blob: 7d89cc90eb491590ce24721dc26457146d18b793 [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>
Olivier Houcharda254a372019-04-05 15:30:12 +020033#include <proto/http_htx.h>
Olivier Houchard9aaf7782017-09-13 18:30:23 +020034#include <proto/mux_pt.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020035#include <proto/pipe.h>
Olivier Houcharda254a372019-04-05 15:30:12 +020036#include <proto/proxy.h>
Willy Tarreau87b09662015-04-03 00:22:06 +020037#include <proto/stream.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020038#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010039#include <proto/task.h>
40
Willy Tarreaufd31e532012-07-23 18:24:25 +020041#include <types/pipe.h>
42
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010043/* functions used by default on a detached stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020044static void stream_int_shutr(struct stream_interface *si);
45static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020046static void stream_int_chk_rcv(struct stream_interface *si);
47static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010048
49/* functions used on a conn_stream-based stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020050static void stream_int_shutr_conn(struct stream_interface *si);
51static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020052static void stream_int_chk_rcv_conn(struct stream_interface *si);
53static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010054
55/* functions used on an applet-based stream-interface */
Willy Tarreaud45b9f82015-04-13 16:30:14 +020056static void stream_int_shutr_applet(struct stream_interface *si);
57static void stream_int_shutw_applet(struct stream_interface *si);
58static void stream_int_chk_rcv_applet(struct stream_interface *si);
59static void stream_int_chk_snd_applet(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010060
61/* last read notification */
62static void stream_int_read0(struct stream_interface *si);
63
64/* post-IO notification callback */
65static void stream_int_notify(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020066
Willy Tarreauc5788912012-08-24 18:12:41 +020067/* stream-interface operations for embedded tasks */
68struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020069 .chk_rcv = stream_int_chk_rcv,
70 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020071 .shutr = stream_int_shutr,
72 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020073};
74
Willy Tarreauc5788912012-08-24 18:12:41 +020075/* stream-interface operations for connections */
76struct si_ops si_conn_ops = {
Willy Tarreauc5788912012-08-24 18:12:41 +020077 .chk_rcv = stream_int_chk_rcv_conn,
78 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020079 .shutr = stream_int_shutr_conn,
80 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020081};
82
Willy Tarreaud45b9f82015-04-13 16:30:14 +020083/* stream-interface operations for connections */
84struct si_ops si_applet_ops = {
Willy Tarreaud45b9f82015-04-13 16:30:14 +020085 .chk_rcv = stream_int_chk_rcv_applet,
86 .chk_snd = stream_int_chk_snd_applet,
87 .shutr = stream_int_shutr_applet,
88 .shutw = stream_int_shutw_applet,
89};
90
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010091
92/* Functions used to communicate with a conn_stream. The first two may be used
93 * directly, the last one is mostly a wake callback.
94 */
95int si_cs_recv(struct conn_stream *cs);
96int si_cs_send(struct conn_stream *cs);
97static int si_cs_process(struct conn_stream *cs);
98
99
Willy Tarreau74beec32012-10-03 00:41:04 +0200100struct data_cb si_conn_cb = {
Olivier Houchard21df6cc2018-09-14 23:21:44 +0200101 .wake = si_cs_process,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100102 .name = "STRM",
Willy Tarreauc5788912012-08-24 18:12:41 +0200103};
104
Willy Tarreaucff64112008-11-03 06:26:53 +0100105/*
106 * This function only has to be called once after a wakeup event in case of
107 * suspected timeout. It controls the stream interface timeouts and sets
108 * si->flags accordingly. It does NOT close anything, as this timeout may
109 * be used for any purpose. It returns 1 if the timeout fired, otherwise
110 * zero.
111 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100112int si_check_timeouts(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100113{
114 if (tick_is_expired(si->exp, now_ms)) {
115 si->flags |= SI_FL_EXP;
116 return 1;
117 }
118 return 0;
119}
120
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100121/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100122void si_report_error(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100123{
124 if (!si->err_type)
125 si->err_type = SI_ET_DATA_ERR;
126
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100127 si_oc(si)->flags |= CF_WRITE_ERROR;
128 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100129}
130
131/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100132 * Returns a message to the client ; the connection is shut down for read,
133 * and the request is cleared so that no server connection can be initiated.
134 * The buffer is marked for read shutdown on the other side to protect the
135 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100136 * "chunk". If it is null, then an empty message is used. The reply buffer does
137 * not need to be empty before this, and its contents will not be overwritten.
138 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100139 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100140void si_retnclose(struct stream_interface *si,
Willy Tarreau83061a82018-07-13 11:56:34 +0200141 const struct buffer *msg)
Willy Tarreaudded32d2008-11-30 19:48:07 +0100142{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100143 struct channel *ic = si_ic(si);
144 struct channel *oc = si_oc(si);
145
146 channel_auto_read(ic);
147 channel_abort(ic);
148 channel_auto_close(ic);
149 channel_erase(ic);
150 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100151
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200152 if (likely(msg && msg->data))
153 co_inject(oc, msg->area, msg->data);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100154
Willy Tarreauafc8a222014-11-28 15:46:27 +0100155 oc->wex = tick_add_ifset(now_ms, oc->wto);
156 channel_auto_read(oc);
157 channel_auto_close(oc);
158 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100159}
160
Willy Tarreau4a36b562012-08-06 19:31:45 +0200161/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200162 * This function performs a shutdown-read on a detached stream interface in a
163 * connected or init state (it does nothing for other states). It either shuts
164 * the read side or marks itself as closed. The buffer flags are updated to
165 * reflect the new state. If the stream interface has SI_FL_NOHALF, we also
166 * forward the close to the write side. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200167 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200168static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200169{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100170 struct channel *ic = si_ic(si);
171
Willy Tarreauabb5d422018-11-14 16:58:52 +0100172 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100173 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200174 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100175 ic->flags |= CF_SHUTR;
176 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200177
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200178 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200179 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200180
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100181 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200182 si->state = SI_ST_DIS;
183 si->exp = TICK_ETERNITY;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200184 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200185 else if (si->flags & SI_FL_NOHALF) {
186 /* we want to immediately forward this close to the write side */
187 return stream_int_shutw(si);
188 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200189
Willy Tarreau4a36b562012-08-06 19:31:45 +0200190 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100191 if (!(si->flags & SI_FL_DONT_WAKE))
192 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200193}
194
Willy Tarreau4a36b562012-08-06 19:31:45 +0200195/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200196 * This function performs a shutdown-write on a detached stream interface in a
197 * connected or init state (it does nothing for other states). It either shuts
198 * the write side or marks itself as closed. The buffer flags are updated to
199 * reflect the new state. It does also close everything if the SI was marked as
200 * being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200201 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200202static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200203{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100204 struct channel *ic = si_ic(si);
205 struct channel *oc = si_oc(si);
206
207 oc->flags &= ~CF_SHUTW_NOW;
208 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200209 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100210 oc->flags |= CF_SHUTW;
211 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +0100212 si_done_get(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200213
Hongbo Longe39683c2017-03-10 18:41:51 +0100214 if (tick_isset(si->hcto)) {
215 ic->rto = si->hcto;
216 ic->rex = tick_add(now_ms, ic->rto);
217 }
218
Willy Tarreaufb90d942009-09-05 20:57:35 +0200219 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200220 case SI_ST_RDY:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200221 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200222 /* we have to shut before closing, otherwise some short messages
223 * may never leave the system, especially when there are remaining
224 * unread data in the socket input buffer, or when nolinger is set.
225 * However, if SI_FL_NOLINGER is explicitly set, we know there is
226 * no risk so we close both sides immediately.
227 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200228 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100229 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200230 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200231
232 /* fall through */
233 case SI_ST_CON:
234 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100235 case SI_ST_QUE:
236 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200237 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200238 si->state = SI_ST_DIS;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200239 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +0100240 si->flags &= ~SI_FL_NOLINGER;
241 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100242 ic->flags |= CF_SHUTR;
243 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244 si->exp = TICK_ETERNITY;
245 }
246
Willy Tarreau4a36b562012-08-06 19:31:45 +0200247 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100248 if (!(si->flags & SI_FL_DONT_WAKE))
249 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250}
251
252/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200253static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200254{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100255 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200256
Willy Tarreauafc8a222014-11-28 15:46:27 +0100257 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200258 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100259 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200260
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200261 if (ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200262 /* stop reading */
Willy Tarreaudb398432018-11-15 11:08:52 +0100263 si_rx_room_blk(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200264 }
265 else {
266 /* (re)start reading */
Willy Tarreau3c39a7d2019-06-14 14:42:29 +0200267 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau07373b82014-11-28 12:08:47 +0100268 if (!(si->flags & SI_FL_DONT_WAKE))
269 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200270 }
271}
272
273/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200274static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200275{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100276 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200277
Willy Tarreauafc8a222014-11-28 15:46:27 +0100278 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200279 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100280 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200281
Willy Tarreauafc8a222014-11-28 15:46:27 +0100282 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200283 return;
284
285 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100286 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200287 return;
288
289 /* Otherwise there are remaining data to be sent in the buffer,
290 * so we tell the handler.
291 */
292 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100293 if (!tick_isset(oc->wex))
294 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200295
Willy Tarreau07373b82014-11-28 12:08:47 +0100296 if (!(si->flags & SI_FL_DONT_WAKE))
297 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200298}
299
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200300/* Register an applet to handle a stream_interface as a new appctx. The SI will
301 * wake it up everytime it is solicited. The appctx must be deleted by the task
302 * handler using si_release_endpoint(), possibly from within the function itself.
303 * It also pre-initializes the applet's context and returns it (or NULL in case
304 * it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200305 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100306struct appctx *si_register_handler(struct stream_interface *si, struct applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200307{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100308 struct appctx *appctx;
309
Willy Tarreau07373b82014-11-28 12:08:47 +0100310 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200311
Willy Tarreaua7513f52015-04-05 00:15:26 +0200312 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100313 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100314 return NULL;
315
Willy Tarreau0cd3bd62018-11-06 18:46:37 +0100316 si_cant_get(si);
Willy Tarreau828824a2015-04-19 17:20:03 +0200317 appctx_wakeup(appctx);
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100318 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200319}
320
Willy Tarreau2c6be842012-07-06 17:12:34 +0200321/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200322 * established. It returns 0 if it fails in a fatal way or needs to poll to go
323 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200324 * flags (the bit is provided in <flag> by the caller). It is designed to be
325 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200326 * Note that it can emit a PROXY line by relying on the other end's address
327 * when the connection is attached to a stream interface, or by resolving the
328 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200329 */
330int conn_si_send_proxy(struct connection *conn, unsigned int flag)
331{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200332 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200333 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200334 goto out_error;
335
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100336 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200337 goto out_error;
338
Willy Tarreau2c6be842012-07-06 17:12:34 +0200339 /* If we have a PROXY line to send, we'll use this to validate the
340 * connection, in which case the connection is validated only once
341 * we've sent the whole proxy line. Otherwise we use connect().
342 */
Tim Duesterhus36839dc2019-02-26 17:09:51 +0100343 if (conn->send_proxy_ofs) {
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100344 const struct conn_stream *cs;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200345 int ret;
346
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100347 cs = cs_get_first(conn);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200348 /* The target server expects a PROXY line to be sent first.
349 * If the send_proxy_ofs is negative, it corresponds to the
350 * offset to start sending from then end of the proxy string
351 * (which is recomputed every time since it's constant). If
352 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200353 * We can only send a "normal" PROXY line when the connection
354 * is attached to a stream interface. Otherwise we can only
355 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200356 */
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100357
358 if (cs && cs->data_cb == &si_conn_cb) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200359 struct stream_interface *si = cs->data;
360 struct conn_stream *remote_cs = objt_cs(si_opposite(si)->end);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200361 ret = make_proxy_line(trash.area, trash.size,
362 objt_server(conn->target),
363 remote_cs ? remote_cs->conn : NULL);
Olivier Houchard661167d2019-05-29 17:08:03 +0200364 /* We may not have a conn_stream yet, if we don't
365 * know which mux to use, because it will be decided
366 * during the SSL handshake. In this case, there should
367 * be a session associated to the connection in
368 * conn->owner, and we know it is the session that
369 * initiated that connection, so we can just use
370 * its origin, which should contain the client
371 * connection.
372 */
373 } else if (!cs && conn->owner) {
374 struct session *sess = conn->owner;
375
376 ret = make_proxy_line(trash.area, trash.size,
377 objt_server(conn->target),
378 objt_conn(sess->origin));
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200379 }
380 else {
381 /* The target server expects a LOCAL line to be sent first. Retrieving
382 * local or remote addresses may fail until the connection is established.
383 */
384 conn_get_from_addr(conn);
385 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
386 goto out_wait;
387
388 conn_get_to_addr(conn);
389 if (!(conn->flags & CO_FL_ADDR_TO_SET))
390 goto out_wait;
391
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200392 ret = make_proxy_line(trash.area, trash.size,
393 objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200394 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200395
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396 if (!ret)
397 goto out_error;
398
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200399 if (conn->send_proxy_ofs > 0)
400 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200401
Willy Tarreaua1a74742012-08-24 12:14:49 +0200402 /* we have to send trash from (ret+sp for -sp bytes). If the
403 * data layer has a pending write, we'll also set MSG_MORE.
404 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200405 ret = conn_sock_send(conn,
406 trash.area + ret + conn->send_proxy_ofs,
407 -conn->send_proxy_ofs,
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200408 (conn->flags & CO_FL_XPRT_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200409
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100410 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200411 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200412
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200413 conn->send_proxy_ofs += ret; /* becomes zero once complete */
414 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415 goto out_wait;
416
417 /* OK we've sent the whole line, we're connected */
418 }
419
Willy Tarreaua1a74742012-08-24 12:14:49 +0200420 /* The connection is ready now, simply return and let the connection
421 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200422 */
423 if (conn->flags & CO_FL_WAIT_L4_CONN)
424 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200425 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200426 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200427
428 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200429 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200430 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200431 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200432
433 out_wait:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200434 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200435}
436
Willy Tarreau27375622013-12-17 00:00:28 +0100437
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100438/* This function is the equivalent to si_update() except that it's
Willy Tarreau615f28b2015-09-23 18:40:09 +0200439 * designed to be called from outside the stream handlers, typically the lower
440 * layers (applets, connections) after I/O completion. After updating the stream
441 * interface and timeouts, it will try to forward what can be forwarded, then to
442 * wake the associated task up if an important event requires special handling.
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100443 * It may update SI_FL_WAIT_DATA and/or SI_FL_RXBLK_ROOM, that the callers are
Willy Tarreau0dfccb22018-10-25 13:55:20 +0200444 * encouraged to watch to take appropriate action.
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100445 * It should not be called from within the stream itself, si_update()
Willy Tarreau615f28b2015-09-23 18:40:09 +0200446 * is designed for this.
447 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100448static void stream_int_notify(struct stream_interface *si)
Willy Tarreau615f28b2015-09-23 18:40:09 +0200449{
450 struct channel *ic = si_ic(si);
451 struct channel *oc = si_oc(si);
Willy Tarreau47baeb82018-11-15 07:46:57 +0100452 struct stream_interface *sio = si_opposite(si);
Christopher Fauletd7607de2019-01-03 16:24:54 +0100453 struct task *task = si_task(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200454
455 /* process consumer side */
456 if (channel_is_empty(oc)) {
Olivier Houcharde9bed532017-11-16 17:49:25 +0100457 struct connection *conn = objt_cs(si->end) ? objt_cs(si->end)->conn : NULL;
458
Willy Tarreau615f28b2015-09-23 18:40:09 +0200459 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Olivier Houcharde9bed532017-11-16 17:49:25 +0100460 (si->state == SI_ST_EST) && (!conn || !(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS))))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200461 si_shutw(si);
462 oc->wex = TICK_ETERNITY;
463 }
464
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100465 /* indicate that we may be waiting for data from the output channel or
466 * we're about to close and can't expect more data if SHUTW_NOW is there.
467 */
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200468 if (!(oc->flags & (CF_SHUTW|CF_SHUTW_NOW)))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200469 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100470 else if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW)
471 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200472
473 /* update OC timeouts and wake the other side up if it's waiting for room */
474 if (oc->flags & CF_WRITE_ACTIVITY) {
475 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
476 !channel_is_empty(oc))
477 if (tick_isset(oc->wex))
478 oc->wex = tick_add_ifset(now_ms, oc->wto);
479
480 if (!(si->flags & SI_FL_INDEP_STR))
481 if (tick_isset(ic->rex))
482 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf26c26c2018-11-12 16:11:08 +0100483 }
Willy Tarreau615f28b2015-09-23 18:40:09 +0200484
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100485 if (oc->flags & CF_DONT_READ)
486 si_rx_chan_blk(sio);
487 else
488 si_rx_chan_rdy(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200489
490 /* Notify the other side when we've injected data into the IC that
491 * needs to be forwarded. We can do fast-forwarding as soon as there
492 * are output data, but we avoid doing this if some of the data are
493 * not yet scheduled for being forwarded, because it is very likely
494 * that it will be done again immediately afterwards once the following
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100495 * data are parsed (eg: HTTP chunking). We only SI_FL_RXBLK_ROOM once
Willy Tarreau615f28b2015-09-23 18:40:09 +0200496 * we've emptied *some* of the output buffer, and not just when there
497 * is available room, because applets are often forced to stop before
498 * the buffer is full. We must not stop based on input data alone because
499 * an HTTP parser might need more data to complete the parsing.
500 */
501 if (!channel_is_empty(ic) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100502 (sio->flags & SI_FL_WAIT_DATA) &&
Willy Tarreau89b6a2b2018-11-18 15:46:10 +0100503 (!(ic->flags & CF_EXPECT_MORE) || c_full(ic) || ci_data(ic) == 0 || ic->pipe)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200504 int new_len, last_len;
505
Willy Tarreau77e478c2018-06-19 07:03:14 +0200506 last_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200507 if (ic->pipe)
508 last_len += ic->pipe->data;
509
Willy Tarreau47baeb82018-11-15 07:46:57 +0100510 si_chk_snd(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200511
Willy Tarreau77e478c2018-06-19 07:03:14 +0200512 new_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200513 if (ic->pipe)
514 new_len += ic->pipe->data;
515
516 /* check if the consumer has freed some space either in the
517 * buffer or in the pipe.
518 */
Willy Tarreau47baeb82018-11-15 07:46:57 +0100519 if (new_len < last_len)
Willy Tarreaudb398432018-11-15 11:08:52 +0100520 si_rx_room_rdy(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200521 }
522
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100523 if (!(ic->flags & CF_DONT_READ))
524 si_rx_chan_rdy(si);
525
Willy Tarreau47baeb82018-11-15 07:46:57 +0100526 si_chk_rcv(si);
527 si_chk_rcv(sio);
528
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100529 if (si_rx_blocked(si)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200530 ic->rex = TICK_ETERNITY;
531 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100532 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL)) == CF_READ_PARTIAL) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200533 /* we must re-enable reading if si_chk_snd() has freed some space */
534 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
535 ic->rex = tick_add_ifset(now_ms, ic->rto);
536 }
537
538 /* wake the task up only when needed */
539 if (/* changes on the production side */
540 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200541 !si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200542 (si->flags & SI_FL_ERR) ||
543 ((ic->flags & CF_READ_PARTIAL) &&
Christopher Faulet297d3e22019-03-22 14:16:14 +0100544 ((ic->flags & CF_EOI) || !ic->to_forward || sio->state != SI_ST_EST)) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200545
546 /* changes on the consumption side */
547 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
Willy Tarreauede3d882018-10-24 17:17:56 +0200548 ((oc->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreau615f28b2015-09-23 18:40:09 +0200549 ((oc->flags & CF_SHUTW) ||
Willy Tarreau78f5ff82018-12-19 11:00:00 +0100550 (((oc->flags & CF_WAKE_WRITE) ||
551 !(oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW|CF_SHUTW))) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100552 (sio->state != SI_ST_EST ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200553 (channel_is_empty(oc) && !oc->to_forward)))))) {
Christopher Fauletd7607de2019-01-03 16:24:54 +0100554 task_wakeup(task, TASK_WOKEN_IO);
555 }
556 else {
557 /* Update expiration date for the task and requeue it */
558 task->expire = tick_first((tick_is_expired(task->expire, now_ms) ? 0 : task->expire),
559 tick_first(tick_first(ic->rex, ic->wex),
560 tick_first(oc->rex, oc->wex)));
Willy Tarreau7343c712019-08-01 18:51:38 +0200561
562 task->expire = tick_first(task->expire, ic->analyse_exp);
563 task->expire = tick_first(task->expire, oc->analyse_exp);
564
565 if (si->exp)
566 task->expire = tick_first(task->expire, si->exp);
567
568 if (sio->exp)
569 task->expire = tick_first(task->expire, sio->exp);
570
Christopher Fauletd7607de2019-01-03 16:24:54 +0100571 task_queue(task);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200572 }
573 if (ic->flags & CF_READ_ACTIVITY)
574 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200575}
576
577
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200578/* Called by I/O handlers after completion.. It propagates
Willy Tarreau651e1822015-09-23 20:06:13 +0200579 * connection flags to the stream interface, updates the stream (which may or
580 * may not take this opportunity to try to forward data), then update the
581 * connection's polling based on the channels and stream interface's final
582 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200583 */
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200584static int si_cs_process(struct conn_stream *cs)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200585{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200586 struct connection *conn = cs->conn;
587 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100588 struct channel *ic = si_ic(si);
589 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200590
Olivier Houchardc7ffa912018-08-28 19:37:41 +0200591 /* If we have data to send, try it now */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100592 if (!channel_is_empty(oc) && !(si->wait_event.events & SUB_RETRY_SEND))
Willy Tarreau908d26f2018-10-25 14:02:47 +0200593 si_cs_send(cs);
594
Willy Tarreau651e1822015-09-23 20:06:13 +0200595 /* First step, report to the stream-int what was detected at the
596 * connection layer : errors and connection establishment.
Olivier Houchard6187cc32019-06-24 16:08:08 +0200597 * Only add SI_FL_ERR if we're connected, or we're attempting to
598 * connect, we may get there because we got woken up, but only run
599 * after process_stream() noticed there were an error, and decided
600 * to retry to connect, the connection may still have CO_FL_ERROR,
601 * and we don't want to add SI_FL_ERR back
Willy Tarreau651e1822015-09-23 20:06:13 +0200602 */
Olivier Houchard6187cc32019-06-24 16:08:08 +0200603 if (si->state >= SI_ST_CON &&
604 (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR))
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200605 si->flags |= SI_FL_ERR;
606
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200607 /* If we had early data, and the handshake ended, then
608 * we can remove the flag, and attempt to wake the task up,
609 * in the event there's an analyser waiting for the end of
610 * the handshake.
611 */
Olivier Houchard6fa63d92017-11-27 18:41:32 +0100612 if (!(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS)) &&
613 (cs->flags & CS_FL_WAIT_FOR_HS)) {
614 cs->flags &= ~CS_FL_WAIT_FOR_HS;
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200615 task_wakeup(si_task(si), TASK_WOKEN_MSG);
616 }
617
Willy Tarreau7ab22adb2019-06-05 14:53:22 +0200618 if (!si_state_in(si->state, SI_SB_EST|SI_SB_DIS|SI_SB_CLO) &&
Willy Tarreau52821e22017-03-18 17:11:37 +0100619 (conn->flags & (CO_FL_CONNECTED | CO_FL_HANDSHAKE)) == CO_FL_CONNECTED) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200620 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100621 oc->flags |= CF_WRITE_NULL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +0200622 if (si->state == SI_ST_CON)
623 si->state = SI_ST_RDY;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200624 }
625
Christopher Faulet297d3e22019-03-22 14:16:14 +0100626 /* Report EOI on the channel if it was reached from the mux point of
627 * view. */
628 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200629 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100630
Willy Tarreau651e1822015-09-23 20:06:13 +0200631 /* Second step : update the stream-int and channels, try to forward any
632 * pending data, then possibly wake the stream up based on the new
633 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200634 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200635 stream_int_notify(si);
Christopher Fauleta73e59b2016-12-09 17:30:18 +0100636 channel_release_buffer(ic, &(si_strm(si)->buffer_wait));
Willy Tarreauea3cc482015-09-23 19:37:00 +0200637
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200638 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200639}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200640
Willy Tarreau5368d802012-08-21 18:22:06 +0200641/*
642 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200643 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800644 * caller to commit polling changes. The caller should check conn->flags
645 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200646 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200647int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200648{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200649 struct connection *conn = cs->conn;
650 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100651 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200652 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200653 int did_send = 0;
654
655 /* We're already waiting to be able to send, give up */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100656 if (si->wait_event.events & SUB_RETRY_SEND)
Olivier Houchardf6535282018-08-31 17:29:12 +0200657 return 0;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200658
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100659 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchard6187cc32019-06-24 16:08:08 +0200660 /* We're probably there because the tasklet was woken up,
661 * but process_stream() ran before, detected there were an
662 * error and put the si back to SI_ST_TAR. There's still
663 * CO_FL_ERROR on the connection but we don't want to add
664 * SI_FL_ERR back, so give up
665 */
666 if (si->state < SI_ST_CON)
667 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100668 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200669 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100670 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200671
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200672 /* we might have been called just after an asynchronous shutw */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100673 if (conn->flags & CO_FL_SOCK_WR_SH || oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200674 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200675
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200676 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
677 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet0059e992019-07-05 11:49:11 +0200678 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200679 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200680
Willy Tarreauafc8a222014-11-28 15:46:27 +0100681 if (!oc->pipe->data) {
682 put_pipe(oc->pipe);
683 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200684 }
685
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100686 if (oc->pipe)
687 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200688 }
689
690 /* At this point, the pipe is empty, but we may still have data pending
691 * in the normal buffer.
692 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100693 if (co_data(oc)) {
694 /* when we're here, we already know that there is no spliced
695 * data left, and that there are sendable buffered data.
696 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200697
Willy Tarreau5368d802012-08-21 18:22:06 +0200698 /* check if we want to inform the kernel that we're interested in
699 * sending more data after this call. We want this if :
700 * - we're about to close after this last send and want to merge
701 * the ongoing FIN with the last segment.
702 * - we know we can't send everything at once and must get back
703 * here because of unaligned data
704 * - there is still a finite amount of data to forward
705 * The test is arranged so that the most common case does only 2
706 * tests.
707 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100708 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200709
Willy Tarreauafc8a222014-11-28 15:46:27 +0100710 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
711 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau4ac49282017-10-17 16:33:46 +0200712 (oc->flags & CF_EXPECT_MORE))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100713 ((oc->flags & CF_ISRESP) &&
714 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100715 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200716
Willy Tarreauafc8a222014-11-28 15:46:27 +0100717 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100718 send_flag |= CO_SFL_STREAMER;
719
Olivier Houcharda254a372019-04-05 15:30:12 +0200720 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
Olivier Houchardce1a0292019-05-17 15:38:29 +0200721 struct stream *s = si_strm(si);
Olivier Houcharda254a372019-04-05 15:30:12 +0200722 /* If we want to be able to do L7 retries, copy
723 * the data we're about to send, so that we are able
724 * to resend them if needed
725 */
726 /* Try to allocate a buffer if we had none.
727 * If it fails, the next test will just
728 * disable the l7 retries by setting
729 * l7_conn_retries to 0.
730 */
Olivier Houchardce1a0292019-05-17 15:38:29 +0200731 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
Olivier Houcharda254a372019-04-05 15:30:12 +0200732 si->flags &= ~SI_FL_L7_RETRY;
733 else {
734 if (b_is_null(&si->l7_buffer))
735 b_alloc(&si->l7_buffer);
736 if (b_is_null(&si->l7_buffer))
737 si->flags &= ~SI_FL_L7_RETRY;
738 else {
739 memcpy(b_orig(&si->l7_buffer),
740 b_orig(&oc->buf),
741 b_size(&oc->buf));
742 si->l7_buffer.head = co_data(oc);
743 b_add(&si->l7_buffer, co_data(oc));
744 }
745
746 }
747 }
748
Olivier Houcharded0f2072018-08-16 15:41:52 +0200749 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800750 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200751 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200752 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200753 c_realign_if_empty(oc);
754
755 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800756 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100757 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800758 }
Godbache68e02d2013-10-11 15:48:29 +0800759 /* if some data remain in the buffer, it's only because the
760 * system buffers are full, we will try next time.
761 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200762 }
Godbache68e02d2013-10-11 15:48:29 +0800763 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100764
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100765 end:
Christopher Faulet0059e992019-07-05 11:49:11 +0200766 if (did_send) {
767 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
768 if (si->state == SI_ST_CON)
769 si->state = SI_ST_RDY;
Christopher Fauletab1300b2019-07-05 13:44:29 +0200770
771 si_rx_room_rdy(si_opposite(si));
Christopher Faulet0059e992019-07-05 11:49:11 +0200772 }
773
774 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
775 si->flags |= SI_FL_ERR;
776 return 1;
777 }
778
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200779 /* We couldn't send all of our data, let the mux know we'd like to send more */
Willy Tarreau691fe392018-11-12 18:48:52 +0100780 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100781 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200782 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200783}
784
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100785/* This is the ->process() function for any stream-interface's wait_event task.
786 * It's assigned during the stream-interface's initialization, for any type of
787 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
788 * stream interface, as the presence of the CS is checked there.
789 */
Olivier Houchard91894cb2018-08-02 18:06:28 +0200790struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned short state)
791{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200792 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200793 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200794 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200795
796 if (!cs)
797 return NULL;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100798
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100799 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200800 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100801 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200802 ret |= si_cs_recv(cs);
803 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200804 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200805
Olivier Houchard91894cb2018-08-02 18:06:28 +0200806 return (NULL);
807}
808
Willy Tarreau25f13102015-09-24 11:32:22 +0200809/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200810 * update the input channel's expiration timer and the stream interface's
811 * Rx flags based on the channel's flags. It needs to be called only once
812 * after the channel's flags have settled down, and before they are cleared,
813 * though it doesn't harm to call it as often as desired (it just slightly
814 * hurts performance). It must not be called from outside of the stream
815 * handler, as what it does will be used to compute the stream task's
816 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200817 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200818void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200819{
820 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200821
Willy Tarreau236c4292019-06-06 08:19:20 +0200822 if (ic->flags & CF_SHUTR) {
823 si_rx_shut_blk(si);
824 return;
825 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100826
Willy Tarreau236c4292019-06-06 08:19:20 +0200827 /* Read not closed, update FD status and timeout for reads */
828 if (ic->flags & CF_DONT_READ)
829 si_rx_chan_blk(si);
830 else
831 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100832
Willy Tarreau236c4292019-06-06 08:19:20 +0200833 if (!channel_is_empty(ic)) {
834 /* stop reading, imposed by channel's policy or contents */
835 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200836 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200837 else {
838 /* (re)start reading and update timeout. Note: we don't recompute the timeout
839 * everytime we get here, otherwise it would risk never to expire. We only
840 * update it if is was not yet set. The stream socket handler will already
841 * have updated it if there has been a completed I/O.
842 */
843 si_rx_room_rdy(si);
844 }
845 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
846 ic->rex = TICK_ETERNITY;
847 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
848 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200849
Willy Tarreau236c4292019-06-06 08:19:20 +0200850 si_chk_rcv(si);
851}
852
853/* This function is designed to be called from within the stream handler to
854 * update the output channel's expiration timer and the stream interface's
855 * Tx flags based on the channel's flags. It needs to be called only once
856 * after the channel's flags have settled down, and before they are cleared,
857 * though it doesn't harm to call it as often as desired (it just slightly
858 * hurts performance). It must not be called from outside of the stream
859 * handler, as what it does will be used to compute the stream task's
860 * expiration.
861 */
862void si_update_tx(struct stream_interface *si)
863{
864 struct channel *oc = si_oc(si);
865 struct channel *ic = si_ic(si);
866
867 if (oc->flags & CF_SHUTW)
868 return;
869
870 /* Write not closed, update FD status and timeout for writes */
871 if (channel_is_empty(oc)) {
872 /* stop writing */
873 if (!(si->flags & SI_FL_WAIT_DATA)) {
874 if ((oc->flags & CF_SHUTW_NOW) == 0)
875 si->flags |= SI_FL_WAIT_DATA;
876 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200877 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200878 return;
879 }
880
881 /* (re)start writing and update timeout. Note: we don't recompute the timeout
882 * everytime we get here, otherwise it would risk never to expire. We only
883 * update it if is was not yet set. The stream socket handler will already
884 * have updated it if there has been a completed I/O.
885 */
886 si->flags &= ~SI_FL_WAIT_DATA;
887 if (!tick_isset(oc->wex)) {
888 oc->wex = tick_add_ifset(now_ms, oc->wto);
889 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
890 /* Note: depending on the protocol, we don't know if we're waiting
891 * for incoming data or not. So in order to prevent the socket from
892 * expiring read timeouts during writes, we refresh the read timeout,
893 * except if it was already infinite or if we have explicitly setup
894 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200895 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200896 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200897 }
898 }
899}
900
Willy Tarreau3b285d72019-06-06 08:20:17 +0200901/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
902 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
903 * be updated in case of success.
904 */
905void si_sync_send(struct stream_interface *si)
906{
907 struct channel *oc = si_oc(si);
908 struct conn_stream *cs;
909
910 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
911
912 if (oc->flags & CF_SHUTW)
913 return;
914
915 if (channel_is_empty(oc))
916 return;
917
918 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
919 return;
920
921 cs = objt_cs(si->end);
922 if (!cs)
923 return;
924
925 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING))
926 return;
927
928 if (cs->conn->flags & CO_FL_ERROR)
929 return;
930
Christopher Fauletab1300b2019-07-05 13:44:29 +0200931 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200932}
933
Willy Tarreaud14844a2018-11-08 18:15:29 +0100934/* Updates at once the channel flags, and timers of both stream interfaces of a
935 * same stream, to complete the work after the analysers, then updates the data
936 * layer below. This will ensure that any synchronous update performed at the
937 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200938 * Note that this does not change the stream interface's current state, though
939 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100940 */
941void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
942{
943 struct channel *req = si_ic(si_f);
944 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100945
946 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
947 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
948
Willy Tarreaud14844a2018-11-08 18:15:29 +0100949 si_f->prev_state = si_f->state;
950 si_b->prev_state = si_b->state;
951
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100952 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200953 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100954 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100955
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200956 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100957 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100958
959 /* stream ints are processed outside of process_stream() and must be
960 * handled at the latest moment.
961 */
962 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100963 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
964 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100965 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100966
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100967 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100968 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
969 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100970 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100971}
972
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200973/*
974 * This function performs a shutdown-read on a stream interface attached to
975 * a connection in a connected or init state (it does nothing for other
976 * states). It either shuts the read side or marks itself as closed. The buffer
977 * flags are updated to reflect the new state. If the stream interface has
978 * SI_FL_NOHALF, we also forward the close to the write side. If a control
979 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200980 * descriptors are then shutdown or closed accordingly. The function
981 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200982 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200983static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200984{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200985 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100986 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200987
Willy Tarreauabb5d422018-11-14 16:58:52 +0100988 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100989 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200990 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100991 ic->flags |= CF_SHUTR;
992 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200993
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200994 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200995 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200996
Willy Tarreau51d0a7e2019-01-31 19:09:59 +0100997 if (si->flags & SI_FL_KILL_CONN)
998 cs->flags |= CS_FL_KILL_CONN;
999
Willy Tarreau2bb4a962014-11-28 11:11:05 +01001000 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02001001 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001002 si->state = SI_ST_DIS;
1003 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001004 }
1005 else if (si->flags & SI_FL_NOHALF) {
1006 /* we want to immediately forward this close to the write side */
1007 return stream_int_shutw_conn(si);
1008 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001009}
1010
1011/*
1012 * This function performs a shutdown-write on a stream interface attached to
1013 * a connection in a connected or init state (it does nothing for other
1014 * states). It either shuts the write side or marks itself as closed. The
1015 * buffer flags are updated to reflect the new state. It does also close
1016 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001017 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001018 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001019static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001020{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001021 struct conn_stream *cs = __objt_cs(si->end);
1022 struct connection *conn = cs->conn;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001023 struct channel *ic = si_ic(si);
1024 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001025
Willy Tarreauafc8a222014-11-28 15:46:27 +01001026 oc->flags &= ~CF_SHUTW_NOW;
1027 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001028 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001029 oc->flags |= CF_SHUTW;
1030 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001031 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001032
Hongbo Longe39683c2017-03-10 18:41:51 +01001033 if (tick_isset(si->hcto)) {
1034 ic->rto = si->hcto;
1035 ic->rex = tick_add(now_ms, ic->rto);
1036 }
1037
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001038 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001039 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001040 case SI_ST_EST:
1041 /* we have to shut before closing, otherwise some short messages
1042 * may never leave the system, especially when there are remaining
1043 * unread data in the socket input buffer, or when nolinger is set.
1044 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1045 * no risk so we close both sides immediately.
1046 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001047 if (si->flags & SI_FL_KILL_CONN)
1048 cs->flags |= CS_FL_KILL_CONN;
1049
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001050 if (si->flags & SI_FL_ERR) {
1051 /* quick close, the socket is alredy shut anyway */
1052 }
1053 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001054 /* unclean data-layer shutdown, typically an aborted request
1055 * or a forwarded shutdown from a client to a server due to
1056 * option abortonclose. No need for the TLS layer to try to
1057 * emit a shutdown message.
1058 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001059 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001060 }
1061 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001062 /* clean data-layer shutdown. This only happens on the
1063 * frontend side, or on the backend side when forwarding
1064 * a client close in TCP mode or in HTTP TUNNEL mode
1065 * while option abortonclose is set. We want the TLS
1066 * layer to try to signal it to the peer before we close.
1067 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001068 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001069
Willy Tarreaua553ae92017-10-05 18:52:17 +02001070 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
1071 /* OK just a shutw, but we want the caller
1072 * to disable polling on this FD if exists.
1073 */
1074 conn_cond_update_polling(conn);
1075 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001076 }
1077 }
1078
1079 /* fall through */
1080 case SI_ST_CON:
1081 /* we may have to close a pending connection, and mark the
1082 * response buffer as shutr
1083 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001084 if (si->flags & SI_FL_KILL_CONN)
1085 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001086 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001087 /* fall through */
1088 case SI_ST_CER:
1089 case SI_ST_QUE:
1090 case SI_ST_TAR:
1091 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001092 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001093 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001094 si->flags &= ~SI_FL_NOLINGER;
1095 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001096 ic->flags |= CF_SHUTR;
1097 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001098 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001099 }
1100}
1101
Willy Tarreau46a8d922012-08-20 12:38:36 +02001102/* This function is used for inter-stream-interface calls. It is called by the
1103 * consumer to inform the producer side that it may be interested in checking
1104 * for free space in the buffer. Note that it intentionally does not update
1105 * timeouts, so that we can still check them later at wake-up. This function is
1106 * dedicated to connection-based stream interfaces.
1107 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001108static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001109{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001110 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001111 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001112 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001113}
1114
1115
Willy Tarreaude5722c2012-08-20 15:01:10 +02001116/* This function is used for inter-stream-interface calls. It is called by the
1117 * producer to inform the consumer side that it may be interested in checking
1118 * for data in the buffer. Note that it intentionally does not update timeouts,
1119 * so that we can still check them later at wake-up.
1120 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001121static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001122{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001123 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001124 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001125
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001126 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001127 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001128 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001129
Willy Tarreauafc8a222014-11-28 15:46:27 +01001130 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001131 return;
1132
Willy Tarreauafc8a222014-11-28 15:46:27 +01001133 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001134 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001135 return;
1136
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001137 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001138 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001139
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001140 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001141 /* Write error on the file descriptor */
Olivier Houchard6187cc32019-06-24 16:08:08 +02001142 if (si->state >= SI_ST_CON)
1143 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001144 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001145 }
1146
1147 /* OK, so now we know that some data might have been sent, and that we may
1148 * have to poll first. We have to do that too if the buffer is not empty.
1149 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001150 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001151 /* the connection is established but we can't write. Either the
1152 * buffer is empty, or we just refrain from sending because the
1153 * ->o limit was reached. Maybe we just wrote the last
1154 * chunk and need to close.
1155 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001156 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001157 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001158 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001159 si_shutw(si);
1160 goto out_wakeup;
1161 }
1162
Willy Tarreauafc8a222014-11-28 15:46:27 +01001163 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001164 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001165 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001166 }
1167 else {
1168 /* Otherwise there are remaining data to be sent in the buffer,
1169 * which means we have to poll before doing so.
1170 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001171 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001172 if (!tick_isset(oc->wex))
1173 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001174 }
1175
Willy Tarreauafc8a222014-11-28 15:46:27 +01001176 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1177 struct channel *ic = si_ic(si);
1178
Willy Tarreaude5722c2012-08-20 15:01:10 +02001179 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001180 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1181 !channel_is_empty(oc))
1182 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001183
Willy Tarreauafc8a222014-11-28 15:46:27 +01001184 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001185 /* Note: to prevent the client from expiring read timeouts
1186 * during writes, we refresh it. We only do this if the
1187 * interface is not configured for "independent streams",
1188 * because for some applications it's better not to do this,
1189 * for instance when continuously exchanging small amounts
1190 * of data which can full the socket buffers long before a
1191 * write timeout is detected.
1192 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001193 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001194 }
1195 }
1196
1197 /* in case of special condition (error, shutdown, end of write...), we
1198 * have to notify the task.
1199 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001200 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1201 ((oc->flags & CF_WAKE_WRITE) &&
1202 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001203 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001204 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001205 if (!(si->flags & SI_FL_DONT_WAKE))
1206 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001207 }
1208}
1209
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001210/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001211 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001212 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001213 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001214 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001215int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001216{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001217 struct connection *conn = cs->conn;
1218 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001219 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001220 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001221 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001222 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001223
Olivier Houchardf6535282018-08-31 17:29:12 +02001224 /* If another call to si_cs_recv() failed, and we subscribed to
1225 * recv events already, give up now.
1226 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001227 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001228 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001229
Willy Tarreauce323de2012-08-20 21:41:06 +02001230 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001231 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001232 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001233
Willy Tarreau54e917c2017-08-30 07:35:35 +02001234 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001235 if (cs->flags & CS_FL_EOS)
Willy Tarreau54e917c2017-08-30 07:35:35 +02001236 goto out_shutdown_r;
1237
Christopher Fauletf061e422018-12-07 14:51:20 +01001238 /* stop immediately on errors. Note that we DON'T want to stop on
1239 * POLL_ERR, as the poller might report a write error while there
1240 * are still data available in the recv buffer. This typically
1241 * happens when we send too large a request to a backend server
1242 * which rejects it before reading it all.
1243 */
1244 if (!(cs->flags & CS_FL_RCV_MORE)) {
1245 if (!conn_xprt_ready(conn))
1246 return 0;
1247 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
1248 return 1; // We want to make sure si_cs_wake() is called, so that process_strema is woken up, on failure
1249 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001250
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001251 /* prepare to detect if the mux needs more room */
1252 cs->flags &= ~CS_FL_WANT_ROOM;
1253
Willy Tarreau77e478c2018-06-19 07:03:14 +02001254 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001255 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001256 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001257 /* The buffer was empty and nothing was transferred for more
1258 * than one second. This was caused by a pause and not by
1259 * congestion. Reset any streaming mode to reduce latency.
1260 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001261 ic->xfer_small = 0;
1262 ic->xfer_large = 0;
1263 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001264 }
1265
Willy Tarreau96199b12012-08-24 00:46:52 +02001266 /* First, let's see if we may splice data across the channel without
1267 * using a buffer.
1268 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001269 if (conn->xprt->rcv_pipe && conn->mux->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001270 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1271 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001272 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001273 /* We're embarrassed, there are already data pending in
1274 * the buffer and we don't want to have them at two
1275 * locations at a time. Let's indicate we need some
1276 * place and ask the consumer to hurry.
1277 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001278 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001279 goto abort_splice;
1280 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001281
Willy Tarreauafc8a222014-11-28 15:46:27 +01001282 if (unlikely(ic->pipe == NULL)) {
1283 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1284 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001285 goto abort_splice;
1286 }
1287 }
1288
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001289 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001290 if (ret < 0) {
1291 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001292 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001293 goto abort_splice;
1294 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001295
Willy Tarreau96199b12012-08-24 00:46:52 +02001296 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001297 if (ic->to_forward != CHN_INFINITE_FORWARD)
1298 ic->to_forward -= ret;
1299 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001300 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001301 ic->flags |= CF_READ_PARTIAL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001302 if (si->state == SI_ST_CON)
1303 si->state = SI_ST_RDY;
Willy Tarreauce323de2012-08-20 21:41:06 +02001304 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001305
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001306 if (cs->flags & CS_FL_EOS)
Willy Tarreau96199b12012-08-24 00:46:52 +02001307 goto out_shutdown_r;
1308
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001309 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001310 return 1;
Willy Tarreau96199b12012-08-24 00:46:52 +02001311
Willy Tarreau61d39a02013-07-18 21:49:32 +02001312 if (conn->flags & CO_FL_WAIT_ROOM) {
1313 /* the pipe is full or we have read enough data that it
1314 * could soon be full. Let's stop before needing to poll.
1315 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001316 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001317 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001318 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001319
Willy Tarreauce323de2012-08-20 21:41:06 +02001320 /* splice not possible (anymore), let's go on on standard copy */
1321 }
Willy Tarreau81464b42018-11-15 15:52:53 +01001322 else {
1323 /* be sure not to block regular receive path below */
1324 conn->flags &= ~CO_FL_WAIT_ROOM;
1325 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001326
1327 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001328 if (ic->pipe && unlikely(!ic->pipe->data)) {
1329 put_pipe(ic->pipe);
1330 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001331 }
1332
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001333 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001334 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001335 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001336
Willy Tarreau61d39a02013-07-18 21:49:32 +02001337 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1338 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1339 * that if such an event is not handled above in splice, it will be handled here by
1340 * recv().
1341 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001342 while ((cs->flags & CS_FL_RCV_MORE) ||
1343 (!(conn->flags & (CO_FL_ERROR | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE)) &&
1344 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001345 /* <max> may be null. This is the mux responsibility to set
1346 * CS_FL_RCV_MORE on the CS if more space is needed.
1347 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001348 max = channel_recv_max(ic);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001349 ret = cs->conn->mux->rcv_buf(cs, &ic->buf, max, flags | (co_data(ic) ? CO_RFL_BUF_WET : 0));
Willy Tarreau674e0ad2018-12-05 13:45:41 +01001350
Olivier Houchardd247be02018-12-06 16:22:29 +01001351 if (cs->flags & CS_FL_WANT_ROOM)
Willy Tarreaudb398432018-11-15 11:08:52 +01001352 si_rx_room_blk(si);
Willy Tarreau6577b482017-12-10 21:19:33 +01001353
Christopher Fauleteffc3752018-10-31 08:53:54 +01001354 if (cs->flags & CS_FL_READ_PARTIAL) {
1355 if (tick_isset(ic->rex))
1356 ic->rex = tick_add_ifset(now_ms, ic->rto);
1357 cs->flags &= ~CS_FL_READ_PARTIAL;
1358 }
1359
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001360 if (ret <= 0) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001361 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001362 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001363
Olivier Houcharda254a372019-04-05 15:30:12 +02001364 if (si->flags & SI_FL_L7_RETRY) {
1365 struct htx *htx;
1366 struct htx_sl *sl;
1367
1368 htx = htxbuf(&ic->buf);
1369 if (htx) {
Christopher Faulet297fbb42019-05-13 14:41:27 +02001370 sl = http_get_stline(htx);
Olivier Houcharda254a372019-04-05 15:30:12 +02001371 if (sl && l7_status_match(si_strm(si)->be,
1372 sl->info.res.status)) {
1373 /* If we got a status for which we would
1374 * like to retry the request, empty
1375 * the buffer and pretend there's an
1376 * error on the channel.
1377 */
1378 ic->flags |= CF_READ_ERROR;
1379 htx_reset(htx);
1380 return 1;
1381 }
1382 }
1383 si->flags &= ~SI_FL_L7_RETRY;
1384 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001385 cur_read += ret;
1386
1387 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001388 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001389 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001390 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1391 if (fwd > ic->to_forward)
1392 fwd = ic->to_forward;
1393 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001394 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001395 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001396 }
1397
Willy Tarreauafc8a222014-11-28 15:46:27 +01001398 ic->flags |= CF_READ_PARTIAL;
1399 ic->total += ret;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001400 if (si->state == SI_ST_CON)
1401 si->state = SI_ST_RDY;
Willy Tarreauce323de2012-08-20 21:41:06 +02001402
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001403 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1404 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001405 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001406 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001407 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001408
1409 /* if too many bytes were missing from last read, it means that
1410 * it's pointless trying to read again because the system does
1411 * not have them in buffers.
1412 */
1413 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001414 /* if a streamer has read few data, it may be because we
1415 * have exhausted system buffers. It's not worth trying
1416 * again.
1417 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001418 if (ic->flags & CF_STREAMER) {
1419 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001420 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001421 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001422 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001423
1424 /* if we read a large block smaller than what we requested,
1425 * it's almost certain we'll never get anything more.
1426 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001427 if (ret >= global.tune.recv_enough) {
1428 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001429 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001430 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001431 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001432 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001433
1434 /* if we are waiting for more space, don't try to read more data
1435 * right now.
1436 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001437 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001438 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001439 } /* while !flags */
1440
Willy Tarreauffb12052018-11-15 16:06:02 +01001441 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001442 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001443 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001444 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001445 ic->xfer_large = 0;
1446 ic->xfer_small++;
1447 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001448 /* we have read less than half of the buffer in
1449 * one pass, and this happened at least 3 times.
1450 * This is definitely not a streamer.
1451 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001452 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001453 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001454 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001455 /* if the buffer has been at least half full twice,
1456 * we receive faster than we send, so at least it
1457 * is not a "fast streamer".
1458 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001459 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001460 }
1461 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001462 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001463 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001464 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001465 ic->xfer_small = 0;
1466 ic->xfer_large++;
1467 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001468 /* we call this buffer a fast streamer if it manages
1469 * to be filled in one call 3 consecutive times.
1470 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001471 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001472 }
1473 }
1474 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001475 ic->xfer_small = 0;
1476 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001477 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001478 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001479 }
1480
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001481 end_recv:
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001482 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001483 return 1;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001484
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001485 if (cs->flags & CS_FL_EOS)
Willy Tarreauce323de2012-08-20 21:41:06 +02001486 /* connection closed */
1487 goto out_shutdown_r;
1488
Willy Tarreau32742fd2018-11-14 14:07:59 +01001489 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001490 if (!si_rx_blocked(si)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001491 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001492 si_rx_endp_done(si);
1493 } else {
1494 si_rx_endp_more(si);
1495 }
Olivier Houchardf6535282018-08-31 17:29:12 +02001496
Olivier Houchard19a2e2d2019-06-07 18:10:52 +02001497 return (cur_read != 0) || si_rx_blocked(si) || (cs->flags & CS_FL_EOI);
Willy Tarreauce323de2012-08-20 21:41:06 +02001498
1499 out_shutdown_r:
Olivier Houchardfd0c2dc2018-12-13 15:38:16 +01001500 if (conn->flags & CO_FL_CONNECTED) {
1501 /* we received a shutdown */
1502 ic->flags |= CF_READ_NULL;
1503 if (ic->flags & CF_AUTO_CLOSE)
1504 channel_shutw_now(ic);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001505 stream_int_read0(si);
Olivier Houchardfd0c2dc2018-12-13 15:38:16 +01001506 }
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001507 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001508}
1509
1510/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001511 * This function propagates a null read received on a socket-based connection.
1512 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001513 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001514 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001515static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001516{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001517 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001518 struct channel *ic = si_ic(si);
1519 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001520
Willy Tarreauabb5d422018-11-14 16:58:52 +01001521 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001522 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001523 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001524 ic->flags |= CF_SHUTR;
1525 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001526
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001527 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001528 return;
1529
Willy Tarreauafc8a222014-11-28 15:46:27 +01001530 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001531 goto do_close;
1532
1533 if (si->flags & SI_FL_NOHALF) {
1534 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001535 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001536 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001537 goto do_close;
1538 }
1539
1540 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001541 return;
1542
1543 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001544 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001545 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001546
Willy Tarreauafc8a222014-11-28 15:46:27 +01001547 oc->flags &= ~CF_SHUTW_NOW;
1548 oc->flags |= CF_SHUTW;
1549 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001550
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001551 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001552
Olivier Houchardaacc4052019-05-21 17:43:50 +02001553 /* Don't change the state to SI_ST_DIS yet if we're still
1554 * in SI_ST_CON, otherwise it means sess_establish() hasn't
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001555 * been called yet, and so the analysers would not run. However
1556 * it's fine to switch to SI_ST_RDY as we have really validated
1557 * the connection.
Olivier Houchardaacc4052019-05-21 17:43:50 +02001558 */
1559 if (si->state == SI_ST_EST)
1560 si->state = SI_ST_DIS;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001561 else if (si->state == SI_ST_CON)
1562 si->state = SI_ST_RDY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001563 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001564 return;
1565}
1566
Willy Tarreau651e1822015-09-23 20:06:13 +02001567/* Callback to be used by applet handlers upon completion. It updates the stream
1568 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001569 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001570 * states.
1571 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001572void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001573{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001574 struct channel *ic = si_ic(si);
1575
1576 /* If the applet wants to write and the channel is closed, it's a
1577 * broken pipe and it must be reported.
1578 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001579 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001580 si->flags |= SI_FL_ERR;
1581
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001582 /* automatically mark the applet having data available if it reported
1583 * begin blocked by the channel.
1584 */
1585 if (si_rx_blocked(si))
1586 si_rx_endp_more(si);
1587
Willy Tarreau651e1822015-09-23 20:06:13 +02001588 /* update the stream-int, channels, and possibly wake the stream up */
1589 stream_int_notify(si);
Willy Tarreaue5f86492015-04-19 15:16:35 +02001590
Willy Tarreau32742fd2018-11-14 14:07:59 +01001591 /* stream_int_notify may have passed through chk_snd and released some
1592 * RXBLK flags. Process_stream will consider those flags to wake up the
1593 * appctx but in the case the task is not in runqueue we may have to
1594 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001595 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001596 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1597 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001598 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001599}
1600
1601/*
1602 * This function performs a shutdown-read on a stream interface attached to an
1603 * applet in a connected or init state (it does nothing for other states). It
1604 * either shuts the read side or marks itself as closed. The buffer flags are
1605 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1606 * we also forward the close to the write side. The owner task is woken up if
1607 * it exists.
1608 */
1609static void stream_int_shutr_applet(struct stream_interface *si)
1610{
1611 struct channel *ic = si_ic(si);
1612
Willy Tarreauabb5d422018-11-14 16:58:52 +01001613 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001614 if (ic->flags & CF_SHUTR)
1615 return;
1616 ic->flags |= CF_SHUTR;
1617 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001618
Willy Tarreau828824a2015-04-19 17:20:03 +02001619 /* Note: on shutr, we don't call the applet */
1620
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001621 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001622 return;
1623
1624 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001625 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001626 si->state = SI_ST_DIS;
1627 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001628 }
1629 else if (si->flags & SI_FL_NOHALF) {
1630 /* we want to immediately forward this close to the write side */
1631 return stream_int_shutw_applet(si);
1632 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001633}
1634
1635/*
1636 * This function performs a shutdown-write on a stream interface attached to an
1637 * applet in a connected or init state (it does nothing for other states). It
1638 * either shuts the write side or marks itself as closed. The buffer flags are
1639 * updated to reflect the new state. It does also close everything if the SI
1640 * was marked as being in error state. The owner task is woken up if it exists.
1641 */
1642static void stream_int_shutw_applet(struct stream_interface *si)
1643{
1644 struct channel *ic = si_ic(si);
1645 struct channel *oc = si_oc(si);
1646
1647 oc->flags &= ~CF_SHUTW_NOW;
1648 if (oc->flags & CF_SHUTW)
1649 return;
1650 oc->flags |= CF_SHUTW;
1651 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001652 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001653
Hongbo Longe39683c2017-03-10 18:41:51 +01001654 if (tick_isset(si->hcto)) {
1655 ic->rto = si->hcto;
1656 ic->rex = tick_add(now_ms, ic->rto);
1657 }
1658
Willy Tarreau828824a2015-04-19 17:20:03 +02001659 /* on shutw we always wake the applet up */
1660 appctx_wakeup(si_appctx(si));
1661
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001662 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001663 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001664 case SI_ST_EST:
1665 /* we have to shut before closing, otherwise some short messages
1666 * may never leave the system, especially when there are remaining
1667 * unread data in the socket input buffer, or when nolinger is set.
1668 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1669 * no risk so we close both sides immediately.
1670 */
1671 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1672 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1673 return;
1674
1675 /* fall through */
1676 case SI_ST_CON:
1677 case SI_ST_CER:
1678 case SI_ST_QUE:
1679 case SI_ST_TAR:
1680 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001681 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001682 si->state = SI_ST_DIS;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001683 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001684 si->flags &= ~SI_FL_NOLINGER;
1685 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001686 ic->flags |= CF_SHUTR;
1687 ic->rex = TICK_ETERNITY;
1688 si->exp = TICK_ETERNITY;
1689 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001690}
1691
1692/* chk_rcv function for applets */
1693static void stream_int_chk_rcv_applet(struct stream_interface *si)
1694{
1695 struct channel *ic = si_ic(si);
1696
1697 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1698 __FUNCTION__,
1699 si, si->state, ic->flags, si_oc(si)->flags);
1700
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001701 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001702 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001703 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001704 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001705}
1706
1707/* chk_snd function for applets */
1708static void stream_int_chk_snd_applet(struct stream_interface *si)
1709{
1710 struct channel *oc = si_oc(si);
1711
1712 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1713 __FUNCTION__,
1714 si, si->state, si_ic(si)->flags, oc->flags);
1715
1716 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1717 return;
1718
Willy Tarreau828824a2015-04-19 17:20:03 +02001719 /* we only wake the applet up if it was waiting for some data */
1720
1721 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001722 return;
1723
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001724 if (!tick_isset(oc->wex))
1725 oc->wex = tick_add_ifset(now_ms, oc->wto);
1726
Willy Tarreau828824a2015-04-19 17:20:03 +02001727 if (!channel_is_empty(oc)) {
1728 /* (re)start sending */
1729 appctx_wakeup(si_appctx(si));
1730 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001731}
1732
Willy Tarreaudded32d2008-11-30 19:48:07 +01001733/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001734 * Local variables:
1735 * c-indent-level: 8
1736 * c-basic-offset: 8
1737 * End:
1738 */