib->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200812 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200813 si->ib->flags |= CF_SHUTR;
814 si->ib->rex = TICK_ETERNITY;
815 si->flags &= ~SI_FL_WAIT_ROOM;
816
817 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200818 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200819
820 if (si->ob->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200821 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200822 si->state = SI_ST_DIS;
823 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200824 }
825 else if (si->flags & SI_FL_NOHALF) {
826 /* we want to immediately forward this close to the write side */
827 return stream_int_shutw_conn(si);
828 }
829 else if (conn->ctrl) {
830 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200831 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200832 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200833}
834
835/*
836 * This function performs a shutdown-write on a stream interface attached to
837 * a connection in a connected or init state (it does nothing for other
838 * states). It either shuts the write side or marks itself as closed. The
839 * buffer flags are updated to reflect the new state. It does also close
840 * everything if the SI was marked as being in error state. If there is a
841 * data-layer shutdown, it is called. If a control layer is defined, then it is
842 * supposed to be a socket layer and file descriptors are then shutdown or
Willy Tarreau6fe15412013-09-29 15:16:03 +0200843 * closed accordingly. The function automatically disables polling if needed.
844 * Note: at the moment, we continue to check conn->ctrl eventhough we *know* it
845 * is valid. This will help selecting the proper shutdown() and setsockopt()
846 * calls if/when we implement remote sockets later.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200847 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200848static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200849{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200850 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200851
852 si->ob->flags &= ~CF_SHUTW_NOW;
853 if (si->ob->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200854 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200855 si->ob->flags |= CF_SHUTW;
856 si->ob->wex = TICK_ETERNITY;
857 si->flags &= ~SI_FL_WAIT_DATA;
858
859 switch (si->state) {
860 case SI_ST_EST:
861 /* we have to shut before closing, otherwise some short messages
862 * may never leave the system, especially when there are remaining
863 * unread data in the socket input buffer, or when nolinger is set.
864 * However, if SI_FL_NOLINGER is explicitly set, we know there is
865 * no risk so we close both sides immediately.
866 */
867 if (si->flags & SI_FL_ERR) {
868 /* quick close, the socket is alredy shut anyway */
869 }
870 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200871 /* unclean data-layer shutdown */
872 if (conn->xprt && conn->xprt->shutw)
873 conn->xprt->shutw(conn, 0);
874 }
875 else {
876 /* clean data-layer shutdown */
877 if (conn->xprt && conn->xprt->shutw)
878 conn->xprt->shutw(conn, 1);
879
880 /* If the stream interface is configured to disable half-open
881 * connections, we'll skip the shutdown(), but only if the
882 * read size is already closed. Otherwise we can't support
883 * closed write with pending read (eg: abortonclose while
884 * waiting for the server).
885 */
886 if (!(si->flags & SI_FL_NOHALF) || !(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
887 /* We shutdown transport layer */
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100888 if (conn_ctrl_ready(conn))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200889 shutdown(conn->t.sock.fd, SHUT_WR);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200890
891 if (!(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
892 /* OK just a shutw, but we want the caller
893 * to disable polling on this FD if exists.
894 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200895 if (conn->ctrl)
896 conn_data_stop_send(conn);
897 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200898 }
899 }
900 }
901
902 /* fall through */
903 case SI_ST_CON:
904 /* we may have to close a pending connection, and mark the
905 * response buffer as shutr
906 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200907 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200908 /* fall through */
909 case SI_ST_CER:
910 case SI_ST_QUE:
911 case SI_ST_TAR:
912 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200913 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200914 default:
915 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
916 si->ib->flags &= ~CF_SHUTR_NOW;
917 si->ib->flags |= CF_SHUTR;
918 si->ib->rex = TICK_ETERNITY;
919 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200920 }
921}
922
Willy Tarreau46a8d922012-08-20 12:38:36 +0200923/* This function is used for inter-stream-interface calls. It is called by the
924 * consumer to inform the producer side that it may be interested in checking
925 * for free space in the buffer. Note that it intentionally does not update
926 * timeouts, so that we can still check them later at wake-up. This function is
927 * dedicated to connection-based stream interfaces.
928 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200929static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200930{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200931 struct channel *ib = si->ib;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200932 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200933
Willy Tarreau34ffd772012-09-03 16:51:27 +0200934 if (unlikely(si->state > SI_ST_EST || (ib->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200935 return;
936
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200937 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100938
Willy Tarreaub31c9712012-11-11 23:05:39 +0100939 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200940 /* stop reading */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100941 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200942 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200943 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200944 }
945 else {
946 /* (re)start reading */
947 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200948 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200949 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200950 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200951}
952
953
Willy Tarreaude5722c2012-08-20 15:01:10 +0200954/* This function is used for inter-stream-interface calls. It is called by the
955 * producer to inform the consumer side that it may be interested in checking
956 * for data in the buffer. Note that it intentionally does not update timeouts,
957 * so that we can still check them later at wake-up.
958 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200959static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200960{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200961 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200962 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200963
Willy Tarreau34ffd772012-09-03 16:51:27 +0200964 if (unlikely(si->state > SI_ST_EST || (ob->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200965 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200966
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200967 if (unlikely(channel_is_empty(ob))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200968 return;
969
970 if (!ob->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100971 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200972 return;
973
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200974 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200975 /* already subscribed to write notifications, will be called
976 * anyway, so let's avoid calling it especially if the reader
977 * is not ready.
978 */
979 return;
980 }
981
Willy Tarreau708e7172014-01-21 10:27:49 +0100982 /* Before calling the data-level operations, we have to prepare
983 * the polling flags to ensure we properly detect changes.
984 */
985 conn_refresh_polling_flags(conn);
986 __conn_data_want_send(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100987
Willy Tarreau708e7172014-01-21 10:27:49 +0100988 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200989 si_conn_send(conn);
Willy Tarreau798c3c92014-01-21 10:30:08 +0100990 if (conn->flags & CO_FL_ERROR) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100991 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200992 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100993 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100994 goto out_wakeup;
995 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200996 }
997
998 /* OK, so now we know that some data might have been sent, and that we may
999 * have to poll first. We have to do that too if the buffer is not empty.
1000 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +02001001 if (channel_is_empty(ob)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001002 /* the connection is established but we can't write. Either the
1003 * buffer is empty, or we just refrain from sending because the
1004 * ->o limit was reached. Maybe we just wrote the last
1005 * chunk and need to close.
1006 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001007 __conn_data_stop_send(conn);
Willy Tarreaub31c9712012-11-11 23:05:39 +01001008 if (((ob->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001009 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +02001010 (si->state == SI_ST_EST)) {
1011 si_shutw(si);
1012 goto out_wakeup;
1013 }
1014
Willy Tarreaub31c9712012-11-11 23:05:39 +01001015 if ((ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001016 si->flags |= SI_FL_WAIT_DATA;
1017 ob->wex = TICK_ETERNITY;
1018 }
1019 else {
1020 /* Otherwise there are remaining data to be sent in the buffer,
1021 * which means we have to poll before doing so.
1022 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001023 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001024 si->flags &= ~SI_FL_WAIT_DATA;
1025 if (!tick_isset(ob->wex))
1026 ob->wex = tick_add_ifset(now_ms, ob->wto);
1027 }
1028
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001029 if (likely(ob->flags & CF_WRITE_ACTIVITY)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001030 /* update timeout if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001031 if ((ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +02001032 !channel_is_empty(ob))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001033 ob->wex = tick_add_ifset(now_ms, ob->wto);
1034
1035 if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
1036 /* Note: to prevent the client from expiring read timeouts
1037 * during writes, we refresh it. We only do this if the
1038 * interface is not configured for "independent streams",
1039 * because for some applications it's better not to do this,
1040 * for instance when continuously exchanging small amounts
1041 * of data which can full the socket buffers long before a
1042 * write timeout is detected.
1043 */
1044 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
1045 }
1046 }
1047
1048 /* in case of special condition (error, shutdown, end of write...), we
1049 * have to notify the task.
1050 */
Willy Tarreaue6300be2014-01-25 02:33:21 +01001051 if (likely((ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1052 ((ob->flags & CF_WAKE_WRITE) &&
1053 ((channel_is_empty(si->ob) && !ob->to_forward) ||
1054 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001055 out_wakeup:
1056 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
1057 task_wakeup(si->owner, TASK_WOKEN_IO);
1058 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001059
1060 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001061 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001062}
1063
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001064/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001065 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001066 * into the buffer from the connection. It iterates over the transport layer's
1067 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001068 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001069static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001070{
Willy Tarreaue603e692012-09-27 22:20:41 +02001071 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001072 struct channel *chn = si->ib;
Willy Tarreauce323de2012-08-20 21:41:06 +02001073 int ret, max, cur_read;
1074 int read_poll = MAX_READ_POLL_LOOPS;
1075
1076 /* stop immediately on errors. Note that we DON'T want to stop on
1077 * POLL_ERR, as the poller might report a write error while there
1078 * are still data available in the recv buffer. This typically
1079 * happens when we send too large a request to a backend server
1080 * which rejects it before reading it all.
1081 */
1082 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001083 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001084
1085 /* stop here if we reached the end of data */
1086 if (conn_data_read0_pending(conn))
1087 goto out_shutdown_r;
1088
1089 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001090 if (chn->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001091 return;
1092
Willy Tarreau96199b12012-08-24 00:46:52 +02001093 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001094
Willy Tarreauc5890e62014-02-09 17:47:01 +01001095 if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !chn->buf->o &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001096 global.tune.idle_timer &&
1097 (unsigned short)(now_ms - chn->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001098 /* The buffer was empty and nothing was transferred for more
1099 * than one second. This was caused by a pause and not by
1100 * congestion. Reset any streaming mode to reduce latency.
1101 */
1102 chn->xfer_small = 0;
1103 chn->xfer_large = 0;
1104 chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
1105 }
1106
Willy Tarreau96199b12012-08-24 00:46:52 +02001107 /* First, let's see if we may splice data across the channel without
1108 * using a buffer.
1109 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001110 if (conn->xprt->rcv_pipe &&
Willy Tarreaufa8e2bc2013-07-18 22:21:54 +02001111 (chn->pipe || chn->to_forward >= MIN_SPLICE_FORWARD) &&
1112 chn->flags & CF_KERN_SPLICING) {
Willy Tarreau9b28e032012-10-12 23:49:43 +02001113 if (buffer_not_empty(chn->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001114 /* We're embarrassed, there are already data pending in
1115 * the buffer and we don't want to have them at two
1116 * locations at a time. Let's indicate we need some
1117 * place and ask the consumer to hurry.
1118 */
1119 goto abort_splice;
1120 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001121
Willy Tarreaucb76e592012-10-12 23:56:57 +02001122 if (unlikely(chn->pipe == NULL)) {
1123 if (pipes_used >= global.maxpipes || !(chn->pipe = get_pipe())) {
1124 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001125 goto abort_splice;
1126 }
1127 }
1128
Willy Tarreaucb76e592012-10-12 23:56:57 +02001129 ret = conn->xprt->rcv_pipe(conn, chn->pipe, chn->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001130 if (ret < 0) {
1131 /* splice not supported on this end, let's disable it */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001132 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001133 goto abort_splice;
1134 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001135
Willy Tarreau96199b12012-08-24 00:46:52 +02001136 if (ret > 0) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001137 if (chn->to_forward != CHN_INFINITE_FORWARD)
1138 chn->to_forward -= ret;
1139 chn->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001140 cur_read += ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001141 chn->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001142 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001143
1144 if (conn_data_read0_pending(conn))
1145 goto out_shutdown_r;
1146
1147 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001148 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001149
Willy Tarreau61d39a02013-07-18 21:49:32 +02001150 if (conn->flags & CO_FL_WAIT_ROOM) {
1151 /* the pipe is full or we have read enough data that it
1152 * could soon be full. Let's stop before needing to poll.
1153 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001154 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001155 __conn_data_stop_recv(conn);
1156 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001157
Willy Tarreauce323de2012-08-20 21:41:06 +02001158 /* splice not possible (anymore), let's go on on standard copy */
1159 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001160
1161 abort_splice:
Willy Tarreau61d39a02013-07-18 21:49:32 +02001162 if (chn->pipe && unlikely(!chn->pipe->data)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001163 put_pipe(chn->pipe);
1164 chn->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001165 }
1166
Willy Tarreau61d39a02013-07-18 21:49:32 +02001167 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1168 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1169 * that if such an event is not handled above in splice, it will be handled here by
1170 * recv().
1171 */
Willy Tarreau310987a2014-01-22 19:46:33 +01001172 while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001173 max = bi_avail(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001174
1175 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001176 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001177 break;
1178 }
1179
Willy Tarreau9b28e032012-10-12 23:49:43 +02001180 ret = conn->xprt->rcv_buf(conn, chn->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001181 if (ret <= 0)
1182 break;
1183
1184 cur_read += ret;
1185
1186 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001187 if (chn->to_forward && !(chn->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001188 unsigned long fwd = ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001189 if (chn->to_forward != CHN_INFINITE_FORWARD) {
1190 if (fwd > chn->to_forward)
1191 fwd = chn->to_forward;
1192 chn->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001193 }
Willy Tarreau9b28e032012-10-12 23:49:43 +02001194 b_adv(chn->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001195 }
1196
Willy Tarreaucb76e592012-10-12 23:56:57 +02001197 chn->flags |= CF_READ_PARTIAL;
1198 chn->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001199
Willy Tarreaucb76e592012-10-12 23:56:57 +02001200 if (channel_full(chn)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001201 si->flags |= SI_FL_WAIT_ROOM;
1202 break;
1203 }
1204
Willy Tarreau5fddab02012-11-09 18:27:26 +01001205 if ((chn->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001206 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001207 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001208 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001209 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001210
1211 /* if too many bytes were missing from last read, it means that
1212 * it's pointless trying to read again because the system does
1213 * not have them in buffers.
1214 */
1215 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001216 /* if a streamer has read few data, it may be because we
1217 * have exhausted system buffers. It's not worth trying
1218 * again.
1219 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001220 if (chn->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001221 break;
1222
1223 /* if we read a large block smaller than what we requested,
1224 * it's almost certain we'll never get anything more.
1225 */
1226 if (ret >= global.tune.recv_enough)
1227 break;
1228 }
1229 } /* while !flags */
1230
Willy Tarreau96199b12012-08-24 00:46:52 +02001231 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001232 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001233
Willy Tarreauc5890e62014-02-09 17:47:01 +01001234 if (cur_read) {
1235 if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
1236 (cur_read <= chn->buf->size / 2)) {
1237 chn->xfer_large = 0;
1238 chn->xfer_small++;
1239 if (chn->xfer_small >= 3) {
1240 /* we have read less than half of the buffer in
1241 * one pass, and this happened at least 3 times.
1242 * This is definitely not a streamer.
1243 */
1244 chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
1245 }
1246 else if (chn->xfer_small >= 2) {
1247 /* if the buffer has been at least half full twice,
1248 * we receive faster than we send, so at least it
1249 * is not a "fast streamer".
1250 */
1251 chn->flags &= ~CF_STREAMER_FAST;
1252 }
1253 }
1254 else if (!(chn->flags & CF_STREAMER_FAST) &&
1255 (cur_read >= chn->buf->size - global.tune.maxrewrite)) {
1256 /* we read a full buffer at once */
1257 chn->xfer_small = 0;
1258 chn->xfer_large++;
1259 if (chn->xfer_large >= 3) {
1260 /* we call this buffer a fast streamer if it manages
1261 * to be filled in one call 3 consecutive times.
1262 */
1263 chn->flags |= (CF_STREAMER | CF_STREAMER_FAST);
1264 }
1265 }
1266 else {
1267 chn->xfer_small = 0;
1268 chn->xfer_large = 0;
1269 }
1270 chn->last_read = now_ms;
1271 }
1272
Willy Tarreauce323de2012-08-20 21:41:06 +02001273 if (conn_data_read0_pending(conn))
1274 /* connection closed */
1275 goto out_shutdown_r;
1276
1277 return;
1278
1279 out_shutdown_r:
1280 /* we received a shutdown */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001281 chn->flags |= CF_READ_NULL;
1282 if (chn->flags & CF_AUTO_CLOSE)
1283 channel_shutw_now(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001284 stream_sock_read0(si);
1285 conn_data_read0(conn);
1286 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001287}
1288
1289/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001290 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001291 * from the buffer to the connection. It iterates over the transport layer's
1292 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001293 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001294static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001295{
Willy Tarreaue603e692012-09-27 22:20:41 +02001296 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001297 struct channel *chn = si->ob;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001298
1299 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001300 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001301
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001302 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001303 /* a handshake was requested */
1304 return;
1305
1306 /* we might have been called just after an asynchronous shutw */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001307 if (chn->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001308 return;
1309
1310 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001311 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001312
1313 /* OK all done */
1314 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001315}
1316
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001317/*
1318 * This function propagates a null read received on a socket-based connection.
1319 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
1320 * the close is also forwarded to the write side as an abort. This function is
1321 * still socket-specific as it handles a setsockopt() call to set the SO_LINGER
1322 * state on the socket.
1323 */
1324void stream_sock_read0(struct stream_interface *si)
1325{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001326 struct connection *conn = __objt_conn(si->end);
1327
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001328 si->ib->flags &= ~CF_SHUTR_NOW;
1329 if (si->ib->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001330 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001331 si->ib->flags |= CF_SHUTR;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001332 si->ib->rex = TICK_ETERNITY;
1333 si->flags &= ~SI_FL_WAIT_ROOM;
1334
1335 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1336 return;
1337
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001338 if (si->ob->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001339 goto do_close;
1340
1341 if (si->flags & SI_FL_NOHALF) {
1342 /* we want to immediately forward this close to the write side */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001343 /* force flag on ssl to keep session in cache */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001344 if (conn->xprt->shutw)
1345 conn->xprt->shutw(conn, 0);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001346 goto do_close;
1347 }
1348
1349 /* otherwise that's just a normal read shutdown */
Willy Tarreauad38ace2013-12-15 14:19:38 +01001350 if (conn_ctrl_ready(conn))
1351 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001352 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001353 return;
1354
1355 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001356 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001357 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001358
1359 si->ib->flags &= ~CF_SHUTR_NOW;
1360 si->ib->flags |= CF_SHUTR;
1361 si->ib->rex = TICK_ETERNITY;
1362
1363 si->ob->flags &= ~CF_SHUTW_NOW;
1364 si->ob->flags |= CF_SHUTW;
1365 si->ob->wex = TICK_ETERNITY;
1366
1367 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1368
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001369 si->state = SI_ST_DIS;
1370 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001371 return;
1372}
1373
Willy Tarreaudded32d2008-11-30 19:48:07 +01001374/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001375 * Local variables:
1376 * c-indent-level: 8
1377 * c-basic-offset: 8
1378 * End:
1379 */