blob: e5da8ea56534ad091d8caee351b53820350e3b8c [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
Thayne McCombs8f0cc5c2021-01-07 21:35:52 -0700298 * wake it up every time it is solicited. The appctx must be deleted by the task
Willy Tarreaua9ff5e62015-07-19 18:46:30 +0200299 * 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 Tarreau827fee72020-12-11 15:26:55 +0100386 ret = conn_ctrl_send(conn,
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200387 trash.area + ret + conn->send_proxy_ofs,
388 -conn->send_proxy_ofs,
Willy Tarreau827fee72020-12-11 15:26:55 +0100389 (conn->subs && conn->subs->events & SUB_RETRY_SEND) ? CO_SFL_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 Faulet89e34c22021-01-21 16:22:01 +0100610 /* Report EOS on the channel if it was reached from the mux point of
611 * 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 */
617 if (cs->flags & CS_FL_EOS && !(ic->flags & CF_SHUTR)) {
618 /* we received a shutdown */
619 ic->flags |= CF_READ_NULL;
620 if (ic->flags & CF_AUTO_CLOSE)
621 channel_shutw_now(ic);
622 stream_int_read0(si);
623 }
624
Christopher Faulet297d3e22019-03-22 14:16:14 +0100625 /* Report EOI on the channel if it was reached from the mux point of
Christopher Faulet36b536d2019-11-20 11:56:33 +0100626 * view.
627 *
628 * Note: This test is only required because si_cs_process is also the SI
629 * wake callback. Otherwise si_cs_recv()/si_cs_send() already take
630 * care of it.
631 */
Christopher Faulet297d3e22019-03-22 14:16:14 +0100632 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI))
Christopher Faulet8e9e3ef2019-05-17 09:14:10 +0200633 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulet203b2b02019-03-08 09:23:46 +0100634
Willy Tarreau651e1822015-09-23 20:06:13 +0200635 /* Second step : update the stream-int and channels, try to forward any
636 * pending data, then possibly wake the stream up based on the new
637 * stream-int status.
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200638 */
Willy Tarreau651e1822015-09-23 20:06:13 +0200639 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +0200640 stream_release_buffers(si_strm(si));
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200641 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200642}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200643
Willy Tarreau5368d802012-08-21 18:22:06 +0200644/*
645 * This function is called to send buffer data to a stream socket.
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200646 * It calls the mux layer's snd_buf function. It relies on the
Godbach4f489902013-12-04 17:24:06 +0800647 * caller to commit polling changes. The caller should check conn->flags
648 * for errors.
Willy Tarreau5368d802012-08-21 18:22:06 +0200649 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +0200650int si_cs_send(struct conn_stream *cs)
Willy Tarreau5368d802012-08-21 18:22:06 +0200651{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200652 struct connection *conn = cs->conn;
653 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100654 struct channel *oc = si_oc(si);
Willy Tarreau5368d802012-08-21 18:22:06 +0200655 int ret;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200656 int did_send = 0;
657
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100658 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
Olivier Houchardc31e2cb2019-06-24 16:08:08 +0200659 /* We're probably there because the tasklet was woken up,
660 * but process_stream() ran before, detected there were an
661 * error and put the si back to SI_ST_TAR. There's still
662 * CO_FL_ERROR on the connection but we don't want to add
663 * SI_FL_ERR back, so give up
664 */
665 if (si->state < SI_ST_CON)
666 return 0;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100667 si->flags |= SI_FL_ERR;
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200668 return 1;
Willy Tarreaubddf7fc2018-12-19 17:17:10 +0100669 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200670
Christopher Faulet328ed222019-09-23 15:57:29 +0200671 /* We're already waiting to be able to send, give up */
672 if (si->wait_event.events & SUB_RETRY_SEND)
673 return 0;
674
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200675 /* we might have been called just after an asynchronous shutw */
Willy Tarreauf22758d2020-01-23 18:25:23 +0100676 if (oc->flags & CF_SHUTW)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200677 return 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200678
Christopher Faulete96993b2020-07-30 09:26:46 +0200679 /* we must wait because the mux is not installed yet */
680 if (!conn->mux)
681 return 0;
682
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200683 if (oc->pipe && conn->xprt->snd_pipe && conn->mux->snd_pipe) {
684 ret = conn->mux->snd_pipe(cs, oc->pipe);
Christopher Faulet86162db2019-07-05 11:49:11 +0200685 if (ret > 0)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200686 did_send = 1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200687
Willy Tarreauafc8a222014-11-28 15:46:27 +0100688 if (!oc->pipe->data) {
689 put_pipe(oc->pipe);
690 oc->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200691 }
692
Christopher Faulet3f76f4c2018-11-20 10:21:08 +0100693 if (oc->pipe)
694 goto end;
Willy Tarreau5368d802012-08-21 18:22:06 +0200695 }
696
697 /* At this point, the pipe is empty, but we may still have data pending
698 * in the normal buffer.
699 */
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100700 if (co_data(oc)) {
701 /* when we're here, we already know that there is no spliced
702 * data left, and that there are sendable buffered data.
703 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200704
Willy Tarreau5368d802012-08-21 18:22:06 +0200705 /* check if we want to inform the kernel that we're interested in
706 * sending more data after this call. We want this if :
707 * - we're about to close after this last send and want to merge
708 * the ongoing FIN with the last segment.
709 * - we know we can't send everything at once and must get back
710 * here because of unaligned data
711 * - there is still a finite amount of data to forward
712 * The test is arranged so that the most common case does only 2
713 * tests.
714 */
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100715 unsigned int send_flag = 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200716
Willy Tarreauafc8a222014-11-28 15:46:27 +0100717 if ((!(oc->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
718 ((oc->to_forward && oc->to_forward != CHN_INFINITE_FORWARD) ||
Willy Tarreau8945bb62020-06-19 17:07:06 +0200719 (oc->flags & CF_EXPECT_MORE) ||
Christopher Faulet9e3dc832020-07-22 16:28:44 +0200720 (IS_HTX_STRM(si_strm(si)) &&
721 (!(oc->flags & (CF_EOI|CF_SHUTR)) && htx_expect_more(htxbuf(&oc->buf)))))) ||
Willy Tarreauecd2e152017-11-07 15:07:25 +0100722 ((oc->flags & CF_ISRESP) &&
723 ((oc->flags & (CF_AUTO_CLOSE|CF_SHUTW_NOW)) == (CF_AUTO_CLOSE|CF_SHUTW_NOW))))
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100724 send_flag |= CO_SFL_MSG_MORE;
Willy Tarreau5368d802012-08-21 18:22:06 +0200725
Willy Tarreauafc8a222014-11-28 15:46:27 +0100726 if (oc->flags & CF_STREAMER)
Willy Tarreau7bed9452014-02-02 02:00:24 +0100727 send_flag |= CO_SFL_STREAMER;
728
Christopher Faulet5bf85852021-05-21 13:46:14 +0200729 if ((si->flags & SI_FL_L7_RETRY) && !b_data(&si->l7_buffer)) {
730 struct stream *s = si_strm(si);
731 /* If we want to be able to do L7 retries, copy
732 * the data we're about to send, so that we are able
733 * to resend them if needed
734 */
735 /* Try to allocate a buffer if we had none.
736 * If it fails, the next test will just
737 * disable the l7 retries by setting
738 * l7_conn_retries to 0.
739 */
740 if (!s->txn || (s->txn->req.msg_state != HTTP_MSG_DONE))
741 si->flags &= ~SI_FL_L7_RETRY;
742 else {
Christopher Fauletd6bbe822021-06-11 15:55:56 +0200743 if (b_alloc(&si->l7_buffer) == NULL)
Christopher Faulet5bf85852021-05-21 13:46:14 +0200744 si->flags &= ~SI_FL_L7_RETRY;
745 else {
746 memcpy(b_orig(&si->l7_buffer),
747 b_orig(&oc->buf),
748 b_size(&oc->buf));
749 si->l7_buffer.head = co_data(oc);
750 b_add(&si->l7_buffer, co_data(oc));
751 }
752
753 }
754 }
755
Olivier Houcharded0f2072018-08-16 15:41:52 +0200756 ret = cs->conn->mux->snd_buf(cs, &oc->buf, co_data(oc), send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800757 if (ret > 0) {
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200758 did_send = 1;
Willy Tarreau77e478c2018-06-19 07:03:14 +0200759 co_set_data(oc, co_data(oc) - ret);
Willy Tarreaudeccd112018-06-14 18:38:55 +0200760 c_realign_if_empty(oc);
761
762 if (!co_data(oc)) {
Godbache68e02d2013-10-11 15:48:29 +0800763 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreauafc8a222014-11-28 15:46:27 +0100764 oc->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Godbache68e02d2013-10-11 15:48:29 +0800765 }
Godbache68e02d2013-10-11 15:48:29 +0800766 /* if some data remain in the buffer, it's only because the
767 * system buffers are full, we will try next time.
768 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200769 }
Godbache68e02d2013-10-11 15:48:29 +0800770 }
Christopher Faulet55dec0d2018-11-20 10:30:02 +0100771
Willy Tarreauf6975aa2018-11-15 14:33:05 +0100772 end:
Christopher Faulet86162db2019-07-05 11:49:11 +0200773 if (did_send) {
774 oc->flags |= CF_WRITE_PARTIAL | CF_WROTE_DATA;
775 if (si->state == SI_ST_CON)
776 si->state = SI_ST_RDY;
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200777
778 si_rx_room_rdy(si_opposite(si));
Christopher Faulet86162db2019-07-05 11:49:11 +0200779 }
780
781 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING)) {
782 si->flags |= SI_FL_ERR;
783 return 1;
784 }
785
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200786 /* 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 +0100787 if (!channel_is_empty(oc))
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100788 conn->mux->subscribe(cs, SUB_RETRY_SEND, &si->wait_event);
Olivier Houchardf6535282018-08-31 17:29:12 +0200789 return did_send;
Willy Tarreau5368d802012-08-21 18:22:06 +0200790}
791
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100792/* This is the ->process() function for any stream-interface's wait_event task.
793 * It's assigned during the stream-interface's initialization, for any type of
794 * stream interface. Thus it is always safe to perform a tasklet_wakeup() on a
795 * stream interface, as the presence of the CS is checked there.
796 */
Willy Tarreau144f84a2021-03-02 16:09:26 +0100797struct task *si_cs_io_cb(struct task *t, void *ctx, unsigned int state)
Olivier Houchard91894cb2018-08-02 18:06:28 +0200798{
Olivier Houchard8f0b4c62018-08-02 18:21:38 +0200799 struct stream_interface *si = ctx;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200800 struct conn_stream *cs = objt_cs(si->end);
Olivier Houchardf6535282018-08-31 17:29:12 +0200801 int ret = 0;
Olivier Houcharda6ff0352018-08-21 15:59:43 +0200802
803 if (!cs)
Willy Tarreau74163142021-03-13 11:30:19 +0100804 return t;
Willy Tarreau8ccd2082018-11-07 07:47:52 +0100805
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100806 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchardf6535282018-08-31 17:29:12 +0200807 ret = si_cs_send(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100808 if (!(si->wait_event.events & SUB_RETRY_RECV))
Olivier Houchardf6535282018-08-31 17:29:12 +0200809 ret |= si_cs_recv(cs);
810 if (ret != 0)
Olivier Houchardc2aa7112018-09-11 18:27:21 +0200811 si_cs_process(cs);
Olivier Houchardf6535282018-08-31 17:29:12 +0200812
Willy Tarreaua64c7032019-08-01 14:17:02 +0200813 stream_release_buffers(si_strm(si));
Willy Tarreau74163142021-03-13 11:30:19 +0100814 return t;
Olivier Houchard91894cb2018-08-02 18:06:28 +0200815}
816
Willy Tarreau25f13102015-09-24 11:32:22 +0200817/* This function is designed to be called from within the stream handler to
Willy Tarreau236c4292019-06-06 08:19:20 +0200818 * update the input channel's expiration timer and the stream interface's
819 * Rx flags based on the channel's flags. It needs to be called only once
820 * after the channel's flags have settled down, and before they are cleared,
821 * though it doesn't harm to call it as often as desired (it just slightly
822 * hurts performance). It must not be called from outside of the stream
823 * handler, as what it does will be used to compute the stream task's
824 * expiration.
Willy Tarreau25f13102015-09-24 11:32:22 +0200825 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200826void si_update_rx(struct stream_interface *si)
Willy Tarreau25f13102015-09-24 11:32:22 +0200827{
828 struct channel *ic = si_ic(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200829
Willy Tarreau236c4292019-06-06 08:19:20 +0200830 if (ic->flags & CF_SHUTR) {
831 si_rx_shut_blk(si);
832 return;
833 }
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100834
Willy Tarreau236c4292019-06-06 08:19:20 +0200835 /* Read not closed, update FD status and timeout for reads */
836 if (ic->flags & CF_DONT_READ)
837 si_rx_chan_blk(si);
838 else
839 si_rx_chan_rdy(si);
Willy Tarreaub26a6f92018-11-14 17:10:36 +0100840
Christopher Faulet62fc11c2021-09-07 14:31:02 +0200841 if (!channel_is_empty(ic)) {
Willy Tarreau236c4292019-06-06 08:19:20 +0200842 /* stop reading, imposed by channel's policy or contents */
843 si_rx_room_blk(si);
Willy Tarreau25f13102015-09-24 11:32:22 +0200844 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200845 else {
846 /* (re)start reading and update timeout. Note: we don't recompute the timeout
Thayne McCombs8f0cc5c2021-01-07 21:35:52 -0700847 * every time we get here, otherwise it would risk never to expire. We only
Willy Tarreau236c4292019-06-06 08:19:20 +0200848 * update it if is was not yet set. The stream socket handler will already
849 * have updated it if there has been a completed I/O.
850 */
851 si_rx_room_rdy(si);
852 }
853 if (si->flags & SI_FL_RXBLK_ANY & ~SI_FL_RX_WAIT_EP)
854 ic->rex = TICK_ETERNITY;
855 else if (!(ic->flags & CF_READ_NOEXP) && !tick_isset(ic->rex))
856 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200857
Willy Tarreau236c4292019-06-06 08:19:20 +0200858 si_chk_rcv(si);
859}
860
861/* This function is designed to be called from within the stream handler to
862 * update the output channel's expiration timer and the stream interface's
863 * Tx flags based on the channel's flags. It needs to be called only once
864 * after the channel's flags have settled down, and before they are cleared,
865 * though it doesn't harm to call it as often as desired (it just slightly
866 * hurts performance). It must not be called from outside of the stream
867 * handler, as what it does will be used to compute the stream task's
868 * expiration.
869 */
870void si_update_tx(struct stream_interface *si)
871{
872 struct channel *oc = si_oc(si);
873 struct channel *ic = si_ic(si);
874
875 if (oc->flags & CF_SHUTW)
876 return;
877
878 /* Write not closed, update FD status and timeout for writes */
879 if (channel_is_empty(oc)) {
880 /* stop writing */
881 if (!(si->flags & SI_FL_WAIT_DATA)) {
882 if ((oc->flags & CF_SHUTW_NOW) == 0)
883 si->flags |= SI_FL_WAIT_DATA;
884 oc->wex = TICK_ETERNITY;
Willy Tarreau25f13102015-09-24 11:32:22 +0200885 }
Willy Tarreau236c4292019-06-06 08:19:20 +0200886 return;
887 }
888
889 /* (re)start writing and update timeout. Note: we don't recompute the timeout
Thayne McCombs8f0cc5c2021-01-07 21:35:52 -0700890 * every time we get here, otherwise it would risk never to expire. We only
Willy Tarreau236c4292019-06-06 08:19:20 +0200891 * update it if is was not yet set. The stream socket handler will already
892 * have updated it if there has been a completed I/O.
893 */
894 si->flags &= ~SI_FL_WAIT_DATA;
895 if (!tick_isset(oc->wex)) {
896 oc->wex = tick_add_ifset(now_ms, oc->wto);
897 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
898 /* Note: depending on the protocol, we don't know if we're waiting
899 * for incoming data or not. So in order to prevent the socket from
900 * expiring read timeouts during writes, we refresh the read timeout,
901 * except if it was already infinite or if we have explicitly setup
902 * independent streams.
Willy Tarreau25f13102015-09-24 11:32:22 +0200903 */
Willy Tarreau236c4292019-06-06 08:19:20 +0200904 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreau25f13102015-09-24 11:32:22 +0200905 }
906 }
907}
908
Willy Tarreau3b285d72019-06-06 08:20:17 +0200909/* perform a synchronous send() for the stream interface. The CF_WRITE_NULL and
910 * CF_WRITE_PARTIAL flags are cleared prior to the attempt, and will possibly
911 * be updated in case of success.
912 */
913void si_sync_send(struct stream_interface *si)
914{
915 struct channel *oc = si_oc(si);
916 struct conn_stream *cs;
917
918 oc->flags &= ~(CF_WRITE_NULL|CF_WRITE_PARTIAL);
919
920 if (oc->flags & CF_SHUTW)
921 return;
922
923 if (channel_is_empty(oc))
924 return;
925
926 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
927 return;
928
929 cs = objt_cs(si->end);
Olivier Houchard8af03b32020-01-22 17:34:54 +0100930 if (!cs || !cs->conn->mux)
Willy Tarreau3b285d72019-06-06 08:20:17 +0200931 return;
932
Christopher Faulet037b3eb2019-07-05 13:44:29 +0200933 si_cs_send(cs);
Willy Tarreau3b285d72019-06-06 08:20:17 +0200934}
935
Willy Tarreaud14844a2018-11-08 18:15:29 +0100936/* Updates at once the channel flags, and timers of both stream interfaces of a
937 * same stream, to complete the work after the analysers, then updates the data
938 * layer below. This will ensure that any synchronous update performed at the
939 * data layer will be reflected in the channel flags and/or stream-interface.
Willy Tarreau829bd472019-06-06 09:17:23 +0200940 * Note that this does not change the stream interface's current state, though
941 * it updates the previous state to the current one.
Willy Tarreaud14844a2018-11-08 18:15:29 +0100942 */
943void si_update_both(struct stream_interface *si_f, struct stream_interface *si_b)
944{
945 struct channel *req = si_ic(si_f);
946 struct channel *res = si_oc(si_f);
Willy Tarreaud14844a2018-11-08 18:15:29 +0100947
948 req->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
949 res->flags &= ~(CF_READ_NULL|CF_READ_PARTIAL|CF_READ_ATTACHED|CF_WRITE_NULL|CF_WRITE_PARTIAL);
950
Willy Tarreaud14844a2018-11-08 18:15:29 +0100951 si_f->prev_state = si_f->state;
952 si_b->prev_state = si_b->state;
953
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100954 /* let's recompute both sides states */
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200955 if (si_state_in(si_f->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100956 si_update(si_f);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100957
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200958 if (si_state_in(si_b->state, SI_SB_RDY|SI_SB_EST))
Willy Tarreau14bfe9a2018-12-19 15:19:27 +0100959 si_update(si_b);
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100960
961 /* stream ints are processed outside of process_stream() and must be
962 * handled at the latest moment.
963 */
964 if (obj_type(si_f->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100965 ((si_rx_endp_ready(si_f) && !si_rx_blocked(si_f)) ||
966 (si_tx_endp_ready(si_f) && !si_tx_blocked(si_f))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100967 appctx_wakeup(si_appctx(si_f));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100968
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100969 if (obj_type(si_b->end) == OBJ_TYPE_APPCTX &&
Willy Tarreau32742fd2018-11-14 14:07:59 +0100970 ((si_rx_endp_ready(si_b) && !si_rx_blocked(si_b)) ||
971 (si_tx_endp_ready(si_b) && !si_tx_blocked(si_b))))
Willy Tarreaubf89ff32018-11-09 14:59:25 +0100972 appctx_wakeup(si_appctx(si_b));
Willy Tarreaud14844a2018-11-08 18:15:29 +0100973}
974
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200975/*
976 * This function performs a shutdown-read on a stream interface attached to
977 * a connection in a connected or init state (it does nothing for other
978 * states). It either shuts the read side or marks itself as closed. The buffer
979 * flags are updated to reflect the new state. If the stream interface has
980 * SI_FL_NOHALF, we also forward the close to the write side. If a control
981 * layer is defined, then it is supposed to be a socket layer and file
Willy Tarreau6fe15412013-09-29 15:16:03 +0200982 * descriptors are then shutdown or closed accordingly. The function
983 * automatically disables polling if needed.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200984 */
Willy Tarreau6fe15412013-09-29 15:16:03 +0200985static void stream_int_shutr_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200986{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200987 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100988 struct channel *ic = si_ic(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200989
Willy Tarreauabb5d422018-11-14 16:58:52 +0100990 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +0100991 if (ic->flags & CF_SHUTR)
Willy Tarreau6fe15412013-09-29 15:16:03 +0200992 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +0100993 ic->flags |= CF_SHUTR;
994 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200995
Willy Tarreau4f283fa2019-06-05 14:34:03 +0200996 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau6fe15412013-09-29 15:16:03 +0200997 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +0200998
Willy Tarreau51d0a7e2019-01-31 19:09:59 +0100999 if (si->flags & SI_FL_KILL_CONN)
1000 cs->flags |= CS_FL_KILL_CONN;
1001
Willy Tarreau2bb4a962014-11-28 11:11:05 +01001002 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02001003 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001004 si->state = SI_ST_DIS;
1005 si->exp = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001006 }
1007 else if (si->flags & SI_FL_NOHALF) {
1008 /* we want to immediately forward this close to the write side */
1009 return stream_int_shutw_conn(si);
1010 }
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001011}
1012
1013/*
1014 * This function performs a shutdown-write on a stream interface attached to
1015 * a connection in a connected or init state (it does nothing for other
1016 * states). It either shuts the write side or marks itself as closed. The
1017 * buffer flags are updated to reflect the new state. It does also close
1018 * everything if the SI was marked as being in error state. If there is a
Willy Tarreau1398aa12015-03-12 23:04:07 +01001019 * data-layer shutdown, it is called.
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001020 */
Willy Tarreau6fe15412013-09-29 15:16:03 +02001021static void stream_int_shutw_conn(struct stream_interface *si)
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001022{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001023 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001024 struct channel *ic = si_ic(si);
1025 struct channel *oc = si_oc(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001026
Willy Tarreauafc8a222014-11-28 15:46:27 +01001027 oc->flags &= ~CF_SHUTW_NOW;
1028 if (oc->flags & CF_SHUTW)
Willy Tarreau6fe15412013-09-29 15:16:03 +02001029 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001030 oc->flags |= CF_SHUTW;
1031 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001032 si_done_get(si);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001033
Hongbo Longe39683c2017-03-10 18:41:51 +01001034 if (tick_isset(si->hcto)) {
1035 ic->rto = si->hcto;
1036 ic->rex = tick_add(now_ms, ic->rto);
1037 }
1038
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001039 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001040 case SI_ST_RDY:
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001041 case SI_ST_EST:
1042 /* we have to shut before closing, otherwise some short messages
1043 * may never leave the system, especially when there are remaining
1044 * unread data in the socket input buffer, or when nolinger is set.
1045 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1046 * no risk so we close both sides immediately.
1047 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001048 if (si->flags & SI_FL_KILL_CONN)
1049 cs->flags |= CS_FL_KILL_CONN;
1050
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001051 if (si->flags & SI_FL_ERR) {
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +05001052 /* quick close, the socket is already shut anyway */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001053 }
1054 else if (si->flags & SI_FL_NOLINGER) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001055 /* unclean data-layer shutdown, typically an aborted request
1056 * or a forwarded shutdown from a client to a server due to
1057 * option abortonclose. No need for the TLS layer to try to
1058 * emit a shutdown message.
1059 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001060 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001061 }
1062 else {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001063 /* clean data-layer shutdown. This only happens on the
1064 * frontend side, or on the backend side when forwarding
1065 * a client close in TCP mode or in HTTP TUNNEL mode
1066 * while option abortonclose is set. We want the TLS
1067 * layer to try to signal it to the peer before we close.
1068 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001069 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001070
Willy Tarreaua5ea7512020-12-11 10:24:05 +01001071 if (!(ic->flags & (CF_SHUTR|CF_DONT_READ)))
Willy Tarreaua553ae92017-10-05 18:52:17 +02001072 return;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001073 }
1074
1075 /* fall through */
1076 case SI_ST_CON:
1077 /* we may have to close a pending connection, and mark the
1078 * response buffer as shutr
1079 */
Willy Tarreau51d0a7e2019-01-31 19:09:59 +01001080 if (si->flags & SI_FL_KILL_CONN)
1081 cs->flags |= CS_FL_KILL_CONN;
Willy Tarreaua553ae92017-10-05 18:52:17 +02001082 cs_close(cs);
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001083 /* fall through */
1084 case SI_ST_CER:
1085 case SI_ST_QUE:
1086 case SI_ST_TAR:
1087 si->state = SI_ST_DIS;
Willy Tarreau4a59f2f2013-10-24 20:10:45 +02001088 /* fall through */
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001089 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001090 si->flags &= ~SI_FL_NOLINGER;
1091 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001092 ic->flags |= CF_SHUTR;
1093 ic->rex = TICK_ETERNITY;
Willy Tarreau8b3d7df2013-09-29 14:51:58 +02001094 si->exp = TICK_ETERNITY;
Willy Tarreau100c4672012-08-20 12:06:26 +02001095 }
1096}
1097
Willy Tarreau46a8d922012-08-20 12:38:36 +02001098/* This function is used for inter-stream-interface calls. It is called by the
1099 * consumer to inform the producer side that it may be interested in checking
1100 * for free space in the buffer. Note that it intentionally does not update
1101 * timeouts, so that we can still check them later at wake-up. This function is
1102 * dedicated to connection-based stream interfaces.
1103 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001104static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +02001105{
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001106 /* (re)start reading */
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001107 if (si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02001108 tasklet_wakeup(si->wait_event.tasklet);
Willy Tarreau46a8d922012-08-20 12:38:36 +02001109}
1110
1111
Willy Tarreaude5722c2012-08-20 15:01:10 +02001112/* This function is used for inter-stream-interface calls. It is called by the
1113 * producer to inform the consumer side that it may be interested in checking
1114 * for data in the buffer. Note that it intentionally does not update timeouts,
1115 * so that we can still check them later at wake-up.
1116 */
Willy Tarreauc5788912012-08-24 18:12:41 +02001117static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001118{
Willy Tarreauafc8a222014-11-28 15:46:27 +01001119 struct channel *oc = si_oc(si);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001120 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001121
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001122 if (unlikely(!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST) ||
Olivier Houchardb2fc04e2019-04-11 13:56:26 +02001123 (oc->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +02001124 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001125
Willy Tarreauafc8a222014-11-28 15:46:27 +01001126 if (unlikely(channel_is_empty(oc))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001127 return;
1128
Willy Tarreauafc8a222014-11-28 15:46:27 +01001129 if (!oc->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +01001130 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001131 return;
1132
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001133 if (!(si->wait_event.events & SUB_RETRY_SEND) && !channel_is_empty(si_oc(si)))
Olivier Houchard31f04e42018-10-22 16:01:09 +02001134 si_cs_send(cs);
Willy Tarreau33a09a52018-10-25 13:49:49 +02001135
Willy Tarreaubddf7fc2018-12-19 17:17:10 +01001136 if (cs->flags & (CS_FL_ERROR|CS_FL_ERR_PENDING) || cs->conn->flags & CO_FL_ERROR) {
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001137 /* Write error on the file descriptor */
Olivier Houchardc31e2cb2019-06-24 16:08:08 +02001138 if (si->state >= SI_ST_CON)
1139 si->flags |= SI_FL_ERR;
Willy Tarreau3b9c8502017-10-25 14:22:28 +02001140 goto out_wakeup;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001141 }
1142
1143 /* OK, so now we know that some data might have been sent, and that we may
1144 * have to poll first. We have to do that too if the buffer is not empty.
1145 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001146 if (channel_is_empty(oc)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001147 /* the connection is established but we can't write. Either the
1148 * buffer is empty, or we just refrain from sending because the
1149 * ->o limit was reached. Maybe we just wrote the last
1150 * chunk and need to close.
1151 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001152 if (((oc->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001153 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001154 si_state_in(si->state, SI_SB_RDY|SI_SB_EST)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001155 si_shutw(si);
1156 goto out_wakeup;
1157 }
1158
Willy Tarreauafc8a222014-11-28 15:46:27 +01001159 if ((oc->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +02001160 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001161 oc->wex = TICK_ETERNITY;
Willy Tarreaude5722c2012-08-20 15:01:10 +02001162 }
1163 else {
1164 /* Otherwise there are remaining data to be sent in the buffer,
1165 * which means we have to poll before doing so.
1166 */
Willy Tarreaude5722c2012-08-20 15:01:10 +02001167 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001168 if (!tick_isset(oc->wex))
1169 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001170 }
1171
Willy Tarreauafc8a222014-11-28 15:46:27 +01001172 if (likely(oc->flags & CF_WRITE_ACTIVITY)) {
1173 struct channel *ic = si_ic(si);
1174
Willy Tarreaude5722c2012-08-20 15:01:10 +02001175 /* update timeout if we have written something */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001176 if ((oc->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
1177 !channel_is_empty(oc))
1178 oc->wex = tick_add_ifset(now_ms, oc->wto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001179
Willy Tarreauafc8a222014-11-28 15:46:27 +01001180 if (tick_isset(ic->rex) && !(si->flags & SI_FL_INDEP_STR)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001181 /* Note: to prevent the client from expiring read timeouts
1182 * during writes, we refresh it. We only do this if the
1183 * interface is not configured for "independent streams",
1184 * because for some applications it's better not to do this,
1185 * for instance when continuously exchanging small amounts
1186 * of data which can full the socket buffers long before a
1187 * write timeout is detected.
1188 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001189 ic->rex = tick_add_ifset(now_ms, ic->rto);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001190 }
1191 }
1192
1193 /* in case of special condition (error, shutdown, end of write...), we
1194 * have to notify the task.
1195 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001196 if (likely((oc->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
1197 ((oc->flags & CF_WAKE_WRITE) &&
1198 ((channel_is_empty(oc) && !oc->to_forward) ||
Willy Tarreau7ab22adb2019-06-05 14:53:22 +02001199 !si_state_in(si->state, SI_SB_EST))))) {
Willy Tarreaude5722c2012-08-20 15:01:10 +02001200 out_wakeup:
Willy Tarreau07373b82014-11-28 12:08:47 +01001201 if (!(si->flags & SI_FL_DONT_WAKE))
1202 task_wakeup(si_task(si), TASK_WOKEN_IO);
Willy Tarreaude5722c2012-08-20 15:01:10 +02001203 }
1204}
1205
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001206/*
Willy Tarreauce323de2012-08-20 21:41:06 +02001207 * This is the callback which is called by the connection layer to receive data
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001208 * into the buffer from the connection. It iterates over the mux layer's
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001209 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +02001210 */
Olivier Houchard0e367bb2018-09-14 19:41:13 +02001211int si_cs_recv(struct conn_stream *cs)
Willy Tarreauce323de2012-08-20 21:41:06 +02001212{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001213 struct connection *conn = cs->conn;
1214 struct stream_interface *si = cs->data;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001215 struct channel *ic = si_ic(si);
Olivier Houchardf6535282018-08-31 17:29:12 +02001216 int ret, max, cur_read = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001217 int read_poll = MAX_READ_POLL_LOOPS;
Christopher Fauletc6618d62018-10-11 15:56:04 +02001218 int flags = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001219
Christopher Faulet04400bc2019-10-25 10:21:01 +02001220 /* If not established yet, do nothing. */
1221 if (si->state != SI_ST_EST)
1222 return 0;
1223
Olivier Houchardf6535282018-08-31 17:29:12 +02001224 /* If another call to si_cs_recv() failed, and we subscribed to
1225 * recv events already, give up now.
1226 */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001227 if (si->wait_event.events & SUB_RETRY_RECV)
Olivier Houchardf6535282018-08-31 17:29:12 +02001228 return 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001229
Willy Tarreauce323de2012-08-20 21:41:06 +02001230 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001231 if (ic->flags & CF_SHUTR)
Olivier Houchardc2aa7112018-09-11 18:27:21 +02001232 return 1;
Willy Tarreauce323de2012-08-20 21:41:06 +02001233
Christopher Faulete96993b2020-07-30 09:26:46 +02001234 /* we must wait because the mux is not installed yet */
1235 if (!conn->mux)
1236 return 0;
1237
Willy Tarreau54e917c2017-08-30 07:35:35 +02001238 /* stop here if we reached the end of data */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001239 if (cs->flags & CS_FL_EOS)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001240 goto end_recv;
Willy Tarreau54e917c2017-08-30 07:35:35 +02001241
Christopher Fauletf061e422018-12-07 14:51:20 +01001242 /* stop immediately on errors. Note that we DON'T want to stop on
1243 * POLL_ERR, as the poller might report a write error while there
1244 * are still data available in the recv buffer. This typically
1245 * happens when we send too large a request to a backend server
1246 * which rejects it before reading it all.
1247 */
1248 if (!(cs->flags & CS_FL_RCV_MORE)) {
1249 if (!conn_xprt_ready(conn))
1250 return 0;
1251 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Christopher Faulet36b536d2019-11-20 11:56:33 +01001252 goto end_recv;
Christopher Fauletf061e422018-12-07 14:51:20 +01001253 }
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001254
Willy Tarreau7ab99a32018-12-18 09:15:43 +01001255 /* prepare to detect if the mux needs more room */
1256 cs->flags &= ~CS_FL_WANT_ROOM;
1257
Willy Tarreau77e478c2018-06-19 07:03:14 +02001258 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !co_data(ic) &&
Willy Tarreau7e312732014-02-12 16:35:14 +01001259 global.tune.idle_timer &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001260 (unsigned short)(now_ms - ic->last_read) >= global.tune.idle_timer) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001261 /* The buffer was empty and nothing was transferred for more
1262 * than one second. This was caused by a pause and not by
1263 * congestion. Reset any streaming mode to reduce latency.
1264 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001265 ic->xfer_small = 0;
1266 ic->xfer_large = 0;
1267 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001268 }
1269
Willy Tarreau96199b12012-08-24 00:46:52 +02001270 /* First, let's see if we may splice data across the channel without
1271 * using a buffer.
1272 */
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001273 if (cs->flags & CS_FL_MAY_SPLICE &&
Willy Tarreauafc8a222014-11-28 15:46:27 +01001274 (ic->pipe || ic->to_forward >= MIN_SPLICE_FORWARD) &&
1275 ic->flags & CF_KERN_SPLICING) {
Willy Tarreaud760eec2018-07-10 09:50:25 +02001276 if (c_data(ic)) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001277 /* We're embarrassed, there are already data pending in
1278 * the buffer and we don't want to have them at two
1279 * locations at a time. Let's indicate we need some
1280 * place and ask the consumer to hurry.
1281 */
Christopher Fauletc6618d62018-10-11 15:56:04 +02001282 flags |= CO_RFL_BUF_FLUSH;
Willy Tarreau96199b12012-08-24 00:46:52 +02001283 goto abort_splice;
1284 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001285
Willy Tarreauafc8a222014-11-28 15:46:27 +01001286 if (unlikely(ic->pipe == NULL)) {
1287 if (pipes_used >= global.maxpipes || !(ic->pipe = get_pipe())) {
1288 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001289 goto abort_splice;
1290 }
1291 }
1292
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001293 ret = conn->mux->rcv_pipe(cs, ic->pipe, ic->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +02001294 if (ret < 0) {
1295 /* splice not supported on this end, let's disable it */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001296 ic->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001297 goto abort_splice;
1298 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001299
Willy Tarreau96199b12012-08-24 00:46:52 +02001300 if (ret > 0) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001301 if (ic->to_forward != CHN_INFINITE_FORWARD)
1302 ic->to_forward -= ret;
1303 ic->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +02001304 cur_read += ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001305 ic->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001306 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001307
Christopher Faulet36b536d2019-11-20 11:56:33 +01001308 if (conn->flags & CO_FL_ERROR || cs->flags & (CS_FL_EOS|CS_FL_ERROR))
1309 goto end_recv;
Willy Tarreau96199b12012-08-24 00:46:52 +02001310
Willy Tarreau61d39a02013-07-18 21:49:32 +02001311 if (conn->flags & CO_FL_WAIT_ROOM) {
1312 /* the pipe is full or we have read enough data that it
1313 * could soon be full. Let's stop before needing to poll.
1314 */
Willy Tarreaudb398432018-11-15 11:08:52 +01001315 si_rx_room_blk(si);
Willy Tarreauffb12052018-11-15 16:06:02 +01001316 goto done_recv;
Willy Tarreau61d39a02013-07-18 21:49:32 +02001317 }
Willy Tarreau56a77e52012-09-02 18:34:44 +02001318
Willy Tarreauce323de2012-08-20 21:41:06 +02001319 /* splice not possible (anymore), let's go on on standard copy */
1320 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001321
1322 abort_splice:
Willy Tarreauafc8a222014-11-28 15:46:27 +01001323 if (ic->pipe && unlikely(!ic->pipe->data)) {
1324 put_pipe(ic->pipe);
1325 ic->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +02001326 }
1327
Willy Tarreau17ccd1a2020-01-17 16:19:34 +01001328 if (ic->pipe && ic->to_forward && !(flags & CO_RFL_BUF_FLUSH) && cs->flags & CS_FL_MAY_SPLICE) {
Willy Tarreauc640ef12019-12-03 18:13:04 +01001329 /* don't break splicing by reading, but still call rcv_buf()
1330 * to pass the flag.
1331 */
1332 goto done_recv;
1333 }
1334
Christopher Fauleta73e59b2016-12-09 17:30:18 +01001335 /* now we'll need a input buffer for the stream */
Willy Tarreau581abd32018-10-25 10:21:41 +02001336 if (!si_alloc_ibuf(si, &(si_strm(si)->buffer_wait)))
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001337 goto end_recv;
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001338
Willy Tarreau61d39a02013-07-18 21:49:32 +02001339 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
1340 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
1341 * that if such an event is not handled above in splice, it will be handled here by
1342 * recv().
1343 */
Olivier Houchardc490efd2018-12-04 15:46:16 +01001344 while ((cs->flags & CS_FL_RCV_MORE) ||
Willy Tarreau93c9f592020-01-17 17:24:30 +01001345 (!(conn->flags & (CO_FL_ERROR | CO_FL_HANDSHAKE)) &&
Olivier Houchardc490efd2018-12-04 15:46:16 +01001346 (!(cs->flags & (CS_FL_ERROR|CS_FL_EOS))) && !(ic->flags & CF_SHUTR))) {
Christopher Faulet4eb7d742018-10-11 15:29:21 +02001347 /* <max> may be null. This is the mux responsibility to set
1348 * CS_FL_RCV_MORE on the CS if more space is needed.
1349 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001350 max = channel_recv_max(ic);
Christopher Fauletd8219b32021-04-08 18:13:25 +02001351 flags |= ((!conn_is_back(conn) && (si_strm(si)->be->options & PR_O_ABRT_CLOSE)) ? CO_RFL_KEEP_RECV : 0);
Christopher Faulet9cdd5032019-05-14 22:46:41 +02001352 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 +01001353
Christopher Faulet47545ca2021-09-23 14:17:20 +02001354 if (cs->flags & CS_FL_WANT_ROOM) {
Willy Tarreaudb398432018-11-15 11:08:52 +01001355 si_rx_room_blk(si);
Christopher Faulet47545ca2021-09-23 14:17:20 +02001356 /* Add READ_PARTIAL because some data are pending but
1357 * cannot be xferred to the channel
1358 */
1359 ic->flags |= CF_READ_PARTIAL;
1360 }
Willy Tarreau6577b482017-12-10 21:19:33 +01001361
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001362 if (ret <= 0) {
Willy Tarreau1ac5f202019-12-03 18:08:45 +01001363 /* if we refrained from reading because we asked for a
1364 * flush to satisfy rcv_pipe(), we must not subscribe
1365 * and instead report that there's not enough room
1366 * here to proceed.
1367 */
1368 if (flags & CO_RFL_BUF_FLUSH)
1369 si_rx_room_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001370 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001371 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001372
1373 cur_read += ret;
1374
1375 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001376 if (ic->to_forward && !(ic->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001377 unsigned long fwd = ret;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001378 if (ic->to_forward != CHN_INFINITE_FORWARD) {
1379 if (fwd > ic->to_forward)
1380 fwd = ic->to_forward;
1381 ic->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001382 }
Willy Tarreaubcbd3932018-06-06 07:13:22 +02001383 c_adv(ic, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001384 }
1385
Willy Tarreauafc8a222014-11-28 15:46:27 +01001386 ic->flags |= CF_READ_PARTIAL;
1387 ic->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001388
Christopher Faulet82219b52021-09-09 10:17:59 +02001389 /* End-of-input reached, we can leave. In this case, it is
1390 * important to break the loop to not block the SI because of
1391 * the channel's policies.This way, we are still able to receive
1392 * shutdowns.
1393 */
1394 if (cs->flags & CS_FL_EOI)
1395 break;
1396
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001397 if ((ic->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
1398 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001399 si_rx_chan_blk(si);
Willy Tarreau62dd6982017-11-18 11:26:20 +01001400 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001401 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001402
1403 /* if too many bytes were missing from last read, it means that
1404 * it's pointless trying to read again because the system does
1405 * not have them in buffers.
1406 */
1407 if (ret < max) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001408 /* if a streamer has read few data, it may be because we
1409 * have exhausted system buffers. It's not worth trying
1410 * again.
1411 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001412 if (ic->flags & CF_STREAMER) {
1413 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001414 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001415 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001416 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001417
1418 /* if we read a large block smaller than what we requested,
1419 * it's almost certain we'll never get anything more.
1420 */
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001421 if (ret >= global.tune.recv_enough) {
1422 /* we're stopped by the channel's policy */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001423 si_rx_chan_blk(si);
Willy Tarreauce323de2012-08-20 21:41:06 +02001424 break;
Willy Tarreauf26c26c2018-11-12 16:11:08 +01001425 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001426 }
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001427
1428 /* if we are waiting for more space, don't try to read more data
1429 * right now.
1430 */
Willy Tarreaub26a6f92018-11-14 17:10:36 +01001431 if (si_rx_blocked(si))
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001432 break;
Willy Tarreauce323de2012-08-20 21:41:06 +02001433 } /* while !flags */
1434
Willy Tarreauffb12052018-11-15 16:06:02 +01001435 done_recv:
Willy Tarreauc5890e62014-02-09 17:47:01 +01001436 if (cur_read) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001437 if ((ic->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001438 (cur_read <= ic->buf.size / 2)) {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001439 ic->xfer_large = 0;
1440 ic->xfer_small++;
1441 if (ic->xfer_small >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001442 /* we have read less than half of the buffer in
1443 * one pass, and this happened at least 3 times.
1444 * This is definitely not a streamer.
1445 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001446 ic->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001447 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001448 else if (ic->xfer_small >= 2) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001449 /* if the buffer has been at least half full twice,
1450 * we receive faster than we send, so at least it
1451 * is not a "fast streamer".
1452 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001453 ic->flags &= ~CF_STREAMER_FAST;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001454 }
1455 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001456 else if (!(ic->flags & CF_STREAMER_FAST) &&
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001457 (cur_read >= ic->buf.size - global.tune.maxrewrite)) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001458 /* we read a full buffer at once */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001459 ic->xfer_small = 0;
1460 ic->xfer_large++;
1461 if (ic->xfer_large >= 3) {
Willy Tarreauc5890e62014-02-09 17:47:01 +01001462 /* we call this buffer a fast streamer if it manages
1463 * to be filled in one call 3 consecutive times.
1464 */
Willy Tarreauafc8a222014-11-28 15:46:27 +01001465 ic->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauc5890e62014-02-09 17:47:01 +01001466 }
1467 }
1468 else {
Willy Tarreauafc8a222014-11-28 15:46:27 +01001469 ic->xfer_small = 0;
1470 ic->xfer_large = 0;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001471 }
Willy Tarreauafc8a222014-11-28 15:46:27 +01001472 ic->last_read = now_ms;
Willy Tarreauc5890e62014-02-09 17:47:01 +01001473 }
1474
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001475 end_recv:
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001476 ret = (cur_read != 0);
1477
Christopher Faulet36b536d2019-11-20 11:56:33 +01001478 /* Report EOI on the channel if it was reached from the mux point of
1479 * view. */
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001480 if ((cs->flags & CS_FL_EOI) && !(ic->flags & CF_EOI)) {
Christopher Faulet36b536d2019-11-20 11:56:33 +01001481 ic->flags |= (CF_EOI|CF_READ_PARTIAL);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001482 ret = 1;
1483 }
Willy Tarreau10fc09e2014-11-25 19:46:36 +01001484
Christopher Faulet36b536d2019-11-20 11:56:33 +01001485 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
1486 cs->flags |= CS_FL_ERROR;
1487 si->flags |= SI_FL_ERR;
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001488 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001489 }
1490 else if (cs->flags & CS_FL_EOS) {
Willy Tarreau18955db2020-01-23 16:32:24 +01001491 /* we received a shutdown */
1492 ic->flags |= CF_READ_NULL;
1493 if (ic->flags & CF_AUTO_CLOSE)
1494 channel_shutw_now(ic);
1495 stream_int_read0(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001496 ret = 1;
Christopher Faulet36b536d2019-11-20 11:56:33 +01001497 }
1498 else if (!si_rx_blocked(si)) {
1499 /* Subscribe to receive events if we're blocking on I/O */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001500 conn->mux->subscribe(cs, SUB_RETRY_RECV, &si->wait_event);
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001501 si_rx_endp_done(si);
1502 } else {
1503 si_rx_endp_more(si);
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001504 ret = 1;
Willy Tarreaudd5621a2018-11-15 16:55:14 +01001505 }
Christopher Faulete6d8cb12019-11-20 16:42:00 +01001506 return ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001507}
1508
1509/*
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001510 * This function propagates a null read received on a socket-based connection.
1511 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
Willy Tarreau11405122015-03-12 22:32:27 +01001512 * the close is also forwarded to the write side as an abort.
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001513 */
Willy Tarreau14bfe9a2018-12-19 15:19:27 +01001514static void stream_int_read0(struct stream_interface *si)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001515{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001516 struct conn_stream *cs = __objt_cs(si->end);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001517 struct channel *ic = si_ic(si);
1518 struct channel *oc = si_oc(si);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001519
Willy Tarreauabb5d422018-11-14 16:58:52 +01001520 si_rx_shut_blk(si);
Willy Tarreauafc8a222014-11-28 15:46:27 +01001521 if (ic->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001522 return;
Willy Tarreauafc8a222014-11-28 15:46:27 +01001523 ic->flags |= CF_SHUTR;
1524 ic->rex = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001525
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001526 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001527 return;
1528
Willy Tarreauafc8a222014-11-28 15:46:27 +01001529 if (oc->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001530 goto do_close;
1531
1532 if (si->flags & SI_FL_NOHALF) {
1533 /* we want to immediately forward this close to the write side */
Willy Tarreau87b09662015-04-03 00:22:06 +02001534 /* force flag on ssl to keep stream in cache */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001535 cs_shutw(cs, CS_SHW_SILENT);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001536 goto do_close;
1537 }
1538
1539 /* otherwise that's just a normal read shutdown */
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001540 return;
1541
1542 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001543 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreaua553ae92017-10-05 18:52:17 +02001544 cs_close(cs);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001545
Willy Tarreauafc8a222014-11-28 15:46:27 +01001546 oc->flags &= ~CF_SHUTW_NOW;
1547 oc->flags |= CF_SHUTW;
1548 oc->wex = TICK_ETERNITY;
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001549
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001550 si_done_get(si);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001551
Christopher Faulet04400bc2019-10-25 10:21:01 +02001552 si->state = SI_ST_DIS;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001553 si->exp = TICK_ETERNITY;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001554 return;
1555}
1556
Willy Tarreau651e1822015-09-23 20:06:13 +02001557/* Callback to be used by applet handlers upon completion. It updates the stream
1558 * (which may or may not take this opportunity to try to forward data), then
Emeric Brun2802b072017-06-30 14:11:56 +02001559 * may re-enable the applet's based on the channels and stream interface's final
Willy Tarreau651e1822015-09-23 20:06:13 +02001560 * states.
1561 */
Willy Tarreauaa977ba2015-09-25 11:45:06 +02001562void si_applet_wake_cb(struct stream_interface *si)
Willy Tarreaue5f86492015-04-19 15:16:35 +02001563{
Willy Tarreaueca572f2015-09-25 19:11:55 +02001564 struct channel *ic = si_ic(si);
1565
1566 /* If the applet wants to write and the channel is closed, it's a
1567 * broken pipe and it must be reported.
1568 */
Willy Tarreau05b9b642018-11-14 13:43:35 +01001569 if (!(si->flags & SI_FL_RX_WAIT_EP) && (ic->flags & CF_SHUTR))
Willy Tarreaueca572f2015-09-25 19:11:55 +02001570 si->flags |= SI_FL_ERR;
1571
Willy Tarreau186dcdd2018-11-16 16:18:34 +01001572 /* automatically mark the applet having data available if it reported
1573 * begin blocked by the channel.
1574 */
1575 if (si_rx_blocked(si))
1576 si_rx_endp_more(si);
1577
Willy Tarreau651e1822015-09-23 20:06:13 +02001578 /* update the stream-int, channels, and possibly wake the stream up */
1579 stream_int_notify(si);
Willy Tarreaua64c7032019-08-01 14:17:02 +02001580 stream_release_buffers(si_strm(si));
Willy Tarreaue5f86492015-04-19 15:16:35 +02001581
Willy Tarreau32742fd2018-11-14 14:07:59 +01001582 /* stream_int_notify may have passed through chk_snd and released some
1583 * RXBLK flags. Process_stream will consider those flags to wake up the
1584 * appctx but in the case the task is not in runqueue we may have to
1585 * wakeup the appctx immediately.
Emeric Brun2802b072017-06-30 14:11:56 +02001586 */
Olivier Houchard51205a12019-04-17 19:29:35 +02001587 if ((si_rx_endp_ready(si) && !si_rx_blocked(si)) ||
1588 (si_tx_endp_ready(si) && !si_tx_blocked(si)))
Willy Tarreau563cc372015-04-19 18:13:56 +02001589 appctx_wakeup(si_appctx(si));
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001590}
1591
1592/*
1593 * This function performs a shutdown-read on a stream interface attached to an
1594 * applet in a connected or init state (it does nothing for other states). It
1595 * either shuts the read side or marks itself as closed. The buffer flags are
1596 * updated to reflect the new state. If the stream interface has SI_FL_NOHALF,
1597 * we also forward the close to the write side. The owner task is woken up if
1598 * it exists.
1599 */
1600static void stream_int_shutr_applet(struct stream_interface *si)
1601{
1602 struct channel *ic = si_ic(si);
1603
Willy Tarreauabb5d422018-11-14 16:58:52 +01001604 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001605 if (ic->flags & CF_SHUTR)
1606 return;
1607 ic->flags |= CF_SHUTR;
1608 ic->rex = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001609
Willy Tarreau828824a2015-04-19 17:20:03 +02001610 /* Note: on shutr, we don't call the applet */
1611
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001612 if (!si_state_in(si->state, SI_SB_CON|SI_SB_RDY|SI_SB_EST))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001613 return;
1614
1615 if (si_oc(si)->flags & CF_SHUTW) {
Willy Tarreau958f0742015-09-25 20:24:26 +02001616 si_applet_release(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001617 si->state = SI_ST_DIS;
1618 si->exp = TICK_ETERNITY;
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001619 }
1620 else if (si->flags & SI_FL_NOHALF) {
1621 /* we want to immediately forward this close to the write side */
1622 return stream_int_shutw_applet(si);
1623 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001624}
1625
1626/*
1627 * This function performs a shutdown-write on a stream interface attached to an
1628 * applet in a connected or init state (it does nothing for other states). It
1629 * either shuts the write side or marks itself as closed. The buffer flags are
1630 * updated to reflect the new state. It does also close everything if the SI
1631 * was marked as being in error state. The owner task is woken up if it exists.
1632 */
1633static void stream_int_shutw_applet(struct stream_interface *si)
1634{
1635 struct channel *ic = si_ic(si);
1636 struct channel *oc = si_oc(si);
1637
1638 oc->flags &= ~CF_SHUTW_NOW;
1639 if (oc->flags & CF_SHUTW)
1640 return;
1641 oc->flags |= CF_SHUTW;
1642 oc->wex = TICK_ETERNITY;
Willy Tarreau43e69dc2018-11-06 19:23:03 +01001643 si_done_get(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001644
Hongbo Longe39683c2017-03-10 18:41:51 +01001645 if (tick_isset(si->hcto)) {
1646 ic->rto = si->hcto;
1647 ic->rex = tick_add(now_ms, ic->rto);
1648 }
1649
Willy Tarreau828824a2015-04-19 17:20:03 +02001650 /* on shutw we always wake the applet up */
1651 appctx_wakeup(si_appctx(si));
1652
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001653 switch (si->state) {
Willy Tarreau4f283fa2019-06-05 14:34:03 +02001654 case SI_ST_RDY:
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001655 case SI_ST_EST:
1656 /* we have to shut before closing, otherwise some short messages
1657 * may never leave the system, especially when there are remaining
1658 * unread data in the socket input buffer, or when nolinger is set.
1659 * However, if SI_FL_NOLINGER is explicitly set, we know there is
1660 * no risk so we close both sides immediately.
1661 */
1662 if (!(si->flags & (SI_FL_ERR | SI_FL_NOLINGER)) &&
1663 !(ic->flags & (CF_SHUTR|CF_DONT_READ)))
1664 return;
1665
1666 /* fall through */
1667 case SI_ST_CON:
1668 case SI_ST_CER:
1669 case SI_ST_QUE:
1670 case SI_ST_TAR:
1671 /* Note that none of these states may happen with applets */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001672 si_applet_release(si);
Willy Tarreau958f0742015-09-25 20:24:26 +02001673 si->state = SI_ST_DIS;
Tim Duesterhus588b3142020-05-29 14:35:51 +02001674 /* fall through */
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001675 default:
Willy Tarreauabb5d422018-11-14 16:58:52 +01001676 si->flags &= ~SI_FL_NOLINGER;
1677 si_rx_shut_blk(si);
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001678 ic->flags |= CF_SHUTR;
1679 ic->rex = TICK_ETERNITY;
1680 si->exp = TICK_ETERNITY;
1681 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001682}
1683
1684/* chk_rcv function for applets */
1685static void stream_int_chk_rcv_applet(struct stream_interface *si)
1686{
1687 struct channel *ic = si_ic(si);
1688
1689 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1690 __FUNCTION__,
1691 si, si->state, ic->flags, si_oc(si)->flags);
1692
Christopher Fauletb3e0de42018-10-11 13:54:13 +02001693 if (!ic->pipe) {
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001694 /* (re)start reading */
Willy Tarreau828824a2015-04-19 17:20:03 +02001695 appctx_wakeup(si_appctx(si));
Thierry FOURNIER5bc2cbf2015-09-04 18:40:36 +02001696 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001697}
1698
1699/* chk_snd function for applets */
1700static void stream_int_chk_snd_applet(struct stream_interface *si)
1701{
1702 struct channel *oc = si_oc(si);
1703
1704 DPRINTF(stderr, "%s: si=%p, si->state=%d ic->flags=%08x oc->flags=%08x\n",
1705 __FUNCTION__,
1706 si, si->state, si_ic(si)->flags, oc->flags);
1707
1708 if (unlikely(si->state != SI_ST_EST || (oc->flags & CF_SHUTW)))
1709 return;
1710
Willy Tarreau828824a2015-04-19 17:20:03 +02001711 /* we only wake the applet up if it was waiting for some data */
1712
1713 if (!(si->flags & SI_FL_WAIT_DATA))
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001714 return;
1715
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001716 if (!tick_isset(oc->wex))
1717 oc->wex = tick_add_ifset(now_ms, oc->wto);
1718
Willy Tarreau828824a2015-04-19 17:20:03 +02001719 if (!channel_is_empty(oc)) {
1720 /* (re)start sending */
1721 appctx_wakeup(si_appctx(si));
1722 }
Willy Tarreaud45b9f82015-04-13 16:30:14 +02001723}
1724
Willy Tarreaudded32d2008-11-30 19:48:07 +01001725/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001726 * Local variables:
1727 * c-indent-level: 8
1728 * c-basic-offset: 8
1729 * End:
1730 */