blob: 6d15edc9cbefb9f6dfac6c5ae7d960e1f87dd334 [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);
369 if (!si)
370 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 Tarreau2c6be842012-07-06 17:12:34 +0200403 /* If we have a PROXY line to send, we'll use this to validate the
404 * connection, in which case the connection is validated only once
405 * we've sent the whole proxy line. Otherwise we use connect().
406 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200407 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200408 int ret;
409
410 /* The target server expects a PROXY line to be sent first.
411 * If the send_proxy_ofs is negative, it corresponds to the
412 * offset to start sending from then end of the proxy string
413 * (which is recomputed every time since it's constant). If
414 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200415 * We can only send a "normal" PROXY line when the connection
416 * is attached to a stream interface. Otherwise we can only
417 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200418 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200419 if (conn->data == &si_conn_cb) {
420 struct stream_interface *si = conn->owner;
421 struct connection *remote = objt_conn(si->ob->prod->end);
422
423 if (remote)
424 ret = make_proxy_line(trash.str, trash.size, &remote->addr.from, &remote->addr.to);
425 else
426 ret = make_proxy_line(trash.str, trash.size, NULL, NULL);
427 }
428 else {
429 /* The target server expects a LOCAL line to be sent first. Retrieving
430 * local or remote addresses may fail until the connection is established.
431 */
432 conn_get_from_addr(conn);
433 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
434 goto out_wait;
435
436 conn_get_to_addr(conn);
437 if (!(conn->flags & CO_FL_ADDR_TO_SET))
438 goto out_wait;
439
440 ret = make_proxy_line(trash.str, trash.size, &conn->addr.from, &conn->addr.to);
441 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200442
Willy Tarreau2c6be842012-07-06 17:12:34 +0200443 if (!ret)
444 goto out_error;
445
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200446 if (conn->send_proxy_ofs > 0)
447 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200448
Willy Tarreaua1a74742012-08-24 12:14:49 +0200449 /* we have to send trash from (ret+sp for -sp bytes). If the
450 * data layer has a pending write, we'll also set MSG_MORE.
451 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200452 ret = send(conn->t.sock.fd, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
Willy Tarreaua1a74742012-08-24 12:14:49 +0200453 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200454
455 if (ret == 0)
456 goto out_wait;
457
458 if (ret < 0) {
Willy Tarreau95742a42013-09-03 09:02:11 +0200459 if (errno == EAGAIN || errno == ENOTCONN)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200460 goto out_wait;
Willy Tarreau7fe45692013-12-04 23:37:56 +0100461 if (errno == EINTR)
462 continue;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100463 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200464 goto out_error;
465 }
466
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200467 conn->send_proxy_ofs += ret; /* becomes zero once complete */
468 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200469 goto out_wait;
470
471 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100472 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200473 }
474
Willy Tarreaua1a74742012-08-24 12:14:49 +0200475 /* The connection is ready now, simply return and let the connection
476 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200477 */
478 if (conn->flags & CO_FL_WAIT_L4_CONN)
479 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200480 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200481 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200482
483 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200484 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200485 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200486 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200487
488 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200489 __conn_sock_stop_recv(conn);
490 __conn_sock_poll_send(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200491 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200492}
493
Willy Tarreau27375622013-12-17 00:00:28 +0100494
495/* Tiny I/O callback called on recv/send I/O events on idle connections.
496 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
497 * is notified and can kill the connection.
498 */
499static void si_idle_conn_null_cb(struct connection *conn)
500{
501 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
502 return;
503
Willy Tarreau46be2e52014-01-20 12:10:52 +0100504 if (fdtab[conn->t.sock.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) {
505 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreau27375622013-12-17 00:00:28 +0100506 conn->flags |= CO_FL_SOCK_RD_SH;
Willy Tarreau46be2e52014-01-20 12:10:52 +0100507 }
508 else {
509 conn_drain(conn);
510 }
Willy Tarreau27375622013-12-17 00:00:28 +0100511
512 /* disable draining if we were called and have no drain function */
513 if (!conn->ctrl->drain)
514 __conn_data_stop_recv(conn);
515}
516
517/* Callback to be used by connection I/O handlers when some activity is detected
518 * on an idle server connection. Its main purpose is to kill the connection once
519 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
520 * it killed the connection.
521 */
522static int si_idle_conn_wake_cb(struct connection *conn)
523{
524 struct stream_interface *si = conn->owner;
525
526 if (!conn_ctrl_ready(conn))
527 return 0;
528
529 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
530 /* warning, we can't do anything on <conn> after this call ! */
531 conn_force_close(conn);
532 conn_free(conn);
533 si->end = NULL;
534 return -1;
535 }
536 return 0;
537}
538
Willy Tarreau100c4672012-08-20 12:06:26 +0200539/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200540 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200541 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200542 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200543 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200544 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200545static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200546{
Willy Tarreaue603e692012-09-27 22:20:41 +0200547 struct stream_interface *si = conn->owner;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200548
549 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
550 __FUNCTION__,
551 si, si->state, si->ib->flags, si->ob->flags);
552
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200553 if (conn->flags & CO_FL_ERROR)
554 si->flags |= SI_FL_ERR;
555
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200556 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200557 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200558 si->exp = TICK_ETERNITY;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200559 si->ob->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200560 }
561
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200562 /* process consumer side */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200563 if (channel_is_empty(si->ob)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100564 if (((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200565 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200566 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200567 __conn_data_stop_send(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200568 si->ob->wex = TICK_ETERNITY;
569 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200570
Willy Tarreaub31c9712012-11-11 23:05:39 +0100571 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200572 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200573
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200574 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200575 /* update timeouts if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200576 if ((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200577 !channel_is_empty(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200578 if (tick_isset(si->ob->wex))
579 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200580
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200581 if (!(si->flags & SI_FL_INDEP_STR))
582 if (tick_isset(si->ib->rex))
583 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200584
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200585 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200586 !channel_full(si->ob) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200587 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
588 si_chk_rcv(si->ob->prod);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200589 }
590
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200591 /* process producer side.
592 * We might have some data the consumer is waiting for.
593 * We can do fast-forwarding, but we avoid doing this for partial
594 * buffers, because it is very likely that it will be done again
595 * immediately afterwards once the following data is parsed (eg:
596 * HTTP chunking).
597 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200598 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200599 (si->ib->pipe /* always try to send spliced data */ ||
Willy Tarreau9b28e032012-10-12 23:49:43 +0200600 (si->ib->buf->i == 0 && (si->ib->cons->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200601 int last_len = si->ib->pipe ? si->ib->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200602
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200603 si_chk_snd(si->ib->cons);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200604
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200605 /* check if the consumer has freed some space either in the
606 * buffer or in the pipe.
607 */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200608 if (!channel_full(si->ib) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200609 (!last_len || !si->ib->pipe || si->ib->pipe->data < last_len))
610 si->flags &= ~SI_FL_WAIT_ROOM;
611 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200612
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200613 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200614 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200615 si->ib->rex = TICK_ETERNITY;
616 }
Willy Tarreau66572762012-12-19 17:34:17 +0100617 else if ((si->ib->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200618 !channel_full(si->ib)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100619 /* we must re-enable reading if si_chk_snd() has freed some space */
620 __conn_data_want_recv(conn);
Willy Tarreau66572762012-12-19 17:34:17 +0100621 if (!(si->ib->flags & CF_READ_NOEXP) && tick_isset(si->ib->rex))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200622 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200623 }
624
625 /* wake the task up only when needed */
626 if (/* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200627 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200628 si->state != SI_ST_EST ||
629 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200630 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreaufd31e532012-07-23 18:24:25 +0200631 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
632
633 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200634 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
635 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreaue6300be2014-01-25 02:33:21 +0100636 ((si->ob->flags & CF_SHUTW) ||
637 ((si->ob->flags & CF_WAKE_WRITE) &&
638 (si->ob->prod->state != SI_ST_EST ||
639 (channel_is_empty(si->ob) && !si->ob->to_forward)))))) {
Willy Tarreaufd31e532012-07-23 18:24:25 +0200640 task_wakeup(si->owner, TASK_WOKEN_IO);
641 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200642 if (si->ib->flags & CF_READ_ACTIVITY)
643 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200644 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200645}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200646
Willy Tarreau5368d802012-08-21 18:22:06 +0200647/*
648 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800649 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800650 * caller to commit polling changes. The caller should check conn->flags
651 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200652 */
Godbach4f489902013-12-04 17:24:06 +0800653static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200654{
Willy Tarreaue603e692012-09-27 22:20:41 +0200655 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200656 struct channel *chn = si->ob;
Willy Tarreau5368d802012-08-21 18:22:06 +0200657 int ret;
658
Willy Tarreaucb76e592012-10-12 23:56:57 +0200659 if (chn->pipe && conn->xprt->snd_pipe) {
660 ret = conn->xprt->snd_pipe(conn, chn->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200661 if (ret > 0)
Willy Tarreaucb76e592012-10-12 23:56:57 +0200662 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200663
Willy Tarreaucb76e592012-10-12 23:56:57 +0200664 if (!chn->pipe->data) {
665 put_pipe(chn->pipe);
666 chn->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200667 }
668
Willy Tarreau96199b12012-08-24 00:46:52 +0200669 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800670 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200671 }
672
673 /* At this point, the pipe is empty, but we may still have data pending
674 * in the normal buffer.
675 */
Willy Tarreau9b28e032012-10-12 23:49:43 +0200676 if (!chn->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800677 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200678
Godbache68e02d2013-10-11 15:48:29 +0800679 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200680 * data left, and that there are sendable buffered data.
681 */
Godbache68e02d2013-10-11 15:48:29 +0800682 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 +0200683 /* check if we want to inform the kernel that we're interested in
684 * sending more data after this call. We want this if :
685 * - we're about to close after this last send and want to merge
686 * the ongoing FIN with the last segment.
687 * - we know we can't send everything at once and must get back
688 * here because of unaligned data
689 * - there is still a finite amount of data to forward
690 * The test is arranged so that the most common case does only 2
691 * tests.
692 */
693 unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
694
Willy Tarreaucb76e592012-10-12 23:56:57 +0200695 if ((!(chn->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
696 ((chn->to_forward && chn->to_forward != CHN_INFINITE_FORWARD) ||
697 (chn->flags & CF_EXPECT_MORE))) ||
Willy Tarreaub31c9712012-11-11 23:05:39 +0100698 ((chn->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau5368d802012-08-21 18:22:06 +0200699 send_flag |= MSG_MORE;
700
Willy Tarreau9b28e032012-10-12 23:49:43 +0200701 ret = conn->xprt->snd_buf(conn, chn->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800702 if (ret > 0) {
703 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200704
Godbache68e02d2013-10-11 15:48:29 +0800705 if (!chn->buf->o) {
706 /* Always clear both flags once everything has been sent, they're one-shot */
707 chn->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
708 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200709
Godbache68e02d2013-10-11 15:48:29 +0800710 /* if some data remain in the buffer, it's only because the
711 * system buffers are full, we will try next time.
712 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200713 }
Godbache68e02d2013-10-11 15:48:29 +0800714 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200715
Godbach4f489902013-12-04 17:24:06 +0800716 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200717}
718
719
Willy Tarreau100c4672012-08-20 12:06:26 +0200720/* Updates the timers and flags of a stream interface attached to a connection,
721 * depending on the buffers' flags. It should only be called once after the
722 * buffer flags have settled down, and before they are cleared. It doesn't
723 * harm to call it as often as desired (it just slightly hurts performance).
724 * It is only meant to be called by upper layers after buffer flags have been
725 * manipulated by analysers.
726 */
727void stream_int_update_conn(struct stream_interface *si)
728{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200729 struct channel *ib = si->ib;
730 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200731 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200732
Willy Tarreau100c4672012-08-20 12:06:26 +0200733 /* Check if we need to close the read side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200734 if (!(ib->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200735 /* Read not closed, update FD status and timeout for reads */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100736 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200737 /* stop reading */
738 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100739 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200740 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200741 conn_data_stop_recv(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200742 ib->rex = TICK_ETERNITY;
743 }
744 }
745 else {
746 /* (re)start reading and update timeout. Note: we don't recompute the timeout
747 * everytime we get here, otherwise it would risk never to expire. We only
748 * update it if is was not yet set. The stream socket handler will already
749 * have updated it if there has been a completed I/O.
750 */
751 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200752 conn_data_want_recv(conn);
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200753 if (!(ib->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ib->rex))
Willy Tarreau100c4672012-08-20 12:06:26 +0200754 ib->rex = tick_add_ifset(now_ms, ib->rto);
755 }
756 }
757
758 /* Check if we need to close the write side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200759 if (!(ob->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200760 /* Write not closed, update FD status and timeout for writes */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200761 if (channel_is_empty(ob)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200762 /* stop writing */
763 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100764 if ((ob->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200765 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200766 conn_data_stop_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200767 ob->wex = TICK_ETERNITY;
768 }
769 }
770 else {
771 /* (re)start writing and update timeout. Note: we don't recompute the timeout
772 * everytime we get here, otherwise it would risk never to expire. We only
773 * update it if is was not yet set. The stream socket handler will already
774 * have updated it if there has been a completed I/O.
775 */
776 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200777 conn_data_want_send(conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200778 if (!tick_isset(ob->wex)) {
779 ob->wex = tick_add_ifset(now_ms, ob->wto);
780 if (tick_isset(ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
781 /* Note: depending on the protocol, we don't know if we're waiting
782 * for incoming data or not. So in order to prevent the socket from
783 * expiring read timeouts during writes, we refresh the read timeout,
784 * except if it was already infinite or if we have explicitly setup
785 * independent streams.
786 */
787 ib->rex = tick_add_ifset(now_ms, ib->rto);
788 }
789 }
790 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200791 }
792}
793
794/*
795 * This function performs a shutdown-read on a stream interface attached to
796 * a connection in a connected or init state (it does nothing for other
797 * states). It either shuts the read side or marks itself as closed. The buffer
798 * flags are updated to reflect the new state. If the stream interface has
799 * SI_FL_NOHALF, we also forward the close to the write side. If a control
800 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200801 * descriptors are then shutdown or closed accordingly. The function
802 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200803 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200804static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200805{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200806 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200807
808 si->ib->flags &= ~CF_SHUTR_NOW;
809 if (si->ib->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200810 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200811 si->ib->flags |= CF_SHUTR;
812 si->ib->rex = TICK_ETERNITY;
813 si->flags &= ~SI_FL_WAIT_ROOM;
814
815 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200816 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200817
818 if (si->ob->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200819 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200820 si->state = SI_ST_DIS;
821 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200822 }
823 else if (si->flags & SI_FL_NOHALF) {
824 /* we want to immediately forward this close to the write side */
825 return stream_int_shutw_conn(si);
826 }
827 else if (conn->ctrl) {
828 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200829 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200830 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200831}
832
833/*
834 * This function performs a shutdown-write on a stream interface attached to
835 * a connection in a connected or init state (it does nothing for other
836 * states). It either shuts the write side or marks itself as closed. The
837 * buffer flags are updated to reflect the new state. It does also close
838 * everything if the SI was marked as being in error state. If there is a
839 * data-layer shutdown, it is called. If a control layer is defined, then it is
840 * supposed to be a socket layer and file descriptors are then shutdown or
Willy Tarreau6fe15412013-09-29 15:16:03 +0200841 * closed accordingly. The function automatically disables polling if needed.
842 * Note: at the moment, we continue to check conn->ctrl eventhough we *know* it
843 * is valid. This will help selecting the proper shutdown() and setsockopt()
844 * calls if/when we implement remote sockets later.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200845 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200846static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200847{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200848 struct connection *conn = __objt_conn(si->end);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200849
850 si->ob->flags &= ~CF_SHUTW_NOW;
851 if (si->ob->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200852 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200853 si->ob->flags |= CF_SHUTW;
854 si->ob->wex = TICK_ETERNITY;
855 si->flags &= ~SI_FL_WAIT_DATA;
856
857 switch (si->state) {
858 case SI_ST_EST:
859 /* we have to shut before closing, otherwise some short messages
860 * may never leave the system, especially when there are remaining
861 * unread data in the socket input buffer, or when nolinger is set.
862 * However, if SI_FL_NOLINGER is explicitly set, we know there is
863 * no risk so we close both sides immediately.
864 */
865 if (si->flags & SI_FL_ERR) {
866 /* quick close, the socket is alredy shut anyway */
867 }
868 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200869 /* unclean data-layer shutdown */
870 if (conn->xprt && conn->xprt->shutw)
871 conn->xprt->shutw(conn, 0);
872 }
873 else {
874 /* clean data-layer shutdown */
875 if (conn->xprt && conn->xprt->shutw)
876 conn->xprt->shutw(conn, 1);
877
878 /* If the stream interface is configured to disable half-open
879 * connections, we'll skip the shutdown(), but only if the
880 * read size is already closed. Otherwise we can't support
881 * closed write with pending read (eg: abortonclose while
882 * waiting for the server).
883 */
884 if (!(si->flags & SI_FL_NOHALF) || !(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
885 /* We shutdown transport layer */
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100886 if (conn_ctrl_ready(conn))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200887 shutdown(conn->t.sock.fd, SHUT_WR);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200888
889 if (!(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
890 /* OK just a shutw, but we want the caller
891 * to disable polling on this FD if exists.
892 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200893 if (conn->ctrl)
894 conn_data_stop_send(conn);
895 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200896 }
897 }
898 }
899
900 /* fall through */
901 case SI_ST_CON:
902 /* we may have to close a pending connection, and mark the
903 * response buffer as shutr
904 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200905 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200906 /* fall through */
907 case SI_ST_CER:
908 case SI_ST_QUE:
909 case SI_ST_TAR:
910 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200911 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200912 default:
913 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
914 si->ib->flags &= ~CF_SHUTR_NOW;
915 si->ib->flags |= CF_SHUTR;
916 si->ib->rex = TICK_ETERNITY;
917 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200918 }
919}
920
Willy Tarreau46a8d922012-08-20 12:38:36 +0200921/* This function is used for inter-stream-interface calls. It is called by the
922 * consumer to inform the producer side that it may be interested in checking
923 * for free space in the buffer. Note that it intentionally does not update
924 * timeouts, so that we can still check them later at wake-up. This function is
925 * dedicated to connection-based stream interfaces.
926 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200927static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200928{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200929 struct channel *ib = si->ib;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200930 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200931
Willy Tarreau34ffd772012-09-03 16:51:27 +0200932 if (unlikely(si->state > SI_ST_EST || (ib->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200933 return;
934
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200935 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100936
Willy Tarreaub31c9712012-11-11 23:05:39 +0100937 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200938 /* stop reading */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100939 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200940 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200941 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200942 }
943 else {
944 /* (re)start reading */
945 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200946 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200947 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200948 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200949}
950
951
Willy Tarreaude5722c2012-08-20 15:01:10 +0200952/* This function is used for inter-stream-interface calls. It is called by the
953 * producer to inform the consumer side that it may be interested in checking
954 * for data in the buffer. Note that it intentionally does not update timeouts,
955 * so that we can still check them later at wake-up.
956 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200957static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200958{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200959 struct channel *ob = si->ob;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200960 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200961
Willy Tarreau34ffd772012-09-03 16:51:27 +0200962 if (unlikely(si->state > SI_ST_EST || (ob->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200963 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200964
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200965 if (unlikely(channel_is_empty(ob))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200966 return;
967
968 if (!ob->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100969 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200970 return;
971
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200972 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200973 /* already subscribed to write notifications, will be called
974 * anyway, so let's avoid calling it especially if the reader
975 * is not ready.
976 */
977 return;
978 }
979
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200980 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100981 /* Before calling the data-level operations, we have to prepare
982 * the polling flags to ensure we properly detect changes.
Willy Tarreaude5722c2012-08-20 15:01:10 +0200983 */
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100984 if (conn_ctrl_ready(conn))
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200985 fd_want_send(conn->t.sock.fd);
Willy Tarreau7d281492012-12-16 19:19:13 +0100986
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200987 conn_refresh_polling_flags(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100988
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200989 si_conn_send(conn);
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100990 if (conn_ctrl_ready(conn) && (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 fd_stop_both(conn->t.sock.fd);
993 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100994 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100995 goto out_wakeup;
996 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200997 }
998
999 /* OK, so now we know that some data might have been sent, and that we may
1000 * have to poll first. We have to do that too if the buffer is not empty.
1001 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +02001002 if (channel_is_empty(ob)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001003 /* the connection is established but we can't write. Either the
1004 * buffer is empty, or we just refrain from sending because the
1005 * ->o limit was reached. Maybe we just wrote the last
1006 * chunk and need to close.
1007 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001008 __conn_data_stop_send(conn);
Willy Tarreaub31c9712012-11-11 23:05:39 +01001009 if (((ob->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001010 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +02001011 (si->state == SI_ST_EST)) {
1012 si_shutw(si);
1013 goto out_wakeup;
1014 }
1015
Willy Tarreaub31c9712012-11-11 23:05:39 +01001016 if ((ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001017 si->flags |= SI_FL_WAIT_DATA;
1018 ob->wex = TICK_ETERNITY;
1019 }
1020 else {
1021 /* Otherwise there are remaining data to be sent in the buffer,
1022 * which means we have to poll before doing so.
1023 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001024 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001025 si->flags &= ~SI_FL_WAIT_DATA;
1026 if (!tick_isset(ob->wex))
1027 ob->wex = tick_add_ifset(now_ms, ob->wto);
1028 }
1029
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001030 if (likely(ob->flags & CF_WRITE_ACTIVITY)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001031 /* update timeout if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001032 if ((ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +02001033 !channel_is_empty(ob))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001034 ob->wex = tick_add_ifset(now_ms, ob->wto);
1035
1036 if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
1037 /* Note: to prevent the client from expiring read timeouts
1038 * during writes, we refresh it. We only do this if the
1039 * interface is not configured for "independent streams",
1040 * because for some applications it's better not to do this,
1041 * for instance when continuously exchanging small amounts
1042 * of data which can full the socket buffers long before a
1043 * write timeout is detected.
1044 */
1045 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
1046 }
1047 }
1048
1049 /* in case of special condition (error, shutdown, end of write...), we
1050 * have to notify the task.
1051 */
Willy Tarreaue6300be2014-01-25 02:33:21 +01001052 if (likely((ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1053 ((ob->flags & CF_WAKE_WRITE) &&
1054 ((channel_is_empty(si->ob) && !ob->to_forward) ||
1055 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001056 out_wakeup:
1057 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
1058 task_wakeup(si->owner, TASK_WOKEN_IO);
1059 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001060
1061 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001062 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001063}
1064
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001065/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001066 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001067 * into the buffer from the connection. It iterates over the transport layer's
1068 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001069 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001070static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001071{
Willy Tarreaue603e692012-09-27 22:20:41 +02001072 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001073 struct channel *chn = si->ib;
Willy Tarreauce323de2012-08-20 21:41:06 +02001074 int ret, max, cur_read;
1075 int read_poll = MAX_READ_POLL_LOOPS;
1076
1077 /* stop immediately on errors. Note that we DON'T want to stop on
1078 * POLL_ERR, as the poller might report a write error while there
1079 * are still data available in the recv buffer. This typically
1080 * happens when we send too large a request to a backend server
1081 * which rejects it before reading it all.
1082 */
1083 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001084 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001085
1086 /* stop here if we reached the end of data */
1087 if (conn_data_read0_pending(conn))
1088 goto out_shutdown_r;
1089
1090 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001091 if (chn->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001092 return;
1093
Willy Tarreau96199b12012-08-24 00:46:52 +02001094 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001095
1096 /* First, let's see if we may splice data across the channel without
1097 * using a buffer.
1098 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001099 if (conn->xprt->rcv_pipe &&
Willy Tarreaufa8e2bc2013-07-18 22:21:54 +02001100 (chn->pipe || chn->to_forward >= MIN_SPLICE_FORWARD) &&
1101 chn->flags & CF_KERN_SPLICING) {
Willy Tarreau9b28e032012-10-12 23:49:43 +02001102 if (buffer_not_empty(chn->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001103 /* We're embarrassed, there are already data pending in
1104 * the buffer and we don't want to have them at two
1105 * locations at a time. Let's indicate we need some
1106 * place and ask the consumer to hurry.
1107 */
1108 goto abort_splice;
1109 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001110
Willy Tarreaucb76e592012-10-12 23:56:57 +02001111 if (unlikely(chn->pipe == NULL)) {
1112 if (pipes_used >= global.maxpipes || !(chn->pipe = get_pipe())) {
1113 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001114 goto abort_splice;
1115 }
1116 }
1117
Willy Tarreaucb76e592012-10-12 23:56:57 +02001118 ret = conn->xprt->rcv_pipe(conn, chn->pipe, chn->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001119 if (ret < 0) {
1120 /* splice not supported on this end, let's disable it */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001121 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001122 goto abort_splice;
1123 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001124
Willy Tarreau96199b12012-08-24 00:46:52 +02001125 if (ret > 0) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001126 if (chn->to_forward != CHN_INFINITE_FORWARD)
1127 chn->to_forward -= ret;
1128 chn->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001129 cur_read += ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001130 chn->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001131 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001132
1133 if (conn_data_read0_pending(conn))
1134 goto out_shutdown_r;
1135
1136 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001137 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001138
Willy Tarreau61d39a02013-07-18 21:49:32 +02001139 if (conn->flags & CO_FL_WAIT_ROOM) {
1140 /* the pipe is full or we have read enough data that it
1141 * could soon be full. Let's stop before needing to poll.
1142 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001143 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001144 __conn_data_stop_recv(conn);
1145 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001146
Willy Tarreauce323de2012-08-20 21:41:06 +02001147 /* splice not possible (anymore), let's go on on standard copy */
1148 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001149
1150 abort_splice:
Willy Tarreau61d39a02013-07-18 21:49:32 +02001151 if (chn->pipe && unlikely(!chn->pipe->data)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001152 put_pipe(chn->pipe);
1153 chn->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001154 }
1155
Willy Tarreau61d39a02013-07-18 21:49:32 +02001156 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1157 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1158 * that if such an event is not handled above in splice, it will be handled here by
1159 * recv().
1160 */
1161 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 +02001162 max = bi_avail(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001163
1164 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001165 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001166 break;
1167 }
1168
Willy Tarreau9b28e032012-10-12 23:49:43 +02001169 ret = conn->xprt->rcv_buf(conn, chn->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001170 if (ret <= 0)
1171 break;
1172
1173 cur_read += ret;
1174
1175 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001176 if (chn->to_forward && !(chn->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001177 unsigned long fwd = ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001178 if (chn->to_forward != CHN_INFINITE_FORWARD) {
1179 if (fwd > chn->to_forward)
1180 fwd = chn->to_forward;
1181 chn->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001182 }
Willy Tarreau9b28e032012-10-12 23:49:43 +02001183 b_adv(chn->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001184 }
1185
Willy Tarreaucb76e592012-10-12 23:56:57 +02001186 chn->flags |= CF_READ_PARTIAL;
1187 chn->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001188
Willy Tarreaucb76e592012-10-12 23:56:57 +02001189 if (channel_full(chn)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001190 /* The buffer is now full, there's no point in going through
1191 * the loop again.
1192 */
Willy Tarreau9b28e032012-10-12 23:49:43 +02001193 if (!(chn->flags & CF_STREAMER_FAST) && (cur_read == buffer_len(chn->buf))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001194 chn->xfer_small = 0;
1195 chn->xfer_large++;
1196 if (chn->xfer_large >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001197 /* we call this buffer a fast streamer if it manages
1198 * to be filled in one call 3 consecutive times.
1199 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001200 chn->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001201 //fputc('+', stderr);
1202 }
1203 }
Willy Tarreaucb76e592012-10-12 23:56:57 +02001204 else if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001205 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001206 chn->xfer_large = 0;
1207 chn->xfer_small++;
1208 if (chn->xfer_small >= 2) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001209 /* if the buffer has been at least half full twice,
1210 * we receive faster than we send, so at least it
1211 * is not a "fast streamer".
1212 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001213 chn->flags &= ~CF_STREAMER_FAST;
Willy Tarreauce323de2012-08-20 21:41:06 +02001214 //fputc('-', stderr);
1215 }
1216 }
1217 else {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001218 chn->xfer_small = 0;
1219 chn->xfer_large = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001220 }
1221
Willy Tarreauce323de2012-08-20 21:41:06 +02001222 si->flags |= SI_FL_WAIT_ROOM;
1223 break;
1224 }
1225
Willy Tarreau5fddab02012-11-09 18:27:26 +01001226 if ((chn->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001227 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001228 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001229 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001230 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001231
1232 /* if too many bytes were missing from last read, it means that
1233 * it's pointless trying to read again because the system does
1234 * not have them in buffers.
1235 */
1236 if (ret < max) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001237 if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001238 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001239 chn->xfer_large = 0;
1240 chn->xfer_small++;
1241 if (chn->xfer_small >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001242 /* we have read less than half of the buffer in
1243 * one pass, and this happened at least 3 times.
1244 * This is definitely not a streamer.
1245 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001246 chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001247 //fputc('!', stderr);
1248 }
1249 }
1250
1251 /* if a streamer has read few data, it may be because we
1252 * have exhausted system buffers. It's not worth trying
1253 * again.
1254 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001255 if (chn->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001256 break;
1257
1258 /* if we read a large block smaller than what we requested,
1259 * it's almost certain we'll never get anything more.
1260 */
1261 if (ret >= global.tune.recv_enough)
1262 break;
1263 }
1264 } /* while !flags */
1265
Willy Tarreau96199b12012-08-24 00:46:52 +02001266 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001267 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001268
Willy Tarreauce323de2012-08-20 21:41:06 +02001269 if (conn_data_read0_pending(conn))
1270 /* connection closed */
1271 goto out_shutdown_r;
1272
1273 return;
1274
1275 out_shutdown_r:
1276 /* we received a shutdown */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001277 chn->flags |= CF_READ_NULL;
1278 if (chn->flags & CF_AUTO_CLOSE)
1279 channel_shutw_now(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001280 stream_sock_read0(si);
1281 conn_data_read0(conn);
1282 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001283}
1284
1285/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001286 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001287 * from the buffer to the connection. It iterates over the transport layer's
1288 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001289 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001290static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001291{
Willy Tarreaue603e692012-09-27 22:20:41 +02001292 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001293 struct channel *chn = si->ob;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001294
1295 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001296 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001297
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001298 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001299 /* a handshake was requested */
1300 return;
1301
1302 /* we might have been called just after an asynchronous shutw */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001303 if (chn->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001304 return;
1305
1306 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001307 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001308
1309 /* OK all done */
1310 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001311}
1312
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001313/*
1314 * This function propagates a null read received on a socket-based connection.
1315 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
1316 * the close is also forwarded to the write side as an abort. This function is
1317 * still socket-specific as it handles a setsockopt() call to set the SO_LINGER
1318 * state on the socket.
1319 */
1320void stream_sock_read0(struct stream_interface *si)
1321{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001322 struct connection *conn = __objt_conn(si->end);
1323
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001324 si->ib->flags &= ~CF_SHUTR_NOW;
1325 if (si->ib->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001326 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001327 si->ib->flags |= CF_SHUTR;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001328 si->ib->rex = TICK_ETERNITY;
1329 si->flags &= ~SI_FL_WAIT_ROOM;
1330
1331 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1332 return;
1333
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001334 if (si->ob->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001335 goto do_close;
1336
1337 if (si->flags & SI_FL_NOHALF) {
1338 /* we want to immediately forward this close to the write side */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001339 /* force flag on ssl to keep session in cache */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001340 if (conn->xprt->shutw)
1341 conn->xprt->shutw(conn, 0);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001342 goto do_close;
1343 }
1344
1345 /* otherwise that's just a normal read shutdown */
Willy Tarreauad38ace2013-12-15 14:19:38 +01001346 if (conn_ctrl_ready(conn))
1347 fdtab[conn->t.sock.fd].linger_risk = 0;
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001348 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001349 return;
1350
1351 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001352 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001353 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001354
1355 si->ib->flags &= ~CF_SHUTR_NOW;
1356 si->ib->flags |= CF_SHUTR;
1357 si->ib->rex = TICK_ETERNITY;
1358
1359 si->ob->flags &= ~CF_SHUTW_NOW;
1360 si->ob->flags |= CF_SHUTW;
1361 si->ob->wex = TICK_ETERNITY;
1362
1363 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1364
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001365 si->state = SI_ST_DIS;
1366 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001367 return;
1368}
1369
Willy Tarreaudded32d2008-11-30 19:48:07 +01001370/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001371 * Local variables:
1372 * c-indent-level: 8
1373 * c-basic-offset: 8
1374 * End:
1375 */