blob: d3080660935af43c9d7aa3751e027417b396bdc6 [file] [log] [blame]
Willy Tarreaucff64112008-11-03 06:26:53 +01001/*
2 * Functions managing stream_interface structures
3 *
Willy Tarreauf873d752012-05-11 17:47:17 +02004 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
Willy Tarreaucff64112008-11-03 06:26:53 +01005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <errno.h>
14#include <fcntl.h>
15#include <stdio.h>
16#include <stdlib.h>
17
18#include <sys/socket.h>
19#include <sys/stat.h>
20#include <sys/types.h>
21
Willy Tarreaubf883e02014-11-25 21:10:35 +010022#include <common/buffer.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010023#include <common/compat.h>
24#include <common/config.h>
25#include <common/debug.h>
26#include <common/standard.h>
27#include <common/ticks.h>
28#include <common/time.h>
29
Willy Tarreau8a8d83b2015-04-13 13:24:54 +020030#include <proto/applet.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020031#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020032#include <proto/connection.h>
Olivier Houcharda254a372019-04-05 15:30:12 +020033#include <proto/http_htx.h>
Olivier Houchard9aaf7782017-09-13 18:30:23 +020034#include <proto/mux_pt.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020035#include <proto/pipe.h>
Olivier Houcharda254a372019-04-05 15:30:12 +020036#include <proto/proxy.h>
Willy Tarreau87b09662015-04-03 00:22:06 +020037#include <proto/stream.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020038#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010039#include <proto/task.h>
40
Willy Tarreaufd31e532012-07-23 18:24:25 +020041#include <types/pipe.h>
42
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010043/* functions used by default on a detached stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020044static void stream_int_shutr(struct stream_interface *si);
45static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020046static void stream_int_chk_rcv(struct stream_interface *si);
47static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010048
49/* functions used on a conn_stream-based stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020050static void stream_int_shutr_conn(struct stream_interface *si);
51static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020052static void stream_int_chk_rcv_conn(struct stream_interface *si);
53static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010054
55/* functions used on an applet-based stream-interface */
Willy Tarreaud45b9f82015-04-13 16:30:14 +020056static void stream_int_shutr_applet(struct stream_interface *si);
57static void stream_int_shutw_applet(struct stream_interface *si);
58static void stream_int_chk_rcv_applet(struct stream_interface *si);
59static void stream_int_chk_snd_applet(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010060
61/* last read notification */
62static void stream_int_read0(struct stream_interface *si);
63
64/* post-IO notification callback */
65static void stream_int_notify(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020066
Willy Tarreauc5788912012-08-24 18:12:41 +020067/* stream-interface operations for embedded tasks */
68struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020069 .chk_rcv = stream_int_chk_rcv,
70 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020071 .shutr = stream_int_shutr,
72 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020073};
74
Willy Tarreauc5788912012-08-24 18:12:41 +020075/* stream-interface operations for connections */
76struct si_ops si_conn_ops = {
Willy Tarreauc5788912012-08-24 18:12:41 +020077 .chk_rcv = stream_int_chk_rcv_conn,
78 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020079 .shutr = stream_int_shutr_conn,
80 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020081};
82
Willy Tarreaud45b9f82015-04-13 16:30:14 +020083/* stream-interface operations for connections */
84struct si_ops si_applet_ops = {
Willy Tarreaud45b9f82015-04-13 16:30:14 +020085 .chk_rcv = stream_int_chk_rcv_applet,
86 .chk_snd = stream_int_chk_snd_applet,
87 .shutr = stream_int_shutr_applet,
88 .shutw = stream_int_shutw_applet,
89};
90
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010091
92/* Functions used to communicate with a conn_stream. The first two may be used
93 * directly, the last one is mostly a wake callback.
94 */
95int si_cs_recv(struct conn_stream *cs);
96int si_cs_send(struct conn_stream *cs);
97static int si_cs_process(struct conn_stream *cs);
98
99
Willy Tarreau74beec32012-10-03 00:41:04 +0200100struct data_cb si_conn_cb = {
Olivier Houchard21df6cc2018-09-14 23:21:44 +0200101 .wake = si_cs_process,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100102 .name = "STRM",
Willy Tarreauc5788912012-08-24 18:12:41 +0200103};
104
Willy Tarreaucff64112008-11-03 06:26:53 +0100105/*
106 * This function only has to be called once after a wakeup event in case of
107 * suspected timeout. It controls the stream interface timeouts and sets
108 * si->flags accordingly. It does NOT close anything, as this timeout may
109 * be used for any purpose. It returns 1 if the timeout fired, otherwise
110 * zero.
111 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100112int si_check_timeouts(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100113{
114 if (tick_is_expired(si->exp, now_ms)) {
115 si->flags |= SI_FL_EXP;
116 return 1;
117 }
118 return 0;
119}
120
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100121/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100122void si_report_error(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100123{
124 if (!si->err_type)
125 si->err_type = SI_ET_DATA_ERR;
126
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100127 si_oc(si)->flags |= CF_WRITE_ERROR;
128 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100129}
130
131/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100132 * Returns a message to the client ; the connection is shut down for read,
133 * and the request is cleared so that no server connection can be initiated.
134 * The buffer is marked for read shutdown on the other side to protect the
135 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100136 * "chunk". If it is null, then an empty message is used. The reply buffer does
137 * not need to be empty before this, and its contents will not be overwritten.
138 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100139 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100140void si_retnclose(struct stream_interface *si,
Willy Tarreau83061a82018-07-13 11:56:34 +0200141 const struct buffer *msg)
Willy Tarreaudded32d2008-11-30 19:48:07 +0100142{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100143 struct channel *ic = si_ic(si);
144 struct channel *oc = si_oc(si);
145
146 channel_auto_read(ic);
147 channel_abort(ic);
148 channel_auto_close(ic);
149 channel_erase(ic);
150 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100151
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200152 if (likely(msg && msg->data))
153 co_inject(oc, msg->area, msg->data);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100154
Willy Tarreauafc8a222014-11-28 15:46:27 +0100155 oc->wex = tick_add_ifset(now_ms, oc->wto);
156 channel_auto_read(oc);
157 channel_auto_close(oc);
158 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100159}
160
Willy Tarreau4a36b562012-08-06 19:31:45 +0200161/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200162 * This function performs a shutdown-read on a detached stream interface in a
163 * connected or init state (it does nothing for other states). It either shuts
164 * the read side or marks itself as closed. The buffer flags are updated to
165 * reflect the new state. If the stream interface has SI_FL_NOHALF, we also
166 * forward the close to the write side. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200167 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200168static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200169{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100170 struct channel *ic = si_ic(si);
171
Willy Tarreauabb5d422018-11-14 16:58:52 +0100172 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100173 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200174 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100175 ic->flags |= CF_SHUTR;
176 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200177
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200178 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200179 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200180
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100181 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200182 si->state = SI_ST_DIS;
183 si->exp = TICK_ETERNITY;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200184 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200185 else if (si->flags & SI_FL_NOHALF) {
186 /* we want to immediately forward this close to the write side */
187 return stream_int_shutw(si);
188 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200189
Willy Tarreau4a36b562012-08-06 19:31:45 +0200190 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100191 if (!(si->flags & SI_FL_DONT_WAKE))
192 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200193}
194
Willy Tarreau4a36b562012-08-06 19:31:45 +0200195/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200196 * This function performs a shutdown-write on a detached stream interface in a
197 * connected or init state (it does nothing for other states). It either shuts
198 * the write side or marks itself as closed. The buffer flags are updated to
199 * reflect the new state. It does also close everything if the SI was marked as
200 * being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200201 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200202static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200203{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100204 struct channel *ic = si_ic(si);
205 struct channel *oc = si_oc(si);
206
207 oc->flags &= ~CF_SHUTW_NOW;
208 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200209 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100210 oc->flags |= CF_SHUTW;
211 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +0100212 si_done_get(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200213
Hongbo Longe39683c2017-03-10 18:41:51 +0100214 if (tick_isset(si->hcto)) {
215 ic->rto = si->hcto;
216 ic->rex = tick_add(now_ms, ic->rto);
217 }
218
Willy Tarreaufb90d942009-09-05 20:57:35 +0200219 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200220 case SI_ST_RDY:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200221 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200222 /* we have to shut before closing, otherwise some short messages
223 * may never leave the system, especially when there are remaining
224 * unread data in the socket input buffer, or when nolinger is set.
225 * However, if SI_FL_NOLINGER is explicitly set, we know there is
226 * no risk so we close both sides immediately.
227 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200228 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100229 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200230 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200231
232 /* fall through */
233 case SI_ST_CON:
234 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100235 case SI_ST_QUE:
236 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200237 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200238 si->state = SI_ST_DIS;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200239 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +0100240 si->flags &= ~SI_FL_NOLINGER;
241 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100242 ic->flags |= CF_SHUTR;
243 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244 si->exp = TICK_ETERNITY;
245 }
246
Willy Tarreau4a36b562012-08-06 19:31:45 +0200247 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100248 if (!(si->flags & SI_FL_DONT_WAKE))
249 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250}
251
252/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200253static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200254{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100255 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200256
Willy Tarreauafc8a222014-11-28 15:46:27 +0100257 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200258 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100259 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200260
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200261 if (ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200262 /* stop reading */
Willy Tarreaudb398432018-11-15 11:08:52 +0100263 si_rx_room_blk(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200264 }
265 else {
266 /* (re)start reading */
Willy Tarreau3c39a7d2019-06-14 14:42:29 +0200267 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau07373b82014-11-28 12:08:47 +0100268 if (!(si->flags & SI_FL_DONT_WAKE))
269 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200270 }
271}
272
273/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200274static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200275{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100276 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200277
Willy Tarreauafc8a222014-11-28 15:46:27 +0100278 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200279 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100280 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200281
Willy Tarreauafc8a222014-11-28 15:46:27 +0100282 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200283 return;
284
285 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100286 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200287 return;
288
289 /* Otherwise there are remaining data to be sent in the buffer,
290 * so we tell the handler.
291 */
292 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100293 if (!tick_isset(oc->wex))
294 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200295
Willy Tarreau07373b82014-11-28 12:08:47 +0100296 if (!(si->flags & SI_FL_DONT_WAKE))
297 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200298}
299
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200300/* Register an applet to handle a stream_interface as a new appctx. The SI will
301 * wake it up everytime it is solicited. The appctx must be deleted by the task
302 * handler using si_release_endpoint(), possibly from within the function itself.
303 * It also pre-initializes the applet's context and returns it (or NULL in case
304 * it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200305 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100306struct appctx *si_register_handler(struct stream_interface *si, struct applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200307{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100308 struct appctx *appctx;
309
Willy Tarreau07373b82014-11-28 12:08:47 +0100310 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200311
Willy Tarreaua7513f52015-04-05 00:15:26 +0200312 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100313 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100314 return NULL;
315
Willy Tarreau0cd3bd62018-11-06 18:46:37 +0100316 si_cant_get(si);
Willy Tarreau828824a2015-04-19 17:20:03 +0200317 appctx_wakeup(appctx);
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100318 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200319}
320
Willy Tarreau2c6be842012-07-06 17:12:34 +0200321/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200322 * established. It returns 0 if it fails in a fatal way or needs to poll to go
323 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200324 * flags (the bit is provided in <flag> by the caller). It is designed to be
325 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200326 * Note that it can emit a PROXY line by relying on the other end's address
327 * when the connection is attached to a stream interface, or by resolving the
328 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200329 */
330int conn_si_send_proxy(struct connection *conn, unsigned int flag)
331{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200332 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200333 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200334 goto out_error;
335
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100336 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200337 goto out_error;
338
Willy Tarreau2c6be842012-07-06 17:12:34 +0200339 /* If we have a PROXY line to send, we'll use this to validate the
340 * connection, in which case the connection is validated only once
341 * we've sent the whole proxy line. Otherwise we use connect().
342 */
Tim Duesterhus36839dc2019-02-26 17:09:51 +0100343 if (conn->send_proxy_ofs) {
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100344 const struct conn_stream *cs;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200345 int ret;
346
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100347 cs = cs_get_first(conn);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200348 /* The target server expects a PROXY line to be sent first.
349 * If the send_proxy_ofs is negative, it corresponds to the
350 * offset to start sending from then end of the proxy string
351 * (which is recomputed every time since it's constant). If
352 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200353 * We can only send a "normal" PROXY line when the connection
354 * is attached to a stream interface. Otherwise we can only
355 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200356 */
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100357
358 if (cs && cs->data_cb == &si_conn_cb) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200359 struct stream_interface *si = cs->data;
360 struct conn_stream *remote_cs = objt_cs(si_opposite(si)->end);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200361 ret = make_proxy_line(trash.area, trash.size,
362 objt_server(conn->target),
363 remote_cs ? remote_cs->conn : NULL);
Olivier Houchard661167d2019-05-29 17:08:03 +0200364 /* We may not have a conn_stream yet, if we don't
365 * know which mux to use, because it will be decided
366 * during the SSL handshake. In this case, there should
367 * be a session associated to the connection in
368 * conn->owner, and we know it is the session that
369 * initiated that connection, so we can just use
370 * its origin, which should contain the client
371 * connection.
372 */
373 } else if (!cs && conn->owner) {
374 struct session *sess = conn->owner;
375
376 ret = make_proxy_line(trash.area, trash.size,
377 objt_server(conn->target),
378 objt_conn(sess->origin));
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200379 }
380 else {
381 /* The target server expects a LOCAL line to be sent first. Retrieving
382 * local or remote addresses may fail until the connection is established.
383 */
Willy Tarreau7bb447c2019-07-17 11:40:51 +0200384 if (!conn_get_src(conn) || !conn_get_dst(conn))
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200385 goto out_wait;
386
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200387 ret = make_proxy_line(trash.area, trash.size,
388 objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200389 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200390
Willy Tarreau2c6be842012-07-06 17:12:34 +0200391 if (!ret)
392 goto out_error;
393
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200394 if (conn->send_proxy_ofs > 0)
395 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396
Willy Tarreaua1a74742012-08-24 12:14:49 +0200397 /* we have to send trash from (ret+sp for -sp bytes). If the
398 * data layer has a pending write, we'll also set MSG_MORE.
399 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200400 ret = conn_sock_send(conn,
401 trash.area + ret + conn->send_proxy_ofs,
402 -conn->send_proxy_ofs,
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200403 (conn->flags & CO_FL_XPRT_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200404
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100405 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200406 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200407
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200408 conn->send_proxy_ofs += ret; /* becomes zero once complete */
409 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200410 goto out_wait;
411
412 /* OK we've sent the whole line, we're connected */
413 }
414
Willy Tarreaua1a74742012-08-24 12:14:49 +0200415 /* The connection is ready now, simply return and let the connection
416 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200417 */
418 if (conn->flags & CO_FL_WAIT_L4_CONN)
419 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200420 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200421 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200422
423 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200424 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200425 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200426 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200427
428 out_wait:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200429 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200430}
431
Willy Tarreau27375622013-12-17 00:00:28 +0100432
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100433/* This function is the equivalent to si_update() except that it's
Willy Tarreau615f28b2015-09-23 18:40:09 +0200434 * designed to be called from outside the stream handlers, typically the lower
435 * layers (applets, connections) after I/O completion. After updating the stream
436 * interface and timeouts, it will try to forward what can be forwarded, then to
437 * wake the associated task up if an important event requires special handling.
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100438 * It may update SI_FL_WAIT_DATA and/or SI_FL_RXBLK_ROOM, that the callers are
Willy Tarreau0dfccb22018-10-25 13:55:20 +0200439 * encouraged to watch to take appropriate action.
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100440 * It should not be called from within the stream itself, si_update()
Willy Tarreau615f28b2015-09-23 18:40:09 +0200441 * is designed for this.
442 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100443static void stream_int_notify(struct stream_interface *si)
Willy Tarreau615f28b2015-09-23 18:40:09 +0200444{
445 struct channel *ic = si_ic(si);
446 struct channel *oc = si_oc(si);
Willy Tarreau47baeb82018-11-15 07:46:57 +0100447 struct stream_interface *sio = si_opposite(si);
Christopher Fauletd7607de2019-01-03 16:24:54 +0100448 struct task *task = si_task(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200449
450 /* process consumer side */
451 if (channel_is_empty(oc)) {
Olivier Houcharde9bed532017-11-16 17:49:25 +0100452 struct connection *conn = objt_cs(si->end) ? objt_cs(si->end)->conn : NULL;
453
Willy Tarreau615f28b2015-09-23 18:40:09 +0200454 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Olivier Houcharde9bed532017-11-16 17:49:25 +0100455 (si->state == SI_ST_EST) && (!conn || !(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS))))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200456 si_shutw(si);
457 oc->wex = TICK_ETERNITY;
458 }
459
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100460 /* indicate that we may be waiting for data from the output channel or
461 * we're about to close and can't expect more data if SHUTW_NOW is there.
462 */
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200463 if (!(oc->flags & (CF_SHUTW|CF_SHUTW_NOW)))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200464 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100465 else if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW)
466 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200467
468 /* update OC timeouts and wake the other side up if it's waiting for room */
469 if (oc->flags & CF_WRITE_ACTIVITY) {
470 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
471 !channel_is_empty(oc))
472 if (tick_isset(oc->wex))
473 oc->wex = tick_add_ifset(now_ms, oc->wto);
474
475 if (!(si->flags & SI_FL_INDEP_STR))
476 if (tick_isset(ic->rex))
477 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf26c26c2018-11-12 16:11:08 +0100478 }
Willy Tarreau615f28b2015-09-23 18:40:09 +0200479
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100480 if (oc->flags & CF_DONT_READ)
481 si_rx_chan_blk(sio);
482 else
483 si_rx_chan_rdy(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200484
485 /* Notify the other side when we've injected data into the IC that
486 * needs to be forwarded. We can do fast-forwarding as soon as there
487 * are output data, but we avoid doing this if some of the data are
488 * not yet scheduled for being forwarded, because it is very likely
489 * that it will be done again immediately afterwards once the following
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100490 * data are parsed (eg: HTTP chunking). We only SI_FL_RXBLK_ROOM once
Willy Tarreau615f28b2015-09-23 18:40:09 +0200491 * we've emptied *some* of the output buffer, and not just when there
492 * is available room, because applets are often forced to stop before
493 * the buffer is full. We must not stop based on input data alone because
494 * an HTTP parser might need more data to complete the parsing.
495 */
496 if (!channel_is_empty(ic) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100497 (sio->flags & SI_FL_WAIT_DATA) &&
Willy Tarreau89b6a2b2018-11-18 15:46:10 +0100498 (!(ic->flags & CF_EXPECT_MORE) || c_full(ic) || ci_data(ic) == 0 || ic->pipe)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200499 int new_len, last_len;
500
Willy Tarreau77e478c2018-06-19 07:03:14 +0200501 last_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200502 if (ic->pipe)
503 last_len += ic->pipe->data;
504
Willy Tarreau47baeb82018-11-15 07:46:57 +0100505 si_chk_snd(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200506
Willy Tarreau77e478c2018-06-19 07:03:14 +0200507 new_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200508 if (ic->pipe)
509 new_len += ic->pipe->data;
510
511 /* check if the consumer has freed some space either in the
512 * buffer or in the pipe.
513 */
Willy Tarreau47baeb82018-11-15 07:46:57 +0100514 if (new_len < last_len)
Willy Tarreaudb398432018-11-15 11:08:52 +0100515 si_rx_room_rdy(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200516 }
517
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100518 if (!(ic->flags & CF_DONT_READ))
519 si_rx_chan_rdy(si);
520
Willy Tarreau47baeb82018-11-15 07:46:57 +0100521 si_chk_rcv(si);
522 si_chk_rcv(sio);
523
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100524 if (si_rx_blocked(si)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200525 ic->rex = TICK_ETERNITY;
526 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100527 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL)) == CF_READ_PARTIAL) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200528 /* we must re-enable reading if si_chk_snd() has freed some space */
529 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
530 ic->rex = tick_add_ifset(now_ms, ic->rto);
531 }
532
533 /* wake the task up only when needed */
534 if (/* changes on the production side */
535 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200536 !si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200537 (si->flags & SI_FL_ERR) ||
538 ((ic->flags & CF_READ_PARTIAL) &&
Christopher Faulet297d3e22019-03-22 14:16:14 +0100539 ((ic->flags & CF_EOI) || !ic->to_forward || sio->state != SI_ST_EST)) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200540
541 /* changes on the consumption side */
542 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
Willy Tarreauede3d882018-10-24 17:17:56 +0200543 ((oc->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreau615f28b2015-09-23 18:40:09 +0200544 ((oc->flags & CF_SHUTW) ||
Willy Tarreau78f5ff82018-12-19 11:00:00 +0100545 (((oc->flags & CF_WAKE_WRITE) ||
546 !(oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW|CF_SHUTW))) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100547 (sio->state != SI_ST_EST ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200548 (channel_is_empty(oc) && !oc->to_forward)))))) {
Christopher Fauletd7607de2019-01-03 16:24:54 +0100549 task_wakeup(task, TASK_WOKEN_IO);
550 }
551 else {
552 /* Update expiration date for the task and requeue it */
553 task->expire = tick_first((tick_is_expired(task->expire, now_ms) ? 0 : task->expire),
554 tick_first(tick_first(ic->rex, ic->wex),
555 tick_first(oc->rex, oc->wex)));
556 task_queue(task);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200557 }
558 if (ic->flags & CF_READ_ACTIVITY)
559 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200560}
561
562
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200563/* Called by I/O handlers after completion.. It propagates
Willy Tarreau651e1822015-09-23 20:06:13 +0200564 * connection flags to the stream interface, updates the stream (which may or
565 * may not take this opportunity to try to forward data), then update the
566 * connection's polling based on the channels and stream interface's final
567 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200568 */
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200569static int si_cs_process(struct conn_stream *cs)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200570{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200571 struct connection *conn = cs->conn;
572 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100573 struct channel *ic = si_ic(si);
574 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200575
Olivier Houchardc7ffa912018-08-28 19:37:41 +0200576 /* If we have data to send, try it now */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100577 if (!channel_is_empty(oc) && !(si->wait_event.events & SUB_RETRY_SEND))
Willy Tarreau908d26f2018-10-25 14:02:47 +0200578 si_cs_send(cs);
579
Willy Tarreau651e1822015-09-23 20:06:13 +0200580 /* First step, report to the stream-int what was detected at the
581 * connection layer : errors and connection establishment.
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200582 * Only add SI_FL_ERR if we're connected, or we're attempting to
583 * connect, we may get there because we got woken up, but only run
584 * after process_stream() noticed there were an error, and decided
585 * to retry to connect, the connection may still have CO_FL_ERROR,
586 * and we don't want to add SI_FL_ERR back
Willy Tarreau651e1822015-09-23 20:06:13 +0200587 */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200588 if (si->state >= SI_ST_CON &&
589 (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR))
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200590 si->flags |= SI_FL_ERR;
591
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200592 /* If we had early data, and the handshake ended, then
593 * we can remove the flag, and attempt to wake the task up,
594 * in the event there's an analyser waiting for the end of
595 * the handshake.
596 */
Olivier Houchard6fa63d92017-11-27 18:41:32 +0100597 if (!(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS)) &&
598 (cs->flags & CS_FL_WAIT_FOR_HS)) {
599 cs->flags &= ~CS_FL_WAIT_FOR_HS;
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200600 task_wakeup(si_task(si), TASK_WOKEN_MSG);
601 }
602
Willy Tarreau7ab22adb2019-06-05 14:53:22 +0200603 if (!si_state_in(si->state, SI_SB_EST|SI_SB_DIS|SI_SB_CLO) &&
Willy Tarreau52821e22017-03-18 17:11:37 +0100604 (conn->flags & (CO_FL_CONNECTED | CO_FL_HANDSHAKE)) == CO_FL_CONNECTED) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200605 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100606 oc->flags |= CF_WRITE_NULL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +0200607 if (si->state == SI_ST_CON)
608 si->state = SI_ST_RDY;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200609 }
610
Christopher Faulet297d3e22019-03-22 14:16:14 +0100611 /* Report EOI on the channel if it was reached from the mux point of
612 * view. */
613 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200614 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100615
Willy Tarreau651e1822015-09-23 20:06:13 +0200616 /* Second step : update the stream-int and channels, try to forward any
617 * pending data, then possibly wake the stream up based on the new
618 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200619 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200620 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +0200621 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200622 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200623}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200624
Willy Tarreau5368d802012-08-21 18:22:06 +0200625/*
626 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200627 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800628 * caller to commit polling changes. The caller should check conn->flags
629 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200630 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200631int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200632{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200633 struct connection *conn = cs->conn;
634 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100635 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200636 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200637 int did_send = 0;
638
639 /* We're already waiting to be able to send, give up */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100640 if (si->wait_event.events & SUB_RETRY_SEND)
Olivier Houchardf6535282018-08-31 17:29:12 +0200641 return 0;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200642
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100643 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200644 /* We're probably there because the tasklet was woken up,
645 * but process_stream() ran before, detected there were an
646 * error and put the si back to SI_ST_TAR. There's still
647 * CO_FL_ERROR on the connection but we don't want to add
648 * SI_FL_ERR back, so give up
649 */
650 if (si->state < SI_ST_CON)
651 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100652 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200653 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100654 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200655
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200656 /* we might have been called just after an asynchronous shutw */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100657 if (conn->flags & CO_FL_SOCK_WR_SH || oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200658 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200659
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200660 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
661 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet86162db2019-07-05 11:49:11 +0200662 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200663 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200664
Willy Tarreauafc8a222014-11-28 15:46:27 +0100665 if (!oc->pipe->data) {
666 put_pipe(oc->pipe);
667 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200668 }
669
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100670 if (oc->pipe)
671 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200672 }
673
674 /* At this point, the pipe is empty, but we may still have data pending
675 * in the normal buffer.
676 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100677 if (co_data(oc)) {
678 /* when we're here, we already know that there is no spliced
679 * data left, and that there are sendable buffered data.
680 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200681
Willy Tarreau5368d802012-08-21 18:22:06 +0200682 /* check if we want to inform the kernel that we're interested in
683 * sending more data after this call. We want this if :
684 * - we're about to close after this last send and want to merge
685 * the ongoing FIN with the last segment.
686 * - we know we can't send everything at once and must get back
687 * here because of unaligned data
688 * - there is still a finite amount of data to forward
689 * The test is arranged so that the most common case does only 2
690 * tests.
691 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100692 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200693
Willy Tarreauafc8a222014-11-28 15:46:27 +0100694 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
695 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau4ac49282017-10-17 16:33:46 +0200696 (oc->flags & CF_EXPECT_MORE))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100697 ((oc->flags & CF_ISRESP) &&
698 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100699 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200700
Willy Tarreauafc8a222014-11-28 15:46:27 +0100701 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100702 send_flag |= CO_SFL_STREAMER;
703
Olivier Houcharda254a372019-04-05 15:30:12 +0200704 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
Olivier Houchardce1a0292019-05-17 15:38:29 +0200705 struct stream *s = si_strm(si);
Olivier Houcharda254a372019-04-05 15:30:12 +0200706 /* If we want to be able to do L7 retries, copy
707 * the data we're about to send, so that we are able
708 * to resend them if needed
709 */
710 /* Try to allocate a buffer if we had none.
711 * If it fails, the next test will just
712 * disable the l7 retries by setting
713 * l7_conn_retries to 0.
714 */
Olivier Houchardce1a0292019-05-17 15:38:29 +0200715 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
Olivier Houcharda254a372019-04-05 15:30:12 +0200716 si->flags &= ~SI_FL_L7_RETRY;
717 else {
718 if (b_is_null(&si->l7_buffer))
719 b_alloc(&si->l7_buffer);
720 if (b_is_null(&si->l7_buffer))
721 si->flags &= ~SI_FL_L7_RETRY;
722 else {
723 memcpy(b_orig(&si->l7_buffer),
724 b_orig(&oc->buf),
725 b_size(&oc->buf));
726 si->l7_buffer.head = co_data(oc);
727 b_add(&si->l7_buffer, co_data(oc));
728 }
729
730 }
731 }
732
Olivier Houcharded0f2072018-08-16 15:41:52 +0200733 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800734 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200735 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200736 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200737 c_realign_if_empty(oc);
738
739 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800740 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100741 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800742 }
Godbache68e02d2013-10-11 15:48:29 +0800743 /* if some data remain in the buffer, it's only because the
744 * system buffers are full, we will try next time.
745 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200746 }
Godbache68e02d2013-10-11 15:48:29 +0800747 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100748
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100749 end:
Christopher Faulet86162db2019-07-05 11:49:11 +0200750 if (did_send) {
751 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
752 if (si->state == SI_ST_CON)
753 si->state = SI_ST_RDY;
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200754
755 si_rx_room_rdy(si_opposite(si));
Christopher Faulet86162db2019-07-05 11:49:11 +0200756 }
757
758 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
759 si->flags |= SI_FL_ERR;
760 return 1;
761 }
762
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200763 /* 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 +0100764 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100765 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200766 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200767}
768
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100769/* This is the ->process() function for any stream-interface's wait_event task.
770 * It's assigned during the stream-interface's initialization, for any type of
771 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
772 * stream interface, as the presence of the CS is checked there.
773 */
Olivier Houchard91894cb2018-08-02 18:06:28 +0200774struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned short state)
775{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200776 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200777 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200778 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200779
780 if (!cs)
781 return NULL;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100782
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100783 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200784 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100785 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200786 ret |= si_cs_recv(cs);
787 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200788 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200789
Willy Tarreaua64c7032019-08-01 14:17:02 +0200790 stream_release_buffers(si_strm(si));
Olivier Houchard91894cb2018-08-02 18:06:28 +0200791 return (NULL);
792}
793
Willy Tarreau25f13102015-09-24 11:32:22 +0200794/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200795 * update the input channel's expiration timer and the stream interface's
796 * Rx flags based on the channel's flags. It needs to be called only once
797 * after the channel's flags have settled down, and before they are cleared,
798 * though it doesn't harm to call it as often as desired (it just slightly
799 * hurts performance). It must not be called from outside of the stream
800 * handler, as what it does will be used to compute the stream task's
801 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200802 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200803void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200804{
805 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200806
Willy Tarreau236c4292019-06-06 08:19:20 +0200807 if (ic->flags & CF_SHUTR) {
808 si_rx_shut_blk(si);
809 return;
810 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100811
Willy Tarreau236c4292019-06-06 08:19:20 +0200812 /* Read not closed, update FD status and timeout for reads */
813 if (ic->flags & CF_DONT_READ)
814 si_rx_chan_blk(si);
815 else
816 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100817
Willy Tarreau236c4292019-06-06 08:19:20 +0200818 if (!channel_is_empty(ic)) {
819 /* stop reading, imposed by channel's policy or contents */
820 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200821 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200822 else {
823 /* (re)start reading and update timeout. Note: we don't recompute the timeout
824 * everytime we get here, otherwise it would risk never to expire. We only
825 * update it if is was not yet set. The stream socket handler will already
826 * have updated it if there has been a completed I/O.
827 */
828 si_rx_room_rdy(si);
829 }
830 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
831 ic->rex = TICK_ETERNITY;
832 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
833 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200834
Willy Tarreau236c4292019-06-06 08:19:20 +0200835 si_chk_rcv(si);
836}
837
838/* This function is designed to be called from within the stream handler to
839 * update the output channel's expiration timer and the stream interface's
840 * Tx flags based on the channel's flags. It needs to be called only once
841 * after the channel's flags have settled down, and before they are cleared,
842 * though it doesn't harm to call it as often as desired (it just slightly
843 * hurts performance). It must not be called from outside of the stream
844 * handler, as what it does will be used to compute the stream task's
845 * expiration.
846 */
847void si_update_tx(struct stream_interface *si)
848{
849 struct channel *oc = si_oc(si);
850 struct channel *ic = si_ic(si);
851
852 if (oc->flags & CF_SHUTW)
853 return;
854
855 /* Write not closed, update FD status and timeout for writes */
856 if (channel_is_empty(oc)) {
857 /* stop writing */
858 if (!(si->flags & SI_FL_WAIT_DATA)) {
859 if ((oc->flags & CF_SHUTW_NOW) == 0)
860 si->flags |= SI_FL_WAIT_DATA;
861 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200862 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200863 return;
864 }
865
866 /* (re)start writing and update timeout. Note: we don't recompute the timeout
867 * everytime we get here, otherwise it would risk never to expire. We only
868 * update it if is was not yet set. The stream socket handler will already
869 * have updated it if there has been a completed I/O.
870 */
871 si->flags &= ~SI_FL_WAIT_DATA;
872 if (!tick_isset(oc->wex)) {
873 oc->wex = tick_add_ifset(now_ms, oc->wto);
874 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
875 /* Note: depending on the protocol, we don't know if we're waiting
876 * for incoming data or not. So in order to prevent the socket from
877 * expiring read timeouts during writes, we refresh the read timeout,
878 * except if it was already infinite or if we have explicitly setup
879 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200880 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200881 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200882 }
883 }
884}
885
Willy Tarreau3b285d72019-06-06 08:20:17 +0200886/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
887 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
888 * be updated in case of success.
889 */
890void si_sync_send(struct stream_interface *si)
891{
892 struct channel *oc = si_oc(si);
893 struct conn_stream *cs;
894
895 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
896
897 if (oc->flags & CF_SHUTW)
898 return;
899
900 if (channel_is_empty(oc))
901 return;
902
903 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
904 return;
905
906 cs = objt_cs(si->end);
907 if (!cs)
908 return;
909
910 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING))
911 return;
912
913 if (cs->conn->flags & CO_FL_ERROR)
914 return;
915
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200916 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200917}
918
Willy Tarreaud14844a2018-11-08 18:15:29 +0100919/* Updates at once the channel flags, and timers of both stream interfaces of a
920 * same stream, to complete the work after the analysers, then updates the data
921 * layer below. This will ensure that any synchronous update performed at the
922 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200923 * Note that this does not change the stream interface's current state, though
924 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100925 */
926void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
927{
928 struct channel *req = si_ic(si_f);
929 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100930
931 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
932 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
933
Willy Tarreaud14844a2018-11-08 18:15:29 +0100934 si_f->prev_state = si_f->state;
935 si_b->prev_state = si_b->state;
936
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100937 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200938 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100939 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100940
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200941 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100942 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100943
944 /* stream ints are processed outside of process_stream() and must be
945 * handled at the latest moment.
946 */
947 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100948 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
949 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100950 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100951
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100952 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100953 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
954 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100955 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100956}
957
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200958/*
959 * This function performs a shutdown-read on a stream interface attached to
960 * a connection in a connected or init state (it does nothing for other
961 * states). It either shuts the read side or marks itself as closed. The buffer
962 * flags are updated to reflect the new state. If the stream interface has
963 * SI_FL_NOHALF, we also forward the close to the write side. If a control
964 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200965 * descriptors are then shutdown or closed accordingly. The function
966 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200967 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200968static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200969{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200970 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100971 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200972
Willy Tarreauabb5d422018-11-14 16:58:52 +0100973 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100974 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200975 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100976 ic->flags |= CF_SHUTR;
977 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200978
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200979 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200980 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200981
Willy Tarreau51d0a7e2019-01-31 19:09:59 +0100982 if (si->flags & SI_FL_KILL_CONN)
983 cs->flags |= CS_FL_KILL_CONN;
984
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100985 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +0200986 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200987 si->state = SI_ST_DIS;
988 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200989 }
990 else if (si->flags & SI_FL_NOHALF) {
991 /* we want to immediately forward this close to the write side */
992 return stream_int_shutw_conn(si);
993 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200994}
995
996/*
997 * This function performs a shutdown-write on a stream interface attached to
998 * a connection in a connected or init state (it does nothing for other
999 * states). It either shuts the write side or marks itself as closed. The
1000 * buffer flags are updated to reflect the new state. It does also close
1001 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001002 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001003 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001004static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001005{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001006 struct conn_stream *cs = __objt_cs(si->end);
1007 struct connection *conn = cs->conn;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001008 struct channel *ic = si_ic(si);
1009 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001010
Willy Tarreauafc8a222014-11-28 15:46:27 +01001011 oc->flags &= ~CF_SHUTW_NOW;
1012 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001013 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001014 oc->flags |= CF_SHUTW;
1015 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001016 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001017
Hongbo Longe39683c2017-03-10 18:41:51 +01001018 if (tick_isset(si->hcto)) {
1019 ic->rto = si->hcto;
1020 ic->rex = tick_add(now_ms, ic->rto);
1021 }
1022
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001023 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001024 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001025 case SI_ST_EST:
1026 /* we have to shut before closing, otherwise some short messages
1027 * may never leave the system, especially when there are remaining
1028 * unread data in the socket input buffer, or when nolinger is set.
1029 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1030 * no risk so we close both sides immediately.
1031 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001032 if (si->flags & SI_FL_KILL_CONN)
1033 cs->flags |= CS_FL_KILL_CONN;
1034
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001035 if (si->flags & SI_FL_ERR) {
1036 /* quick close, the socket is alredy shut anyway */
1037 }
1038 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001039 /* unclean data-layer shutdown, typically an aborted request
1040 * or a forwarded shutdown from a client to a server due to
1041 * option abortonclose. No need for the TLS layer to try to
1042 * emit a shutdown message.
1043 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001044 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001045 }
1046 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001047 /* clean data-layer shutdown. This only happens on the
1048 * frontend side, or on the backend side when forwarding
1049 * a client close in TCP mode or in HTTP TUNNEL mode
1050 * while option abortonclose is set. We want the TLS
1051 * layer to try to signal it to the peer before we close.
1052 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001053 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001054
Willy Tarreaua553ae92017-10-05 18:52:17 +02001055 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
1056 /* OK just a shutw, but we want the caller
1057 * to disable polling on this FD if exists.
1058 */
1059 conn_cond_update_polling(conn);
1060 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001061 }
1062 }
1063
1064 /* fall through */
1065 case SI_ST_CON:
1066 /* we may have to close a pending connection, and mark the
1067 * response buffer as shutr
1068 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001069 if (si->flags & SI_FL_KILL_CONN)
1070 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001071 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001072 /* fall through */
1073 case SI_ST_CER:
1074 case SI_ST_QUE:
1075 case SI_ST_TAR:
1076 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001077 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001078 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001079 si->flags &= ~SI_FL_NOLINGER;
1080 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001081 ic->flags |= CF_SHUTR;
1082 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001083 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001084 }
1085}
1086
Willy Tarreau46a8d922012-08-20 12:38:36 +02001087/* This function is used for inter-stream-interface calls. It is called by the
1088 * consumer to inform the producer side that it may be interested in checking
1089 * for free space in the buffer. Note that it intentionally does not update
1090 * timeouts, so that we can still check them later at wake-up. This function is
1091 * dedicated to connection-based stream interfaces.
1092 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001093static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001094{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001095 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001096 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001097 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001098}
1099
1100
Willy Tarreaude5722c2012-08-20 15:01:10 +02001101/* This function is used for inter-stream-interface calls. It is called by the
1102 * producer to inform the consumer side that it may be interested in checking
1103 * for data in the buffer. Note that it intentionally does not update timeouts,
1104 * so that we can still check them later at wake-up.
1105 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001106static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001107{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001108 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001109 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001110
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001111 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001112 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001113 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001114
Willy Tarreauafc8a222014-11-28 15:46:27 +01001115 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001116 return;
1117
Willy Tarreauafc8a222014-11-28 15:46:27 +01001118 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001119 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001120 return;
1121
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001122 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001123 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001124
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001125 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001126 /* Write error on the file descriptor */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +02001127 if (si->state >= SI_ST_CON)
1128 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001129 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001130 }
1131
1132 /* OK, so now we know that some data might have been sent, and that we may
1133 * have to poll first. We have to do that too if the buffer is not empty.
1134 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001135 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001136 /* the connection is established but we can't write. Either the
1137 * buffer is empty, or we just refrain from sending because the
1138 * ->o limit was reached. Maybe we just wrote the last
1139 * chunk and need to close.
1140 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001141 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001142 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001143 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001144 si_shutw(si);
1145 goto out_wakeup;
1146 }
1147
Willy Tarreauafc8a222014-11-28 15:46:27 +01001148 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001149 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001150 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001151 }
1152 else {
1153 /* Otherwise there are remaining data to be sent in the buffer,
1154 * which means we have to poll before doing so.
1155 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001156 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001157 if (!tick_isset(oc->wex))
1158 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001159 }
1160
Willy Tarreauafc8a222014-11-28 15:46:27 +01001161 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1162 struct channel *ic = si_ic(si);
1163
Willy Tarreaude5722c2012-08-20 15:01:10 +02001164 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001165 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1166 !channel_is_empty(oc))
1167 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001168
Willy Tarreauafc8a222014-11-28 15:46:27 +01001169 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001170 /* Note: to prevent the client from expiring read timeouts
1171 * during writes, we refresh it. We only do this if the
1172 * interface is not configured for "independent streams",
1173 * because for some applications it's better not to do this,
1174 * for instance when continuously exchanging small amounts
1175 * of data which can full the socket buffers long before a
1176 * write timeout is detected.
1177 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001178 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001179 }
1180 }
1181
1182 /* in case of special condition (error, shutdown, end of write...), we
1183 * have to notify the task.
1184 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001185 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1186 ((oc->flags & CF_WAKE_WRITE) &&
1187 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001188 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001189 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001190 if (!(si->flags & SI_FL_DONT_WAKE))
1191 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001192 }
1193}
1194
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001195/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001196 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001197 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001198 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001199 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001200int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001201{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001202 struct connection *conn = cs->conn;
1203 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001204 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001205 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001206 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001207 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001208
Olivier Houchardf6535282018-08-31 17:29:12 +02001209 /* If another call to si_cs_recv() failed, and we subscribed to
1210 * recv events already, give up now.
1211 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001212 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001213 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001214
Willy Tarreauce323de2012-08-20 21:41:06 +02001215 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001216 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001217 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001218
Willy Tarreau54e917c2017-08-30 07:35:35 +02001219 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001220 if (cs->flags & CS_FL_EOS)
Willy Tarreau54e917c2017-08-30 07:35:35 +02001221 goto out_shutdown_r;
1222
Christopher Fauletf061e422018-12-07 14:51:20 +01001223 /* stop immediately on errors. Note that we DON'T want to stop on
1224 * POLL_ERR, as the poller might report a write error while there
1225 * are still data available in the recv buffer. This typically
1226 * happens when we send too large a request to a backend server
1227 * which rejects it before reading it all.
1228 */
1229 if (!(cs->flags & CS_FL_RCV_MORE)) {
1230 if (!conn_xprt_ready(conn))
1231 return 0;
1232 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
1233 return 1; // We want to make sure si_cs_wake() is called, so that process_strema is woken up, on failure
1234 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001235
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001236 /* prepare to detect if the mux needs more room */
1237 cs->flags &= ~CS_FL_WANT_ROOM;
1238
Willy Tarreau77e478c2018-06-19 07:03:14 +02001239 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001240 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001241 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001242 /* The buffer was empty and nothing was transferred for more
1243 * than one second. This was caused by a pause and not by
1244 * congestion. Reset any streaming mode to reduce latency.
1245 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001246 ic->xfer_small = 0;
1247 ic->xfer_large = 0;
1248 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001249 }
1250
Willy Tarreau96199b12012-08-24 00:46:52 +02001251 /* First, let's see if we may splice data across the channel without
1252 * using a buffer.
1253 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001254 if (conn->xprt->rcv_pipe && conn->mux->rcv_pipe &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001255 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1256 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001257 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001258 /* We're embarrassed, there are already data pending in
1259 * the buffer and we don't want to have them at two
1260 * locations at a time. Let's indicate we need some
1261 * place and ask the consumer to hurry.
1262 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001263 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001264 goto abort_splice;
1265 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001266
Willy Tarreauafc8a222014-11-28 15:46:27 +01001267 if (unlikely(ic->pipe == NULL)) {
1268 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1269 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001270 goto abort_splice;
1271 }
1272 }
1273
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001274 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001275 if (ret < 0) {
1276 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001277 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001278 goto abort_splice;
1279 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001280
Willy Tarreau96199b12012-08-24 00:46:52 +02001281 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001282 if (ic->to_forward != CHN_INFINITE_FORWARD)
1283 ic->to_forward -= ret;
1284 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001285 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001286 ic->flags |= CF_READ_PARTIAL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001287 if (si->state == SI_ST_CON)
1288 si->state = SI_ST_RDY;
Willy Tarreauce323de2012-08-20 21:41:06 +02001289 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001290
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001291 if (cs->flags & CS_FL_EOS)
Willy Tarreau96199b12012-08-24 00:46:52 +02001292 goto out_shutdown_r;
1293
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001294 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001295 return 1;
Willy Tarreau96199b12012-08-24 00:46:52 +02001296
Willy Tarreau61d39a02013-07-18 21:49:32 +02001297 if (conn->flags & CO_FL_WAIT_ROOM) {
1298 /* the pipe is full or we have read enough data that it
1299 * could soon be full. Let's stop before needing to poll.
1300 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001301 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001302 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001303 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001304
Willy Tarreauce323de2012-08-20 21:41:06 +02001305 /* splice not possible (anymore), let's go on on standard copy */
1306 }
Willy Tarreau81464b42018-11-15 15:52:53 +01001307 else {
1308 /* be sure not to block regular receive path below */
1309 conn->flags &= ~CO_FL_WAIT_ROOM;
1310 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001311
1312 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001313 if (ic->pipe && unlikely(!ic->pipe->data)) {
1314 put_pipe(ic->pipe);
1315 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001316 }
1317
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001318 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001319 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001320 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001321
Willy Tarreau61d39a02013-07-18 21:49:32 +02001322 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1323 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1324 * that if such an event is not handled above in splice, it will be handled here by
1325 * recv().
1326 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001327 while ((cs->flags & CS_FL_RCV_MORE) ||
1328 (!(conn->flags & (CO_FL_ERROR | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE)) &&
1329 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001330 /* <max> may be null. This is the mux responsibility to set
1331 * CS_FL_RCV_MORE on the CS if more space is needed.
1332 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001333 max = channel_recv_max(ic);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001334 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 +01001335
Olivier Houchardd247be02018-12-06 16:22:29 +01001336 if (cs->flags & CS_FL_WANT_ROOM)
Willy Tarreaudb398432018-11-15 11:08:52 +01001337 si_rx_room_blk(si);
Willy Tarreau6577b482017-12-10 21:19:33 +01001338
Christopher Fauleteffc3752018-10-31 08:53:54 +01001339 if (cs->flags & CS_FL_READ_PARTIAL) {
1340 if (tick_isset(ic->rex))
1341 ic->rex = tick_add_ifset(now_ms, ic->rto);
1342 cs->flags &= ~CS_FL_READ_PARTIAL;
1343 }
1344
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001345 if (ret <= 0) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001346 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001347 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001348
Olivier Houcharda254a372019-04-05 15:30:12 +02001349 if (si->flags & SI_FL_L7_RETRY) {
1350 struct htx *htx;
1351 struct htx_sl *sl;
1352
1353 htx = htxbuf(&ic->buf);
1354 if (htx) {
Christopher Faulet297fbb42019-05-13 14:41:27 +02001355 sl = http_get_stline(htx);
Olivier Houcharda254a372019-04-05 15:30:12 +02001356 if (sl && l7_status_match(si_strm(si)->be,
1357 sl->info.res.status)) {
1358 /* If we got a status for which we would
1359 * like to retry the request, empty
1360 * the buffer and pretend there's an
1361 * error on the channel.
1362 */
1363 ic->flags |= CF_READ_ERROR;
1364 htx_reset(htx);
1365 return 1;
1366 }
1367 }
1368 si->flags &= ~SI_FL_L7_RETRY;
1369 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001370 cur_read += ret;
1371
1372 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001373 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001374 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001375 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1376 if (fwd > ic->to_forward)
1377 fwd = ic->to_forward;
1378 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001379 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001380 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001381 }
1382
Willy Tarreauafc8a222014-11-28 15:46:27 +01001383 ic->flags |= CF_READ_PARTIAL;
1384 ic->total += ret;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001385 if (si->state == SI_ST_CON)
1386 si->state = SI_ST_RDY;
Willy Tarreauce323de2012-08-20 21:41:06 +02001387
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001388 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1389 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001390 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001391 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001392 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001393
1394 /* if too many bytes were missing from last read, it means that
1395 * it's pointless trying to read again because the system does
1396 * not have them in buffers.
1397 */
1398 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001399 /* if a streamer has read few data, it may be because we
1400 * have exhausted system buffers. It's not worth trying
1401 * again.
1402 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001403 if (ic->flags & CF_STREAMER) {
1404 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001405 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001406 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001407 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001408
1409 /* if we read a large block smaller than what we requested,
1410 * it's almost certain we'll never get anything more.
1411 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001412 if (ret >= global.tune.recv_enough) {
1413 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001414 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001415 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001416 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001417 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001418
1419 /* if we are waiting for more space, don't try to read more data
1420 * right now.
1421 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001422 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001423 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001424 } /* while !flags */
1425
Willy Tarreauffb12052018-11-15 16:06:02 +01001426 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001427 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001428 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001429 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001430 ic->xfer_large = 0;
1431 ic->xfer_small++;
1432 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001433 /* we have read less than half of the buffer in
1434 * one pass, and this happened at least 3 times.
1435 * This is definitely not a streamer.
1436 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001437 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001438 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001439 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001440 /* if the buffer has been at least half full twice,
1441 * we receive faster than we send, so at least it
1442 * is not a "fast streamer".
1443 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001444 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001445 }
1446 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001447 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001448 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001449 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001450 ic->xfer_small = 0;
1451 ic->xfer_large++;
1452 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001453 /* we call this buffer a fast streamer if it manages
1454 * to be filled in one call 3 consecutive times.
1455 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001456 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001457 }
1458 }
1459 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001460 ic->xfer_small = 0;
1461 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001462 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001463 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001464 }
1465
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001466 end_recv:
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001467 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001468 return 1;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001469
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001470 if (cs->flags & CS_FL_EOS)
Willy Tarreauce323de2012-08-20 21:41:06 +02001471 /* connection closed */
1472 goto out_shutdown_r;
1473
Willy Tarreau32742fd2018-11-14 14:07:59 +01001474 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001475 if (!si_rx_blocked(si)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001476 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001477 si_rx_endp_done(si);
1478 } else {
1479 si_rx_endp_more(si);
1480 }
Olivier Houchardf6535282018-08-31 17:29:12 +02001481
Olivier Houchard19a2e2d2019-06-07 18:10:52 +02001482 return (cur_read != 0) || si_rx_blocked(si) || (cs->flags & CS_FL_EOI);
Willy Tarreauce323de2012-08-20 21:41:06 +02001483
1484 out_shutdown_r:
Olivier Houchardfd0c2dc2018-12-13 15:38:16 +01001485 if (conn->flags & CO_FL_CONNECTED) {
1486 /* we received a shutdown */
1487 ic->flags |= CF_READ_NULL;
1488 if (ic->flags & CF_AUTO_CLOSE)
1489 channel_shutw_now(ic);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001490 stream_int_read0(si);
Olivier Houchardfd0c2dc2018-12-13 15:38:16 +01001491 }
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001492 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001493}
1494
1495/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001496 * This function propagates a null read received on a socket-based connection.
1497 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001498 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001499 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001500static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001501{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001502 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001503 struct channel *ic = si_ic(si);
1504 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001505
Willy Tarreauabb5d422018-11-14 16:58:52 +01001506 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001507 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001508 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001509 ic->flags |= CF_SHUTR;
1510 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001511
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001512 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001513 return;
1514
Willy Tarreauafc8a222014-11-28 15:46:27 +01001515 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001516 goto do_close;
1517
1518 if (si->flags & SI_FL_NOHALF) {
1519 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001520 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001521 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001522 goto do_close;
1523 }
1524
1525 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001526 return;
1527
1528 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001529 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001530 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001531
Willy Tarreauafc8a222014-11-28 15:46:27 +01001532 oc->flags &= ~CF_SHUTW_NOW;
1533 oc->flags |= CF_SHUTW;
1534 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001535
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001536 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001537
Olivier Houchardaacc4052019-05-21 17:43:50 +02001538 /* Don't change the state to SI_ST_DIS yet if we're still
1539 * in SI_ST_CON, otherwise it means sess_establish() hasn't
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001540 * been called yet, and so the analysers would not run. However
1541 * it's fine to switch to SI_ST_RDY as we have really validated
1542 * the connection.
Olivier Houchardaacc4052019-05-21 17:43:50 +02001543 */
1544 if (si->state == SI_ST_EST)
1545 si->state = SI_ST_DIS;
Willy Tarreaub27f54a2019-06-05 16:43:44 +02001546 else if (si->state == SI_ST_CON)
1547 si->state = SI_ST_RDY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001548 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001549 return;
1550}
1551
Willy Tarreau651e1822015-09-23 20:06:13 +02001552/* Callback to be used by applet handlers upon completion. It updates the stream
1553 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001554 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001555 * states.
1556 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001557void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001558{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001559 struct channel *ic = si_ic(si);
1560
1561 /* If the applet wants to write and the channel is closed, it's a
1562 * broken pipe and it must be reported.
1563 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001564 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001565 si->flags |= SI_FL_ERR;
1566
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001567 /* automatically mark the applet having data available if it reported
1568 * begin blocked by the channel.
1569 */
1570 if (si_rx_blocked(si))
1571 si_rx_endp_more(si);
1572
Willy Tarreau651e1822015-09-23 20:06:13 +02001573 /* update the stream-int, channels, and possibly wake the stream up */
1574 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +02001575 stream_release_buffers(si_strm(si));
Willy Tarreaue5f86492015-04-19 15:16:35 +02001576
Willy Tarreau32742fd2018-11-14 14:07:59 +01001577 /* stream_int_notify may have passed through chk_snd and released some
1578 * RXBLK flags. Process_stream will consider those flags to wake up the
1579 * appctx but in the case the task is not in runqueue we may have to
1580 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001581 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001582 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1583 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001584 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001585}
1586
1587/*
1588 * This function performs a shutdown-read on a stream interface attached to an
1589 * applet in a connected or init state (it does nothing for other states). It
1590 * either shuts the read side or marks itself as closed. The buffer flags are
1591 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1592 * we also forward the close to the write side. The owner task is woken up if
1593 * it exists.
1594 */
1595static void stream_int_shutr_applet(struct stream_interface *si)
1596{
1597 struct channel *ic = si_ic(si);
1598
Willy Tarreauabb5d422018-11-14 16:58:52 +01001599 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001600 if (ic->flags & CF_SHUTR)
1601 return;
1602 ic->flags |= CF_SHUTR;
1603 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001604
Willy Tarreau828824a2015-04-19 17:20:03 +02001605 /* Note: on shutr, we don't call the applet */
1606
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001607 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001608 return;
1609
1610 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001611 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001612 si->state = SI_ST_DIS;
1613 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001614 }
1615 else if (si->flags & SI_FL_NOHALF) {
1616 /* we want to immediately forward this close to the write side */
1617 return stream_int_shutw_applet(si);
1618 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001619}
1620
1621/*
1622 * This function performs a shutdown-write on a stream interface attached to an
1623 * applet in a connected or init state (it does nothing for other states). It
1624 * either shuts the write side or marks itself as closed. The buffer flags are
1625 * updated to reflect the new state. It does also close everything if the SI
1626 * was marked as being in error state. The owner task is woken up if it exists.
1627 */
1628static void stream_int_shutw_applet(struct stream_interface *si)
1629{
1630 struct channel *ic = si_ic(si);
1631 struct channel *oc = si_oc(si);
1632
1633 oc->flags &= ~CF_SHUTW_NOW;
1634 if (oc->flags & CF_SHUTW)
1635 return;
1636 oc->flags |= CF_SHUTW;
1637 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001638 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001639
Hongbo Longe39683c2017-03-10 18:41:51 +01001640 if (tick_isset(si->hcto)) {
1641 ic->rto = si->hcto;
1642 ic->rex = tick_add(now_ms, ic->rto);
1643 }
1644
Willy Tarreau828824a2015-04-19 17:20:03 +02001645 /* on shutw we always wake the applet up */
1646 appctx_wakeup(si_appctx(si));
1647
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001648 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001649 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001650 case SI_ST_EST:
1651 /* we have to shut before closing, otherwise some short messages
1652 * may never leave the system, especially when there are remaining
1653 * unread data in the socket input buffer, or when nolinger is set.
1654 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1655 * no risk so we close both sides immediately.
1656 */
1657 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1658 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1659 return;
1660
1661 /* fall through */
1662 case SI_ST_CON:
1663 case SI_ST_CER:
1664 case SI_ST_QUE:
1665 case SI_ST_TAR:
1666 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001667 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001668 si->state = SI_ST_DIS;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001669 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001670 si->flags &= ~SI_FL_NOLINGER;
1671 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001672 ic->flags |= CF_SHUTR;
1673 ic->rex = TICK_ETERNITY;
1674 si->exp = TICK_ETERNITY;
1675 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001676}
1677
1678/* chk_rcv function for applets */
1679static void stream_int_chk_rcv_applet(struct stream_interface *si)
1680{
1681 struct channel *ic = si_ic(si);
1682
1683 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1684 __FUNCTION__,
1685 si, si->state, ic->flags, si_oc(si)->flags);
1686
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001687 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001688 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001689 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001690 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001691}
1692
1693/* chk_snd function for applets */
1694static void stream_int_chk_snd_applet(struct stream_interface *si)
1695{
1696 struct channel *oc = si_oc(si);
1697
1698 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1699 __FUNCTION__,
1700 si, si->state, si_ic(si)->flags, oc->flags);
1701
1702 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1703 return;
1704
Willy Tarreau828824a2015-04-19 17:20:03 +02001705 /* we only wake the applet up if it was waiting for some data */
1706
1707 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001708 return;
1709
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001710 if (!tick_isset(oc->wex))
1711 oc->wex = tick_add_ifset(now_ms, oc->wto);
1712
Willy Tarreau828824a2015-04-19 17:20:03 +02001713 if (!channel_is_empty(oc)) {
1714 /* (re)start sending */
1715 appctx_wakeup(si_appctx(si));
1716 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001717}
1718
Willy Tarreaudded32d2008-11-30 19:48:07 +01001719/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001720 * Local variables:
1721 * c-indent-level: 8
1722 * c-basic-offset: 8
1723 * End:
1724 */