blob: a6be2c9a0ac2cb9f6891dc7d643ccca13129b35c [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 Tarreau4c7e4b72020-05-27 12:58:42 +020022#include <haproxy/api.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020023#include <haproxy/applet.h>
Willy Tarreauf1d32c42020-06-04 21:07:02 +020024#include <haproxy/channel.h>
Willy Tarreau7ea393d2020-06-04 18:02:10 +020025#include <haproxy/connection.h>
Willy Tarreau2741c8c2020-06-02 11:28:02 +020026#include <haproxy/dynbuf.h>
Willy Tarreau87735332020-06-04 09:08:41 +020027#include <haproxy/http_htx.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020028#include <haproxy/pipe-t.h>
29#include <haproxy/pipe.h>
Willy Tarreaua264d962020-06-04 22:29:18 +020030#include <haproxy/proxy.h>
Willy Tarreaudfd3de82020-06-04 23:46:14 +020031#include <haproxy/stream-t.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020032#include <haproxy/stream_interface.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020033#include <haproxy/task.h>
Willy Tarreauc2f7c582020-06-02 18:15:32 +020034#include <haproxy/ticks.h>
Willy Tarreau92b4f132020-06-01 11:05:15 +020035#include <haproxy/time.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020036#include <haproxy/tools.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010037
Willy Tarreaufd31e532012-07-23 18:24:25 +020038
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010039/* functions used by default on a detached stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020040static void stream_int_shutr(struct stream_interface *si);
41static void stream_int_shutw(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020042static void stream_int_chk_rcv(struct stream_interface *si);
43static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010044
45/* functions used on a conn_stream-based stream-interface */
Willy Tarreau6fe15412013-09-29 15:16:03 +020046static void stream_int_shutr_conn(struct stream_interface *si);
47static void stream_int_shutw_conn(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020048static void stream_int_chk_rcv_conn(struct stream_interface *si);
49static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010050
51/* functions used on an applet-based stream-interface */
Willy Tarreaud45b9f82015-04-13 16:30:14 +020052static void stream_int_shutr_applet(struct stream_interface *si);
53static void stream_int_shutw_applet(struct stream_interface *si);
54static void stream_int_chk_rcv_applet(struct stream_interface *si);
55static void stream_int_chk_snd_applet(struct stream_interface *si);
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010056
57/* last read notification */
58static void stream_int_read0(struct stream_interface *si);
59
60/* post-IO notification callback */
61static void stream_int_notify(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020062
Willy Tarreauc5788912012-08-24 18:12:41 +020063/* stream-interface operations for embedded tasks */
64struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020065 .chk_rcv = stream_int_chk_rcv,
66 .chk_snd = stream_int_chk_snd,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020067 .shutr = stream_int_shutr,
68 .shutw = stream_int_shutw,
Willy Tarreau5c979a92012-05-07 17:15:39 +020069};
70
Willy Tarreauc5788912012-08-24 18:12:41 +020071/* stream-interface operations for connections */
72struct si_ops si_conn_ops = {
Willy Tarreauc5788912012-08-24 18:12:41 +020073 .chk_rcv = stream_int_chk_rcv_conn,
74 .chk_snd = stream_int_chk_snd_conn,
Willy Tarreau8b3d7df2013-09-29 14:51:58 +020075 .shutr = stream_int_shutr_conn,
76 .shutw = stream_int_shutw_conn,
Willy Tarreauc5788912012-08-24 18:12:41 +020077};
78
Willy Tarreaud45b9f82015-04-13 16:30:14 +020079/* stream-interface operations for connections */
80struct si_ops si_applet_ops = {
Willy Tarreaud45b9f82015-04-13 16:30:14 +020081 .chk_rcv = stream_int_chk_rcv_applet,
82 .chk_snd = stream_int_chk_snd_applet,
83 .shutr = stream_int_shutr_applet,
84 .shutw = stream_int_shutw_applet,
85};
86
Willy Tarreau14bfe9a2018-12-19 15:19:27 +010087
88/* Functions used to communicate with a conn_stream. The first two may be used
89 * directly, the last one is mostly a wake callback.
90 */
91int si_cs_recv(struct conn_stream *cs);
92int si_cs_send(struct conn_stream *cs);
93static int si_cs_process(struct conn_stream *cs);
94
95
Willy Tarreau74beec32012-10-03 00:41:04 +020096struct data_cb si_conn_cb = {
Olivier Houchard21df6cc2018-09-14 23:21:44 +020097 .wake = si_cs_process,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +010098 .name = "STRM",
Willy Tarreauc5788912012-08-24 18:12:41 +020099};
100
Willy Tarreaucff64112008-11-03 06:26:53 +0100101/*
102 * This function only has to be called once after a wakeup event in case of
103 * suspected timeout. It controls the stream interface timeouts and sets
104 * si->flags accordingly. It does NOT close anything, as this timeout may
105 * be used for any purpose. It returns 1 if the timeout fired, otherwise
106 * zero.
107 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100108int si_check_timeouts(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100109{
110 if (tick_is_expired(si->exp, now_ms)) {
111 si->flags |= SI_FL_EXP;
112 return 1;
113 }
114 return 0;
115}
116
Willy Tarreaufe3718a2008-11-30 18:14:12 +0100117/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100118void si_report_error(struct stream_interface *si)
Willy Tarreaucff64112008-11-03 06:26:53 +0100119{
120 if (!si->err_type)
121 si->err_type = SI_ET_DATA_ERR;
122
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100123 si_oc(si)->flags |= CF_WRITE_ERROR;
124 si_ic(si)->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +0100125}
126
127/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100128 * Returns a message to the client ; the connection is shut down for read,
129 * and the request is cleared so that no server connection can be initiated.
130 * The buffer is marked for read shutdown on the other side to protect the
131 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100132 * "chunk". If it is null, then an empty message is used. The reply buffer does
133 * not need to be empty before this, and its contents will not be overwritten.
134 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100135 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100136void si_retnclose(struct stream_interface *si,
Willy Tarreau83061a82018-07-13 11:56:34 +0200137 const struct buffer *msg)
Willy Tarreaudded32d2008-11-30 19:48:07 +0100138{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100139 struct channel *ic = si_ic(si);
140 struct channel *oc = si_oc(si);
141
142 channel_auto_read(ic);
143 channel_abort(ic);
144 channel_auto_close(ic);
145 channel_erase(ic);
146 channel_truncate(oc);
Willy Tarreau798e1282010-12-12 13:06:00 +0100147
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200148 if (likely(msg && msg->data))
149 co_inject(oc, msg->area, msg->data);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100150
Willy Tarreauafc8a222014-11-28 15:46:27 +0100151 oc->wex = tick_add_ifset(now_ms, oc->wto);
152 channel_auto_read(oc);
153 channel_auto_close(oc);
154 channel_shutr_now(oc);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100155}
156
Willy Tarreau4a36b562012-08-06 19:31:45 +0200157/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200158 * This function performs a shutdown-read on a detached stream interface in a
159 * connected or init state (it does nothing for other states). It either shuts
160 * the read side or marks itself as closed. The buffer flags are updated to
161 * reflect the new state. If the stream interface has SI_FL_NOHALF, we also
162 * forward the close to the write side. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200163 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200164static void stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200165{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100166 struct channel *ic = si_ic(si);
167
Willy Tarreauabb5d422018-11-14 16:58:52 +0100168 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100169 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200170 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100171 ic->flags |= CF_SHUTR;
172 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200173
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200174 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200175 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200176
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100177 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200178 si->state = SI_ST_DIS;
179 si->exp = TICK_ETERNITY;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200180 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200181 else if (si->flags & SI_FL_NOHALF) {
182 /* we want to immediately forward this close to the write side */
183 return stream_int_shutw(si);
184 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200185
Willy Tarreau4a36b562012-08-06 19:31:45 +0200186 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100187 if (!(si->flags & SI_FL_DONT_WAKE))
188 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200189}
190
Willy Tarreau4a36b562012-08-06 19:31:45 +0200191/*
Willy Tarreaud45b9f82015-04-13 16:30:14 +0200192 * This function performs a shutdown-write on a detached stream interface in a
193 * connected or init state (it does nothing for other states). It either shuts
194 * the write side or marks itself as closed. The buffer flags are updated to
195 * reflect the new state. It does also close everything if the SI was marked as
196 * being in error state. The owner task is woken up if it exists.
Willy Tarreau4a36b562012-08-06 19:31:45 +0200197 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200198static void stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200199{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100200 struct channel *ic = si_ic(si);
201 struct channel *oc = si_oc(si);
202
203 oc->flags &= ~CF_SHUTW_NOW;
204 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200205 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100206 oc->flags |= CF_SHUTW;
207 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +0100208 si_done_get(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200209
Hongbo Longe39683c2017-03-10 18:41:51 +0100210 if (tick_isset(si->hcto)) {
211 ic->rto = si->hcto;
212 ic->rex = tick_add(now_ms, ic->rto);
213 }
214
Willy Tarreaufb90d942009-09-05 20:57:35 +0200215 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200216 case SI_ST_RDY:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200217 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200218 /* we have to shut before closing, otherwise some short messages
219 * may never leave the system, especially when there are remaining
220 * unread data in the socket input buffer, or when nolinger is set.
221 * However, if SI_FL_NOLINGER is explicitly set, we know there is
222 * no risk so we close both sides immediately.
223 */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200224 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
Willy Tarreauafc8a222014-11-28 15:46:27 +0100225 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200226 return;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200227
228 /* fall through */
229 case SI_ST_CON:
230 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100231 case SI_ST_QUE:
232 case SI_ST_TAR:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200233 /* Note that none of these states may happen with applets */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200234 si->state = SI_ST_DIS;
Tim Duesterhus588b3142020-05-29 14:35:51 +0200235 /* fall through */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200236 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +0100237 si->flags &= ~SI_FL_NOLINGER;
238 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100239 ic->flags |= CF_SHUTR;
240 ic->rex = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200241 si->exp = TICK_ETERNITY;
242 }
243
Willy Tarreau4a36b562012-08-06 19:31:45 +0200244 /* note that if the task exists, it must unregister itself once it runs */
Willy Tarreau07373b82014-11-28 12:08:47 +0100245 if (!(si->flags & SI_FL_DONT_WAKE))
246 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200247}
248
249/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200250static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200251{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100252 struct channel *ic = si_ic(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200253
Willy Tarreauafc8a222014-11-28 15:46:27 +0100254 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200255 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100256 si, si->state, ic->flags, si_oc(si)->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200257
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200258 if (ic->pipe) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200259 /* stop reading */
Willy Tarreaudb398432018-11-15 11:08:52 +0100260 si_rx_room_blk(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200261 }
262 else {
263 /* (re)start reading */
Willy Tarreau3c39a7d2019-06-14 14:42:29 +0200264 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau07373b82014-11-28 12:08:47 +0100265 if (!(si->flags & SI_FL_DONT_WAKE))
266 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200267 }
268}
269
270/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200271static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200272{
Willy Tarreauafc8a222014-11-28 15:46:27 +0100273 struct channel *oc = si_oc(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200274
Willy Tarreauafc8a222014-11-28 15:46:27 +0100275 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
Willy Tarreaufb90d942009-09-05 20:57:35 +0200276 __FUNCTION__,
Willy Tarreauafc8a222014-11-28 15:46:27 +0100277 si, si->state, si_ic(si)->flags, oc->flags);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200278
Willy Tarreauafc8a222014-11-28 15:46:27 +0100279 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200280 return;
281
282 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100283 channel_is_empty(oc)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200284 return;
285
286 /* Otherwise there are remaining data to be sent in the buffer,
287 * so we tell the handler.
288 */
289 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100290 if (!tick_isset(oc->wex))
291 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200292
Willy Tarreau07373b82014-11-28 12:08:47 +0100293 if (!(si->flags & SI_FL_DONT_WAKE))
294 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200295}
296
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200297/* Register an applet to handle a stream_interface as a new appctx. The SI will
298 * wake it up everytime it is solicited. The appctx must be deleted by the task
299 * handler using si_release_endpoint(), possibly from within the function itself.
300 * It also pre-initializes the applet's context and returns it (or NULL in case
301 * it could not be allocated).
Willy Tarreaufb90d942009-09-05 20:57:35 +0200302 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100303struct appctx *si_register_handler(struct stream_interface *si, struct applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200304{
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100305 struct appctx *appctx;
306
Willy Tarreau07373b82014-11-28 12:08:47 +0100307 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si_task(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200308
Willy Tarreaua7513f52015-04-05 00:15:26 +0200309 appctx = si_alloc_appctx(si, app);
Willy Tarreaua69fc9f2014-12-22 19:34:00 +0100310 if (!appctx)
Willy Tarreau0a23bcb2013-12-01 11:31:38 +0100311 return NULL;
312
Willy Tarreau0cd3bd62018-11-06 18:46:37 +0100313 si_cant_get(si);
Willy Tarreau828824a2015-04-19 17:20:03 +0200314 appctx_wakeup(appctx);
Willy Tarreau1fbe1c92013-12-01 09:35:41 +0100315 return si_appctx(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200316}
317
Willy Tarreau2c6be842012-07-06 17:12:34 +0200318/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200319 * established. It returns 0 if it fails in a fatal way or needs to poll to go
320 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200321 * flags (the bit is provided in <flag> by the caller). It is designed to be
322 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200323 * Note that it can emit a PROXY line by relying on the other end's address
324 * when the connection is attached to a stream interface, or by resolving the
325 * local address otherwise (also called a LOCAL line).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200326 */
327int conn_si_send_proxy(struct connection *conn, unsigned int flag)
328{
Willy Tarreaud02cdd22013-12-15 10:23:20 +0100329 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200330 goto out_error;
331
Willy Tarreau2c6be842012-07-06 17:12:34 +0200332 /* If we have a PROXY line to send, we'll use this to validate the
333 * connection, in which case the connection is validated only once
334 * we've sent the whole proxy line. Otherwise we use connect().
335 */
Tim Duesterhus36839dc2019-02-26 17:09:51 +0100336 if (conn->send_proxy_ofs) {
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100337 const struct conn_stream *cs;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200338 int ret;
339
Christopher Fauletd82056c2020-05-26 16:08:49 +0200340 /* If there is no mux attached to the connection, it means the
341 * connection context is a conn-stream.
342 */
343 cs = (conn->mux ? cs_get_first(conn) : conn->ctx);
344
Willy Tarreau2c6be842012-07-06 17:12:34 +0200345 /* The target server expects a PROXY line to be sent first.
346 * If the send_proxy_ofs is negative, it corresponds to the
347 * offset to start sending from then end of the proxy string
348 * (which is recomputed every time since it's constant). If
349 * it is positive, it means we have to send from the start.
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200350 * We can only send a "normal" PROXY line when the connection
351 * is attached to a stream interface. Otherwise we can only
352 * send a LOCAL line (eg: for use with health checks).
Willy Tarreau2c6be842012-07-06 17:12:34 +0200353 */
Willy Tarreau6b1379f2018-11-18 21:38:19 +0100354
355 if (cs && cs->data_cb == &si_conn_cb) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200356 struct stream_interface *si = cs->data;
357 struct conn_stream *remote_cs = objt_cs(si_opposite(si)->end);
Tim Duesterhuscf6e0c82020-03-13 12:34:24 +0100358 struct stream *strm = si_strm(si);
359
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200360 ret = make_proxy_line(trash.area, trash.size,
361 objt_server(conn->target),
Tim Duesterhuscf6e0c82020-03-13 12:34:24 +0100362 remote_cs ? remote_cs->conn : NULL,
363 strm);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200364 }
365 else {
366 /* The target server expects a LOCAL line to be sent first. Retrieving
367 * local or remote addresses may fail until the connection is established.
368 */
Willy Tarreau7bb447c2019-07-17 11:40:51 +0200369 if (!conn_get_src(conn) || !conn_get_dst(conn))
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200370 goto out_wait;
371
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200372 ret = make_proxy_line(trash.area, trash.size,
Tim Duesterhuscf6e0c82020-03-13 12:34:24 +0100373 objt_server(conn->target), conn,
374 NULL);
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200375 }
Willy Tarreaub363a1f2013-10-01 10:45:07 +0200376
Willy Tarreau2c6be842012-07-06 17:12:34 +0200377 if (!ret)
378 goto out_error;
379
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200380 if (conn->send_proxy_ofs > 0)
381 conn->send_proxy_ofs = -ret; /* first call */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200382
Willy Tarreaua1a74742012-08-24 12:14:49 +0200383 /* we have to send trash from (ret+sp for -sp bytes). If the
384 * data layer has a pending write, we'll also set MSG_MORE.
385 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200386 ret = conn_sock_send(conn,
387 trash.area + ret + conn->send_proxy_ofs,
388 -conn->send_proxy_ofs,
Willy Tarreau19bc2012020-02-21 08:46:19 +0100389 (conn->subs && conn->subs->events & SUB_RETRY_SEND) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200390
Willy Tarreau0a03c0f2015-03-13 00:05:28 +0100391 if (ret < 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200392 goto out_error;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200393
Willy Tarreaub8020ce2013-10-24 21:10:08 +0200394 conn->send_proxy_ofs += ret; /* becomes zero once complete */
395 if (conn->send_proxy_ofs != 0)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200396 goto out_wait;
397
398 /* OK we've sent the whole line, we're connected */
399 }
400
Willy Tarreaua1a74742012-08-24 12:14:49 +0200401 /* The connection is ready now, simply return and let the connection
402 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200403 */
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100404 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200405 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200406 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200407
408 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200409 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200410 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200411 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200412
413 out_wait:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200414 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200415}
416
Willy Tarreau27375622013-12-17 00:00:28 +0100417
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100418/* This function is the equivalent to si_update() except that it's
Willy Tarreau615f28b2015-09-23 18:40:09 +0200419 * designed to be called from outside the stream handlers, typically the lower
420 * layers (applets, connections) after I/O completion. After updating the stream
421 * interface and timeouts, it will try to forward what can be forwarded, then to
422 * wake the associated task up if an important event requires special handling.
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100423 * It may update SI_FL_WAIT_DATA and/or SI_FL_RXBLK_ROOM, that the callers are
Willy Tarreau0dfccb22018-10-25 13:55:20 +0200424 * encouraged to watch to take appropriate action.
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100425 * It should not be called from within the stream itself, si_update()
Willy Tarreau615f28b2015-09-23 18:40:09 +0200426 * is designed for this.
427 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100428static void stream_int_notify(struct stream_interface *si)
Willy Tarreau615f28b2015-09-23 18:40:09 +0200429{
430 struct channel *ic = si_ic(si);
431 struct channel *oc = si_oc(si);
Willy Tarreau47baeb82018-11-15 07:46:57 +0100432 struct stream_interface *sio = si_opposite(si);
Christopher Fauletd7607de2019-01-03 16:24:54 +0100433 struct task *task = si_task(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200434
435 /* process consumer side */
436 if (channel_is_empty(oc)) {
Olivier Houcharde9bed532017-11-16 17:49:25 +0100437 struct connection *conn = objt_cs(si->end) ? objt_cs(si->end)->conn : NULL;
438
Willy Tarreau615f28b2015-09-23 18:40:09 +0200439 if (((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +0100440 (si->state == SI_ST_EST) && (!conn || !(conn->flags & (CO_FL_WAIT_XPRT | CO_FL_EARLY_SSL_HS))))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200441 si_shutw(si);
442 oc->wex = TICK_ETERNITY;
443 }
444
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100445 /* indicate that we may be waiting for data from the output channel or
446 * we're about to close and can't expect more data if SHUTW_NOW is there.
447 */
Christopher Fauletb3e0de42018-10-11 13:54:13 +0200448 if (!(oc->flags & (CF_SHUTW|CF_SHUTW_NOW)))
Willy Tarreau615f28b2015-09-23 18:40:09 +0200449 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreau8cf9c8e2016-12-13 15:21:25 +0100450 else if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW)
451 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200452
453 /* update OC timeouts and wake the other side up if it's waiting for room */
454 if (oc->flags & CF_WRITE_ACTIVITY) {
455 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
456 !channel_is_empty(oc))
457 if (tick_isset(oc->wex))
458 oc->wex = tick_add_ifset(now_ms, oc->wto);
459
460 if (!(si->flags & SI_FL_INDEP_STR))
461 if (tick_isset(ic->rex))
462 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreauf26c26c2018-11-12 16:11:08 +0100463 }
Willy Tarreau615f28b2015-09-23 18:40:09 +0200464
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100465 if (oc->flags & CF_DONT_READ)
466 si_rx_chan_blk(sio);
467 else
468 si_rx_chan_rdy(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200469
470 /* Notify the other side when we've injected data into the IC that
471 * needs to be forwarded. We can do fast-forwarding as soon as there
472 * are output data, but we avoid doing this if some of the data are
473 * not yet scheduled for being forwarded, because it is very likely
474 * that it will be done again immediately afterwards once the following
Willy Tarreaud0f5bbc2018-11-14 11:10:26 +0100475 * data are parsed (eg: HTTP chunking). We only SI_FL_RXBLK_ROOM once
Willy Tarreau615f28b2015-09-23 18:40:09 +0200476 * we've emptied *some* of the output buffer, and not just when there
477 * is available room, because applets are often forced to stop before
478 * the buffer is full. We must not stop based on input data alone because
479 * an HTTP parser might need more data to complete the parsing.
480 */
481 if (!channel_is_empty(ic) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100482 (sio->flags & SI_FL_WAIT_DATA) &&
Willy Tarreau89b6a2b2018-11-18 15:46:10 +0100483 (!(ic->flags & CF_EXPECT_MORE) || c_full(ic) || ci_data(ic) == 0 || ic->pipe)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200484 int new_len, last_len;
485
Willy Tarreau77e478c2018-06-19 07:03:14 +0200486 last_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200487 if (ic->pipe)
488 last_len += ic->pipe->data;
489
Willy Tarreau47baeb82018-11-15 07:46:57 +0100490 si_chk_snd(sio);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200491
Willy Tarreau77e478c2018-06-19 07:03:14 +0200492 new_len = co_data(ic);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200493 if (ic->pipe)
494 new_len += ic->pipe->data;
495
496 /* check if the consumer has freed some space either in the
497 * buffer or in the pipe.
498 */
Willy Tarreau47baeb82018-11-15 07:46:57 +0100499 if (new_len < last_len)
Willy Tarreaudb398432018-11-15 11:08:52 +0100500 si_rx_room_rdy(si);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200501 }
502
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100503 if (!(ic->flags & CF_DONT_READ))
504 si_rx_chan_rdy(si);
505
Willy Tarreau47baeb82018-11-15 07:46:57 +0100506 si_chk_rcv(si);
507 si_chk_rcv(sio);
508
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100509 if (si_rx_blocked(si)) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200510 ic->rex = TICK_ETERNITY;
511 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100512 else if ((ic->flags & (CF_SHUTR|CF_READ_PARTIAL)) == CF_READ_PARTIAL) {
Willy Tarreau615f28b2015-09-23 18:40:09 +0200513 /* we must re-enable reading if si_chk_snd() has freed some space */
514 if (!(ic->flags & CF_READ_NOEXP) && tick_isset(ic->rex))
515 ic->rex = tick_add_ifset(now_ms, ic->rto);
516 }
517
518 /* wake the task up only when needed */
519 if (/* changes on the production side */
520 (ic->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200521 !si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200522 (si->flags & SI_FL_ERR) ||
523 ((ic->flags & CF_READ_PARTIAL) &&
Christopher Faulet297d3e22019-03-22 14:16:14 +0100524 ((ic->flags & CF_EOI) || !ic->to_forward || sio->state != SI_ST_EST)) ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200525
526 /* changes on the consumption side */
527 (oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
Willy Tarreauede3d882018-10-24 17:17:56 +0200528 ((oc->flags & CF_WRITE_ACTIVITY) &&
Willy Tarreau615f28b2015-09-23 18:40:09 +0200529 ((oc->flags & CF_SHUTW) ||
Willy Tarreau78f5ff82018-12-19 11:00:00 +0100530 (((oc->flags & CF_WAKE_WRITE) ||
531 !(oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW|CF_SHUTW))) &&
Willy Tarreau47baeb82018-11-15 07:46:57 +0100532 (sio->state != SI_ST_EST ||
Willy Tarreau615f28b2015-09-23 18:40:09 +0200533 (channel_is_empty(oc) && !oc->to_forward)))))) {
Christopher Fauletd7607de2019-01-03 16:24:54 +0100534 task_wakeup(task, TASK_WOKEN_IO);
535 }
536 else {
537 /* Update expiration date for the task and requeue it */
538 task->expire = tick_first((tick_is_expired(task->expire, now_ms) ? 0 : task->expire),
539 tick_first(tick_first(ic->rex, ic->wex),
540 tick_first(oc->rex, oc->wex)));
Willy Tarreau45bcb372019-08-01 18:51:38 +0200541
542 task->expire = tick_first(task->expire, ic->analyse_exp);
543 task->expire = tick_first(task->expire, oc->analyse_exp);
544
545 if (si->exp)
546 task->expire = tick_first(task->expire, si->exp);
547
548 if (sio->exp)
549 task->expire = tick_first(task->expire, sio->exp);
550
Christopher Fauletd7607de2019-01-03 16:24:54 +0100551 task_queue(task);
Willy Tarreau615f28b2015-09-23 18:40:09 +0200552 }
553 if (ic->flags & CF_READ_ACTIVITY)
554 ic->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau615f28b2015-09-23 18:40:09 +0200555}
556
557
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200558/* Called by I/O handlers after completion.. It propagates
Willy Tarreau651e1822015-09-23 20:06:13 +0200559 * connection flags to the stream interface, updates the stream (which may or
560 * may not take this opportunity to try to forward data), then update the
561 * connection's polling based on the channels and stream interface's final
562 * states. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200563 */
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200564static int si_cs_process(struct conn_stream *cs)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200565{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200566 struct connection *conn = cs->conn;
567 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100568 struct channel *ic = si_ic(si);
569 struct channel *oc = si_oc(si);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200570
Olivier Houchardc7ffa912018-08-28 19:37:41 +0200571 /* If we have data to send, try it now */
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100572 if (!channel_is_empty(oc) && !(si->wait_event.events & SUB_RETRY_SEND))
Willy Tarreau908d26f2018-10-25 14:02:47 +0200573 si_cs_send(cs);
574
Willy Tarreau651e1822015-09-23 20:06:13 +0200575 /* First step, report to the stream-int what was detected at the
576 * connection layer : errors and connection establishment.
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200577 * Only add SI_FL_ERR if we're connected, or we're attempting to
578 * connect, we may get there because we got woken up, but only run
579 * after process_stream() noticed there were an error, and decided
580 * to retry to connect, the connection may still have CO_FL_ERROR,
581 * and we don't want to add SI_FL_ERR back
Christopher Faulet36b536d2019-11-20 11:56:33 +0100582 *
583 * Note: This test is only required because si_cs_process is also the SI
584 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
585 * care of it.
Willy Tarreau651e1822015-09-23 20:06:13 +0200586 */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200587 if (si->state >= SI_ST_CON &&
588 (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR))
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200589 si->flags |= SI_FL_ERR;
590
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200591 /* If we had early data, and the handshake ended, then
592 * we can remove the flag, and attempt to wake the task up,
593 * in the event there's an analyser waiting for the end of
594 * the handshake.
595 */
Willy Tarreau911db9b2020-01-23 16:27:54 +0100596 if (!(conn->flags & (CO_FL_WAIT_XPRT | CO_FL_EARLY_SSL_HS)) &&
Olivier Houchard6fa63d92017-11-27 18:41:32 +0100597 (cs->flags & CS_FL_WAIT_FOR_HS)) {
598 cs->flags &= ~CS_FL_WAIT_FOR_HS;
Olivier Houchardccaa7de2017-10-02 11:51:03 +0200599 task_wakeup(si_task(si), TASK_WOKEN_MSG);
600 }
601
Willy Tarreau7ab22adb2019-06-05 14:53:22 +0200602 if (!si_state_in(si->state, SI_SB_EST|SI_SB_DIS|SI_SB_CLO) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +0100603 (conn->flags & CO_FL_WAIT_XPRT) == 0) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200604 si->exp = TICK_ETERNITY;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100605 oc->flags |= CF_WRITE_NULL;
Willy Tarreaub27f54a2019-06-05 16:43:44 +0200606 if (si->state == SI_ST_CON)
607 si->state = SI_ST_RDY;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200608 }
609
Christopher Faulet297d3e22019-03-22 14:16:14 +0100610 /* Report EOI on the channel if it was reached from the mux point of
Christopher Faulet36b536d2019-11-20 11:56:33 +0100611 * view.
612 *
613 * Note: This test is only required because si_cs_process is also the SI
614 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
615 * care of it.
616 */
Christopher Faulet297d3e22019-03-22 14:16:14 +0100617 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200618 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100619
Willy Tarreau651e1822015-09-23 20:06:13 +0200620 /* Second step : update the stream-int and channels, try to forward any
621 * pending data, then possibly wake the stream up based on the new
622 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200623 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200624 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +0200625 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200626 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200627}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200628
Willy Tarreau5368d802012-08-21 18:22:06 +0200629/*
630 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200631 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800632 * caller to commit polling changes. The caller should check conn->flags
633 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200634 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200635int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200636{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200637 struct connection *conn = cs->conn;
638 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100639 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200640 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200641 int did_send = 0;
642
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100643 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200644 /* We're probably there because the tasklet was woken up,
645 * but process_stream() ran before, detected there were an
646 * error and put the si back to SI_ST_TAR. There's still
647 * CO_FL_ERROR on the connection but we don't want to add
648 * SI_FL_ERR back, so give up
649 */
650 if (si->state < SI_ST_CON)
651 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100652 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200653 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100654 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200655
Christopher Faulet328ed222019-09-23 15:57:29 +0200656 /* We're already waiting to be able to send, give up */
657 if (si->wait_event.events & SUB_RETRY_SEND)
658 return 0;
659
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200660 /* we might have been called just after an asynchronous shutw */
Willy Tarreauf22758d2020-01-23 18:25:23 +0100661 if (oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200662 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200663
Christopher Faulete96993b2020-07-30 09:26:46 +0200664 /* we must wait because the mux is not installed yet */
665 if (!conn->mux)
666 return 0;
667
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200668 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
669 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet86162db2019-07-05 11:49:11 +0200670 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200671 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200672
Willy Tarreauafc8a222014-11-28 15:46:27 +0100673 if (!oc->pipe->data) {
674 put_pipe(oc->pipe);
675 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200676 }
677
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100678 if (oc->pipe)
679 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200680 }
681
682 /* At this point, the pipe is empty, but we may still have data pending
683 * in the normal buffer.
684 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100685 if (co_data(oc)) {
686 /* when we're here, we already know that there is no spliced
687 * data left, and that there are sendable buffered data.
688 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200689
Willy Tarreau5368d802012-08-21 18:22:06 +0200690 /* check if we want to inform the kernel that we're interested in
691 * sending more data after this call. We want this if :
692 * - we're about to close after this last send and want to merge
693 * the ongoing FIN with the last segment.
694 * - we know we can't send everything at once and must get back
695 * here because of unaligned data
696 * - there is still a finite amount of data to forward
697 * The test is arranged so that the most common case does only 2
698 * tests.
699 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100700 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200701
Willy Tarreauafc8a222014-11-28 15:46:27 +0100702 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
703 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau8945bb62020-06-19 17:07:06 +0200704 (oc->flags & CF_EXPECT_MORE) ||
Christopher Faulet9e3dc832020-07-22 16:28:44 +0200705 (IS_HTX_STRM(si_strm(si)) &&
706 (!(oc->flags & (CF_EOI|CF_SHUTR)) && htx_expect_more(htxbuf(&oc->buf)))))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100707 ((oc->flags & CF_ISRESP) &&
708 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100709 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200710
Willy Tarreauafc8a222014-11-28 15:46:27 +0100711 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100712 send_flag |= CO_SFL_STREAMER;
713
Olivier Houcharda254a372019-04-05 15:30:12 +0200714 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
Olivier Houchardce1a0292019-05-17 15:38:29 +0200715 struct stream *s = si_strm(si);
Olivier Houcharda254a372019-04-05 15:30:12 +0200716 /* If we want to be able to do L7 retries, copy
717 * the data we're about to send, so that we are able
718 * to resend them if needed
719 */
720 /* Try to allocate a buffer if we had none.
721 * If it fails, the next test will just
722 * disable the l7 retries by setting
723 * l7_conn_retries to 0.
724 */
Olivier Houchardce1a0292019-05-17 15:38:29 +0200725 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
Olivier Houcharda254a372019-04-05 15:30:12 +0200726 si->flags &= ~SI_FL_L7_RETRY;
727 else {
728 if (b_is_null(&si->l7_buffer))
729 b_alloc(&si->l7_buffer);
730 if (b_is_null(&si->l7_buffer))
731 si->flags &= ~SI_FL_L7_RETRY;
732 else {
733 memcpy(b_orig(&si->l7_buffer),
734 b_orig(&oc->buf),
735 b_size(&oc->buf));
736 si->l7_buffer.head = co_data(oc);
737 b_add(&si->l7_buffer, co_data(oc));
738 }
739
740 }
741 }
742
Olivier Houcharded0f2072018-08-16 15:41:52 +0200743 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800744 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200745 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200746 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200747 c_realign_if_empty(oc);
748
749 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800750 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100751 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800752 }
Godbache68e02d2013-10-11 15:48:29 +0800753 /* if some data remain in the buffer, it's only because the
754 * system buffers are full, we will try next time.
755 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200756 }
Godbache68e02d2013-10-11 15:48:29 +0800757 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100758
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100759 end:
Christopher Faulet86162db2019-07-05 11:49:11 +0200760 if (did_send) {
761 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
762 if (si->state == SI_ST_CON)
763 si->state = SI_ST_RDY;
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200764
765 si_rx_room_rdy(si_opposite(si));
Christopher Faulet86162db2019-07-05 11:49:11 +0200766 }
767
768 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
769 si->flags |= SI_FL_ERR;
770 return 1;
771 }
772
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200773 /* 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 +0100774 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100775 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200776 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200777}
778
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100779/* This is the ->process() function for any stream-interface's wait_event task.
780 * It's assigned during the stream-interface's initialization, for any type of
781 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
782 * stream interface, as the presence of the CS is checked there.
783 */
Olivier Houchard91894cb2018-08-02 18:06:28 +0200784struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned short state)
785{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200786 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200787 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200788 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200789
790 if (!cs)
791 return NULL;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100792
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100793 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200794 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100795 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200796 ret |= si_cs_recv(cs);
797 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200798 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200799
Willy Tarreaua64c7032019-08-01 14:17:02 +0200800 stream_release_buffers(si_strm(si));
Olivier Houchard91894cb2018-08-02 18:06:28 +0200801 return (NULL);
802}
803
Willy Tarreau25f13102015-09-24 11:32:22 +0200804/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200805 * update the input channel's expiration timer and the stream interface's
806 * Rx flags based on the channel's flags. It needs to be called only once
807 * after the channel's flags have settled down, and before they are cleared,
808 * though it doesn't harm to call it as often as desired (it just slightly
809 * hurts performance). It must not be called from outside of the stream
810 * handler, as what it does will be used to compute the stream task's
811 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200812 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200813void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200814{
815 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200816
Willy Tarreau236c4292019-06-06 08:19:20 +0200817 if (ic->flags & CF_SHUTR) {
818 si_rx_shut_blk(si);
819 return;
820 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100821
Willy Tarreau236c4292019-06-06 08:19:20 +0200822 /* Read not closed, update FD status and timeout for reads */
823 if (ic->flags & CF_DONT_READ)
824 si_rx_chan_blk(si);
825 else
826 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100827
Willy Tarreau236c4292019-06-06 08:19:20 +0200828 if (!channel_is_empty(ic)) {
829 /* stop reading, imposed by channel's policy or contents */
830 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200831 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200832 else {
833 /* (re)start reading and update timeout. Note: we don't recompute the timeout
834 * everytime we get here, otherwise it would risk never to expire. We only
835 * update it if is was not yet set. The stream socket handler will already
836 * have updated it if there has been a completed I/O.
837 */
838 si_rx_room_rdy(si);
839 }
840 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
841 ic->rex = TICK_ETERNITY;
842 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
843 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200844
Willy Tarreau236c4292019-06-06 08:19:20 +0200845 si_chk_rcv(si);
846}
847
848/* This function is designed to be called from within the stream handler to
849 * update the output channel's expiration timer and the stream interface's
850 * Tx flags based on the channel's flags. It needs to be called only once
851 * after the channel's flags have settled down, and before they are cleared,
852 * though it doesn't harm to call it as often as desired (it just slightly
853 * hurts performance). It must not be called from outside of the stream
854 * handler, as what it does will be used to compute the stream task's
855 * expiration.
856 */
857void si_update_tx(struct stream_interface *si)
858{
859 struct channel *oc = si_oc(si);
860 struct channel *ic = si_ic(si);
861
862 if (oc->flags & CF_SHUTW)
863 return;
864
865 /* Write not closed, update FD status and timeout for writes */
866 if (channel_is_empty(oc)) {
867 /* stop writing */
868 if (!(si->flags & SI_FL_WAIT_DATA)) {
869 if ((oc->flags & CF_SHUTW_NOW) == 0)
870 si->flags |= SI_FL_WAIT_DATA;
871 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200872 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200873 return;
874 }
875
876 /* (re)start writing and update timeout. Note: we don't recompute the timeout
877 * everytime we get here, otherwise it would risk never to expire. We only
878 * update it if is was not yet set. The stream socket handler will already
879 * have updated it if there has been a completed I/O.
880 */
881 si->flags &= ~SI_FL_WAIT_DATA;
882 if (!tick_isset(oc->wex)) {
883 oc->wex = tick_add_ifset(now_ms, oc->wto);
884 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
885 /* Note: depending on the protocol, we don't know if we're waiting
886 * for incoming data or not. So in order to prevent the socket from
887 * expiring read timeouts during writes, we refresh the read timeout,
888 * except if it was already infinite or if we have explicitly setup
889 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200890 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200891 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200892 }
893 }
894}
895
Willy Tarreau3b285d72019-06-06 08:20:17 +0200896/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
897 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
898 * be updated in case of success.
899 */
900void si_sync_send(struct stream_interface *si)
901{
902 struct channel *oc = si_oc(si);
903 struct conn_stream *cs;
904
905 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
906
907 if (oc->flags & CF_SHUTW)
908 return;
909
910 if (channel_is_empty(oc))
911 return;
912
913 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
914 return;
915
916 cs = objt_cs(si->end);
Olivier Houchard8af03b32020-01-22 17:34:54 +0100917 if (!cs || !cs->conn->mux)
Willy Tarreau3b285d72019-06-06 08:20:17 +0200918 return;
919
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200920 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200921}
922
Willy Tarreaud14844a2018-11-08 18:15:29 +0100923/* Updates at once the channel flags, and timers of both stream interfaces of a
924 * same stream, to complete the work after the analysers, then updates the data
925 * layer below. This will ensure that any synchronous update performed at the
926 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200927 * Note that this does not change the stream interface's current state, though
928 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100929 */
930void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
931{
932 struct channel *req = si_ic(si_f);
933 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100934
935 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
936 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
937
Willy Tarreaud14844a2018-11-08 18:15:29 +0100938 si_f->prev_state = si_f->state;
939 si_b->prev_state = si_b->state;
940
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100941 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200942 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100943 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100944
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200945 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100946 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100947
948 /* stream ints are processed outside of process_stream() and must be
949 * handled at the latest moment.
950 */
951 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100952 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
953 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100954 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100955
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100956 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100957 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
958 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100959 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100960}
961
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200962/*
963 * This function performs a shutdown-read on a stream interface attached to
964 * a connection in a connected or init state (it does nothing for other
965 * states). It either shuts the read side or marks itself as closed. The buffer
966 * flags are updated to reflect the new state. If the stream interface has
967 * SI_FL_NOHALF, we also forward the close to the write side. If a control
968 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200969 * descriptors are then shutdown or closed accordingly. The function
970 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200971 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200972static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200973{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200974 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100975 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200976
Willy Tarreauabb5d422018-11-14 16:58:52 +0100977 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100978 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200979 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100980 ic->flags |= CF_SHUTR;
981 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200982
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200983 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200984 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200985
Willy Tarreau51d0a7e2019-01-31 19:09:59 +0100986 if (si->flags & SI_FL_KILL_CONN)
987 cs->flags |= CS_FL_KILL_CONN;
988
Willy Tarreau2bb4a962014-11-28 11:11:05 +0100989 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +0200990 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200991 si->state = SI_ST_DIS;
992 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200993 }
994 else if (si->flags & SI_FL_NOHALF) {
995 /* we want to immediately forward this close to the write side */
996 return stream_int_shutw_conn(si);
997 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200998}
999
1000/*
1001 * This function performs a shutdown-write on a stream interface attached to
1002 * a connection in a connected or init state (it does nothing for other
1003 * states). It either shuts the write side or marks itself as closed. The
1004 * buffer flags are updated to reflect the new state. It does also close
1005 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001006 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001007 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001008static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001009{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001010 struct conn_stream *cs = __objt_cs(si->end);
1011 struct connection *conn = cs->conn;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001012 struct channel *ic = si_ic(si);
1013 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001014
Willy Tarreauafc8a222014-11-28 15:46:27 +01001015 oc->flags &= ~CF_SHUTW_NOW;
1016 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001017 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001018 oc->flags |= CF_SHUTW;
1019 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001020 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001021
Hongbo Longe39683c2017-03-10 18:41:51 +01001022 if (tick_isset(si->hcto)) {
1023 ic->rto = si->hcto;
1024 ic->rex = tick_add(now_ms, ic->rto);
1025 }
1026
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001027 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001028 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001029 case SI_ST_EST:
1030 /* we have to shut before closing, otherwise some short messages
1031 * may never leave the system, especially when there are remaining
1032 * unread data in the socket input buffer, or when nolinger is set.
1033 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1034 * no risk so we close both sides immediately.
1035 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001036 if (si->flags & SI_FL_KILL_CONN)
1037 cs->flags |= CS_FL_KILL_CONN;
1038
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001039 if (si->flags & SI_FL_ERR) {
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +05001040 /* quick close, the socket is already shut anyway */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001041 }
1042 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001043 /* unclean data-layer shutdown, typically an aborted request
1044 * or a forwarded shutdown from a client to a server due to
1045 * option abortonclose. No need for the TLS layer to try to
1046 * emit a shutdown message.
1047 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001048 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001049 }
1050 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001051 /* clean data-layer shutdown. This only happens on the
1052 * frontend side, or on the backend side when forwarding
1053 * a client close in TCP mode or in HTTP TUNNEL mode
1054 * while option abortonclose is set. We want the TLS
1055 * layer to try to signal it to the peer before we close.
1056 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001057 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001058
Willy Tarreaua553ae92017-10-05 18:52:17 +02001059 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ))) {
1060 /* OK just a shutw, but we want the caller
1061 * to disable polling on this FD if exists.
1062 */
1063 conn_cond_update_polling(conn);
1064 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001065 }
1066 }
1067
1068 /* fall through */
1069 case SI_ST_CON:
1070 /* we may have to close a pending connection, and mark the
1071 * response buffer as shutr
1072 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001073 if (si->flags & SI_FL_KILL_CONN)
1074 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001075 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001076 /* fall through */
1077 case SI_ST_CER:
1078 case SI_ST_QUE:
1079 case SI_ST_TAR:
1080 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001081 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001082 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001083 si->flags &= ~SI_FL_NOLINGER;
1084 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001085 ic->flags |= CF_SHUTR;
1086 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001087 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001088 }
1089}
1090
Willy Tarreau46a8d922012-08-20 12:38:36 +02001091/* This function is used for inter-stream-interface calls. It is called by the
1092 * consumer to inform the producer side that it may be interested in checking
1093 * for free space in the buffer. Note that it intentionally does not update
1094 * timeouts, so that we can still check them later at wake-up. This function is
1095 * dedicated to connection-based stream interfaces.
1096 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001097static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001098{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001099 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001100 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001101 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001102}
1103
1104
Willy Tarreaude5722c2012-08-20 15:01:10 +02001105/* This function is used for inter-stream-interface calls. It is called by the
1106 * producer to inform the consumer side that it may be interested in checking
1107 * for data in the buffer. Note that it intentionally does not update timeouts,
1108 * so that we can still check them later at wake-up.
1109 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001110static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001111{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001112 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001113 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001114
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001115 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001116 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001117 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001118
Willy Tarreauafc8a222014-11-28 15:46:27 +01001119 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001120 return;
1121
Willy Tarreauafc8a222014-11-28 15:46:27 +01001122 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001123 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001124 return;
1125
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001126 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001127 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001128
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001129 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001130 /* Write error on the file descriptor */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +02001131 if (si->state >= SI_ST_CON)
1132 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001133 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001134 }
1135
1136 /* OK, so now we know that some data might have been sent, and that we may
1137 * have to poll first. We have to do that too if the buffer is not empty.
1138 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001139 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001140 /* the connection is established but we can't write. Either the
1141 * buffer is empty, or we just refrain from sending because the
1142 * ->o limit was reached. Maybe we just wrote the last
1143 * chunk and need to close.
1144 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001145 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001146 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001147 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001148 si_shutw(si);
1149 goto out_wakeup;
1150 }
1151
Willy Tarreauafc8a222014-11-28 15:46:27 +01001152 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001153 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001154 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001155 }
1156 else {
1157 /* Otherwise there are remaining data to be sent in the buffer,
1158 * which means we have to poll before doing so.
1159 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001160 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001161 if (!tick_isset(oc->wex))
1162 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001163 }
1164
Willy Tarreauafc8a222014-11-28 15:46:27 +01001165 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1166 struct channel *ic = si_ic(si);
1167
Willy Tarreaude5722c2012-08-20 15:01:10 +02001168 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001169 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1170 !channel_is_empty(oc))
1171 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001172
Willy Tarreauafc8a222014-11-28 15:46:27 +01001173 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001174 /* Note: to prevent the client from expiring read timeouts
1175 * during writes, we refresh it. We only do this if the
1176 * interface is not configured for "independent streams",
1177 * because for some applications it's better not to do this,
1178 * for instance when continuously exchanging small amounts
1179 * of data which can full the socket buffers long before a
1180 * write timeout is detected.
1181 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001182 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001183 }
1184 }
1185
1186 /* in case of special condition (error, shutdown, end of write...), we
1187 * have to notify the task.
1188 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001189 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1190 ((oc->flags & CF_WAKE_WRITE) &&
1191 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001192 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001193 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001194 if (!(si->flags & SI_FL_DONT_WAKE))
1195 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001196 }
1197}
1198
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001199/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001200 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001201 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001202 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001203 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001204int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001205{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001206 struct connection *conn = cs->conn;
1207 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001208 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001209 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001210 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001211 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001212
Christopher Faulet04400bc2019-10-25 10:21:01 +02001213 /* If not established yet, do nothing. */
1214 if (si->state != SI_ST_EST)
1215 return 0;
1216
Olivier Houchardf6535282018-08-31 17:29:12 +02001217 /* If another call to si_cs_recv() failed, and we subscribed to
1218 * recv events already, give up now.
1219 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001220 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001221 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001222
Willy Tarreauce323de2012-08-20 21:41:06 +02001223 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001224 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001225 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001226
Christopher Faulete96993b2020-07-30 09:26:46 +02001227 /* we must wait because the mux is not installed yet */
1228 if (!conn->mux)
1229 return 0;
1230
Willy Tarreau54e917c2017-08-30 07:35:35 +02001231 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001232 if (cs->flags & CS_FL_EOS)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001233 goto end_recv;
Willy Tarreau54e917c2017-08-30 07:35:35 +02001234
Christopher Fauletf061e422018-12-07 14:51:20 +01001235 /* stop immediately on errors. Note that we DON'T want to stop on
1236 * POLL_ERR, as the poller might report a write error while there
1237 * are still data available in the recv buffer. This typically
1238 * happens when we send too large a request to a backend server
1239 * which rejects it before reading it all.
1240 */
1241 if (!(cs->flags & CS_FL_RCV_MORE)) {
1242 if (!conn_xprt_ready(conn))
1243 return 0;
1244 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001245 goto end_recv;
Christopher Fauletf061e422018-12-07 14:51:20 +01001246 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001247
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001248 /* prepare to detect if the mux needs more room */
1249 cs->flags &= ~CS_FL_WANT_ROOM;
1250
Willy Tarreau77e478c2018-06-19 07:03:14 +02001251 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001252 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001253 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001254 /* The buffer was empty and nothing was transferred for more
1255 * than one second. This was caused by a pause and not by
1256 * congestion. Reset any streaming mode to reduce latency.
1257 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001258 ic->xfer_small = 0;
1259 ic->xfer_large = 0;
1260 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001261 }
1262
Willy Tarreau96199b12012-08-24 00:46:52 +02001263 /* First, let's see if we may splice data across the channel without
1264 * using a buffer.
1265 */
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001266 if (cs->flags & CS_FL_MAY_SPLICE &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001267 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1268 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001269 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001270 /* We're embarrassed, there are already data pending in
1271 * the buffer and we don't want to have them at two
1272 * locations at a time. Let's indicate we need some
1273 * place and ask the consumer to hurry.
1274 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001275 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001276 goto abort_splice;
1277 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001278
Willy Tarreauafc8a222014-11-28 15:46:27 +01001279 if (unlikely(ic->pipe == NULL)) {
1280 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1281 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001282 goto abort_splice;
1283 }
1284 }
1285
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001286 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001287 if (ret < 0) {
1288 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001289 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001290 goto abort_splice;
1291 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001292
Willy Tarreau96199b12012-08-24 00:46:52 +02001293 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001294 if (ic->to_forward != CHN_INFINITE_FORWARD)
1295 ic->to_forward -= ret;
1296 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001297 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001298 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001299 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001300
Christopher Faulet36b536d2019-11-20 11:56:33 +01001301 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_EOS|CS_FL_ERROR))
1302 goto end_recv;
Willy Tarreau96199b12012-08-24 00:46:52 +02001303
Willy Tarreau61d39a02013-07-18 21:49:32 +02001304 if (conn->flags & CO_FL_WAIT_ROOM) {
1305 /* the pipe is full or we have read enough data that it
1306 * could soon be full. Let's stop before needing to poll.
1307 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001308 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001309 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001310 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001311
Willy Tarreauce323de2012-08-20 21:41:06 +02001312 /* splice not possible (anymore), let's go on on standard copy */
1313 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001314
1315 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001316 if (ic->pipe && unlikely(!ic->pipe->data)) {
1317 put_pipe(ic->pipe);
1318 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001319 }
1320
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001321 if (ic->pipe && ic->to_forward && !(flags & CO_RFL_BUF_FLUSH) && cs->flags & CS_FL_MAY_SPLICE) {
Willy Tarreauc640ef12019-12-03 18:13:04 +01001322 /* don't break splicing by reading, but still call rcv_buf()
1323 * to pass the flag.
1324 */
1325 goto done_recv;
1326 }
1327
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001328 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001329 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001330 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001331
Willy Tarreau61d39a02013-07-18 21:49:32 +02001332 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1333 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1334 * that if such an event is not handled above in splice, it will be handled here by
1335 * recv().
1336 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001337 while ((cs->flags & CS_FL_RCV_MORE) ||
Willy Tarreau93c9f592020-01-17 17:24:30 +01001338 (!(conn->flags & (CO_FL_ERROR | CO_FL_HANDSHAKE)) &&
Olivier Houchardc490efd2018-12-04 15:46:16 +01001339 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001340 /* <max> may be null. This is the mux responsibility to set
1341 * CS_FL_RCV_MORE on the CS if more space is needed.
1342 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001343 max = channel_recv_max(ic);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001344 ret = cs->conn->mux->rcv_buf(cs, &ic->buf, max, flags | (co_data(ic) ? CO_RFL_BUF_WET : 0));
Willy Tarreau674e0ad2018-12-05 13:45:41 +01001345
Olivier Houchardd247be02018-12-06 16:22:29 +01001346 if (cs->flags & CS_FL_WANT_ROOM)
Willy Tarreaudb398432018-11-15 11:08:52 +01001347 si_rx_room_blk(si);
Willy Tarreau6577b482017-12-10 21:19:33 +01001348
Christopher Fauleteffc3752018-10-31 08:53:54 +01001349 if (cs->flags & CS_FL_READ_PARTIAL) {
1350 if (tick_isset(ic->rex))
1351 ic->rex = tick_add_ifset(now_ms, ic->rto);
1352 cs->flags &= ~CS_FL_READ_PARTIAL;
1353 }
1354
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001355 if (ret <= 0) {
Willy Tarreau1ac5f202019-12-03 18:08:45 +01001356 /* if we refrained from reading because we asked for a
1357 * flush to satisfy rcv_pipe(), we must not subscribe
1358 * and instead report that there's not enough room
1359 * here to proceed.
1360 */
1361 if (flags & CO_RFL_BUF_FLUSH)
1362 si_rx_room_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001363 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001364 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001365
Christopher Faulet48726b72020-01-09 14:31:13 +01001366 /* L7 retries enabled and maximum connection retries not reached */
1367 if ((si->flags & SI_FL_L7_RETRY) && si->conn_retries) {
Olivier Houcharda254a372019-04-05 15:30:12 +02001368 struct htx *htx;
1369 struct htx_sl *sl;
1370
1371 htx = htxbuf(&ic->buf);
1372 if (htx) {
Christopher Faulet297fbb42019-05-13 14:41:27 +02001373 sl = http_get_stline(htx);
Olivier Houcharda254a372019-04-05 15:30:12 +02001374 if (sl && l7_status_match(si_strm(si)->be,
1375 sl->info.res.status)) {
1376 /* If we got a status for which we would
1377 * like to retry the request, empty
1378 * the buffer and pretend there's an
1379 * error on the channel.
1380 */
1381 ic->flags |= CF_READ_ERROR;
1382 htx_reset(htx);
1383 return 1;
1384 }
1385 }
1386 si->flags &= ~SI_FL_L7_RETRY;
1387 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001388 cur_read += ret;
1389
1390 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001391 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001392 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001393 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1394 if (fwd > ic->to_forward)
1395 fwd = ic->to_forward;
1396 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001397 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001398 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001399 }
1400
Willy Tarreauafc8a222014-11-28 15:46:27 +01001401 ic->flags |= CF_READ_PARTIAL;
1402 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001403
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001404 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1405 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001406 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001407 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001408 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001409
1410 /* if too many bytes were missing from last read, it means that
1411 * it's pointless trying to read again because the system does
1412 * not have them in buffers.
1413 */
1414 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001415 /* if a streamer has read few data, it may be because we
1416 * have exhausted system buffers. It's not worth trying
1417 * again.
1418 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001419 if (ic->flags & CF_STREAMER) {
1420 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001421 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001422 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001423 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001424
1425 /* if we read a large block smaller than what we requested,
1426 * it's almost certain we'll never get anything more.
1427 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001428 if (ret >= global.tune.recv_enough) {
1429 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001430 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001431 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001432 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001433 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001434
1435 /* if we are waiting for more space, don't try to read more data
1436 * right now.
1437 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001438 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001439 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001440 } /* while !flags */
1441
Willy Tarreauffb12052018-11-15 16:06:02 +01001442 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001443 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001444 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001445 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001446 ic->xfer_large = 0;
1447 ic->xfer_small++;
1448 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001449 /* we have read less than half of the buffer in
1450 * one pass, and this happened at least 3 times.
1451 * This is definitely not a streamer.
1452 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001453 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001454 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001455 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001456 /* if the buffer has been at least half full twice,
1457 * we receive faster than we send, so at least it
1458 * is not a "fast streamer".
1459 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001460 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001461 }
1462 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001463 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001464 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001465 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001466 ic->xfer_small = 0;
1467 ic->xfer_large++;
1468 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001469 /* we call this buffer a fast streamer if it manages
1470 * to be filled in one call 3 consecutive times.
1471 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001472 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001473 }
1474 }
1475 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001476 ic->xfer_small = 0;
1477 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001478 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001479 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001480 }
1481
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001482 end_recv:
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001483 ret = (cur_read != 0);
1484
Christopher Faulet36b536d2019-11-20 11:56:33 +01001485 /* Report EOI on the channel if it was reached from the mux point of
1486 * view. */
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001487 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI)) {
Christopher Faulet36b536d2019-11-20 11:56:33 +01001488 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001489 ret = 1;
1490 }
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001491
Christopher Faulet36b536d2019-11-20 11:56:33 +01001492 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
1493 cs->flags |= CS_FL_ERROR;
1494 si->flags |= SI_FL_ERR;
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001495 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001496 }
1497 else if (cs->flags & CS_FL_EOS) {
Willy Tarreau18955db2020-01-23 16:32:24 +01001498 /* we received a shutdown */
1499 ic->flags |= CF_READ_NULL;
1500 if (ic->flags & CF_AUTO_CLOSE)
1501 channel_shutw_now(ic);
1502 stream_int_read0(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001503 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001504 }
1505 else if (!si_rx_blocked(si)) {
1506 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001507 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001508 si_rx_endp_done(si);
1509 } else {
1510 si_rx_endp_more(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001511 ret = 1;
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001512 }
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001513 return ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001514}
1515
1516/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001517 * This function propagates a null read received on a socket-based connection.
1518 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001519 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001520 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001521static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001522{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001523 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001524 struct channel *ic = si_ic(si);
1525 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001526
Willy Tarreauabb5d422018-11-14 16:58:52 +01001527 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001528 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001529 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001530 ic->flags |= CF_SHUTR;
1531 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001532
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001533 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001534 return;
1535
Willy Tarreauafc8a222014-11-28 15:46:27 +01001536 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001537 goto do_close;
1538
1539 if (si->flags & SI_FL_NOHALF) {
1540 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001541 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001542 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001543 goto do_close;
1544 }
1545
1546 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001547 return;
1548
1549 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001550 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001551 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001552
Willy Tarreauafc8a222014-11-28 15:46:27 +01001553 oc->flags &= ~CF_SHUTW_NOW;
1554 oc->flags |= CF_SHUTW;
1555 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001556
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001557 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001558
Christopher Faulet04400bc2019-10-25 10:21:01 +02001559 si->state = SI_ST_DIS;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001560 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001561 return;
1562}
1563
Willy Tarreau651e1822015-09-23 20:06:13 +02001564/* Callback to be used by applet handlers upon completion. It updates the stream
1565 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001566 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001567 * states.
1568 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001569void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001570{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001571 struct channel *ic = si_ic(si);
1572
1573 /* If the applet wants to write and the channel is closed, it's a
1574 * broken pipe and it must be reported.
1575 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001576 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001577 si->flags |= SI_FL_ERR;
1578
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001579 /* automatically mark the applet having data available if it reported
1580 * begin blocked by the channel.
1581 */
1582 if (si_rx_blocked(si))
1583 si_rx_endp_more(si);
1584
Willy Tarreau651e1822015-09-23 20:06:13 +02001585 /* update the stream-int, channels, and possibly wake the stream up */
1586 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +02001587 stream_release_buffers(si_strm(si));
Willy Tarreaue5f86492015-04-19 15:16:35 +02001588
Willy Tarreau32742fd2018-11-14 14:07:59 +01001589 /* stream_int_notify may have passed through chk_snd and released some
1590 * RXBLK flags. Process_stream will consider those flags to wake up the
1591 * appctx but in the case the task is not in runqueue we may have to
1592 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001593 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001594 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1595 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001596 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001597}
1598
1599/*
1600 * This function performs a shutdown-read on a stream interface attached to an
1601 * applet in a connected or init state (it does nothing for other states). It
1602 * either shuts the read side or marks itself as closed. The buffer flags are
1603 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1604 * we also forward the close to the write side. The owner task is woken up if
1605 * it exists.
1606 */
1607static void stream_int_shutr_applet(struct stream_interface *si)
1608{
1609 struct channel *ic = si_ic(si);
1610
Willy Tarreauabb5d422018-11-14 16:58:52 +01001611 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001612 if (ic->flags & CF_SHUTR)
1613 return;
1614 ic->flags |= CF_SHUTR;
1615 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001616
Willy Tarreau828824a2015-04-19 17:20:03 +02001617 /* Note: on shutr, we don't call the applet */
1618
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001619 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001620 return;
1621
1622 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001623 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001624 si->state = SI_ST_DIS;
1625 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001626 }
1627 else if (si->flags & SI_FL_NOHALF) {
1628 /* we want to immediately forward this close to the write side */
1629 return stream_int_shutw_applet(si);
1630 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001631}
1632
1633/*
1634 * This function performs a shutdown-write on a stream interface attached to an
1635 * applet in a connected or init state (it does nothing for other states). It
1636 * either shuts the write side or marks itself as closed. The buffer flags are
1637 * updated to reflect the new state. It does also close everything if the SI
1638 * was marked as being in error state. The owner task is woken up if it exists.
1639 */
1640static void stream_int_shutw_applet(struct stream_interface *si)
1641{
1642 struct channel *ic = si_ic(si);
1643 struct channel *oc = si_oc(si);
1644
1645 oc->flags &= ~CF_SHUTW_NOW;
1646 if (oc->flags & CF_SHUTW)
1647 return;
1648 oc->flags |= CF_SHUTW;
1649 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001650 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001651
Hongbo Longe39683c2017-03-10 18:41:51 +01001652 if (tick_isset(si->hcto)) {
1653 ic->rto = si->hcto;
1654 ic->rex = tick_add(now_ms, ic->rto);
1655 }
1656
Willy Tarreau828824a2015-04-19 17:20:03 +02001657 /* on shutw we always wake the applet up */
1658 appctx_wakeup(si_appctx(si));
1659
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001660 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001661 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001662 case SI_ST_EST:
1663 /* we have to shut before closing, otherwise some short messages
1664 * may never leave the system, especially when there are remaining
1665 * unread data in the socket input buffer, or when nolinger is set.
1666 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1667 * no risk so we close both sides immediately.
1668 */
1669 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1670 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1671 return;
1672
1673 /* fall through */
1674 case SI_ST_CON:
1675 case SI_ST_CER:
1676 case SI_ST_QUE:
1677 case SI_ST_TAR:
1678 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001679 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001680 si->state = SI_ST_DIS;
Tim Duesterhus588b3142020-05-29 14:35:51 +02001681 /* fall through */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001682 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001683 si->flags &= ~SI_FL_NOLINGER;
1684 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001685 ic->flags |= CF_SHUTR;
1686 ic->rex = TICK_ETERNITY;
1687 si->exp = TICK_ETERNITY;
1688 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001689}
1690
1691/* chk_rcv function for applets */
1692static void stream_int_chk_rcv_applet(struct stream_interface *si)
1693{
1694 struct channel *ic = si_ic(si);
1695
1696 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1697 __FUNCTION__,
1698 si, si->state, ic->flags, si_oc(si)->flags);
1699
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001700 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001701 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001702 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001703 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001704}
1705
1706/* chk_snd function for applets */
1707static void stream_int_chk_snd_applet(struct stream_interface *si)
1708{
1709 struct channel *oc = si_oc(si);
1710
1711 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1712 __FUNCTION__,
1713 si, si->state, si_ic(si)->flags, oc->flags);
1714
1715 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1716 return;
1717
Willy Tarreau828824a2015-04-19 17:20:03 +02001718 /* we only wake the applet up if it was waiting for some data */
1719
1720 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001721 return;
1722
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001723 if (!tick_isset(oc->wex))
1724 oc->wex = tick_add_ifset(now_ms, oc->wto);
1725
Willy Tarreau828824a2015-04-19 17:20:03 +02001726 if (!channel_is_empty(oc)) {
1727 /* (re)start sending */
1728 appctx_wakeup(si_appctx(si));
1729 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001730}
1731
Willy Tarreaudded32d2008-11-30 19:48:07 +01001732/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001733 * Local variables:
1734 * c-indent-level: 8
1735 * c-basic-offset: 8
1736 * End:
1737 */