blob: 765d04feab2d1cd9f5105b9f5a13ebc03a52d39d [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 Tarreau2c6be842012-07-06 17:12:34 +020032#include <proto/frontend.h>
Willy Tarreau96199b12012-08-24 00:46:52 +020033#include <proto/pipe.h>
Willy Tarreau269358d2009-09-20 20:14:49 +020034#include <proto/stream_interface.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010035#include <proto/task.h>
36
Willy Tarreaufd31e532012-07-23 18:24:25 +020037#include <types/pipe.h>
38
Willy Tarreauf873d752012-05-11 17:47:17 +020039/* socket functions used when running a stream interface as a task */
40static void stream_int_update(struct stream_interface *si);
41static void stream_int_update_embedded(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 Tarreauc5788912012-08-24 18:12:41 +020044static void stream_int_update_conn(struct stream_interface *si);
45static void stream_int_chk_rcv_conn(struct stream_interface *si);
46static void stream_int_chk_snd_conn(struct stream_interface *si);
Willy Tarreauf873d752012-05-11 17:47:17 +020047
Willy Tarreauc5788912012-08-24 18:12:41 +020048/* stream-interface operations for embedded tasks */
49struct si_ops si_embedded_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020050 .update = stream_int_update_embedded,
Willy Tarreau5c979a92012-05-07 17:15:39 +020051 .chk_rcv = stream_int_chk_rcv,
52 .chk_snd = stream_int_chk_snd,
Willy Tarreau5c979a92012-05-07 17:15:39 +020053};
54
Willy Tarreauc5788912012-08-24 18:12:41 +020055/* stream-interface operations for external tasks */
56struct si_ops si_task_ops = {
Willy Tarreau5c979a92012-05-07 17:15:39 +020057 .update = stream_int_update,
Willy Tarreau5c979a92012-05-07 17:15:39 +020058 .chk_rcv = stream_int_chk_rcv,
59 .chk_snd = stream_int_chk_snd,
Willy Tarreau5c979a92012-05-07 17:15:39 +020060};
61
Willy Tarreauc5788912012-08-24 18:12:41 +020062/* stream-interface operations for connections */
63struct si_ops si_conn_ops = {
64 .update = stream_int_update_conn,
65 .chk_rcv = stream_int_chk_rcv_conn,
66 .chk_snd = stream_int_chk_snd_conn,
67};
68
69struct app_cb si_conn_cb = {
70 .recv = si_conn_recv_cb,
71 .send = si_conn_send_cb,
72};
73
Willy Tarreaucff64112008-11-03 06:26:53 +010074/*
75 * This function only has to be called once after a wakeup event in case of
76 * suspected timeout. It controls the stream interface timeouts and sets
77 * si->flags accordingly. It does NOT close anything, as this timeout may
78 * be used for any purpose. It returns 1 if the timeout fired, otherwise
79 * zero.
80 */
81int stream_int_check_timeouts(struct stream_interface *si)
82{
83 if (tick_is_expired(si->exp, now_ms)) {
84 si->flags |= SI_FL_EXP;
85 return 1;
86 }
87 return 0;
88}
89
Willy Tarreaufe3718a2008-11-30 18:14:12 +010090/* to be called only when in SI_ST_DIS with SI_FL_ERR */
Willy Tarreaucff64112008-11-03 06:26:53 +010091void stream_int_report_error(struct stream_interface *si)
92{
93 if (!si->err_type)
94 si->err_type = SI_ET_DATA_ERR;
95
Willy Tarreau03cdb7c2012-08-27 23:14:58 +020096 si->ob->flags |= CF_WRITE_ERROR;
97 si->ib->flags |= CF_READ_ERROR;
Willy Tarreaucff64112008-11-03 06:26:53 +010098}
99
100/*
Willy Tarreaudded32d2008-11-30 19:48:07 +0100101 * Returns a message to the client ; the connection is shut down for read,
102 * and the request is cleared so that no server connection can be initiated.
103 * The buffer is marked for read shutdown on the other side to protect the
104 * message, and the buffer write is enabled. The message is contained in a
Willy Tarreau148d0992010-01-10 10:21:21 +0100105 * "chunk". If it is null, then an empty message is used. The reply buffer does
106 * not need to be empty before this, and its contents will not be overwritten.
107 * The primary goal of this function is to return error messages to a client.
Willy Tarreaudded32d2008-11-30 19:48:07 +0100108 */
109void stream_int_retnclose(struct stream_interface *si, const struct chunk *msg)
110{
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200111 channel_auto_read(si->ib);
112 channel_abort(si->ib);
113 channel_auto_close(si->ib);
114 channel_erase(si->ib);
Willy Tarreau798e1282010-12-12 13:06:00 +0100115
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200116 bi_erase(si->ob);
Willy Tarreau148d0992010-01-10 10:21:21 +0100117 if (likely(msg && msg->len))
Willy Tarreau9dab5fc2012-05-07 11:56:55 +0200118 bo_inject(si->ob, msg->str, msg->len);
Willy Tarreaudded32d2008-11-30 19:48:07 +0100119
120 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200121 channel_auto_read(si->ob);
122 channel_auto_close(si->ob);
123 channel_shutr_now(si->ob);
Willy Tarreau5d881d02009-12-27 22:51:06 +0100124}
125
Willy Tarreaufb90d942009-09-05 20:57:35 +0200126/* default update function for scheduled tasks, not used for embedded tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200127static void stream_int_update(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200128{
129 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
130 __FUNCTION__,
131 si, si->state, si->ib->flags, si->ob->flags);
132
133 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
134 task_wakeup(si->owner, TASK_WOKEN_IO);
135}
136
137/* default update function for embedded tasks, to be used at the end of the i/o handler */
Willy Tarreauf873d752012-05-11 17:47:17 +0200138static void stream_int_update_embedded(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200139{
Willy Tarreau3488e252010-08-09 16:24:56 +0200140 int old_flags = si->flags;
141
Willy Tarreaufb90d942009-09-05 20:57:35 +0200142 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
143 __FUNCTION__,
144 si, si->state, si->ib->flags, si->ob->flags);
145
146 if (si->state != SI_ST_EST)
147 return;
148
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200149 if ((si->ob->flags & (CF_SHUTW|CF_HIJACK|CF_SHUTW_NOW)) == CF_SHUTW_NOW &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200150 channel_is_empty(si->ob))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200151 si_shutw(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200152
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200153 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW|CF_HIJACK)) == 0 && !channel_full(si->ob))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200154 si->flags |= SI_FL_WAIT_DATA;
155
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200156 /* we're almost sure that we need some space if the buffer is not
157 * empty, even if it's not full, because the applets can't fill it.
158 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200159 if ((si->ib->flags & (CF_SHUTR|CF_DONT_READ)) == 0 && !channel_is_empty(si->ib))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200160 si->flags |= SI_FL_WAIT_ROOM;
161
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200162 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200163 if (tick_isset(si->ob->wex))
164 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
165 }
166
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200167 if (si->ib->flags & CF_READ_ACTIVITY ||
168 (si->ob->flags & CF_WRITE_ACTIVITY && !(si->flags & SI_FL_INDEP_STR))) {
Willy Tarreauf27b5ea2009-10-03 22:01:18 +0200169 if (tick_isset(si->ib->rex))
170 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
171 }
172
Willy Tarreau3488e252010-08-09 16:24:56 +0200173 /* save flags to detect changes */
174 old_flags = si->flags;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200175 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200176 !channel_full(si->ob) &&
Willy Tarreau96fd4b52009-10-04 17:18:35 +0200177 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
Willy Tarreau73b013b2012-05-21 16:31:45 +0200178 si_chk_rcv(si->ob->prod);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200179
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200180 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200181 (si->ib->cons->flags & SI_FL_WAIT_DATA)) {
Willy Tarreau73b013b2012-05-21 16:31:45 +0200182 si_chk_snd(si->ib->cons);
Willy Tarreau3488e252010-08-09 16:24:56 +0200183 /* check if the consumer has freed some space */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200184 if (!channel_full(si->ib))
Willy Tarreau3488e252010-08-09 16:24:56 +0200185 si->flags &= ~SI_FL_WAIT_ROOM;
186 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200187
188 /* Note that we're trying to wake up in two conditions here :
189 * - special event, which needs the holder task attention
190 * - status indicating that the applet can go on working. This
191 * is rather hard because we might be blocking on output and
192 * don't want to wake up on input and vice-versa. The idea is
Willy Tarreau3488e252010-08-09 16:24:56 +0200193 * to only rely on the changes the chk_* might have performed.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200194 */
195 if (/* check stream interface changes */
Willy Tarreau3488e252010-08-09 16:24:56 +0200196 ((old_flags & ~si->flags) & (SI_FL_WAIT_ROOM|SI_FL_WAIT_DATA)) ||
197
198 /* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200199 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200200 si->state != SI_ST_EST ||
201 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200202 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreau3488e252010-08-09 16:24:56 +0200203 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
204
205 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200206 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
207 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
208 ((si->ob->flags & CF_SHUTW) ||
Willy Tarreau3488e252010-08-09 16:24:56 +0200209 si->ob->prod->state != SI_ST_EST ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200210 (channel_is_empty(si->ob) && !si->ob->to_forward)))) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200211 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
212 task_wakeup(si->owner, TASK_WOKEN_IO);
213 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200214 if (si->ib->flags & CF_READ_ACTIVITY)
215 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200216}
217
Willy Tarreau4a36b562012-08-06 19:31:45 +0200218/*
219 * This function performs a shutdown-read on a stream interface in a connected
220 * or init state (it does nothing for other states). It either shuts the read
221 * side or marks itself as closed. The buffer flags are updated to reflect the
222 * new state. If the stream interface has SI_FL_NOHALF, we also forward the
223 * close to the write side. If a control layer is defined, then it is supposed
224 * to be a socket layer and file descriptors are then shutdown or closed
225 * accordingly. If no control layer is defined, then the SI is supposed to be
226 * an embedded one and the owner task is woken up if it exists. The function
227 * does not disable polling on the FD by itself, it returns non-zero instead
228 * if the caller needs to do so (except when the FD is deleted where this is
229 * implicit).
230 */
231int stream_int_shutr(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200232{
Willy Tarreau4a36b562012-08-06 19:31:45 +0200233 struct connection *conn = &si->conn;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200234
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200235 si->ib->flags &= ~CF_SHUTR_NOW;
236 if (si->ib->flags & CF_SHUTR)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200237 return 0;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200238 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200239 si->ib->rex = TICK_ETERNITY;
240 si->flags &= ~SI_FL_WAIT_ROOM;
241
242 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200243 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200244
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200245 if (si->ob->flags & CF_SHUTW) {
Willy Tarreau4a36b562012-08-06 19:31:45 +0200246 conn_data_close(&si->conn);
247 if (conn->ctrl)
248 fd_delete(si_fd(si));
Willy Tarreaufb90d942009-09-05 20:57:35 +0200249 si->state = SI_ST_DIS;
250 si->exp = TICK_ETERNITY;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200251
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200252 if (si->release)
253 si->release(si);
254 }
Willy Tarreau4a36b562012-08-06 19:31:45 +0200255 else if (si->flags & SI_FL_NOHALF) {
256 /* we want to immediately forward this close to the write side */
257 return stream_int_shutw(si);
258 }
259 else if (conn->ctrl) {
260 /* we want the caller to disable polling on this FD */
261 return 1;
262 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200263
Willy Tarreau4a36b562012-08-06 19:31:45 +0200264 /* note that if the task exists, it must unregister itself once it runs */
265 if (!conn->ctrl && !(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200266 task_wakeup(si->owner, TASK_WOKEN_IO);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200267 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200268}
269
Willy Tarreau4a36b562012-08-06 19:31:45 +0200270/*
271 * This function performs a shutdown-write on a stream interface in a connected or
272 * init state (it does nothing for other states). It either shuts the write side
273 * or marks itself as closed. The buffer flags are updated to reflect the new state.
274 * It does also close everything if the SI was marked as being in error state. If
275 * there is a data-layer shutdown, it is called. If a control layer is defined, then
276 * it is supposed to be a socket layer and file descriptors are then shutdown or
277 * closed accordingly. If no control layer is defined, then the SI is supposed to
278 * be an embedded one and the owner task is woken up if it exists. The function
279 * does not disable polling on the FD by itself, it returns non-zero instead if
280 * the caller needs to do so (except when the FD is deleted where this is implicit).
281 */
282int stream_int_shutw(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200283{
Willy Tarreau4a36b562012-08-06 19:31:45 +0200284 struct connection *conn = &si->conn;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200285
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200286 si->ob->flags &= ~CF_SHUTW_NOW;
287 if (si->ob->flags & CF_SHUTW)
Willy Tarreau4a36b562012-08-06 19:31:45 +0200288 return 0;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200289 si->ob->flags |= CF_SHUTW;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200290 si->ob->wex = TICK_ETERNITY;
291 si->flags &= ~SI_FL_WAIT_DATA;
292
293 switch (si->state) {
294 case SI_ST_EST:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200295 /* we have to shut before closing, otherwise some short messages
296 * may never leave the system, especially when there are remaining
297 * unread data in the socket input buffer, or when nolinger is set.
298 * However, if SI_FL_NOLINGER is explicitly set, we know there is
299 * no risk so we close both sides immediately.
300 */
301 if (si->flags & SI_FL_ERR) {
302 /* quick close, the socket is already shut. Remove pending flags. */
303 si->flags &= ~SI_FL_NOLINGER;
304 } else if (si->flags & SI_FL_NOLINGER) {
305 si->flags &= ~SI_FL_NOLINGER;
306 if (conn->ctrl) {
307 setsockopt(si_fd(si), SOL_SOCKET, SO_LINGER,
308 (struct linger *) &nolinger, sizeof(struct linger));
309 }
310 /* unclean data-layer shutdown */
311 if (conn->data && conn->data->shutw)
312 conn->data->shutw(conn, 0);
313 } else {
314 /* clean data-layer shutdown */
315 if (conn->data && conn->data->shutw)
316 conn->data->shutw(conn, 1);
317
318 if (!(si->flags & SI_FL_NOHALF)) {
319 /* We shutdown transport layer */
320 if (conn->ctrl)
321 shutdown(si_fd(si), SHUT_WR);
322
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200323 if (!(si->ib->flags & (CF_SHUTR|CF_DONT_READ))) {
Willy Tarreau4a36b562012-08-06 19:31:45 +0200324 /* OK just a shutw, but we want the caller
325 * to disable polling on this FD if exists.
326 */
327 return !!conn->ctrl;
328 }
329 }
330 }
Willy Tarreaufb90d942009-09-05 20:57:35 +0200331
332 /* fall through */
333 case SI_ST_CON:
Willy Tarreau4a36b562012-08-06 19:31:45 +0200334 /* we may have to close a pending connection, and mark the
335 * response buffer as shutr
336 */
337 conn_data_close(&si->conn);
338 if (conn->ctrl)
339 fd_delete(si_fd(si));
340 /* fall through */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200341 case SI_ST_CER:
Willy Tarreau32d3ee92010-12-29 14:03:02 +0100342 case SI_ST_QUE:
343 case SI_ST_TAR:
Willy Tarreaufb90d942009-09-05 20:57:35 +0200344 si->state = SI_ST_DIS;
Willy Tarreaud8ccffe2010-09-07 16:16:50 +0200345
346 if (si->release)
347 si->release(si);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200348 default:
349 si->flags &= ~SI_FL_WAIT_ROOM;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200350 si->ib->flags |= CF_SHUTR;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200351 si->ib->rex = TICK_ETERNITY;
352 si->exp = TICK_ETERNITY;
353 }
354
Willy Tarreau4a36b562012-08-06 19:31:45 +0200355 /* note that if the task exists, it must unregister itself once it runs */
356 if (!conn->ctrl && !(si->flags & SI_FL_DONT_WAKE) && si->owner)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200357 task_wakeup(si->owner, TASK_WOKEN_IO);
Willy Tarreau4a36b562012-08-06 19:31:45 +0200358 return 0;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200359}
360
361/* default chk_rcv function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200362static void stream_int_chk_rcv(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200363{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200364 struct channel *ib = si->ib;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200365
366 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
367 __FUNCTION__,
368 si, si->state, si->ib->flags, si->ob->flags);
369
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200370 if (unlikely(si->state != SI_ST_EST || (ib->flags & (CF_SHUTR|CF_HIJACK|CF_DONT_READ))))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200371 return;
372
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200373 if (channel_full(ib)) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200374 /* stop reading */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200375 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200376 }
377 else {
378 /* (re)start reading */
379 si->flags &= ~SI_FL_WAIT_ROOM;
380 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
381 task_wakeup(si->owner, TASK_WOKEN_IO);
382 }
383}
384
385/* default chk_snd function for scheduled tasks */
Willy Tarreauf873d752012-05-11 17:47:17 +0200386static void stream_int_chk_snd(struct stream_interface *si)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200387{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200388 struct channel *ob = si->ob;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200389
390 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
391 __FUNCTION__,
392 si, si->state, si->ib->flags, si->ob->flags);
393
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200394 if (unlikely(si->state != SI_ST_EST || (si->ob->flags & CF_SHUTW)))
Willy Tarreaufb90d942009-09-05 20:57:35 +0200395 return;
396
397 if (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200398 channel_is_empty(ob)) /* called with nothing to send ! */
Willy Tarreaufb90d942009-09-05 20:57:35 +0200399 return;
400
401 /* Otherwise there are remaining data to be sent in the buffer,
402 * so we tell the handler.
403 */
404 si->flags &= ~SI_FL_WAIT_DATA;
405 if (!tick_isset(ob->wex))
406 ob->wex = tick_add_ifset(now_ms, ob->wto);
407
408 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
409 task_wakeup(si->owner, TASK_WOKEN_IO);
410}
411
Willy Tarreaub24281b2011-02-13 13:16:36 +0100412/* Register an applet to handle a stream_interface as part of the stream
Willy Tarreaufb90d942009-09-05 20:57:35 +0200413 * interface's owner task, which is returned. The SI will wake it up everytime
Willy Tarreaub24281b2011-02-13 13:16:36 +0100414 * it is solicited. The task's processing function must call the applet's
Willy Tarreaufb90d942009-09-05 20:57:35 +0200415 * function before returning. It must be deleted by the task handler using
Willy Tarreaub24281b2011-02-13 13:16:36 +0100416 * stream_int_unregister_handler(), possibly from within the function itself.
Willy Tarreaufa6bac62012-05-31 14:16:59 +0200417 * It also pre-initializes applet.state to zero and the connection context
418 * to NULL.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200419 */
Willy Tarreaub24281b2011-02-13 13:16:36 +0100420struct task *stream_int_register_handler(struct stream_interface *si, struct si_applet *app)
Willy Tarreaufb90d942009-09-05 20:57:35 +0200421{
Simon Horman7abd00d2011-08-13 08:03:51 +0900422 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", app, si, si->owner);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200423
Willy Tarreauc5788912012-08-24 18:12:41 +0200424 si_prepare_embedded(si);
Willy Tarreau9e000c62011-03-10 14:03:36 +0100425 set_target_applet(&si->target, app);
Aman Gupta9a13e842012-04-02 18:57:53 -0700426 si->release = app->release;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200427 si->flags |= SI_FL_WAIT_DATA;
428 return si->owner;
429}
430
431/* Register a function to handle a stream_interface as a standalone task. The
432 * new task itself is returned and is assigned as si->owner. The stream_interface
433 * pointer will be pointed to by the task's context. The handler can be detached
434 * by using stream_int_unregister_handler().
Willy Tarreau7c0a1512011-03-10 11:17:02 +0100435 * FIXME: the code should be updated to ensure that we don't change si->owner
436 * anymore as this is not needed. However, process_session still relies on it.
Willy Tarreaufb90d942009-09-05 20:57:35 +0200437 */
438struct task *stream_int_register_handler_task(struct stream_interface *si,
439 struct task *(*fct)(struct task *))
440{
441 struct task *t;
442
443 DPRINTF(stderr, "registering handler %p for si %p (was %p)\n", fct, si, si->owner);
444
Willy Tarreauc5788912012-08-24 18:12:41 +0200445 si_prepare_task(si);
Willy Tarreau9e000c62011-03-10 14:03:36 +0100446 clear_target(&si->target);
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200447 si->release = NULL;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200448 si->flags |= SI_FL_WAIT_DATA;
449
450 t = task_new();
451 si->owner = t;
452 if (!t)
453 return t;
Willy Tarreau7c0a1512011-03-10 11:17:02 +0100454
Willy Tarreau9e000c62011-03-10 14:03:36 +0100455 set_target_task(&si->target, t);
Willy Tarreau7c0a1512011-03-10 11:17:02 +0100456
Willy Tarreaufb90d942009-09-05 20:57:35 +0200457 t->process = fct;
458 t->context = si;
459 task_wakeup(si->owner, TASK_WOKEN_INIT);
460
461 return t;
462}
463
464/* Unregister a stream interface handler. This must be called by the handler task
465 * itself when it detects that it is in the SI_ST_DIS state. This function can
466 * both detach standalone handlers and embedded handlers.
467 */
468void stream_int_unregister_handler(struct stream_interface *si)
469{
Willy Tarreau7c0a1512011-03-10 11:17:02 +0100470 if (si->target.type == TARG_TYPE_TASK) {
Willy Tarreaufb90d942009-09-05 20:57:35 +0200471 /* external handler : kill the task */
Willy Tarreau7c0a1512011-03-10 11:17:02 +0100472 task_delete(si->target.ptr.t);
473 task_free(si->target.ptr.t);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200474 }
Willy Tarreau0bd05ea2010-07-02 11:18:03 +0200475 si->release = NULL;
Willy Tarreaufb90d942009-09-05 20:57:35 +0200476 si->owner = NULL;
Willy Tarreau9e000c62011-03-10 14:03:36 +0100477 clear_target(&si->target);
Willy Tarreaufb90d942009-09-05 20:57:35 +0200478}
479
Willy Tarreau2c6be842012-07-06 17:12:34 +0200480/* This callback is used to send a valid PROXY protocol line to a socket being
Willy Tarreauafad0e02012-08-09 14:45:22 +0200481 * established. It returns 0 if it fails in a fatal way or needs to poll to go
482 * further, otherwise it returns non-zero and removes itself from the connection's
Willy Tarreaua1a74742012-08-24 12:14:49 +0200483 * flags (the bit is provided in <flag> by the caller). It is designed to be
484 * called by the connection handler and relies on it to commit polling changes.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200485 */
486int conn_si_send_proxy(struct connection *conn, unsigned int flag)
487{
Willy Tarreau2c6be842012-07-06 17:12:34 +0200488 struct stream_interface *si = container_of(conn, struct stream_interface, conn);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200489
490 /* we might have been called just after an asynchronous shutw */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200491 if (conn->flags & CO_FL_SOCK_WR_SH)
Willy Tarreau2c6be842012-07-06 17:12:34 +0200492 goto out_error;
493
494 /* If we have a PROXY line to send, we'll use this to validate the
495 * connection, in which case the connection is validated only once
496 * we've sent the whole proxy line. Otherwise we use connect().
497 */
498 if (si->send_proxy_ofs) {
499 int ret;
500
501 /* The target server expects a PROXY line to be sent first.
502 * If the send_proxy_ofs is negative, it corresponds to the
503 * offset to start sending from then end of the proxy string
504 * (which is recomputed every time since it's constant). If
505 * it is positive, it means we have to send from the start.
506 */
Willy Tarreaua1a74742012-08-24 12:14:49 +0200507 ret = make_proxy_line(trash, trashlen, &si->ob->prod->addr.from, &si->ob->prod->addr.to);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200508 if (!ret)
509 goto out_error;
510
511 if (si->send_proxy_ofs > 0)
512 si->send_proxy_ofs = -ret; /* first call */
513
Willy Tarreaua1a74742012-08-24 12:14:49 +0200514 /* we have to send trash from (ret+sp for -sp bytes). If the
515 * data layer has a pending write, we'll also set MSG_MORE.
516 */
517 ret = send(conn->t.sock.fd, trash + ret + si->send_proxy_ofs, -si->send_proxy_ofs,
518 (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau2c6be842012-07-06 17:12:34 +0200519
520 if (ret == 0)
521 goto out_wait;
522
523 if (ret < 0) {
524 if (errno == EAGAIN)
525 goto out_wait;
526 goto out_error;
527 }
528
529 si->send_proxy_ofs += ret; /* becomes zero once complete */
530 if (si->send_proxy_ofs != 0)
531 goto out_wait;
532
533 /* OK we've sent the whole line, we're connected */
534 }
535
Willy Tarreaua1a74742012-08-24 12:14:49 +0200536 /* The connection is ready now, simply return and let the connection
537 * handler notify upper layers if needed.
Willy Tarreau2c6be842012-07-06 17:12:34 +0200538 */
539 if (conn->flags & CO_FL_WAIT_L4_CONN)
540 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200541 conn->flags &= ~flag;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200542 return 1;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200543
544 out_error:
Willy Tarreauafad0e02012-08-09 14:45:22 +0200545 /* Write error on the file descriptor */
Willy Tarreau2c6be842012-07-06 17:12:34 +0200546 conn->flags |= CO_FL_ERROR;
Willy Tarreauafad0e02012-08-09 14:45:22 +0200547 conn->flags &= ~flag;
Willy Tarreaua1a74742012-08-24 12:14:49 +0200548 __conn_sock_stop_both(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200549 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200550
551 out_wait:
Willy Tarreaua1a74742012-08-24 12:14:49 +0200552 __conn_sock_stop_recv(conn);
553 __conn_sock_poll_send(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200554 return 0;
Willy Tarreau2c6be842012-07-06 17:12:34 +0200555}
556
Willy Tarreau100c4672012-08-20 12:06:26 +0200557/* Callback to be used by connection I/O handlers upon completion. It differs from
558 * the function below in that it is designed to be called by lower layers after I/O
559 * events have been completed. It will also try to wake the associated task up if
Willy Tarreauf16723e2012-08-24 12:52:22 +0200560 * an important event requires special handling. It relies on the connection handler
561 * to commit any polling updates.
Willy Tarreau100c4672012-08-20 12:06:26 +0200562 */
563void conn_notify_si(struct connection *conn)
Willy Tarreaufd31e532012-07-23 18:24:25 +0200564{
Willy Tarreaufd31e532012-07-23 18:24:25 +0200565 struct stream_interface *si = container_of(conn, struct stream_interface, conn);
566
567 DPRINTF(stderr, "%s: si=%p, si->state=%d ib->flags=%08x ob->flags=%08x\n",
568 __FUNCTION__,
569 si, si->state, si->ib->flags, si->ob->flags);
570
Willy Tarreau3c55ec22012-07-23 19:19:51 +0200571 if (conn->flags & CO_FL_ERROR)
572 si->flags |= SI_FL_ERR;
573
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200574 /* check for recent connection establishment */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200575 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED)))) {
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200576 si->exp = TICK_ETERNITY;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200577 si->ob->flags |= CF_WRITE_NULL;
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200578 }
579
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200580 /* process consumer side */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200581 if (channel_is_empty(si->ob)) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200582 if (((si->ob->flags & (CF_SHUTW|CF_HIJACK|CF_SHUTW_NOW)) == CF_SHUTW_NOW) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200583 (si->state == SI_ST_EST))
584 stream_int_shutw(si);
Willy Tarreauf16723e2012-08-24 12:52:22 +0200585 __conn_data_stop_send(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200586 si->ob->wex = TICK_ETERNITY;
587 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200588
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200589 if ((si->ob->flags & (CF_SHUTW|CF_SHUTW_NOW|CF_HIJACK)) == 0 && !channel_full(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200590 si->flags |= SI_FL_WAIT_DATA;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200591
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200592 if (si->ob->flags & CF_WRITE_ACTIVITY) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200593 /* update timeouts if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200594 if ((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200595 !channel_is_empty(si->ob))
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200596 if (tick_isset(si->ob->wex))
597 si->ob->wex = tick_add_ifset(now_ms, si->ob->wto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200598
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200599 if (!(si->flags & SI_FL_INDEP_STR))
600 if (tick_isset(si->ib->rex))
601 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200602
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200603 if (likely((si->ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL|CF_DONT_READ)) == CF_WRITE_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200604 !channel_full(si->ob) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200605 (si->ob->prod->flags & SI_FL_WAIT_ROOM)))
606 si_chk_rcv(si->ob->prod);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200607 }
608
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200609 /* process producer side.
610 * We might have some data the consumer is waiting for.
611 * We can do fast-forwarding, but we avoid doing this for partial
612 * buffers, because it is very likely that it will be done again
613 * immediately afterwards once the following data is parsed (eg:
614 * HTTP chunking).
615 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200616 if (((si->ib->flags & CF_READ_PARTIAL) && !channel_is_empty(si->ib)) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200617 (si->ib->pipe /* always try to send spliced data */ ||
618 (si->ib->buf.i == 0 && (si->ib->cons->flags & SI_FL_WAIT_DATA)))) {
619 int last_len = si->ib->pipe ? si->ib->pipe->data : 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200620
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200621 si_chk_snd(si->ib->cons);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200622
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200623 /* check if the consumer has freed some space either in the
624 * buffer or in the pipe.
625 */
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200626 if (!channel_full(si->ib) &&
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200627 (!last_len || !si->ib->pipe || si->ib->pipe->data < last_len))
628 si->flags &= ~SI_FL_WAIT_ROOM;
629 }
Willy Tarreaufd31e532012-07-23 18:24:25 +0200630
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200631 if (si->flags & SI_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200632 __conn_data_stop_recv(conn);
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200633 si->ib->rex = TICK_ETERNITY;
634 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200635 else if ((si->ib->flags & (CF_SHUTR|CF_READ_PARTIAL|CF_DONT_READ|CF_READ_NOEXP)) == CF_READ_PARTIAL &&
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200636 !channel_full(si->ib)) {
Willy Tarreau44b5dc62012-08-24 12:12:53 +0200637 if (tick_isset(si->ib->rex))
638 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
Willy Tarreaufd31e532012-07-23 18:24:25 +0200639 }
640
641 /* wake the task up only when needed */
642 if (/* changes on the production side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200643 (si->ib->flags & (CF_READ_NULL|CF_READ_ERROR)) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200644 si->state != SI_ST_EST ||
645 (si->flags & SI_FL_ERR) ||
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200646 ((si->ib->flags & CF_READ_PARTIAL) &&
Willy Tarreaufd31e532012-07-23 18:24:25 +0200647 (!si->ib->to_forward || si->ib->cons->state != SI_ST_EST)) ||
648
649 /* changes on the consumption side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200650 (si->ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR)) ||
651 ((si->ob->flags & CF_WRITE_ACTIVITY) &&
652 ((si->ob->flags & CF_SHUTW) ||
Willy Tarreaufd31e532012-07-23 18:24:25 +0200653 si->ob->prod->state != SI_ST_EST ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200654 (channel_is_empty(si->ob) && !si->ob->to_forward)))) {
Willy Tarreaufd31e532012-07-23 18:24:25 +0200655 task_wakeup(si->owner, TASK_WOKEN_IO);
656 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200657 if (si->ib->flags & CF_READ_ACTIVITY)
658 si->ib->flags &= ~CF_READ_DONTWAIT;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200659}
Willy Tarreau2c6be842012-07-06 17:12:34 +0200660
Willy Tarreau5368d802012-08-21 18:22:06 +0200661/*
662 * This function is called to send buffer data to a stream socket.
663 * It returns -1 in case of unrecoverable error, otherwise zero.
Willy Tarreauf16723e2012-08-24 12:52:22 +0200664 * It iterates the data layer's snd_buf function. It relies on the
665 * caller to commit polling changes.
Willy Tarreau5368d802012-08-21 18:22:06 +0200666 */
667static int si_conn_send_loop(struct connection *conn)
668{
669 struct stream_interface *si = container_of(conn, struct stream_interface, conn);
670 struct channel *b = si->ob;
671 int write_poll = MAX_WRITE_POLL_LOOPS;
672 int ret;
673
Willy Tarreau96199b12012-08-24 00:46:52 +0200674 conn->flags &= ~(CO_FL_WAIT_DATA | CO_FL_WAIT_ROOM);
Willy Tarreau5368d802012-08-21 18:22:06 +0200675
Willy Tarreau96199b12012-08-24 00:46:52 +0200676 if (b->pipe && conn->data->snd_pipe) {
677 ret = conn->data->snd_pipe(conn, b->pipe);
678 if (ret > 0)
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200679 b->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200680
681 if (!b->pipe->data) {
682 put_pipe(b->pipe);
683 b->pipe = NULL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200684 }
685
Willy Tarreau96199b12012-08-24 00:46:52 +0200686 if (conn->flags & CO_FL_ERROR)
687 return -1;
Willy Tarreau5368d802012-08-21 18:22:06 +0200688
Willy Tarreau96199b12012-08-24 00:46:52 +0200689 if (conn->flags & CO_FL_WAIT_ROOM) {
Willy Tarreauf16723e2012-08-24 12:52:22 +0200690 __conn_data_poll_send(conn);
Willy Tarreau96199b12012-08-24 00:46:52 +0200691 return 0;
692 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200693 }
694
695 /* At this point, the pipe is empty, but we may still have data pending
696 * in the normal buffer.
697 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200698 if (!b->buf.o)
Willy Tarreau5368d802012-08-21 18:22:06 +0200699 return 0;
Willy Tarreau5368d802012-08-21 18:22:06 +0200700
701 /* when we're in this loop, 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 while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH | CO_FL_WAIT_DATA | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
705 /* 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 */
715 unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
716
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200717 if ((!(b->flags & (CF_NEVER_WAIT|CF_SEND_DONTWAIT)) &&
718 ((b->to_forward && b->to_forward != CHN_INFINITE_FORWARD) ||
719 (b->flags & CF_EXPECT_MORE))) ||
720 ((b->flags & (CF_SHUTW|CF_SHUTW_NOW|CF_HIJACK)) == CF_SHUTW_NOW))
Willy Tarreau5368d802012-08-21 18:22:06 +0200721 send_flag |= MSG_MORE;
722
723 ret = conn->data->snd_buf(conn, &b->buf, send_flag);
724 if (ret <= 0)
725 break;
726
727 if (si->conn.flags & CO_FL_WAIT_L4_CONN)
728 si->conn.flags &= ~CO_FL_WAIT_L4_CONN;
729
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200730 b->flags |= CF_WRITE_PARTIAL;
Willy Tarreau5368d802012-08-21 18:22:06 +0200731
Willy Tarreau5368d802012-08-21 18:22:06 +0200732 if (!b->buf.o) {
733 /* Always clear both flags once everything has been sent, they're one-shot */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200734 b->flags &= ~(CF_EXPECT_MORE | CF_SEND_DONTWAIT);
Willy Tarreau5368d802012-08-21 18:22:06 +0200735 break;
736 }
737
738 if (--write_poll <= 0)
739 break;
740 } /* while */
741
742 if (conn->flags & CO_FL_ERROR)
743 return -1;
744
745 if (conn->flags & CO_FL_WAIT_ROOM) {
746 /* we need to poll before going on */
Willy Tarreauf16723e2012-08-24 12:52:22 +0200747 __conn_data_poll_send(&si->conn);
Willy Tarreau5368d802012-08-21 18:22:06 +0200748 return 0;
749 }
750 return 0;
751}
752
753
Willy Tarreau100c4672012-08-20 12:06:26 +0200754/* Updates the timers and flags of a stream interface attached to a connection,
755 * depending on the buffers' flags. It should only be called once after the
756 * buffer flags have settled down, and before they are cleared. It doesn't
757 * harm to call it as often as desired (it just slightly hurts performance).
758 * It is only meant to be called by upper layers after buffer flags have been
759 * manipulated by analysers.
760 */
761void stream_int_update_conn(struct stream_interface *si)
762{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200763 struct channel *ib = si->ib;
764 struct channel *ob = si->ob;
Willy Tarreau100c4672012-08-20 12:06:26 +0200765
766 if (si->conn.flags & CO_FL_HANDSHAKE) {
767 /* a handshake is in progress */
Willy Tarreau100c4672012-08-20 12:06:26 +0200768 return;
769 }
770
771 /* Check if we need to close the read side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200772 if (!(ib->flags & CF_SHUTR)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200773 /* Read not closed, update FD status and timeout for reads */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200774 if ((ib->flags & (CF_HIJACK|CF_DONT_READ)) || channel_full(ib)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200775 /* stop reading */
776 if (!(si->flags & SI_FL_WAIT_ROOM)) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200777 if (!(ib->flags & (CF_HIJACK|CF_DONT_READ))) /* full */
Willy Tarreau100c4672012-08-20 12:06:26 +0200778 si->flags |= SI_FL_WAIT_ROOM;
779 conn_data_stop_recv(&si->conn);
780 ib->rex = TICK_ETERNITY;
781 }
782 }
783 else {
784 /* (re)start reading and update timeout. Note: we don't recompute the timeout
785 * everytime we get here, otherwise it would risk never to expire. We only
786 * update it if is was not yet set. The stream socket handler will already
787 * have updated it if there has been a completed I/O.
788 */
789 si->flags &= ~SI_FL_WAIT_ROOM;
790 conn_data_want_recv(&si->conn);
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200791 if (!(ib->flags & (CF_READ_NOEXP|CF_DONT_READ)) && !tick_isset(ib->rex))
Willy Tarreau100c4672012-08-20 12:06:26 +0200792 ib->rex = tick_add_ifset(now_ms, ib->rto);
793 }
794 }
795
796 /* Check if we need to close the write side */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200797 if (!(ob->flags & CF_SHUTW)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200798 /* Write not closed, update FD status and timeout for writes */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200799 if (channel_is_empty(ob)) {
Willy Tarreau100c4672012-08-20 12:06:26 +0200800 /* stop writing */
801 if (!(si->flags & SI_FL_WAIT_DATA)) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200802 if ((ob->flags & (CF_HIJACK|CF_SHUTW_NOW)) == 0)
Willy Tarreau100c4672012-08-20 12:06:26 +0200803 si->flags |= SI_FL_WAIT_DATA;
804 conn_data_stop_send(&si->conn);
805 ob->wex = TICK_ETERNITY;
806 }
807 }
808 else {
809 /* (re)start writing and update timeout. Note: we don't recompute the timeout
810 * everytime we get here, otherwise it would risk never to expire. We only
811 * update it if is was not yet set. The stream socket handler will already
812 * have updated it if there has been a completed I/O.
813 */
814 si->flags &= ~SI_FL_WAIT_DATA;
815 conn_data_want_send(&si->conn);
816 if (!tick_isset(ob->wex)) {
817 ob->wex = tick_add_ifset(now_ms, ob->wto);
818 if (tick_isset(ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
819 /* Note: depending on the protocol, we don't know if we're waiting
820 * for incoming data or not. So in order to prevent the socket from
821 * expiring read timeouts during writes, we refresh the read timeout,
822 * except if it was already infinite or if we have explicitly setup
823 * independent streams.
824 */
825 ib->rex = tick_add_ifset(now_ms, ib->rto);
826 }
827 }
828 }
829 }
830}
831
Willy Tarreau46a8d922012-08-20 12:38:36 +0200832/* This function is used for inter-stream-interface calls. It is called by the
833 * consumer to inform the producer side that it may be interested in checking
834 * for free space in the buffer. Note that it intentionally does not update
835 * timeouts, so that we can still check them later at wake-up. This function is
836 * dedicated to connection-based stream interfaces.
837 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200838static void stream_int_chk_rcv_conn(struct stream_interface *si)
Willy Tarreau46a8d922012-08-20 12:38:36 +0200839{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200840 struct channel *ib = si->ib;
Willy Tarreau46a8d922012-08-20 12:38:36 +0200841
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200842 if (unlikely(si->state != SI_ST_EST || (ib->flags & CF_SHUTR)))
Willy Tarreau46a8d922012-08-20 12:38:36 +0200843 return;
844
845 if (si->conn.flags & CO_FL_HANDSHAKE) {
846 /* a handshake is in progress */
847 return;
848 }
849
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200850 if ((ib->flags & (CF_HIJACK|CF_DONT_READ)) || channel_full(ib)) {
Willy Tarreau46a8d922012-08-20 12:38:36 +0200851 /* stop reading */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200852 if (!(ib->flags & (CF_HIJACK|CF_DONT_READ))) /* full */
Willy Tarreau46a8d922012-08-20 12:38:36 +0200853 si->flags |= SI_FL_WAIT_ROOM;
854 conn_data_stop_recv(&si->conn);
855 }
856 else {
857 /* (re)start reading */
858 si->flags &= ~SI_FL_WAIT_ROOM;
859 conn_data_want_recv(&si->conn);
860 }
861}
862
863
Willy Tarreaude5722c2012-08-20 15:01:10 +0200864/* This function is used for inter-stream-interface calls. It is called by the
865 * producer to inform the consumer side that it may be interested in checking
866 * for data in the buffer. Note that it intentionally does not update timeouts,
867 * so that we can still check them later at wake-up.
868 */
Willy Tarreauc5788912012-08-24 18:12:41 +0200869static void stream_int_chk_snd_conn(struct stream_interface *si)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200870{
Willy Tarreau7421efb2012-07-02 15:11:27 +0200871 struct channel *ob = si->ob;
Willy Tarreaude5722c2012-08-20 15:01:10 +0200872
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200873 if (unlikely(si->state != SI_ST_EST || (ob->flags & CF_SHUTW)))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200874 return;
875
876 /* handshake running on producer */
877 if (si->conn.flags & CO_FL_HANDSHAKE) {
878 /* a handshake is in progress */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200879 return;
880 }
881
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200882 if (unlikely(channel_is_empty(ob))) /* called with nothing to send ! */
Willy Tarreaude5722c2012-08-20 15:01:10 +0200883 return;
884
885 if (!ob->pipe && /* spliced data wants to be forwarded ASAP */
886 (!(si->flags & SI_FL_WAIT_DATA) || /* not waiting for data */
887 (fdtab[si_fd(si)].ev & FD_POLL_OUT))) /* we'll be called anyway */
888 return;
889
Willy Tarreau5368d802012-08-21 18:22:06 +0200890 if (si_conn_send_loop(&si->conn) < 0) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200891 /* Write error on the file descriptor. We mark the FD as STERROR so
892 * that we don't use it anymore and we notify the task.
893 */
894 fdtab[si_fd(si)].ev &= ~FD_POLL_STICKY;
Willy Tarreauf16723e2012-08-24 12:52:22 +0200895 __conn_data_stop_both(&si->conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200896 si->flags |= SI_FL_ERR;
897 si->conn.flags |= CO_FL_ERROR;
898 goto out_wakeup;
899 }
900
901 /* OK, so now we know that some data might have been sent, and that we may
902 * have to poll first. We have to do that too if the buffer is not empty.
903 */
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200904 if (channel_is_empty(ob)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200905 /* the connection is established but we can't write. Either the
906 * buffer is empty, or we just refrain from sending because the
907 * ->o limit was reached. Maybe we just wrote the last
908 * chunk and need to close.
909 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200910 if (((ob->flags & (CF_SHUTW|CF_HIJACK|CF_AUTO_CLOSE|CF_SHUTW_NOW)) ==
911 (CF_AUTO_CLOSE|CF_SHUTW_NOW)) &&
Willy Tarreaude5722c2012-08-20 15:01:10 +0200912 (si->state == SI_ST_EST)) {
913 si_shutw(si);
914 goto out_wakeup;
915 }
916
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200917 if ((ob->flags & (CF_SHUTW|CF_SHUTW_NOW|CF_HIJACK)) == 0)
Willy Tarreaude5722c2012-08-20 15:01:10 +0200918 si->flags |= SI_FL_WAIT_DATA;
919 ob->wex = TICK_ETERNITY;
920 }
921 else {
922 /* Otherwise there are remaining data to be sent in the buffer,
923 * which means we have to poll before doing so.
924 */
Willy Tarreauf16723e2012-08-24 12:52:22 +0200925 __conn_data_want_send(&si->conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200926 si->flags &= ~SI_FL_WAIT_DATA;
927 if (!tick_isset(ob->wex))
928 ob->wex = tick_add_ifset(now_ms, ob->wto);
929 }
930
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200931 if (likely(ob->flags & CF_WRITE_ACTIVITY)) {
Willy Tarreaude5722c2012-08-20 15:01:10 +0200932 /* update timeout if we have written something */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200933 if ((ob->flags & (CF_SHUTW|CF_WRITE_PARTIAL)) == CF_WRITE_PARTIAL &&
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200934 !channel_is_empty(ob))
Willy Tarreaude5722c2012-08-20 15:01:10 +0200935 ob->wex = tick_add_ifset(now_ms, ob->wto);
936
937 if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
938 /* Note: to prevent the client from expiring read timeouts
939 * during writes, we refresh it. We only do this if the
940 * interface is not configured for "independent streams",
941 * because for some applications it's better not to do this,
942 * for instance when continuously exchanging small amounts
943 * of data which can full the socket buffers long before a
944 * write timeout is detected.
945 */
946 si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
947 }
948 }
949
950 /* in case of special condition (error, shutdown, end of write...), we
951 * have to notify the task.
952 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200953 if (likely((ob->flags & (CF_WRITE_NULL|CF_WRITE_ERROR|CF_SHUTW)) ||
Willy Tarreau8e21bb92012-08-24 22:40:29 +0200954 (channel_is_empty(ob) && !ob->to_forward) ||
Willy Tarreaude5722c2012-08-20 15:01:10 +0200955 si->state != SI_ST_EST)) {
956 out_wakeup:
957 if (!(si->flags & SI_FL_DONT_WAKE) && si->owner)
958 task_wakeup(si->owner, TASK_WOKEN_IO);
959 }
Willy Tarreauf16723e2012-08-24 12:52:22 +0200960
961 /* commit possible polling changes */
962 conn_cond_update_polling(&si->conn);
Willy Tarreaude5722c2012-08-20 15:01:10 +0200963}
964
Willy Tarreaueecf6ca2012-08-20 15:09:53 +0200965/*
Willy Tarreauce323de2012-08-20 21:41:06 +0200966 * This is the callback which is called by the connection layer to receive data
967 * into the buffer from the connection. It iterates over the data layer's rcv_buf
968 * function.
969 */
970void si_conn_recv_cb(struct connection *conn)
971{
972 struct stream_interface *si = container_of(conn, struct stream_interface, conn);
973 struct channel *b = si->ib;
974 int ret, max, cur_read;
975 int read_poll = MAX_READ_POLL_LOOPS;
976
977 /* stop immediately on errors. Note that we DON'T want to stop on
978 * POLL_ERR, as the poller might report a write error while there
979 * are still data available in the recv buffer. This typically
980 * happens when we send too large a request to a backend server
981 * which rejects it before reading it all.
982 */
983 if (conn->flags & CO_FL_ERROR)
984 goto out_error;
985
986 /* stop here if we reached the end of data */
987 if (conn_data_read0_pending(conn))
988 goto out_shutdown_r;
989
990 /* maybe we were called immediately after an asynchronous shutr */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +0200991 if (b->flags & CF_SHUTR)
Willy Tarreauce323de2012-08-20 21:41:06 +0200992 return;
993
Willy Tarreau96199b12012-08-24 00:46:52 +0200994 cur_read = 0;
995 conn->flags &= ~(CO_FL_WAIT_DATA | CO_FL_WAIT_ROOM);
996
997 /* First, let's see if we may splice data across the channel without
998 * using a buffer.
999 */
1000 if (conn->data->rcv_pipe &&
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001001 b->to_forward >= MIN_SPLICE_FORWARD && b->flags & CF_KERN_SPLICING) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001002 if (buffer_not_empty(&b->buf)) {
1003 /* We're embarrassed, there are already data pending in
1004 * the buffer and we don't want to have them at two
1005 * locations at a time. Let's indicate we need some
1006 * place and ask the consumer to hurry.
1007 */
1008 goto abort_splice;
1009 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001010
Willy Tarreau96199b12012-08-24 00:46:52 +02001011 if (unlikely(b->pipe == NULL)) {
1012 if (pipes_used >= global.maxpipes || !(b->pipe = get_pipe())) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001013 b->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001014 goto abort_splice;
1015 }
1016 }
1017
1018 ret = conn->data->rcv_pipe(conn, b->pipe, b->to_forward);
1019 if (ret < 0) {
1020 /* splice not supported on this end, let's disable it */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001021 b->flags &= ~CF_KERN_SPLICING;
Willy Tarreau96199b12012-08-24 00:46:52 +02001022 si->flags &= ~SI_FL_CAP_SPLICE;
1023 goto abort_splice;
1024 }
Willy Tarreauce323de2012-08-20 21:41:06 +02001025
Willy Tarreau96199b12012-08-24 00:46:52 +02001026 if (ret > 0) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001027 if (b->to_forward != CHN_INFINITE_FORWARD)
Willy Tarreau96199b12012-08-24 00:46:52 +02001028 b->to_forward -= ret;
1029 b->total += ret;
1030 cur_read += ret;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001031 b->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001032 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001033
1034 if (conn_data_read0_pending(conn))
1035 goto out_shutdown_r;
1036
1037 if (conn->flags & CO_FL_ERROR)
1038 goto out_error;
1039
Willy Tarreauce323de2012-08-20 21:41:06 +02001040 /* splice not possible (anymore), let's go on on standard copy */
1041 }
Willy Tarreau96199b12012-08-24 00:46:52 +02001042
1043 abort_splice:
1044 /* release the pipe if we can, which is almost always the case */
1045 if (b->pipe && !b->pipe->data) {
1046 put_pipe(b->pipe);
1047 b->pipe = NULL;
1048 }
1049
1050 while (!b->pipe && !(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_DATA | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001051 max = bi_avail(b);
1052
1053 if (!max) {
Willy Tarreau96199b12012-08-24 00:46:52 +02001054 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreauce323de2012-08-20 21:41:06 +02001055 break;
1056 }
1057
1058 ret = conn->data->rcv_buf(conn, &b->buf, max);
1059 if (ret <= 0)
1060 break;
1061
1062 cur_read += ret;
1063
1064 /* if we're allowed to directly forward data, we must update ->o */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001065 if (b->to_forward && !(b->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001066 unsigned long fwd = ret;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001067 if (b->to_forward != CHN_INFINITE_FORWARD) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001068 if (fwd > b->to_forward)
1069 fwd = b->to_forward;
1070 b->to_forward -= fwd;
1071 }
Willy Tarreaua75bcef2012-08-24 22:56:11 +02001072 b_adv(&b->buf, fwd);
Willy Tarreauce323de2012-08-20 21:41:06 +02001073 }
1074
1075 if (conn->flags & CO_FL_WAIT_L4_CONN)
1076 conn->flags &= ~CO_FL_WAIT_L4_CONN;
1077
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001078 b->flags |= CF_READ_PARTIAL;
Willy Tarreauce323de2012-08-20 21:41:06 +02001079 b->total += ret;
1080
Willy Tarreauad1cc3d2012-08-27 18:54:20 +02001081 if (channel_full(b)) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001082 /* The buffer is now full, there's no point in going through
1083 * the loop again.
1084 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001085 if (!(b->flags & CF_STREAMER_FAST) && (cur_read == buffer_len(&b->buf))) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001086 b->xfer_small = 0;
1087 b->xfer_large++;
1088 if (b->xfer_large >= 3) {
1089 /* we call this buffer a fast streamer if it manages
1090 * to be filled in one call 3 consecutive times.
1091 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001092 b->flags |= (CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001093 //fputc('+', stderr);
1094 }
1095 }
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001096 else if ((b->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauce323de2012-08-20 21:41:06 +02001097 (cur_read <= b->buf.size / 2)) {
1098 b->xfer_large = 0;
1099 b->xfer_small++;
1100 if (b->xfer_small >= 2) {
1101 /* if the buffer has been at least half full twice,
1102 * we receive faster than we send, so at least it
1103 * is not a "fast streamer".
1104 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001105 b->flags &= ~CF_STREAMER_FAST;
Willy Tarreauce323de2012-08-20 21:41:06 +02001106 //fputc('-', stderr);
1107 }
1108 }
1109 else {
1110 b->xfer_small = 0;
1111 b->xfer_large = 0;
1112 }
1113
Willy Tarreauce323de2012-08-20 21:41:06 +02001114 si->flags |= SI_FL_WAIT_ROOM;
1115 break;
1116 }
1117
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001118 if ((b->flags & CF_READ_DONTWAIT) || --read_poll <= 0)
Willy Tarreauce323de2012-08-20 21:41:06 +02001119 break;
1120
1121 /* if too many bytes were missing from last read, it means that
1122 * it's pointless trying to read again because the system does
1123 * not have them in buffers.
1124 */
1125 if (ret < max) {
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001126 if ((b->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
Willy Tarreauce323de2012-08-20 21:41:06 +02001127 (cur_read <= b->buf.size / 2)) {
1128 b->xfer_large = 0;
1129 b->xfer_small++;
1130 if (b->xfer_small >= 3) {
1131 /* we have read less than half of the buffer in
1132 * one pass, and this happened at least 3 times.
1133 * This is definitely not a streamer.
1134 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001135 b->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
Willy Tarreauce323de2012-08-20 21:41:06 +02001136 //fputc('!', stderr);
1137 }
1138 }
1139
1140 /* if a streamer has read few data, it may be because we
1141 * have exhausted system buffers. It's not worth trying
1142 * again.
1143 */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001144 if (b->flags & CF_STREAMER)
Willy Tarreauce323de2012-08-20 21:41:06 +02001145 break;
1146
1147 /* if we read a large block smaller than what we requested,
1148 * it's almost certain we'll never get anything more.
1149 */
1150 if (ret >= global.tune.recv_enough)
1151 break;
1152 }
1153 } /* while !flags */
1154
Willy Tarreau96199b12012-08-24 00:46:52 +02001155 if (conn->flags & CO_FL_ERROR)
1156 goto out_error;
1157
1158 if (conn->flags & CO_FL_WAIT_ROOM) {
Willy Tarreau2c052082012-08-24 12:53:56 +02001159 si->flags |= SI_FL_WAIT_ROOM;
Willy Tarreau96199b12012-08-24 00:46:52 +02001160 }
1161 else if (conn->flags & CO_FL_WAIT_DATA) {
Willy Tarreauce323de2012-08-20 21:41:06 +02001162 /* we don't automatically ask for polling if we have
1163 * read enough data, as it saves some syscalls with
1164 * speculative pollers.
1165 */
1166 if (cur_read < MIN_RET_FOR_READ_LOOP)
1167 __conn_data_poll_recv(conn);
1168 else
1169 __conn_data_want_recv(conn);
1170 }
1171
Willy Tarreauce323de2012-08-20 21:41:06 +02001172 if (conn_data_read0_pending(conn))
1173 /* connection closed */
1174 goto out_shutdown_r;
1175
1176 return;
1177
1178 out_shutdown_r:
1179 /* we received a shutdown */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001180 b->flags |= CF_READ_NULL;
1181 if (b->flags & CF_AUTO_CLOSE)
Willy Tarreau8263d2b2012-08-28 00:06:31 +02001182 channel_shutw_now(b);
Willy Tarreauce323de2012-08-20 21:41:06 +02001183 stream_sock_read0(si);
1184 conn_data_read0(conn);
1185 return;
1186
1187 out_error:
1188 /* Read error on the connection, report the error and stop I/O */
1189 conn->flags |= CO_FL_ERROR;
Willy Tarreauf16723e2012-08-24 12:52:22 +02001190 __conn_data_stop_both(conn);
Willy Tarreauce323de2012-08-20 21:41:06 +02001191}
1192
1193/*
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001194 * This is the callback which is called by the connection layer to send data
1195 * from the buffer to the connection. It iterates over the data layer's snd_buf
1196 * function.
1197 */
1198void si_conn_send_cb(struct connection *conn)
1199{
1200 struct stream_interface *si = container_of(conn, struct stream_interface, conn);
Willy Tarreau7421efb2012-07-02 15:11:27 +02001201 struct channel *b = si->ob;
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001202
1203 if (conn->flags & CO_FL_ERROR)
1204 goto out_error;
1205
1206 if (si->conn.flags & CO_FL_HANDSHAKE)
1207 /* a handshake was requested */
1208 return;
1209
1210 /* we might have been called just after an asynchronous shutw */
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001211 if (b->flags & CF_SHUTW)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001212 return;
1213
1214 /* OK there are data waiting to be sent */
Willy Tarreau5368d802012-08-21 18:22:06 +02001215 if (si_conn_send_loop(conn) < 0)
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001216 goto out_error;
1217
1218 /* OK all done */
1219 return;
1220
1221 out_error:
1222 /* Write error on the connection, report the error and stop I/O */
1223 conn->flags |= CO_FL_ERROR;
Willy Tarreauf16723e2012-08-24 12:52:22 +02001224 __conn_data_stop_both(conn);
Willy Tarreaueecf6ca2012-08-20 15:09:53 +02001225}
1226
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001227/*
1228 * This function propagates a null read received on a socket-based connection.
1229 * It updates the stream interface. If the stream interface has SI_FL_NOHALF,
1230 * the close is also forwarded to the write side as an abort. This function is
1231 * still socket-specific as it handles a setsockopt() call to set the SO_LINGER
1232 * state on the socket.
1233 */
1234void stream_sock_read0(struct stream_interface *si)
1235{
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001236 si->ib->flags &= ~CF_SHUTR_NOW;
1237 if (si->ib->flags & CF_SHUTR)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001238 return;
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001239 si->ib->flags |= CF_SHUTR;
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001240 si->ib->rex = TICK_ETERNITY;
1241 si->flags &= ~SI_FL_WAIT_ROOM;
1242
1243 if (si->state != SI_ST_EST && si->state != SI_ST_CON)
1244 return;
1245
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001246 if (si->ob->flags & CF_SHUTW)
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001247 goto do_close;
1248
1249 if (si->flags & SI_FL_NOHALF) {
1250 /* we want to immediately forward this close to the write side */
1251 if (si->flags & SI_FL_NOLINGER) {
1252 si->flags &= ~SI_FL_NOLINGER;
1253 setsockopt(si_fd(si), SOL_SOCKET, SO_LINGER,
1254 (struct linger *) &nolinger, sizeof(struct linger));
1255 }
1256 /* force flag on ssl to keep session in cache */
1257 if (si->conn.data->shutw)
1258 si->conn.data->shutw(&si->conn, 0);
1259 goto do_close;
1260 }
1261
1262 /* otherwise that's just a normal read shutdown */
Willy Tarreauf16723e2012-08-24 12:52:22 +02001263 __conn_data_stop_recv(&si->conn);
Willy Tarreau9bf9c142012-08-20 15:38:41 +02001264 return;
1265
1266 do_close:
1267 conn_data_close(&si->conn);
1268 fd_delete(si_fd(si));
1269 si->state = SI_ST_DIS;
1270 si->exp = TICK_ETERNITY;
1271 if (si->release)
1272 si->release(si);
1273 return;
1274}
1275
Willy Tarreaudded32d2008-11-30 19:48:07 +01001276/*
Willy Tarreaucff64112008-11-03 06:26:53 +01001277 * Local variables:
1278 * c-indent-level: 8
1279 * c-basic-offset: 8
1280 * End:
1281 */