blob: 47cbd9693acab54f2f34ca1a82c3953628819eae [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 Tarreaud02cdd22013-12-15 10:23:20 +0100332 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200333 goto out_error;
334
Willy Tarreau2c6be842012-07-06 17:12:34 +0200335 /* If we have a PROXY line to send, we'll use this to validate the
336 * connection, in which case the connection is validated only once
337 * we've sent the whole proxy line. Otherwise we use connect().
338 */
Tim Duesterhus36839dc2019-02-26 17:09:51 +0100339 if (conn->send_proxy_ofs) {
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100340 const struct conn_stream *cs;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200341 int ret;
342
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100343 cs = cs_get_first(conn);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200344 /* The target server expects a PROXY line to be sent first.
345 * If the send_proxy_ofs is negative, it corresponds to the
346 * offset to start sending from then end of the proxy string
347 * (which is recomputed every time since it's constant). If
348 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200349 * We can only send a "normal" PROXY line when the connection
350 * is attached to a stream interface. Otherwise we can only
351 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200352 */
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100353
354 if (cs && cs->data_cb == &si_conn_cb) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200355 struct stream_interface *si = cs->data;
356 struct conn_stream *remote_cs = objt_cs(si_opposite(si)->end);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200357 ret = make_proxy_line(trash.area, trash.size,
358 objt_server(conn->target),
359 remote_cs ? remote_cs->conn : NULL);
Olivier Houchard661167d2019-05-29 17:08:03 +0200360 /* We may not have a conn_stream yet, if we don't
361 * know which mux to use, because it will be decided
362 * during the SSL handshake. In this case, there should
363 * be a session associated to the connection in
364 * conn->owner, and we know it is the session that
365 * initiated that connection, so we can just use
366 * its origin, which should contain the client
367 * connection.
368 */
369 } else if (!cs && conn->owner) {
370 struct session *sess = conn->owner;
371
372 ret = make_proxy_line(trash.area, trash.size,
373 objt_server(conn->target),
374 objt_conn(sess->origin));
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200375 }
376 else {
377 /* The target server expects a LOCAL line to be sent first. Retrieving
378 * local or remote addresses may fail until the connection is established.
379 */
Willy Tarreau7bb447c2019-07-17 11:40:51 +0200380 if (!conn_get_src(conn) || !conn_get_dst(conn))
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200381 goto out_wait;
382
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200383 ret = make_proxy_line(trash.area, trash.size,
384 objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200385 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200386
Willy Tarreau2c6be842012-07-06 17:12:34 +0200387 if (!ret)
388 goto out_error;
389
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200390 if (conn->send_proxy_ofs > 0)
391 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200392
Willy Tarreaua1a74742012-08-24 12:14:49 +0200393 /* we have to send trash from (ret+sp for -sp bytes). If the
394 * data layer has a pending write, we'll also set MSG_MORE.
395 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200396 ret = conn_sock_send(conn,
397 trash.area + ret + conn->send_proxy_ofs,
398 -conn->send_proxy_ofs,
Willy Tarreau19bc2012020-02-21 08:46:19 +0100399 (conn->subs && conn->subs->events & SUB_RETRY_SEND) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200400
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100401 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200402 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200403
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200404 conn->send_proxy_ofs += ret; /* becomes zero once complete */
405 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200406 goto out_wait;
407
408 /* OK we've sent the whole line, we're connected */
409 }
410
Willy Tarreaua1a74742012-08-24 12:14:49 +0200411 /* The connection is ready now, simply return and let the connection
412 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200413 */
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100414 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200416 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200417
418 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200419 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200420 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200421 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200422
423 out_wait:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200424 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200425}
426
Willy Tarreau27375622013-12-17 00:00:28 +0100427
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100428/* This function is the equivalent to si_update() except that it's
Willy Tarreau615f28b2015-09-23 18:40:09 +0200429 * designed to be called from outside the stream handlers, typically the lower
430 * layers (applets, connections) after I/O completion. After updating the stream
431 * interface and timeouts, it will try to forward what can be forwarded, then to
432 * wake the associated task up if an important event requires special handling.
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100433 * It may update SI_FL_WAIT_DATA and/or SI_FL_RXBLK_ROOM, that the callers are
Willy Tarreau0dfccb22018-10-25 13:55:20 +0200434 * encouraged to watch to take appropriate action.
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100435 * It should not be called from within the stream itself, si_update()
Willy Tarreau615f28b2015-09-23 18:40:09 +0200436 * is designed for this.
437 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100438static void stream_int_notify(struct stream_interface *si)
Willy Tarreau615f28b2015-09-23 18:40:09 +0200439{
440 struct channel *ic = si_ic(si);
441 struct channel *oc = si_oc(si);
Willy Tarreau47baeb82018-11-15 07:46:57 +0100442 struct stream_interface *sio = si_opposite(si);
Christopher Fauletd7607de2019-01-03 16:24:54 +0100443 struct task *task = si_task(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200444
445 /* process consumer side */
446 if (channel_is_empty(oc)) {
Olivier Houcharde9bed532017-11-16 17:49:25 +0100447 struct connection *conn = objt_cs(si->end) ? objt_cs(si->end)->conn : NULL;
448
Willy Tarreau615f28b2015-09-23 18:40:09 +0200449 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +0100450 (si->state == SI_ST_EST) && (!conn || !(conn->flags & (CO_FL_WAIT_XPRT | CO_FL_EARLY_SSL_HS))))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200451 si_shutw(si);
452 oc->wex = TICK_ETERNITY;
453 }
454
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100455 /* indicate that we may be waiting for data from the output channel or
456 * we're about to close and can't expect more data if SHUTW_NOW is there.
457 */
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200458 if (!(oc->flags & (CF_SHUTW|CF_SHUTW_NOW)))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200459 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100460 else if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW)
461 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200462
463 /* update OC timeouts and wake the other side up if it's waiting for room */
464 if (oc->flags & CF_WRITE_ACTIVITY) {
465 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
466 !channel_is_empty(oc))
467 if (tick_isset(oc->wex))
468 oc->wex = tick_add_ifset(now_ms, oc->wto);
469
470 if (!(si->flags & SI_FL_INDEP_STR))
471 if (tick_isset(ic->rex))
472 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf26c26c2018-11-12 16:11:08 +0100473 }
Willy Tarreau615f28b2015-09-23 18:40:09 +0200474
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100475 if (oc->flags & CF_DONT_READ)
476 si_rx_chan_blk(sio);
477 else
478 si_rx_chan_rdy(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200479
480 /* Notify the other side when we've injected data into the IC that
481 * needs to be forwarded. We can do fast-forwarding as soon as there
482 * are output data, but we avoid doing this if some of the data are
483 * not yet scheduled for being forwarded, because it is very likely
484 * that it will be done again immediately afterwards once the following
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100485 * data are parsed (eg: HTTP chunking). We only SI_FL_RXBLK_ROOM once
Willy Tarreau615f28b2015-09-23 18:40:09 +0200486 * we've emptied *some* of the output buffer, and not just when there
487 * is available room, because applets are often forced to stop before
488 * the buffer is full. We must not stop based on input data alone because
489 * an HTTP parser might need more data to complete the parsing.
490 */
491 if (!channel_is_empty(ic) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100492 (sio->flags & SI_FL_WAIT_DATA) &&
Willy Tarreau89b6a2b2018-11-18 15:46:10 +0100493 (!(ic->flags & CF_EXPECT_MORE) || c_full(ic) || ci_data(ic) == 0 || ic->pipe)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200494 int new_len, last_len;
495
Willy Tarreau77e478c2018-06-19 07:03:14 +0200496 last_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200497 if (ic->pipe)
498 last_len += ic->pipe->data;
499
Willy Tarreau47baeb82018-11-15 07:46:57 +0100500 si_chk_snd(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200501
Willy Tarreau77e478c2018-06-19 07:03:14 +0200502 new_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200503 if (ic->pipe)
504 new_len += ic->pipe->data;
505
506 /* check if the consumer has freed some space either in the
507 * buffer or in the pipe.
508 */
Willy Tarreau47baeb82018-11-15 07:46:57 +0100509 if (new_len < last_len)
Willy Tarreaudb398432018-11-15 11:08:52 +0100510 si_rx_room_rdy(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200511 }
512
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100513 if (!(ic->flags & CF_DONT_READ))
514 si_rx_chan_rdy(si);
515
Willy Tarreau47baeb82018-11-15 07:46:57 +0100516 si_chk_rcv(si);
517 si_chk_rcv(sio);
518
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100519 if (si_rx_blocked(si)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200520 ic->rex = TICK_ETERNITY;
521 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100522 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL)) == CF_READ_PARTIAL) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200523 /* we must re-enable reading if si_chk_snd() has freed some space */
524 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
525 ic->rex = tick_add_ifset(now_ms, ic->rto);
526 }
527
528 /* wake the task up only when needed */
529 if (/* changes on the production side */
530 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200531 !si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200532 (si->flags & SI_FL_ERR) ||
533 ((ic->flags & CF_READ_PARTIAL) &&
Christopher Faulet297d3e22019-03-22 14:16:14 +0100534 ((ic->flags & CF_EOI) || !ic->to_forward || sio->state != SI_ST_EST)) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200535
536 /* changes on the consumption side */
537 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
Willy Tarreauede3d882018-10-24 17:17:56 +0200538 ((oc->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreau615f28b2015-09-23 18:40:09 +0200539 ((oc->flags & CF_SHUTW) ||
Willy Tarreau78f5ff82018-12-19 11:00:00 +0100540 (((oc->flags & CF_WAKE_WRITE) ||
541 !(oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW|CF_SHUTW))) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100542 (sio->state != SI_ST_EST ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200543 (channel_is_empty(oc) && !oc->to_forward)))))) {
Christopher Fauletd7607de2019-01-03 16:24:54 +0100544 task_wakeup(task, TASK_WOKEN_IO);
545 }
546 else {
547 /* Update expiration date for the task and requeue it */
548 task->expire = tick_first((tick_is_expired(task->expire, now_ms) ? 0 : task->expire),
549 tick_first(tick_first(ic->rex, ic->wex),
550 tick_first(oc->rex, oc->wex)));
Willy Tarreau45bcb372019-08-01 18:51:38 +0200551
552 task->expire = tick_first(task->expire, ic->analyse_exp);
553 task->expire = tick_first(task->expire, oc->analyse_exp);
554
555 if (si->exp)
556 task->expire = tick_first(task->expire, si->exp);
557
558 if (sio->exp)
559 task->expire = tick_first(task->expire, sio->exp);
560
Christopher Fauletd7607de2019-01-03 16:24:54 +0100561 task_queue(task);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200562 }
563 if (ic->flags & CF_READ_ACTIVITY)
564 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200565}
566
567
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200568/* Called by I/O handlers after completion.. It propagates
Willy Tarreau651e1822015-09-23 20:06:13 +0200569 * connection flags to the stream interface, updates the stream (which may or
570 * may not take this opportunity to try to forward data), then update the
571 * connection's polling based on the channels and stream interface's final
572 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200573 */
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200574static int si_cs_process(struct conn_stream *cs)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200575{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200576 struct connection *conn = cs->conn;
577 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100578 struct channel *ic = si_ic(si);
579 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200580
Olivier Houchardc7ffa912018-08-28 19:37:41 +0200581 /* If we have data to send, try it now */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100582 if (!channel_is_empty(oc) && !(si->wait_event.events & SUB_RETRY_SEND))
Willy Tarreau908d26f2018-10-25 14:02:47 +0200583 si_cs_send(cs);
584
Willy Tarreau651e1822015-09-23 20:06:13 +0200585 /* First step, report to the stream-int what was detected at the
586 * connection layer : errors and connection establishment.
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200587 * Only add SI_FL_ERR if we're connected, or we're attempting to
588 * connect, we may get there because we got woken up, but only run
589 * after process_stream() noticed there were an error, and decided
590 * to retry to connect, the connection may still have CO_FL_ERROR,
591 * and we don't want to add SI_FL_ERR back
Christopher Faulet36b536d2019-11-20 11:56:33 +0100592 *
593 * Note: This test is only required because si_cs_process is also the SI
594 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
595 * care of it.
Willy Tarreau651e1822015-09-23 20:06:13 +0200596 */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200597 if (si->state >= SI_ST_CON &&
598 (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR))
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200599 si->flags |= SI_FL_ERR;
600
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200601 /* If we had early data, and the handshake ended, then
602 * we can remove the flag, and attempt to wake the task up,
603 * in the event there's an analyser waiting for the end of
604 * the handshake.
605 */
Willy Tarreau911db9b2020-01-23 16:27:54 +0100606 if (!(conn->flags & (CO_FL_WAIT_XPRT | CO_FL_EARLY_SSL_HS)) &&
Olivier Houchard6fa63d92017-11-27 18:41:32 +0100607 (cs->flags & CS_FL_WAIT_FOR_HS)) {
608 cs->flags &= ~CS_FL_WAIT_FOR_HS;
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200609 task_wakeup(si_task(si), TASK_WOKEN_MSG);
610 }
611
Willy Tarreau7ab22adb2019-06-05 14:53:22 +0200612 if (!si_state_in(si->state, SI_SB_EST|SI_SB_DIS|SI_SB_CLO) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +0100613 (conn->flags & CO_FL_WAIT_XPRT) == 0) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200614 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100615 oc->flags |= CF_WRITE_NULL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +0200616 if (si->state == SI_ST_CON)
617 si->state = SI_ST_RDY;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200618 }
619
Christopher Faulet297d3e22019-03-22 14:16:14 +0100620 /* Report EOI on the channel if it was reached from the mux point of
Christopher Faulet36b536d2019-11-20 11:56:33 +0100621 * view.
622 *
623 * Note: This test is only required because si_cs_process is also the SI
624 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
625 * care of it.
626 */
Christopher Faulet297d3e22019-03-22 14:16:14 +0100627 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200628 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100629
Willy Tarreau651e1822015-09-23 20:06:13 +0200630 /* Second step : update the stream-int and channels, try to forward any
631 * pending data, then possibly wake the stream up based on the new
632 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200633 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200634 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +0200635 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200636 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200637}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200638
Willy Tarreau5368d802012-08-21 18:22:06 +0200639/*
640 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200641 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800642 * caller to commit polling changes. The caller should check conn->flags
643 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200644 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200645int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200646{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200647 struct connection *conn = cs->conn;
648 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100649 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200650 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200651 int did_send = 0;
652
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100653 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200654 /* We're probably there because the tasklet was woken up,
655 * but process_stream() ran before, detected there were an
656 * error and put the si back to SI_ST_TAR. There's still
657 * CO_FL_ERROR on the connection but we don't want to add
658 * SI_FL_ERR back, so give up
659 */
660 if (si->state < SI_ST_CON)
661 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100662 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200663 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100664 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200665
Christopher Faulet328ed222019-09-23 15:57:29 +0200666 /* We're already waiting to be able to send, give up */
667 if (si->wait_event.events & SUB_RETRY_SEND)
668 return 0;
669
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200670 /* we might have been called just after an asynchronous shutw */
Willy Tarreauf22758d2020-01-23 18:25:23 +0100671 if (oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200672 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200673
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200674 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
675 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet86162db2019-07-05 11:49:11 +0200676 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200677 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200678
Willy Tarreauafc8a222014-11-28 15:46:27 +0100679 if (!oc->pipe->data) {
680 put_pipe(oc->pipe);
681 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200682 }
683
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100684 if (oc->pipe)
685 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200686 }
687
688 /* At this point, the pipe is empty, but we may still have data pending
689 * in the normal buffer.
690 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100691 if (co_data(oc)) {
692 /* when we're here, we already know that there is no spliced
693 * data left, and that there are sendable buffered data.
694 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200695
Willy Tarreau5368d802012-08-21 18:22:06 +0200696 /* check if we want to inform the kernel that we're interested in
697 * sending more data after this call. We want this if :
698 * - we're about to close after this last send and want to merge
699 * the ongoing FIN with the last segment.
700 * - we know we can't send everything at once and must get back
701 * here because of unaligned data
702 * - there is still a finite amount of data to forward
703 * The test is arranged so that the most common case does only 2
704 * tests.
705 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100706 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200707
Willy Tarreauafc8a222014-11-28 15:46:27 +0100708 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
709 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau4ac49282017-10-17 16:33:46 +0200710 (oc->flags & CF_EXPECT_MORE))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100711 ((oc->flags & CF_ISRESP) &&
712 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100713 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200714
Willy Tarreauafc8a222014-11-28 15:46:27 +0100715 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100716 send_flag |= CO_SFL_STREAMER;
717
Olivier Houcharda254a372019-04-05 15:30:12 +0200718 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
Olivier Houchardce1a0292019-05-17 15:38:29 +0200719 struct stream *s = si_strm(si);
Olivier Houcharda254a372019-04-05 15:30:12 +0200720 /* If we want to be able to do L7 retries, copy
721 * the data we're about to send, so that we are able
722 * to resend them if needed
723 */
724 /* Try to allocate a buffer if we had none.
725 * If it fails, the next test will just
726 * disable the l7 retries by setting
727 * l7_conn_retries to 0.
728 */
Olivier Houchardce1a0292019-05-17 15:38:29 +0200729 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
Olivier Houcharda254a372019-04-05 15:30:12 +0200730 si->flags &= ~SI_FL_L7_RETRY;
731 else {
732 if (b_is_null(&si->l7_buffer))
733 b_alloc(&si->l7_buffer);
734 if (b_is_null(&si->l7_buffer))
735 si->flags &= ~SI_FL_L7_RETRY;
736 else {
737 memcpy(b_orig(&si->l7_buffer),
738 b_orig(&oc->buf),
739 b_size(&oc->buf));
740 si->l7_buffer.head = co_data(oc);
741 b_add(&si->l7_buffer, co_data(oc));
742 }
743
744 }
745 }
746
Olivier Houcharded0f2072018-08-16 15:41:52 +0200747 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800748 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200749 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200750 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200751 c_realign_if_empty(oc);
752
753 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800754 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100755 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800756 }
Godbache68e02d2013-10-11 15:48:29 +0800757 /* if some data remain in the buffer, it's only because the
758 * system buffers are full, we will try next time.
759 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200760 }
Godbache68e02d2013-10-11 15:48:29 +0800761 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100762
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100763 end:
Christopher Faulet86162db2019-07-05 11:49:11 +0200764 if (did_send) {
765 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
766 if (si->state == SI_ST_CON)
767 si->state = SI_ST_RDY;
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200768
769 si_rx_room_rdy(si_opposite(si));
Christopher Faulet86162db2019-07-05 11:49:11 +0200770 }
771
772 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
773 si->flags |= SI_FL_ERR;
774 return 1;
775 }
776
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200777 /* 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 +0100778 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100779 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200780 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200781}
782
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100783/* This is the ->process() function for any stream-interface's wait_event task.
784 * It's assigned during the stream-interface's initialization, for any type of
785 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
786 * stream interface, as the presence of the CS is checked there.
787 */
Olivier Houchard91894cb2018-08-02 18:06:28 +0200788struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned short state)
789{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200790 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200791 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200792 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200793
794 if (!cs)
795 return NULL;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100796
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100797 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200798 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100799 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200800 ret |= si_cs_recv(cs);
801 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200802 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200803
Willy Tarreaua64c7032019-08-01 14:17:02 +0200804 stream_release_buffers(si_strm(si));
Olivier Houchard91894cb2018-08-02 18:06:28 +0200805 return (NULL);
806}
807
Willy Tarreau25f13102015-09-24 11:32:22 +0200808/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200809 * update the input channel's expiration timer and the stream interface's
810 * Rx flags based on the channel's flags. It needs to be called only once
811 * after the channel's flags have settled down, and before they are cleared,
812 * though it doesn't harm to call it as often as desired (it just slightly
813 * hurts performance). It must not be called from outside of the stream
814 * handler, as what it does will be used to compute the stream task's
815 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200816 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200817void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200818{
819 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200820
Willy Tarreau236c4292019-06-06 08:19:20 +0200821 if (ic->flags & CF_SHUTR) {
822 si_rx_shut_blk(si);
823 return;
824 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100825
Willy Tarreau236c4292019-06-06 08:19:20 +0200826 /* Read not closed, update FD status and timeout for reads */
827 if (ic->flags & CF_DONT_READ)
828 si_rx_chan_blk(si);
829 else
830 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100831
Willy Tarreau236c4292019-06-06 08:19:20 +0200832 if (!channel_is_empty(ic)) {
833 /* stop reading, imposed by channel's policy or contents */
834 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200835 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200836 else {
837 /* (re)start reading and update timeout. Note: we don't recompute the timeout
838 * everytime we get here, otherwise it would risk never to expire. We only
839 * update it if is was not yet set. The stream socket handler will already
840 * have updated it if there has been a completed I/O.
841 */
842 si_rx_room_rdy(si);
843 }
844 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
845 ic->rex = TICK_ETERNITY;
846 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
847 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200848
Willy Tarreau236c4292019-06-06 08:19:20 +0200849 si_chk_rcv(si);
850}
851
852/* This function is designed to be called from within the stream handler to
853 * update the output channel's expiration timer and the stream interface's
854 * Tx flags based on the channel's flags. It needs to be called only once
855 * after the channel's flags have settled down, and before they are cleared,
856 * though it doesn't harm to call it as often as desired (it just slightly
857 * hurts performance). It must not be called from outside of the stream
858 * handler, as what it does will be used to compute the stream task's
859 * expiration.
860 */
861void si_update_tx(struct stream_interface *si)
862{
863 struct channel *oc = si_oc(si);
864 struct channel *ic = si_ic(si);
865
866 if (oc->flags & CF_SHUTW)
867 return;
868
869 /* Write not closed, update FD status and timeout for writes */
870 if (channel_is_empty(oc)) {
871 /* stop writing */
872 if (!(si->flags & SI_FL_WAIT_DATA)) {
873 if ((oc->flags & CF_SHUTW_NOW) == 0)
874 si->flags |= SI_FL_WAIT_DATA;
875 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200876 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200877 return;
878 }
879
880 /* (re)start writing and update timeout. Note: we don't recompute the timeout
881 * everytime we get here, otherwise it would risk never to expire. We only
882 * update it if is was not yet set. The stream socket handler will already
883 * have updated it if there has been a completed I/O.
884 */
885 si->flags &= ~SI_FL_WAIT_DATA;
886 if (!tick_isset(oc->wex)) {
887 oc->wex = tick_add_ifset(now_ms, oc->wto);
888 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
889 /* Note: depending on the protocol, we don't know if we're waiting
890 * for incoming data or not. So in order to prevent the socket from
891 * expiring read timeouts during writes, we refresh the read timeout,
892 * except if it was already infinite or if we have explicitly setup
893 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200894 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200895 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200896 }
897 }
898}
899
Willy Tarreau3b285d72019-06-06 08:20:17 +0200900/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
901 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
902 * be updated in case of success.
903 */
904void si_sync_send(struct stream_interface *si)
905{
906 struct channel *oc = si_oc(si);
907 struct conn_stream *cs;
908
909 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
910
911 if (oc->flags & CF_SHUTW)
912 return;
913
914 if (channel_is_empty(oc))
915 return;
916
917 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
918 return;
919
920 cs = objt_cs(si->end);
Olivier Houchard8af03b32020-01-22 17:34:54 +0100921 if (!cs || !cs->conn->mux)
Willy Tarreau3b285d72019-06-06 08:20:17 +0200922 return;
923
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200924 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200925}
926
Willy Tarreaud14844a2018-11-08 18:15:29 +0100927/* Updates at once the channel flags, and timers of both stream interfaces of a
928 * same stream, to complete the work after the analysers, then updates the data
929 * layer below. This will ensure that any synchronous update performed at the
930 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200931 * Note that this does not change the stream interface's current state, though
932 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100933 */
934void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
935{
936 struct channel *req = si_ic(si_f);
937 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100938
939 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
940 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
941
Willy Tarreaud14844a2018-11-08 18:15:29 +0100942 si_f->prev_state = si_f->state;
943 si_b->prev_state = si_b->state;
944
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100945 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200946 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100947 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100948
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200949 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100950 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100951
952 /* stream ints are processed outside of process_stream() and must be
953 * handled at the latest moment.
954 */
955 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100956 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
957 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100958 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100959
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100960 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100961 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
962 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100963 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100964}
965
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200966/*
967 * This function performs a shutdown-read on a stream interface attached to
968 * a connection in a connected or init state (it does nothing for other
969 * states). It either shuts the read side or marks itself as closed. The buffer
970 * flags are updated to reflect the new state. If the stream interface has
971 * SI_FL_NOHALF, we also forward the close to the write side. If a control
972 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200973 * descriptors are then shutdown or closed accordingly. The function
974 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200975 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200976static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200977{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200978 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100979 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200980
Willy Tarreauabb5d422018-11-14 16:58:52 +0100981 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100982 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200983 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100984 ic->flags |= CF_SHUTR;
985 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200986
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200987 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200988 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200989
Willy Tarreau51d0a7e2019-01-31 19:09:59 +0100990 if (si->flags & SI_FL_KILL_CONN)
991 cs->flags |= CS_FL_KILL_CONN;
992
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100993 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +0200994 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200995 si->state = SI_ST_DIS;
996 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200997 }
998 else if (si->flags & SI_FL_NOHALF) {
999 /* we want to immediately forward this close to the write side */
1000 return stream_int_shutw_conn(si);
1001 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001002}
1003
1004/*
1005 * This function performs a shutdown-write on a stream interface attached to
1006 * a connection in a connected or init state (it does nothing for other
1007 * states). It either shuts the write side or marks itself as closed. The
1008 * buffer flags are updated to reflect the new state. It does also close
1009 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001010 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001011 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001012static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001013{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001014 struct conn_stream *cs = __objt_cs(si->end);
1015 struct connection *conn = cs->conn;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001016 struct channel *ic = si_ic(si);
1017 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001018
Willy Tarreauafc8a222014-11-28 15:46:27 +01001019 oc->flags &= ~CF_SHUTW_NOW;
1020 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001021 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001022 oc->flags |= CF_SHUTW;
1023 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001024 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001025
Hongbo Longe39683c2017-03-10 18:41:51 +01001026 if (tick_isset(si->hcto)) {
1027 ic->rto = si->hcto;
1028 ic->rex = tick_add(now_ms, ic->rto);
1029 }
1030
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001031 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001032 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001033 case SI_ST_EST:
1034 /* we have to shut before closing, otherwise some short messages
1035 * may never leave the system, especially when there are remaining
1036 * unread data in the socket input buffer, or when nolinger is set.
1037 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1038 * no risk so we close both sides immediately.
1039 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001040 if (si->flags & SI_FL_KILL_CONN)
1041 cs->flags |= CS_FL_KILL_CONN;
1042
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001043 if (si->flags & SI_FL_ERR) {
1044 /* quick close, the socket is alredy shut anyway */
1045 }
1046 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001047 /* unclean data-layer shutdown, typically an aborted request
1048 * or a forwarded shutdown from a client to a server due to
1049 * option abortonclose. No need for the TLS layer to try to
1050 * emit a shutdown message.
1051 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001052 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001053 }
1054 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001055 /* clean data-layer shutdown. This only happens on the
1056 * frontend side, or on the backend side when forwarding
1057 * a client close in TCP mode or in HTTP TUNNEL mode
1058 * while option abortonclose is set. We want the TLS
1059 * layer to try to signal it to the peer before we close.
1060 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001061 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001062
Willy Tarreaua553ae92017-10-05 18:52:17 +02001063 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
1064 /* OK just a shutw, but we want the caller
1065 * to disable polling on this FD if exists.
1066 */
1067 conn_cond_update_polling(conn);
1068 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001069 }
1070 }
1071
1072 /* fall through */
1073 case SI_ST_CON:
1074 /* we may have to close a pending connection, and mark the
1075 * response buffer as shutr
1076 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001077 if (si->flags & SI_FL_KILL_CONN)
1078 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001079 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001080 /* fall through */
1081 case SI_ST_CER:
1082 case SI_ST_QUE:
1083 case SI_ST_TAR:
1084 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001085 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001086 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001087 si->flags &= ~SI_FL_NOLINGER;
1088 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001089 ic->flags |= CF_SHUTR;
1090 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001091 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001092 }
1093}
1094
Willy Tarreau46a8d922012-08-20 12:38:36 +02001095/* This function is used for inter-stream-interface calls. It is called by the
1096 * consumer to inform the producer side that it may be interested in checking
1097 * for free space in the buffer. Note that it intentionally does not update
1098 * timeouts, so that we can still check them later at wake-up. This function is
1099 * dedicated to connection-based stream interfaces.
1100 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001101static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001102{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001103 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001104 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001105 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001106}
1107
1108
Willy Tarreaude5722c2012-08-20 15:01:10 +02001109/* This function is used for inter-stream-interface calls. It is called by the
1110 * producer to inform the consumer side that it may be interested in checking
1111 * for data in the buffer. Note that it intentionally does not update timeouts,
1112 * so that we can still check them later at wake-up.
1113 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001114static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001115{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001116 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001117 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001118
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001119 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001120 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001121 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001122
Willy Tarreauafc8a222014-11-28 15:46:27 +01001123 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001124 return;
1125
Willy Tarreauafc8a222014-11-28 15:46:27 +01001126 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001127 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001128 return;
1129
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001130 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001131 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001132
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001133 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001134 /* Write error on the file descriptor */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +02001135 if (si->state >= SI_ST_CON)
1136 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001137 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001138 }
1139
1140 /* OK, so now we know that some data might have been sent, and that we may
1141 * have to poll first. We have to do that too if the buffer is not empty.
1142 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001143 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001144 /* the connection is established but we can't write. Either the
1145 * buffer is empty, or we just refrain from sending because the
1146 * ->o limit was reached. Maybe we just wrote the last
1147 * chunk and need to close.
1148 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001149 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001150 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001151 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001152 si_shutw(si);
1153 goto out_wakeup;
1154 }
1155
Willy Tarreauafc8a222014-11-28 15:46:27 +01001156 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001157 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001158 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001159 }
1160 else {
1161 /* Otherwise there are remaining data to be sent in the buffer,
1162 * which means we have to poll before doing so.
1163 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001164 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001165 if (!tick_isset(oc->wex))
1166 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001167 }
1168
Willy Tarreauafc8a222014-11-28 15:46:27 +01001169 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1170 struct channel *ic = si_ic(si);
1171
Willy Tarreaude5722c2012-08-20 15:01:10 +02001172 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001173 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1174 !channel_is_empty(oc))
1175 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001176
Willy Tarreauafc8a222014-11-28 15:46:27 +01001177 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001178 /* Note: to prevent the client from expiring read timeouts
1179 * during writes, we refresh it. We only do this if the
1180 * interface is not configured for "independent streams",
1181 * because for some applications it's better not to do this,
1182 * for instance when continuously exchanging small amounts
1183 * of data which can full the socket buffers long before a
1184 * write timeout is detected.
1185 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001186 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001187 }
1188 }
1189
1190 /* in case of special condition (error, shutdown, end of write...), we
1191 * have to notify the task.
1192 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001193 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1194 ((oc->flags & CF_WAKE_WRITE) &&
1195 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001196 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001197 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001198 if (!(si->flags & SI_FL_DONT_WAKE))
1199 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001200 }
1201}
1202
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001203/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001204 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001205 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001206 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001207 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001208int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001209{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001210 struct connection *conn = cs->conn;
1211 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001212 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001213 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001214 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001215 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001216
Christopher Faulet04400bc2019-10-25 10:21:01 +02001217 /* If not established yet, do nothing. */
1218 if (si->state != SI_ST_EST)
1219 return 0;
1220
Olivier Houchardf6535282018-08-31 17:29:12 +02001221 /* If another call to si_cs_recv() failed, and we subscribed to
1222 * recv events already, give up now.
1223 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001224 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001225 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001226
Willy Tarreauce323de2012-08-20 21:41:06 +02001227 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001228 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001229 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001230
Willy Tarreau54e917c2017-08-30 07:35:35 +02001231 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001232 if (cs->flags & CS_FL_EOS)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001233 goto end_recv;
Willy Tarreau54e917c2017-08-30 07:35:35 +02001234
Christopher Fauletf061e422018-12-07 14:51:20 +01001235 /* stop immediately on errors. Note that we DON'T want to stop on
1236 * POLL_ERR, as the poller might report a write error while there
1237 * are still data available in the recv buffer. This typically
1238 * happens when we send too large a request to a backend server
1239 * which rejects it before reading it all.
1240 */
1241 if (!(cs->flags & CS_FL_RCV_MORE)) {
1242 if (!conn_xprt_ready(conn))
1243 return 0;
1244 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001245 goto end_recv;
Christopher Fauletf061e422018-12-07 14:51:20 +01001246 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001247
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001248 /* prepare to detect if the mux needs more room */
1249 cs->flags &= ~CS_FL_WANT_ROOM;
1250
Willy Tarreau77e478c2018-06-19 07:03:14 +02001251 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001252 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001253 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001254 /* The buffer was empty and nothing was transferred for more
1255 * than one second. This was caused by a pause and not by
1256 * congestion. Reset any streaming mode to reduce latency.
1257 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001258 ic->xfer_small = 0;
1259 ic->xfer_large = 0;
1260 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001261 }
1262
Willy Tarreau96199b12012-08-24 00:46:52 +02001263 /* First, let's see if we may splice data across the channel without
1264 * using a buffer.
1265 */
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001266 if (cs->flags & CS_FL_MAY_SPLICE &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001267 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1268 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001269 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001270 /* We're embarrassed, there are already data pending in
1271 * the buffer and we don't want to have them at two
1272 * locations at a time. Let's indicate we need some
1273 * place and ask the consumer to hurry.
1274 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001275 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001276 goto abort_splice;
1277 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001278
Willy Tarreauafc8a222014-11-28 15:46:27 +01001279 if (unlikely(ic->pipe == NULL)) {
1280 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1281 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001282 goto abort_splice;
1283 }
1284 }
1285
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001286 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001287 if (ret < 0) {
1288 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001289 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001290 goto abort_splice;
1291 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001292
Willy Tarreau96199b12012-08-24 00:46:52 +02001293 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001294 if (ic->to_forward != CHN_INFINITE_FORWARD)
1295 ic->to_forward -= ret;
1296 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001297 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001298 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001299 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001300
Christopher Faulet36b536d2019-11-20 11:56:33 +01001301 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_EOS|CS_FL_ERROR))
1302 goto end_recv;
Willy Tarreau96199b12012-08-24 00:46:52 +02001303
Willy Tarreau61d39a02013-07-18 21:49:32 +02001304 if (conn->flags & CO_FL_WAIT_ROOM) {
1305 /* the pipe is full or we have read enough data that it
1306 * could soon be full. Let's stop before needing to poll.
1307 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001308 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001309 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001310 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001311
Willy Tarreauce323de2012-08-20 21:41:06 +02001312 /* splice not possible (anymore), let's go on on standard copy */
1313 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001314
1315 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001316 if (ic->pipe && unlikely(!ic->pipe->data)) {
1317 put_pipe(ic->pipe);
1318 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001319 }
1320
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001321 if (ic->pipe && ic->to_forward && !(flags & CO_RFL_BUF_FLUSH) && cs->flags & CS_FL_MAY_SPLICE) {
Willy Tarreauc640ef12019-12-03 18:13:04 +01001322 /* don't break splicing by reading, but still call rcv_buf()
1323 * to pass the flag.
1324 */
1325 goto done_recv;
1326 }
1327
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001328 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001329 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001330 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001331
Willy Tarreau61d39a02013-07-18 21:49:32 +02001332 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1333 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1334 * that if such an event is not handled above in splice, it will be handled here by
1335 * recv().
1336 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001337 while ((cs->flags & CS_FL_RCV_MORE) ||
Willy Tarreau93c9f592020-01-17 17:24:30 +01001338 (!(conn->flags & (CO_FL_ERROR | CO_FL_HANDSHAKE)) &&
Olivier Houchardc490efd2018-12-04 15:46:16 +01001339 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001340 /* <max> may be null. This is the mux responsibility to set
1341 * CS_FL_RCV_MORE on the CS if more space is needed.
1342 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001343 max = channel_recv_max(ic);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001344 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 +01001345
Olivier Houchardd247be02018-12-06 16:22:29 +01001346 if (cs->flags & CS_FL_WANT_ROOM)
Willy Tarreaudb398432018-11-15 11:08:52 +01001347 si_rx_room_blk(si);
Willy Tarreau6577b482017-12-10 21:19:33 +01001348
Christopher Fauleteffc3752018-10-31 08:53:54 +01001349 if (cs->flags & CS_FL_READ_PARTIAL) {
1350 if (tick_isset(ic->rex))
1351 ic->rex = tick_add_ifset(now_ms, ic->rto);
1352 cs->flags &= ~CS_FL_READ_PARTIAL;
1353 }
1354
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001355 if (ret <= 0) {
Willy Tarreau1ac5f202019-12-03 18:08:45 +01001356 /* if we refrained from reading because we asked for a
1357 * flush to satisfy rcv_pipe(), we must not subscribe
1358 * and instead report that there's not enough room
1359 * here to proceed.
1360 */
1361 if (flags & CO_RFL_BUF_FLUSH)
1362 si_rx_room_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001363 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001364 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001365
Christopher Faulet48726b72020-01-09 14:31:13 +01001366 /* L7 retries enabled and maximum connection retries not reached */
1367 if ((si->flags & SI_FL_L7_RETRY) && si->conn_retries) {
Olivier Houcharda254a372019-04-05 15:30:12 +02001368 struct htx *htx;
1369 struct htx_sl *sl;
1370
1371 htx = htxbuf(&ic->buf);
1372 if (htx) {
Christopher Faulet297fbb42019-05-13 14:41:27 +02001373 sl = http_get_stline(htx);
Olivier Houcharda254a372019-04-05 15:30:12 +02001374 if (sl && l7_status_match(si_strm(si)->be,
1375 sl->info.res.status)) {
1376 /* If we got a status for which we would
1377 * like to retry the request, empty
1378 * the buffer and pretend there's an
1379 * error on the channel.
1380 */
1381 ic->flags |= CF_READ_ERROR;
1382 htx_reset(htx);
1383 return 1;
1384 }
1385 }
1386 si->flags &= ~SI_FL_L7_RETRY;
1387 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001388 cur_read += ret;
1389
1390 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001391 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001392 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001393 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1394 if (fwd > ic->to_forward)
1395 fwd = ic->to_forward;
1396 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001397 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001398 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001399 }
1400
Willy Tarreauafc8a222014-11-28 15:46:27 +01001401 ic->flags |= CF_READ_PARTIAL;
1402 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001403
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001404 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1405 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001406 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001407 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001408 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001409
1410 /* if too many bytes were missing from last read, it means that
1411 * it's pointless trying to read again because the system does
1412 * not have them in buffers.
1413 */
1414 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001415 /* if a streamer has read few data, it may be because we
1416 * have exhausted system buffers. It's not worth trying
1417 * again.
1418 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001419 if (ic->flags & CF_STREAMER) {
1420 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001421 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001422 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001423 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001424
1425 /* if we read a large block smaller than what we requested,
1426 * it's almost certain we'll never get anything more.
1427 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001428 if (ret >= global.tune.recv_enough) {
1429 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001430 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001431 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001432 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001433 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001434
1435 /* if we are waiting for more space, don't try to read more data
1436 * right now.
1437 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001438 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001439 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001440 } /* while !flags */
1441
Willy Tarreauffb12052018-11-15 16:06:02 +01001442 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001443 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001444 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001445 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001446 ic->xfer_large = 0;
1447 ic->xfer_small++;
1448 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001449 /* we have read less than half of the buffer in
1450 * one pass, and this happened at least 3 times.
1451 * This is definitely not a streamer.
1452 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001453 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001454 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001455 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001456 /* if the buffer has been at least half full twice,
1457 * we receive faster than we send, so at least it
1458 * is not a "fast streamer".
1459 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001460 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001461 }
1462 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001463 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001464 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001465 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001466 ic->xfer_small = 0;
1467 ic->xfer_large++;
1468 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001469 /* we call this buffer a fast streamer if it manages
1470 * to be filled in one call 3 consecutive times.
1471 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001472 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001473 }
1474 }
1475 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001476 ic->xfer_small = 0;
1477 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001478 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001479 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001480 }
1481
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001482 end_recv:
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001483 ret = (cur_read != 0);
1484
Christopher Faulet36b536d2019-11-20 11:56:33 +01001485 /* Report EOI on the channel if it was reached from the mux point of
1486 * view. */
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001487 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI)) {
Christopher Faulet36b536d2019-11-20 11:56:33 +01001488 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001489 ret = 1;
1490 }
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001491
Christopher Faulet36b536d2019-11-20 11:56:33 +01001492 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
1493 cs->flags |= CS_FL_ERROR;
1494 si->flags |= SI_FL_ERR;
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001495 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001496 }
1497 else if (cs->flags & CS_FL_EOS) {
Willy Tarreau18955db2020-01-23 16:32:24 +01001498 /* we received a shutdown */
1499 ic->flags |= CF_READ_NULL;
1500 if (ic->flags & CF_AUTO_CLOSE)
1501 channel_shutw_now(ic);
1502 stream_int_read0(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001503 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001504 }
1505 else if (!si_rx_blocked(si)) {
1506 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001507 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001508 si_rx_endp_done(si);
1509 } else {
1510 si_rx_endp_more(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001511 ret = 1;
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001512 }
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001513 return ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001514}
1515
1516/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001517 * This function propagates a null read received on a socket-based connection.
1518 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001519 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001520 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001521static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001522{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001523 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001524 struct channel *ic = si_ic(si);
1525 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001526
Willy Tarreauabb5d422018-11-14 16:58:52 +01001527 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001528 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001529 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001530 ic->flags |= CF_SHUTR;
1531 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001532
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001533 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001534 return;
1535
Willy Tarreauafc8a222014-11-28 15:46:27 +01001536 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001537 goto do_close;
1538
1539 if (si->flags & SI_FL_NOHALF) {
1540 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001541 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001542 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001543 goto do_close;
1544 }
1545
1546 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001547 return;
1548
1549 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001550 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001551 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001552
Willy Tarreauafc8a222014-11-28 15:46:27 +01001553 oc->flags &= ~CF_SHUTW_NOW;
1554 oc->flags |= CF_SHUTW;
1555 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001556
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001557 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001558
Christopher Faulet04400bc2019-10-25 10:21:01 +02001559 si->state = SI_ST_DIS;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001560 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001561 return;
1562}
1563
Willy Tarreau651e1822015-09-23 20:06:13 +02001564/* Callback to be used by applet handlers upon completion. It updates the stream
1565 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001566 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001567 * states.
1568 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001569void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001570{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001571 struct channel *ic = si_ic(si);
1572
1573 /* If the applet wants to write and the channel is closed, it's a
1574 * broken pipe and it must be reported.
1575 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001576 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001577 si->flags |= SI_FL_ERR;
1578
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001579 /* automatically mark the applet having data available if it reported
1580 * begin blocked by the channel.
1581 */
1582 if (si_rx_blocked(si))
1583 si_rx_endp_more(si);
1584
Willy Tarreau651e1822015-09-23 20:06:13 +02001585 /* update the stream-int, channels, and possibly wake the stream up */
1586 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +02001587 stream_release_buffers(si_strm(si));
Willy Tarreaue5f86492015-04-19 15:16:35 +02001588
Willy Tarreau32742fd2018-11-14 14:07:59 +01001589 /* stream_int_notify may have passed through chk_snd and released some
1590 * RXBLK flags. Process_stream will consider those flags to wake up the
1591 * appctx but in the case the task is not in runqueue we may have to
1592 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001593 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001594 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1595 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001596 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001597}
1598
1599/*
1600 * This function performs a shutdown-read on a stream interface attached to an
1601 * applet in a connected or init state (it does nothing for other states). It
1602 * either shuts the read side or marks itself as closed. The buffer flags are
1603 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1604 * we also forward the close to the write side. The owner task is woken up if
1605 * it exists.
1606 */
1607static void stream_int_shutr_applet(struct stream_interface *si)
1608{
1609 struct channel *ic = si_ic(si);
1610
Willy Tarreauabb5d422018-11-14 16:58:52 +01001611 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001612 if (ic->flags & CF_SHUTR)
1613 return;
1614 ic->flags |= CF_SHUTR;
1615 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001616
Willy Tarreau828824a2015-04-19 17:20:03 +02001617 /* Note: on shutr, we don't call the applet */
1618
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001619 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001620 return;
1621
1622 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001623 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001624 si->state = SI_ST_DIS;
1625 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001626 }
1627 else if (si->flags & SI_FL_NOHALF) {
1628 /* we want to immediately forward this close to the write side */
1629 return stream_int_shutw_applet(si);
1630 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001631}
1632
1633/*
1634 * This function performs a shutdown-write on a stream interface attached to an
1635 * applet in a connected or init state (it does nothing for other states). It
1636 * either shuts the write side or marks itself as closed. The buffer flags are
1637 * updated to reflect the new state. It does also close everything if the SI
1638 * was marked as being in error state. The owner task is woken up if it exists.
1639 */
1640static void stream_int_shutw_applet(struct stream_interface *si)
1641{
1642 struct channel *ic = si_ic(si);
1643 struct channel *oc = si_oc(si);
1644
1645 oc->flags &= ~CF_SHUTW_NOW;
1646 if (oc->flags & CF_SHUTW)
1647 return;
1648 oc->flags |= CF_SHUTW;
1649 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001650 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001651
Hongbo Longe39683c2017-03-10 18:41:51 +01001652 if (tick_isset(si->hcto)) {
1653 ic->rto = si->hcto;
1654 ic->rex = tick_add(now_ms, ic->rto);
1655 }
1656
Willy Tarreau828824a2015-04-19 17:20:03 +02001657 /* on shutw we always wake the applet up */
1658 appctx_wakeup(si_appctx(si));
1659
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001660 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001661 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001662 case SI_ST_EST:
1663 /* we have to shut before closing, otherwise some short messages
1664 * may never leave the system, especially when there are remaining
1665 * unread data in the socket input buffer, or when nolinger is set.
1666 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1667 * no risk so we close both sides immediately.
1668 */
1669 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1670 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1671 return;
1672
1673 /* fall through */
1674 case SI_ST_CON:
1675 case SI_ST_CER:
1676 case SI_ST_QUE:
1677 case SI_ST_TAR:
1678 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001679 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001680 si->state = SI_ST_DIS;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001681 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001682 si->flags &= ~SI_FL_NOLINGER;
1683 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001684 ic->flags |= CF_SHUTR;
1685 ic->rex = TICK_ETERNITY;
1686 si->exp = TICK_ETERNITY;
1687 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001688}
1689
1690/* chk_rcv function for applets */
1691static void stream_int_chk_rcv_applet(struct stream_interface *si)
1692{
1693 struct channel *ic = si_ic(si);
1694
1695 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1696 __FUNCTION__,
1697 si, si->state, ic->flags, si_oc(si)->flags);
1698
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001699 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001700 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001701 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001702 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001703}
1704
1705/* chk_snd function for applets */
1706static void stream_int_chk_snd_applet(struct stream_interface *si)
1707{
1708 struct channel *oc = si_oc(si);
1709
1710 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1711 __FUNCTION__,
1712 si, si->state, si_ic(si)->flags, oc->flags);
1713
1714 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1715 return;
1716
Willy Tarreau828824a2015-04-19 17:20:03 +02001717 /* we only wake the applet up if it was waiting for some data */
1718
1719 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001720 return;
1721
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001722 if (!tick_isset(oc->wex))
1723 oc->wex = tick_add_ifset(now_ms, oc->wto);
1724
Willy Tarreau828824a2015-04-19 17:20:03 +02001725 if (!channel_is_empty(oc)) {
1726 /* (re)start sending */
1727 appctx_wakeup(si_appctx(si));
1728 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001729}
1730
Willy Tarreaudded32d2008-11-30 19:48:07 +01001731/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001732 * Local variables:
1733 * c-indent-level: 8
1734 * c-basic-offset: 8
1735 * End:
1736 */