blob: c76c7ae5e890078459c76dc3f56ec1c67be3ed40 [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 Tarreaud7ad9f52013-12-31 17:26:25 +0100208 (si->ob->flags & (CF_SHUTW|CF_WAKE_WRITE)))) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200209 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
210 task_wakeup(si->owner, TASK_WOKEN_IO);
211 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200212 if (si->ib->flags & CF_READ_ACTIVITY)
213 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200214}
215
Willy Tarreau4a36b562012-08-06 19:31:45 +0200216/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200217 * This function performs a shutdown-read on a stream interface attached to an
218 * applet in a connected or init state (it does nothing for other states). It
219 * either shuts the read side or marks itself as closed. The buffer flags are
220 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
221 * we also forward the close to the write side. The owner task is woken up if
Willy Tarreau6fe15412013-09-29 15:16:03 +0200222 * it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200223 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200224static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200225{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200226 si->ib->flags &= ~CF_SHUTR_NOW;
227 if (si->ib->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200228 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200229 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200230 si->ib->rex = TICK_ETERNITY;
231 si->flags &= ~SI_FL_WAIT_ROOM;
232
233 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200234 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200235
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200236 if (si->ob->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200237 si->state = SI_ST_DIS;
238 si->exp = TICK_ETERNITY;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200239 si_applet_release(si);
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200240 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200241 else if (si->flags & SI_FL_NOHALF) {
242 /* we want to immediately forward this close to the write side */
243 return stream_int_shutw(si);
244 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200245
Willy Tarreau4a36b562012-08-06 19:31:45 +0200246 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200247 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200248 task_wakeup(si->owner, TASK_WOKEN_IO);
249}
250
Willy Tarreau4a36b562012-08-06 19:31:45 +0200251/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200252 * This function performs a shutdown-write on a stream interface attached to an
253 * applet in a connected or init state (it does nothing for other states). It
254 * either shuts the write side or marks itself as closed. The buffer flags are
255 * updated to reflect the new state. It does also close everything if the SI
256 * was marked as being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200257 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200258static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200259{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200260 si->ob->flags &= ~CF_SHUTW_NOW;
261 if (si->ob->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200262 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200263 si->ob->flags |= CF_SHUTW;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200264 si->ob->wex = TICK_ETERNITY;
265 si->flags &= ~SI_FL_WAIT_DATA;
266
267 switch (si->state) {
268 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200269 /* we have to shut before closing, otherwise some short messages
270 * may never leave the system, especially when there are remaining
271 * unread data in the socket input buffer, or when nolinger is set.
272 * However, if SI_FL_NOLINGER is explicitly set, we know there is
273 * no risk so we close both sides immediately.
274 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200275 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
276 !(si->ib->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200277 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200278
279 /* fall through */
280 case SI_ST_CON:
281 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100282 case SI_ST_QUE:
283 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200284 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200285 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200286 si_applet_release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200287 default:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200288 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
289 si->ib->flags &= ~CF_SHUTR_NOW;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200290 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200291 si->ib->rex = TICK_ETERNITY;
292 si->exp = TICK_ETERNITY;
293 }
294
Willy Tarreau4a36b562012-08-06 19:31:45 +0200295 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200296 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200297 task_wakeup(si->owner, TASK_WOKEN_IO);
298}
299
300/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200301static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200302{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200303 struct channel *ib = si->ib;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200304
305 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
306 __FUNCTION__,
307 si, si->state, si->ib->flags, si->ob->flags);
308
Willy Tarreaub31c9712012-11-11 23:05:39 +0100309 if (unlikely(si->state != SI_ST_EST || (ib->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200310 return;
311
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200312 if (channel_full(ib)) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200313 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200314 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200315 }
316 else {
317 /* (re)start reading */
318 si->flags &= ~SI_FL_WAIT_ROOM;
319 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
320 task_wakeup(si->owner, TASK_WOKEN_IO);
321 }
322}
323
324/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200325static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200326{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200327 struct channel *ob = si->ob;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200328
329 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
330 __FUNCTION__,
331 si, si->state, si->ib->flags, si->ob->flags);
332
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200333 if (unlikely(si->state != SI_ST_EST || (si->ob->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200334 return;
335
336 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200337 channel_is_empty(ob)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200338 return;
339
340 /* Otherwise there are remaining data to be sent in the buffer,
341 * so we tell the handler.
342 */
343 si->flags &= ~SI_FL_WAIT_DATA;
344 if (!tick_isset(ob->wex))
345 ob->wex = tick_add_ifset(now_ms, ob->wto);
346
347 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
348 task_wakeup(si->owner, TASK_WOKEN_IO);
349}
350
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100351/* Register an applet to handle a stream_interface as part of the
352 * stream interface's owner task. The SI will wake it up everytime it
353 * is solicited. The task's processing function must call the applet's
354 * function before returning. It must be deleted by the task handler
355 * using stream_int_unregister_handler(), possibly from within the
356 * function itself. It also pre-initializes the applet's context and
357 * returns it (or NULL in case it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200358 */
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100359struct appctx *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200360{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100361 struct appctx *appctx;
362
Simon Horman7abd00d2011-08-13 08:03:51 +0900363 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si->owner);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200364
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100365 appctx = si_alloc_appctx(si);
366 if (!si)
367 return NULL;
368
369 appctx_set_applet(appctx, app);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200370 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100371 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200372}
373
Willy Tarreaufb90d942009-09-05 20:57:35 +0200374/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100375 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200376 */
377void stream_int_unregister_handler(struct stream_interface *si)
378{
Willy Tarreau32e3c6a2013-10-11 19:34:20 +0200379 si_detach(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200380}
381
Willy Tarreau2c6be842012-07-06 17:12:34 +0200382/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200383 * established. It returns 0 if it fails in a fatal way or needs to poll to go
384 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200385 * flags (the bit is provided in <flag> by the caller). It is designed to be
386 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200387 * Note that it can emit a PROXY line by relying on the other end's address
388 * when the connection is attached to a stream interface, or by resolving the
389 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200390 */
391int conn_si_send_proxy(struct connection *conn, unsigned int flag)
392{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200393 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200394 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200395 goto out_error;
396
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100397 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200398 goto out_error;
399
Willy Tarreau2c6be842012-07-06 17:12:34 +0200400 /* If we have a PROXY line to send, we'll use this to validate the
401 * connection, in which case the connection is validated only once
402 * we've sent the whole proxy line. Otherwise we use connect().
403 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200404 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200405 int ret;
406
407 /* The target server expects a PROXY line to be sent first.
408 * If the send_proxy_ofs is negative, it corresponds to the
409 * offset to start sending from then end of the proxy string
410 * (which is recomputed every time since it's constant). If
411 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200412 * We can only send a "normal" PROXY line when the connection
413 * is attached to a stream interface. Otherwise we can only
414 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200416 if (conn->data == &si_conn_cb) {
417 struct stream_interface *si = conn->owner;
418 struct connection *remote = objt_conn(si->ob->prod->end);
419
420 if (remote)
421 ret = make_proxy_line(trash.str, trash.size, &remote->addr.from, &remote->addr.to);
422 else
423 ret = make_proxy_line(trash.str, trash.size, NULL, NULL);
424 }
425 else {
426 /* The target server expects a LOCAL line to be sent first. Retrieving
427 * local or remote addresses may fail until the connection is established.
428 */
429 conn_get_from_addr(conn);
430 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
431 goto out_wait;
432
433 conn_get_to_addr(conn);
434 if (!(conn->flags & CO_FL_ADDR_TO_SET))
435 goto out_wait;
436
437 ret = make_proxy_line(trash.str, trash.size, &conn->addr.from, &conn->addr.to);
438 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200439
Willy Tarreau2c6be842012-07-06 17:12:34 +0200440 if (!ret)
441 goto out_error;
442
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200443 if (conn->send_proxy_ofs > 0)
444 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200445
Willy Tarreaua1a74742012-08-24 12:14:49 +0200446 /* we have to send trash from (ret+sp for -sp bytes). If the
447 * data layer has a pending write, we'll also set MSG_MORE.
448 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200449 ret = send(conn->t.sock.fd, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
Willy Tarreaua1a74742012-08-24 12:14:49 +0200450 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200451
452 if (ret == 0)
453 goto out_wait;
454
455 if (ret < 0) {
Willy Tarreau95742a42013-09-03 09:02:11 +0200456 if (errno == EAGAIN || errno == ENOTCONN)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200457 goto out_wait;
Willy Tarreau7fe45692013-12-04 23:37:56 +0100458 if (errno == EINTR)
459 continue;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100460 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200461 goto out_error;
462 }
463
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200464 conn->send_proxy_ofs += ret; /* becomes zero once complete */
465 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200466 goto out_wait;
467
468 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100469 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200470 }
471
Willy Tarreaua1a74742012-08-24 12:14:49 +0200472 /* The connection is ready now, simply return and let the connection
473 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200474 */
475 if (conn->flags & CO_FL_WAIT_L4_CONN)
476 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200477 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200478 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200479
480 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200481 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200482 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200483 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200484
485 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200486 __conn_sock_stop_recv(conn);
487 __conn_sock_poll_send(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200488 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200489}
490
Willy Tarreau27375622013-12-17 00:00:28 +0100491
492/* Tiny I/O callback called on recv/send I/O events on idle connections.
493 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
494 * is notified and can kill the connection.
495 */
496static void si_idle_conn_null_cb(struct connection *conn)
497{
498 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
499 return;
500
Willy Tarreau46be2e52014-01-20 12:10:52 +0100501 if (fdtab[conn->t.sock.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) {
502 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreau27375622013-12-17 00:00:28 +0100503 conn->flags |= CO_FL_SOCK_RD_SH;
Willy Tarreau46be2e52014-01-20 12:10:52 +0100504 }
505 else {
506 conn_drain(conn);
507 }
Willy Tarreau27375622013-12-17 00:00:28 +0100508
509 /* disable draining if we were called and have no drain function */
510 if (!conn->ctrl->drain)
511 __conn_data_stop_recv(conn);
512}
513
514/* Callback to be used by connection I/O handlers when some activity is detected
515 * on an idle server connection. Its main purpose is to kill the connection once
516 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
517 * it killed the connection.
518 */
519static int si_idle_conn_wake_cb(struct connection *conn)
520{
521 struct stream_interface *si = conn->owner;
522
523 if (!conn_ctrl_ready(conn))
524 return 0;
525
526 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
527 /* warning, we can't do anything on <conn> after this call ! */
528 conn_force_close(conn);
529 conn_free(conn);
530 si->end = NULL;
531 return -1;
532 }
533 return 0;
534}
535
Willy Tarreau100c4672012-08-20 12:06:26 +0200536/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200537 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200538 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200539 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200540 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200541 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200542static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200543{
Willy Tarreaue603e692012-09-27 22:20:41 +0200544 struct stream_interface *si = conn->owner;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200545
546 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
547 __FUNCTION__,
548 si, si->state, si->ib->flags, si->ob->flags);
549
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200550 if (conn->flags & CO_FL_ERROR)
551 si->flags |= SI_FL_ERR;
552
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200553 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200554 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200555 si->exp = TICK_ETERNITY;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200556 si->ob->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200557 }
558
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200559 /* process consumer side */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200560 if (channel_is_empty(si->ob)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100561 if (((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200562 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200563 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200564 __conn_data_stop_send(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200565 si->ob->wex = TICK_ETERNITY;
566 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200567
Willy Tarreaub31c9712012-11-11 23:05:39 +0100568 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200569 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200570
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200571 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200572 /* update timeouts if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200573 if ((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200574 !channel_is_empty(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200575 if (tick_isset(si->ob->wex))
576 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200577
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200578 if (!(si->flags & SI_FL_INDEP_STR))
579 if (tick_isset(si->ib->rex))
580 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200581
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200582 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200583 !channel_full(si->ob) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200584 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
585 si_chk_rcv(si->ob->prod);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200586 }
587
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200588 /* process producer side.
589 * We might have some data the consumer is waiting for.
590 * We can do fast-forwarding, but we avoid doing this for partial
591 * buffers, because it is very likely that it will be done again
592 * immediately afterwards once the following data is parsed (eg:
593 * HTTP chunking).
594 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200595 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200596 (si->ib->pipe /* always try to send spliced data */ ||
Willy Tarreau9b28e032012-10-12 23:49:43 +0200597 (si->ib->buf->i == 0 && (si->ib->cons->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200598 int last_len = si->ib->pipe ? si->ib->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200599
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200600 si_chk_snd(si->ib->cons);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200601
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200602 /* check if the consumer has freed some space either in the
603 * buffer or in the pipe.
604 */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200605 if (!channel_full(si->ib) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200606 (!last_len || !si->ib->pipe || si->ib->pipe->data < last_len))
607 si->flags &= ~SI_FL_WAIT_ROOM;
608 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200609
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200610 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200611 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200612 si->ib->rex = TICK_ETERNITY;
613 }
Willy Tarreau66572762012-12-19 17:34:17 +0100614 else if ((si->ib->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200615 !channel_full(si->ib)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100616 /* we must re-enable reading if si_chk_snd() has freed some space */
617 __conn_data_want_recv(conn);
Willy Tarreau66572762012-12-19 17:34:17 +0100618 if (!(si->ib->flags & CF_READ_NOEXP) && tick_isset(si->ib->rex))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200619 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200620 }
621
622 /* wake the task up only when needed */
623 if (/* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200624 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200625 si->state != SI_ST_EST ||
626 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200627 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreaufd31e532012-07-23 18:24:25 +0200628 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
629
630 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200631 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
632 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreaud7ad9f52013-12-31 17:26:25 +0100633 (si->ob->flags & (CF_SHUTW|CF_WAKE_WRITE)))) {
Willy Tarreaufd31e532012-07-23 18:24:25 +0200634 task_wakeup(si->owner, TASK_WOKEN_IO);
635 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200636 if (si->ib->flags & CF_READ_ACTIVITY)
637 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200638 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200639}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200640
Willy Tarreau5368d802012-08-21 18:22:06 +0200641/*
642 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800643 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800644 * caller to commit polling changes. The caller should check conn->flags
645 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200646 */
Godbach4f489902013-12-04 17:24:06 +0800647static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200648{
Willy Tarreaue603e692012-09-27 22:20:41 +0200649 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200650 struct channel *chn = si->ob;
Willy Tarreau5368d802012-08-21 18:22:06 +0200651 int ret;
652
Willy Tarreaucb76e592012-10-12 23:56:57 +0200653 if (chn->pipe && conn->xprt->snd_pipe) {
654 ret = conn->xprt->snd_pipe(conn, chn->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200655 if (ret > 0)
Willy Tarreaucb76e592012-10-12 23:56:57 +0200656 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200657
Willy Tarreaucb76e592012-10-12 23:56:57 +0200658 if (!chn->pipe->data) {
659 put_pipe(chn->pipe);
660 chn->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200661 }
662
Willy Tarreau96199b12012-08-24 00:46:52 +0200663 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800664 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200665 }
666
667 /* At this point, the pipe is empty, but we may still have data pending
668 * in the normal buffer.
669 */
Willy Tarreau9b28e032012-10-12 23:49:43 +0200670 if (!chn->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800671 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200672
Godbache68e02d2013-10-11 15:48:29 +0800673 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200674 * data left, and that there are sendable buffered data.
675 */
Godbache68e02d2013-10-11 15:48:29 +0800676 if (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH | CO_FL_WAIT_DATA | CO_FL_WAIT_WR | CO_FL_HANDSHAKE))) {
Willy Tarreau5368d802012-08-21 18:22:06 +0200677 /* check if we want to inform the kernel that we're interested in
678 * sending more data after this call. We want this if :
679 * - we're about to close after this last send and want to merge
680 * the ongoing FIN with the last segment.
681 * - we know we can't send everything at once and must get back
682 * here because of unaligned data
683 * - there is still a finite amount of data to forward
684 * The test is arranged so that the most common case does only 2
685 * tests.
686 */
687 unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
688
Willy Tarreaucb76e592012-10-12 23:56:57 +0200689 if ((!(chn->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
690 ((chn->to_forward && chn->to_forward != CHN_INFINITE_FORWARD) ||
691 (chn->flags & CF_EXPECT_MORE))) ||
Willy Tarreaub31c9712012-11-11 23:05:39 +0100692 ((chn->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau5368d802012-08-21 18:22:06 +0200693 send_flag |= MSG_MORE;
694
Willy Tarreau9b28e032012-10-12 23:49:43 +0200695 ret = conn->xprt->snd_buf(conn, chn->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800696 if (ret > 0) {
697 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200698
Godbache68e02d2013-10-11 15:48:29 +0800699 if (!chn->buf->o) {
700 /* Always clear both flags once everything has been sent, they're one-shot */
701 chn->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
702 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200703
Godbache68e02d2013-10-11 15:48:29 +0800704 /* if some data remain in the buffer, it's only because the
705 * system buffers are full, we will try next time.
706 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200707 }
Godbache68e02d2013-10-11 15:48:29 +0800708 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200709
Godbach4f489902013-12-04 17:24:06 +0800710 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200711}
712
713
Willy Tarreau100c4672012-08-20 12:06:26 +0200714/* Updates the timers and flags of a stream interface attached to a connection,
715 * depending on the buffers' flags. It should only be called once after the
716 * buffer flags have settled down, and before they are cleared. It doesn't
717 * harm to call it as often as desired (it just slightly hurts performance).
718 * It is only meant to be called by upper layers after buffer flags have been
719 * manipulated by analysers.
720 */
721void stream_int_update_conn(struct stream_interface *si)
722{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200723 struct channel *ib = si->ib;
724 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200725 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200726
Willy Tarreau100c4672012-08-20 12:06:26 +0200727 /* Check if we need to close the read side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200728 if (!(ib->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200729 /* Read not closed, update FD status and timeout for reads */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100730 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200731 /* stop reading */
732 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100733 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200734 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200735 conn_data_stop_recv(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200736 ib->rex = TICK_ETERNITY;
737 }
738 }
739 else {
740 /* (re)start reading and update timeout. Note: we don't recompute the timeout
741 * everytime we get here, otherwise it would risk never to expire. We only
742 * update it if is was not yet set. The stream socket handler will already
743 * have updated it if there has been a completed I/O.
744 */
745 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200746 conn_data_want_recv(conn);
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200747 if (!(ib->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ib->rex))
Willy Tarreau100c4672012-08-20 12:06:26 +0200748 ib->rex = tick_add_ifset(now_ms, ib->rto);
749 }
750 }
751
752 /* Check if we need to close the write side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200753 if (!(ob->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200754 /* Write not closed, update FD status and timeout for writes */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200755 if (channel_is_empty(ob)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200756 /* stop writing */
757 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100758 if ((ob->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200759 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200760 conn_data_stop_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200761 ob->wex = TICK_ETERNITY;
762 }
763 }
764 else {
765 /* (re)start writing and update timeout. Note: we don't recompute the timeout
766 * everytime we get here, otherwise it would risk never to expire. We only
767 * update it if is was not yet set. The stream socket handler will already
768 * have updated it if there has been a completed I/O.
769 */
770 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200771 conn_data_want_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200772 if (!tick_isset(ob->wex)) {
773 ob->wex = tick_add_ifset(now_ms, ob->wto);
774 if (tick_isset(ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
775 /* Note: depending on the protocol, we don't know if we're waiting
776 * for incoming data or not. So in order to prevent the socket from
777 * expiring read timeouts during writes, we refresh the read timeout,
778 * except if it was already infinite or if we have explicitly setup
779 * independent streams.
780 */
781 ib->rex = tick_add_ifset(now_ms, ib->rto);
782 }
783 }
784 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200785 }
786}
787
788/*
789 * This function performs a shutdown-read on a stream interface attached to
790 * a connection in a connected or init state (it does nothing for other
791 * states). It either shuts the read side or marks itself as closed. The buffer
792 * flags are updated to reflect the new state. If the stream interface has
793 * SI_FL_NOHALF, we also forward the close to the write side. If a control
794 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200795 * descriptors are then shutdown or closed accordingly. The function
796 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200797 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200798static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200799{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200800 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200801
802 si->ib->flags &= ~CF_SHUTR_NOW;
803 if (si->ib->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200804 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200805 si->ib->flags |= CF_SHUTR;
806 si->ib->rex = TICK_ETERNITY;
807 si->flags &= ~SI_FL_WAIT_ROOM;
808
809 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200810 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200811
812 if (si->ob->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200813 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200814 si->state = SI_ST_DIS;
815 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200816 }
817 else if (si->flags & SI_FL_NOHALF) {
818 /* we want to immediately forward this close to the write side */
819 return stream_int_shutw_conn(si);
820 }
821 else if (conn->ctrl) {
822 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200823 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200824 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200825}
826
827/*
828 * This function performs a shutdown-write on a stream interface attached to
829 * a connection in a connected or init state (it does nothing for other
830 * states). It either shuts the write side or marks itself as closed. The
831 * buffer flags are updated to reflect the new state. It does also close
832 * everything if the SI was marked as being in error state. If there is a
833 * data-layer shutdown, it is called. If a control layer is defined, then it is
834 * supposed to be a socket layer and file descriptors are then shutdown or
Willy Tarreau6fe15412013-09-29 15:16:03 +0200835 * closed accordingly. The function automatically disables polling if needed.
836 * Note: at the moment, we continue to check conn->ctrl eventhough we *know* it
837 * is valid. This will help selecting the proper shutdown() and setsockopt()
838 * calls if/when we implement remote sockets later.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200839 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200840static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200841{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200842 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200843
844 si->ob->flags &= ~CF_SHUTW_NOW;
845 if (si->ob->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200846 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200847 si->ob->flags |= CF_SHUTW;
848 si->ob->wex = TICK_ETERNITY;
849 si->flags &= ~SI_FL_WAIT_DATA;
850
851 switch (si->state) {
852 case SI_ST_EST:
853 /* we have to shut before closing, otherwise some short messages
854 * may never leave the system, especially when there are remaining
855 * unread data in the socket input buffer, or when nolinger is set.
856 * However, if SI_FL_NOLINGER is explicitly set, we know there is
857 * no risk so we close both sides immediately.
858 */
859 if (si->flags & SI_FL_ERR) {
860 /* quick close, the socket is alredy shut anyway */
861 }
862 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200863 /* unclean data-layer shutdown */
864 if (conn->xprt && conn->xprt->shutw)
865 conn->xprt->shutw(conn, 0);
866 }
867 else {
868 /* clean data-layer shutdown */
869 if (conn->xprt && conn->xprt->shutw)
870 conn->xprt->shutw(conn, 1);
871
872 /* If the stream interface is configured to disable half-open
873 * connections, we'll skip the shutdown(), but only if the
874 * read size is already closed. Otherwise we can't support
875 * closed write with pending read (eg: abortonclose while
876 * waiting for the server).
877 */
878 if (!(si->flags & SI_FL_NOHALF) || !(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
879 /* We shutdown transport layer */
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100880 if (conn_ctrl_ready(conn))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200881 shutdown(conn->t.sock.fd, SHUT_WR);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200882
883 if (!(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
884 /* OK just a shutw, but we want the caller
885 * to disable polling on this FD if exists.
886 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200887 if (conn->ctrl)
888 conn_data_stop_send(conn);
889 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200890 }
891 }
892 }
893
894 /* fall through */
895 case SI_ST_CON:
896 /* we may have to close a pending connection, and mark the
897 * response buffer as shutr
898 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200899 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200900 /* fall through */
901 case SI_ST_CER:
902 case SI_ST_QUE:
903 case SI_ST_TAR:
904 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200905 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200906 default:
907 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
908 si->ib->flags &= ~CF_SHUTR_NOW;
909 si->ib->flags |= CF_SHUTR;
910 si->ib->rex = TICK_ETERNITY;
911 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200912 }
913}
914
Willy Tarreau46a8d922012-08-20 12:38:36 +0200915/* This function is used for inter-stream-interface calls. It is called by the
916 * consumer to inform the producer side that it may be interested in checking
917 * for free space in the buffer. Note that it intentionally does not update
918 * timeouts, so that we can still check them later at wake-up. This function is
919 * dedicated to connection-based stream interfaces.
920 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200921static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200922{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200923 struct channel *ib = si->ib;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200924 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200925
Willy Tarreau34ffd772012-09-03 16:51:27 +0200926 if (unlikely(si->state > SI_ST_EST || (ib->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200927 return;
928
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200929 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100930
Willy Tarreaub31c9712012-11-11 23:05:39 +0100931 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200932 /* stop reading */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100933 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200934 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200935 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200936 }
937 else {
938 /* (re)start reading */
939 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200940 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200941 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200942 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200943}
944
945
Willy Tarreaude5722c2012-08-20 15:01:10 +0200946/* This function is used for inter-stream-interface calls. It is called by the
947 * producer to inform the consumer side that it may be interested in checking
948 * for data in the buffer. Note that it intentionally does not update timeouts,
949 * so that we can still check them later at wake-up.
950 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200951static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200952{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200953 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200954 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200955
Willy Tarreau34ffd772012-09-03 16:51:27 +0200956 if (unlikely(si->state > SI_ST_EST || (ob->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200957 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200958
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200959 if (unlikely(channel_is_empty(ob))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200960 return;
961
962 if (!ob->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100963 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200964 return;
965
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200966 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200967 /* already subscribed to write notifications, will be called
968 * anyway, so let's avoid calling it especially if the reader
969 * is not ready.
970 */
971 return;
972 }
973
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200974 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100975 /* Before calling the data-level operations, we have to prepare
976 * the polling flags to ensure we properly detect changes.
Willy Tarreaude5722c2012-08-20 15:01:10 +0200977 */
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100978 if (conn_ctrl_ready(conn))
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200979 fd_want_send(conn->t.sock.fd);
Willy Tarreau7d281492012-12-16 19:19:13 +0100980
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200981 conn_refresh_polling_flags(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100982
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200983 si_conn_send(conn);
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100984 if (conn_ctrl_ready(conn) && (conn->flags & CO_FL_ERROR)) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100985 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200986 fd_stop_both(conn->t.sock.fd);
987 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100988 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100989 goto out_wakeup;
990 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200991 }
992
993 /* OK, so now we know that some data might have been sent, and that we may
994 * have to poll first. We have to do that too if the buffer is not empty.
995 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200996 if (channel_is_empty(ob)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200997 /* the connection is established but we can't write. Either the
998 * buffer is empty, or we just refrain from sending because the
999 * ->o limit was reached. Maybe we just wrote the last
1000 * chunk and need to close.
1001 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001002 __conn_data_stop_send(conn);
Willy Tarreaub31c9712012-11-11 23:05:39 +01001003 if (((ob->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001004 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +02001005 (si->state == SI_ST_EST)) {
1006 si_shutw(si);
1007 goto out_wakeup;
1008 }
1009
Willy Tarreaub31c9712012-11-11 23:05:39 +01001010 if ((ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001011 si->flags |= SI_FL_WAIT_DATA;
1012 ob->wex = TICK_ETERNITY;
1013 }
1014 else {
1015 /* Otherwise there are remaining data to be sent in the buffer,
1016 * which means we have to poll before doing so.
1017 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001018 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001019 si->flags &= ~SI_FL_WAIT_DATA;
1020 if (!tick_isset(ob->wex))
1021 ob->wex = tick_add_ifset(now_ms, ob->wto);
1022 }
1023
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001024 if (likely(ob->flags & CF_WRITE_ACTIVITY)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001025 /* update timeout if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001026 if ((ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +02001027 !channel_is_empty(ob))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001028 ob->wex = tick_add_ifset(now_ms, ob->wto);
1029
1030 if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
1031 /* Note: to prevent the client from expiring read timeouts
1032 * during writes, we refresh it. We only do this if the
1033 * interface is not configured for "independent streams",
1034 * because for some applications it's better not to do this,
1035 * for instance when continuously exchanging small amounts
1036 * of data which can full the socket buffers long before a
1037 * write timeout is detected.
1038 */
1039 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
1040 }
1041 }
1042
1043 /* in case of special condition (error, shutdown, end of write...), we
1044 * have to notify the task.
1045 */
Willy Tarreaud7ad9f52013-12-31 17:26:25 +01001046 if (likely(ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW|CF_WAKE_WRITE))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001047 out_wakeup:
1048 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
1049 task_wakeup(si->owner, TASK_WOKEN_IO);
1050 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001051
1052 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001053 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001054}
1055
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001056/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001057 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001058 * into the buffer from the connection. It iterates over the transport layer's
1059 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001060 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001061static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001062{
Willy Tarreaue603e692012-09-27 22:20:41 +02001063 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001064 struct channel *chn = si->ib;
Willy Tarreauce323de2012-08-20 21:41:06 +02001065 int ret, max, cur_read;
1066 int read_poll = MAX_READ_POLL_LOOPS;
1067
1068 /* stop immediately on errors. Note that we DON'T want to stop on
1069 * POLL_ERR, as the poller might report a write error while there
1070 * are still data available in the recv buffer. This typically
1071 * happens when we send too large a request to a backend server
1072 * which rejects it before reading it all.
1073 */
1074 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001075 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001076
1077 /* stop here if we reached the end of data */
1078 if (conn_data_read0_pending(conn))
1079 goto out_shutdown_r;
1080
1081 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001082 if (chn->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001083 return;
1084
Willy Tarreau96199b12012-08-24 00:46:52 +02001085 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001086
1087 /* First, let's see if we may splice data across the channel without
1088 * using a buffer.
1089 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001090 if (conn->xprt->rcv_pipe &&
Willy Tarreaufa8e2bc2013-07-18 22:21:54 +02001091 (chn->pipe || chn->to_forward >= MIN_SPLICE_FORWARD) &&
1092 chn->flags & CF_KERN_SPLICING) {
Willy Tarreau9b28e032012-10-12 23:49:43 +02001093 if (buffer_not_empty(chn->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001094 /* We're embarrassed, there are already data pending in
1095 * the buffer and we don't want to have them at two
1096 * locations at a time. Let's indicate we need some
1097 * place and ask the consumer to hurry.
1098 */
1099 goto abort_splice;
1100 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001101
Willy Tarreaucb76e592012-10-12 23:56:57 +02001102 if (unlikely(chn->pipe == NULL)) {
1103 if (pipes_used >= global.maxpipes || !(chn->pipe = get_pipe())) {
1104 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001105 goto abort_splice;
1106 }
1107 }
1108
Willy Tarreaucb76e592012-10-12 23:56:57 +02001109 ret = conn->xprt->rcv_pipe(conn, chn->pipe, chn->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001110 if (ret < 0) {
1111 /* splice not supported on this end, let's disable it */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001112 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001113 goto abort_splice;
1114 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001115
Willy Tarreau96199b12012-08-24 00:46:52 +02001116 if (ret > 0) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001117 if (chn->to_forward != CHN_INFINITE_FORWARD)
1118 chn->to_forward -= ret;
1119 chn->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001120 cur_read += ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001121 chn->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001122 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001123
1124 if (conn_data_read0_pending(conn))
1125 goto out_shutdown_r;
1126
1127 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001128 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001129
Willy Tarreau61d39a02013-07-18 21:49:32 +02001130 if (conn->flags & CO_FL_WAIT_ROOM) {
1131 /* the pipe is full or we have read enough data that it
1132 * could soon be full. Let's stop before needing to poll.
1133 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001134 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001135 __conn_data_stop_recv(conn);
1136 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001137
Willy Tarreauce323de2012-08-20 21:41:06 +02001138 /* splice not possible (anymore), let's go on on standard copy */
1139 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001140
1141 abort_splice:
Willy Tarreau61d39a02013-07-18 21:49:32 +02001142 if (chn->pipe && unlikely(!chn->pipe->data)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001143 put_pipe(chn->pipe);
1144 chn->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001145 }
1146
Willy Tarreau61d39a02013-07-18 21:49:32 +02001147 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1148 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1149 * that if such an event is not handled above in splice, it will be handled here by
1150 * recv().
1151 */
1152 while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_RD | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001153 max = bi_avail(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001154
1155 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001156 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001157 break;
1158 }
1159
Willy Tarreau9b28e032012-10-12 23:49:43 +02001160 ret = conn->xprt->rcv_buf(conn, chn->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001161 if (ret <= 0)
1162 break;
1163
1164 cur_read += ret;
1165
1166 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001167 if (chn->to_forward && !(chn->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001168 unsigned long fwd = ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001169 if (chn->to_forward != CHN_INFINITE_FORWARD) {
1170 if (fwd > chn->to_forward)
1171 fwd = chn->to_forward;
1172 chn->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001173 }
Willy Tarreau9b28e032012-10-12 23:49:43 +02001174 b_adv(chn->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001175 }
1176
Willy Tarreaucb76e592012-10-12 23:56:57 +02001177 chn->flags |= CF_READ_PARTIAL;
1178 chn->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001179
Willy Tarreaucb76e592012-10-12 23:56:57 +02001180 if (channel_full(chn)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001181 /* The buffer is now full, there's no point in going through
1182 * the loop again.
1183 */
Willy Tarreau9b28e032012-10-12 23:49:43 +02001184 if (!(chn->flags & CF_STREAMER_FAST) && (cur_read == buffer_len(chn->buf))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001185 chn->xfer_small = 0;
1186 chn->xfer_large++;
1187 if (chn->xfer_large >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001188 /* we call this buffer a fast streamer if it manages
1189 * to be filled in one call 3 consecutive times.
1190 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001191 chn->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001192 //fputc('+', stderr);
1193 }
1194 }
Willy Tarreaucb76e592012-10-12 23:56:57 +02001195 else if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001196 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001197 chn->xfer_large = 0;
1198 chn->xfer_small++;
1199 if (chn->xfer_small >= 2) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001200 /* if the buffer has been at least half full twice,
1201 * we receive faster than we send, so at least it
1202 * is not a "fast streamer".
1203 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001204 chn->flags &= ~CF_STREAMER_FAST;
Willy Tarreauce323de2012-08-20 21:41:06 +02001205 //fputc('-', stderr);
1206 }
1207 }
1208 else {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001209 chn->xfer_small = 0;
1210 chn->xfer_large = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001211 }
1212
Willy Tarreauce323de2012-08-20 21:41:06 +02001213 si->flags |= SI_FL_WAIT_ROOM;
1214 break;
1215 }
1216
Willy Tarreau5fddab02012-11-09 18:27:26 +01001217 if ((chn->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001218 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001219 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001220 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001221 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001222
1223 /* if too many bytes were missing from last read, it means that
1224 * it's pointless trying to read again because the system does
1225 * not have them in buffers.
1226 */
1227 if (ret < max) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001228 if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001229 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001230 chn->xfer_large = 0;
1231 chn->xfer_small++;
1232 if (chn->xfer_small >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001233 /* we have read less than half of the buffer in
1234 * one pass, and this happened at least 3 times.
1235 * This is definitely not a streamer.
1236 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001237 chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001238 //fputc('!', stderr);
1239 }
1240 }
1241
1242 /* if a streamer has read few data, it may be because we
1243 * have exhausted system buffers. It's not worth trying
1244 * again.
1245 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001246 if (chn->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001247 break;
1248
1249 /* if we read a large block smaller than what we requested,
1250 * it's almost certain we'll never get anything more.
1251 */
1252 if (ret >= global.tune.recv_enough)
1253 break;
1254 }
1255 } /* while !flags */
1256
Willy Tarreau96199b12012-08-24 00:46:52 +02001257 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001258 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001259
Willy Tarreauce323de2012-08-20 21:41:06 +02001260 if (conn_data_read0_pending(conn))
1261 /* connection closed */
1262 goto out_shutdown_r;
1263
1264 return;
1265
1266 out_shutdown_r:
1267 /* we received a shutdown */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001268 chn->flags |= CF_READ_NULL;
1269 if (chn->flags & CF_AUTO_CLOSE)
1270 channel_shutw_now(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001271 stream_sock_read0(si);
1272 conn_data_read0(conn);
1273 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001274}
1275
1276/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001277 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001278 * from the buffer to the connection. It iterates over the transport layer's
1279 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001280 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001281static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001282{
Willy Tarreaue603e692012-09-27 22:20:41 +02001283 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001284 struct channel *chn = si->ob;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001285
1286 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001287 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001288
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001289 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001290 /* a handshake was requested */
1291 return;
1292
1293 /* we might have been called just after an asynchronous shutw */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001294 if (chn->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001295 return;
1296
1297 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001298 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001299
1300 /* OK all done */
1301 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001302}
1303
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001304/*
1305 * This function propagates a null read received on a socket-based connection.
1306 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
1307 * the close is also forwarded to the write side as an abort. This function is
1308 * still socket-specific as it handles a setsockopt() call to set the SO_LINGER
1309 * state on the socket.
1310 */
1311void stream_sock_read0(struct stream_interface *si)
1312{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001313 struct connection *conn = __objt_conn(si->end);
1314
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001315 si->ib->flags &= ~CF_SHUTR_NOW;
1316 if (si->ib->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001317 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001318 si->ib->flags |= CF_SHUTR;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001319 si->ib->rex = TICK_ETERNITY;
1320 si->flags &= ~SI_FL_WAIT_ROOM;
1321
1322 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1323 return;
1324
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001325 if (si->ob->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001326 goto do_close;
1327
1328 if (si->flags & SI_FL_NOHALF) {
1329 /* we want to immediately forward this close to the write side */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001330 /* force flag on ssl to keep session in cache */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001331 if (conn->xprt->shutw)
1332 conn->xprt->shutw(conn, 0);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001333 goto do_close;
1334 }
1335
1336 /* otherwise that's just a normal read shutdown */
Willy Tarreauad38ace2013-12-15 14:19:38 +01001337 if (conn_ctrl_ready(conn))
1338 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001339 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001340 return;
1341
1342 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001343 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001344 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001345
1346 si->ib->flags &= ~CF_SHUTR_NOW;
1347 si->ib->flags |= CF_SHUTR;
1348 si->ib->rex = TICK_ETERNITY;
1349
1350 si->ob->flags &= ~CF_SHUTW_NOW;
1351 si->ob->flags |= CF_SHUTW;
1352 si->ob->wex = TICK_ETERNITY;
1353
1354 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1355
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001356 si->state = SI_ST_DIS;
1357 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001358 return;
1359}
1360
Willy Tarreaudded32d2008-11-30 19:48:07 +01001361/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001362 * Local variables:
1363 * c-indent-level: 8
1364 * c-basic-offset: 8
1365 * End:
1366 */