blob: 8df5008b82862821afc2234e641d300abfa85c11 [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 Tarreau96199b12012-08-24 00:46:52 +020032#include <proto/pipe.h>
Willy Tarreau87b09662015-04-03 00:22:06 +020033#include <proto/stream.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020034#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010035#include <proto/task.h>
36
Willy Tarreaufd31e532012-07-23 18:24:25 +020037#include <types/pipe.h>
38
Willy Tarreauf873d752012-05-11 17:47:17 +020039/* socket functions used when running a stream interface as a task */
Willy Tarreauf873d752012-05-11 17:47:17 +020040static void stream_int_update_embedded(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020041static void stream_int_shutr(struct stream_interface *si);
42static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020043static void stream_int_chk_rcv(struct stream_interface *si);
44static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020045static void stream_int_update_conn(struct stream_interface *si);
Willy Tarreau6fe15412013-09-29 15:16:03 +020046static void stream_int_shutr_conn(struct stream_interface *si);
47static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020048static void stream_int_chk_rcv_conn(struct stream_interface *si);
49static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau4aa36832012-10-02 20:07:22 +020050static void si_conn_recv_cb(struct connection *conn);
51static void si_conn_send_cb(struct connection *conn);
Willy Tarreau2396c1c2012-10-03 21:12:16 +020052static int si_conn_wake_cb(struct connection *conn);
Willy Tarreau27375622013-12-17 00:00:28 +010053static int si_idle_conn_wake_cb(struct connection *conn);
54static void si_idle_conn_null_cb(struct connection *conn);
Willy Tarreauf873d752012-05-11 17:47:17 +020055
Willy Tarreauc5788912012-08-24 18:12:41 +020056/* stream-interface operations for embedded tasks */
57struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020058 .update = stream_int_update_embedded,
Willy Tarreau5c979a92012-05-07 17:15:39 +020059 .chk_rcv = stream_int_chk_rcv,
60 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020061 .shutr = stream_int_shutr,
62 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020063};
64
Willy Tarreauc5788912012-08-24 18:12:41 +020065/* stream-interface operations for connections */
66struct si_ops si_conn_ops = {
67 .update = stream_int_update_conn,
68 .chk_rcv = stream_int_chk_rcv_conn,
69 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020070 .shutr = stream_int_shutr_conn,
71 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020072};
73
Willy Tarreau74beec32012-10-03 00:41:04 +020074struct data_cb si_conn_cb = {
Willy Tarreauc5788912012-08-24 18:12:41 +020075 .recv = si_conn_recv_cb,
76 .send = si_conn_send_cb,
Willy Tarreau4aa36832012-10-02 20:07:22 +020077 .wake = si_conn_wake_cb,
Willy Tarreauc5788912012-08-24 18:12:41 +020078};
79
Willy Tarreau27375622013-12-17 00:00:28 +010080struct data_cb si_idle_conn_cb = {
81 .recv = si_idle_conn_null_cb,
82 .send = si_idle_conn_null_cb,
83 .wake = si_idle_conn_wake_cb,
84};
85
Willy Tarreaucff64112008-11-03 06:26:53 +010086/*
87 * This function only has to be called once after a wakeup event in case of
88 * suspected timeout. It controls the stream interface timeouts and sets
89 * si->flags accordingly. It does NOT close anything, as this timeout may
90 * be used for any purpose. It returns 1 if the timeout fired, otherwise
91 * zero.
92 */
93int stream_int_check_timeouts(struct stream_interface *si)
94{
95 if (tick_is_expired(si->exp, now_ms)) {
96 si->flags |= SI_FL_EXP;
97 return 1;
98 }
99 return 0;
100}
101
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100102/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +0100103void stream_int_report_error(struct stream_interface *si)
104{
105 if (!si->err_type)
106 si->err_type = SI_ET_DATA_ERR;
107
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100108 si_oc(si)->flags |= CF_WRITE_ERROR;
109 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100110}
111
112/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100113 * Returns a message to the client ; the connection is shut down for read,
114 * and the request is cleared so that no server connection can be initiated.
115 * The buffer is marked for read shutdown on the other side to protect the
116 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100117 * "chunk". If it is null, then an empty message is used. The reply buffer does
118 * not need to be empty before this, and its contents will not be overwritten.
119 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100120 */
121void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
122{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100123 struct channel *ic = si_ic(si);
124 struct channel *oc = si_oc(si);
125
126 channel_auto_read(ic);
127 channel_abort(ic);
128 channel_auto_close(ic);
129 channel_erase(ic);
130 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100131
Willy Tarreau148d0992010-01-10 10:21:21 +0100132 if (likely(msg && msg->len))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100133 bo_inject(oc, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100134
Willy Tarreauafc8a222014-11-28 15:46:27 +0100135 oc->wex = tick_add_ifset(now_ms, oc->wto);
136 channel_auto_read(oc);
137 channel_auto_close(oc);
138 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100139}
140
Willy Tarreaufb90d942009-09-05 20:57:35 +0200141/* default update function for embedded tasks, to be used at the end of the i/o handler */
Willy Tarreauf873d752012-05-11 17:47:17 +0200142static void stream_int_update_embedded(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200143{
Willy Tarreau3488e252010-08-09 16:24:56 +0200144 int old_flags = si->flags;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100145 struct channel *ic = si_ic(si);
146 struct channel *oc = si_oc(si);
Willy Tarreau3488e252010-08-09 16:24:56 +0200147
Willy Tarreauafc8a222014-11-28 15:46:27 +0100148 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200149 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100150 si, si->state, ic->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200151
152 if (si->state != SI_ST_EST)
153 return;
154
Willy Tarreauafc8a222014-11-28 15:46:27 +0100155 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW &&
156 channel_is_empty(oc))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200157 si_shutw(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200158
Willy Tarreauafc8a222014-11-28 15:46:27 +0100159 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200160 si->flags |= SI_FL_WAIT_DATA;
161
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200162 /* we're almost sure that we need some space if the buffer is not
163 * empty, even if it's not full, because the applets can't fill it.
164 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100165 if ((ic->flags & (CF_SHUTR|CF_DONT_READ)) == 0 && !channel_is_empty(ic))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200166 si->flags |= SI_FL_WAIT_ROOM;
167
Willy Tarreauafc8a222014-11-28 15:46:27 +0100168 if (oc->flags & CF_WRITE_ACTIVITY) {
169 if (tick_isset(oc->wex))
170 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200171 }
172
Willy Tarreauafc8a222014-11-28 15:46:27 +0100173 if (ic->flags & CF_READ_ACTIVITY ||
174 (oc->flags & CF_WRITE_ACTIVITY && !(si->flags & SI_FL_INDEP_STR))) {
175 if (tick_isset(ic->rex))
176 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf27b5ea2009-10-03 22:01:18 +0200177 }
178
Willy Tarreau3488e252010-08-09 16:24:56 +0200179 /* save flags to detect changes */
180 old_flags = si->flags;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100181 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
182 channel_may_recv(oc) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100183 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
184 si_chk_rcv(si_opposite(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200185
Willy Tarreauafc8a222014-11-28 15:46:27 +0100186 if (((ic->flags & CF_READ_PARTIAL) && !channel_is_empty(ic)) &&
187 (ic->pipe /* always try to send spliced data */ ||
188 (ic->buf->i == 0 && (si_opposite(si)->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100189 si_chk_snd(si_opposite(si));
Willy Tarreau3488e252010-08-09 16:24:56 +0200190 /* check if the consumer has freed some space */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100191 if (channel_may_recv(ic) && !ic->pipe)
Willy Tarreau3488e252010-08-09 16:24:56 +0200192 si->flags &= ~SI_FL_WAIT_ROOM;
193 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200194
195 /* Note that we're trying to wake up in two conditions here :
196 * - special event, which needs the holder task attention
197 * - status indicating that the applet can go on working. This
198 * is rather hard because we might be blocking on output and
199 * don't want to wake up on input and vice-versa. The idea is
Willy Tarreau3488e252010-08-09 16:24:56 +0200200 * to only rely on the changes the chk_* might have performed.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200201 */
202 if (/* check stream interface changes */
Willy Tarreau3488e252010-08-09 16:24:56 +0200203 ((old_flags & ~si->flags) & (SI_FL_WAIT_ROOM|SI_FL_WAIT_DATA)) ||
204
205 /* changes on the production side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100206 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200207 si->state != SI_ST_EST ||
208 (si->flags & SI_FL_ERR) ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100209 ((ic->flags & CF_READ_PARTIAL) &&
210 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200211
212 /* changes on the consumption side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100213 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
214 ((oc->flags & CF_WRITE_ACTIVITY) &&
215 ((oc->flags & CF_SHUTW) ||
216 ((oc->flags & CF_WAKE_WRITE) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100217 (si_opposite(si)->state != SI_ST_EST ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100218 (channel_is_empty(oc) && !oc->to_forward)))))) {
Willy Tarreau07373b82014-11-28 12:08:47 +0100219 if (!(si->flags & SI_FL_DONT_WAKE))
220 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200221 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100222 if (ic->flags & CF_READ_ACTIVITY)
223 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200224}
225
Willy Tarreau4a36b562012-08-06 19:31:45 +0200226/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200227 * This function performs a shutdown-read on a stream interface attached to an
228 * applet in a connected or init state (it does nothing for other states). It
229 * either shuts the read side or marks itself as closed. The buffer flags are
230 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
231 * we also forward the close to the write side. The owner task is woken up if
Willy Tarreau6fe15412013-09-29 15:16:03 +0200232 * it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200233 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200234static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200235{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100236 struct channel *ic = si_ic(si);
237
238 ic->flags &= ~CF_SHUTR_NOW;
239 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200240 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100241 ic->flags |= CF_SHUTR;
242 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200243 si->flags &= ~SI_FL_WAIT_ROOM;
244
245 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200246 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200247
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100248 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200249 si->state = SI_ST_DIS;
250 si->exp = TICK_ETERNITY;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200251 si_applet_release(si);
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200252 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200253 else if (si->flags & SI_FL_NOHALF) {
254 /* we want to immediately forward this close to the write side */
255 return stream_int_shutw(si);
256 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200257
Willy Tarreau4a36b562012-08-06 19:31:45 +0200258 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100259 if (!(si->flags & SI_FL_DONT_WAKE))
260 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200261}
262
Willy Tarreau4a36b562012-08-06 19:31:45 +0200263/*
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200264 * This function performs a shutdown-write on a stream interface attached to an
265 * applet in a connected or init state (it does nothing for other states). It
266 * either shuts the write side or marks itself as closed. The buffer flags are
267 * updated to reflect the new state. It does also close everything if the SI
268 * was marked as being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200269 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200270static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200271{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100272 struct channel *ic = si_ic(si);
273 struct channel *oc = si_oc(si);
274
275 oc->flags &= ~CF_SHUTW_NOW;
276 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200277 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100278 oc->flags |= CF_SHUTW;
279 oc->wex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200280 si->flags &= ~SI_FL_WAIT_DATA;
281
282 switch (si->state) {
283 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200284 /* we have to shut before closing, otherwise some short messages
285 * may never leave the system, especially when there are remaining
286 * unread data in the socket input buffer, or when nolinger is set.
287 * However, if SI_FL_NOLINGER is explicitly set, we know there is
288 * no risk so we close both sides immediately.
289 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200290 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100291 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200292 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200293
294 /* fall through */
295 case SI_ST_CON:
296 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100297 case SI_ST_QUE:
298 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200299 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200300 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200301 si_applet_release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200302 default:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200303 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100304 ic->flags &= ~CF_SHUTR_NOW;
305 ic->flags |= CF_SHUTR;
306 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200307 si->exp = TICK_ETERNITY;
308 }
309
Willy Tarreau4a36b562012-08-06 19:31:45 +0200310 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100311 if (!(si->flags & SI_FL_DONT_WAKE))
312 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200313}
314
315/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200316static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200317{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100318 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200319
Willy Tarreauafc8a222014-11-28 15:46:27 +0100320 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200321 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100322 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200323
Willy Tarreauafc8a222014-11-28 15:46:27 +0100324 if (unlikely(si->state != SI_ST_EST || (ic->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200325 return;
326
Willy Tarreauafc8a222014-11-28 15:46:27 +0100327 if (!channel_may_recv(ic) || ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200328 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200329 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200330 }
331 else {
332 /* (re)start reading */
333 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreau07373b82014-11-28 12:08:47 +0100334 if (!(si->flags & SI_FL_DONT_WAKE))
335 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200336 }
337}
338
339/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200340static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200341{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100342 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200343
Willy Tarreauafc8a222014-11-28 15:46:27 +0100344 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200345 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100346 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200347
Willy Tarreauafc8a222014-11-28 15:46:27 +0100348 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200349 return;
350
351 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100352 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200353 return;
354
355 /* Otherwise there are remaining data to be sent in the buffer,
356 * so we tell the handler.
357 */
358 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100359 if (!tick_isset(oc->wex))
360 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200361
Willy Tarreau07373b82014-11-28 12:08:47 +0100362 if (!(si->flags & SI_FL_DONT_WAKE))
363 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200364}
365
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100366/* Register an applet to handle a stream_interface as part of the
367 * stream interface's owner task. The SI will wake it up everytime it
368 * is solicited. The task's processing function must call the applet's
369 * function before returning. It must be deleted by the task handler
370 * using stream_int_unregister_handler(), possibly from within the
371 * function itself. It also pre-initializes the applet's context and
372 * returns it (or NULL in case it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200373 */
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100374struct appctx *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200375{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100376 struct appctx *appctx;
377
Willy Tarreau07373b82014-11-28 12:08:47 +0100378 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200379
Willy Tarreaua7513f52015-04-05 00:15:26 +0200380 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100381 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100382 return NULL;
383
Willy Tarreaufb90d942009-09-05 20:57:35 +0200384 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100385 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200386}
387
Willy Tarreaufb90d942009-09-05 20:57:35 +0200388/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100389 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200390 */
391void stream_int_unregister_handler(struct stream_interface *si)
392{
Willy Tarreau32e3c6a2013-10-11 19:34:20 +0200393 si_detach(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200394}
395
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200397 * established. It returns 0 if it fails in a fatal way or needs to poll to go
398 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200399 * flags (the bit is provided in <flag> by the caller). It is designed to be
400 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200401 * Note that it can emit a PROXY line by relying on the other end's address
402 * when the connection is attached to a stream interface, or by resolving the
403 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200404 */
405int conn_si_send_proxy(struct connection *conn, unsigned int flag)
406{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200407 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200408 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200409 goto out_error;
410
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100411 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200412 goto out_error;
413
Willy Tarreau2c6be842012-07-06 17:12:34 +0200414 /* If we have a PROXY line to send, we'll use this to validate the
415 * connection, in which case the connection is validated only once
416 * we've sent the whole proxy line. Otherwise we use connect().
417 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200418 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200419 int ret;
420
421 /* The target server expects a PROXY line to be sent first.
422 * If the send_proxy_ofs is negative, it corresponds to the
423 * offset to start sending from then end of the proxy string
424 * (which is recomputed every time since it's constant). If
425 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200426 * We can only send a "normal" PROXY line when the connection
427 * is attached to a stream interface. Otherwise we can only
428 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200429 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200430 if (conn->data == &si_conn_cb) {
431 struct stream_interface *si = conn->owner;
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100432 struct connection *remote = objt_conn(si_opposite(si)->end);
David Safb76832014-05-08 23:42:08 -0400433 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), remote);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200434 }
435 else {
436 /* The target server expects a LOCAL line to be sent first. Retrieving
437 * local or remote addresses may fail until the connection is established.
438 */
439 conn_get_from_addr(conn);
440 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
441 goto out_wait;
442
443 conn_get_to_addr(conn);
444 if (!(conn->flags & CO_FL_ADDR_TO_SET))
445 goto out_wait;
446
David Safb76832014-05-08 23:42:08 -0400447 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200448 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200449
Willy Tarreau2c6be842012-07-06 17:12:34 +0200450 if (!ret)
451 goto out_error;
452
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200453 if (conn->send_proxy_ofs > 0)
454 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200455
Willy Tarreaua1a74742012-08-24 12:14:49 +0200456 /* we have to send trash from (ret+sp for -sp bytes). If the
457 * data layer has a pending write, we'll also set MSG_MORE.
458 */
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100459 ret = conn_sock_send(conn, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
460 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200461
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100462 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200463 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200464
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200465 conn->send_proxy_ofs += ret; /* becomes zero once complete */
466 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200467 goto out_wait;
468
469 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100470 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200471 }
472
Willy Tarreaua1a74742012-08-24 12:14:49 +0200473 /* The connection is ready now, simply return and let the connection
474 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200475 */
476 if (conn->flags & CO_FL_WAIT_L4_CONN)
477 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200478 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200479 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200480
481 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200482 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200483 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200484 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200485
486 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200487 __conn_sock_stop_recv(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200488 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200489}
490
Willy Tarreau27375622013-12-17 00:00:28 +0100491
492/* Tiny I/O callback called on recv/send I/O events on idle connections.
493 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
494 * is notified and can kill the connection.
495 */
496static void si_idle_conn_null_cb(struct connection *conn)
497{
Willy Tarreaud85c4852015-03-13 00:40:28 +0100498 conn_sock_drain(conn);
Willy Tarreau27375622013-12-17 00:00:28 +0100499}
500
501/* Callback to be used by connection I/O handlers when some activity is detected
502 * on an idle server connection. Its main purpose is to kill the connection once
503 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
504 * it killed the connection.
505 */
506static int si_idle_conn_wake_cb(struct connection *conn)
507{
508 struct stream_interface *si = conn->owner;
509
510 if (!conn_ctrl_ready(conn))
511 return 0;
512
513 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
514 /* warning, we can't do anything on <conn> after this call ! */
515 conn_force_close(conn);
516 conn_free(conn);
517 si->end = NULL;
518 return -1;
519 }
520 return 0;
521}
522
Willy Tarreau100c4672012-08-20 12:06:26 +0200523/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200524 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200525 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200526 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200527 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200528 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200529static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200530{
Willy Tarreaue603e692012-09-27 22:20:41 +0200531 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100532 struct channel *ic = si_ic(si);
533 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200534
Willy Tarreauafc8a222014-11-28 15:46:27 +0100535 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufd31e532012-07-23 18:24:25 +0200536 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100537 si, si->state, ic->flags, oc->flags);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200538
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200539 if (conn->flags & CO_FL_ERROR)
540 si->flags |= SI_FL_ERR;
541
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200542 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200543 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200544 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100545 oc->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200546 }
547
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200548 /* process consumer side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100549 if (channel_is_empty(oc)) {
550 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200551 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200552 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200553 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100554 oc->wex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200555 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200556
Willy Tarreauafc8a222014-11-28 15:46:27 +0100557 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200558 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200559
Willy Tarreauafc8a222014-11-28 15:46:27 +0100560 if (oc->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200561 /* update timeouts if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100562 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
563 !channel_is_empty(oc))
564 if (tick_isset(oc->wex))
565 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200566
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200567 if (!(si->flags & SI_FL_INDEP_STR))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100568 if (tick_isset(ic->rex))
569 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200570
Willy Tarreauafc8a222014-11-28 15:46:27 +0100571 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
572 channel_may_recv(oc) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100573 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
574 si_chk_rcv(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200575 }
576
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200577 /* process producer side.
578 * We might have some data the consumer is waiting for.
579 * We can do fast-forwarding, but we avoid doing this for partial
580 * buffers, because it is very likely that it will be done again
581 * immediately afterwards once the following data is parsed (eg:
582 * HTTP chunking).
583 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100584 if (((ic->flags & CF_READ_PARTIAL) && !channel_is_empty(ic)) &&
585 (ic->pipe /* always try to send spliced data */ ||
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100586 (si_ib(si)->i == 0 && (si_opposite(si)->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100587 int last_len = ic->pipe ? ic->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200588
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100589 si_chk_snd(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200590
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200591 /* check if the consumer has freed some space either in the
592 * buffer or in the pipe.
593 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100594 if (channel_may_recv(ic) &&
595 (!last_len || !ic->pipe || ic->pipe->data < last_len))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200596 si->flags &= ~SI_FL_WAIT_ROOM;
597 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200598
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200599 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200600 __conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100601 ic->rex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200602 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100603 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
604 channel_may_recv(ic)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100605 /* we must re-enable reading if si_chk_snd() has freed some space */
606 __conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100607 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
608 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200609 }
610
611 /* wake the task up only when needed */
612 if (/* changes on the production side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100613 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200614 si->state != SI_ST_EST ||
615 (si->flags & SI_FL_ERR) ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100616 ((ic->flags & CF_READ_PARTIAL) &&
617 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200618
619 /* changes on the consumption side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100620 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
621 ((oc->flags & CF_WRITE_ACTIVITY) &&
622 ((oc->flags & CF_SHUTW) ||
623 ((oc->flags & CF_WAKE_WRITE) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100624 (si_opposite(si)->state != SI_ST_EST ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100625 (channel_is_empty(oc) && !oc->to_forward)))))) {
Willy Tarreau07373b82014-11-28 12:08:47 +0100626 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200627 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100628 if (ic->flags & CF_READ_ACTIVITY)
629 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau10fc09e2014-11-25 19:46:36 +0100630
Willy Tarreau87b09662015-04-03 00:22:06 +0200631 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200632 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200633}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200634
Willy Tarreau5368d802012-08-21 18:22:06 +0200635/*
636 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800637 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800638 * caller to commit polling changes. The caller should check conn->flags
639 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200640 */
Godbach4f489902013-12-04 17:24:06 +0800641static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200642{
Willy Tarreaue603e692012-09-27 22:20:41 +0200643 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100644 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200645 int ret;
646
Willy Tarreauafc8a222014-11-28 15:46:27 +0100647 if (oc->pipe && conn->xprt->snd_pipe) {
648 ret = conn->xprt->snd_pipe(conn, oc->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200649 if (ret > 0)
Willy Tarreauafc8a222014-11-28 15:46:27 +0100650 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200651
Willy Tarreauafc8a222014-11-28 15:46:27 +0100652 if (!oc->pipe->data) {
653 put_pipe(oc->pipe);
654 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200655 }
656
Willy Tarreau96199b12012-08-24 00:46:52 +0200657 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800658 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200659 }
660
661 /* At this point, the pipe is empty, but we may still have data pending
662 * in the normal buffer.
663 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100664 if (!oc->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800665 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200666
Godbache68e02d2013-10-11 15:48:29 +0800667 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200668 * data left, and that there are sendable buffered data.
669 */
Willy Tarreau310987a2014-01-22 19:46:33 +0100670 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 +0200671 /* check if we want to inform the kernel that we're interested in
672 * sending more data after this call. We want this if :
673 * - we're about to close after this last send and want to merge
674 * the ongoing FIN with the last segment.
675 * - we know we can't send everything at once and must get back
676 * here because of unaligned data
677 * - there is still a finite amount of data to forward
678 * The test is arranged so that the most common case does only 2
679 * tests.
680 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100681 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200682
Willy Tarreauafc8a222014-11-28 15:46:27 +0100683 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
684 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
685 (oc->flags & CF_EXPECT_MORE))) ||
686 ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100687 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200688
Willy Tarreauafc8a222014-11-28 15:46:27 +0100689 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100690 send_flag |= CO_SFL_STREAMER;
691
Willy Tarreauafc8a222014-11-28 15:46:27 +0100692 ret = conn->xprt->snd_buf(conn, oc->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800693 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100694 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200695
Willy Tarreauafc8a222014-11-28 15:46:27 +0100696 if (!oc->buf->o) {
Godbache68e02d2013-10-11 15:48:29 +0800697 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100698 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800699 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200700
Godbache68e02d2013-10-11 15:48:29 +0800701 /* if some data remain in the buffer, it's only because the
702 * system buffers are full, we will try next time.
703 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200704 }
Godbache68e02d2013-10-11 15:48:29 +0800705 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200706}
707
708
Willy Tarreau100c4672012-08-20 12:06:26 +0200709/* Updates the timers and flags of a stream interface attached to a connection,
710 * depending on the buffers' flags. It should only be called once after the
711 * buffer flags have settled down, and before they are cleared. It doesn't
712 * harm to call it as often as desired (it just slightly hurts performance).
713 * It is only meant to be called by upper layers after buffer flags have been
714 * manipulated by analysers.
715 */
716void stream_int_update_conn(struct stream_interface *si)
717{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100718 struct channel *ic = si_ic(si);
719 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200720 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200721
Willy Tarreau100c4672012-08-20 12:06:26 +0200722 /* Check if we need to close the read side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100723 if (!(ic->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200724 /* Read not closed, update FD status and timeout for reads */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100725 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200726 /* stop reading */
727 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100728 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200729 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200730 conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100731 ic->rex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200732 }
733 }
734 else {
735 /* (re)start reading and update timeout. Note: we don't recompute the timeout
736 * everytime we get here, otherwise it would risk never to expire. We only
737 * update it if is was not yet set. The stream socket handler will already
738 * have updated it if there has been a completed I/O.
739 */
740 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200741 conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100742 if (!(ic->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ic->rex))
743 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200744 }
745 }
746
747 /* Check if we need to close the write side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100748 if (!(oc->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200749 /* Write not closed, update FD status and timeout for writes */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100750 if (channel_is_empty(oc)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200751 /* stop writing */
752 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100753 if ((oc->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200754 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200755 conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100756 oc->wex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200757 }
758 }
759 else {
760 /* (re)start writing and update timeout. Note: we don't recompute the timeout
761 * everytime we get here, otherwise it would risk never to expire. We only
762 * update it if is was not yet set. The stream socket handler will already
763 * have updated it if there has been a completed I/O.
764 */
765 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200766 conn_data_want_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100767 if (!tick_isset(oc->wex)) {
768 oc->wex = tick_add_ifset(now_ms, oc->wto);
769 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200770 /* Note: depending on the protocol, we don't know if we're waiting
771 * for incoming data or not. So in order to prevent the socket from
772 * expiring read timeouts during writes, we refresh the read timeout,
773 * except if it was already infinite or if we have explicitly setup
774 * independent streams.
775 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100776 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200777 }
778 }
779 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200780 }
781}
782
783/*
784 * This function performs a shutdown-read on a stream interface attached to
785 * a connection in a connected or init state (it does nothing for other
786 * states). It either shuts the read side or marks itself as closed. The buffer
787 * flags are updated to reflect the new state. If the stream interface has
788 * SI_FL_NOHALF, we also forward the close to the write side. If a control
789 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200790 * descriptors are then shutdown or closed accordingly. The function
791 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200792 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200793static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200794{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200795 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100796 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200797
Willy Tarreauafc8a222014-11-28 15:46:27 +0100798 ic->flags &= ~CF_SHUTR_NOW;
799 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200800 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100801 ic->flags |= CF_SHUTR;
802 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200803 si->flags &= ~SI_FL_WAIT_ROOM;
804
805 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200806 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200807
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100808 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200809 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200810 si->state = SI_ST_DIS;
811 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200812 }
813 else if (si->flags & SI_FL_NOHALF) {
814 /* we want to immediately forward this close to the write side */
815 return stream_int_shutw_conn(si);
816 }
817 else if (conn->ctrl) {
818 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200819 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200820 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200821}
822
823/*
824 * This function performs a shutdown-write on a stream interface attached to
825 * a connection in a connected or init state (it does nothing for other
826 * states). It either shuts the write side or marks itself as closed. The
827 * buffer flags are updated to reflect the new state. It does also close
828 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +0100829 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200830 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200831static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200832{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200833 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100834 struct channel *ic = si_ic(si);
835 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200836
Willy Tarreauafc8a222014-11-28 15:46:27 +0100837 oc->flags &= ~CF_SHUTW_NOW;
838 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200839 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100840 oc->flags |= CF_SHUTW;
841 oc->wex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200842 si->flags &= ~SI_FL_WAIT_DATA;
843
844 switch (si->state) {
845 case SI_ST_EST:
846 /* we have to shut before closing, otherwise some short messages
847 * may never leave the system, especially when there are remaining
848 * unread data in the socket input buffer, or when nolinger is set.
849 * However, if SI_FL_NOLINGER is explicitly set, we know there is
850 * no risk so we close both sides immediately.
851 */
852 if (si->flags & SI_FL_ERR) {
853 /* quick close, the socket is alredy shut anyway */
854 }
855 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200856 /* unclean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100857 conn_data_shutw_hard(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200858 }
859 else {
860 /* clean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100861 conn_data_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200862
863 /* If the stream interface is configured to disable half-open
864 * connections, we'll skip the shutdown(), but only if the
865 * read size is already closed. Otherwise we can't support
866 * closed write with pending read (eg: abortonclose while
867 * waiting for the server).
868 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100869 if (!(si->flags & SI_FL_NOHALF) || !(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200870 /* We shutdown transport layer */
Willy Tarreau4dfd54f2015-03-12 22:44:53 +0100871 conn_sock_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200872
Willy Tarreauafc8a222014-11-28 15:46:27 +0100873 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200874 /* OK just a shutw, but we want the caller
875 * to disable polling on this FD if exists.
876 */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100877 conn_cond_update_polling(conn);
Willy Tarreau6fe15412013-09-29 15:16:03 +0200878 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200879 }
880 }
881 }
882
883 /* fall through */
884 case SI_ST_CON:
885 /* we may have to close a pending connection, and mark the
886 * response buffer as shutr
887 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200888 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200889 /* fall through */
890 case SI_ST_CER:
891 case SI_ST_QUE:
892 case SI_ST_TAR:
893 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200894 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200895 default:
896 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100897 ic->flags &= ~CF_SHUTR_NOW;
898 ic->flags |= CF_SHUTR;
899 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200900 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200901 }
902}
903
Willy Tarreau46a8d922012-08-20 12:38:36 +0200904/* This function is used for inter-stream-interface calls. It is called by the
905 * consumer to inform the producer side that it may be interested in checking
906 * for free space in the buffer. Note that it intentionally does not update
907 * timeouts, so that we can still check them later at wake-up. This function is
908 * dedicated to connection-based stream interfaces.
909 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200910static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200911{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100912 struct channel *ic = si_ic(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200913 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200914
Willy Tarreauafc8a222014-11-28 15:46:27 +0100915 if (unlikely(si->state > SI_ST_EST || (ic->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200916 return;
917
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200918 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100919
Willy Tarreauafc8a222014-11-28 15:46:27 +0100920 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200921 /* stop reading */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100922 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200923 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200924 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200925 }
926 else {
927 /* (re)start reading */
928 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200929 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200930 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200931 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200932}
933
934
Willy Tarreaude5722c2012-08-20 15:01:10 +0200935/* This function is used for inter-stream-interface calls. It is called by the
936 * producer to inform the consumer side that it may be interested in checking
937 * for data in the buffer. Note that it intentionally does not update timeouts,
938 * so that we can still check them later at wake-up.
939 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200940static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200941{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100942 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200943 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200944
Willy Tarreauafc8a222014-11-28 15:46:27 +0100945 if (unlikely(si->state > SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200946 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200947
Willy Tarreauafc8a222014-11-28 15:46:27 +0100948 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200949 return;
950
Willy Tarreauafc8a222014-11-28 15:46:27 +0100951 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100952 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200953 return;
954
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200955 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200956 /* already subscribed to write notifications, will be called
957 * anyway, so let's avoid calling it especially if the reader
958 * is not ready.
959 */
960 return;
961 }
962
Willy Tarreau708e7172014-01-21 10:27:49 +0100963 /* Before calling the data-level operations, we have to prepare
964 * the polling flags to ensure we properly detect changes.
965 */
966 conn_refresh_polling_flags(conn);
967 __conn_data_want_send(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100968
Willy Tarreau708e7172014-01-21 10:27:49 +0100969 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200970 si_conn_send(conn);
Willy Tarreau798c3c92014-01-21 10:30:08 +0100971 if (conn->flags & CO_FL_ERROR) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100972 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200973 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100974 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100975 goto out_wakeup;
976 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200977 }
978
979 /* OK, so now we know that some data might have been sent, and that we may
980 * have to poll first. We have to do that too if the buffer is not empty.
981 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100982 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200983 /* the connection is established but we can't write. Either the
984 * buffer is empty, or we just refrain from sending because the
985 * ->o limit was reached. Maybe we just wrote the last
986 * chunk and need to close.
987 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200988 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100989 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200990 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200991 (si->state == SI_ST_EST)) {
992 si_shutw(si);
993 goto out_wakeup;
994 }
995
Willy Tarreauafc8a222014-11-28 15:46:27 +0100996 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200997 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100998 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200999 }
1000 else {
1001 /* Otherwise there are remaining data to be sent in the buffer,
1002 * which means we have to poll before doing so.
1003 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001004 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001005 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001006 if (!tick_isset(oc->wex))
1007 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001008 }
1009
Willy Tarreauafc8a222014-11-28 15:46:27 +01001010 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1011 struct channel *ic = si_ic(si);
1012
Willy Tarreaude5722c2012-08-20 15:01:10 +02001013 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001014 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1015 !channel_is_empty(oc))
1016 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001017
Willy Tarreauafc8a222014-11-28 15:46:27 +01001018 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001019 /* Note: to prevent the client from expiring read timeouts
1020 * during writes, we refresh it. We only do this if the
1021 * interface is not configured for "independent streams",
1022 * because for some applications it's better not to do this,
1023 * for instance when continuously exchanging small amounts
1024 * of data which can full the socket buffers long before a
1025 * write timeout is detected.
1026 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001027 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001028 }
1029 }
1030
1031 /* in case of special condition (error, shutdown, end of write...), we
1032 * have to notify the task.
1033 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001034 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1035 ((oc->flags & CF_WAKE_WRITE) &&
1036 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreaue6300be2014-01-25 02:33:21 +01001037 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001038 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001039 if (!(si->flags & SI_FL_DONT_WAKE))
1040 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001041 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001042
1043 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001044 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001045}
1046
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001047/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001048 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001049 * into the buffer from the connection. It iterates over the transport layer's
1050 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001051 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001052static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001053{
Willy Tarreaue603e692012-09-27 22:20:41 +02001054 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001055 struct channel *ic = si_ic(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001056 int ret, max, cur_read;
1057 int read_poll = MAX_READ_POLL_LOOPS;
1058
1059 /* stop immediately on errors. Note that we DON'T want to stop on
1060 * POLL_ERR, as the poller might report a write error while there
1061 * are still data available in the recv buffer. This typically
1062 * happens when we send too large a request to a backend server
1063 * which rejects it before reading it all.
1064 */
1065 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001066 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001067
1068 /* stop here if we reached the end of data */
1069 if (conn_data_read0_pending(conn))
1070 goto out_shutdown_r;
1071
1072 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001073 if (ic->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001074 return;
1075
Willy Tarreau96199b12012-08-24 00:46:52 +02001076 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001077
Willy Tarreauafc8a222014-11-28 15:46:27 +01001078 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !ic->buf->o &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001079 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001080 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001081 /* The buffer was empty and nothing was transferred for more
1082 * than one second. This was caused by a pause and not by
1083 * congestion. Reset any streaming mode to reduce latency.
1084 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001085 ic->xfer_small = 0;
1086 ic->xfer_large = 0;
1087 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001088 }
1089
Willy Tarreau96199b12012-08-24 00:46:52 +02001090 /* First, let's see if we may splice data across the channel without
1091 * using a buffer.
1092 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001093 if (conn->xprt->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001094 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1095 ic->flags & CF_KERN_SPLICING) {
1096 if (buffer_not_empty(ic->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001097 /* We're embarrassed, there are already data pending in
1098 * the buffer and we don't want to have them at two
1099 * locations at a time. Let's indicate we need some
1100 * place and ask the consumer to hurry.
1101 */
1102 goto abort_splice;
1103 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001104
Willy Tarreauafc8a222014-11-28 15:46:27 +01001105 if (unlikely(ic->pipe == NULL)) {
1106 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1107 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001108 goto abort_splice;
1109 }
1110 }
1111
Willy Tarreauafc8a222014-11-28 15:46:27 +01001112 ret = conn->xprt->rcv_pipe(conn, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001113 if (ret < 0) {
1114 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001115 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001116 goto abort_splice;
1117 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001118
Willy Tarreau96199b12012-08-24 00:46:52 +02001119 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001120 if (ic->to_forward != CHN_INFINITE_FORWARD)
1121 ic->to_forward -= ret;
1122 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001123 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001124 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001125 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001126
1127 if (conn_data_read0_pending(conn))
1128 goto out_shutdown_r;
1129
1130 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001131 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001132
Willy Tarreau61d39a02013-07-18 21:49:32 +02001133 if (conn->flags & CO_FL_WAIT_ROOM) {
1134 /* the pipe is full or we have read enough data that it
1135 * could soon be full. Let's stop before needing to poll.
1136 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001137 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001138 __conn_data_stop_recv(conn);
1139 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001140
Willy Tarreauce323de2012-08-20 21:41:06 +02001141 /* splice not possible (anymore), let's go on on standard copy */
1142 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001143
1144 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001145 if (ic->pipe && unlikely(!ic->pipe->data)) {
1146 put_pipe(ic->pipe);
1147 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001148 }
1149
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001150 /* now we'll need a buffer */
Willy Tarreau87b09662015-04-03 00:22:06 +02001151 if (!stream_alloc_recv_buffer(ic)) {
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001152 si->flags |= SI_FL_WAIT_ROOM;
1153 goto end_recv;
1154 }
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 */
Willy Tarreau310987a2014-01-22 19:46:33 +01001161 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 +01001162 max = channel_recv_max(ic);
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 Tarreauafc8a222014-11-28 15:46:27 +01001169 ret = conn->xprt->rcv_buf(conn, ic->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 Tarreauafc8a222014-11-28 15:46:27 +01001176 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001177 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001178 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1179 if (fwd > ic->to_forward)
1180 fwd = ic->to_forward;
1181 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001182 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001183 b_adv(ic->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001184 }
1185
Willy Tarreauafc8a222014-11-28 15:46:27 +01001186 ic->flags |= CF_READ_PARTIAL;
1187 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001188
Willy Tarreauafc8a222014-11-28 15:46:27 +01001189 if (!channel_may_recv(ic)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001190 si->flags |= SI_FL_WAIT_ROOM;
1191 break;
1192 }
1193
Willy Tarreauafc8a222014-11-28 15:46:27 +01001194 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001195 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001196 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001197 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001198 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001199
1200 /* if too many bytes were missing from last read, it means that
1201 * it's pointless trying to read again because the system does
1202 * not have them in buffers.
1203 */
1204 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001205 /* if a streamer has read few data, it may be because we
1206 * have exhausted system buffers. It's not worth trying
1207 * again.
1208 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001209 if (ic->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001210 break;
1211
1212 /* if we read a large block smaller than what we requested,
1213 * it's almost certain we'll never get anything more.
1214 */
1215 if (ret >= global.tune.recv_enough)
1216 break;
1217 }
1218 } /* while !flags */
1219
Willy Tarreauc5890e62014-02-09 17:47:01 +01001220 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001221 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
1222 (cur_read <= ic->buf->size / 2)) {
1223 ic->xfer_large = 0;
1224 ic->xfer_small++;
1225 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001226 /* we have read less than half of the buffer in
1227 * one pass, and this happened at least 3 times.
1228 * This is definitely not a streamer.
1229 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001230 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001231 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001232 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001233 /* if the buffer has been at least half full twice,
1234 * we receive faster than we send, so at least it
1235 * is not a "fast streamer".
1236 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001237 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001238 }
1239 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001240 else if (!(ic->flags & CF_STREAMER_FAST) &&
1241 (cur_read >= ic->buf->size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001242 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001243 ic->xfer_small = 0;
1244 ic->xfer_large++;
1245 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001246 /* we call this buffer a fast streamer if it manages
1247 * to be filled in one call 3 consecutive times.
1248 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001249 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001250 }
1251 }
1252 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001253 ic->xfer_small = 0;
1254 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001255 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001256 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001257 }
1258
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001259 end_recv:
1260 if (conn->flags & CO_FL_ERROR)
1261 return;
1262
Willy Tarreauce323de2012-08-20 21:41:06 +02001263 if (conn_data_read0_pending(conn))
1264 /* connection closed */
1265 goto out_shutdown_r;
1266
1267 return;
1268
1269 out_shutdown_r:
1270 /* we received a shutdown */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001271 ic->flags |= CF_READ_NULL;
1272 if (ic->flags & CF_AUTO_CLOSE)
1273 channel_shutw_now(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001274 stream_sock_read0(si);
1275 conn_data_read0(conn);
1276 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001277}
1278
1279/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001280 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001281 * from the buffer to the connection. It iterates over the transport layer's
1282 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001283 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001284static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001285{
Willy Tarreaue603e692012-09-27 22:20:41 +02001286 struct stream_interface *si = conn->owner;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001287
1288 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001289 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001290
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001291 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001292 /* a handshake was requested */
1293 return;
1294
1295 /* we might have been called just after an asynchronous shutw */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001296 if (si_oc(si)->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001297 return;
1298
1299 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001300 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001301
1302 /* OK all done */
1303 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001304}
1305
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001306/*
1307 * This function propagates a null read received on a socket-based connection.
1308 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001309 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001310 */
1311void stream_sock_read0(struct stream_interface *si)
1312{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001313 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001314 struct channel *ic = si_ic(si);
1315 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001316
Willy Tarreauafc8a222014-11-28 15:46:27 +01001317 ic->flags &= ~CF_SHUTR_NOW;
1318 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001319 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001320 ic->flags |= CF_SHUTR;
1321 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001322 si->flags &= ~SI_FL_WAIT_ROOM;
1323
1324 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1325 return;
1326
Willy Tarreauafc8a222014-11-28 15:46:27 +01001327 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001328 goto do_close;
1329
1330 if (si->flags & SI_FL_NOHALF) {
1331 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001332 /* force flag on ssl to keep stream in cache */
Willy Tarreau1398aa12015-03-12 23:04:07 +01001333 conn_data_shutw_hard(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001334 goto do_close;
1335 }
1336
1337 /* otherwise that's just a normal read shutdown */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001338 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001339 return;
1340
1341 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001342 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001343 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001344
Willy Tarreauafc8a222014-11-28 15:46:27 +01001345 ic->flags &= ~CF_SHUTR_NOW;
1346 ic->flags |= CF_SHUTR;
1347 ic->rex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001348
Willy Tarreauafc8a222014-11-28 15:46:27 +01001349 oc->flags &= ~CF_SHUTW_NOW;
1350 oc->flags |= CF_SHUTW;
1351 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001352
1353 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1354
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001355 si->state = SI_ST_DIS;
1356 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001357 return;
1358}
1359
Willy Tarreaudded32d2008-11-30 19:48:07 +01001360/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001361 * Local variables:
1362 * c-indent-level: 8
1363 * c-basic-offset: 8
1364 * End:
1365 */