blob: 524fac7ddc700971c3cf917141ff2dc3defec2ab [file] [log] [blame]
Willy Tarreaucff64112008-11-03 06:26:53 +01001/*
2 * Functions managing stream_interface structures
3 *
Willy Tarreauf873d752012-05-11 17:47:17 +02004 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
Willy Tarreaucff64112008-11-03 06:26:53 +01005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <errno.h>
14#include <fcntl.h>
15#include <stdio.h>
16#include <stdlib.h>
17
18#include <sys/socket.h>
19#include <sys/stat.h>
20#include <sys/types.h>
21
Willy Tarreaubf883e02014-11-25 21:10:35 +010022#include <common/buffer.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010023#include <common/compat.h>
24#include <common/config.h>
25#include <common/debug.h>
26#include <common/standard.h>
27#include <common/ticks.h>
28#include <common/time.h>
29
Willy Tarreau8a8d83b2015-04-13 13:24:54 +020030#include <proto/applet.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020031#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020032#include <proto/connection.h>
Olivier Houcharda254a372019-04-05 15:30:12 +020033#include <proto/http_htx.h>
Olivier Houchard9aaf7782017-09-13 18:30:23 +020034#include <proto/mux_pt.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020035#include <proto/pipe.h>
Olivier Houcharda254a372019-04-05 15:30:12 +020036#include <proto/proxy.h>
Willy Tarreau87b09662015-04-03 00:22:06 +020037#include <proto/stream.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020038#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010039#include <proto/task.h>
40
Willy Tarreaufd31e532012-07-23 18:24:25 +020041#include <types/pipe.h>
42
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010043/* functions used by default on a detached stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020044static void stream_int_shutr(struct stream_interface *si);
45static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020046static void stream_int_chk_rcv(struct stream_interface *si);
47static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010048
49/* functions used on a conn_stream-based stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020050static void stream_int_shutr_conn(struct stream_interface *si);
51static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020052static void stream_int_chk_rcv_conn(struct stream_interface *si);
53static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010054
55/* functions used on an applet-based stream-interface */
Willy Tarreaud45b9f82015-04-13 16:30:14 +020056static void stream_int_shutr_applet(struct stream_interface *si);
57static void stream_int_shutw_applet(struct stream_interface *si);
58static void stream_int_chk_rcv_applet(struct stream_interface *si);
59static void stream_int_chk_snd_applet(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010060
61/* last read notification */
62static void stream_int_read0(struct stream_interface *si);
63
64/* post-IO notification callback */
65static void stream_int_notify(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020066
Willy Tarreauc5788912012-08-24 18:12:41 +020067/* stream-interface operations for embedded tasks */
68struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020069 .chk_rcv = stream_int_chk_rcv,
70 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020071 .shutr = stream_int_shutr,
72 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020073};
74
Willy Tarreauc5788912012-08-24 18:12:41 +020075/* stream-interface operations for connections */
76struct si_ops si_conn_ops = {
Willy Tarreauc5788912012-08-24 18:12:41 +020077 .chk_rcv = stream_int_chk_rcv_conn,
78 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020079 .shutr = stream_int_shutr_conn,
80 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020081};
82
Willy Tarreaud45b9f82015-04-13 16:30:14 +020083/* stream-interface operations for connections */
84struct si_ops si_applet_ops = {
Willy Tarreaud45b9f82015-04-13 16:30:14 +020085 .chk_rcv = stream_int_chk_rcv_applet,
86 .chk_snd = stream_int_chk_snd_applet,
87 .shutr = stream_int_shutr_applet,
88 .shutw = stream_int_shutw_applet,
89};
90
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010091
92/* Functions used to communicate with a conn_stream. The first two may be used
93 * directly, the last one is mostly a wake callback.
94 */
95int si_cs_recv(struct conn_stream *cs);
96int si_cs_send(struct conn_stream *cs);
97static int si_cs_process(struct conn_stream *cs);
98
99
Willy Tarreau74beec32012-10-03 00:41:04 +0200100struct data_cb si_conn_cb = {
Olivier Houchard21df6cc2018-09-14 23:21:44 +0200101 .wake = si_cs_process,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100102 .name = "STRM",
Willy Tarreauc5788912012-08-24 18:12:41 +0200103};
104
Willy Tarreaucff64112008-11-03 06:26:53 +0100105/*
106 * This function only has to be called once after a wakeup event in case of
107 * suspected timeout. It controls the stream interface timeouts and sets
108 * si->flags accordingly. It does NOT close anything, as this timeout may
109 * be used for any purpose. It returns 1 if the timeout fired, otherwise
110 * zero.
111 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100112int si_check_timeouts(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100113{
114 if (tick_is_expired(si->exp, now_ms)) {
115 si->flags |= SI_FL_EXP;
116 return 1;
117 }
118 return 0;
119}
120
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100121/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100122void si_report_error(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100123{
124 if (!si->err_type)
125 si->err_type = SI_ET_DATA_ERR;
126
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100127 si_oc(si)->flags |= CF_WRITE_ERROR;
128 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100129}
130
131/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100132 * Returns a message to the client ; the connection is shut down for read,
133 * and the request is cleared so that no server connection can be initiated.
134 * The buffer is marked for read shutdown on the other side to protect the
135 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100136 * "chunk". If it is null, then an empty message is used. The reply buffer does
137 * not need to be empty before this, and its contents will not be overwritten.
138 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100139 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100140void si_retnclose(struct stream_interface *si,
Willy Tarreau83061a82018-07-13 11:56:34 +0200141 const struct buffer *msg)
Willy Tarreaudded32d2008-11-30 19:48:07 +0100142{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100143 struct channel *ic = si_ic(si);
144 struct channel *oc = si_oc(si);
145
146 channel_auto_read(ic);
147 channel_abort(ic);
148 channel_auto_close(ic);
149 channel_erase(ic);
150 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100151
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200152 if (likely(msg && msg->data))
153 co_inject(oc, msg->area, msg->data);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100154
Willy Tarreauafc8a222014-11-28 15:46:27 +0100155 oc->wex = tick_add_ifset(now_ms, oc->wto);
156 channel_auto_read(oc);
157 channel_auto_close(oc);
158 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100159}
160
Willy Tarreau4a36b562012-08-06 19:31:45 +0200161/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200162 * This function performs a shutdown-read on a detached stream interface in a
163 * connected or init state (it does nothing for other states). It either shuts
164 * the read side or marks itself as closed. The buffer flags are updated to
165 * reflect the new state. If the stream interface has SI_FL_NOHALF, we also
166 * forward the close to the write side. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200167 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200168static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200169{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100170 struct channel *ic = si_ic(si);
171
Willy Tarreauabb5d422018-11-14 16:58:52 +0100172 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100173 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200174 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100175 ic->flags |= CF_SHUTR;
176 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200177
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200178 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200179 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200180
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100181 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200182 si->state = SI_ST_DIS;
183 si->exp = TICK_ETERNITY;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200184 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200185 else if (si->flags & SI_FL_NOHALF) {
186 /* we want to immediately forward this close to the write side */
187 return stream_int_shutw(si);
188 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200189
Willy Tarreau4a36b562012-08-06 19:31:45 +0200190 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100191 if (!(si->flags & SI_FL_DONT_WAKE))
192 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200193}
194
Willy Tarreau4a36b562012-08-06 19:31:45 +0200195/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200196 * This function performs a shutdown-write on a detached stream interface in a
197 * connected or init state (it does nothing for other states). It either shuts
198 * the write side or marks itself as closed. The buffer flags are updated to
199 * reflect the new state. It does also close everything if the SI was marked as
200 * being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200201 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200202static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200203{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100204 struct channel *ic = si_ic(si);
205 struct channel *oc = si_oc(si);
206
207 oc->flags &= ~CF_SHUTW_NOW;
208 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200209 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100210 oc->flags |= CF_SHUTW;
211 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +0100212 si_done_get(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200213
Hongbo Longe39683c2017-03-10 18:41:51 +0100214 if (tick_isset(si->hcto)) {
215 ic->rto = si->hcto;
216 ic->rex = tick_add(now_ms, ic->rto);
217 }
218
Willy Tarreaufb90d942009-09-05 20:57:35 +0200219 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200220 case SI_ST_RDY:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200221 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200222 /* we have to shut before closing, otherwise some short messages
223 * may never leave the system, especially when there are remaining
224 * unread data in the socket input buffer, or when nolinger is set.
225 * However, if SI_FL_NOLINGER is explicitly set, we know there is
226 * no risk so we close both sides immediately.
227 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200228 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100229 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200230 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200231
232 /* fall through */
233 case SI_ST_CON:
234 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100235 case SI_ST_QUE:
236 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200237 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200238 si->state = SI_ST_DIS;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200239 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +0100240 si->flags &= ~SI_FL_NOLINGER;
241 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100242 ic->flags |= CF_SHUTR;
243 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244 si->exp = TICK_ETERNITY;
245 }
246
Willy Tarreau4a36b562012-08-06 19:31:45 +0200247 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100248 if (!(si->flags & SI_FL_DONT_WAKE))
249 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250}
251
252/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200253static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200254{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100255 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200256
Willy Tarreauafc8a222014-11-28 15:46:27 +0100257 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200258 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100259 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200260
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200261 if (ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200262 /* stop reading */
Willy Tarreaudb398432018-11-15 11:08:52 +0100263 si_rx_room_blk(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200264 }
265 else {
266 /* (re)start reading */
Willy Tarreau3c39a7d2019-06-14 14:42:29 +0200267 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau07373b82014-11-28 12:08:47 +0100268 if (!(si->flags & SI_FL_DONT_WAKE))
269 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200270 }
271}
272
273/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200274static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200275{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100276 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200277
Willy Tarreauafc8a222014-11-28 15:46:27 +0100278 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200279 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100280 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200281
Willy Tarreauafc8a222014-11-28 15:46:27 +0100282 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200283 return;
284
285 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100286 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200287 return;
288
289 /* Otherwise there are remaining data to be sent in the buffer,
290 * so we tell the handler.
291 */
292 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100293 if (!tick_isset(oc->wex))
294 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200295
Willy Tarreau07373b82014-11-28 12:08:47 +0100296 if (!(si->flags & SI_FL_DONT_WAKE))
297 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200298}
299
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200300/* Register an applet to handle a stream_interface as a new appctx. The SI will
301 * wake it up everytime it is solicited. The appctx must be deleted by the task
302 * handler using si_release_endpoint(), possibly from within the function itself.
303 * It also pre-initializes the applet's context and returns it (or NULL in case
304 * it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200305 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100306struct appctx *si_register_handler(struct stream_interface *si, struct applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200307{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100308 struct appctx *appctx;
309
Willy Tarreau07373b82014-11-28 12:08:47 +0100310 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200311
Willy Tarreaua7513f52015-04-05 00:15:26 +0200312 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100313 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100314 return NULL;
315
Willy Tarreau0cd3bd62018-11-06 18:46:37 +0100316 si_cant_get(si);
Willy Tarreau828824a2015-04-19 17:20:03 +0200317 appctx_wakeup(appctx);
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100318 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200319}
320
Willy Tarreau2c6be842012-07-06 17:12:34 +0200321/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200322 * established. It returns 0 if it fails in a fatal way or needs to poll to go
323 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200324 * flags (the bit is provided in <flag> by the caller). It is designed to be
325 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200326 * Note that it can emit a PROXY line by relying on the other end's address
327 * when the connection is attached to a stream interface, or by resolving the
328 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200329 */
330int conn_si_send_proxy(struct connection *conn, unsigned int flag)
331{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200332 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200333 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200334 goto out_error;
335
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100336 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200337 goto out_error;
338
Willy Tarreau2c6be842012-07-06 17:12:34 +0200339 /* If we have a PROXY line to send, we'll use this to validate the
340 * connection, in which case the connection is validated only once
341 * we've sent the whole proxy line. Otherwise we use connect().
342 */
Tim Duesterhus36839dc2019-02-26 17:09:51 +0100343 if (conn->send_proxy_ofs) {
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100344 const struct conn_stream *cs;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200345 int ret;
346
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100347 cs = cs_get_first(conn);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200348 /* The target server expects a PROXY line to be sent first.
349 * If the send_proxy_ofs is negative, it corresponds to the
350 * offset to start sending from then end of the proxy string
351 * (which is recomputed every time since it's constant). If
352 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200353 * We can only send a "normal" PROXY line when the connection
354 * is attached to a stream interface. Otherwise we can only
355 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200356 */
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100357
358 if (cs && cs->data_cb == &si_conn_cb) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200359 struct stream_interface *si = cs->data;
360 struct conn_stream *remote_cs = objt_cs(si_opposite(si)->end);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200361 ret = make_proxy_line(trash.area, trash.size,
362 objt_server(conn->target),
363 remote_cs ? remote_cs->conn : NULL);
Olivier Houchard661167d2019-05-29 17:08:03 +0200364 /* We may not have a conn_stream yet, if we don't
365 * know which mux to use, because it will be decided
366 * during the SSL handshake. In this case, there should
367 * be a session associated to the connection in
368 * conn->owner, and we know it is the session that
369 * initiated that connection, so we can just use
370 * its origin, which should contain the client
371 * connection.
372 */
373 } else if (!cs && conn->owner) {
374 struct session *sess = conn->owner;
375
376 ret = make_proxy_line(trash.area, trash.size,
377 objt_server(conn->target),
378 objt_conn(sess->origin));
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200379 }
380 else {
381 /* The target server expects a LOCAL line to be sent first. Retrieving
382 * local or remote addresses may fail until the connection is established.
383 */
384 conn_get_from_addr(conn);
385 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
386 goto out_wait;
387
388 conn_get_to_addr(conn);
389 if (!(conn->flags & CO_FL_ADDR_TO_SET))
390 goto out_wait;
391
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200392 ret = make_proxy_line(trash.area, trash.size,
393 objt_server(conn->target), conn);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200394 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200395
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396 if (!ret)
397 goto out_error;
398
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200399 if (conn->send_proxy_ofs > 0)
400 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200401
Willy Tarreaua1a74742012-08-24 12:14:49 +0200402 /* we have to send trash from (ret+sp for -sp bytes). If the
403 * data layer has a pending write, we'll also set MSG_MORE.
404 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200405 ret = conn_sock_send(conn,
406 trash.area + ret + conn->send_proxy_ofs,
407 -conn->send_proxy_ofs,
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200408 (conn->flags & CO_FL_XPRT_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200409
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100410 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200411 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200412
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200413 conn->send_proxy_ofs += ret; /* becomes zero once complete */
414 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415 goto out_wait;
416
417 /* OK we've sent the whole line, we're connected */
418 }
419
Willy Tarreaua1a74742012-08-24 12:14:49 +0200420 /* The connection is ready now, simply return and let the connection
421 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200422 */
423 if (conn->flags & CO_FL_WAIT_L4_CONN)
424 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200425 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200426 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200427
428 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200429 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200430 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200431 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200432
433 out_wait:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200434 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200435}
436
Willy Tarreau27375622013-12-17 00:00:28 +0100437
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100438/* This function is the equivalent to si_update() except that it's
Willy Tarreau615f28b2015-09-23 18:40:09 +0200439 * designed to be called from outside the stream handlers, typically the lower
440 * layers (applets, connections) after I/O completion. After updating the stream
441 * interface and timeouts, it will try to forward what can be forwarded, then to
442 * wake the associated task up if an important event requires special handling.
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100443 * It may update SI_FL_WAIT_DATA and/or SI_FL_RXBLK_ROOM, that the callers are
Willy Tarreau0dfccb22018-10-25 13:55:20 +0200444 * encouraged to watch to take appropriate action.
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100445 * It should not be called from within the stream itself, si_update()
Willy Tarreau615f28b2015-09-23 18:40:09 +0200446 * is designed for this.
447 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100448static void stream_int_notify(struct stream_interface *si)
Willy Tarreau615f28b2015-09-23 18:40:09 +0200449{
450 struct channel *ic = si_ic(si);
451 struct channel *oc = si_oc(si);
Willy Tarreau47baeb82018-11-15 07:46:57 +0100452 struct stream_interface *sio = si_opposite(si);
Christopher Fauletd7607de2019-01-03 16:24:54 +0100453 struct task *task = si_task(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200454
455 /* process consumer side */
456 if (channel_is_empty(oc)) {
Olivier Houcharde9bed532017-11-16 17:49:25 +0100457 struct connection *conn = objt_cs(si->end) ? objt_cs(si->end)->conn : NULL;
458
Willy Tarreau615f28b2015-09-23 18:40:09 +0200459 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Olivier Houcharde9bed532017-11-16 17:49:25 +0100460 (si->state == SI_ST_EST) && (!conn || !(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS))))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200461 si_shutw(si);
462 oc->wex = TICK_ETERNITY;
463 }
464
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100465 /* indicate that we may be waiting for data from the output channel or
466 * we're about to close and can't expect more data if SHUTW_NOW is there.
467 */
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200468 if (!(oc->flags & (CF_SHUTW|CF_SHUTW_NOW)))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200469 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100470 else if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW)
471 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200472
473 /* update OC timeouts and wake the other side up if it's waiting for room */
474 if (oc->flags & CF_WRITE_ACTIVITY) {
475 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
476 !channel_is_empty(oc))
477 if (tick_isset(oc->wex))
478 oc->wex = tick_add_ifset(now_ms, oc->wto);
479
480 if (!(si->flags & SI_FL_INDEP_STR))
481 if (tick_isset(ic->rex))
482 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf26c26c2018-11-12 16:11:08 +0100483 }
Willy Tarreau615f28b2015-09-23 18:40:09 +0200484
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100485 if (oc->flags & CF_DONT_READ)
486 si_rx_chan_blk(sio);
487 else
488 si_rx_chan_rdy(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200489
490 /* Notify the other side when we've injected data into the IC that
491 * needs to be forwarded. We can do fast-forwarding as soon as there
492 * are output data, but we avoid doing this if some of the data are
493 * not yet scheduled for being forwarded, because it is very likely
494 * that it will be done again immediately afterwards once the following
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100495 * data are parsed (eg: HTTP chunking). We only SI_FL_RXBLK_ROOM once
Willy Tarreau615f28b2015-09-23 18:40:09 +0200496 * we've emptied *some* of the output buffer, and not just when there
497 * is available room, because applets are often forced to stop before
498 * the buffer is full. We must not stop based on input data alone because
499 * an HTTP parser might need more data to complete the parsing.
500 */
501 if (!channel_is_empty(ic) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100502 (sio->flags & SI_FL_WAIT_DATA) &&
Willy Tarreau89b6a2b2018-11-18 15:46:10 +0100503 (!(ic->flags & CF_EXPECT_MORE) || c_full(ic) || ci_data(ic) == 0 || ic->pipe)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200504 int new_len, last_len;
505
Willy Tarreau77e478c2018-06-19 07:03:14 +0200506 last_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200507 if (ic->pipe)
508 last_len += ic->pipe->data;
509
Willy Tarreau47baeb82018-11-15 07:46:57 +0100510 si_chk_snd(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200511
Willy Tarreau77e478c2018-06-19 07:03:14 +0200512 new_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200513 if (ic->pipe)
514 new_len += ic->pipe->data;
515
516 /* check if the consumer has freed some space either in the
517 * buffer or in the pipe.
518 */
Willy Tarreau47baeb82018-11-15 07:46:57 +0100519 if (new_len < last_len)
Willy Tarreaudb398432018-11-15 11:08:52 +0100520 si_rx_room_rdy(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200521 }
522
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100523 if (!(ic->flags & CF_DONT_READ))
524 si_rx_chan_rdy(si);
525
Willy Tarreau47baeb82018-11-15 07:46:57 +0100526 si_chk_rcv(si);
527 si_chk_rcv(sio);
528
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100529 if (si_rx_blocked(si)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200530 ic->rex = TICK_ETERNITY;
531 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100532 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL)) == CF_READ_PARTIAL) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200533 /* we must re-enable reading if si_chk_snd() has freed some space */
534 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
535 ic->rex = tick_add_ifset(now_ms, ic->rto);
536 }
537
538 /* wake the task up only when needed */
539 if (/* changes on the production side */
540 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200541 !si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200542 (si->flags & SI_FL_ERR) ||
543 ((ic->flags & CF_READ_PARTIAL) &&
Christopher Faulet297d3e22019-03-22 14:16:14 +0100544 ((ic->flags & CF_EOI) || !ic->to_forward || sio->state != SI_ST_EST)) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200545
546 /* changes on the consumption side */
547 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
Willy Tarreauede3d882018-10-24 17:17:56 +0200548 ((oc->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreau615f28b2015-09-23 18:40:09 +0200549 ((oc->flags & CF_SHUTW) ||
Willy Tarreau78f5ff82018-12-19 11:00:00 +0100550 (((oc->flags & CF_WAKE_WRITE) ||
551 !(oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW|CF_SHUTW))) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100552 (sio->state != SI_ST_EST ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200553 (channel_is_empty(oc) && !oc->to_forward)))))) {
Christopher Fauletd7607de2019-01-03 16:24:54 +0100554 task_wakeup(task, TASK_WOKEN_IO);
555 }
556 else {
557 /* Update expiration date for the task and requeue it */
558 task->expire = tick_first((tick_is_expired(task->expire, now_ms) ? 0 : task->expire),
559 tick_first(tick_first(ic->rex, ic->wex),
560 tick_first(oc->rex, oc->wex)));
Willy Tarreau7343c712019-08-01 18:51:38 +0200561
562 task->expire = tick_first(task->expire, ic->analyse_exp);
563 task->expire = tick_first(task->expire, oc->analyse_exp);
564
565 if (si->exp)
566 task->expire = tick_first(task->expire, si->exp);
567
568 if (sio->exp)
569 task->expire = tick_first(task->expire, sio->exp);
570
Christopher Fauletd7607de2019-01-03 16:24:54 +0100571 task_queue(task);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200572 }
573 if (ic->flags & CF_READ_ACTIVITY)
574 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200575}
576
577
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200578/* Called by I/O handlers after completion.. It propagates
Willy Tarreau651e1822015-09-23 20:06:13 +0200579 * connection flags to the stream interface, updates the stream (which may or
580 * may not take this opportunity to try to forward data), then update the
581 * connection's polling based on the channels and stream interface's final
582 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200583 */
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200584static int si_cs_process(struct conn_stream *cs)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200585{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200586 struct connection *conn = cs->conn;
587 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100588 struct channel *ic = si_ic(si);
589 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200590
Olivier Houchardc7ffa912018-08-28 19:37:41 +0200591 /* If we have data to send, try it now */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100592 if (!channel_is_empty(oc) && !(si->wait_event.events & SUB_RETRY_SEND))
Willy Tarreau908d26f2018-10-25 14:02:47 +0200593 si_cs_send(cs);
594
Willy Tarreau651e1822015-09-23 20:06:13 +0200595 /* First step, report to the stream-int what was detected at the
596 * connection layer : errors and connection establishment.
Olivier Houchard6187cc32019-06-24 16:08:08 +0200597 * Only add SI_FL_ERR if we're connected, or we're attempting to
598 * connect, we may get there because we got woken up, but only run
599 * after process_stream() noticed there were an error, and decided
600 * to retry to connect, the connection may still have CO_FL_ERROR,
601 * and we don't want to add SI_FL_ERR back
Christopher Fauletf7374102019-11-20 11:56:33 +0100602 *
603 * Note: This test is only required because si_cs_process is also the SI
604 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
605 * care of it.
Willy Tarreau651e1822015-09-23 20:06:13 +0200606 */
Olivier Houchard6187cc32019-06-24 16:08:08 +0200607 if (si->state >= SI_ST_CON &&
608 (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR))
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200609 si->flags |= SI_FL_ERR;
610
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200611 /* If we had early data, and the handshake ended, then
612 * we can remove the flag, and attempt to wake the task up,
613 * in the event there's an analyser waiting for the end of
614 * the handshake.
615 */
Olivier Houchard6fa63d92017-11-27 18:41:32 +0100616 if (!(conn->flags & (CO_FL_HANDSHAKE | CO_FL_EARLY_SSL_HS)) &&
617 (cs->flags & CS_FL_WAIT_FOR_HS)) {
618 cs->flags &= ~CS_FL_WAIT_FOR_HS;
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200619 task_wakeup(si_task(si), TASK_WOKEN_MSG);
620 }
621
Willy Tarreau7ab22adb2019-06-05 14:53:22 +0200622 if (!si_state_in(si->state, SI_SB_EST|SI_SB_DIS|SI_SB_CLO) &&
Willy Tarreau52821e22017-03-18 17:11:37 +0100623 (conn->flags & (CO_FL_CONNECTED | CO_FL_HANDSHAKE)) == CO_FL_CONNECTED) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200624 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100625 oc->flags |= CF_WRITE_NULL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +0200626 if (si->state == SI_ST_CON)
627 si->state = SI_ST_RDY;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200628 }
629
Christopher Faulet297d3e22019-03-22 14:16:14 +0100630 /* Report EOI on the channel if it was reached from the mux point of
Christopher Fauletf7374102019-11-20 11:56:33 +0100631 * view.
632 *
633 * Note: This test is only required because si_cs_process is also the SI
634 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
635 * care of it.
636 */
Christopher Faulet297d3e22019-03-22 14:16:14 +0100637 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200638 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100639
Willy Tarreau651e1822015-09-23 20:06:13 +0200640 /* Second step : update the stream-int and channels, try to forward any
641 * pending data, then possibly wake the stream up based on the new
642 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200643 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200644 stream_int_notify(si);
Christopher Fauleta73e59b2016-12-09 17:30:18 +0100645 channel_release_buffer(ic, &(si_strm(si)->buffer_wait));
Willy Tarreauea3cc482015-09-23 19:37:00 +0200646
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200647 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200648}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200649
Willy Tarreau5368d802012-08-21 18:22:06 +0200650/*
651 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200652 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800653 * caller to commit polling changes. The caller should check conn->flags
654 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200655 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200656int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200657{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200658 struct connection *conn = cs->conn;
659 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100660 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200661 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200662 int did_send = 0;
663
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100664 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchard6187cc32019-06-24 16:08:08 +0200665 /* We're probably there because the tasklet was woken up,
666 * but process_stream() ran before, detected there were an
667 * error and put the si back to SI_ST_TAR. There's still
668 * CO_FL_ERROR on the connection but we don't want to add
669 * SI_FL_ERR back, so give up
670 */
671 if (si->state < SI_ST_CON)
672 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100673 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200674 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100675 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200676
Christopher Fauletcfdef9f2019-09-23 15:57:29 +0200677 /* We're already waiting to be able to send, give up */
678 if (si->wait_event.events & SUB_RETRY_SEND)
679 return 0;
680
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200681 /* we might have been called just after an asynchronous shutw */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100682 if (conn->flags & CO_FL_SOCK_WR_SH || oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200683 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200684
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200685 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
686 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet0059e992019-07-05 11:49:11 +0200687 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200688 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200689
Willy Tarreauafc8a222014-11-28 15:46:27 +0100690 if (!oc->pipe->data) {
691 put_pipe(oc->pipe);
692 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200693 }
694
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100695 if (oc->pipe)
696 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200697 }
698
699 /* At this point, the pipe is empty, but we may still have data pending
700 * in the normal buffer.
701 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100702 if (co_data(oc)) {
703 /* when we're here, we already know that there is no spliced
704 * data left, and that there are sendable buffered data.
705 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200706
Willy Tarreau5368d802012-08-21 18:22:06 +0200707 /* check if we want to inform the kernel that we're interested in
708 * sending more data after this call. We want this if :
709 * - we're about to close after this last send and want to merge
710 * the ongoing FIN with the last segment.
711 * - we know we can't send everything at once and must get back
712 * here because of unaligned data
713 * - there is still a finite amount of data to forward
714 * The test is arranged so that the most common case does only 2
715 * tests.
716 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100717 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200718
Willy Tarreauafc8a222014-11-28 15:46:27 +0100719 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
720 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau4ac49282017-10-17 16:33:46 +0200721 (oc->flags & CF_EXPECT_MORE))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100722 ((oc->flags & CF_ISRESP) &&
723 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100724 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200725
Willy Tarreauafc8a222014-11-28 15:46:27 +0100726 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100727 send_flag |= CO_SFL_STREAMER;
728
Olivier Houcharda254a372019-04-05 15:30:12 +0200729 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
Olivier Houchardce1a0292019-05-17 15:38:29 +0200730 struct stream *s = si_strm(si);
Olivier Houcharda254a372019-04-05 15:30:12 +0200731 /* If we want to be able to do L7 retries, copy
732 * the data we're about to send, so that we are able
733 * to resend them if needed
734 */
735 /* Try to allocate a buffer if we had none.
736 * If it fails, the next test will just
737 * disable the l7 retries by setting
738 * l7_conn_retries to 0.
739 */
Olivier Houchardce1a0292019-05-17 15:38:29 +0200740 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
Olivier Houcharda254a372019-04-05 15:30:12 +0200741 si->flags &= ~SI_FL_L7_RETRY;
742 else {
743 if (b_is_null(&si->l7_buffer))
744 b_alloc(&si->l7_buffer);
745 if (b_is_null(&si->l7_buffer))
746 si->flags &= ~SI_FL_L7_RETRY;
747 else {
748 memcpy(b_orig(&si->l7_buffer),
749 b_orig(&oc->buf),
750 b_size(&oc->buf));
751 si->l7_buffer.head = co_data(oc);
752 b_add(&si->l7_buffer, co_data(oc));
753 }
754
755 }
756 }
757
Olivier Houcharded0f2072018-08-16 15:41:52 +0200758 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800759 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200760 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200761 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200762 c_realign_if_empty(oc);
763
764 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800765 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100766 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800767 }
Godbache68e02d2013-10-11 15:48:29 +0800768 /* if some data remain in the buffer, it's only because the
769 * system buffers are full, we will try next time.
770 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200771 }
Godbache68e02d2013-10-11 15:48:29 +0800772 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100773
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100774 end:
Christopher Faulet0059e992019-07-05 11:49:11 +0200775 if (did_send) {
776 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
777 if (si->state == SI_ST_CON)
778 si->state = SI_ST_RDY;
Christopher Fauletab1300b2019-07-05 13:44:29 +0200779
780 si_rx_room_rdy(si_opposite(si));
Christopher Faulet0059e992019-07-05 11:49:11 +0200781 }
782
783 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
784 si->flags |= SI_FL_ERR;
785 return 1;
786 }
787
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200788 /* 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 +0100789 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100790 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200791 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200792}
793
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100794/* This is the ->process() function for any stream-interface's wait_event task.
795 * It's assigned during the stream-interface's initialization, for any type of
796 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
797 * stream interface, as the presence of the CS is checked there.
798 */
Olivier Houchard91894cb2018-08-02 18:06:28 +0200799struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned short state)
800{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200801 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200802 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200803 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200804
805 if (!cs)
806 return NULL;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100807
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100808 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200809 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100810 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200811 ret |= si_cs_recv(cs);
812 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200813 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200814
Olivier Houchard91894cb2018-08-02 18:06:28 +0200815 return (NULL);
816}
817
Willy Tarreau25f13102015-09-24 11:32:22 +0200818/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200819 * update the input channel's expiration timer and the stream interface's
820 * Rx flags based on the channel's flags. It needs to be called only once
821 * after the channel's flags have settled down, and before they are cleared,
822 * though it doesn't harm to call it as often as desired (it just slightly
823 * hurts performance). It must not be called from outside of the stream
824 * handler, as what it does will be used to compute the stream task's
825 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200826 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200827void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200828{
829 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200830
Willy Tarreau236c4292019-06-06 08:19:20 +0200831 if (ic->flags & CF_SHUTR) {
832 si_rx_shut_blk(si);
833 return;
834 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100835
Willy Tarreau236c4292019-06-06 08:19:20 +0200836 /* Read not closed, update FD status and timeout for reads */
837 if (ic->flags & CF_DONT_READ)
838 si_rx_chan_blk(si);
839 else
840 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100841
Willy Tarreau236c4292019-06-06 08:19:20 +0200842 if (!channel_is_empty(ic)) {
843 /* stop reading, imposed by channel's policy or contents */
844 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200845 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200846 else {
847 /* (re)start reading and update timeout. Note: we don't recompute the timeout
848 * everytime we get here, otherwise it would risk never to expire. We only
849 * update it if is was not yet set. The stream socket handler will already
850 * have updated it if there has been a completed I/O.
851 */
852 si_rx_room_rdy(si);
853 }
854 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
855 ic->rex = TICK_ETERNITY;
856 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
857 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200858
Willy Tarreau236c4292019-06-06 08:19:20 +0200859 si_chk_rcv(si);
860}
861
862/* This function is designed to be called from within the stream handler to
863 * update the output channel's expiration timer and the stream interface's
864 * Tx flags based on the channel's flags. It needs to be called only once
865 * after the channel's flags have settled down, and before they are cleared,
866 * though it doesn't harm to call it as often as desired (it just slightly
867 * hurts performance). It must not be called from outside of the stream
868 * handler, as what it does will be used to compute the stream task's
869 * expiration.
870 */
871void si_update_tx(struct stream_interface *si)
872{
873 struct channel *oc = si_oc(si);
874 struct channel *ic = si_ic(si);
875
876 if (oc->flags & CF_SHUTW)
877 return;
878
879 /* Write not closed, update FD status and timeout for writes */
880 if (channel_is_empty(oc)) {
881 /* stop writing */
882 if (!(si->flags & SI_FL_WAIT_DATA)) {
883 if ((oc->flags & CF_SHUTW_NOW) == 0)
884 si->flags |= SI_FL_WAIT_DATA;
885 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200886 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200887 return;
888 }
889
890 /* (re)start writing and update timeout. Note: we don't recompute the timeout
891 * everytime we get here, otherwise it would risk never to expire. We only
892 * update it if is was not yet set. The stream socket handler will already
893 * have updated it if there has been a completed I/O.
894 */
895 si->flags &= ~SI_FL_WAIT_DATA;
896 if (!tick_isset(oc->wex)) {
897 oc->wex = tick_add_ifset(now_ms, oc->wto);
898 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
899 /* Note: depending on the protocol, we don't know if we're waiting
900 * for incoming data or not. So in order to prevent the socket from
901 * expiring read timeouts during writes, we refresh the read timeout,
902 * except if it was already infinite or if we have explicitly setup
903 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200904 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200905 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200906 }
907 }
908}
909
Willy Tarreau3b285d72019-06-06 08:20:17 +0200910/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
911 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
912 * be updated in case of success.
913 */
914void si_sync_send(struct stream_interface *si)
915{
916 struct channel *oc = si_oc(si);
917 struct conn_stream *cs;
918
919 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
920
921 if (oc->flags & CF_SHUTW)
922 return;
923
924 if (channel_is_empty(oc))
925 return;
926
927 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
928 return;
929
930 cs = objt_cs(si->end);
931 if (!cs)
932 return;
933
Christopher Fauletab1300b2019-07-05 13:44:29 +0200934 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200935}
936
Willy Tarreaud14844a2018-11-08 18:15:29 +0100937/* Updates at once the channel flags, and timers of both stream interfaces of a
938 * same stream, to complete the work after the analysers, then updates the data
939 * layer below. This will ensure that any synchronous update performed at the
940 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200941 * Note that this does not change the stream interface's current state, though
942 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100943 */
944void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
945{
946 struct channel *req = si_ic(si_f);
947 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100948
949 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
950 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
951
Willy Tarreaud14844a2018-11-08 18:15:29 +0100952 si_f->prev_state = si_f->state;
953 si_b->prev_state = si_b->state;
954
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100955 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200956 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100957 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100958
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200959 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100960 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100961
962 /* stream ints are processed outside of process_stream() and must be
963 * handled at the latest moment.
964 */
965 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100966 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
967 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100968 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100969
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100970 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100971 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
972 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100973 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100974}
975
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200976/*
977 * This function performs a shutdown-read on a stream interface attached to
978 * a connection in a connected or init state (it does nothing for other
979 * states). It either shuts the read side or marks itself as closed. The buffer
980 * flags are updated to reflect the new state. If the stream interface has
981 * SI_FL_NOHALF, we also forward the close to the write side. If a control
982 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200983 * descriptors are then shutdown or closed accordingly. The function
984 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200985 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200986static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200987{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200988 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100989 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200990
Willy Tarreauabb5d422018-11-14 16:58:52 +0100991 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100992 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200993 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100994 ic->flags |= CF_SHUTR;
995 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200996
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200997 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200998 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200999
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001000 if (si->flags & SI_FL_KILL_CONN)
1001 cs->flags |= CS_FL_KILL_CONN;
1002
Willy Tarreau2bb4a962014-11-28 11:11:05 +01001003 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02001004 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001005 si->state = SI_ST_DIS;
1006 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001007 }
1008 else if (si->flags & SI_FL_NOHALF) {
1009 /* we want to immediately forward this close to the write side */
1010 return stream_int_shutw_conn(si);
1011 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001012}
1013
1014/*
1015 * This function performs a shutdown-write on a stream interface attached to
1016 * a connection in a connected or init state (it does nothing for other
1017 * states). It either shuts the write side or marks itself as closed. The
1018 * buffer flags are updated to reflect the new state. It does also close
1019 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001020 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001021 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001022static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001023{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001024 struct conn_stream *cs = __objt_cs(si->end);
1025 struct connection *conn = cs->conn;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001026 struct channel *ic = si_ic(si);
1027 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001028
Willy Tarreauafc8a222014-11-28 15:46:27 +01001029 oc->flags &= ~CF_SHUTW_NOW;
1030 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001031 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001032 oc->flags |= CF_SHUTW;
1033 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001034 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001035
Hongbo Longe39683c2017-03-10 18:41:51 +01001036 if (tick_isset(si->hcto)) {
1037 ic->rto = si->hcto;
1038 ic->rex = tick_add(now_ms, ic->rto);
1039 }
1040
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001041 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001042 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001043 case SI_ST_EST:
1044 /* we have to shut before closing, otherwise some short messages
1045 * may never leave the system, especially when there are remaining
1046 * unread data in the socket input buffer, or when nolinger is set.
1047 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1048 * no risk so we close both sides immediately.
1049 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001050 if (si->flags & SI_FL_KILL_CONN)
1051 cs->flags |= CS_FL_KILL_CONN;
1052
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001053 if (si->flags & SI_FL_ERR) {
1054 /* quick close, the socket is alredy shut anyway */
1055 }
1056 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001057 /* unclean data-layer shutdown, typically an aborted request
1058 * or a forwarded shutdown from a client to a server due to
1059 * option abortonclose. No need for the TLS layer to try to
1060 * emit a shutdown message.
1061 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001062 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001063 }
1064 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001065 /* clean data-layer shutdown. This only happens on the
1066 * frontend side, or on the backend side when forwarding
1067 * a client close in TCP mode or in HTTP TUNNEL mode
1068 * while option abortonclose is set. We want the TLS
1069 * layer to try to signal it to the peer before we close.
1070 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001071 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001072
Willy Tarreaua553ae92017-10-05 18:52:17 +02001073 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
1074 /* OK just a shutw, but we want the caller
1075 * to disable polling on this FD if exists.
1076 */
1077 conn_cond_update_polling(conn);
1078 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001079 }
1080 }
1081
1082 /* fall through */
1083 case SI_ST_CON:
1084 /* we may have to close a pending connection, and mark the
1085 * response buffer as shutr
1086 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001087 if (si->flags & SI_FL_KILL_CONN)
1088 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001089 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001090 /* fall through */
1091 case SI_ST_CER:
1092 case SI_ST_QUE:
1093 case SI_ST_TAR:
1094 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001095 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001096 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001097 si->flags &= ~SI_FL_NOLINGER;
1098 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001099 ic->flags |= CF_SHUTR;
1100 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001101 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001102 }
1103}
1104
Willy Tarreau46a8d922012-08-20 12:38:36 +02001105/* This function is used for inter-stream-interface calls. It is called by the
1106 * consumer to inform the producer side that it may be interested in checking
1107 * for free space in the buffer. Note that it intentionally does not update
1108 * timeouts, so that we can still check them later at wake-up. This function is
1109 * dedicated to connection-based stream interfaces.
1110 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001111static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001112{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001113 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001114 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001115 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001116}
1117
1118
Willy Tarreaude5722c2012-08-20 15:01:10 +02001119/* This function is used for inter-stream-interface calls. It is called by the
1120 * producer to inform the consumer side that it may be interested in checking
1121 * for data in the buffer. Note that it intentionally does not update timeouts,
1122 * so that we can still check them later at wake-up.
1123 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001124static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001125{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001126 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001127 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001128
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001129 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001130 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001131 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001132
Willy Tarreauafc8a222014-11-28 15:46:27 +01001133 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001134 return;
1135
Willy Tarreauafc8a222014-11-28 15:46:27 +01001136 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001137 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001138 return;
1139
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001140 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001141 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001142
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001143 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001144 /* Write error on the file descriptor */
Olivier Houchard6187cc32019-06-24 16:08:08 +02001145 if (si->state >= SI_ST_CON)
1146 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001147 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001148 }
1149
1150 /* OK, so now we know that some data might have been sent, and that we may
1151 * have to poll first. We have to do that too if the buffer is not empty.
1152 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001153 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001154 /* the connection is established but we can't write. Either the
1155 * buffer is empty, or we just refrain from sending because the
1156 * ->o limit was reached. Maybe we just wrote the last
1157 * chunk and need to close.
1158 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001159 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001160 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001161 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001162 si_shutw(si);
1163 goto out_wakeup;
1164 }
1165
Willy Tarreauafc8a222014-11-28 15:46:27 +01001166 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001167 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001168 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001169 }
1170 else {
1171 /* Otherwise there are remaining data to be sent in the buffer,
1172 * which means we have to poll before doing so.
1173 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001174 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001175 if (!tick_isset(oc->wex))
1176 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001177 }
1178
Willy Tarreauafc8a222014-11-28 15:46:27 +01001179 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1180 struct channel *ic = si_ic(si);
1181
Willy Tarreaude5722c2012-08-20 15:01:10 +02001182 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001183 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1184 !channel_is_empty(oc))
1185 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001186
Willy Tarreauafc8a222014-11-28 15:46:27 +01001187 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001188 /* Note: to prevent the client from expiring read timeouts
1189 * during writes, we refresh it. We only do this if the
1190 * interface is not configured for "independent streams",
1191 * because for some applications it's better not to do this,
1192 * for instance when continuously exchanging small amounts
1193 * of data which can full the socket buffers long before a
1194 * write timeout is detected.
1195 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001196 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001197 }
1198 }
1199
1200 /* in case of special condition (error, shutdown, end of write...), we
1201 * have to notify the task.
1202 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001203 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1204 ((oc->flags & CF_WAKE_WRITE) &&
1205 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001206 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001207 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001208 if (!(si->flags & SI_FL_DONT_WAKE))
1209 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001210 }
1211}
1212
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001213/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001214 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001215 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001216 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001217 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001218int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001219{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001220 struct connection *conn = cs->conn;
1221 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001222 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001223 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001224 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001225 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001226
Christopher Faulet27ebcef2019-10-25 10:21:01 +02001227 /* If not established yet, do nothing. */
1228 if (si->state != SI_ST_EST)
1229 return 0;
1230
Olivier Houchardf6535282018-08-31 17:29:12 +02001231 /* If another call to si_cs_recv() failed, and we subscribed to
1232 * recv events already, give up now.
1233 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001234 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001235 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001236
Willy Tarreauce323de2012-08-20 21:41:06 +02001237 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001238 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001239 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001240
Willy Tarreau54e917c2017-08-30 07:35:35 +02001241 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001242 if (cs->flags & CS_FL_EOS)
Christopher Fauletf7374102019-11-20 11:56:33 +01001243 goto end_recv;
Willy Tarreau54e917c2017-08-30 07:35:35 +02001244
Christopher Fauletf061e422018-12-07 14:51:20 +01001245 /* stop immediately on errors. Note that we DON'T want to stop on
1246 * POLL_ERR, as the poller might report a write error while there
1247 * are still data available in the recv buffer. This typically
1248 * happens when we send too large a request to a backend server
1249 * which rejects it before reading it all.
1250 */
1251 if (!(cs->flags & CS_FL_RCV_MORE)) {
1252 if (!conn_xprt_ready(conn))
1253 return 0;
1254 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Christopher Fauletf7374102019-11-20 11:56:33 +01001255 goto end_recv;
Christopher Fauletf061e422018-12-07 14:51:20 +01001256 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001257
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001258 /* prepare to detect if the mux needs more room */
1259 cs->flags &= ~CS_FL_WANT_ROOM;
1260
Willy Tarreau77e478c2018-06-19 07:03:14 +02001261 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001262 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001263 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001264 /* The buffer was empty and nothing was transferred for more
1265 * than one second. This was caused by a pause and not by
1266 * congestion. Reset any streaming mode to reduce latency.
1267 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001268 ic->xfer_small = 0;
1269 ic->xfer_large = 0;
1270 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001271 }
1272
Willy Tarreau96199b12012-08-24 00:46:52 +02001273 /* First, let's see if we may splice data across the channel without
1274 * using a buffer.
1275 */
Willy Tarreau7e9dd732020-01-17 16:19:34 +01001276 if (cs->flags & CS_FL_MAY_SPLICE &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001277 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1278 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001279 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001280 /* We're embarrassed, there are already data pending in
1281 * the buffer and we don't want to have them at two
1282 * locations at a time. Let's indicate we need some
1283 * place and ask the consumer to hurry.
1284 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001285 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001286 goto abort_splice;
1287 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001288
Willy Tarreauafc8a222014-11-28 15:46:27 +01001289 if (unlikely(ic->pipe == NULL)) {
1290 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1291 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001292 goto abort_splice;
1293 }
1294 }
1295
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001296 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001297 if (ret < 0) {
1298 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001299 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001300 goto abort_splice;
1301 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001302
Willy Tarreau96199b12012-08-24 00:46:52 +02001303 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001304 if (ic->to_forward != CHN_INFINITE_FORWARD)
1305 ic->to_forward -= ret;
1306 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001307 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001308 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001309 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001310
Christopher Fauletf7374102019-11-20 11:56:33 +01001311 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_EOS|CS_FL_ERROR))
1312 goto end_recv;
Willy Tarreau96199b12012-08-24 00:46:52 +02001313
Willy Tarreau61d39a02013-07-18 21:49:32 +02001314 if (conn->flags & CO_FL_WAIT_ROOM) {
1315 /* the pipe is full or we have read enough data that it
1316 * could soon be full. Let's stop before needing to poll.
1317 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001318 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001319 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001320 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001321
Willy Tarreauce323de2012-08-20 21:41:06 +02001322 /* splice not possible (anymore), let's go on on standard copy */
1323 }
Willy Tarreau81464b42018-11-15 15:52:53 +01001324 else {
1325 /* be sure not to block regular receive path below */
1326 conn->flags &= ~CO_FL_WAIT_ROOM;
1327 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001328
1329 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001330 if (ic->pipe && unlikely(!ic->pipe->data)) {
1331 put_pipe(ic->pipe);
1332 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001333 }
1334
Willy Tarreau7e9dd732020-01-17 16:19:34 +01001335 if (ic->pipe && ic->to_forward && !(flags & CO_RFL_BUF_FLUSH) && cs->flags & CS_FL_MAY_SPLICE) {
Willy Tarreau791b18b2019-12-03 18:13:04 +01001336 /* don't break splicing by reading, but still call rcv_buf()
1337 * to pass the flag.
1338 */
1339 goto done_recv;
1340 }
1341
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001342 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001343 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001344 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001345
Willy Tarreau61d39a02013-07-18 21:49:32 +02001346 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1347 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1348 * that if such an event is not handled above in splice, it will be handled here by
1349 * recv().
1350 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001351 while ((cs->flags & CS_FL_RCV_MORE) ||
1352 (!(conn->flags & (CO_FL_ERROR | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE)) &&
1353 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001354 /* <max> may be null. This is the mux responsibility to set
1355 * CS_FL_RCV_MORE on the CS if more space is needed.
1356 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001357 max = channel_recv_max(ic);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001358 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 +01001359
Olivier Houchardd247be02018-12-06 16:22:29 +01001360 if (cs->flags & CS_FL_WANT_ROOM)
Willy Tarreaudb398432018-11-15 11:08:52 +01001361 si_rx_room_blk(si);
Willy Tarreau6577b482017-12-10 21:19:33 +01001362
Christopher Fauleteffc3752018-10-31 08:53:54 +01001363 if (cs->flags & CS_FL_READ_PARTIAL) {
1364 if (tick_isset(ic->rex))
1365 ic->rex = tick_add_ifset(now_ms, ic->rto);
1366 cs->flags &= ~CS_FL_READ_PARTIAL;
1367 }
1368
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001369 if (ret <= 0) {
Willy Tarreau9dc67722019-12-03 18:08:45 +01001370 /* if we refrained from reading because we asked for a
1371 * flush to satisfy rcv_pipe(), we must not subscribe
1372 * and instead report that there's not enough room
1373 * here to proceed.
1374 */
1375 if (flags & CO_RFL_BUF_FLUSH)
1376 si_rx_room_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001377 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001378 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001379
Christopher Faulete7810372020-01-09 14:31:13 +01001380 /* L7 retries enabled and maximum connection retries not reached */
1381 if ((si->flags & SI_FL_L7_RETRY) && si->conn_retries) {
Olivier Houcharda254a372019-04-05 15:30:12 +02001382 struct htx *htx;
1383 struct htx_sl *sl;
1384
1385 htx = htxbuf(&ic->buf);
1386 if (htx) {
Christopher Faulet297fbb42019-05-13 14:41:27 +02001387 sl = http_get_stline(htx);
Olivier Houcharda254a372019-04-05 15:30:12 +02001388 if (sl && l7_status_match(si_strm(si)->be,
1389 sl->info.res.status)) {
1390 /* If we got a status for which we would
1391 * like to retry the request, empty
1392 * the buffer and pretend there's an
1393 * error on the channel.
1394 */
1395 ic->flags |= CF_READ_ERROR;
1396 htx_reset(htx);
1397 return 1;
1398 }
1399 }
1400 si->flags &= ~SI_FL_L7_RETRY;
1401 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001402 cur_read += ret;
1403
1404 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001405 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001406 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001407 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1408 if (fwd > ic->to_forward)
1409 fwd = ic->to_forward;
1410 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001411 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001412 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001413 }
1414
Willy Tarreauafc8a222014-11-28 15:46:27 +01001415 ic->flags |= CF_READ_PARTIAL;
1416 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001417
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001418 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1419 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001420 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001421 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001422 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001423
1424 /* if too many bytes were missing from last read, it means that
1425 * it's pointless trying to read again because the system does
1426 * not have them in buffers.
1427 */
1428 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001429 /* if a streamer has read few data, it may be because we
1430 * have exhausted system buffers. It's not worth trying
1431 * again.
1432 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001433 if (ic->flags & CF_STREAMER) {
1434 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001435 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001436 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001437 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001438
1439 /* if we read a large block smaller than what we requested,
1440 * it's almost certain we'll never get anything more.
1441 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001442 if (ret >= global.tune.recv_enough) {
1443 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001444 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001445 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001446 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001447 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001448
1449 /* if we are waiting for more space, don't try to read more data
1450 * right now.
1451 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001452 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001453 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001454 } /* while !flags */
1455
Willy Tarreauffb12052018-11-15 16:06:02 +01001456 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001457 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001458 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001459 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001460 ic->xfer_large = 0;
1461 ic->xfer_small++;
1462 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001463 /* we have read less than half of the buffer in
1464 * one pass, and this happened at least 3 times.
1465 * This is definitely not a streamer.
1466 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001467 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001468 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001469 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001470 /* if the buffer has been at least half full twice,
1471 * we receive faster than we send, so at least it
1472 * is not a "fast streamer".
1473 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001474 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001475 }
1476 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001477 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001478 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001479 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001480 ic->xfer_small = 0;
1481 ic->xfer_large++;
1482 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001483 /* we call this buffer a fast streamer if it manages
1484 * to be filled in one call 3 consecutive times.
1485 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001486 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001487 }
1488 }
1489 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001490 ic->xfer_small = 0;
1491 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001492 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001493 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001494 }
1495
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001496 end_recv:
Christopher Fauleta3279522019-11-20 16:42:00 +01001497 ret = (cur_read != 0);
1498
Christopher Fauletf7374102019-11-20 11:56:33 +01001499 /* Report EOI on the channel if it was reached from the mux point of
1500 * view. */
Christopher Fauleta3279522019-11-20 16:42:00 +01001501 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI)) {
Christopher Fauletf7374102019-11-20 11:56:33 +01001502 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Fauleta3279522019-11-20 16:42:00 +01001503 ret = 1;
1504 }
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001505
Christopher Fauletf7374102019-11-20 11:56:33 +01001506 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
1507 cs->flags |= CS_FL_ERROR;
1508 si->flags |= SI_FL_ERR;
Christopher Fauleta3279522019-11-20 16:42:00 +01001509 ret = 1;
Christopher Fauletf7374102019-11-20 11:56:33 +01001510 }
1511 else if (cs->flags & CS_FL_EOS) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001512 /* connection closed */
Christopher Fauletf7374102019-11-20 11:56:33 +01001513 if (conn->flags & CO_FL_CONNECTED) {
1514 /* we received a shutdown */
1515 ic->flags |= CF_READ_NULL;
1516 if (ic->flags & CF_AUTO_CLOSE)
1517 channel_shutw_now(ic);
1518 stream_int_read0(si);
1519 }
Christopher Fauleta3279522019-11-20 16:42:00 +01001520 ret = 1;
Christopher Fauletf7374102019-11-20 11:56:33 +01001521 }
1522 else if (!si_rx_blocked(si)) {
1523 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001524 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001525 si_rx_endp_done(si);
1526 } else {
1527 si_rx_endp_more(si);
Christopher Fauleta3279522019-11-20 16:42:00 +01001528 ret = 1;
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001529 }
Christopher Fauleta3279522019-11-20 16:42:00 +01001530 return ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001531}
1532
1533/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001534 * This function propagates a null read received on a socket-based connection.
1535 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001536 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001537 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001538static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001539{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001540 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001541 struct channel *ic = si_ic(si);
1542 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001543
Willy Tarreauabb5d422018-11-14 16:58:52 +01001544 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001545 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001546 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001547 ic->flags |= CF_SHUTR;
1548 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001549
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001550 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001551 return;
1552
Willy Tarreauafc8a222014-11-28 15:46:27 +01001553 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001554 goto do_close;
1555
1556 if (si->flags & SI_FL_NOHALF) {
1557 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001558 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001559 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001560 goto do_close;
1561 }
1562
1563 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001564 return;
1565
1566 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001567 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001568 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001569
Willy Tarreauafc8a222014-11-28 15:46:27 +01001570 oc->flags &= ~CF_SHUTW_NOW;
1571 oc->flags |= CF_SHUTW;
1572 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001573
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001574 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001575
Christopher Faulet27ebcef2019-10-25 10:21:01 +02001576 si->state = SI_ST_DIS;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001577 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001578 return;
1579}
1580
Willy Tarreau651e1822015-09-23 20:06:13 +02001581/* Callback to be used by applet handlers upon completion. It updates the stream
1582 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001583 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001584 * states.
1585 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001586void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001587{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001588 struct channel *ic = si_ic(si);
1589
1590 /* If the applet wants to write and the channel is closed, it's a
1591 * broken pipe and it must be reported.
1592 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001593 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001594 si->flags |= SI_FL_ERR;
1595
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001596 /* automatically mark the applet having data available if it reported
1597 * begin blocked by the channel.
1598 */
1599 if (si_rx_blocked(si))
1600 si_rx_endp_more(si);
1601
Willy Tarreau651e1822015-09-23 20:06:13 +02001602 /* update the stream-int, channels, and possibly wake the stream up */
1603 stream_int_notify(si);
Willy Tarreaue5f86492015-04-19 15:16:35 +02001604
Willy Tarreau32742fd2018-11-14 14:07:59 +01001605 /* stream_int_notify may have passed through chk_snd and released some
1606 * RXBLK flags. Process_stream will consider those flags to wake up the
1607 * appctx but in the case the task is not in runqueue we may have to
1608 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001609 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001610 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1611 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001612 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001613}
1614
1615/*
1616 * This function performs a shutdown-read on a stream interface attached to an
1617 * applet in a connected or init state (it does nothing for other states). It
1618 * either shuts the read side or marks itself as closed. The buffer flags are
1619 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1620 * we also forward the close to the write side. The owner task is woken up if
1621 * it exists.
1622 */
1623static void stream_int_shutr_applet(struct stream_interface *si)
1624{
1625 struct channel *ic = si_ic(si);
1626
Willy Tarreauabb5d422018-11-14 16:58:52 +01001627 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001628 if (ic->flags & CF_SHUTR)
1629 return;
1630 ic->flags |= CF_SHUTR;
1631 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001632
Willy Tarreau828824a2015-04-19 17:20:03 +02001633 /* Note: on shutr, we don't call the applet */
1634
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001635 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001636 return;
1637
1638 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001639 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001640 si->state = SI_ST_DIS;
1641 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001642 }
1643 else if (si->flags & SI_FL_NOHALF) {
1644 /* we want to immediately forward this close to the write side */
1645 return stream_int_shutw_applet(si);
1646 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001647}
1648
1649/*
1650 * This function performs a shutdown-write on a stream interface attached to an
1651 * applet in a connected or init state (it does nothing for other states). It
1652 * either shuts the write side or marks itself as closed. The buffer flags are
1653 * updated to reflect the new state. It does also close everything if the SI
1654 * was marked as being in error state. The owner task is woken up if it exists.
1655 */
1656static void stream_int_shutw_applet(struct stream_interface *si)
1657{
1658 struct channel *ic = si_ic(si);
1659 struct channel *oc = si_oc(si);
1660
1661 oc->flags &= ~CF_SHUTW_NOW;
1662 if (oc->flags & CF_SHUTW)
1663 return;
1664 oc->flags |= CF_SHUTW;
1665 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001666 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001667
Hongbo Longe39683c2017-03-10 18:41:51 +01001668 if (tick_isset(si->hcto)) {
1669 ic->rto = si->hcto;
1670 ic->rex = tick_add(now_ms, ic->rto);
1671 }
1672
Willy Tarreau828824a2015-04-19 17:20:03 +02001673 /* on shutw we always wake the applet up */
1674 appctx_wakeup(si_appctx(si));
1675
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001676 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001677 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001678 case SI_ST_EST:
1679 /* we have to shut before closing, otherwise some short messages
1680 * may never leave the system, especially when there are remaining
1681 * unread data in the socket input buffer, or when nolinger is set.
1682 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1683 * no risk so we close both sides immediately.
1684 */
1685 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1686 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1687 return;
1688
1689 /* fall through */
1690 case SI_ST_CON:
1691 case SI_ST_CER:
1692 case SI_ST_QUE:
1693 case SI_ST_TAR:
1694 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001695 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001696 si->state = SI_ST_DIS;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001697 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001698 si->flags &= ~SI_FL_NOLINGER;
1699 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001700 ic->flags |= CF_SHUTR;
1701 ic->rex = TICK_ETERNITY;
1702 si->exp = TICK_ETERNITY;
1703 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001704}
1705
1706/* chk_rcv function for applets */
1707static void stream_int_chk_rcv_applet(struct stream_interface *si)
1708{
1709 struct channel *ic = si_ic(si);
1710
1711 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1712 __FUNCTION__,
1713 si, si->state, ic->flags, si_oc(si)->flags);
1714
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001715 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001716 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001717 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001718 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001719}
1720
1721/* chk_snd function for applets */
1722static void stream_int_chk_snd_applet(struct stream_interface *si)
1723{
1724 struct channel *oc = si_oc(si);
1725
1726 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1727 __FUNCTION__,
1728 si, si->state, si_ic(si)->flags, oc->flags);
1729
1730 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1731 return;
1732
Willy Tarreau828824a2015-04-19 17:20:03 +02001733 /* we only wake the applet up if it was waiting for some data */
1734
1735 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001736 return;
1737
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001738 if (!tick_isset(oc->wex))
1739 oc->wex = tick_add_ifset(now_ms, oc->wto);
1740
Willy Tarreau828824a2015-04-19 17:20:03 +02001741 if (!channel_is_empty(oc)) {
1742 /* (re)start sending */
1743 appctx_wakeup(si_appctx(si));
1744 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001745}
1746
Willy Tarreaudded32d2008-11-30 19:48:07 +01001747/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001748 * Local variables:
1749 * c-indent-level: 8
1750 * c-basic-offset: 8
1751 * End:
1752 */