blob: 8450117c236ca3483fa4c994d541fd7d912f21dd [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
Willy Tarreaubf883e02014-11-25 21:10:35 +010022#include <common/buffer.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010023#include <common/compat.h>
24#include <common/config.h>
25#include <common/debug.h>
26#include <common/standard.h>
27#include <common/ticks.h>
28#include <common/time.h>
29
Willy Tarreauc7e42382012-08-24 19:22:53 +020030#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020031#include <proto/connection.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010032#include <proto/fd.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020033#include <proto/pipe.h>
Willy Tarreaubf883e02014-11-25 21:10:35 +010034#include <proto/session.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020035#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010036#include <proto/task.h>
37
Willy Tarreaufd31e532012-07-23 18:24:25 +020038#include <types/pipe.h>
39
Willy Tarreauf873d752012-05-11 17:47:17 +020040/* socket functions used when running a stream interface as a task */
Willy Tarreauf873d752012-05-11 17:47:17 +020041static void stream_int_update_embedded(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020042static void stream_int_shutr(struct stream_interface *si);
43static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020044static void stream_int_chk_rcv(struct stream_interface *si);
45static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020046static void stream_int_update_conn(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020047static void stream_int_shutr_conn(struct stream_interface *si);
48static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020049static void stream_int_chk_rcv_conn(struct stream_interface *si);
50static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau4aa36832012-10-02 20:07:22 +020051static void si_conn_recv_cb(struct connection *conn);
52static void si_conn_send_cb(struct connection *conn);
Willy Tarreau2396c1c2012-10-03 21:12:16 +020053static int si_conn_wake_cb(struct connection *conn);
Willy Tarreau27375622013-12-17 00:00:28 +010054static int si_idle_conn_wake_cb(struct connection *conn);
55static void si_idle_conn_null_cb(struct connection *conn);
Willy Tarreauf873d752012-05-11 17:47:17 +020056
Willy Tarreauc5788912012-08-24 18:12:41 +020057/* stream-interface operations for embedded tasks */
58struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020059 .update = stream_int_update_embedded,
Willy Tarreau5c979a92012-05-07 17:15:39 +020060 .chk_rcv = stream_int_chk_rcv,
61 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020062 .shutr = stream_int_shutr,
63 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020064};
65
Willy Tarreauc5788912012-08-24 18:12:41 +020066/* stream-interface operations for connections */
67struct si_ops si_conn_ops = {
68 .update = stream_int_update_conn,
69 .chk_rcv = stream_int_chk_rcv_conn,
70 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020071 .shutr = stream_int_shutr_conn,
72 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020073};
74
Willy Tarreau74beec32012-10-03 00:41:04 +020075struct data_cb si_conn_cb = {
Willy Tarreauc5788912012-08-24 18:12:41 +020076 .recv = si_conn_recv_cb,
77 .send = si_conn_send_cb,
Willy Tarreau4aa36832012-10-02 20:07:22 +020078 .wake = si_conn_wake_cb,
Willy Tarreauc5788912012-08-24 18:12:41 +020079};
80
Willy Tarreau27375622013-12-17 00:00:28 +010081struct data_cb si_idle_conn_cb = {
82 .recv = si_idle_conn_null_cb,
83 .send = si_idle_conn_null_cb,
84 .wake = si_idle_conn_wake_cb,
85};
86
Willy Tarreaucff64112008-11-03 06:26:53 +010087/*
88 * This function only has to be called once after a wakeup event in case of
89 * suspected timeout. It controls the stream interface timeouts and sets
90 * si->flags accordingly. It does NOT close anything, as this timeout may
91 * be used for any purpose. It returns 1 if the timeout fired, otherwise
92 * zero.
93 */
94int stream_int_check_timeouts(struct stream_interface *si)
95{
96 if (tick_is_expired(si->exp, now_ms)) {
97 si->flags |= SI_FL_EXP;
98 return 1;
99 }
100 return 0;
101}
102
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100103/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +0100104void stream_int_report_error(struct stream_interface *si)
105{
106 if (!si->err_type)
107 si->err_type = SI_ET_DATA_ERR;
108
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100109 si_oc(si)->flags |= CF_WRITE_ERROR;
110 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100111}
112
113/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100114 * Returns a message to the client ; the connection is shut down for read,
115 * and the request is cleared so that no server connection can be initiated.
116 * The buffer is marked for read shutdown on the other side to protect the
117 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100118 * "chunk". If it is null, then an empty message is used. The reply buffer does
119 * not need to be empty before this, and its contents will not be overwritten.
120 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100121 */
122void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
123{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100124 struct channel *ic = si_ic(si);
125 struct channel *oc = si_oc(si);
126
127 channel_auto_read(ic);
128 channel_abort(ic);
129 channel_auto_close(ic);
130 channel_erase(ic);
131 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100132
Willy Tarreau148d0992010-01-10 10:21:21 +0100133 if (likely(msg && msg->len))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100134 bo_inject(oc, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100135
Willy Tarreauafc8a222014-11-28 15:46:27 +0100136 oc->wex = tick_add_ifset(now_ms, oc->wto);
137 channel_auto_read(oc);
138 channel_auto_close(oc);
139 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100140}
141
Willy Tarreaufb90d942009-09-05 20:57:35 +0200142/* default update function for embedded tasks, to be used at the end of the i/o handler */
Willy Tarreauf873d752012-05-11 17:47:17 +0200143static void stream_int_update_embedded(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200144{
Willy Tarreau3488e252010-08-09 16:24:56 +0200145 int old_flags = si->flags;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100146 struct channel *ic = si_ic(si);
147 struct channel *oc = si_oc(si);
Willy Tarreau3488e252010-08-09 16:24:56 +0200148
Willy Tarreauafc8a222014-11-28 15:46:27 +0100149 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200150 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100151 si, si->state, ic->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200152
153 if (si->state != SI_ST_EST)
154 return;
155
Willy Tarreauafc8a222014-11-28 15:46:27 +0100156 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW &&
157 channel_is_empty(oc))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200158 si_shutw(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200159
Willy Tarreauafc8a222014-11-28 15:46:27 +0100160 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200161 si->flags |= SI_FL_WAIT_DATA;
162
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200163 /* we're almost sure that we need some space if the buffer is not
164 * empty, even if it's not full, because the applets can't fill it.
165 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100166 if ((ic->flags & (CF_SHUTR|CF_DONT_READ)) == 0 && !channel_is_empty(ic))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200167 si->flags |= SI_FL_WAIT_ROOM;
168
Willy Tarreauafc8a222014-11-28 15:46:27 +0100169 if (oc->flags & CF_WRITE_ACTIVITY) {
170 if (tick_isset(oc->wex))
171 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200172 }
173
Willy Tarreauafc8a222014-11-28 15:46:27 +0100174 if (ic->flags & CF_READ_ACTIVITY ||
175 (oc->flags & CF_WRITE_ACTIVITY && !(si->flags & SI_FL_INDEP_STR))) {
176 if (tick_isset(ic->rex))
177 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf27b5ea2009-10-03 22:01:18 +0200178 }
179
Willy Tarreau3488e252010-08-09 16:24:56 +0200180 /* save flags to detect changes */
181 old_flags = si->flags;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100182 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
183 channel_may_recv(oc) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100184 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
185 si_chk_rcv(si_opposite(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200186
Willy Tarreauafc8a222014-11-28 15:46:27 +0100187 if (((ic->flags & CF_READ_PARTIAL) && !channel_is_empty(ic)) &&
188 (ic->pipe /* always try to send spliced data */ ||
189 (ic->buf->i == 0 && (si_opposite(si)->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100190 si_chk_snd(si_opposite(si));
Willy Tarreau3488e252010-08-09 16:24:56 +0200191 /* check if the consumer has freed some space */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100192 if (channel_may_recv(ic) && !ic->pipe)
Willy Tarreau3488e252010-08-09 16:24:56 +0200193 si->flags &= ~SI_FL_WAIT_ROOM;
194 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200195
196 /* Note that we're trying to wake up in two conditions here :
197 * - special event, which needs the holder task attention
198 * - status indicating that the applet can go on working. This
199 * is rather hard because we might be blocking on output and
200 * don't want to wake up on input and vice-versa. The idea is
Willy Tarreau3488e252010-08-09 16:24:56 +0200201 * to only rely on the changes the chk_* might have performed.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200202 */
203 if (/* check stream interface changes */
Willy Tarreau3488e252010-08-09 16:24:56 +0200204 ((old_flags & ~si->flags) & (SI_FL_WAIT_ROOM|SI_FL_WAIT_DATA)) ||
205
206 /* changes on the production side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100207 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200208 si->state != SI_ST_EST ||
209 (si->flags & SI_FL_ERR) ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100210 ((ic->flags & CF_READ_PARTIAL) &&
211 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200212
213 /* changes on the consumption side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100214 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
215 ((oc->flags & CF_WRITE_ACTIVITY) &&
216 ((oc->flags & CF_SHUTW) ||
217 ((oc->flags & CF_WAKE_WRITE) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100218 (si_opposite(si)->state != SI_ST_EST ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100219 (channel_is_empty(oc) && !oc->to_forward)))))) {
Willy Tarreau07373b82014-11-28 12:08:47 +0100220 if (!(si->flags & SI_FL_DONT_WAKE))
221 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200222 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100223 if (ic->flags & CF_READ_ACTIVITY)
224 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200225}
226
Willy Tarreau4a36b562012-08-06 19:31:45 +0200227/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200228 * This function performs a shutdown-read on a stream interface attached to an
229 * applet in a connected or init state (it does nothing for other states). It
230 * either shuts the read side or marks itself as closed. The buffer flags are
231 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
232 * we also forward the close to the write side. The owner task is woken up if
Willy Tarreau6fe15412013-09-29 15:16:03 +0200233 * it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200234 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200235static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200236{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100237 struct channel *ic = si_ic(si);
238
239 ic->flags &= ~CF_SHUTR_NOW;
240 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200241 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100242 ic->flags |= CF_SHUTR;
243 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244 si->flags &= ~SI_FL_WAIT_ROOM;
245
246 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200247 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200248
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100249 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250 si->state = SI_ST_DIS;
251 si->exp = TICK_ETERNITY;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200252 si_applet_release(si);
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200253 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200254 else if (si->flags & SI_FL_NOHALF) {
255 /* we want to immediately forward this close to the write side */
256 return stream_int_shutw(si);
257 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200258
Willy Tarreau4a36b562012-08-06 19:31:45 +0200259 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100260 if (!(si->flags & SI_FL_DONT_WAKE))
261 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200262}
263
Willy Tarreau4a36b562012-08-06 19:31:45 +0200264/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200265 * This function performs a shutdown-write on a stream interface attached to an
266 * applet in a connected or init state (it does nothing for other states). It
267 * either shuts the write side or marks itself as closed. The buffer flags are
268 * updated to reflect the new state. It does also close everything if the SI
269 * was marked as being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200270 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200271static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200272{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100273 struct channel *ic = si_ic(si);
274 struct channel *oc = si_oc(si);
275
276 oc->flags &= ~CF_SHUTW_NOW;
277 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200278 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100279 oc->flags |= CF_SHUTW;
280 oc->wex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200281 si->flags &= ~SI_FL_WAIT_DATA;
282
283 switch (si->state) {
284 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200285 /* we have to shut before closing, otherwise some short messages
286 * may never leave the system, especially when there are remaining
287 * unread data in the socket input buffer, or when nolinger is set.
288 * However, if SI_FL_NOLINGER is explicitly set, we know there is
289 * no risk so we close both sides immediately.
290 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200291 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100292 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200293 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200294
295 /* fall through */
296 case SI_ST_CON:
297 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100298 case SI_ST_QUE:
299 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200300 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200301 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200302 si_applet_release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200303 default:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200304 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100305 ic->flags &= ~CF_SHUTR_NOW;
306 ic->flags |= CF_SHUTR;
307 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200308 si->exp = TICK_ETERNITY;
309 }
310
Willy Tarreau4a36b562012-08-06 19:31:45 +0200311 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100312 if (!(si->flags & SI_FL_DONT_WAKE))
313 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200314}
315
316/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200317static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200318{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100319 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200320
Willy Tarreauafc8a222014-11-28 15:46:27 +0100321 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200322 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100323 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200324
Willy Tarreauafc8a222014-11-28 15:46:27 +0100325 if (unlikely(si->state != SI_ST_EST || (ic->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200326 return;
327
Willy Tarreauafc8a222014-11-28 15:46:27 +0100328 if (!channel_may_recv(ic) || ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200329 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200330 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200331 }
332 else {
333 /* (re)start reading */
334 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreau07373b82014-11-28 12:08:47 +0100335 if (!(si->flags & SI_FL_DONT_WAKE))
336 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200337 }
338}
339
340/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200341static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200342{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100343 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200344
Willy Tarreauafc8a222014-11-28 15:46:27 +0100345 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200346 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100347 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200348
Willy Tarreauafc8a222014-11-28 15:46:27 +0100349 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200350 return;
351
352 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100353 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200354 return;
355
356 /* Otherwise there are remaining data to be sent in the buffer,
357 * so we tell the handler.
358 */
359 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100360 if (!tick_isset(oc->wex))
361 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200362
Willy Tarreau07373b82014-11-28 12:08:47 +0100363 if (!(si->flags & SI_FL_DONT_WAKE))
364 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200365}
366
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100367/* Register an applet to handle a stream_interface as part of the
368 * stream interface's owner task. The SI will wake it up everytime it
369 * is solicited. The task's processing function must call the applet's
370 * function before returning. It must be deleted by the task handler
371 * using stream_int_unregister_handler(), possibly from within the
372 * function itself. It also pre-initializes the applet's context and
373 * returns it (or NULL in case it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200374 */
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100375struct appctx *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200376{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100377 struct appctx *appctx;
378
Willy Tarreau07373b82014-11-28 12:08:47 +0100379 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200380
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100381 appctx = si_alloc_appctx(si);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100382 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100383 return NULL;
384
385 appctx_set_applet(appctx, app);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200386 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100387 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200388}
389
Willy Tarreaufb90d942009-09-05 20:57:35 +0200390/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100391 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200392 */
393void stream_int_unregister_handler(struct stream_interface *si)
394{
Willy Tarreau32e3c6a2013-10-11 19:34:20 +0200395 si_detach(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200396}
397
Willy Tarreau2c6be842012-07-06 17:12:34 +0200398/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200399 * established. It returns 0 if it fails in a fatal way or needs to poll to go
400 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200401 * flags (the bit is provided in <flag> by the caller). It is designed to be
402 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200403 * Note that it can emit a PROXY line by relying on the other end's address
404 * when the connection is attached to a stream interface, or by resolving the
405 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200406 */
407int conn_si_send_proxy(struct connection *conn, unsigned int flag)
408{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200409 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200410 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200411 goto out_error;
412
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100413 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200414 goto out_error;
415
Willy Tarreau2c6be842012-07-06 17:12:34 +0200416 /* If we have a PROXY line to send, we'll use this to validate the
417 * connection, in which case the connection is validated only once
418 * we've sent the whole proxy line. Otherwise we use connect().
419 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200420 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200421 int ret;
422
423 /* The target server expects a PROXY line to be sent first.
424 * If the send_proxy_ofs is negative, it corresponds to the
425 * offset to start sending from then end of the proxy string
426 * (which is recomputed every time since it's constant). If
427 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200428 * We can only send a "normal" PROXY line when the connection
429 * is attached to a stream interface. Otherwise we can only
430 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200431 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200432 if (conn->data == &si_conn_cb) {
433 struct stream_interface *si = conn->owner;
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100434 struct connection *remote = objt_conn(si_opposite(si)->end);
David Safb76832014-05-08 23:42:08 -0400435 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), remote);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200436 }
437 else {
438 /* The target server expects a LOCAL line to be sent first. Retrieving
439 * local or remote addresses may fail until the connection is established.
440 */
441 conn_get_from_addr(conn);
442 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
443 goto out_wait;
444
445 conn_get_to_addr(conn);
446 if (!(conn->flags & CO_FL_ADDR_TO_SET))
447 goto out_wait;
448
David Safb76832014-05-08 23:42:08 -0400449 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200450 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200451
Willy Tarreau2c6be842012-07-06 17:12:34 +0200452 if (!ret)
453 goto out_error;
454
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200455 if (conn->send_proxy_ofs > 0)
456 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200457
Willy Tarreaua1a74742012-08-24 12:14:49 +0200458 /* we have to send trash from (ret+sp for -sp bytes). If the
459 * data layer has a pending write, we'll also set MSG_MORE.
460 */
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100461 ret = conn_sock_send(conn, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
462 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200463
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100464 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200465 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200466
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);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200490 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200491}
492
Willy Tarreau27375622013-12-17 00:00:28 +0100493
494/* Tiny I/O callback called on recv/send I/O events on idle connections.
495 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
496 * is notified and can kill the connection.
497 */
498static void si_idle_conn_null_cb(struct connection *conn)
499{
Willy Tarreauf31fb072015-03-13 00:32:20 +0100500 conn_drain(conn);
Willy Tarreau27375622013-12-17 00:00:28 +0100501}
502
503/* Callback to be used by connection I/O handlers when some activity is detected
504 * on an idle server connection. Its main purpose is to kill the connection once
505 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
506 * it killed the connection.
507 */
508static int si_idle_conn_wake_cb(struct connection *conn)
509{
510 struct stream_interface *si = conn->owner;
511
512 if (!conn_ctrl_ready(conn))
513 return 0;
514
515 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
516 /* warning, we can't do anything on <conn> after this call ! */
517 conn_force_close(conn);
518 conn_free(conn);
519 si->end = NULL;
520 return -1;
521 }
522 return 0;
523}
524
Willy Tarreau100c4672012-08-20 12:06:26 +0200525/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200526 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200527 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200528 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200529 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200530 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200531static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200532{
Willy Tarreaue603e692012-09-27 22:20:41 +0200533 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100534 struct channel *ic = si_ic(si);
535 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200536
Willy Tarreauafc8a222014-11-28 15:46:27 +0100537 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufd31e532012-07-23 18:24:25 +0200538 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100539 si, si->state, ic->flags, oc->flags);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200540
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200541 if (conn->flags & CO_FL_ERROR)
542 si->flags |= SI_FL_ERR;
543
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200544 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200545 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200546 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100547 oc->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200548 }
549
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200550 /* process consumer side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100551 if (channel_is_empty(oc)) {
552 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200553 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200554 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200555 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100556 oc->wex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200557 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200558
Willy Tarreauafc8a222014-11-28 15:46:27 +0100559 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200560 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200561
Willy Tarreauafc8a222014-11-28 15:46:27 +0100562 if (oc->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200563 /* update timeouts if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100564 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
565 !channel_is_empty(oc))
566 if (tick_isset(oc->wex))
567 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200568
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200569 if (!(si->flags & SI_FL_INDEP_STR))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100570 if (tick_isset(ic->rex))
571 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200572
Willy Tarreauafc8a222014-11-28 15:46:27 +0100573 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
574 channel_may_recv(oc) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100575 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
576 si_chk_rcv(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200577 }
578
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200579 /* process producer side.
580 * We might have some data the consumer is waiting for.
581 * We can do fast-forwarding, but we avoid doing this for partial
582 * buffers, because it is very likely that it will be done again
583 * immediately afterwards once the following data is parsed (eg:
584 * HTTP chunking).
585 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100586 if (((ic->flags & CF_READ_PARTIAL) && !channel_is_empty(ic)) &&
587 (ic->pipe /* always try to send spliced data */ ||
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100588 (si_ib(si)->i == 0 && (si_opposite(si)->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100589 int last_len = ic->pipe ? ic->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200590
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100591 si_chk_snd(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200592
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200593 /* check if the consumer has freed some space either in the
594 * buffer or in the pipe.
595 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100596 if (channel_may_recv(ic) &&
597 (!last_len || !ic->pipe || ic->pipe->data < last_len))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200598 si->flags &= ~SI_FL_WAIT_ROOM;
599 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200600
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200601 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200602 __conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100603 ic->rex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200604 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100605 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
606 channel_may_recv(ic)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100607 /* we must re-enable reading if si_chk_snd() has freed some space */
608 __conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100609 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
610 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200611 }
612
613 /* wake the task up only when needed */
614 if (/* changes on the production side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100615 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200616 si->state != SI_ST_EST ||
617 (si->flags & SI_FL_ERR) ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100618 ((ic->flags & CF_READ_PARTIAL) &&
619 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200620
621 /* changes on the consumption side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100622 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
623 ((oc->flags & CF_WRITE_ACTIVITY) &&
624 ((oc->flags & CF_SHUTW) ||
625 ((oc->flags & CF_WAKE_WRITE) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100626 (si_opposite(si)->state != SI_ST_EST ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100627 (channel_is_empty(oc) && !oc->to_forward)))))) {
Willy Tarreau07373b82014-11-28 12:08:47 +0100628 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200629 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100630 if (ic->flags & CF_READ_ACTIVITY)
631 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau10fc09e2014-11-25 19:46:36 +0100632
633 session_release_buffers(si_sess(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200634 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200635}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200636
Willy Tarreau5368d802012-08-21 18:22:06 +0200637/*
638 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800639 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800640 * caller to commit polling changes. The caller should check conn->flags
641 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200642 */
Godbach4f489902013-12-04 17:24:06 +0800643static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200644{
Willy Tarreaue603e692012-09-27 22:20:41 +0200645 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100646 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200647 int ret;
648
Willy Tarreauafc8a222014-11-28 15:46:27 +0100649 if (oc->pipe && conn->xprt->snd_pipe) {
650 ret = conn->xprt->snd_pipe(conn, oc->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200651 if (ret > 0)
Willy Tarreauafc8a222014-11-28 15:46:27 +0100652 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200653
Willy Tarreauafc8a222014-11-28 15:46:27 +0100654 if (!oc->pipe->data) {
655 put_pipe(oc->pipe);
656 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200657 }
658
Willy Tarreau96199b12012-08-24 00:46:52 +0200659 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800660 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200661 }
662
663 /* At this point, the pipe is empty, but we may still have data pending
664 * in the normal buffer.
665 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100666 if (!oc->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800667 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200668
Godbache68e02d2013-10-11 15:48:29 +0800669 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200670 * data left, and that there are sendable buffered data.
671 */
Willy Tarreau310987a2014-01-22 19:46:33 +0100672 if (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH | CO_FL_WAIT_DATA | CO_FL_HANDSHAKE))) {
Willy Tarreau5368d802012-08-21 18:22:06 +0200673 /* check if we want to inform the kernel that we're interested in
674 * sending more data after this call. We want this if :
675 * - we're about to close after this last send and want to merge
676 * the ongoing FIN with the last segment.
677 * - we know we can't send everything at once and must get back
678 * here because of unaligned data
679 * - there is still a finite amount of data to forward
680 * The test is arranged so that the most common case does only 2
681 * tests.
682 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100683 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200684
Willy Tarreauafc8a222014-11-28 15:46:27 +0100685 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
686 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
687 (oc->flags & CF_EXPECT_MORE))) ||
688 ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100689 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200690
Willy Tarreauafc8a222014-11-28 15:46:27 +0100691 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100692 send_flag |= CO_SFL_STREAMER;
693
Willy Tarreauafc8a222014-11-28 15:46:27 +0100694 ret = conn->xprt->snd_buf(conn, oc->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800695 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100696 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200697
Willy Tarreauafc8a222014-11-28 15:46:27 +0100698 if (!oc->buf->o) {
Godbache68e02d2013-10-11 15:48:29 +0800699 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100700 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800701 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200702
Godbache68e02d2013-10-11 15:48:29 +0800703 /* if some data remain in the buffer, it's only because the
704 * system buffers are full, we will try next time.
705 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200706 }
Godbache68e02d2013-10-11 15:48:29 +0800707 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200708}
709
710
Willy Tarreau100c4672012-08-20 12:06:26 +0200711/* Updates the timers and flags of a stream interface attached to a connection,
712 * depending on the buffers' flags. It should only be called once after the
713 * buffer flags have settled down, and before they are cleared. It doesn't
714 * harm to call it as often as desired (it just slightly hurts performance).
715 * It is only meant to be called by upper layers after buffer flags have been
716 * manipulated by analysers.
717 */
718void stream_int_update_conn(struct stream_interface *si)
719{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100720 struct channel *ic = si_ic(si);
721 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200722 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200723
Willy Tarreau100c4672012-08-20 12:06:26 +0200724 /* Check if we need to close the read side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100725 if (!(ic->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200726 /* Read not closed, update FD status and timeout for reads */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100727 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200728 /* stop reading */
729 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100730 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200731 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200732 conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100733 ic->rex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200734 }
735 }
736 else {
737 /* (re)start reading and update timeout. Note: we don't recompute the timeout
738 * everytime we get here, otherwise it would risk never to expire. We only
739 * update it if is was not yet set. The stream socket handler will already
740 * have updated it if there has been a completed I/O.
741 */
742 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200743 conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100744 if (!(ic->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ic->rex))
745 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200746 }
747 }
748
749 /* Check if we need to close the write side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100750 if (!(oc->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200751 /* Write not closed, update FD status and timeout for writes */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100752 if (channel_is_empty(oc)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200753 /* stop writing */
754 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100755 if ((oc->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200756 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200757 conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100758 oc->wex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200759 }
760 }
761 else {
762 /* (re)start writing and update timeout. Note: we don't recompute the timeout
763 * everytime we get here, otherwise it would risk never to expire. We only
764 * update it if is was not yet set. The stream socket handler will already
765 * have updated it if there has been a completed I/O.
766 */
767 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200768 conn_data_want_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100769 if (!tick_isset(oc->wex)) {
770 oc->wex = tick_add_ifset(now_ms, oc->wto);
771 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200772 /* Note: depending on the protocol, we don't know if we're waiting
773 * for incoming data or not. So in order to prevent the socket from
774 * expiring read timeouts during writes, we refresh the read timeout,
775 * except if it was already infinite or if we have explicitly setup
776 * independent streams.
777 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100778 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200779 }
780 }
781 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200782 }
783}
784
785/*
786 * This function performs a shutdown-read on a stream interface attached to
787 * a connection in a connected or init state (it does nothing for other
788 * states). It either shuts the read side or marks itself as closed. The buffer
789 * flags are updated to reflect the new state. If the stream interface has
790 * SI_FL_NOHALF, we also forward the close to the write side. If a control
791 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200792 * descriptors are then shutdown or closed accordingly. The function
793 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200794 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200795static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200796{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200797 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100798 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200799
Willy Tarreauafc8a222014-11-28 15:46:27 +0100800 ic->flags &= ~CF_SHUTR_NOW;
801 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200802 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100803 ic->flags |= CF_SHUTR;
804 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200805 si->flags &= ~SI_FL_WAIT_ROOM;
806
807 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200808 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200809
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100810 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200811 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200812 si->state = SI_ST_DIS;
813 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200814 }
815 else if (si->flags & SI_FL_NOHALF) {
816 /* we want to immediately forward this close to the write side */
817 return stream_int_shutw_conn(si);
818 }
819 else if (conn->ctrl) {
820 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200821 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200822 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200823}
824
825/*
826 * This function performs a shutdown-write on a stream interface attached to
827 * a connection in a connected or init state (it does nothing for other
828 * states). It either shuts the write side or marks itself as closed. The
829 * buffer flags are updated to reflect the new state. It does also close
830 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +0100831 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200832 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200833static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200834{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200835 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100836 struct channel *ic = si_ic(si);
837 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200838
Willy Tarreauafc8a222014-11-28 15:46:27 +0100839 oc->flags &= ~CF_SHUTW_NOW;
840 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200841 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100842 oc->flags |= CF_SHUTW;
843 oc->wex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200844 si->flags &= ~SI_FL_WAIT_DATA;
845
846 switch (si->state) {
847 case SI_ST_EST:
848 /* we have to shut before closing, otherwise some short messages
849 * may never leave the system, especially when there are remaining
850 * unread data in the socket input buffer, or when nolinger is set.
851 * However, if SI_FL_NOLINGER is explicitly set, we know there is
852 * no risk so we close both sides immediately.
853 */
854 if (si->flags & SI_FL_ERR) {
855 /* quick close, the socket is alredy shut anyway */
856 }
857 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200858 /* unclean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100859 conn_data_shutw_hard(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200860 }
861 else {
862 /* clean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100863 conn_data_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200864
865 /* If the stream interface is configured to disable half-open
866 * connections, we'll skip the shutdown(), but only if the
867 * read size is already closed. Otherwise we can't support
868 * closed write with pending read (eg: abortonclose while
869 * waiting for the server).
870 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100871 if (!(si->flags & SI_FL_NOHALF) || !(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200872 /* We shutdown transport layer */
Willy Tarreau4dfd54f2015-03-12 22:44:53 +0100873 conn_sock_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200874
Willy Tarreauafc8a222014-11-28 15:46:27 +0100875 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200876 /* OK just a shutw, but we want the caller
877 * to disable polling on this FD if exists.
878 */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100879 conn_cond_update_polling(conn);
Willy Tarreau6fe15412013-09-29 15:16:03 +0200880 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200881 }
882 }
883 }
884
885 /* fall through */
886 case SI_ST_CON:
887 /* we may have to close a pending connection, and mark the
888 * response buffer as shutr
889 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200890 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200891 /* fall through */
892 case SI_ST_CER:
893 case SI_ST_QUE:
894 case SI_ST_TAR:
895 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200896 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200897 default:
898 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100899 ic->flags &= ~CF_SHUTR_NOW;
900 ic->flags |= CF_SHUTR;
901 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200902 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200903 }
904}
905
Willy Tarreau46a8d922012-08-20 12:38:36 +0200906/* This function is used for inter-stream-interface calls. It is called by the
907 * consumer to inform the producer side that it may be interested in checking
908 * for free space in the buffer. Note that it intentionally does not update
909 * timeouts, so that we can still check them later at wake-up. This function is
910 * dedicated to connection-based stream interfaces.
911 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200912static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200913{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100914 struct channel *ic = si_ic(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200915 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200916
Willy Tarreauafc8a222014-11-28 15:46:27 +0100917 if (unlikely(si->state > SI_ST_EST || (ic->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200918 return;
919
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200920 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100921
Willy Tarreauafc8a222014-11-28 15:46:27 +0100922 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200923 /* stop reading */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100924 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200925 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200926 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200927 }
928 else {
929 /* (re)start reading */
930 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200931 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200932 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200933 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200934}
935
936
Willy Tarreaude5722c2012-08-20 15:01:10 +0200937/* This function is used for inter-stream-interface calls. It is called by the
938 * producer to inform the consumer side that it may be interested in checking
939 * for data in the buffer. Note that it intentionally does not update timeouts,
940 * so that we can still check them later at wake-up.
941 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200942static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200943{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100944 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200945 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200946
Willy Tarreauafc8a222014-11-28 15:46:27 +0100947 if (unlikely(si->state > SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200948 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200949
Willy Tarreauafc8a222014-11-28 15:46:27 +0100950 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200951 return;
952
Willy Tarreauafc8a222014-11-28 15:46:27 +0100953 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100954 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200955 return;
956
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200957 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200958 /* already subscribed to write notifications, will be called
959 * anyway, so let's avoid calling it especially if the reader
960 * is not ready.
961 */
962 return;
963 }
964
Willy Tarreau708e7172014-01-21 10:27:49 +0100965 /* Before calling the data-level operations, we have to prepare
966 * the polling flags to ensure we properly detect changes.
967 */
968 conn_refresh_polling_flags(conn);
969 __conn_data_want_send(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100970
Willy Tarreau708e7172014-01-21 10:27:49 +0100971 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200972 si_conn_send(conn);
Willy Tarreau798c3c92014-01-21 10:30:08 +0100973 if (conn->flags & CO_FL_ERROR) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100974 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200975 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100976 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100977 goto out_wakeup;
978 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200979 }
980
981 /* OK, so now we know that some data might have been sent, and that we may
982 * have to poll first. We have to do that too if the buffer is not empty.
983 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100984 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200985 /* the connection is established but we can't write. Either the
986 * buffer is empty, or we just refrain from sending because the
987 * ->o limit was reached. Maybe we just wrote the last
988 * chunk and need to close.
989 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200990 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100991 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200992 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200993 (si->state == SI_ST_EST)) {
994 si_shutw(si);
995 goto out_wakeup;
996 }
997
Willy Tarreauafc8a222014-11-28 15:46:27 +0100998 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200999 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001000 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001001 }
1002 else {
1003 /* Otherwise there are remaining data to be sent in the buffer,
1004 * which means we have to poll before doing so.
1005 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001006 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001007 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001008 if (!tick_isset(oc->wex))
1009 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001010 }
1011
Willy Tarreauafc8a222014-11-28 15:46:27 +01001012 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1013 struct channel *ic = si_ic(si);
1014
Willy Tarreaude5722c2012-08-20 15:01:10 +02001015 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001016 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1017 !channel_is_empty(oc))
1018 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001019
Willy Tarreauafc8a222014-11-28 15:46:27 +01001020 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001021 /* Note: to prevent the client from expiring read timeouts
1022 * during writes, we refresh it. We only do this if the
1023 * interface is not configured for "independent streams",
1024 * because for some applications it's better not to do this,
1025 * for instance when continuously exchanging small amounts
1026 * of data which can full the socket buffers long before a
1027 * write timeout is detected.
1028 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001029 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001030 }
1031 }
1032
1033 /* in case of special condition (error, shutdown, end of write...), we
1034 * have to notify the task.
1035 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001036 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1037 ((oc->flags & CF_WAKE_WRITE) &&
1038 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreaue6300be2014-01-25 02:33:21 +01001039 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001040 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001041 if (!(si->flags & SI_FL_DONT_WAKE))
1042 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001043 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001044
1045 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001046 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001047}
1048
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001049/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001050 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001051 * into the buffer from the connection. It iterates over the transport layer's
1052 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001053 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001054static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001055{
Willy Tarreaue603e692012-09-27 22:20:41 +02001056 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001057 struct channel *ic = si_ic(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001058 int ret, max, cur_read;
1059 int read_poll = MAX_READ_POLL_LOOPS;
1060
1061 /* stop immediately on errors. Note that we DON'T want to stop on
1062 * POLL_ERR, as the poller might report a write error while there
1063 * are still data available in the recv buffer. This typically
1064 * happens when we send too large a request to a backend server
1065 * which rejects it before reading it all.
1066 */
1067 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001068 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001069
1070 /* stop here if we reached the end of data */
1071 if (conn_data_read0_pending(conn))
1072 goto out_shutdown_r;
1073
1074 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001075 if (ic->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001076 return;
1077
Willy Tarreau96199b12012-08-24 00:46:52 +02001078 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001079
Willy Tarreauafc8a222014-11-28 15:46:27 +01001080 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !ic->buf->o &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001081 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001082 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001083 /* The buffer was empty and nothing was transferred for more
1084 * than one second. This was caused by a pause and not by
1085 * congestion. Reset any streaming mode to reduce latency.
1086 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001087 ic->xfer_small = 0;
1088 ic->xfer_large = 0;
1089 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001090 }
1091
Willy Tarreau96199b12012-08-24 00:46:52 +02001092 /* First, let's see if we may splice data across the channel without
1093 * using a buffer.
1094 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001095 if (conn->xprt->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001096 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1097 ic->flags & CF_KERN_SPLICING) {
1098 if (buffer_not_empty(ic->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001099 /* We're embarrassed, there are already data pending in
1100 * the buffer and we don't want to have them at two
1101 * locations at a time. Let's indicate we need some
1102 * place and ask the consumer to hurry.
1103 */
1104 goto abort_splice;
1105 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001106
Willy Tarreauafc8a222014-11-28 15:46:27 +01001107 if (unlikely(ic->pipe == NULL)) {
1108 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1109 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001110 goto abort_splice;
1111 }
1112 }
1113
Willy Tarreauafc8a222014-11-28 15:46:27 +01001114 ret = conn->xprt->rcv_pipe(conn, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001115 if (ret < 0) {
1116 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001117 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001118 goto abort_splice;
1119 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001120
Willy Tarreau96199b12012-08-24 00:46:52 +02001121 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001122 if (ic->to_forward != CHN_INFINITE_FORWARD)
1123 ic->to_forward -= ret;
1124 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001125 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001126 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001127 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001128
1129 if (conn_data_read0_pending(conn))
1130 goto out_shutdown_r;
1131
1132 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001133 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001134
Willy Tarreau61d39a02013-07-18 21:49:32 +02001135 if (conn->flags & CO_FL_WAIT_ROOM) {
1136 /* the pipe is full or we have read enough data that it
1137 * could soon be full. Let's stop before needing to poll.
1138 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001139 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001140 __conn_data_stop_recv(conn);
1141 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001142
Willy Tarreauce323de2012-08-20 21:41:06 +02001143 /* splice not possible (anymore), let's go on on standard copy */
1144 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001145
1146 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001147 if (ic->pipe && unlikely(!ic->pipe->data)) {
1148 put_pipe(ic->pipe);
1149 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001150 }
1151
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001152 /* now we'll need a buffer */
Willy Tarreau78955f42014-12-28 13:09:02 +01001153 if (!session_alloc_recv_buffer(ic)) {
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001154 si->flags |= SI_FL_WAIT_ROOM;
1155 goto end_recv;
1156 }
1157
Willy Tarreau61d39a02013-07-18 21:49:32 +02001158 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1159 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1160 * that if such an event is not handled above in splice, it will be handled here by
1161 * recv().
1162 */
Willy Tarreau310987a2014-01-22 19:46:33 +01001163 while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001164 max = channel_recv_max(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001165
1166 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001167 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001168 break;
1169 }
1170
Willy Tarreauafc8a222014-11-28 15:46:27 +01001171 ret = conn->xprt->rcv_buf(conn, ic->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001172 if (ret <= 0)
1173 break;
1174
1175 cur_read += ret;
1176
1177 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001178 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001179 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001180 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1181 if (fwd > ic->to_forward)
1182 fwd = ic->to_forward;
1183 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001184 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001185 b_adv(ic->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001186 }
1187
Willy Tarreauafc8a222014-11-28 15:46:27 +01001188 ic->flags |= CF_READ_PARTIAL;
1189 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001190
Willy Tarreauafc8a222014-11-28 15:46:27 +01001191 if (!channel_may_recv(ic)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001192 si->flags |= SI_FL_WAIT_ROOM;
1193 break;
1194 }
1195
Willy Tarreauafc8a222014-11-28 15:46:27 +01001196 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001197 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001198 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001199 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001200 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001201
1202 /* if too many bytes were missing from last read, it means that
1203 * it's pointless trying to read again because the system does
1204 * not have them in buffers.
1205 */
1206 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001207 /* if a streamer has read few data, it may be because we
1208 * have exhausted system buffers. It's not worth trying
1209 * again.
1210 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001211 if (ic->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001212 break;
1213
1214 /* if we read a large block smaller than what we requested,
1215 * it's almost certain we'll never get anything more.
1216 */
1217 if (ret >= global.tune.recv_enough)
1218 break;
1219 }
1220 } /* while !flags */
1221
Willy Tarreauc5890e62014-02-09 17:47:01 +01001222 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001223 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
1224 (cur_read <= ic->buf->size / 2)) {
1225 ic->xfer_large = 0;
1226 ic->xfer_small++;
1227 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001228 /* we have read less than half of the buffer in
1229 * one pass, and this happened at least 3 times.
1230 * This is definitely not a streamer.
1231 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001232 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001233 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001234 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001235 /* if the buffer has been at least half full twice,
1236 * we receive faster than we send, so at least it
1237 * is not a "fast streamer".
1238 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001239 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001240 }
1241 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001242 else if (!(ic->flags & CF_STREAMER_FAST) &&
1243 (cur_read >= ic->buf->size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001244 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001245 ic->xfer_small = 0;
1246 ic->xfer_large++;
1247 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001248 /* we call this buffer a fast streamer if it manages
1249 * to be filled in one call 3 consecutive times.
1250 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001251 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001252 }
1253 }
1254 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001255 ic->xfer_small = 0;
1256 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001257 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001258 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001259 }
1260
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001261 end_recv:
1262 if (conn->flags & CO_FL_ERROR)
1263 return;
1264
Willy Tarreauce323de2012-08-20 21:41:06 +02001265 if (conn_data_read0_pending(conn))
1266 /* connection closed */
1267 goto out_shutdown_r;
1268
1269 return;
1270
1271 out_shutdown_r:
1272 /* we received a shutdown */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001273 ic->flags |= CF_READ_NULL;
1274 if (ic->flags & CF_AUTO_CLOSE)
1275 channel_shutw_now(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001276 stream_sock_read0(si);
1277 conn_data_read0(conn);
1278 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001279}
1280
1281/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001282 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001283 * from the buffer to the connection. It iterates over the transport layer's
1284 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001285 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001286static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001287{
Willy Tarreaue603e692012-09-27 22:20:41 +02001288 struct stream_interface *si = conn->owner;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001289
1290 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001291 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001292
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001293 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001294 /* a handshake was requested */
1295 return;
1296
1297 /* we might have been called just after an asynchronous shutw */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001298 if (si_oc(si)->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001299 return;
1300
1301 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001302 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001303
1304 /* OK all done */
1305 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001306}
1307
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001308/*
1309 * This function propagates a null read received on a socket-based connection.
1310 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001311 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001312 */
1313void stream_sock_read0(struct stream_interface *si)
1314{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001315 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001316 struct channel *ic = si_ic(si);
1317 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001318
Willy Tarreauafc8a222014-11-28 15:46:27 +01001319 ic->flags &= ~CF_SHUTR_NOW;
1320 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001321 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001322 ic->flags |= CF_SHUTR;
1323 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001324 si->flags &= ~SI_FL_WAIT_ROOM;
1325
1326 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1327 return;
1328
Willy Tarreauafc8a222014-11-28 15:46:27 +01001329 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001330 goto do_close;
1331
1332 if (si->flags & SI_FL_NOHALF) {
1333 /* we want to immediately forward this close to the write side */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001334 /* force flag on ssl to keep session in cache */
Willy Tarreau1398aa12015-03-12 23:04:07 +01001335 conn_data_shutw_hard(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001336 goto do_close;
1337 }
1338
1339 /* otherwise that's just a normal read shutdown */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001340 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001341 return;
1342
1343 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001344 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001345 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001346
Willy Tarreauafc8a222014-11-28 15:46:27 +01001347 ic->flags &= ~CF_SHUTR_NOW;
1348 ic->flags |= CF_SHUTR;
1349 ic->rex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001350
Willy Tarreauafc8a222014-11-28 15:46:27 +01001351 oc->flags &= ~CF_SHUTW_NOW;
1352 oc->flags |= CF_SHUTW;
1353 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001354
1355 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1356
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001357 si->state = SI_ST_DIS;
1358 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001359 return;
1360}
1361
Willy Tarreaudded32d2008-11-30 19:48:07 +01001362/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001363 * Local variables:
1364 * c-indent-level: 8
1365 * c-basic-offset: 8
1366 * End:
1367 */