blob: 4f93a2e2a7a47f2bf580071cbf69d3814114ddb6 [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 Tarreau8a8d83b2015-04-13 13:24:54 +020030#include <proto/applet.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020031#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020032#include <proto/connection.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020033#include <proto/pipe.h>
Willy Tarreau87b09662015-04-03 00:22:06 +020034#include <proto/stream.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020035#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010036#include <proto/task.h>
37
Willy Tarreaufd31e532012-07-23 18:24:25 +020038#include <types/pipe.h>
39
Willy Tarreauf873d752012-05-11 17:47:17 +020040/* socket functions used when running a stream interface as a task */
Willy 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 Tarreau6fe15412013-09-29 15:16:03 +020045static void stream_int_shutr_conn(struct stream_interface *si);
46static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020047static void stream_int_chk_rcv_conn(struct stream_interface *si);
48static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +020049static void stream_int_shutr_applet(struct stream_interface *si);
50static void stream_int_shutw_applet(struct stream_interface *si);
51static void stream_int_chk_rcv_applet(struct stream_interface *si);
52static void stream_int_chk_snd_applet(struct stream_interface *si);
Willy Tarreau4aa36832012-10-02 20:07:22 +020053static void si_conn_recv_cb(struct connection *conn);
54static void si_conn_send_cb(struct connection *conn);
Willy Tarreau2396c1c2012-10-03 21:12:16 +020055static int si_conn_wake_cb(struct connection *conn);
Willy Tarreau27375622013-12-17 00:00:28 +010056static int si_idle_conn_wake_cb(struct connection *conn);
57static void si_idle_conn_null_cb(struct connection *conn);
Willy Tarreauf873d752012-05-11 17:47:17 +020058
Willy Tarreauc5788912012-08-24 18:12:41 +020059/* stream-interface operations for embedded tasks */
60struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020061 .chk_rcv = stream_int_chk_rcv,
62 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020063 .shutr = stream_int_shutr,
64 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020065};
66
Willy Tarreauc5788912012-08-24 18:12:41 +020067/* stream-interface operations for connections */
68struct si_ops si_conn_ops = {
69 .update = stream_int_update_conn,
70 .chk_rcv = stream_int_chk_rcv_conn,
71 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020072 .shutr = stream_int_shutr_conn,
73 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020074};
75
Willy Tarreaud45b9f82015-04-13 16:30:14 +020076/* stream-interface operations for connections */
77struct si_ops si_applet_ops = {
78 .update = stream_int_update_applet,
79 .chk_rcv = stream_int_chk_rcv_applet,
80 .chk_snd = stream_int_chk_snd_applet,
81 .shutr = stream_int_shutr_applet,
82 .shutw = stream_int_shutw_applet,
83};
84
Willy Tarreau74beec32012-10-03 00:41:04 +020085struct data_cb si_conn_cb = {
Willy Tarreauc5788912012-08-24 18:12:41 +020086 .recv = si_conn_recv_cb,
87 .send = si_conn_send_cb,
Willy Tarreau4aa36832012-10-02 20:07:22 +020088 .wake = si_conn_wake_cb,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +010089 .name = "STRM",
Willy Tarreauc5788912012-08-24 18:12:41 +020090};
91
Willy Tarreau27375622013-12-17 00:00:28 +010092struct data_cb si_idle_conn_cb = {
93 .recv = si_idle_conn_null_cb,
94 .send = si_idle_conn_null_cb,
95 .wake = si_idle_conn_wake_cb,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +010096 .name = "IDLE",
Willy Tarreau27375622013-12-17 00:00:28 +010097};
98
Willy Tarreaucff64112008-11-03 06:26:53 +010099/*
100 * This function only has to be called once after a wakeup event in case of
101 * suspected timeout. It controls the stream interface timeouts and sets
102 * si->flags accordingly. It does NOT close anything, as this timeout may
103 * be used for any purpose. It returns 1 if the timeout fired, otherwise
104 * zero.
105 */
106int stream_int_check_timeouts(struct stream_interface *si)
107{
108 if (tick_is_expired(si->exp, now_ms)) {
109 si->flags |= SI_FL_EXP;
110 return 1;
111 }
112 return 0;
113}
114
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100115/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +0100116void stream_int_report_error(struct stream_interface *si)
117{
118 if (!si->err_type)
119 si->err_type = SI_ET_DATA_ERR;
120
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100121 si_oc(si)->flags |= CF_WRITE_ERROR;
122 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100123}
124
125/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100126 * Returns a message to the client ; the connection is shut down for read,
127 * and the request is cleared so that no server connection can be initiated.
128 * The buffer is marked for read shutdown on the other side to protect the
129 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100130 * "chunk". If it is null, then an empty message is used. The reply buffer does
131 * not need to be empty before this, and its contents will not be overwritten.
132 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100133 */
134void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
135{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100136 struct channel *ic = si_ic(si);
137 struct channel *oc = si_oc(si);
138
139 channel_auto_read(ic);
140 channel_abort(ic);
141 channel_auto_close(ic);
142 channel_erase(ic);
143 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100144
Willy Tarreau148d0992010-01-10 10:21:21 +0100145 if (likely(msg && msg->len))
Willy Tarreauafc8a222014-11-28 15:46:27 +0100146 bo_inject(oc, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100147
Willy Tarreauafc8a222014-11-28 15:46:27 +0100148 oc->wex = tick_add_ifset(now_ms, oc->wto);
149 channel_auto_read(oc);
150 channel_auto_close(oc);
151 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100152}
153
Willy Tarreau4a36b562012-08-06 19:31:45 +0200154/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200155 * This function performs a shutdown-read on a detached stream interface in a
156 * connected or init state (it does nothing for other states). It either shuts
157 * the read side or marks itself as closed. The buffer flags are updated to
158 * reflect the new state. If the stream interface has SI_FL_NOHALF, we also
159 * forward the close to the write side. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200160 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200161static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200162{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100163 struct channel *ic = si_ic(si);
164
165 ic->flags &= ~CF_SHUTR_NOW;
166 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200167 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100168 ic->flags |= CF_SHUTR;
169 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200170 si->flags &= ~SI_FL_WAIT_ROOM;
171
172 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200173 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200174
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100175 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200176 si->state = SI_ST_DIS;
177 si->exp = TICK_ETERNITY;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200178 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200179 else if (si->flags & SI_FL_NOHALF) {
180 /* we want to immediately forward this close to the write side */
181 return stream_int_shutw(si);
182 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200183
Willy Tarreau4a36b562012-08-06 19:31:45 +0200184 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100185 if (!(si->flags & SI_FL_DONT_WAKE))
186 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200187}
188
Willy Tarreau4a36b562012-08-06 19:31:45 +0200189/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200190 * This function performs a shutdown-write on a detached stream interface in a
191 * connected or init state (it does nothing for other states). It either shuts
192 * the write side or marks itself as closed. The buffer flags are updated to
193 * reflect the new state. It does also close everything if the SI was marked as
194 * being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200195 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200196static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200197{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100198 struct channel *ic = si_ic(si);
199 struct channel *oc = si_oc(si);
200
201 oc->flags &= ~CF_SHUTW_NOW;
202 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200203 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100204 oc->flags |= CF_SHUTW;
205 oc->wex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200206 si->flags &= ~SI_FL_WAIT_DATA;
207
208 switch (si->state) {
209 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200210 /* we have to shut before closing, otherwise some short messages
211 * may never leave the system, especially when there are remaining
212 * unread data in the socket input buffer, or when nolinger is set.
213 * However, if SI_FL_NOLINGER is explicitly set, we know there is
214 * no risk so we close both sides immediately.
215 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200216 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100217 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200218 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200219
220 /* fall through */
221 case SI_ST_CON:
222 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100223 case SI_ST_QUE:
224 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200225 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200226 si->state = SI_ST_DIS;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200227 default:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200228 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100229 ic->flags &= ~CF_SHUTR_NOW;
230 ic->flags |= CF_SHUTR;
231 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200232 si->exp = TICK_ETERNITY;
233 }
234
Willy Tarreau4a36b562012-08-06 19:31:45 +0200235 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100236 if (!(si->flags & SI_FL_DONT_WAKE))
237 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200238}
239
240/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200241static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200242{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100243 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244
Willy Tarreauafc8a222014-11-28 15:46:27 +0100245 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200246 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100247 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200248
Willy Tarreauafc8a222014-11-28 15:46:27 +0100249 if (unlikely(si->state != SI_ST_EST || (ic->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250 return;
251
Willy Tarreauafc8a222014-11-28 15:46:27 +0100252 if (!channel_may_recv(ic) || ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200253 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200254 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200255 }
256 else {
257 /* (re)start reading */
258 si->flags &= ~SI_FL_WAIT_ROOM;
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}
263
264/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200265static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200266{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100267 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200268
Willy Tarreauafc8a222014-11-28 15:46:27 +0100269 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200270 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100271 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200272
Willy Tarreauafc8a222014-11-28 15:46:27 +0100273 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200274 return;
275
276 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100277 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200278 return;
279
280 /* Otherwise there are remaining data to be sent in the buffer,
281 * so we tell the handler.
282 */
283 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100284 if (!tick_isset(oc->wex))
285 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200286
Willy Tarreau07373b82014-11-28 12:08:47 +0100287 if (!(si->flags & SI_FL_DONT_WAKE))
288 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200289}
290
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200291/* Register an applet to handle a stream_interface as a new appctx. The SI will
292 * wake it up everytime it is solicited. The appctx must be deleted by the task
293 * handler using si_release_endpoint(), possibly from within the function itself.
294 * It also pre-initializes the applet's context and returns it (or NULL in case
295 * it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200296 */
Willy Tarreau30576452015-04-13 13:50:30 +0200297struct appctx *stream_int_register_handler(struct stream_interface *si, struct applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200298{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100299 struct appctx *appctx;
300
Willy Tarreau07373b82014-11-28 12:08:47 +0100301 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200302
Willy Tarreaua7513f52015-04-05 00:15:26 +0200303 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100304 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100305 return NULL;
306
Willy Tarreaufe127932015-04-21 19:23:39 +0200307 si_applet_cant_get(si);
Willy Tarreau828824a2015-04-19 17:20:03 +0200308 appctx_wakeup(appctx);
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100309 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200310}
311
Willy Tarreau2c6be842012-07-06 17:12:34 +0200312/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200313 * established. It returns 0 if it fails in a fatal way or needs to poll to go
314 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200315 * flags (the bit is provided in <flag> by the caller). It is designed to be
316 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200317 * Note that it can emit a PROXY line by relying on the other end's address
318 * when the connection is attached to a stream interface, or by resolving the
319 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200320 */
321int conn_si_send_proxy(struct connection *conn, unsigned int flag)
322{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200323 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200324 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200325 goto out_error;
326
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100327 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200328 goto out_error;
329
Willy Tarreau2c6be842012-07-06 17:12:34 +0200330 /* If we have a PROXY line to send, we'll use this to validate the
331 * connection, in which case the connection is validated only once
332 * we've sent the whole proxy line. Otherwise we use connect().
333 */
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200334 while (conn->send_proxy_ofs) {
Willy Tarreau2c6be842012-07-06 17:12:34 +0200335 int ret;
336
337 /* The target server expects a PROXY line to be sent first.
338 * If the send_proxy_ofs is negative, it corresponds to the
339 * offset to start sending from then end of the proxy string
340 * (which is recomputed every time since it's constant). If
341 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200342 * We can only send a "normal" PROXY line when the connection
343 * is attached to a stream interface. Otherwise we can only
344 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200345 */
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200346 if (conn->data == &si_conn_cb) {
347 struct stream_interface *si = conn->owner;
Willy Tarreau50fe03b2014-11-28 13:59:31 +0100348 struct connection *remote = objt_conn(si_opposite(si)->end);
David Safb76832014-05-08 23:42:08 -0400349 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), remote);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200350 }
351 else {
352 /* The target server expects a LOCAL line to be sent first. Retrieving
353 * local or remote addresses may fail until the connection is established.
354 */
355 conn_get_from_addr(conn);
356 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
357 goto out_wait;
358
359 conn_get_to_addr(conn);
360 if (!(conn->flags & CO_FL_ADDR_TO_SET))
361 goto out_wait;
362
David Safb76832014-05-08 23:42:08 -0400363 ret = make_proxy_line(trash.str, trash.size, objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200364 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200365
Willy Tarreau2c6be842012-07-06 17:12:34 +0200366 if (!ret)
367 goto out_error;
368
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200369 if (conn->send_proxy_ofs > 0)
370 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200371
Willy Tarreaua1a74742012-08-24 12:14:49 +0200372 /* we have to send trash from (ret+sp for -sp bytes). If the
373 * data layer has a pending write, we'll also set MSG_MORE.
374 */
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100375 ret = conn_sock_send(conn, trash.str + ret + conn->send_proxy_ofs, -conn->send_proxy_ofs,
376 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200377
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100378 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200379 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200380
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200381 conn->send_proxy_ofs += ret; /* becomes zero once complete */
382 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200383 goto out_wait;
384
385 /* OK we've sent the whole line, we're connected */
Willy Tarreau7fe45692013-12-04 23:37:56 +0100386 break;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200387 }
388
Willy Tarreaua1a74742012-08-24 12:14:49 +0200389 /* The connection is ready now, simply return and let the connection
390 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200391 */
392 if (conn->flags & CO_FL_WAIT_L4_CONN)
393 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200394 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200395 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396
397 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200398 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200399 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200400 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200401
402 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200403 __conn_sock_stop_recv(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200404 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200405}
406
Willy Tarreau27375622013-12-17 00:00:28 +0100407
408/* Tiny I/O callback called on recv/send I/O events on idle connections.
409 * It simply sets the CO_FL_SOCK_RD_SH flag so that si_idle_conn_wake_cb()
410 * is notified and can kill the connection.
411 */
412static void si_idle_conn_null_cb(struct connection *conn)
413{
Willy Tarreaud85c4852015-03-13 00:40:28 +0100414 conn_sock_drain(conn);
Willy Tarreau27375622013-12-17 00:00:28 +0100415}
416
417/* Callback to be used by connection I/O handlers when some activity is detected
418 * on an idle server connection. Its main purpose is to kill the connection once
419 * a close was detected on it. It returns 0 if it did nothing serious, or -1 if
420 * it killed the connection.
421 */
422static int si_idle_conn_wake_cb(struct connection *conn)
423{
424 struct stream_interface *si = conn->owner;
425
426 if (!conn_ctrl_ready(conn))
427 return 0;
428
429 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH)) {
430 /* warning, we can't do anything on <conn> after this call ! */
Willy Tarreauc4b56e42015-09-23 17:56:02 +0200431 si_release_endpoint(si);
Willy Tarreau27375622013-12-17 00:00:28 +0100432 return -1;
433 }
434 return 0;
435}
436
Willy Tarreau615f28b2015-09-23 18:40:09 +0200437/* This function is the equivalent to stream_int_update() except that it's
438 * designed to be called from outside the stream handlers, typically the lower
439 * layers (applets, connections) after I/O completion. After updating the stream
440 * interface and timeouts, it will try to forward what can be forwarded, then to
441 * wake the associated task up if an important event requires special handling.
442 * It should not be called from within the stream itself, stream_int_update()
443 * is designed for this.
444 */
445void stream_int_notify(struct stream_interface *si)
446{
447 struct channel *ic = si_ic(si);
448 struct channel *oc = si_oc(si);
449
450 /* process consumer side */
451 if (channel_is_empty(oc)) {
452 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
453 (si->state == SI_ST_EST))
454 si_shutw(si);
455 oc->wex = TICK_ETERNITY;
456 }
457
458 /* indicate that we may be waiting for data from the output channel */
459 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && channel_may_recv(oc))
460 si->flags |= SI_FL_WAIT_DATA;
461
462 /* update OC timeouts and wake the other side up if it's waiting for room */
463 if (oc->flags & CF_WRITE_ACTIVITY) {
464 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
465 !channel_is_empty(oc))
466 if (tick_isset(oc->wex))
467 oc->wex = tick_add_ifset(now_ms, oc->wto);
468
469 if (!(si->flags & SI_FL_INDEP_STR))
470 if (tick_isset(ic->rex))
471 ic->rex = tick_add_ifset(now_ms, ic->rto);
472
473 if (likely((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
474 channel_may_recv(oc) &&
475 (si_opposite(si)->flags & SI_FL_WAIT_ROOM)))
476 si_chk_rcv(si_opposite(si));
477 }
478
479 /* Notify the other side when we've injected data into the IC that
480 * needs to be forwarded. We can do fast-forwarding as soon as there
481 * are output data, but we avoid doing this if some of the data are
482 * not yet scheduled for being forwarded, because it is very likely
483 * that it will be done again immediately afterwards once the following
484 * data are parsed (eg: HTTP chunking). We only SI_FL_WAIT_ROOM once
485 * we've emptied *some* of the output buffer, and not just when there
486 * is available room, because applets are often forced to stop before
487 * the buffer is full. We must not stop based on input data alone because
488 * an HTTP parser might need more data to complete the parsing.
489 */
490 if (!channel_is_empty(ic) &&
491 (si_opposite(si)->flags & SI_FL_WAIT_DATA) &&
492 (ic->buf->i == 0 || ic->pipe)) {
493 int new_len, last_len;
494
495 last_len = ic->buf->o;
496 if (ic->pipe)
497 last_len += ic->pipe->data;
498
499 si_chk_snd(si_opposite(si));
500
501 new_len = ic->buf->o;
502 if (ic->pipe)
503 new_len += ic->pipe->data;
504
505 /* check if the consumer has freed some space either in the
506 * buffer or in the pipe.
507 */
508 if (channel_may_recv(ic) && new_len < last_len)
509 si->flags &= ~SI_FL_WAIT_ROOM;
510 }
511
512 if (si->flags & SI_FL_WAIT_ROOM) {
513 ic->rex = TICK_ETERNITY;
514 }
515 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
516 channel_may_recv(ic)) {
517 /* we must re-enable reading if si_chk_snd() has freed some space */
518 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
519 ic->rex = tick_add_ifset(now_ms, ic->rto);
520 }
521
522 /* wake the task up only when needed */
523 if (/* changes on the production side */
524 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
525 si->state != SI_ST_EST ||
526 (si->flags & SI_FL_ERR) ||
527 ((ic->flags & CF_READ_PARTIAL) &&
528 (!ic->to_forward || si_opposite(si)->state != SI_ST_EST)) ||
529
530 /* changes on the consumption side */
531 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
532 ((oc->flags & CF_WRITE_ACTIVITY) &&
533 ((oc->flags & CF_SHUTW) ||
534 ((oc->flags & CF_WAKE_WRITE) &&
535 (si_opposite(si)->state != SI_ST_EST ||
536 (channel_is_empty(oc) && !oc->to_forward)))))) {
537 task_wakeup(si_task(si), TASK_WOKEN_IO);
538 }
539 if (ic->flags & CF_READ_ACTIVITY)
540 ic->flags &= ~CF_READ_DONTWAIT;
541
542 stream_release_buffers(si_strm(si));
543}
544
545
Willy Tarreau651e1822015-09-23 20:06:13 +0200546/* Callback to be used by connection I/O handlers upon completion. It propagates
547 * connection flags to the stream interface, updates the stream (which may or
548 * may not take this opportunity to try to forward data), then update the
549 * connection's polling based on the channels and stream interface's final
550 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200551 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200552static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200553{
Willy Tarreaue603e692012-09-27 22:20:41 +0200554 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100555 struct channel *ic = si_ic(si);
556 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200557
Willy Tarreau651e1822015-09-23 20:06:13 +0200558 /* First step, report to the stream-int what was detected at the
559 * connection layer : errors and connection establishment.
560 */
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200561 if (conn->flags & CO_FL_ERROR)
562 si->flags |= SI_FL_ERR;
563
Willy Tarreauc76ae332012-07-12 15:32:13 +0200564 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200565 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100566 oc->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200567 }
568
Willy Tarreau651e1822015-09-23 20:06:13 +0200569 /* Second step : update the stream-int and channels, try to forward any
570 * pending data, then possibly wake the stream up based on the new
571 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200572 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200573 stream_int_notify(si);
Willy Tarreauea3cc482015-09-23 19:37:00 +0200574
Willy Tarreau651e1822015-09-23 20:06:13 +0200575 /* Third step : update the connection's polling status based on what
576 * was done above (eg: maybe some buffers got emptied).
577 */
578 if (channel_is_empty(oc))
579 __conn_data_stop_send(conn);
Willy Tarreauea3cc482015-09-23 19:37:00 +0200580
Willy Tarreaufd31e532012-07-23 18:24:25 +0200581
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200582 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200583 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200584 }
Willy Tarreauafc8a222014-11-28 15:46:27 +0100585 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
586 channel_may_recv(ic)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100587 __conn_data_want_recv(conn);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200588 }
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200589 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200590}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200591
Willy Tarreau5368d802012-08-21 18:22:06 +0200592/*
593 * This function is called to send buffer data to a stream socket.
Godbache68e02d2013-10-11 15:48:29 +0800594 * It calls the transport layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800595 * caller to commit polling changes. The caller should check conn->flags
596 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200597 */
Godbach4f489902013-12-04 17:24:06 +0800598static void si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200599{
Willy Tarreaue603e692012-09-27 22:20:41 +0200600 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100601 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200602 int ret;
603
Willy Tarreauafc8a222014-11-28 15:46:27 +0100604 if (oc->pipe && conn->xprt->snd_pipe) {
605 ret = conn->xprt->snd_pipe(conn, oc->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200606 if (ret > 0)
Willy Tarreauafc8a222014-11-28 15:46:27 +0100607 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
Willy Tarreau5368d802012-08-21 18:22:06 +0200608
Willy Tarreauafc8a222014-11-28 15:46:27 +0100609 if (!oc->pipe->data) {
610 put_pipe(oc->pipe);
611 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200612 }
613
Willy Tarreau96199b12012-08-24 00:46:52 +0200614 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +0800615 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200616 }
617
618 /* At this point, the pipe is empty, but we may still have data pending
619 * in the normal buffer.
620 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100621 if (!oc->buf->o)
Godbach4f489902013-12-04 17:24:06 +0800622 return;
Willy Tarreau5368d802012-08-21 18:22:06 +0200623
Godbache68e02d2013-10-11 15:48:29 +0800624 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200625 * data left, and that there are sendable buffered data.
626 */
Willy Tarreau310987a2014-01-22 19:46:33 +0100627 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 +0200628 /* check if we want to inform the kernel that we're interested in
629 * sending more data after this call. We want this if :
630 * - we're about to close after this last send and want to merge
631 * the ongoing FIN with the last segment.
632 * - we know we can't send everything at once and must get back
633 * here because of unaligned data
634 * - there is still a finite amount of data to forward
635 * The test is arranged so that the most common case does only 2
636 * tests.
637 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100638 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200639
Willy Tarreauafc8a222014-11-28 15:46:27 +0100640 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
641 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
642 (oc->flags & CF_EXPECT_MORE))) ||
643 ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100644 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200645
Willy Tarreauafc8a222014-11-28 15:46:27 +0100646 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100647 send_flag |= CO_SFL_STREAMER;
648
Willy Tarreauafc8a222014-11-28 15:46:27 +0100649 ret = conn->xprt->snd_buf(conn, oc->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800650 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->buf->o) {
Godbache68e02d2013-10-11 15:48:29 +0800654 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100655 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800656 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200657
Godbache68e02d2013-10-11 15:48:29 +0800658 /* if some data remain in the buffer, it's only because the
659 * system buffers are full, we will try next time.
660 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200661 }
Godbache68e02d2013-10-11 15:48:29 +0800662 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200663}
664
Willy Tarreau25f13102015-09-24 11:32:22 +0200665/* This function is designed to be called from within the stream handler to
666 * update the channels' expiration timers and the stream interface's flags
667 * based on the channels' flags. It needs to be called only once after the
668 * channels' flags have settled down, and before they are cleared, though it
669 * doesn't harm to call it as often as desired (it just slightly hurts
670 * performance). It must not be called from outside of the stream handler,
671 * as what it does will be used to compute the stream task's expiration.
672 */
673void stream_int_update(struct stream_interface *si)
674{
675 struct channel *ic = si_ic(si);
676 struct channel *oc = si_oc(si);
677
678 if (!(ic->flags & CF_SHUTR)) {
679 /* Read not closed, update FD status and timeout for reads */
680 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
681 /* stop reading */
682 if (!(si->flags & SI_FL_WAIT_ROOM)) {
683 if (!(ic->flags & CF_DONT_READ)) /* full */
684 si->flags |= SI_FL_WAIT_ROOM;
685 ic->rex = TICK_ETERNITY;
686 }
687 }
688 else {
689 /* (re)start reading and update timeout. Note: we don't recompute the timeout
690 * everytime we get here, otherwise it would risk never to expire. We only
691 * update it if is was not yet set. The stream socket handler will already
692 * have updated it if there has been a completed I/O.
693 */
694 si->flags &= ~SI_FL_WAIT_ROOM;
695 if (!(ic->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ic->rex))
696 ic->rex = tick_add_ifset(now_ms, ic->rto);
697 }
698 }
699
700 if (!(oc->flags & CF_SHUTW)) {
701 /* Write not closed, update FD status and timeout for writes */
702 if (channel_is_empty(oc)) {
703 /* stop writing */
704 if (!(si->flags & SI_FL_WAIT_DATA)) {
705 if ((oc->flags & CF_SHUTW_NOW) == 0)
706 si->flags |= SI_FL_WAIT_DATA;
707 oc->wex = TICK_ETERNITY;
708 }
709 }
710 else {
711 /* (re)start writing and update timeout. Note: we don't recompute the timeout
712 * everytime we get here, otherwise it would risk never to expire. We only
713 * update it if is was not yet set. The stream socket handler will already
714 * have updated it if there has been a completed I/O.
715 */
716 si->flags &= ~SI_FL_WAIT_DATA;
717 if (!tick_isset(oc->wex)) {
718 oc->wex = tick_add_ifset(now_ms, oc->wto);
719 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
720 /* Note: depending on the protocol, we don't know if we're waiting
721 * for incoming data or not. So in order to prevent the socket from
722 * expiring read timeouts during writes, we refresh the read timeout,
723 * except if it was already infinite or if we have explicitly setup
724 * independent streams.
725 */
726 ic->rex = tick_add_ifset(now_ms, ic->rto);
727 }
728 }
729 }
730 }
731}
732
Willy Tarreau452c7d52015-09-25 10:39:16 +0200733/* Updates the polling status of a connection outside of the connection handler
734 * based on the channel's flags and the stream interface's flags. It needs to be
735 * called once after the channels' flags have settled down and the stream has
736 * been updated. It is not designed to be called from within the connection
737 * handler itself.
Willy Tarreau100c4672012-08-20 12:06:26 +0200738 */
739void stream_int_update_conn(struct stream_interface *si)
740{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100741 struct channel *ic = si_ic(si);
742 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200743 struct connection *conn = __objt_conn(si->end);
Willy Tarreau100c4672012-08-20 12:06:26 +0200744
Willy Tarreau2f4e7022015-09-25 10:50:59 +0200745 if (!(ic->flags & CF_SHUTR)) {
746 /* Read not closed */
747 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic))
748 __conn_data_stop_recv(conn);
749 else
750 __conn_data_want_recv(conn);
751 }
752
753 if (!(oc->flags & CF_SHUTW)) {
754 /* Write not closed */
755 if (channel_is_empty(oc))
756 __conn_data_stop_send(conn);
757 else
758 __conn_data_want_send(conn);
759 }
760
761 conn_cond_update_data_polling(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200762}
763
764/*
765 * This function performs a shutdown-read on a stream interface attached to
766 * a connection in a connected or init state (it does nothing for other
767 * states). It either shuts the read side or marks itself as closed. The buffer
768 * flags are updated to reflect the new state. If the stream interface has
769 * SI_FL_NOHALF, we also forward the close to the write side. If a control
770 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200771 * descriptors are then shutdown or closed accordingly. The function
772 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200773 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200774static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200775{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200776 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100777 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200778
Willy Tarreauafc8a222014-11-28 15:46:27 +0100779 ic->flags &= ~CF_SHUTR_NOW;
780 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200781 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100782 ic->flags |= CF_SHUTR;
783 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200784 si->flags &= ~SI_FL_WAIT_ROOM;
785
786 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200787 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200788
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100789 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau6fe15412013-09-29 15:16:03 +0200790 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200791 si->state = SI_ST_DIS;
792 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200793 }
794 else if (si->flags & SI_FL_NOHALF) {
795 /* we want to immediately forward this close to the write side */
796 return stream_int_shutw_conn(si);
797 }
798 else if (conn->ctrl) {
799 /* we want the caller to disable polling on this FD */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200800 conn_data_stop_recv(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200801 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200802}
803
804/*
805 * This function performs a shutdown-write on a stream interface attached to
806 * a connection in a connected or init state (it does nothing for other
807 * states). It either shuts the write side or marks itself as closed. The
808 * buffer flags are updated to reflect the new state. It does also close
809 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +0100810 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200811 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200812static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200813{
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200814 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100815 struct channel *ic = si_ic(si);
816 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200817
Willy Tarreauafc8a222014-11-28 15:46:27 +0100818 oc->flags &= ~CF_SHUTW_NOW;
819 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200820 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100821 oc->flags |= CF_SHUTW;
822 oc->wex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200823 si->flags &= ~SI_FL_WAIT_DATA;
824
825 switch (si->state) {
826 case SI_ST_EST:
827 /* we have to shut before closing, otherwise some short messages
828 * may never leave the system, especially when there are remaining
829 * unread data in the socket input buffer, or when nolinger is set.
830 * However, if SI_FL_NOLINGER is explicitly set, we know there is
831 * no risk so we close both sides immediately.
832 */
833 if (si->flags & SI_FL_ERR) {
834 /* quick close, the socket is alredy shut anyway */
835 }
836 else if (si->flags & SI_FL_NOLINGER) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200837 /* unclean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100838 conn_data_shutw_hard(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200839 }
840 else {
841 /* clean data-layer shutdown */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100842 conn_data_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200843
844 /* If the stream interface is configured to disable half-open
845 * connections, we'll skip the shutdown(), but only if the
846 * read size is already closed. Otherwise we can't support
847 * closed write with pending read (eg: abortonclose while
848 * waiting for the server).
849 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100850 if (!(si->flags & SI_FL_NOHALF) || !(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200851 /* We shutdown transport layer */
Willy Tarreau4dfd54f2015-03-12 22:44:53 +0100852 conn_sock_shutw(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200853
Willy Tarreauafc8a222014-11-28 15:46:27 +0100854 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200855 /* OK just a shutw, but we want the caller
856 * to disable polling on this FD if exists.
857 */
Willy Tarreau1398aa12015-03-12 23:04:07 +0100858 conn_cond_update_polling(conn);
Willy Tarreau6fe15412013-09-29 15:16:03 +0200859 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200860 }
861 }
862 }
863
864 /* fall through */
865 case SI_ST_CON:
866 /* we may have to close a pending connection, and mark the
867 * response buffer as shutr
868 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200869 conn_full_close(conn);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200870 /* fall through */
871 case SI_ST_CER:
872 case SI_ST_QUE:
873 case SI_ST_TAR:
874 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +0200875 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200876 default:
877 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100878 ic->flags &= ~CF_SHUTR_NOW;
879 ic->flags |= CF_SHUTR;
880 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200881 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +0200882 }
883}
884
Willy Tarreau46a8d922012-08-20 12:38:36 +0200885/* This function is used for inter-stream-interface calls. It is called by the
886 * consumer to inform the producer side that it may be interested in checking
887 * for free space in the buffer. Note that it intentionally does not update
888 * timeouts, so that we can still check them later at wake-up. This function is
889 * dedicated to connection-based stream interfaces.
890 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200891static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200892{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100893 struct channel *ic = si_ic(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200894 struct connection *conn = __objt_conn(si->end);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200895
Willy Tarreauafc8a222014-11-28 15:46:27 +0100896 if (unlikely(si->state > SI_ST_EST || (ic->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200897 return;
898
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200899 conn_refresh_polling_flags(conn);
Willy Tarreau7d281492012-12-16 19:19:13 +0100900
Willy Tarreauafc8a222014-11-28 15:46:27 +0100901 if ((ic->flags & CF_DONT_READ) || !channel_may_recv(ic)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200902 /* stop reading */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100903 if (!(ic->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200904 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200905 __conn_data_stop_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200906 }
907 else {
908 /* (re)start reading */
909 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200910 __conn_data_want_recv(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200911 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200912 conn_cond_update_data_polling(conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200913}
914
915
Willy Tarreaude5722c2012-08-20 15:01:10 +0200916/* This function is used for inter-stream-interface calls. It is called by the
917 * producer to inform the consumer side that it may be interested in checking
918 * for data in the buffer. Note that it intentionally does not update timeouts,
919 * so that we can still check them later at wake-up.
920 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200921static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200922{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100923 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200924 struct connection *conn = __objt_conn(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200925
Willy Tarreauafc8a222014-11-28 15:46:27 +0100926 if (unlikely(si->state > SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200927 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200928
Willy Tarreauafc8a222014-11-28 15:46:27 +0100929 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200930 return;
931
Willy Tarreauafc8a222014-11-28 15:46:27 +0100932 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100933 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200934 return;
935
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200936 if (conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200937 /* already subscribed to write notifications, will be called
938 * anyway, so let's avoid calling it especially if the reader
939 * is not ready.
940 */
941 return;
942 }
943
Willy Tarreau708e7172014-01-21 10:27:49 +0100944 /* Before calling the data-level operations, we have to prepare
945 * the polling flags to ensure we properly detect changes.
946 */
947 conn_refresh_polling_flags(conn);
948 __conn_data_want_send(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100949
Willy Tarreau708e7172014-01-21 10:27:49 +0100950 if (!(conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200951 si_conn_send(conn);
Willy Tarreau798c3c92014-01-21 10:30:08 +0100952 if (conn->flags & CO_FL_ERROR) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100953 /* Write error on the file descriptor */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200954 __conn_data_stop_both(conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100955 si->flags |= SI_FL_ERR;
Willy Tarreaud29a0662012-12-10 16:33:38 +0100956 goto out_wakeup;
957 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200958 }
959
960 /* OK, so now we know that some data might have been sent, and that we may
961 * have to poll first. We have to do that too if the buffer is not empty.
962 */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100963 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200964 /* the connection is established but we can't write. Either the
965 * buffer is empty, or we just refrain from sending because the
966 * ->o limit was reached. Maybe we just wrote the last
967 * chunk and need to close.
968 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200969 __conn_data_stop_send(conn);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100970 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200971 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200972 (si->state == SI_ST_EST)) {
973 si_shutw(si);
974 goto out_wakeup;
975 }
976
Willy Tarreauafc8a222014-11-28 15:46:27 +0100977 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200978 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100979 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200980 }
981 else {
982 /* Otherwise there are remaining data to be sent in the buffer,
983 * which means we have to poll before doing so.
984 */
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200985 __conn_data_want_send(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200986 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100987 if (!tick_isset(oc->wex))
988 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200989 }
990
Willy Tarreauafc8a222014-11-28 15:46:27 +0100991 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
992 struct channel *ic = si_ic(si);
993
Willy Tarreaude5722c2012-08-20 15:01:10 +0200994 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100995 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
996 !channel_is_empty(oc))
997 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200998
Willy Tarreauafc8a222014-11-28 15:46:27 +0100999 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001000 /* Note: to prevent the client from expiring read timeouts
1001 * during writes, we refresh it. We only do this if the
1002 * interface is not configured for "independent streams",
1003 * because for some applications it's better not to do this,
1004 * for instance when continuously exchanging small amounts
1005 * of data which can full the socket buffers long before a
1006 * write timeout is detected.
1007 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001008 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001009 }
1010 }
1011
1012 /* in case of special condition (error, shutdown, end of write...), we
1013 * have to notify the task.
1014 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001015 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1016 ((oc->flags & CF_WAKE_WRITE) &&
1017 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreaue6300be2014-01-25 02:33:21 +01001018 si->state != SI_ST_EST)))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001019 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001020 if (!(si->flags & SI_FL_DONT_WAKE))
1021 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001022 }
Willy Tarreauf16723e2012-08-24 12:52:22 +02001023
1024 /* commit possible polling changes */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001025 conn_cond_update_polling(conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001026}
1027
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001028/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001029 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001030 * into the buffer from the connection. It iterates over the transport layer's
1031 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001032 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001033static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +02001034{
Willy Tarreaue603e692012-09-27 22:20:41 +02001035 struct stream_interface *si = conn->owner;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001036 struct channel *ic = si_ic(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001037 int ret, max, cur_read;
1038 int read_poll = MAX_READ_POLL_LOOPS;
1039
1040 /* stop immediately on errors. Note that we DON'T want to stop on
1041 * POLL_ERR, as the poller might report a write error while there
1042 * are still data available in the recv buffer. This typically
1043 * happens when we send too large a request to a backend server
1044 * which rejects it before reading it all.
1045 */
1046 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001047 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001048
1049 /* stop here if we reached the end of data */
1050 if (conn_data_read0_pending(conn))
1051 goto out_shutdown_r;
1052
1053 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001054 if (ic->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +02001055 return;
1056
Willy Tarreau96199b12012-08-24 00:46:52 +02001057 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +02001058
Willy Tarreauafc8a222014-11-28 15:46:27 +01001059 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !ic->buf->o &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001060 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001061 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001062 /* The buffer was empty and nothing was transferred for more
1063 * than one second. This was caused by a pause and not by
1064 * congestion. Reset any streaming mode to reduce latency.
1065 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001066 ic->xfer_small = 0;
1067 ic->xfer_large = 0;
1068 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001069 }
1070
Willy Tarreau96199b12012-08-24 00:46:52 +02001071 /* First, let's see if we may splice data across the channel without
1072 * using a buffer.
1073 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001074 if (conn->xprt->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001075 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1076 ic->flags & CF_KERN_SPLICING) {
1077 if (buffer_not_empty(ic->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001078 /* We're embarrassed, there are already data pending in
1079 * the buffer and we don't want to have them at two
1080 * locations at a time. Let's indicate we need some
1081 * place and ask the consumer to hurry.
1082 */
1083 goto abort_splice;
1084 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001085
Willy Tarreauafc8a222014-11-28 15:46:27 +01001086 if (unlikely(ic->pipe == NULL)) {
1087 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1088 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001089 goto abort_splice;
1090 }
1091 }
1092
Willy Tarreauafc8a222014-11-28 15:46:27 +01001093 ret = conn->xprt->rcv_pipe(conn, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001094 if (ret < 0) {
1095 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001096 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001097 goto abort_splice;
1098 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001099
Willy Tarreau96199b12012-08-24 00:46:52 +02001100 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001101 if (ic->to_forward != CHN_INFINITE_FORWARD)
1102 ic->to_forward -= ret;
1103 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001104 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001105 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001106 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001107
1108 if (conn_data_read0_pending(conn))
1109 goto out_shutdown_r;
1110
1111 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001112 return;
Willy Tarreau96199b12012-08-24 00:46:52 +02001113
Willy Tarreau61d39a02013-07-18 21:49:32 +02001114 if (conn->flags & CO_FL_WAIT_ROOM) {
1115 /* the pipe is full or we have read enough data that it
1116 * could soon be full. Let's stop before needing to poll.
1117 */
Willy Tarreau56a77e52012-09-02 18:34:44 +02001118 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001119 __conn_data_stop_recv(conn);
1120 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001121
Willy Tarreauce323de2012-08-20 21:41:06 +02001122 /* splice not possible (anymore), let's go on on standard copy */
1123 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001124
1125 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001126 if (ic->pipe && unlikely(!ic->pipe->data)) {
1127 put_pipe(ic->pipe);
1128 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001129 }
1130
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001131 /* now we'll need a buffer */
Willy Tarreau87b09662015-04-03 00:22:06 +02001132 if (!stream_alloc_recv_buffer(ic)) {
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001133 si->flags |= SI_FL_WAIT_ROOM;
1134 goto end_recv;
1135 }
1136
Willy Tarreau61d39a02013-07-18 21:49:32 +02001137 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1138 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1139 * that if such an event is not handled above in splice, it will be handled here by
1140 * recv().
1141 */
Willy Tarreau310987a2014-01-22 19:46:33 +01001142 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 +01001143 max = channel_recv_max(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001144
1145 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +02001146 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001147 break;
1148 }
1149
Willy Tarreauafc8a222014-11-28 15:46:27 +01001150 ret = conn->xprt->rcv_buf(conn, ic->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001151 if (ret <= 0)
1152 break;
1153
1154 cur_read += ret;
1155
1156 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001157 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001158 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001159 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1160 if (fwd > ic->to_forward)
1161 fwd = ic->to_forward;
1162 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001163 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001164 b_adv(ic->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001165 }
1166
Willy Tarreauafc8a222014-11-28 15:46:27 +01001167 ic->flags |= CF_READ_PARTIAL;
1168 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001169
Willy Tarreauafc8a222014-11-28 15:46:27 +01001170 if (!channel_may_recv(ic)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001171 si->flags |= SI_FL_WAIT_ROOM;
1172 break;
1173 }
1174
Willy Tarreauafc8a222014-11-28 15:46:27 +01001175 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001176 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001177 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001178 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001179 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001180
1181 /* if too many bytes were missing from last read, it means that
1182 * it's pointless trying to read again because the system does
1183 * not have them in buffers.
1184 */
1185 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001186 /* if a streamer has read few data, it may be because we
1187 * have exhausted system buffers. It's not worth trying
1188 * again.
1189 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001190 if (ic->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001191 break;
1192
1193 /* if we read a large block smaller than what we requested,
1194 * it's almost certain we'll never get anything more.
1195 */
1196 if (ret >= global.tune.recv_enough)
1197 break;
1198 }
1199 } /* while !flags */
1200
Willy Tarreauc5890e62014-02-09 17:47:01 +01001201 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001202 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
1203 (cur_read <= ic->buf->size / 2)) {
1204 ic->xfer_large = 0;
1205 ic->xfer_small++;
1206 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001207 /* we have read less than half of the buffer in
1208 * one pass, and this happened at least 3 times.
1209 * This is definitely not a streamer.
1210 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001211 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001212 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001213 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001214 /* if the buffer has been at least half full twice,
1215 * we receive faster than we send, so at least it
1216 * is not a "fast streamer".
1217 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001218 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001219 }
1220 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001221 else if (!(ic->flags & CF_STREAMER_FAST) &&
1222 (cur_read >= ic->buf->size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001223 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001224 ic->xfer_small = 0;
1225 ic->xfer_large++;
1226 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001227 /* we call this buffer a fast streamer if it manages
1228 * to be filled in one call 3 consecutive times.
1229 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001230 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001231 }
1232 }
1233 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001234 ic->xfer_small = 0;
1235 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001236 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001237 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001238 }
1239
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001240 end_recv:
1241 if (conn->flags & CO_FL_ERROR)
1242 return;
1243
Willy Tarreauce323de2012-08-20 21:41:06 +02001244 if (conn_data_read0_pending(conn))
1245 /* connection closed */
1246 goto out_shutdown_r;
1247
1248 return;
1249
1250 out_shutdown_r:
1251 /* we received a shutdown */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001252 ic->flags |= CF_READ_NULL;
1253 if (ic->flags & CF_AUTO_CLOSE)
1254 channel_shutw_now(ic);
Willy Tarreauce323de2012-08-20 21:41:06 +02001255 stream_sock_read0(si);
1256 conn_data_read0(conn);
1257 return;
Willy Tarreauce323de2012-08-20 21:41:06 +02001258}
1259
1260/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001261 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001262 * from the buffer to the connection. It iterates over the transport layer's
1263 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001264 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001265static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001266{
Willy Tarreaue603e692012-09-27 22:20:41 +02001267 struct stream_interface *si = conn->owner;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001268
1269 if (conn->flags & CO_FL_ERROR)
Godbach4f489902013-12-04 17:24:06 +08001270 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001271
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001272 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001273 /* a handshake was requested */
1274 return;
1275
1276 /* we might have been called just after an asynchronous shutw */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001277 if (si_oc(si)->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001278 return;
1279
1280 /* OK there are data waiting to be sent */
Godbach4f489902013-12-04 17:24:06 +08001281 si_conn_send(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001282
1283 /* OK all done */
1284 return;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001285}
1286
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001287/*
1288 * This function propagates a null read received on a socket-based connection.
1289 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001290 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001291 */
1292void stream_sock_read0(struct stream_interface *si)
1293{
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001294 struct connection *conn = __objt_conn(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001295 struct channel *ic = si_ic(si);
1296 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001297
Willy Tarreauafc8a222014-11-28 15:46:27 +01001298 ic->flags &= ~CF_SHUTR_NOW;
1299 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001300 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001301 ic->flags |= CF_SHUTR;
1302 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001303 si->flags &= ~SI_FL_WAIT_ROOM;
1304
1305 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1306 return;
1307
Willy Tarreauafc8a222014-11-28 15:46:27 +01001308 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001309 goto do_close;
1310
1311 if (si->flags & SI_FL_NOHALF) {
1312 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001313 /* force flag on ssl to keep stream in cache */
Willy Tarreau1398aa12015-03-12 23:04:07 +01001314 conn_data_shutw_hard(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001315 goto do_close;
1316 }
1317
1318 /* otherwise that's just a normal read shutdown */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001319 __conn_data_stop_recv(conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001320 return;
1321
1322 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001323 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001324 conn_full_close(conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001325
Willy Tarreauafc8a222014-11-28 15:46:27 +01001326 ic->flags &= ~CF_SHUTR_NOW;
1327 ic->flags |= CF_SHUTR;
1328 ic->rex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001329
Willy Tarreauafc8a222014-11-28 15:46:27 +01001330 oc->flags &= ~CF_SHUTW_NOW;
1331 oc->flags |= CF_SHUTW;
1332 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001333
1334 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1335
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001336 si->state = SI_ST_DIS;
1337 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001338 return;
1339}
1340
Willy Tarreau651e1822015-09-23 20:06:13 +02001341/* Callback to be used by applet handlers upon completion. It updates the stream
1342 * (which may or may not take this opportunity to try to forward data), then
1343 * may disable the applet's based on the channels and stream interface's final
1344 * states.
1345 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001346void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001347{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001348 struct channel *ic = si_ic(si);
1349
1350 /* If the applet wants to write and the channel is closed, it's a
1351 * broken pipe and it must be reported.
1352 */
1353 if ((si->flags & SI_FL_WANT_PUT) && (ic->flags & CF_SHUTR))
1354 si->flags |= SI_FL_ERR;
1355
Willy Tarreau651e1822015-09-23 20:06:13 +02001356 /* update the stream-int, channels, and possibly wake the stream up */
1357 stream_int_notify(si);
Willy Tarreau388a2382015-09-23 19:55:42 +02001358
1359 /* Get away from the active list if we can't work anymore.
1360 * We also do that if the main task has already scheduled, because it
1361 * saves a useless wakeup/pause/wakeup cycle causing one useless call
1362 * per session on average.
1363 */
1364 if (task_in_rq(si_task(si)) ||
1365 (((si->flags & (SI_FL_WANT_PUT|SI_FL_WAIT_ROOM)) != SI_FL_WANT_PUT) &&
1366 ((si->flags & (SI_FL_WANT_GET|SI_FL_WAIT_DATA)) != SI_FL_WANT_GET)))
1367 appctx_pause(si_appctx(si));
Willy Tarreaue5f86492015-04-19 15:16:35 +02001368}
1369
Willy Tarreau452c7d52015-09-25 10:39:16 +02001370
1371/* Updates the activity status of an applet outside of the applet handler based
1372 * on the channel's flags and the stream interface's flags. It needs to be
1373 * called once after the channels' flags have settled down and the stream has
1374 * been updated. It is not designed to be called from within the applet handler
1375 * itself.
Willy Tarreau563cc372015-04-19 18:13:56 +02001376 */
1377void stream_int_update_applet(struct stream_interface *si)
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001378{
Willy Tarreaufe127932015-04-21 19:23:39 +02001379 if (((si->flags & (SI_FL_WANT_PUT|SI_FL_WAIT_ROOM)) == SI_FL_WANT_PUT) ||
1380 ((si->flags & (SI_FL_WANT_GET|SI_FL_WAIT_DATA)) == SI_FL_WANT_GET))
Willy Tarreau563cc372015-04-19 18:13:56 +02001381 appctx_wakeup(si_appctx(si));
Willy Tarreaufe127932015-04-21 19:23:39 +02001382 else
1383 appctx_pause(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001384}
1385
1386/*
1387 * This function performs a shutdown-read on a stream interface attached to an
1388 * applet in a connected or init state (it does nothing for other states). It
1389 * either shuts the read side or marks itself as closed. The buffer flags are
1390 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1391 * we also forward the close to the write side. The owner task is woken up if
1392 * it exists.
1393 */
1394static void stream_int_shutr_applet(struct stream_interface *si)
1395{
1396 struct channel *ic = si_ic(si);
1397
1398 ic->flags &= ~CF_SHUTR_NOW;
1399 if (ic->flags & CF_SHUTR)
1400 return;
1401 ic->flags |= CF_SHUTR;
1402 ic->rex = TICK_ETERNITY;
1403 si->flags &= ~SI_FL_WAIT_ROOM;
1404
Willy Tarreau828824a2015-04-19 17:20:03 +02001405 /* Note: on shutr, we don't call the applet */
1406
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001407 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1408 return;
1409
1410 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001411 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001412 si->state = SI_ST_DIS;
1413 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001414 }
1415 else if (si->flags & SI_FL_NOHALF) {
1416 /* we want to immediately forward this close to the write side */
1417 return stream_int_shutw_applet(si);
1418 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001419}
1420
1421/*
1422 * This function performs a shutdown-write on a stream interface attached to an
1423 * applet in a connected or init state (it does nothing for other states). It
1424 * either shuts the write side or marks itself as closed. The buffer flags are
1425 * updated to reflect the new state. It does also close everything if the SI
1426 * was marked as being in error state. The owner task is woken up if it exists.
1427 */
1428static void stream_int_shutw_applet(struct stream_interface *si)
1429{
1430 struct channel *ic = si_ic(si);
1431 struct channel *oc = si_oc(si);
1432
1433 oc->flags &= ~CF_SHUTW_NOW;
1434 if (oc->flags & CF_SHUTW)
1435 return;
1436 oc->flags |= CF_SHUTW;
1437 oc->wex = TICK_ETERNITY;
1438 si->flags &= ~SI_FL_WAIT_DATA;
1439
Willy Tarreau828824a2015-04-19 17:20:03 +02001440 /* on shutw we always wake the applet up */
1441 appctx_wakeup(si_appctx(si));
1442
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001443 switch (si->state) {
1444 case SI_ST_EST:
1445 /* we have to shut before closing, otherwise some short messages
1446 * may never leave the system, especially when there are remaining
1447 * unread data in the socket input buffer, or when nolinger is set.
1448 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1449 * no risk so we close both sides immediately.
1450 */
1451 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1452 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1453 return;
1454
1455 /* fall through */
1456 case SI_ST_CON:
1457 case SI_ST_CER:
1458 case SI_ST_QUE:
1459 case SI_ST_TAR:
1460 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001461 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001462 si->state = SI_ST_DIS;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001463 default:
1464 si->flags &= ~(SI_FL_WAIT_ROOM | SI_FL_NOLINGER);
1465 ic->flags &= ~CF_SHUTR_NOW;
1466 ic->flags |= CF_SHUTR;
1467 ic->rex = TICK_ETERNITY;
1468 si->exp = TICK_ETERNITY;
1469 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001470}
1471
1472/* chk_rcv function for applets */
1473static void stream_int_chk_rcv_applet(struct stream_interface *si)
1474{
1475 struct channel *ic = si_ic(si);
1476
1477 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1478 __FUNCTION__,
1479 si, si->state, ic->flags, si_oc(si)->flags);
1480
1481 if (unlikely(si->state != SI_ST_EST || (ic->flags & (CF_SHUTR|CF_DONT_READ))))
1482 return;
Willy Tarreau828824a2015-04-19 17:20:03 +02001483 /* here we only wake the applet up if it was waiting for some room */
1484 if (!(si->flags & SI_FL_WAIT_ROOM))
1485 return;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001486
Willy Tarreau828824a2015-04-19 17:20:03 +02001487 if (channel_may_recv(ic) && !ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001488 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001489 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001490 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001491}
1492
1493/* chk_snd function for applets */
1494static void stream_int_chk_snd_applet(struct stream_interface *si)
1495{
1496 struct channel *oc = si_oc(si);
1497
1498 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1499 __FUNCTION__,
1500 si, si->state, si_ic(si)->flags, oc->flags);
1501
1502 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1503 return;
1504
Willy Tarreau828824a2015-04-19 17:20:03 +02001505 /* we only wake the applet up if it was waiting for some data */
1506
1507 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001508 return;
1509
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001510 if (!tick_isset(oc->wex))
1511 oc->wex = tick_add_ifset(now_ms, oc->wto);
1512
Willy Tarreau828824a2015-04-19 17:20:03 +02001513 if (!channel_is_empty(oc)) {
1514 /* (re)start sending */
1515 appctx_wakeup(si_appctx(si));
1516 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001517}
1518
Willy Tarreaudded32d2008-11-30 19:48:07 +01001519/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001520 * Local variables:
1521 * c-indent-level: 8
1522 * c-basic-offset: 8
1523 * End:
1524 */