blob: 9f0c26ad10bdeed9f02d7fba903ab934b7da9ec5 [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
22#include <common/compat.h>
23#include <common/config.h>
24#include <common/debug.h>
25#include <common/standard.h>
26#include <common/ticks.h>
27#include <common/time.h>
28
Willy Tarreauc7e42382012-08-24 19:22:53 +020029#include <proto/channel.h>
Willy Tarreau8b117082012-08-06 15:06:49 +020030#include <proto/connection.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010031#include <proto/fd.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020032#include <proto/pipe.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020033#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010034#include <proto/task.h>
35
Willy Tarreaufd31e532012-07-23 18:24:25 +020036#include <types/pipe.h>
37
Willy Tarreauf873d752012-05-11 17:47:17 +020038/* socket functions used when running a stream interface as a task */
Willy Tarreauf873d752012-05-11 17:47:17 +020039static void stream_int_update_embedded(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020040static void stream_int_chk_rcv(struct stream_interface *si);
41static void stream_int_chk_snd(struct stream_interface *si);
Willy Tarreauc5788912012-08-24 18:12:41 +020042static void stream_int_update_conn(struct stream_interface *si);
43static void stream_int_chk_rcv_conn(struct stream_interface *si);
44static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreau4aa36832012-10-02 20:07:22 +020045static void si_conn_recv_cb(struct connection *conn);
46static void si_conn_send_cb(struct connection *conn);
Willy Tarreau2396c1c2012-10-03 21:12:16 +020047static int si_conn_wake_cb(struct connection *conn);
Willy Tarreauf873d752012-05-11 17:47:17 +020048
Willy Tarreauc5788912012-08-24 18:12:41 +020049/* stream-interface operations for embedded tasks */
50struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020051 .update = stream_int_update_embedded,
Willy Tarreau5c979a92012-05-07 17:15:39 +020052 .chk_rcv = stream_int_chk_rcv,
53 .chk_snd = stream_int_chk_snd,
Willy Tarreau5c979a92012-05-07 17:15:39 +020054};
55
Willy Tarreauc5788912012-08-24 18:12:41 +020056/* stream-interface operations for connections */
57struct si_ops si_conn_ops = {
58 .update = stream_int_update_conn,
59 .chk_rcv = stream_int_chk_rcv_conn,
60 .chk_snd = stream_int_chk_snd_conn,
61};
62
Willy Tarreau74beec32012-10-03 00:41:04 +020063struct data_cb si_conn_cb = {
Willy Tarreauc5788912012-08-24 18:12:41 +020064 .recv = si_conn_recv_cb,
65 .send = si_conn_send_cb,
Willy Tarreau4aa36832012-10-02 20:07:22 +020066 .wake = si_conn_wake_cb,
Willy Tarreauc5788912012-08-24 18:12:41 +020067};
68
Willy Tarreaucff64112008-11-03 06:26:53 +010069/*
70 * This function only has to be called once after a wakeup event in case of
71 * suspected timeout. It controls the stream interface timeouts and sets
72 * si->flags accordingly. It does NOT close anything, as this timeout may
73 * be used for any purpose. It returns 1 if the timeout fired, otherwise
74 * zero.
75 */
76int stream_int_check_timeouts(struct stream_interface *si)
77{
78 if (tick_is_expired(si->exp, now_ms)) {
79 si->flags |= SI_FL_EXP;
80 return 1;
81 }
82 return 0;
83}
84
Willy Tarreaufe3718a2008-11-30 18:14:12 +010085/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +010086void stream_int_report_error(struct stream_interface *si)
87{
88 if (!si->err_type)
89 si->err_type = SI_ET_DATA_ERR;
90
Willy Tarreau03cdb7c2012-08-27 23:14:58 +020091 si->ob->flags |= CF_WRITE_ERROR;
92 si->ib->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +010093}
94
95/*
Willy Tarreaudded32d2008-11-30 19:48:07 +010096 * Returns a message to the client ; the connection is shut down for read,
97 * and the request is cleared so that no server connection can be initiated.
98 * The buffer is marked for read shutdown on the other side to protect the
99 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100100 * "chunk". If it is null, then an empty message is used. The reply buffer does
101 * not need to be empty before this, and its contents will not be overwritten.
102 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100103 */
104void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
105{
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200106 channel_auto_read(si->ib);
107 channel_abort(si->ib);
108 channel_auto_close(si->ib);
109 channel_erase(si->ib);
Willy Tarreau798e1282010-12-12 13:06:00 +0100110
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200111 bi_erase(si->ob);
Willy Tarreau148d0992010-01-10 10:21:21 +0100112 if (likely(msg && msg->len))
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200113 bo_inject(si->ob, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100114
115 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200116 channel_auto_read(si->ob);
117 channel_auto_close(si->ob);
118 channel_shutr_now(si->ob);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100119}
120
Willy Tarreaufb90d942009-09-05 20:57:35 +0200121/* default update function for embedded tasks, to be used at the end of the i/o handler */
Willy Tarreauf873d752012-05-11 17:47:17 +0200122static void stream_int_update_embedded(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200123{
Willy Tarreau3488e252010-08-09 16:24:56 +0200124 int old_flags = si->flags;
125
Willy Tarreaufb90d942009-09-05 20:57:35 +0200126 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
127 __FUNCTION__,
128 si, si->state, si->ib->flags, si->ob->flags);
129
130 if (si->state != SI_ST_EST)
131 return;
132
Willy Tarreaub31c9712012-11-11 23:05:39 +0100133 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200134 channel_is_empty(si->ob))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200135 si_shutw(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200136
Willy Tarreaub31c9712012-11-11 23:05:39 +0100137 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200138 si->flags |= SI_FL_WAIT_DATA;
139
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200140 /* we're almost sure that we need some space if the buffer is not
141 * empty, even if it's not full, because the applets can't fill it.
142 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200143 if ((si->ib->flags & (CF_SHUTR|CF_DONT_READ)) == 0 && !channel_is_empty(si->ib))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200144 si->flags |= SI_FL_WAIT_ROOM;
145
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200146 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200147 if (tick_isset(si->ob->wex))
148 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
149 }
150
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200151 if (si->ib->flags & CF_READ_ACTIVITY ||
152 (si->ob->flags & CF_WRITE_ACTIVITY && !(si->flags & SI_FL_INDEP_STR))) {
Willy Tarreauf27b5ea2009-10-03 22:01:18 +0200153 if (tick_isset(si->ib->rex))
154 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
155 }
156
Willy Tarreau3488e252010-08-09 16:24:56 +0200157 /* save flags to detect changes */
158 old_flags = si->flags;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200159 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200160 !channel_full(si->ob) &&
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200161 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200162 si_chk_rcv(si->ob->prod);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200163
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200164 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200165 (si->ib->cons->flags & SI_FL_WAIT_DATA)) {
Willy Tarreau73b013b2012-05-21 16:31:45 +0200166 si_chk_snd(si->ib->cons);
Willy Tarreau3488e252010-08-09 16:24:56 +0200167 /* check if the consumer has freed some space */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200168 if (!channel_full(si->ib))
Willy Tarreau3488e252010-08-09 16:24:56 +0200169 si->flags &= ~SI_FL_WAIT_ROOM;
170 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200171
172 /* Note that we're trying to wake up in two conditions here :
173 * - special event, which needs the holder task attention
174 * - status indicating that the applet can go on working. This
175 * is rather hard because we might be blocking on output and
176 * don't want to wake up on input and vice-versa. The idea is
Willy Tarreau3488e252010-08-09 16:24:56 +0200177 * to only rely on the changes the chk_* might have performed.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200178 */
179 if (/* check stream interface changes */
Willy Tarreau3488e252010-08-09 16:24:56 +0200180 ((old_flags & ~si->flags) & (SI_FL_WAIT_ROOM|SI_FL_WAIT_DATA)) ||
181
182 /* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200183 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200184 si->state != SI_ST_EST ||
185 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200186 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200187 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
188
189 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200190 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
191 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
192 ((si->ob->flags & CF_SHUTW) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200193 si->ob->prod->state != SI_ST_EST ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200194 (channel_is_empty(si->ob) && !si->ob->to_forward)))) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200195 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
196 task_wakeup(si->owner, TASK_WOKEN_IO);
197 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200198 if (si->ib->flags & CF_READ_ACTIVITY)
199 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200200}
201
Willy Tarreau4a36b562012-08-06 19:31:45 +0200202/*
203 * This function performs a shutdown-read on a stream interface in a connected
204 * or init state (it does nothing for other states). It either shuts the read
205 * side or marks itself as closed. The buffer flags are updated to reflect the
206 * new state. If the stream interface has SI_FL_NOHALF, we also forward the
207 * close to the write side. If a control layer is defined, then it is supposed
208 * to be a socket layer and file descriptors are then shutdown or closed
209 * accordingly. If no control layer is defined, then the SI is supposed to be
210 * an embedded one and the owner task is woken up if it exists. The function
211 * does not disable polling on the FD by itself, it returns non-zero instead
212 * if the caller needs to do so (except when the FD is deleted where this is
213 * implicit).
214 */
215int stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200216{
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200217 struct connection *conn = si->conn;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200218
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200219 si->ib->flags &= ~CF_SHUTR_NOW;
220 if (si->ib->flags & CF_SHUTR)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200221 return 0;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200222 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200223 si->ib->rex = TICK_ETERNITY;
224 si->flags &= ~SI_FL_WAIT_ROOM;
225
226 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200227 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200228
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200229 if (si->ob->flags & CF_SHUTW) {
Willy Tarreau2b199c92012-11-23 17:32:21 +0100230 conn_full_close(si->conn);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200231 si->state = SI_ST_DIS;
232 si->exp = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200233
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200234 if (si->release)
235 si->release(si);
236 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200237 else if (si->flags & SI_FL_NOHALF) {
238 /* we want to immediately forward this close to the write side */
239 return stream_int_shutw(si);
240 }
241 else if (conn->ctrl) {
242 /* we want the caller to disable polling on this FD */
243 return 1;
244 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200245
Willy Tarreau4a36b562012-08-06 19:31:45 +0200246 /* note that if the task exists, it must unregister itself once it runs */
247 if (!conn->ctrl && !(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200248 task_wakeup(si->owner, TASK_WOKEN_IO);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200249 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200250}
251
Willy Tarreau4a36b562012-08-06 19:31:45 +0200252/*
253 * This function performs a shutdown-write on a stream interface in a connected or
254 * init state (it does nothing for other states). It either shuts the write side
255 * or marks itself as closed. The buffer flags are updated to reflect the new state.
256 * It does also close everything if the SI was marked as being in error state. If
257 * there is a data-layer shutdown, it is called. If a control layer is defined, then
258 * it is supposed to be a socket layer and file descriptors are then shutdown or
259 * closed accordingly. If no control layer is defined, then the SI is supposed to
260 * be an embedded one and the owner task is woken up if it exists. The function
261 * does not disable polling on the FD by itself, it returns non-zero instead if
262 * the caller needs to do so (except when the FD is deleted where this is implicit).
263 */
264int stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200265{
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200266 struct connection *conn = si->conn;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200267
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200268 si->ob->flags &= ~CF_SHUTW_NOW;
269 if (si->ob->flags & CF_SHUTW)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200270 return 0;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200271 si->ob->flags |= CF_SHUTW;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200272 si->ob->wex = TICK_ETERNITY;
273 si->flags &= ~SI_FL_WAIT_DATA;
274
275 switch (si->state) {
276 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200277 /* we have to shut before closing, otherwise some short messages
278 * may never leave the system, especially when there are remaining
279 * unread data in the socket input buffer, or when nolinger is set.
280 * However, if SI_FL_NOLINGER is explicitly set, we know there is
281 * no risk so we close both sides immediately.
282 */
283 if (si->flags & SI_FL_ERR) {
284 /* quick close, the socket is already shut. Remove pending flags. */
285 si->flags &= ~SI_FL_NOLINGER;
286 } else if (si->flags & SI_FL_NOLINGER) {
287 si->flags &= ~SI_FL_NOLINGER;
288 if (conn->ctrl) {
Willy Tarreau7f7ad912012-11-11 19:27:15 +0100289 setsockopt(si->conn->t.sock.fd, SOL_SOCKET, SO_LINGER,
Willy Tarreau4a36b562012-08-06 19:31:45 +0200290 (struct linger *) &nolinger, sizeof(struct linger));
291 }
292 /* unclean data-layer shutdown */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200293 if (conn->xprt && conn->xprt->shutw)
294 conn->xprt->shutw(conn, 0);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200295 } else {
296 /* clean data-layer shutdown */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200297 if (conn->xprt && conn->xprt->shutw)
298 conn->xprt->shutw(conn, 1);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200299
Willy Tarreau9568d712012-12-30 01:39:37 +0100300 /* If the stream interface is configured to disable half-open
301 * connections, we'll skip the shutdown(), but only if the
302 * read size is already closed. Otherwise we can't support
303 * closed write with pending read (eg: abortonclose while
304 * waiting for the server).
305 */
306 if (!(si->flags & SI_FL_NOHALF) || !(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau4a36b562012-08-06 19:31:45 +0200307 /* We shutdown transport layer */
308 if (conn->ctrl)
Willy Tarreau7f7ad912012-11-11 19:27:15 +0100309 shutdown(si->conn->t.sock.fd, SHUT_WR);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200310
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200311 if (!(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau4a36b562012-08-06 19:31:45 +0200312 /* OK just a shutw, but we want the caller
313 * to disable polling on this FD if exists.
314 */
315 return !!conn->ctrl;
316 }
317 }
318 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200319
320 /* fall through */
321 case SI_ST_CON:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200322 /* we may have to close a pending connection, and mark the
323 * response buffer as shutr
324 */
Willy Tarreau2b199c92012-11-23 17:32:21 +0100325 conn_full_close(si->conn);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200326 /* fall through */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200327 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100328 case SI_ST_QUE:
329 case SI_ST_TAR:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200330 si->state = SI_ST_DIS;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200331
332 if (si->release)
333 si->release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200334 default:
335 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200336 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200337 si->ib->rex = TICK_ETERNITY;
338 si->exp = TICK_ETERNITY;
339 }
340
Willy Tarreau4a36b562012-08-06 19:31:45 +0200341 /* note that if the task exists, it must unregister itself once it runs */
342 if (!conn->ctrl && !(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200343 task_wakeup(si->owner, TASK_WOKEN_IO);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200344 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200345}
346
347/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200348static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200349{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200350 struct channel *ib = si->ib;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200351
352 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
353 __FUNCTION__,
354 si, si->state, si->ib->flags, si->ob->flags);
355
Willy Tarreaub31c9712012-11-11 23:05:39 +0100356 if (unlikely(si->state != SI_ST_EST || (ib->flags & (CF_SHUTR|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200357 return;
358
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200359 if (channel_full(ib)) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200360 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200361 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200362 }
363 else {
364 /* (re)start reading */
365 si->flags &= ~SI_FL_WAIT_ROOM;
366 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
367 task_wakeup(si->owner, TASK_WOKEN_IO);
368 }
369}
370
371/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200372static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200373{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200374 struct channel *ob = si->ob;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200375
376 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
377 __FUNCTION__,
378 si, si->state, si->ib->flags, si->ob->flags);
379
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200380 if (unlikely(si->state != SI_ST_EST || (si->ob->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200381 return;
382
383 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200384 channel_is_empty(ob)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200385 return;
386
387 /* Otherwise there are remaining data to be sent in the buffer,
388 * so we tell the handler.
389 */
390 si->flags &= ~SI_FL_WAIT_DATA;
391 if (!tick_isset(ob->wex))
392 ob->wex = tick_add_ifset(now_ms, ob->wto);
393
394 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
395 task_wakeup(si->owner, TASK_WOKEN_IO);
396}
397
Willy Tarreaub24281b2011-02-13 13:16:36 +0100398/* Register an applet to handle a stream_interface as part of the stream
Willy Tarreaufb90d942009-09-05 20:57:35 +0200399 * interface's owner task, which is returned. The SI will wake it up everytime
Willy Tarreaub24281b2011-02-13 13:16:36 +0100400 * it is solicited. The task's processing function must call the applet's
Willy Tarreaufb90d942009-09-05 20:57:35 +0200401 * function before returning. It must be deleted by the task handler using
Willy Tarreaub24281b2011-02-13 13:16:36 +0100402 * stream_int_unregister_handler(), possibly from within the function itself.
Willy Tarreaufa6bac62012-05-31 14:16:59 +0200403 * It also pre-initializes applet.state to zero and the connection context
404 * to NULL.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200405 */
Willy Tarreaub24281b2011-02-13 13:16:36 +0100406struct task *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200407{
Simon Horman7abd00d2011-08-13 08:03:51 +0900408 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si->owner);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200409
Willy Tarreauc5788912012-08-24 18:12:41 +0200410 si_prepare_embedded(si);
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100411 si->conn->target = &app->obj_type;
Willy Tarreau128b03c2012-11-11 23:14:16 +0100412 si->release = app->release;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200413 si->flags |= SI_FL_WAIT_DATA;
414 return si->owner;
415}
416
Willy Tarreaufb90d942009-09-05 20:57:35 +0200417/* Unregister a stream interface handler. This must be called by the handler task
Willy Tarreau128b03c2012-11-11 23:14:16 +0100418 * itself when it detects that it is in the SI_ST_DIS state.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200419 */
420void stream_int_unregister_handler(struct stream_interface *si)
421{
Willy Tarreau128b03c2012-11-11 23:14:16 +0100422 si->release = NULL;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200423 si->owner = NULL;
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100424 si->conn->target = NULL;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200425}
426
Willy Tarreau2c6be842012-07-06 17:12:34 +0200427/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200428 * established. It returns 0 if it fails in a fatal way or needs to poll to go
429 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200430 * flags (the bit is provided in <flag> by the caller). It is designed to be
431 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200432 */
433int conn_si_send_proxy(struct connection *conn, unsigned int flag)
434{
Willy Tarreaue603e692012-09-27 22:20:41 +0200435 struct stream_interface *si = conn->owner;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200436
437 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200438 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200439 goto out_error;
440
441 /* If we have a PROXY line to send, we'll use this to validate the
442 * connection, in which case the connection is validated only once
443 * we've sent the whole proxy line. Otherwise we use connect().
444 */
445 if (si->send_proxy_ofs) {
446 int ret;
447
448 /* The target server expects a PROXY line to be sent first.
449 * If the send_proxy_ofs is negative, it corresponds to the
450 * offset to start sending from then end of the proxy string
451 * (which is recomputed every time since it's constant). If
452 * it is positive, it means we have to send from the start.
453 */
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100454 ret = make_proxy_line(trash.str, trash.size, &si->ob->prod->conn->addr.from, &si->ob->prod->conn->addr.to);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200455 if (!ret)
456 goto out_error;
457
458 if (si->send_proxy_ofs > 0)
459 si->send_proxy_ofs = -ret; /* first call */
460
Willy Tarreaua1a74742012-08-24 12:14:49 +0200461 /* we have to send trash from (ret+sp for -sp bytes). If the
462 * data layer has a pending write, we'll also set MSG_MORE.
463 */
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100464 ret = send(conn->t.sock.fd, trash.str + ret + si->send_proxy_ofs, -si->send_proxy_ofs,
Willy Tarreaua1a74742012-08-24 12:14:49 +0200465 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200466
467 if (ret == 0)
468 goto out_wait;
469
470 if (ret < 0) {
Willy Tarreau95742a42013-09-03 09:02:11 +0200471 if (errno == EAGAIN || errno == ENOTCONN)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200472 goto out_wait;
473 goto out_error;
474 }
475
476 si->send_proxy_ofs += ret; /* becomes zero once complete */
477 if (si->send_proxy_ofs != 0)
478 goto out_wait;
479
480 /* OK we've sent the whole line, we're connected */
481 }
482
Willy Tarreaua1a74742012-08-24 12:14:49 +0200483 /* The connection is ready now, simply return and let the connection
484 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200485 */
486 if (conn->flags & CO_FL_WAIT_L4_CONN)
487 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200488 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200489 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200490
491 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200492 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200493 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200494 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200495
496 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200497 __conn_sock_stop_recv(conn);
498 __conn_sock_poll_send(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200499 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200500}
501
Willy Tarreau100c4672012-08-20 12:06:26 +0200502/* Callback to be used by connection I/O handlers upon completion. It differs from
Willy Tarreau4aa36832012-10-02 20:07:22 +0200503 * the update function in that it is designed to be called by lower layers after I/O
Willy Tarreau100c4672012-08-20 12:06:26 +0200504 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200505 * an important event requires special handling. It relies on the connection handler
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200506 * to commit any polling updates. The function always returns 0.
Willy Tarreau100c4672012-08-20 12:06:26 +0200507 */
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200508static int si_conn_wake_cb(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200509{
Willy Tarreaue603e692012-09-27 22:20:41 +0200510 struct stream_interface *si = conn->owner;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200511
512 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
513 __FUNCTION__,
514 si, si->state, si->ib->flags, si->ob->flags);
515
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200516 if (conn->flags & CO_FL_ERROR)
517 si->flags |= SI_FL_ERR;
518
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200519 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200520 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200521 si->exp = TICK_ETERNITY;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200522 si->ob->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200523 }
524
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200525 /* process consumer side */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200526 if (channel_is_empty(si->ob)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100527 if (((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200528 (si->state == SI_ST_EST))
529 stream_int_shutw(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200530 __conn_data_stop_send(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200531 si->ob->wex = TICK_ETERNITY;
532 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200533
Willy Tarreaub31c9712012-11-11 23:05:39 +0100534 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0 && !channel_full(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200535 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200536
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200537 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200538 /* update timeouts if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200539 if ((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200540 !channel_is_empty(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200541 if (tick_isset(si->ob->wex))
542 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200543
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200544 if (!(si->flags & SI_FL_INDEP_STR))
545 if (tick_isset(si->ib->rex))
546 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200547
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200548 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200549 !channel_full(si->ob) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200550 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
551 si_chk_rcv(si->ob->prod);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200552 }
553
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200554 /* process producer side.
555 * We might have some data the consumer is waiting for.
556 * We can do fast-forwarding, but we avoid doing this for partial
557 * buffers, because it is very likely that it will be done again
558 * immediately afterwards once the following data is parsed (eg:
559 * HTTP chunking).
560 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200561 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200562 (si->ib->pipe /* always try to send spliced data */ ||
Willy Tarreau9b28e032012-10-12 23:49:43 +0200563 (si->ib->buf->i == 0 && (si->ib->cons->flags & SI_FL_WAIT_DATA)))) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200564 int last_len = si->ib->pipe ? si->ib->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200565
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200566 si_chk_snd(si->ib->cons);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200567
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200568 /* check if the consumer has freed some space either in the
569 * buffer or in the pipe.
570 */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200571 if (!channel_full(si->ib) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200572 (!last_len || !si->ib->pipe || si->ib->pipe->data < last_len))
573 si->flags &= ~SI_FL_WAIT_ROOM;
574 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200575
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200576 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200577 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200578 si->ib->rex = TICK_ETERNITY;
579 }
Willy Tarreau66572762012-12-19 17:34:17 +0100580 else if ((si->ib->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ)) == CF_READ_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200581 !channel_full(si->ib)) {
Willy Tarreau9f7c6a12012-11-19 16:43:14 +0100582 /* we must re-enable reading if si_chk_snd() has freed some space */
583 __conn_data_want_recv(conn);
Willy Tarreau66572762012-12-19 17:34:17 +0100584 if (!(si->ib->flags & CF_READ_NOEXP) && tick_isset(si->ib->rex))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200585 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200586 }
587
588 /* wake the task up only when needed */
589 if (/* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200590 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200591 si->state != SI_ST_EST ||
592 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200593 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreaufd31e532012-07-23 18:24:25 +0200594 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
595
596 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200597 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
598 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
599 ((si->ob->flags & CF_SHUTW) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200600 si->ob->prod->state != SI_ST_EST ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200601 (channel_is_empty(si->ob) && !si->ob->to_forward)))) {
Willy Tarreaufd31e532012-07-23 18:24:25 +0200602 task_wakeup(si->owner, TASK_WOKEN_IO);
603 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200604 if (si->ib->flags & CF_READ_ACTIVITY)
605 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200606 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200607}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200608
Willy Tarreau5368d802012-08-21 18:22:06 +0200609/*
610 * This function is called to send buffer data to a stream socket.
611 * It returns -1 in case of unrecoverable error, otherwise zero.
Godbache68e02d2013-10-11 15:48:29 +0800612 * It calls the transport layer's snd_buf function. It relies on the
Willy Tarreauf16723e2012-08-24 12:52:22 +0200613 * caller to commit polling changes.
Willy Tarreau5368d802012-08-21 18:22:06 +0200614 */
Godbache68e02d2013-10-11 15:48:29 +0800615static int si_conn_send(struct connection *conn)
Willy Tarreau5368d802012-08-21 18:22:06 +0200616{
Willy Tarreaue603e692012-09-27 22:20:41 +0200617 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200618 struct channel *chn = si->ob;
Willy Tarreau5368d802012-08-21 18:22:06 +0200619 int ret;
620
Willy Tarreaucb76e592012-10-12 23:56:57 +0200621 if (chn->pipe && conn->xprt->snd_pipe) {
622 ret = conn->xprt->snd_pipe(conn, chn->pipe);
Willy Tarreau96199b12012-08-24 00:46:52 +0200623 if (ret > 0)
Willy Tarreaucb76e592012-10-12 23:56:57 +0200624 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200625
Willy Tarreaucb76e592012-10-12 23:56:57 +0200626 if (!chn->pipe->data) {
627 put_pipe(chn->pipe);
628 chn->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200629 }
630
Willy Tarreau96199b12012-08-24 00:46:52 +0200631 if (conn->flags & CO_FL_ERROR)
632 return -1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200633 }
634
635 /* At this point, the pipe is empty, but we may still have data pending
636 * in the normal buffer.
637 */
Willy Tarreau9b28e032012-10-12 23:49:43 +0200638 if (!chn->buf->o)
Willy Tarreau5368d802012-08-21 18:22:06 +0200639 return 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200640
Godbache68e02d2013-10-11 15:48:29 +0800641 /* when we're here, we already know that there is no spliced
Willy Tarreau5368d802012-08-21 18:22:06 +0200642 * data left, and that there are sendable buffered data.
643 */
Godbache68e02d2013-10-11 15:48:29 +0800644 if (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH | CO_FL_WAIT_DATA | CO_FL_WAIT_WR | CO_FL_HANDSHAKE))) {
Willy Tarreau5368d802012-08-21 18:22:06 +0200645 /* check if we want to inform the kernel that we're interested in
646 * sending more data after this call. We want this if :
647 * - we're about to close after this last send and want to merge
648 * the ongoing FIN with the last segment.
649 * - we know we can't send everything at once and must get back
650 * here because of unaligned data
651 * - there is still a finite amount of data to forward
652 * The test is arranged so that the most common case does only 2
653 * tests.
654 */
655 unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
656
Willy Tarreaucb76e592012-10-12 23:56:57 +0200657 if ((!(chn->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
658 ((chn->to_forward && chn->to_forward != CHN_INFINITE_FORWARD) ||
659 (chn->flags & CF_EXPECT_MORE))) ||
Willy Tarreaub31c9712012-11-11 23:05:39 +0100660 ((chn->flags & (CF_SHUTW|CF_SHUTW_NOW)) == CF_SHUTW_NOW))
Willy Tarreau5368d802012-08-21 18:22:06 +0200661 send_flag |= MSG_MORE;
662
Willy Tarreau9b28e032012-10-12 23:49:43 +0200663 ret = conn->xprt->snd_buf(conn, chn->buf, send_flag);
Godbache68e02d2013-10-11 15:48:29 +0800664 if (ret > 0) {
665 chn->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200666
Godbache68e02d2013-10-11 15:48:29 +0800667 if (!chn->buf->o) {
668 /* Always clear both flags once everything has been sent, they're one-shot */
669 chn->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
670 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200671
Godbache68e02d2013-10-11 15:48:29 +0800672 /* if some data remain in the buffer, it's only because the
673 * system buffers are full, we will try next time.
674 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200675 }
Godbache68e02d2013-10-11 15:48:29 +0800676 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200677
678 if (conn->flags & CO_FL_ERROR)
679 return -1;
680
Willy Tarreau5368d802012-08-21 18:22:06 +0200681 return 0;
682}
683
684
Willy Tarreau100c4672012-08-20 12:06:26 +0200685/* Updates the timers and flags of a stream interface attached to a connection,
686 * depending on the buffers' flags. It should only be called once after the
687 * buffer flags have settled down, and before they are cleared. It doesn't
688 * harm to call it as often as desired (it just slightly hurts performance).
689 * It is only meant to be called by upper layers after buffer flags have been
690 * manipulated by analysers.
691 */
692void stream_int_update_conn(struct stream_interface *si)
693{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200694 struct channel *ib = si->ib;
695 struct channel *ob = si->ob;
Willy Tarreau100c4672012-08-20 12:06:26 +0200696
Willy Tarreau100c4672012-08-20 12:06:26 +0200697 /* Check if we need to close the read side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200698 if (!(ib->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200699 /* Read not closed, update FD status and timeout for reads */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100700 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200701 /* stop reading */
702 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100703 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200704 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200705 conn_data_stop_recv(si->conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200706 ib->rex = TICK_ETERNITY;
707 }
708 }
709 else {
710 /* (re)start reading and update timeout. Note: we don't recompute the timeout
711 * everytime we get here, otherwise it would risk never to expire. We only
712 * update it if is was not yet set. The stream socket handler will already
713 * have updated it if there has been a completed I/O.
714 */
715 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200716 conn_data_want_recv(si->conn);
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200717 if (!(ib->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ib->rex))
Willy Tarreau100c4672012-08-20 12:06:26 +0200718 ib->rex = tick_add_ifset(now_ms, ib->rto);
719 }
720 }
721
722 /* Check if we need to close the write side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200723 if (!(ob->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200724 /* Write not closed, update FD status and timeout for writes */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200725 if (channel_is_empty(ob)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200726 /* stop writing */
727 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreaub31c9712012-11-11 23:05:39 +0100728 if ((ob->flags & CF_SHUTW_NOW) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200729 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200730 conn_data_stop_send(si->conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200731 ob->wex = TICK_ETERNITY;
732 }
733 }
734 else {
735 /* (re)start writing and update timeout. Note: we don't recompute the timeout
736 * everytime we get here, otherwise it would risk never to expire. We only
737 * update it if is was not yet set. The stream socket handler will already
738 * have updated it if there has been a completed I/O.
739 */
740 si->flags &= ~SI_FL_WAIT_DATA;
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200741 conn_data_want_send(si->conn);
Willy Tarreau100c4672012-08-20 12:06:26 +0200742 if (!tick_isset(ob->wex)) {
743 ob->wex = tick_add_ifset(now_ms, ob->wto);
744 if (tick_isset(ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
745 /* Note: depending on the protocol, we don't know if we're waiting
746 * for incoming data or not. So in order to prevent the socket from
747 * expiring read timeouts during writes, we refresh the read timeout,
748 * except if it was already infinite or if we have explicitly setup
749 * independent streams.
750 */
751 ib->rex = tick_add_ifset(now_ms, ib->rto);
752 }
753 }
754 }
755 }
756}
757
Willy Tarreau46a8d922012-08-20 12:38:36 +0200758/* This function is used for inter-stream-interface calls. It is called by the
759 * consumer to inform the producer side that it may be interested in checking
760 * for free space in the buffer. Note that it intentionally does not update
761 * timeouts, so that we can still check them later at wake-up. This function is
762 * dedicated to connection-based stream interfaces.
763 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200764static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200765{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200766 struct channel *ib = si->ib;
Willy Tarreau46a8d922012-08-20 12:38:36 +0200767
Willy Tarreau34ffd772012-09-03 16:51:27 +0200768 if (unlikely(si->state > SI_ST_EST || (ib->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200769 return;
770
Willy Tarreau7d281492012-12-16 19:19:13 +0100771 conn_refresh_polling_flags(si->conn);
772
Willy Tarreaub31c9712012-11-11 23:05:39 +0100773 if ((ib->flags & CF_DONT_READ) || channel_full(ib)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200774 /* stop reading */
Willy Tarreaub31c9712012-11-11 23:05:39 +0100775 if (!(ib->flags & CF_DONT_READ)) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200776 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau7d281492012-12-16 19:19:13 +0100777 __conn_data_stop_recv(si->conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200778 }
779 else {
780 /* (re)start reading */
781 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreau7d281492012-12-16 19:19:13 +0100782 __conn_data_want_recv(si->conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200783 }
Willy Tarreau7d281492012-12-16 19:19:13 +0100784 conn_cond_update_data_polling(si->conn);
Willy Tarreau46a8d922012-08-20 12:38:36 +0200785}
786
787
Willy Tarreaude5722c2012-08-20 15:01:10 +0200788/* This function is used for inter-stream-interface calls. It is called by the
789 * producer to inform the consumer side that it may be interested in checking
790 * for data in the buffer. Note that it intentionally does not update timeouts,
791 * so that we can still check them later at wake-up.
792 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200793static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200794{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200795 struct channel *ob = si->ob;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200796
Willy Tarreau34ffd772012-09-03 16:51:27 +0200797 if (unlikely(si->state > SI_ST_EST || (ob->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200798 return;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200799
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200800 if (unlikely(channel_is_empty(ob))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200801 return;
802
803 if (!ob->pipe && /* spliced data wants to be forwarded ASAP */
Willy Tarreaub0165872012-12-15 10:12:39 +0100804 !(si->flags & SI_FL_WAIT_DATA)) /* not waiting for data */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200805 return;
806
Willy Tarreau5007d2a2013-07-18 22:09:48 +0200807 if (si->conn->flags & (CO_FL_DATA_WR_ENA|CO_FL_CURR_WR_ENA)) {
808 /* already subscribed to write notifications, will be called
809 * anyway, so let's avoid calling it especially if the reader
810 * is not ready.
811 */
812 return;
813 }
814
Willy Tarreaud29a0662012-12-10 16:33:38 +0100815 if (!(si->conn->flags & (CO_FL_HANDSHAKE|CO_FL_WAIT_L4_CONN|CO_FL_WAIT_L6_CONN))) {
816 /* Before calling the data-level operations, we have to prepare
817 * the polling flags to ensure we properly detect changes.
Willy Tarreaude5722c2012-08-20 15:01:10 +0200818 */
Willy Tarreaud29a0662012-12-10 16:33:38 +0100819 if (si->conn->ctrl)
820 fd_want_send(si->conn->t.sock.fd);
Willy Tarreau7d281492012-12-16 19:19:13 +0100821
822 conn_refresh_polling_flags(si->conn);
Willy Tarreaud29a0662012-12-10 16:33:38 +0100823
Godbache68e02d2013-10-11 15:48:29 +0800824 if (si_conn_send(si->conn) < 0) {
Willy Tarreaud29a0662012-12-10 16:33:38 +0100825 /* Write error on the file descriptor */
826 fd_stop_both(si->conn->t.sock.fd);
827 __conn_data_stop_both(si->conn);
828 si->flags |= SI_FL_ERR;
829 si->conn->flags |= CO_FL_ERROR;
830 goto out_wakeup;
831 }
Willy Tarreaude5722c2012-08-20 15:01:10 +0200832 }
833
834 /* OK, so now we know that some data might have been sent, and that we may
835 * have to poll first. We have to do that too if the buffer is not empty.
836 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200837 if (channel_is_empty(ob)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200838 /* the connection is established but we can't write. Either the
839 * buffer is empty, or we just refrain from sending because the
840 * ->o limit was reached. Maybe we just wrote the last
841 * chunk and need to close.
842 */
Willy Tarreaud29a0662012-12-10 16:33:38 +0100843 __conn_data_stop_send(si->conn);
Willy Tarreaub31c9712012-11-11 23:05:39 +0100844 if (((ob->flags & (CF_SHUTW|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200845 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200846 (si->state == SI_ST_EST)) {
847 si_shutw(si);
848 goto out_wakeup;
849 }
850
Willy Tarreaub31c9712012-11-11 23:05:39 +0100851 if ((ob->flags & (CF_SHUTW|CF_SHUTW_NOW)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200852 si->flags |= SI_FL_WAIT_DATA;
853 ob->wex = TICK_ETERNITY;
854 }
855 else {
856 /* Otherwise there are remaining data to be sent in the buffer,
857 * which means we have to poll before doing so.
858 */
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200859 __conn_data_want_send(si->conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200860 si->flags &= ~SI_FL_WAIT_DATA;
861 if (!tick_isset(ob->wex))
862 ob->wex = tick_add_ifset(now_ms, ob->wto);
863 }
864
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200865 if (likely(ob->flags & CF_WRITE_ACTIVITY)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200866 /* update timeout if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200867 if ((ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200868 !channel_is_empty(ob))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200869 ob->wex = tick_add_ifset(now_ms, ob->wto);
870
871 if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
872 /* Note: to prevent the client from expiring read timeouts
873 * during writes, we refresh it. We only do this if the
874 * interface is not configured for "independent streams",
875 * because for some applications it's better not to do this,
876 * for instance when continuously exchanging small amounts
877 * of data which can full the socket buffers long before a
878 * write timeout is detected.
879 */
880 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
881 }
882 }
883
884 /* in case of special condition (error, shutdown, end of write...), we
885 * have to notify the task.
886 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200887 if (likely((ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200888 (channel_is_empty(ob) && !ob->to_forward) ||
Willy Tarreaude5722c2012-08-20 15:01:10 +0200889 si->state != SI_ST_EST)) {
890 out_wakeup:
891 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
892 task_wakeup(si->owner, TASK_WOKEN_IO);
893 }
Willy Tarreauf16723e2012-08-24 12:52:22 +0200894
895 /* commit possible polling changes */
Willy Tarreauf2943dc2012-10-26 20:10:28 +0200896 conn_cond_update_polling(si->conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200897}
898
Willy Tarreaueecf6ca2012-08-20 15:09:53 +0200899/*
Willy Tarreauce323de2012-08-20 21:41:06 +0200900 * This is the callback which is called by the connection layer to receive data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200901 * into the buffer from the connection. It iterates over the transport layer's
902 * rcv_buf function.
Willy Tarreauce323de2012-08-20 21:41:06 +0200903 */
Willy Tarreau4aa36832012-10-02 20:07:22 +0200904static void si_conn_recv_cb(struct connection *conn)
Willy Tarreauce323de2012-08-20 21:41:06 +0200905{
Willy Tarreaue603e692012-09-27 22:20:41 +0200906 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200907 struct channel *chn = si->ib;
Willy Tarreauce323de2012-08-20 21:41:06 +0200908 int ret, max, cur_read;
909 int read_poll = MAX_READ_POLL_LOOPS;
910
911 /* stop immediately on errors. Note that we DON'T want to stop on
912 * POLL_ERR, as the poller might report a write error while there
913 * are still data available in the recv buffer. This typically
914 * happens when we send too large a request to a backend server
915 * which rejects it before reading it all.
916 */
917 if (conn->flags & CO_FL_ERROR)
918 goto out_error;
919
920 /* stop here if we reached the end of data */
921 if (conn_data_read0_pending(conn))
922 goto out_shutdown_r;
923
924 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreaucb76e592012-10-12 23:56:57 +0200925 if (chn->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +0200926 return;
927
Willy Tarreau96199b12012-08-24 00:46:52 +0200928 cur_read = 0;
Willy Tarreau96199b12012-08-24 00:46:52 +0200929
930 /* First, let's see if we may splice data across the channel without
931 * using a buffer.
932 */
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200933 if (conn->xprt->rcv_pipe &&
Willy Tarreaufa8e2bc2013-07-18 22:21:54 +0200934 (chn->pipe || chn->to_forward >= MIN_SPLICE_FORWARD) &&
935 chn->flags & CF_KERN_SPLICING) {
Willy Tarreau9b28e032012-10-12 23:49:43 +0200936 if (buffer_not_empty(chn->buf)) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200937 /* We're embarrassed, there are already data pending in
938 * the buffer and we don't want to have them at two
939 * locations at a time. Let's indicate we need some
940 * place and ask the consumer to hurry.
941 */
942 goto abort_splice;
943 }
Willy Tarreauce323de2012-08-20 21:41:06 +0200944
Willy Tarreaucb76e592012-10-12 23:56:57 +0200945 if (unlikely(chn->pipe == NULL)) {
946 if (pipes_used >= global.maxpipes || !(chn->pipe = get_pipe())) {
947 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +0200948 goto abort_splice;
949 }
950 }
951
Willy Tarreaucb76e592012-10-12 23:56:57 +0200952 ret = conn->xprt->rcv_pipe(conn, chn->pipe, chn->to_forward);
Willy Tarreau96199b12012-08-24 00:46:52 +0200953 if (ret < 0) {
954 /* splice not supported on this end, let's disable it */
Willy Tarreaucb76e592012-10-12 23:56:57 +0200955 chn->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +0200956 goto abort_splice;
957 }
Willy Tarreauce323de2012-08-20 21:41:06 +0200958
Willy Tarreau96199b12012-08-24 00:46:52 +0200959 if (ret > 0) {
Willy Tarreaucb76e592012-10-12 23:56:57 +0200960 if (chn->to_forward != CHN_INFINITE_FORWARD)
961 chn->to_forward -= ret;
962 chn->total += ret;
Willy Tarreau96199b12012-08-24 00:46:52 +0200963 cur_read += ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +0200964 chn->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +0200965 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200966
967 if (conn_data_read0_pending(conn))
968 goto out_shutdown_r;
969
970 if (conn->flags & CO_FL_ERROR)
971 goto out_error;
972
Willy Tarreau61d39a02013-07-18 21:49:32 +0200973 if (conn->flags & CO_FL_WAIT_ROOM) {
974 /* the pipe is full or we have read enough data that it
975 * could soon be full. Let's stop before needing to poll.
976 */
Willy Tarreau56a77e52012-09-02 18:34:44 +0200977 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau61d39a02013-07-18 21:49:32 +0200978 __conn_data_stop_recv(conn);
979 }
Willy Tarreau56a77e52012-09-02 18:34:44 +0200980
Willy Tarreauce323de2012-08-20 21:41:06 +0200981 /* splice not possible (anymore), let's go on on standard copy */
982 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200983
984 abort_splice:
Willy Tarreau61d39a02013-07-18 21:49:32 +0200985 if (chn->pipe && unlikely(!chn->pipe->data)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +0200986 put_pipe(chn->pipe);
987 chn->pipe = NULL;
Willy Tarreau96199b12012-08-24 00:46:52 +0200988 }
989
Willy Tarreau61d39a02013-07-18 21:49:32 +0200990 /* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
991 * was enabled, which implies that the recv buffer was not full. So we have a guarantee
992 * that if such an event is not handled above in splice, it will be handled here by
993 * recv().
994 */
995 while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_RD | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +0200996 max = bi_avail(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +0200997
998 if (!max) {
Willy Tarreau56a77e52012-09-02 18:34:44 +0200999 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001000 break;
1001 }
1002
Willy Tarreau9b28e032012-10-12 23:49:43 +02001003 ret = conn->xprt->rcv_buf(conn, chn->buf, max);
Willy Tarreauce323de2012-08-20 21:41:06 +02001004 if (ret <= 0)
1005 break;
1006
1007 cur_read += ret;
1008
1009 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001010 if (chn->to_forward && !(chn->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001011 unsigned long fwd = ret;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001012 if (chn->to_forward != CHN_INFINITE_FORWARD) {
1013 if (fwd > chn->to_forward)
1014 fwd = chn->to_forward;
1015 chn->to_forward -= fwd;
Willy Tarreauce323de2012-08-20 21:41:06 +02001016 }
Willy Tarreau9b28e032012-10-12 23:49:43 +02001017 b_adv(chn->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001018 }
1019
Willy Tarreaucb76e592012-10-12 23:56:57 +02001020 chn->flags |= CF_READ_PARTIAL;
1021 chn->total += ret;
Willy Tarreauce323de2012-08-20 21:41:06 +02001022
Willy Tarreaucb76e592012-10-12 23:56:57 +02001023 if (channel_full(chn)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001024 /* The buffer is now full, there's no point in going through
1025 * the loop again.
1026 */
Willy Tarreau9b28e032012-10-12 23:49:43 +02001027 if (!(chn->flags & CF_STREAMER_FAST) && (cur_read == buffer_len(chn->buf))) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001028 chn->xfer_small = 0;
1029 chn->xfer_large++;
1030 if (chn->xfer_large >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001031 /* we call this buffer a fast streamer if it manages
1032 * to be filled in one call 3 consecutive times.
1033 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001034 chn->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001035 //fputc('+', stderr);
1036 }
1037 }
Willy Tarreaucb76e592012-10-12 23:56:57 +02001038 else if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001039 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001040 chn->xfer_large = 0;
1041 chn->xfer_small++;
1042 if (chn->xfer_small >= 2) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001043 /* if the buffer has been at least half full twice,
1044 * we receive faster than we send, so at least it
1045 * is not a "fast streamer".
1046 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001047 chn->flags &= ~CF_STREAMER_FAST;
Willy Tarreauce323de2012-08-20 21:41:06 +02001048 //fputc('-', stderr);
1049 }
1050 }
1051 else {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001052 chn->xfer_small = 0;
1053 chn->xfer_large = 0;
Willy Tarreauce323de2012-08-20 21:41:06 +02001054 }
1055
Willy Tarreauce323de2012-08-20 21:41:06 +02001056 si->flags |= SI_FL_WAIT_ROOM;
1057 break;
1058 }
1059
Willy Tarreau5fddab02012-11-09 18:27:26 +01001060 if ((chn->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
Willy Tarreau34ac5662012-12-19 18:01:02 +01001061 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaud486ef52012-12-10 17:03:52 +01001062 __conn_data_stop_recv(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001063 break;
Willy Tarreau5fddab02012-11-09 18:27:26 +01001064 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001065
1066 /* if too many bytes were missing from last read, it means that
1067 * it's pointless trying to read again because the system does
1068 * not have them in buffers.
1069 */
1070 if (ret < max) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001071 if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreau9b28e032012-10-12 23:49:43 +02001072 (cur_read <= chn->buf->size / 2)) {
Willy Tarreaucb76e592012-10-12 23:56:57 +02001073 chn->xfer_large = 0;
1074 chn->xfer_small++;
1075 if (chn->xfer_small >= 3) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001076 /* we have read less than half of the buffer in
1077 * one pass, and this happened at least 3 times.
1078 * This is definitely not a streamer.
1079 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001080 chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001081 //fputc('!', stderr);
1082 }
1083 }
1084
1085 /* if a streamer has read few data, it may be because we
1086 * have exhausted system buffers. It's not worth trying
1087 * again.
1088 */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001089 if (chn->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001090 break;
1091
1092 /* if we read a large block smaller than what we requested,
1093 * it's almost certain we'll never get anything more.
1094 */
1095 if (ret >= global.tune.recv_enough)
1096 break;
1097 }
1098 } /* while !flags */
1099
Willy Tarreau96199b12012-08-24 00:46:52 +02001100 if (conn->flags & CO_FL_ERROR)
1101 goto out_error;
1102
Willy Tarreauce323de2012-08-20 21:41:06 +02001103 if (conn_data_read0_pending(conn))
1104 /* connection closed */
1105 goto out_shutdown_r;
1106
1107 return;
1108
1109 out_shutdown_r:
1110 /* we received a shutdown */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001111 chn->flags |= CF_READ_NULL;
1112 if (chn->flags & CF_AUTO_CLOSE)
1113 channel_shutw_now(chn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001114 stream_sock_read0(si);
1115 conn_data_read0(conn);
1116 return;
1117
1118 out_error:
1119 /* Read error on the connection, report the error and stop I/O */
1120 conn->flags |= CO_FL_ERROR;
Willy Tarreauce323de2012-08-20 21:41:06 +02001121}
1122
1123/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001124 * This is the callback which is called by the connection layer to send data
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02001125 * from the buffer to the connection. It iterates over the transport layer's
1126 * snd_buf function.
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001127 */
Willy Tarreau4aa36832012-10-02 20:07:22 +02001128static void si_conn_send_cb(struct connection *conn)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001129{
Willy Tarreaue603e692012-09-27 22:20:41 +02001130 struct stream_interface *si = conn->owner;
Willy Tarreaucb76e592012-10-12 23:56:57 +02001131 struct channel *chn = si->ob;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001132
1133 if (conn->flags & CO_FL_ERROR)
1134 goto out_error;
1135
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001136 if (si->conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001137 /* a handshake was requested */
1138 return;
1139
1140 /* we might have been called just after an asynchronous shutw */
Willy Tarreaucb76e592012-10-12 23:56:57 +02001141 if (chn->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001142 return;
1143
1144 /* OK there are data waiting to be sent */
Godbache68e02d2013-10-11 15:48:29 +08001145 if (si_conn_send(conn) < 0)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001146 goto out_error;
1147
1148 /* OK all done */
1149 return;
1150
1151 out_error:
1152 /* Write error on the connection, report the error and stop I/O */
1153 conn->flags |= CO_FL_ERROR;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001154}
1155
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001156/*
1157 * This function propagates a null read received on a socket-based connection.
1158 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
1159 * the close is also forwarded to the write side as an abort. This function is
1160 * still socket-specific as it handles a setsockopt() call to set the SO_LINGER
1161 * state on the socket.
1162 */
1163void stream_sock_read0(struct stream_interface *si)
1164{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001165 si->ib->flags &= ~CF_SHUTR_NOW;
1166 if (si->ib->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001167 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001168 si->ib->flags |= CF_SHUTR;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001169 si->ib->rex = TICK_ETERNITY;
1170 si->flags &= ~SI_FL_WAIT_ROOM;
1171
1172 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1173 return;
1174
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001175 if (si->ob->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001176 goto do_close;
1177
1178 if (si->flags & SI_FL_NOHALF) {
1179 /* we want to immediately forward this close to the write side */
1180 if (si->flags & SI_FL_NOLINGER) {
1181 si->flags &= ~SI_FL_NOLINGER;
Willy Tarreau7f7ad912012-11-11 19:27:15 +01001182 setsockopt(si->conn->t.sock.fd, SOL_SOCKET, SO_LINGER,
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001183 (struct linger *) &nolinger, sizeof(struct linger));
1184 }
1185 /* force flag on ssl to keep session in cache */
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001186 if (si->conn->xprt->shutw)
1187 si->conn->xprt->shutw(si->conn, 0);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001188 goto do_close;
1189 }
1190
1191 /* otherwise that's just a normal read shutdown */
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001192 __conn_data_stop_recv(si->conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001193 return;
1194
1195 do_close:
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001196 /* OK we completely close the socket here just as if we went through si_shut[rw]() */
Willy Tarreau2b199c92012-11-23 17:32:21 +01001197 conn_full_close(si->conn);
Willy Tarreauf9fbfe82012-11-21 21:51:53 +01001198
1199 si->ib->flags &= ~CF_SHUTR_NOW;
1200 si->ib->flags |= CF_SHUTR;
1201 si->ib->rex = TICK_ETERNITY;
1202
1203 si->ob->flags &= ~CF_SHUTW_NOW;
1204 si->ob->flags |= CF_SHUTW;
1205 si->ob->wex = TICK_ETERNITY;
1206
1207 si->flags &= ~(SI_FL_WAIT_DATA | SI_FL_WAIT_ROOM);
1208
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001209 si->state = SI_ST_DIS;
1210 si->exp = TICK_ETERNITY;
1211 if (si->release)
1212 si->release(si);
1213 return;
1214}
1215
Willy Tarreaudded32d2008-11-30 19:48:07 +01001216/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001217 * Local variables:
1218 * c-indent-level: 8
1219 * c-basic-offset: 8
1220 * End:
1221 */