blob: b8ed071aef458972beaf15ac767cc756ce14a5d8 [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 Tarreau0a23bcb2013-12-01 11:31:38 +0100380 appctx = si_alloc_appctx(si);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100381 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100382 return NULL;
383
384 appctx_set_applet(appctx, app);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200385 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100386 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200387}
388
Willy Tarreaufb90d942009-09-05 20:57:35 +0200389/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100390 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200391 */
392void stream_int_unregister_handler(struct stream_interface *si)
393{
Willy Tarreau32e3c6a2013-10-11 19:34:20 +0200394 si_detach(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200395}
396
Willy Tarreau2c6be842012-07-06 17:12:34 +0200397/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200398 * established. It returns 0 if it fails in a fatal way or needs to poll to go
399 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200400 * flags (the bit is provided in <flag> by the caller). It is designed to be
401 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200402 * Note that it can emit a PROXY line by relying on the other end's address
403 * when the connection is attached to a stream interface, or by resolving the
404 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200405 */
406int conn_si_send_proxy(struct connection *conn, unsigned int flag)
407{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200408 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200409 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200410 goto out_error;
411
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100412 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200413 goto out_error;
414
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415 /* If we have a PROXY line to send, we'll use this to validate the
416 * connection, in which case the connection is validated only once
417 * we've sent the whole proxy line. Otherwise we use connect().
418 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200419 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200420 int ret;
421
422 /* The target server expects a PROXY line to be sent first.
423 * If the send_proxy_ofs is negative, it corresponds to the
424 * offset to start sending from then end of the proxy string
425 * (which is recomputed every time since it's constant). If
426 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200427 * We can only send a "normal" PROXY line when the connection
428 * is attached to a stream interface. Otherwise we can only
429 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200430 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200431 if (conn->data == &si_conn_cb) {
432 struct stream_interface *si = conn->owner;
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100433 struct connection *remote = objt_conn(si_opposite(si)->end);
David Safb76832014-05-08 23:42:08 -0400434 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), remote);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200435 }
436 else {
437 /* The target server expects a LOCAL line to be sent first. Retrieving
438 * local or remote addresses may fail until the connection is established.
439 */
440 conn_get_from_addr(conn);
441 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
442 goto out_wait;
443
444 conn_get_to_addr(conn);
445 if (!(conn->flags & CO_FL_ADDR_TO_SET))
446 goto out_wait;
447
David Safb76832014-05-08 23:42:08 -0400448 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200449 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200450
Willy Tarreau2c6be842012-07-06 17:12:34 +0200451 if (!ret)
452 goto out_error;
453
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200454 if (conn->send_proxy_ofs > 0)
455 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200456
Willy Tarreaua1a74742012-08-24 12:14:49 +0200457 /* we have to send trash from (ret+sp for -sp bytes). If the
458 * data layer has a pending write, we'll also set MSG_MORE.
459 */
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100460 ret = conn_sock_send(conn, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
461 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200462
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100463 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200464 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200465
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200466 conn->send_proxy_ofs += ret; /* becomes zero once complete */
467 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200468 goto out_wait;
469
470 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100471 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200472 }
473
Willy Tarreaua1a74742012-08-24 12:14:49 +0200474 /* The connection is ready now, simply return and let the connection
475 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200476 */
477 if (conn->flags & CO_FL_WAIT_L4_CONN)
478 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200479 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200480 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200481
482 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200483 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200484 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200485 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200486
487 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200488 __conn_sock_stop_recv(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200489 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200490}
491
Willy Tarreau27375622013-12-17 00:00:28 +0100492
493/* Tiny I/O callback called on recv/send I/O events on idle connections.
494 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
495 * is notified and can kill the connection.
496 */
497static void si_idle_conn_null_cb(struct connection *conn)
498{
Willy Tarreaud85c4852015-03-13 00:40:28 +0100499 conn_sock_drain(conn);
Willy Tarreau27375622013-12-17 00:00:28 +0100500}
501
502/* Callback to be used by connection I/O handlers when some activity is detected
503 * on an idle server connection. Its main purpose is to kill the connection once
504 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
505 * it killed the connection.
506 */
507static int si_idle_conn_wake_cb(struct connection *conn)
508{
509 struct stream_interface *si = conn->owner;
510
511 if (!conn_ctrl_ready(conn))
512 return 0;
513
514 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
515 /* warning, we can't do anything on <conn> after this call ! */
516 conn_force_close(conn);
517 conn_free(conn);
518 si->end = NULL;
519 return -1;
520 }
521 return 0;
522}
523
Willy Tarreau100c4672012-08-20 12:06:26 +0200524/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200525 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200526 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200527 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200528 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200529 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200530static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200531{
Willy Tarreaue603e692012-09-27 22:20:41 +0200532 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100533 struct channel *ic = si_ic(si);
534 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200535
Willy Tarreauafc8a222014-11-28 15:46:27 +0100536 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufd31e532012-07-23 18:24:25 +0200537 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100538 si, si->state, ic->flags, oc->flags);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200539
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200540 if (conn->flags & CO_FL_ERROR)
541 si->flags |= SI_FL_ERR;
542
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200543 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200544 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200545 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100546 oc->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200547 }
548
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200549 /* process consumer side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100550 if (channel_is_empty(oc)) {
551 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200552 (si->state == SI_ST_EST))
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200553 stream_int_shutw_conn(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200554 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100555 oc->wex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200556 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200557
Willy Tarreauafc8a222014-11-28 15:46:27 +0100558 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200559 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200560
Willy Tarreauafc8a222014-11-28 15:46:27 +0100561 if (oc->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200562 /* update timeouts if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100563 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
564 !channel_is_empty(oc))
565 if (tick_isset(oc->wex))
566 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200567
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200568 if (!(si->flags & SI_FL_INDEP_STR))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100569 if (tick_isset(ic->rex))
570 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200571
Willy Tarreauafc8a222014-11-28 15:46:27 +0100572 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
573 channel_may_recv(oc) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100574 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
575 si_chk_rcv(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200576 }
577
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200578 /* process producer side.
579 * We might have some data the consumer is waiting for.
580 * We can do fast-forwarding, but we avoid doing this for partial
581 * buffers, because it is very likely that it will be done again
582 * immediately afterwards once the following data is parsed (eg:
583 * HTTP chunking).
584 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100585 if (((ic->flags & CF_READ_PARTIAL) && !channel_is_empty(ic)) &&
586 (ic->pipe /* always try to send spliced data */ ||
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100587 (si_ib(si)->i == 0 && (si_opposite(si)->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100588 int last_len = ic->pipe ? ic->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200589
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100590 si_chk_snd(si_opposite(si));
Willy Tarreaufd31e532012-07-23 18:24:25 +0200591
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200592 /* check if the consumer has freed some space either in the
593 * buffer or in the pipe.
594 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100595 if (channel_may_recv(ic) &&
596 (!last_len || !ic->pipe || ic->pipe->data < last_len))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200597 si->flags &= ~SI_FL_WAIT_ROOM;
598 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200599
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200600 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200601 __conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100602 ic->rex = TICK_ETERNITY;
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200603 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100604 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
605 channel_may_recv(ic)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100606 /* we must re-enable reading if si_chk_snd() has freed some space */
607 __conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100608 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
609 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200610 }
611
612 /* wake the task up only when needed */
613 if (/* changes on the production side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100614 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200615 si->state != SI_ST_EST ||
616 (si->flags & SI_FL_ERR) ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100617 ((ic->flags & CF_READ_PARTIAL) &&
618 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200619
620 /* changes on the consumption side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100621 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
622 ((oc->flags & CF_WRITE_ACTIVITY) &&
623 ((oc->flags & CF_SHUTW) ||
624 ((oc->flags & CF_WAKE_WRITE) &&
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100625 (si_opposite(si)->state != SI_ST_EST ||
Willy Tarreauafc8a222014-11-28 15:46:27 +0100626 (channel_is_empty(oc) && !oc->to_forward)))))) {
Willy Tarreau07373b82014-11-28 12:08:47 +0100627 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200628 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100629 if (ic->flags & CF_READ_ACTIVITY)
630 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau10fc09e2014-11-25 19:46:36 +0100631
Willy Tarreau87b09662015-04-03 00:22:06 +0200632 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200633 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200634}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200635
Willy Tarreau5368d802012-08-21 18:22:06 +0200636/*
637 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800638 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800639 * caller to commit polling changes. The caller should check conn->flags
640 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200641 */
Godbach4f489902013-12-04 17:24:06 +0800642static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200643{
Willy Tarreaue603e692012-09-27 22:20:41 +0200644 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100645 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200646 int ret;
647
Willy Tarreauafc8a222014-11-28 15:46:27 +0100648 if (oc->pipe && conn->xprt->snd_pipe) {
649 ret = conn->xprt->snd_pipe(conn, oc->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200650 if (ret > 0)
Willy Tarreauafc8a222014-11-28 15:46:27 +0100651 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200652
Willy Tarreauafc8a222014-11-28 15:46:27 +0100653 if (!oc->pipe->data) {
654 put_pipe(oc->pipe);
655 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200656 }
657
Willy Tarreau96199b12012-08-24 00:46:52 +0200658 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800659 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200660 }
661
662 /* At this point, the pipe is empty, but we may still have data pending
663 * in the normal buffer.
664 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100665 if (!oc->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800666 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200667
Godbache68e02d2013-10-11 15:48:29 +0800668 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200669 * data left, and that there are sendable buffered data.
670 */
Willy Tarreau310987a2014-01-22 19:46:33 +0100671 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 +0200672 /* check if we want to inform the kernel that we're interested in
673 * sending more data after this call. We want this if :
674 * - we're about to close after this last send and want to merge
675 * the ongoing FIN with the last segment.
676 * - we know we can't send everything at once and must get back
677 * here because of unaligned data
678 * - there is still a finite amount of data to forward
679 * The test is arranged so that the most common case does only 2
680 * tests.
681 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100682 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200683
Willy Tarreauafc8a222014-11-28 15:46:27 +0100684 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
685 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
686 (oc->flags & CF_EXPECT_MORE))) ||
687 ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100688 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200689
Willy Tarreauafc8a222014-11-28 15:46:27 +0100690 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100691 send_flag |= CO_SFL_STREAMER;
692
Willy Tarreauafc8a222014-11-28 15:46:27 +0100693 ret = conn->xprt->snd_buf(conn, oc->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800694 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100695 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200696
Willy Tarreauafc8a222014-11-28 15:46:27 +0100697 if (!oc->buf->o) {
Godbache68e02d2013-10-11 15:48:29 +0800698 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100699 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800700 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200701
Godbache68e02d2013-10-11 15:48:29 +0800702 /* if some data remain in the buffer, it's only because the
703 * system buffers are full, we will try next time.
704 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200705 }
Godbache68e02d2013-10-11 15:48:29 +0800706 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200707}
708
709
Willy Tarreau100c4672012-08-20 12:06:26 +0200710/* Updates the timers and flags of a stream interface attached to a connection,
711 * depending on the buffers' flags. It should only be called once after the
712 * buffer flags have settled down, and before they are cleared. It doesn't
713 * harm to call it as often as desired (it just slightly hurts performance).
714 * It is only meant to be called by upper layers after buffer flags have been
715 * manipulated by analysers.
716 */
717void stream_int_update_conn(struct stream_interface *si)
718{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100719 struct channel *ic = si_ic(si);
720 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200721 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200722
Willy Tarreau100c4672012-08-20 12:06:26 +0200723 /* Check if we need to close the read side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100724 if (!(ic->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200725 /* Read not closed, update FD status and timeout for reads */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100726 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200727 /* stop reading */
728 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100729 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200730 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200731 conn_data_stop_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100732 ic->rex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200733 }
734 }
735 else {
736 /* (re)start reading and update timeout. Note: we don't recompute the timeout
737 * everytime we get here, otherwise it would risk never to expire. We only
738 * update it if is was not yet set. The stream socket handler will already
739 * have updated it if there has been a completed I/O.
740 */
741 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200742 conn_data_want_recv(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100743 if (!(ic->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ic->rex))
744 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200745 }
746 }
747
748 /* Check if we need to close the write side */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100749 if (!(oc->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200750 /* Write not closed, update FD status and timeout for writes */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100751 if (channel_is_empty(oc)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200752 /* stop writing */
753 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +0100754 if ((oc->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200755 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200756 conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100757 oc->wex = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200758 }
759 }
760 else {
761 /* (re)start writing and update timeout. Note: we don't recompute the timeout
762 * everytime we get here, otherwise it would risk never to expire. We only
763 * update it if is was not yet set. The stream socket handler will already
764 * have updated it if there has been a completed I/O.
765 */
766 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200767 conn_data_want_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100768 if (!tick_isset(oc->wex)) {
769 oc->wex = tick_add_ifset(now_ms, oc->wto);
770 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200771 /* Note: depending on the protocol, we don't know if we're waiting
772 * for incoming data or not. So in order to prevent the socket from
773 * expiring read timeouts during writes, we refresh the read timeout,
774 * except if it was already infinite or if we have explicitly setup
775 * independent streams.
776 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100777 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau100c4672012-08-20 12:06:26 +0200778 }
779 }
780 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200781 }
782}
783
784/*
785 * This function performs a shutdown-read on a stream interface attached to
786 * a connection in a connected or init state (it does nothing for other
787 * states). It either shuts the read side or marks itself as closed. The buffer
788 * flags are updated to reflect the new state. If the stream interface has
789 * SI_FL_NOHALF, we also forward the close to the write side. If a control
790 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200791 * descriptors are then shutdown or closed accordingly. The function
792 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200793 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200794static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200795{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200796 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100797 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200798
Willy Tarreauafc8a222014-11-28 15:46:27 +0100799 ic->flags &= ~CF_SHUTR_NOW;
800 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200801 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100802 ic->flags |= CF_SHUTR;
803 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200804 si->flags &= ~SI_FL_WAIT_ROOM;
805
806 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200807 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200808
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100809 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200810 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200811 si->state = SI_ST_DIS;
812 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200813 }
814 else if (si->flags & SI_FL_NOHALF) {
815 /* we want to immediately forward this close to the write side */
816 return stream_int_shutw_conn(si);
817 }
818 else if (conn->ctrl) {
819 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200820 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200821 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200822}
823
824/*
825 * This function performs a shutdown-write on a stream interface attached to
826 * a connection in a connected or init state (it does nothing for other
827 * states). It either shuts the write side or marks itself as closed. The
828 * buffer flags are updated to reflect the new state. It does also close
829 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +0100830 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200831 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200832static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200833{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200834 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100835 struct channel *ic = si_ic(si);
836 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200837
Willy Tarreauafc8a222014-11-28 15:46:27 +0100838 oc->flags &= ~CF_SHUTW_NOW;
839 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200840 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100841 oc->flags |= CF_SHUTW;
842 oc->wex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200843 si->flags &= ~SI_FL_WAIT_DATA;
844
845 switch (si->state) {
846 case SI_ST_EST:
847 /* we have to shut before closing, otherwise some short messages
848 * may never leave the system, especially when there are remaining
849 * unread data in the socket input buffer, or when nolinger is set.
850 * However, if SI_FL_NOLINGER is explicitly set, we know there is
851 * no risk so we close both sides immediately.
852 */
853 if (si->flags & SI_FL_ERR) {
854 /* quick close, the socket is alredy shut anyway */
855 }
856 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200857 /* unclean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100858 conn_data_shutw_hard(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200859 }
860 else {
861 /* clean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100862 conn_data_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200863
864 /* If the stream interface is configured to disable half-open
865 * connections, we'll skip the shutdown(), but only if the
866 * read size is already closed. Otherwise we can't support
867 * closed write with pending read (eg: abortonclose while
868 * waiting for the server).
869 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100870 if (!(si->flags & SI_FL_NOHALF) || !(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200871 /* We shutdown transport layer */
Willy Tarreau4dfd54f2015-03-12 22:44:53 +0100872 conn_sock_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200873
Willy Tarreauafc8a222014-11-28 15:46:27 +0100874 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200875 /* OK just a shutw, but we want the caller
876 * to disable polling on this FD if exists.
877 */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100878 conn_cond_update_polling(conn);
Willy Tarreau6fe15412013-09-29 15:16:03 +0200879 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200880 }
881 }
882 }
883
884 /* fall through */
885 case SI_ST_CON:
886 /* we may have to close a pending connection, and mark the
887 * response buffer as shutr
888 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200889 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200890 /* fall through */
891 case SI_ST_CER:
892 case SI_ST_QUE:
893 case SI_ST_TAR:
894 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200895 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200896 default:
897 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100898 ic->flags &= ~CF_SHUTR_NOW;
899 ic->flags |= CF_SHUTR;
900 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200901 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200902 }
903}
904
Willy Tarreau46a8d922012-08-20 12:38:36 +0200905/* This function is used for inter-stream-interface calls. It is called by the
906 * consumer to inform the producer side that it may be interested in checking
907 * for free space in the buffer. Note that it intentionally does not update
908 * timeouts, so that we can still check them later at wake-up. This function is
909 * dedicated to connection-based stream interfaces.
910 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200911static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200912{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100913 struct channel *ic = si_ic(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200914 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200915
Willy Tarreauafc8a222014-11-28 15:46:27 +0100916 if (unlikely(si->state > SI_ST_EST || (ic->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200917 return;
918
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200919 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100920
Willy Tarreauafc8a222014-11-28 15:46:27 +0100921 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200922 /* stop reading */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100923 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200924 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200925 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200926 }
927 else {
928 /* (re)start reading */
929 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200930 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200931 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200932 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200933}
934
935
Willy Tarreaude5722c2012-08-20 15:01:10 +0200936/* This function is used for inter-stream-interface calls. It is called by the
937 * producer to inform the consumer side that it may be interested in checking
938 * for data in the buffer. Note that it intentionally does not update timeouts,
939 * so that we can still check them later at wake-up.
940 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200941static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200942{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100943 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200944 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200945
Willy Tarreauafc8a222014-11-28 15:46:27 +0100946 if (unlikely(si->state > SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200947 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200948
Willy Tarreauafc8a222014-11-28 15:46:27 +0100949 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200950 return;
951
Willy Tarreauafc8a222014-11-28 15:46:27 +0100952 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100953 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200954 return;
955
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200956 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200957 /* already subscribed to write notifications, will be called
958 * anyway, so let's avoid calling it especially if the reader
959 * is not ready.
960 */
961 return;
962 }
963
Willy Tarreau708e7172014-01-21 10:27:49 +0100964 /* Before calling the data-level operations, we have to prepare
965 * the polling flags to ensure we properly detect changes.
966 */
967 conn_refresh_polling_flags(conn);
968 __conn_data_want_send(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100969
Willy Tarreau708e7172014-01-21 10:27:49 +0100970 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200971 si_conn_send(conn);
Willy Tarreau798c3c92014-01-21 10:30:08 +0100972 if (conn->flags & CO_FL_ERROR) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100973 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200974 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100975 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100976 goto out_wakeup;
977 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200978 }
979
980 /* OK, so now we know that some data might have been sent, and that we may
981 * have to poll first. We have to do that too if the buffer is not empty.
982 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100983 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200984 /* the connection is established but we can't write. Either the
985 * buffer is empty, or we just refrain from sending because the
986 * ->o limit was reached. Maybe we just wrote the last
987 * chunk and need to close.
988 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200989 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100990 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200991 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200992 (si->state == SI_ST_EST)) {
993 si_shutw(si);
994 goto out_wakeup;
995 }
996
Willy Tarreauafc8a222014-11-28 15:46:27 +0100997 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200998 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100999 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001000 }
1001 else {
1002 /* Otherwise there are remaining data to be sent in the buffer,
1003 * which means we have to poll before doing so.
1004 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001005 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001006 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001007 if (!tick_isset(oc->wex))
1008 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001009 }
1010
Willy Tarreauafc8a222014-11-28 15:46:27 +01001011 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1012 struct channel *ic = si_ic(si);
1013
Willy Tarreaude5722c2012-08-20 15:01:10 +02001014 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001015 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1016 !channel_is_empty(oc))
1017 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001018
Willy Tarreauafc8a222014-11-28 15:46:27 +01001019 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001020 /* Note: to prevent the client from expiring read timeouts
1021 * during writes, we refresh it. We only do this if the
1022 * interface is not configured for "independent streams",
1023 * because for some applications it's better not to do this,
1024 * for instance when continuously exchanging small amounts
1025 * of data which can full the socket buffers long before a
1026 * write timeout is detected.
1027 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001028 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001029 }
1030 }
1031
1032 /* in case of special condition (error, shutdown, end of write...), we
1033 * have to notify the task.
1034 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001035 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1036 ((oc->flags & CF_WAKE_WRITE) &&
1037 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreaue6300be2014-01-25 02:33:21 +01001038 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001039 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001040 if (!(si->flags & SI_FL_DONT_WAKE))
1041 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001042 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001043
1044 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001045 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001046}
1047
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001048/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001049 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001050 * into the buffer from the connection. It iterates over the transport layer's
1051 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001052 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001053static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001054{
Willy Tarreaue603e692012-09-27 22:20:41 +02001055 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001056 struct channel *ic = si_ic(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001057 int ret, max, cur_read;
1058 int read_poll = MAX_READ_POLL_LOOPS;
1059
1060 /* stop immediately on errors. Note that we DON'T want to stop on
1061 * POLL_ERR, as the poller might report a write error while there
1062 * are still data available in the recv buffer. This typically
1063 * happens when we send too large a request to a backend server
1064 * which rejects it before reading it all.
1065 */
1066 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001067 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001068
1069 /* stop here if we reached the end of data */
1070 if (conn_data_read0_pending(conn))
1071 goto out_shutdown_r;
1072
1073 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001074 if (ic->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001075 return;
1076
Willy Tarreau96199b12012-08-24 00:46:52 +02001077 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001078
Willy Tarreauafc8a222014-11-28 15:46:27 +01001079 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !ic->buf->o &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001080 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001081 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001082 /* The buffer was empty and nothing was transferred for more
1083 * than one second. This was caused by a pause and not by
1084 * congestion. Reset any streaming mode to reduce latency.
1085 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001086 ic->xfer_small = 0;
1087 ic->xfer_large = 0;
1088 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001089 }
1090
Willy Tarreau96199b12012-08-24 00:46:52 +02001091 /* First, let's see if we may splice data across the channel without
1092 * using a buffer.
1093 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001094 if (conn->xprt->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001095 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1096 ic->flags & CF_KERN_SPLICING) {
1097 if (buffer_not_empty(ic->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001098 /* We're embarrassed, there are already data pending in
1099 * the buffer and we don't want to have them at two
1100 * locations at a time. Let's indicate we need some
1101 * place and ask the consumer to hurry.
1102 */
1103 goto abort_splice;
1104 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001105
Willy Tarreauafc8a222014-11-28 15:46:27 +01001106 if (unlikely(ic->pipe == NULL)) {
1107 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1108 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001109 goto abort_splice;
1110 }
1111 }
1112
Willy Tarreauafc8a222014-11-28 15:46:27 +01001113 ret = conn->xprt->rcv_pipe(conn, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001114 if (ret < 0) {
1115 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001116 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001117 goto abort_splice;
1118 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001119
Willy Tarreau96199b12012-08-24 00:46:52 +02001120 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001121 if (ic->to_forward != CHN_INFINITE_FORWARD)
1122 ic->to_forward -= ret;
1123 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001124 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001125 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001126 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001127
1128 if (conn_data_read0_pending(conn))
1129 goto out_shutdown_r;
1130
1131 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001132 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001133
Willy Tarreau61d39a02013-07-18 21:49:32 +02001134 if (conn->flags & CO_FL_WAIT_ROOM) {
1135 /* the pipe is full or we have read enough data that it
1136 * could soon be full. Let's stop before needing to poll.
1137 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001138 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001139 __conn_data_stop_recv(conn);
1140 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001141
Willy Tarreauce323de2012-08-20 21:41:06 +02001142 /* splice not possible (anymore), let's go on on standard copy */
1143 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001144
1145 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001146 if (ic->pipe && unlikely(!ic->pipe->data)) {
1147 put_pipe(ic->pipe);
1148 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001149 }
1150
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001151 /* now we'll need a buffer */
Willy Tarreau87b09662015-04-03 00:22:06 +02001152 if (!stream_alloc_recv_buffer(ic)) {
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001153 si->flags |= SI_FL_WAIT_ROOM;
1154 goto end_recv;
1155 }
1156
Willy Tarreau61d39a02013-07-18 21:49:32 +02001157 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1158 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1159 * that if such an event is not handled above in splice, it will be handled here by
1160 * recv().
1161 */
Willy Tarreau310987a2014-01-22 19:46:33 +01001162 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 +01001163 max = channel_recv_max(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001164
1165 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001166 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001167 break;
1168 }
1169
Willy Tarreauafc8a222014-11-28 15:46:27 +01001170 ret = conn->xprt->rcv_buf(conn, ic->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001171 if (ret <= 0)
1172 break;
1173
1174 cur_read += ret;
1175
1176 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001177 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001178 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001179 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1180 if (fwd > ic->to_forward)
1181 fwd = ic->to_forward;
1182 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001183 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001184 b_adv(ic->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001185 }
1186
Willy Tarreauafc8a222014-11-28 15:46:27 +01001187 ic->flags |= CF_READ_PARTIAL;
1188 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001189
Willy Tarreauafc8a222014-11-28 15:46:27 +01001190 if (!channel_may_recv(ic)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001191 si->flags |= SI_FL_WAIT_ROOM;
1192 break;
1193 }
1194
Willy Tarreauafc8a222014-11-28 15:46:27 +01001195 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001196 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001197 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001198 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001199 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001200
1201 /* if too many bytes were missing from last read, it means that
1202 * it's pointless trying to read again because the system does
1203 * not have them in buffers.
1204 */
1205 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001206 /* if a streamer has read few data, it may be because we
1207 * have exhausted system buffers. It's not worth trying
1208 * again.
1209 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001210 if (ic->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001211 break;
1212
1213 /* if we read a large block smaller than what we requested,
1214 * it's almost certain we'll never get anything more.
1215 */
1216 if (ret >= global.tune.recv_enough)
1217 break;
1218 }
1219 } /* while !flags */
1220
Willy Tarreauc5890e62014-02-09 17:47:01 +01001221 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001222 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
1223 (cur_read <= ic->buf->size / 2)) {
1224 ic->xfer_large = 0;
1225 ic->xfer_small++;
1226 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001227 /* we have read less than half of the buffer in
1228 * one pass, and this happened at least 3 times.
1229 * This is definitely not a streamer.
1230 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001231 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001232 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001233 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001234 /* if the buffer has been at least half full twice,
1235 * we receive faster than we send, so at least it
1236 * is not a "fast streamer".
1237 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001238 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001239 }
1240 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001241 else if (!(ic->flags & CF_STREAMER_FAST) &&
1242 (cur_read >= ic->buf->size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001243 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001244 ic->xfer_small = 0;
1245 ic->xfer_large++;
1246 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001247 /* we call this buffer a fast streamer if it manages
1248 * to be filled in one call 3 consecutive times.
1249 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001250 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001251 }
1252 }
1253 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001254 ic->xfer_small = 0;
1255 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001256 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001257 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001258 }
1259
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001260 end_recv:
1261 if (conn->flags & CO_FL_ERROR)
1262 return;
1263
Willy Tarreauce323de2012-08-20 21:41:06 +02001264 if (conn_data_read0_pending(conn))
1265 /* connection closed */
1266 goto out_shutdown_r;
1267
1268 return;
1269
1270 out_shutdown_r:
1271 /* we received a shutdown */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001272 ic->flags |= CF_READ_NULL;
1273 if (ic->flags & CF_AUTO_CLOSE)
1274 channel_shutw_now(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001275 stream_sock_read0(si);
1276 conn_data_read0(conn);
1277 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001278}
1279
1280/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001281 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001282 * from the buffer to the connection. It iterates over the transport layer's
1283 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001284 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001285static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001286{
Willy Tarreaue603e692012-09-27 22:20:41 +02001287 struct stream_interface *si = conn->owner;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001288
1289 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001290 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001291
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001292 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001293 /* a handshake was requested */
1294 return;
1295
1296 /* we might have been called just after an asynchronous shutw */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001297 if (si_oc(si)->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001298 return;
1299
1300 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001301 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001302
1303 /* OK all done */
1304 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001305}
1306
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001307/*
1308 * This function propagates a null read received on a socket-based connection.
1309 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001310 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001311 */
1312void stream_sock_read0(struct stream_interface *si)
1313{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001314 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001315 struct channel *ic = si_ic(si);
1316 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001317
Willy Tarreauafc8a222014-11-28 15:46:27 +01001318 ic->flags &= ~CF_SHUTR_NOW;
1319 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001320 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001321 ic->flags |= CF_SHUTR;
1322 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001323 si->flags &= ~SI_FL_WAIT_ROOM;
1324
1325 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1326 return;
1327
Willy Tarreauafc8a222014-11-28 15:46:27 +01001328 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001329 goto do_close;
1330
1331 if (si->flags & SI_FL_NOHALF) {
1332 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001333 /* force flag on ssl to keep stream in cache */
Willy Tarreau1398aa12015-03-12 23:04:07 +01001334 conn_data_shutw_hard(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001335 goto do_close;
1336 }
1337
1338 /* otherwise that's just a normal read shutdown */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001339 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001340 return;
1341
1342 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001343 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001344 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001345
Willy Tarreauafc8a222014-11-28 15:46:27 +01001346 ic->flags &= ~CF_SHUTR_NOW;
1347 ic->flags |= CF_SHUTR;
1348 ic->rex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001349
Willy Tarreauafc8a222014-11-28 15:46:27 +01001350 oc->flags &= ~CF_SHUTW_NOW;
1351 oc->flags |= CF_SHUTW;
1352 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001353
1354 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1355
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001356 si->state = SI_ST_DIS;
1357 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001358 return;
1359}
1360
Willy Tarreaudded32d2008-11-30 19:48:07 +01001361/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001362 * Local variables:
1363 * c-indent-level: 8
1364 * c-basic-offset: 8
1365 * End:
1366 */