blob: b088da2e70cb6200751ec2dc43030259b31db713 [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
Willy Tarreau9f6dc722019-03-01 11:15:10 +010025#include <sys/resource.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020026#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090028#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020029#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020030#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020031#include <arpa/inet.h>
32
Willy Tarreauc7e42382012-08-24 19:22:53 +020033#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/compat.h>
35#include <common/config.h>
36#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020037#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020038#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020039#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040
41#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020042#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010043#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020044
45#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020046#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010047#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020048#include <proto/fd.h>
49#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020050#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020052#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020063#include <proto/ssl_sock.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020064
Willy Tarreaubd741542010-03-16 18:46:54 +010065static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090066static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020067static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020068static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020069static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020070static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020071static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Willy Tarreau8ceae722018-11-26 11:58:30 +010073DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
74DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020075
76
Simon Horman63a4a822012-03-19 07:24:41 +090077static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010078 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
79 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020080 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020081
Willy Tarreau23964182014-05-20 20:56:30 +020082 /* Below we have finished checks */
83 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010085
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
89 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
90 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
93 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
94 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020095
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010096 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
97 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020099 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100101 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
102 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
103 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900104
105 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
106 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200107 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200108};
109
Cyril Bontéac92a062014-12-27 22:28:38 +0100110const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
111 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
122};
123
Simon Horman63a4a822012-03-19 07:24:41 +0900124static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100125 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
126
127 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
128 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
129
130 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
131 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
132 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
133 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
134
135 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
136 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
138};
139
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200140/*
141 * Convert check_status code to description
142 */
143const char *get_check_status_description(short check_status) {
144
145 const char *desc;
146
147 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200148 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200149 else
150 desc = NULL;
151
152 if (desc && *desc)
153 return desc;
154 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200155 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200156}
157
158/*
159 * Convert check_status code to short info
160 */
161const char *get_check_status_info(short check_status) {
162
163 const char *info;
164
165 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200166 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200167 else
168 info = NULL;
169
170 if (info && *info)
171 return info;
172 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200173 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200174}
175
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100176const char *get_analyze_status(short analyze_status) {
177
178 const char *desc;
179
180 if (analyze_status < HANA_STATUS_SIZE)
181 desc = analyze_statuses[analyze_status].desc;
182 else
183 desc = NULL;
184
185 if (desc && *desc)
186 return desc;
187 else
188 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
189}
190
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200191/*
Simon Horman4a741432013-02-23 15:35:38 +0900192 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200193 * an adequate CHK_RES_* value. The new check->health is computed based
194 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200195 *
196 * Show information in logs about failed health check if server is UP
197 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200198 */
Simon Horman4a741432013-02-23 15:35:38 +0900199static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100200{
Simon Horman4a741432013-02-23 15:35:38 +0900201 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200202 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200203 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900204
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200205 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100206 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900207 check->desc[0] = '\0';
208 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200209 return;
210 }
211
Simon Horman4a741432013-02-23 15:35:38 +0900212 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200213 return;
214
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200215 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900216 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
217 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200218 } else
Simon Horman4a741432013-02-23 15:35:38 +0900219 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200220
Simon Horman4a741432013-02-23 15:35:38 +0900221 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200222 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200224
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100225 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900226 check->duration = -1;
227 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200228 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900229 check->duration = tv_ms_elapsed(&check->start, &now);
230 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200231 }
232
Willy Tarreau23964182014-05-20 20:56:30 +0200233 /* no change is expected if no state change occurred */
234 if (check->result == CHK_RES_NEUTRAL)
235 return;
236
Olivier Houchard0923fa42019-01-11 18:43:04 +0100237 /* If the check was really just sending a mail, it won't have an
238 * associated server, so we're done now.
239 */
240 if (!s)
241 return;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200242 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200243
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200244 switch (check->result) {
245 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200246 /* Failure to connect to the agent as a secondary check should not
247 * cause the server to be marked down.
248 */
249 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900250 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200251 (check->health > 0)) {
Olivier Houchard7059c552019-03-08 18:49:32 +0100252 _HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200253 report = 1;
254 check->health--;
255 if (check->health < check->rise)
256 check->health = 0;
257 }
258 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200259
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200260 case CHK_RES_PASSED:
261 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
262 if ((check->health < check->rise + check->fall - 1) &&
263 (check->result == CHK_RES_PASSED || check->health > 0)) {
264 report = 1;
265 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200266
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200267 if (check->health >= check->rise)
268 check->health = check->rise + check->fall - 1; /* OK now */
269 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200270
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200271 /* clear consecutive_errors if observing is enabled */
272 if (s->onerror)
273 s->consecutive_errors = 0;
274 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100275
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200276 default:
277 break;
278 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200279
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200280 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
281 (status != prev_status || report)) {
282 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200283 "%s check for %sserver %s/%s %s%s",
284 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200285 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100286 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100287 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200288 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200289
Emeric Brun5a133512017-10-19 14:42:30 +0200290 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200291
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100292 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200293 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
294 (check->health >= check->rise) ? check->fall : check->rise,
295 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200296
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200297 ha_warning("%s.\n", trash.area);
298 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
299 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200300 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200301}
302
Willy Tarreau4eec5472014-05-20 22:32:27 +0200303/* Marks the check <check>'s server down if the current check is already failed
304 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200305 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200306static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200307{
Simon Horman4a741432013-02-23 15:35:38 +0900308 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900309
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200310 /* The agent secondary check should only cause a server to be marked
311 * as down if check->status is HCHK_STATUS_L7STS, which indicates
312 * that the agent returned "fail", "stopped" or "down".
313 * The implication here is that failure to connect to the agent
314 * as a secondary check should not cause the server to be marked
315 * down. */
316 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
317 return;
318
Willy Tarreau4eec5472014-05-20 22:32:27 +0200319 if (check->health > 0)
320 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100321
Willy Tarreau4eec5472014-05-20 22:32:27 +0200322 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200323 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200324}
325
Willy Tarreauaf549582014-05-16 17:37:50 +0200326/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200327 * it isn't in maintenance, it is not tracking a down server and other checks
328 * comply. The rule is simple : by default, a server is up, unless any of the
329 * following conditions is true :
330 * - health check failed (check->health < rise)
331 * - agent check failed (agent->health < rise)
332 * - the server tracks a down server (track && track->state == STOPPED)
333 * Note that if the server has a slowstart, it will switch to STARTING instead
334 * of RUNNING. Also, only the health checks support the nolb mode, so the
335 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200336 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200337static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200338{
Simon Horman4a741432013-02-23 15:35:38 +0900339 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100340
Emeric Brun52a91d32017-08-31 14:41:55 +0200341 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200342 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100343
Emeric Brun52a91d32017-08-31 14:41:55 +0200344 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200345 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100346
Willy Tarreau3e048382014-05-21 10:30:54 +0200347 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
348 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100349
Willy Tarreau3e048382014-05-21 10:30:54 +0200350 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
351 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200352
Emeric Brun52a91d32017-08-31 14:41:55 +0200353 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200354 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100355
Emeric Brun5a133512017-10-19 14:42:30 +0200356 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100357}
358
Willy Tarreaudb58b792014-05-21 13:57:23 +0200359/* Marks the check <check> as valid and tries to set its server into stopping mode
360 * if it was running or starting, and provided it isn't in maintenance and other
361 * checks comply. The conditions for the server to be marked in stopping mode are
362 * the same as for it to be turned up. Also, only the health checks support the
363 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200364 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200365static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200366{
Simon Horman4a741432013-02-23 15:35:38 +0900367 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100368
Emeric Brun52a91d32017-08-31 14:41:55 +0200369 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200370 return;
371
Willy Tarreaudb58b792014-05-21 13:57:23 +0200372 if (check->state & CHK_ST_AGENT)
373 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100374
Emeric Brun52a91d32017-08-31 14:41:55 +0200375 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200376 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100377
Willy Tarreaudb58b792014-05-21 13:57:23 +0200378 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
379 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100380
Willy Tarreaudb58b792014-05-21 13:57:23 +0200381 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
382 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100383
Willy Tarreaub26881a2017-12-23 11:16:49 +0100384 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100385}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200386
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100387/* note: use health_adjust() only, which first checks that the observe mode is
388 * enabled.
389 */
390void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100391{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100392 int failed;
393 int expire;
394
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100395 if (s->observe >= HANA_OBS_SIZE)
396 return;
397
Willy Tarreaubb956662013-01-24 00:37:39 +0100398 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100399 return;
400
401 switch (analyze_statuses[status].lr[s->observe - 1]) {
402 case 1:
403 failed = 1;
404 break;
405
406 case 2:
407 failed = 0;
408 break;
409
410 default:
411 return;
412 }
413
414 if (!failed) {
415 /* good: clear consecutive_errors */
416 s->consecutive_errors = 0;
417 return;
418 }
419
Olivier Houchard7059c552019-03-08 18:49:32 +0100420 _HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100421
422 if (s->consecutive_errors < s->consecutive_errors_limit)
423 return;
424
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100425 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
426 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100427
428 switch (s->onerror) {
429 case HANA_ONERR_FASTINTER:
430 /* force fastinter - nothing to do here as all modes force it */
431 break;
432
433 case HANA_ONERR_SUDDTH:
434 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900435 if (s->check.health > s->check.rise)
436 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100437
438 /* no break - fall through */
439
440 case HANA_ONERR_FAILCHK:
441 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200442 set_server_check_status(&s->check, HCHK_STATUS_HANA,
443 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200444 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100445 break;
446
447 case HANA_ONERR_MARKDWN:
448 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900449 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200450 set_server_check_status(&s->check, HCHK_STATUS_HANA,
451 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200452 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100453 break;
454
455 default:
456 /* write a warning? */
457 break;
458 }
459
460 s->consecutive_errors = 0;
Olivier Houchard7059c552019-03-08 18:49:32 +0100461 _HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100462
Simon Horman66183002013-02-23 10:16:43 +0900463 if (s->check.fastinter) {
464 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300465 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200466 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300467 /* requeue check task with new expire */
468 task_queue(s->check.task);
469 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100470 }
471}
472
Willy Tarreaua1dab552014-04-14 15:04:54 +0200473static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100474{
475 int sv_state;
476 int ratio;
477 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800478 char addr[46];
479 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100480 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
481 "UP %d/%d", "UP",
482 "NOLB %d/%d", "NOLB",
483 "no check" };
484
485 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
486 hlen += 24;
487
Willy Tarreauff5ae352013-12-11 20:36:34 +0100488 if (!(s->check.state & CHK_ST_ENABLED))
489 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200490 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900491 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100492 sv_state = 3; /* UP */
493 else
494 sv_state = 2; /* going down */
495
Emeric Brun52a91d32017-08-31 14:41:55 +0200496 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100497 sv_state += 2;
498 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900499 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100500 sv_state = 1; /* going up */
501 else
502 sv_state = 0; /* DOWN */
503 }
504
Willy Tarreaua1dab552014-04-14 15:04:54 +0200505 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100506 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200507 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
508 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100509
Joseph Lynch514061c2015-01-15 17:52:59 -0800510 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100511 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
512 snprintf(port, sizeof(port), "%u", s->svc_port);
513 else
514 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800515
516 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
517 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100518 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200519 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100520 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
521 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
522 s->nbpend);
523
Emeric Brun52a91d32017-08-31 14:41:55 +0200524 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100525 now.tv_sec < s->last_change + s->slowstart &&
526 now.tv_sec >= s->last_change) {
527 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200528 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100529 }
530
531 buffer[hlen++] = '\r';
532 buffer[hlen++] = '\n';
533
534 return hlen;
535}
536
Willy Tarreau20a18342013-12-05 00:31:46 +0100537/* Check the connection. If an error has already been reported or the socket is
538 * closed, keep errno intact as it is supposed to contain the valid error code.
539 * If no error is reported, check the socket's error queue using getsockopt().
540 * Warning, this must be done only once when returning from poll, and never
541 * after an I/O error was attempted, otherwise the error queue might contain
542 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
543 * socket. Returns non-zero if an error was reported, zero if everything is
544 * clean (including a properly closed socket).
545 */
546static int retrieve_errno_from_socket(struct connection *conn)
547{
548 int skerr;
549 socklen_t lskerr = sizeof(skerr);
550
551 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
552 return 1;
553
Willy Tarreau3c728722014-01-23 13:50:42 +0100554 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100555 return 0;
556
Willy Tarreau585744b2017-08-24 14:31:19 +0200557 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100558 errno = skerr;
559
560 if (errno == EAGAIN)
561 errno = 0;
562
563 if (!errno) {
564 /* we could not retrieve an error, that does not mean there is
565 * none. Just don't change anything and only report the prior
566 * error if any.
567 */
568 if (conn->flags & CO_FL_ERROR)
569 return 1;
570 else
571 return 0;
572 }
573
574 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
575 return 1;
576}
577
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100578/* Try to collect as much information as possible on the connection status,
579 * and adjust the server status accordingly. It may make use of <errno_bck>
580 * if non-null when the caller is absolutely certain of its validity (eg:
581 * checked just after a syscall). If the caller doesn't have a valid errno,
582 * it can pass zero, and retrieve_errno_from_socket() will be called to try
583 * to extract errno from the socket. If no error is reported, it will consider
584 * the <expired> flag. This is intended to be used when a connection error was
585 * reported in conn->flags or when a timeout was reported in <expired>. The
586 * function takes care of not updating a server status which was already set.
587 * All situations where at least one of <expired> or CO_FL_ERROR are set
588 * produce a status.
589 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200590static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100591{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200592 struct conn_stream *cs = check->cs;
593 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100594 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200595 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200596 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200597 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100598
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100599 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100600 return;
601
602 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200603 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100604 retrieve_errno_from_socket(conn);
605
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200606 if (conn && !(conn->flags & CO_FL_ERROR) &&
607 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100608 return;
609
610 /* we'll try to build a meaningful error message depending on the
611 * context of the error possibly present in conn->err_code, and the
612 * socket error possibly collected above. This is useful to know the
613 * exact step of the L6 layer (eg: SSL handshake).
614 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200615 chk = get_trash_chunk();
616
617 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900618 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200619 if (!step)
620 chunk_printf(chk, " at initial connection step of tcp-check");
621 else {
622 chunk_printf(chk, " at step %d of tcp-check", step);
623 /* we were looking for a string */
624 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
625 if (check->last_started_step->port)
626 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
627 else
628 chunk_appendf(chk, " (connect)");
629 }
630 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
631 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200632 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200633 else if (check->last_started_step->expect_regex)
634 chunk_appendf(chk, " (expect regex)");
635 }
636 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
637 chunk_appendf(chk, " (send)");
638 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200639
640 comment = tcpcheck_get_step_comment(check, step);
641 if (comment)
642 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200643 }
644 }
645
Willy Tarreau00149122017-10-04 18:05:01 +0200646 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100647 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200648 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
649 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100650 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200651 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
652 chk->area);
653 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100654 }
655 else {
656 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200657 chunk_printf(&trash, "%s%s", strerror(errno),
658 chk->area);
659 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100660 }
661 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200662 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100663 }
664 }
665
Willy Tarreau00149122017-10-04 18:05:01 +0200666 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200667 /* NOTE: this is reported after <fall> tries */
668 chunk_printf(chk, "No port available for the TCP connection");
669 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
670 }
671
Willy Tarreau00149122017-10-04 18:05:01 +0200672 if (!conn) {
673 /* connection allocation error before the connection was established */
674 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
675 }
676 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100677 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200678 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100679 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
680 else if (expired)
681 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200682
683 /*
684 * might be due to a server IP change.
685 * Let's trigger a DNS resolution if none are currently running.
686 */
Olivier Houchard0923fa42019-01-11 18:43:04 +0100687 if (check->server)
688 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200689
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100690 }
691 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
692 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200693 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100694 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
695 else if (expired)
696 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
697 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200698 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100699 /* I/O error after connection was established and before we could diagnose */
700 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
701 }
702 else if (expired) {
703 /* connection established but expired check */
704 if (check->type == PR_O2_SSL3_CHK)
705 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
706 else /* HTTP, SMTP, ... */
707 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
708 }
709
710 return;
711}
712
Olivier Houchard5c110b92018-08-14 17:04:58 +0200713/* This function checks if any I/O is wanted, and if so, attempts to do so */
714static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200715{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200716 struct check *check = ctx;
717 struct conn_stream *cs = check->cs;
Olivier Houchard0923fa42019-01-11 18:43:04 +0100718 struct email_alertq *q = container_of(check, typeof(*q), check);
Olivier Houchard81badc42019-07-09 17:28:51 +0200719 int ret = 0;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200720
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100721 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard81badc42019-07-09 17:28:51 +0200722 ret = wake_srv_chk(cs);
723 if (ret == 0 && !(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100724 if (check->server)
725 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
726 else
727 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200728 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100729 if (check->server)
730 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
731 else
732 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200733 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200734 return NULL;
735}
736
737/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100738 *
739 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200740 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
741 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200742 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200743static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200744{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200745 struct connection *conn = cs->conn;
746 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900747 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900748 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200749
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100750 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100751 goto out_wakeup;
752
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200753 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100754 if (!(conn->flags & CO_FL_ERROR))
755 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200756 goto out;
757 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100758
Willy Tarreau20a18342013-12-05 00:31:46 +0100759 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200760 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100761 goto out_wakeup;
762 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100763
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200764 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100765 /* if the output is closed, we can't do anything */
766 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200767 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100768 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200769 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200770
Willy Tarreau06559ac2013-12-05 01:53:08 +0100771 /* here, we know that the connection is established. That's enough for
772 * a pure TCP check.
773 */
774 if (!check->type)
775 goto out_wakeup;
776
Willy Tarreauc09572f2017-10-04 11:58:22 +0200777 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100778 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200779 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200780
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200781 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200782 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200783 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200784 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200785 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100786 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200787 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200788 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100789 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200790 goto out;
791 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100792 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200793
Willy Tarreau100cc4b2019-09-05 18:24:46 +0200794 if (!b_data(&check->bo))
795 conn_xprt_stop_send(conn);
796
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100797 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
798 if (s->proxy->timeout.check) {
799 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
800 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200801 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200802 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100803
Willy Tarreau83749182007-04-15 20:56:27 +0200804 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200805 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200806 out:
807 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200808}
809
Willy Tarreaubaaee002006-06-26 02:48:02 +0200810/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200811 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200812 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900813 * set_server_check_status() to update check->status, check->duration
814 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200815
816 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
817 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
818 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
819 * response to an SSL HELLO (the principle is that this is enough to
820 * distinguish between an SSL server and a pure TCP relay). All other cases will
821 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
822 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100823 *
824 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200825 * via the out label.
826 *
827 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200828 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200829static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200830{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200831 struct connection *conn = cs->conn;
832 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900833 struct server *s = check->server;
834 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200835 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100836 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200837 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200838
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100839 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200840 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200841
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200842 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100843 if (!(conn->flags & CO_FL_ERROR))
844 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200845 goto out;
846 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200847
Willy Tarreauc09572f2017-10-04 11:58:22 +0200848 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100849 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200850 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200851
Willy Tarreau83749182007-04-15 20:56:27 +0200852 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
853 * but the connection was closed on the remote end. Fortunately, recv still
854 * works correctly and we don't need to do the getsockopt() on linux.
855 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000856
857 /* Set buffer to point to the end of the data already read, and check
858 * that there is free space remaining. If the buffer is full, proceed
859 * with running the checks without attempting another socket read.
860 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000861
Willy Tarreau03938182010-03-17 21:52:07 +0100862 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000863
Olivier Houchard511efea2018-08-16 15:30:32 +0200864 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200865 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100866 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200867 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200868 /* Report network errors only if we got no other data. Otherwise
869 * we'll let the upper layers decide whether the response is OK
870 * or not. It is very common that an RST sent by the server is
871 * reported as an error just after the last data chunk.
872 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200873 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100874 goto out_wakeup;
875 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200876 }
877
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100878
Willy Tarreau03938182010-03-17 21:52:07 +0100879 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200880 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100881 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200882 if (b_data(&check->bi) < b_size(&check->bi))
883 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100884 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200885 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100886 done = 1; /* buffer full, don't wait for more data */
887 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200888
Nick Chalk57b1bf72010-03-16 15:50:46 +0000889 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900890 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200891 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200892 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100893 goto wait_more_data;
894
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100895 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200896 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
897 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
898 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
899 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
900 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
901 cut_crlf(b_head(&check->bi));
902 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200903
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100904 goto out_wakeup;
905 }
906
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200907 check->code = str2uic(b_head(&check->bi) + 9);
908 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200909
Willy Tarreaubd741542010-03-16 18:46:54 +0100910 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200911 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000912 /* 404 may be accepted as "stopping" only if the server was up */
913 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900914 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000915 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100916 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
917 /* Run content verification check... We know we have at least 13 chars */
918 if (!httpchk_expect(s, done))
919 goto wait_more_data;
920 }
921 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200922 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100923 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900924 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100925 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000926 else {
927 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900928 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000929 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200930 break;
931
932 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200933 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100934 goto wait_more_data;
935
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100936 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200937 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900938 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200939 else
Simon Horman4a741432013-02-23 15:35:38 +0900940 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200941 break;
942
943 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200944 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100945 goto wait_more_data;
946
Willy Tarreau55706782019-08-06 16:26:31 +0200947 /* do not reset when closing, servers don't like this */
948 if (conn_ctrl_ready(cs->conn))
949 fdtab[cs->conn->handle.fd].linger_risk = 0;
950
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200951 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200952 if ((b_data(&check->bi) < strlen("000\r")) ||
953 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
954 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
955 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
956 cut_crlf(b_head(&check->bi));
957 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200958 goto out_wakeup;
959 }
960
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200961 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200962
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200963 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200964 cut_crlf(desc);
965
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100966 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200967 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900968 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200969 else
Simon Horman4a741432013-02-23 15:35:38 +0900970 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200971 break;
972
Simon Hormana2b9dad2013-02-12 10:45:54 +0900973 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100974 int status = HCHK_STATUS_CHECKED;
975 const char *hs = NULL; /* health status */
976 const char *as = NULL; /* admin status */
977 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200978 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100979 const char *err = NULL; /* first error to report */
980 const char *wrn = NULL; /* first warning to report */
981 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900982
Willy Tarreau81f5d942013-12-09 20:51:51 +0100983 /* We're getting an agent check response. The agent could
984 * have been disabled in the mean time with a long check
985 * still pending. It is important that we ignore the whole
986 * response.
987 */
988 if (!(check->server->agent.state & CHK_ST_ENABLED))
989 break;
990
991 /* The agent supports strings made of a single line ended by the
992 * first CR ('\r') or LF ('\n'). This line is composed of words
993 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
994 * line may optionally contained a description of a state change
995 * after a sharp ('#'), which is only considered if a health state
996 * is announced.
997 *
998 * Words may be composed of :
999 * - a numeric weight suffixed by the percent character ('%').
1000 * - a health status among "up", "down", "stopped", and "fail".
1001 * - an admin status among "ready", "drain", "maint".
1002 *
1003 * These words may appear in any order. If multiple words of the
1004 * same category appear, the last one wins.
1005 */
1006
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001007 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001008 while (*p && *p != '\n' && *p != '\r')
1009 p++;
1010
1011 if (!*p) {
1012 if (!done)
1013 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001014
Willy Tarreau9809b782013-12-11 21:40:11 +01001015 /* at least inform the admin that the agent is mis-behaving */
1016 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1017 break;
1018 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001019
Willy Tarreau9809b782013-12-11 21:40:11 +01001020 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001021 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001022
Willy Tarreau81f5d942013-12-09 20:51:51 +01001023 while (*cmd) {
1024 /* look for next word */
1025 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1026 cmd++;
1027 continue;
1028 }
Simon Horman671b6f02013-11-25 10:46:39 +09001029
Willy Tarreau81f5d942013-12-09 20:51:51 +01001030 if (*cmd == '#') {
1031 /* this is the beginning of a health status description,
1032 * skip the sharp and blanks.
1033 */
1034 cmd++;
1035 while (*cmd == '\t' || *cmd == ' ')
1036 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001037 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001038 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001039
1040 /* find the end of the word so that we have a null-terminated
1041 * word between <cmd> and <p>.
1042 */
1043 p = cmd + 1;
1044 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1045 p++;
1046 if (*p)
1047 *p++ = 0;
1048
1049 /* first, health statuses */
1050 if (strcasecmp(cmd, "up") == 0) {
1051 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001052 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001053 hs = cmd;
1054 }
1055 else if (strcasecmp(cmd, "down") == 0) {
1056 check->health = 0;
1057 status = HCHK_STATUS_L7STS;
1058 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001059 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001060 else if (strcasecmp(cmd, "stopped") == 0) {
1061 check->health = 0;
1062 status = HCHK_STATUS_L7STS;
1063 hs = cmd;
1064 }
1065 else if (strcasecmp(cmd, "fail") == 0) {
1066 check->health = 0;
1067 status = HCHK_STATUS_L7STS;
1068 hs = cmd;
1069 }
1070 /* admin statuses */
1071 else if (strcasecmp(cmd, "ready") == 0) {
1072 as = cmd;
1073 }
1074 else if (strcasecmp(cmd, "drain") == 0) {
1075 as = cmd;
1076 }
1077 else if (strcasecmp(cmd, "maint") == 0) {
1078 as = cmd;
1079 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001080 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001081 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1082 ps = cmd;
1083 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001084 /* try to parse a maxconn here */
1085 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1086 cs = cmd;
1087 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001088 else {
1089 /* keep a copy of the first error */
1090 if (!err)
1091 err = cmd;
1092 }
1093 /* skip to next word */
1094 cmd = p;
1095 }
1096 /* here, cmd points either to \0 or to the beginning of a
1097 * description. Skip possible leading spaces.
1098 */
1099 while (*cmd == ' ' || *cmd == '\n')
1100 cmd++;
1101
1102 /* First, update the admin status so that we avoid sending other
1103 * possibly useless warnings and can also update the health if
1104 * present after going back up.
1105 */
1106 if (as) {
1107 if (strcasecmp(as, "drain") == 0)
1108 srv_adm_set_drain(check->server);
1109 else if (strcasecmp(as, "maint") == 0)
1110 srv_adm_set_maint(check->server);
1111 else
1112 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001113 }
1114
Willy Tarreau81f5d942013-12-09 20:51:51 +01001115 /* now change weights */
1116 if (ps) {
1117 const char *msg;
1118
1119 msg = server_parse_weight_change_request(s, ps);
1120 if (!wrn || !*wrn)
1121 wrn = msg;
1122 }
1123
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001124 if (cs) {
1125 const char *msg;
1126
1127 cs += strlen("maxconn:");
1128
1129 msg = server_parse_maxconn_change_request(s, cs);
1130 if (!wrn || !*wrn)
1131 wrn = msg;
1132 }
1133
Willy Tarreau81f5d942013-12-09 20:51:51 +01001134 /* and finally health status */
1135 if (hs) {
1136 /* We'll report some of the warnings and errors we have
1137 * here. Down reports are critical, we leave them untouched.
1138 * Lack of report, or report of 'UP' leaves the room for
1139 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001140 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001141 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001142 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001143
1144 if (!*msg || status == HCHK_STATUS_L7OKD) {
1145 if (err && *err)
1146 msg = err;
1147 else if (wrn && *wrn)
1148 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001149 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001150
1151 t = get_trash_chunk();
1152 chunk_printf(t, "via agent : %s%s%s%s",
1153 hs, *msg ? " (" : "",
1154 msg, *msg ? ")" : "");
1155
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001156 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001157 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001158 else if (err && *err) {
1159 /* No status change but we'd like to report something odd.
1160 * Just report the current state and copy the message.
1161 */
1162 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001163 set_server_check_status(check, status/*check->status*/,
1164 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001165
Willy Tarreau81f5d942013-12-09 20:51:51 +01001166 }
1167 else if (wrn && *wrn) {
1168 /* No status change but we'd like to report something odd.
1169 * Just report the current state and copy the message.
1170 */
1171 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001172 set_server_check_status(check, status/*check->status*/,
1173 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001174 }
1175 else
1176 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001177 break;
1178 }
1179
Willy Tarreau1620ec32011-08-06 17:05:02 +02001180 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001181 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001182 goto wait_more_data;
1183
Willy Tarreau55706782019-08-06 16:26:31 +02001184 /* do not reset when closing, servers don't like this */
1185 if (conn_ctrl_ready(cs->conn))
1186 fdtab[cs->conn->handle.fd].linger_risk = 0;
1187
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001188 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001189 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001190 }
1191 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001192 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1193 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001194 else
1195 desc = "PostgreSQL unknown error";
1196
Simon Horman4a741432013-02-23 15:35:38 +09001197 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001198 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001199 break;
1200
1201 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001202 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001203 goto wait_more_data;
1204
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001206 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001207 }
1208 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001209 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001210 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001211 break;
1212
1213 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001214 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001215 goto wait_more_data;
1216
Willy Tarreau55706782019-08-06 16:26:31 +02001217 /* do not reset when closing, servers don't like this */
1218 if (conn_ctrl_ready(cs->conn))
1219 fdtab[cs->conn->handle.fd].linger_risk = 0;
1220
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001221 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001222 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001223 /* We set the MySQL Version in description for information purpose
1224 * FIXME : it can be cool to use MySQL Version for other purpose,
1225 * like mark as down old MySQL server.
1226 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001227 if (b_data(&check->bi) > 51) {
1228 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001229 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001230 }
1231 else {
1232 if (!done)
1233 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001234
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001235 /* it seems we have a OK packet but without a valid length,
1236 * it must be a protocol error
1237 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001238 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001240 }
1241 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001242 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001243 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001244 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 }
1246 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001247 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1248 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1249 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001250
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001251 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001252 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001253 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001254 /* We have only one MySQL packet and it is a Handshake Initialization packet
1255 * but we need to have a second packet to know if it is alright
1256 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001257 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001258 goto wait_more_data;
1259 }
1260 else {
1261 /* We have only one packet and it is an Error packet,
1262 * an error message is attached, so we can display it
1263 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001264 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001265 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001266 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001267 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001268 } else if (b_data(&check->bi) > first_packet_len + 4) {
1269 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1270 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1271 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001272
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001273 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001274 /* We have 2 packets and that's good */
1275 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001276 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001277 /* No error packet */
1278 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001279 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001280 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001281 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001282 }
1283 else {
1284 /* An error message is attached in the Error packet
1285 * so we can display it ! :)
1286 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001287 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001288 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001289 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001290 }
1291 }
1292 }
1293 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001294 if (!done)
1295 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001296
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001297 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001298 * it must be a protocol error
1299 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001300 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001301 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001302 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001303 }
1304 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001305 break;
1306
1307 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001308 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001309 goto wait_more_data;
1310
1311 /* Check if the server speaks LDAP (ASN.1/BER)
1312 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1313 * http://tools.ietf.org/html/rfc4511
1314 */
1315
1316 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1317 * LDAPMessage: 0x30: SEQUENCE
1318 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001319 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001320 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001321 }
1322 else {
1323 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001324 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001325
1326 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1327 * messageID: 0x02 0x01 0x01: INTEGER 1
1328 * protocolOp: 0x61: bindResponse
1329 */
1330 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001331 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001332 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001333 goto out_wakeup;
1334 }
1335
1336 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001337 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001338
1339 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1340 * ldapResult: 0x0a 0x01: ENUMERATION
1341 */
1342 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001343 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001344 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001345 goto out_wakeup;
1346 }
1347
1348 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1349 * resultCode
1350 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001351 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001352 if (check->code) {
1353 set_server_check_status(check, HCHK_STATUS_L7STS, "See RFC: http://tools.ietf.org/html/rfc4511#section-4.1.9");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001354 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001355 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001356 }
1357 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001358 break;
1359
Christopher Fauletba7bc162016-11-07 21:07:38 +01001360 case PR_O2_SPOP_CHK: {
1361 unsigned int framesz;
1362 char err[HCHK_DESC_LEN];
1363
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001364 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001365 goto wait_more_data;
1366
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001367 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001368 framesz = ntohl(framesz);
1369
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001370 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001371 goto wait_more_data;
1372
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001373 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001374 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1375 else
1376 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1377 break;
1378 }
1379
Willy Tarreau1620ec32011-08-06 17:05:02 +02001380 default:
Willy Tarreaufbeee552019-09-05 18:43:22 +02001381 /* good connection is enough for pure TCP check */
1382 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
1383 if (check->use_ssl)
1384 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
1385 else
1386 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
1387 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001388 break;
1389 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001390
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001391 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001392 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001393 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001394 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001395
Nick Chalk57b1bf72010-03-16 15:50:46 +00001396 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001397 *b_head(&check->bi) = '\0';
1398 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001399
Steven Davidovitz544d4812017-03-08 11:06:20 -08001400 /* Close the connection... We still attempt to nicely close if,
1401 * for instance, SSL needs to send a "close notify." Later, we perform
1402 * a hard close and reset the connection if some data are pending,
1403 * otherwise we end up with many TIME_WAITs and eat all the source port
1404 * range quickly. To avoid sending RSTs all the time, we first try to
1405 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001406 */
Olivier Houchard8aa445d2019-07-02 16:35:18 +02001407 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1408 * connection, to make sure cs_shutw() will not lead to a shutdown()
1409 * that would provoke TIME_WAITs.
1410 */
1411 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001412 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001413
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001414 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001415 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001416 conn->flags |= CO_FL_ERROR;
1417
Willy Tarreaufdccded2008-08-29 18:19:04 +02001418 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001419out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001420 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001421
1422 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001423 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001424 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001425}
1426
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001427/*
1428 * This function is used only for server health-checks. It handles connection
1429 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001430 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1431 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001432 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001433static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001434{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001435 struct connection *conn = cs->conn;
1436 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001437 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001438 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001439
Olivier Houchard0923fa42019-01-11 18:43:04 +01001440 if (check->server)
1441 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1442 else
1443 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001444
Willy Tarreauc09572f2017-10-04 11:58:22 +02001445 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001446 if (check->type == PR_O2_TCPCHK_CHK) {
1447 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001448 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001449 conn = cs->conn;
Willy Tarreauab160a42019-09-05 17:38:40 +02001450 } else {
1451 if (!(check->wait_list.events & SUB_RETRY_SEND))
1452 __event_srv_chk_w(cs);
1453 if (!(check->wait_list.events & SUB_RETRY_RECV))
1454 __event_srv_chk_r(cs);
1455 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001456
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001457 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001458 /* We may get error reports bypassing the I/O handlers, typically
1459 * the case when sending a pure TCP check which fails, then the I/O
1460 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001461 * main processing task so let's simply wake it up. If we get here,
1462 * we expect errno to still be valid.
1463 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001464 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001465 task_wakeup(check->task, TASK_WOKEN_IO);
1466 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001467 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001468 /* we may get here if only a connection probe was required : we
1469 * don't have any data to send nor anything expected in response,
1470 * so the completion of the connection establishment is enough.
1471 */
1472 task_wakeup(check->task, TASK_WOKEN_IO);
1473 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001474
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001475 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001476 /* Check complete or aborted. If connection not yet closed do it
1477 * now and wake the check task up to be sure the result is
1478 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001479 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001480 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001481 ret = -1;
Olivier Houchard865ed9b2019-07-02 17:42:22 +02001482 /* We may have been scheduled to run, and the
1483 * I/O handler expects to have a cs, so remove
1484 * the tasklet
1485 */
1486 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001487 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001488 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001489
Olivier Houchard0923fa42019-01-11 18:43:04 +01001490 if (check->server)
1491 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1492 else
1493 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001494
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001495 /* if a connection got replaced, we must absolutely prevent the connection
1496 * handler from touching its fd, and perform the FD polling updates ourselves
1497 */
1498 if (ret < 0)
1499 conn_cond_update_polling(conn);
1500
1501 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001502}
1503
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001504struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001505 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001506 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001507};
1508
Willy Tarreaubaaee002006-06-26 02:48:02 +02001509/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001510 * updates the server's weight during a warmup stage. Once the final weight is
1511 * reached, the task automatically stops. Note that any server status change
1512 * must have updated s->last_change accordingly.
1513 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001514static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001515{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001516 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001517
1518 /* by default, plan on stopping the task */
1519 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001520 if ((s->next_admin & SRV_ADMF_MAINT) ||
1521 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001522 return t;
1523
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001524 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1525
Willy Tarreau892337c2014-05-13 23:41:20 +02001526 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001527 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001528
1529 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001530 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001531
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001532 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1533
Willy Tarreau2e993902011-10-31 11:53:20 +01001534 /* get back there in 1 second or 1/20th of the slowstart interval,
1535 * whichever is greater, resulting in small 5% steps.
1536 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001537 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001538 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1539 return t;
1540}
1541
Willy Tarreau894c6422017-10-04 15:58:52 +02001542/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1543 * none was found.
1544 */
1545static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1546{
1547 struct tcpcheck_rule *r;
1548
1549 list_for_each_entry(r, list, list) {
1550 if (r->action != TCPCHK_ACT_COMMENT)
1551 return r;
1552 }
1553 return NULL;
1554}
1555
Willy Tarreau2e993902011-10-31 11:53:20 +01001556/*
Simon Horman98637e52014-06-20 12:30:16 +09001557 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001558 *
1559 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001560 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1561 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1562 * - SF_ERR_SRVTO if there are no more servers
1563 * - SF_ERR_SRVCL if the connection was refused by the server
1564 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1565 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1566 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001567 * - SF_ERR_CHK_PORT if no port could be found to run a health check on an AF_INET* socket
Tim Düsterhus4896c442016-11-29 02:15:19 +01001568 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001569 * Note that we try to prevent the network stack from sending the ACK during the
1570 * connect() when a pure TCP check is used (without PROXY protocol).
1571 */
Simon Horman98637e52014-06-20 12:30:16 +09001572static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001573{
1574 struct check *check = t->context;
1575 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001576 struct conn_stream *cs = check->cs;
1577 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001578 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001579 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001581 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001582
Willy Tarreau00149122017-10-04 18:05:01 +02001583 /* we cannot have a connection here */
1584 if (conn)
1585 return SF_ERR_INTERNAL;
1586
Simon Hormanb00d17a2014-06-13 16:18:16 +09001587 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001588 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001589 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001590 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1591 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001592
1593 /* prepare the check buffer.
1594 * This should not be used if check is the secondary agent check
1595 * of a server as s->proxy->check_req will relate to the
1596 * configuration of the primary check. Similarly, tcp-check uses
1597 * its own strings.
1598 */
1599 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001600 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001601
1602 /* we want to check if this host replies to HTTP or SSLv3 requests
1603 * so we'll send the request, and won't wake the checker up now.
1604 */
1605 if ((check->type) == PR_O2_SSL3_CHK) {
1606 /* SSL requires that we put Unix time in the request */
1607 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001608 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001609 }
1610 else if ((check->type) == PR_O2_HTTP_CHK) {
1611 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001612 b_putblk(&check->bo, trash.area,
1613 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001614 /* prevent HTTP keep-alive when "http-check expect" is used */
1615 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001616 b_putist(&check->bo, ist("Connection: close\r\n"));
1617 b_putist(&check->bo, ist("\r\n"));
1618 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001619 }
1620 }
1621
James Brown55f9ff12015-10-21 18:19:05 -07001622 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001623 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001624 }
1625
Willy Tarreauf411cce2017-10-04 16:21:19 +02001626 /* for tcp-checks, the initial connection setup is handled separately as
1627 * it may be sent to a specific port and not to the server's.
1628 */
1629 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1630 tcpcheck_main(check);
1631 return SF_ERR_UP;
1632 }
1633
Simon Hormanb00d17a2014-06-13 16:18:16 +09001634 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001635 cs = check->cs = cs_new(NULL);
1636 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001637 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001638 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001639 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001640 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001641
Simon Horman41f58762015-01-30 11:22:56 +09001642 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001643 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001644 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001645 }
1646 else {
1647 /* we'll connect to the addr on the server */
1648 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001649 }
1650
Alexander Liu2a54bb72019-05-22 19:44:48 +08001651 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1652 conn->send_proxy_ofs = 1;
1653 conn->flags |= CO_FL_SOCKS4;
1654 }
1655
Olivier Houchard6377a002017-12-01 22:04:05 +01001656 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001657 conn->target = &s->obj_type;
1658
Willy Tarreaube373152018-09-06 11:45:30 +02001659 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001660 int i = 0;
1661
1662 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001663 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001664 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001665
1666 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001667 }
1668
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001669 /* no client address */
1670 clear_addr(&conn->addr.from);
1671
Willy Tarreaube373152018-09-06 11:45:30 +02001672 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001673 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1674 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001675 cs_attach(cs, check, &check_conn_cb);
1676
Willy Tarreauf3d34822014-12-08 12:11:28 +01001677 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001678 if (check->type != 0)
1679 connflags |= CONNECT_HAS_DATA;
1680 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1681 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1682 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001683
Willy Tarreaue7dff022015-04-03 01:14:29 +02001684 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001685 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001686 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001687
Willy Tarreau16257f62017-11-02 15:45:00 +01001688
Olivier Houchard9130a962017-10-17 17:33:43 +02001689#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001690 if (ret == SF_ERR_NONE) {
1691 if (s->check.sni)
1692 ssl_sock_set_servername(conn, s->check.sni);
1693 if (s->check.alpn_str)
1694 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1695 s->check.alpn_len);
1696 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001697#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001698 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001699 conn->send_proxy_ofs = 1;
1700 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001701 if (xprt_add_hs(conn) < 0)
1702 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001703 }
1704
1705 return ret;
1706}
1707
Simon Horman98637e52014-06-20 12:30:16 +09001708static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001709static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001710__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001711
1712void block_sigchld(void)
1713{
1714 sigset_t set;
1715 sigemptyset(&set);
1716 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001717 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001718}
1719
1720void unblock_sigchld(void)
1721{
1722 sigset_t set;
1723 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001724 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001725 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001726}
1727
Simon Horman98637e52014-06-20 12:30:16 +09001728static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1729{
1730 struct pid_list *elem;
1731 struct check *check = t->context;
1732
Willy Tarreaubafbe012017-11-24 17:34:44 +01001733 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001734 if (!elem)
1735 return NULL;
1736 elem->pid = pid;
1737 elem->t = t;
1738 elem->exited = 0;
1739 check->curpid = elem;
1740 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001741
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001742 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001743 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001744 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001745
Simon Horman98637e52014-06-20 12:30:16 +09001746 return elem;
1747}
1748
Simon Horman98637e52014-06-20 12:30:16 +09001749static void pid_list_del(struct pid_list *elem)
1750{
1751 struct check *check;
1752
1753 if (!elem)
1754 return;
1755
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001756 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001757 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001758 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001759
Simon Horman98637e52014-06-20 12:30:16 +09001760 if (!elem->exited)
1761 kill(elem->pid, SIGTERM);
1762
1763 check = elem->t->context;
1764 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001765 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001766}
1767
1768/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1769static void pid_list_expire(pid_t pid, int status)
1770{
1771 struct pid_list *elem;
1772
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001773 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001774 list_for_each_entry(elem, &pid_list, list) {
1775 if (elem->pid == pid) {
1776 elem->t->expire = now_ms;
1777 elem->status = status;
1778 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001779 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001780 break;
Simon Horman98637e52014-06-20 12:30:16 +09001781 }
1782 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001783 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001784}
1785
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001786static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001787{
1788 pid_t pid;
1789 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001790
Simon Horman98637e52014-06-20 12:30:16 +09001791 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1792 pid_list_expire(pid, status);
1793}
1794
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001795static int init_pid_list(void)
1796{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001797 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001798 /* Nothing to do */
1799 return 0;
1800
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001801 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001802 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1803 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001804 return 1;
1805 }
1806
Willy Tarreaubafbe012017-11-24 17:34:44 +01001807 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1808 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001809 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1810 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001811 return 1;
1812 }
1813
1814 return 0;
1815}
1816
Cyril Bontéac92a062014-12-27 22:28:38 +01001817/* helper macro to set an environment variable and jump to a specific label on failure. */
1818#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001819
1820/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001821 * helper function to allocate enough memory to store an environment variable.
1822 * It will also check that the environment variable is updatable, and silently
1823 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001824 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001825static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001826{
1827 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001828 char *envname;
1829 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001830
Cyril Bontéac92a062014-12-27 22:28:38 +01001831 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001832 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001833 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001834 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001835
1836 envname = extcheck_envs[idx].name;
1837 vmaxlen = extcheck_envs[idx].vmaxlen;
1838
1839 /* Check if the environment variable is already set, and silently reject
1840 * the update if this one is not updatable. */
1841 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1842 return 0;
1843
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001844 /* Instead of sending NOT_USED, sending an empty value is preferable */
1845 if (strcmp(value, "NOT_USED") == 0) {
1846 value = "";
1847 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001848
1849 len = strlen(envname) + 1;
1850 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1851 len += strlen(value);
1852 else
1853 len += vmaxlen;
1854
1855 if (!check->envp[idx])
1856 check->envp[idx] = malloc(len + 1);
1857
1858 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001859 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001860 return 1;
1861 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001862 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001863 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001864 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001865 return 1;
1866 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001867 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001868 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001869 return 1;
1870 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001871 return 0;
1872}
Simon Horman98637e52014-06-20 12:30:16 +09001873
1874static int prepare_external_check(struct check *check)
1875{
1876 struct server *s = check->server;
1877 struct proxy *px = s->proxy;
1878 struct listener *listener = NULL, *l;
1879 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001880 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001881 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001882
1883 list_for_each_entry(l, &px->conf.listeners, by_fe)
1884 /* Use the first INET, INET6 or UNIX listener */
1885 if (l->addr.ss_family == AF_INET ||
1886 l->addr.ss_family == AF_INET6 ||
1887 l->addr.ss_family == AF_UNIX) {
1888 listener = l;
1889 break;
1890 }
1891
Simon Horman98637e52014-06-20 12:30:16 +09001892 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001893 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1894 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001895 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001896 goto err;
1897 }
Simon Horman98637e52014-06-20 12:30:16 +09001898
Cyril Bontéac92a062014-12-27 22:28:38 +01001899 check->argv = calloc(6, sizeof(char *));
1900 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001901 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001902 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001903 }
Simon Horman98637e52014-06-20 12:30:16 +09001904
1905 check->argv[0] = px->check_command;
1906
Cyril Bonté777be862014-12-02 21:21:35 +01001907 if (!listener) {
1908 check->argv[1] = strdup("NOT_USED");
1909 check->argv[2] = strdup("NOT_USED");
1910 }
1911 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001912 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001913 addr_to_str(&listener->addr, buf, sizeof(buf));
1914 check->argv[1] = strdup(buf);
1915 port_to_str(&listener->addr, buf, sizeof(buf));
1916 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001917 }
1918 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001919 const struct sockaddr_un *un;
1920
1921 un = (struct sockaddr_un *)&listener->addr;
1922 check->argv[1] = strdup(un->sun_path);
1923 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001924 }
1925 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001926 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001927 goto err;
1928 }
1929
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001930 addr_to_str(&s->addr, buf, sizeof(buf));
1931 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001932
1933 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1934 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1935 else
1936 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001937 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001938
Cyril Bontéac92a062014-12-27 22:28:38 +01001939 for (i = 0; i < 5; i++) {
1940 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001941 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001942 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001943 }
1944 }
Simon Horman98637e52014-06-20 12:30:16 +09001945
Cyril Bontéac92a062014-12-27 22:28:38 +01001946 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001947 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001948 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1949 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1950 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1951 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001952 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001953 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1954 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1955 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1956 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1957 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1958 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1959
1960 /* Ensure that we don't leave any hole in check->envp */
1961 for (i = 0; i < EXTCHK_SIZE; i++)
1962 if (!check->envp[i])
1963 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001964
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001965 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001966err:
1967 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001968 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001969 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001970 free(check->envp);
1971 check->envp = NULL;
1972 }
1973
1974 if (check->argv) {
1975 for (i = 1; i < 5; i++)
1976 free(check->argv[i]);
1977 free(check->argv);
1978 check->argv = NULL;
1979 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001980 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001981}
1982
Simon Hormanb00d17a2014-06-13 16:18:16 +09001983/*
Simon Horman98637e52014-06-20 12:30:16 +09001984 * establish a server health-check that makes use of a process.
1985 *
1986 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001987 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001988 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001989 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001990 *
1991 * Blocks and then unblocks SIGCHLD
1992 */
1993static int connect_proc_chk(struct task *t)
1994{
Cyril Bontéac92a062014-12-27 22:28:38 +01001995 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001996 struct check *check = t->context;
1997 struct server *s = check->server;
1998 struct proxy *px = s->proxy;
1999 int status;
2000 pid_t pid;
2001
Willy Tarreaue7dff022015-04-03 01:14:29 +02002002 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002003
2004 block_sigchld();
2005
2006 pid = fork();
2007 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002008 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
2009 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002010 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2011 goto out;
2012 }
2013 if (pid == 0) {
2014 /* Child */
2015 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002016 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002017 int fd;
2018
2019 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2020 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2021
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002022 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002023
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002024 /* restore the initial FD limits */
2025 limit.rlim_cur = rlim_fd_cur_at_boot;
2026 limit.rlim_max = rlim_fd_max_at_boot;
2027 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2028 getrlimit(RLIMIT_NOFILE, &limit);
2029 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2030 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2031 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2032 }
2033
Simon Horman98637e52014-06-20 12:30:16 +09002034 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01002035 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Willy Tarreauafc313e2019-07-01 07:51:29 +02002036 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002037 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002038 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2039 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002040 exit(-1);
2041 }
2042
2043 /* Parent */
2044 if (check->result == CHK_RES_UNKNOWN) {
2045 if (pid_list_add(pid, t) != NULL) {
2046 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2047
2048 if (px->timeout.check && px->timeout.connect) {
2049 int t_con = tick_add(now_ms, px->timeout.connect);
2050 t->expire = tick_first(t->expire, t_con);
2051 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002052 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002053 goto out;
2054 }
2055 else {
2056 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2057 }
2058 kill(pid, SIGTERM); /* process creation error */
2059 }
2060 else
2061 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2062
2063out:
2064 unblock_sigchld();
2065 return status;
2066}
2067
2068/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002069 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002070 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002071 *
2072 * Please do NOT place any return statement in this function and only leave
2073 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002074 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002075static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002076{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002077 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002078 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002079 int rv;
2080 int ret;
2081 int expired = tick_is_expired(t->expire, now_ms);
2082
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002083 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002084 if (!(check->state & CHK_ST_INPROGRESS)) {
2085 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002086 if (!expired) /* woke up too early */
2087 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002088
2089 /* we don't send any health-checks when the proxy is
2090 * stopped, the server should not be checked or the check
2091 * is disabled.
2092 */
2093 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2094 s->proxy->state == PR_STSTOPPED)
2095 goto reschedule;
2096
2097 /* we'll initiate a new check */
2098 set_server_check_status(check, HCHK_STATUS_START, NULL);
2099
2100 check->state |= CHK_ST_INPROGRESS;
2101
Simon Hormandbf70192015-01-30 11:22:53 +09002102 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002103 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002104 /* the process was forked, we allow up to min(inter,
2105 * timeout.connect) for it to report its status, but
2106 * only when timeout.check is set as it may be to short
2107 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002108 */
2109 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2110
2111 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2112 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2113 t->expire = tick_first(t->expire, t_con);
2114 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002115 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002116 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002117 }
2118
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002119 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002120
2121 check->state &= ~CHK_ST_INPROGRESS;
2122 check_notify_failure(check);
2123
2124 /* we allow up to min(inter, timeout.connect) for a connection
2125 * to establish but only when timeout.check is set
2126 * as it may be to short for a full check otherwise
2127 */
2128 while (tick_is_expired(t->expire, now_ms)) {
2129 int t_con;
2130
2131 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2132 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2133
2134 if (s->proxy->timeout.check)
2135 t->expire = tick_first(t->expire, t_con);
2136 }
2137 }
2138 else {
2139 /* there was a test running.
2140 * First, let's check whether there was an uncaught error,
2141 * which can happen on connect timeout or error.
2142 */
2143 if (check->result == CHK_RES_UNKNOWN) {
2144 /* good connection is enough for pure TCP check */
2145 struct pid_list *elem = check->curpid;
2146 int status = HCHK_STATUS_UNKNOWN;
2147
2148 if (elem->exited) {
2149 status = elem->status; /* Save in case the process exits between use below */
2150 if (!WIFEXITED(status))
2151 check->code = -1;
2152 else
2153 check->code = WEXITSTATUS(status);
2154 if (!WIFEXITED(status) || WEXITSTATUS(status))
2155 status = HCHK_STATUS_PROCERR;
2156 else
2157 status = HCHK_STATUS_PROCOK;
2158 } else if (expired) {
2159 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002160 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002161 kill(elem->pid, SIGTERM);
2162 }
2163 set_server_check_status(check, status, NULL);
2164 }
2165
2166 if (check->result == CHK_RES_FAILED) {
2167 /* a failure or timeout detected */
2168 check_notify_failure(check);
2169 }
2170 else if (check->result == CHK_RES_CONDPASS) {
2171 /* check is OK but asks for stopping mode */
2172 check_notify_stopping(check);
2173 }
2174 else if (check->result == CHK_RES_PASSED) {
2175 /* a success was detected */
2176 check_notify_success(check);
2177 }
Willy Tarreaub1437112019-09-03 18:55:02 +02002178 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002179 check->state &= ~CHK_ST_INPROGRESS;
2180
2181 pid_list_del(check->curpid);
2182
2183 rv = 0;
2184 if (global.spread_checks > 0) {
2185 rv = srv_getinter(check) * global.spread_checks / 100;
2186 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2187 }
2188 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2189 }
2190
2191 reschedule:
2192 while (tick_is_expired(t->expire, now_ms))
2193 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002194
2195 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002196 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002197 return t;
2198}
2199
2200/*
2201 * manages a server health-check that uses a connection. Returns
2202 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002203 *
2204 * Please do NOT place any return statement in this function and only leave
2205 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002206 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002207static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002208{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002209 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002210 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002211 struct conn_stream *cs = check->cs;
2212 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002213 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002214 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002215 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002216
Olivier Houchard0923fa42019-01-11 18:43:04 +01002217 if (check->server)
2218 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002219 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002220 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002221 if (!expired) /* woke up too early */
2222 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002223
Simon Horman671b6f02013-11-25 10:46:39 +09002224 /* we don't send any health-checks when the proxy is
2225 * stopped, the server should not be checked or the check
2226 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002227 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002228 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002229 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002230 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002231
2232 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002233 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002234
Willy Tarreau2c115e52013-12-11 19:41:16 +01002235 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002236 b_reset(&check->bi);
2237 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002238
Simon Hormandbf70192015-01-30 11:22:53 +09002239 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002240 cs = check->cs;
2241 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002242
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002243 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002244 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002245 goto out_unlock;
2246
Willy Tarreaue7dff022015-04-03 01:14:29 +02002247 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002248 /* we allow up to min(inter, timeout.connect) for a connection
2249 * to establish but only when timeout.check is set
2250 * as it may be to short for a full check otherwise
2251 */
Simon Horman4a741432013-02-23 15:35:38 +09002252 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002253
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002254 if (proxy->timeout.check && proxy->timeout.connect) {
2255 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002256 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002257 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002258
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002259 if (check->type) {
2260 /* send the request if we have one. We avoid receiving
2261 * if not connected, unless we didn't subscribe for
2262 * sending since otherwise we won't be woken up.
2263 */
2264 __event_srv_chk_w(cs);
Willy Tarreauab160a42019-09-05 17:38:40 +02002265 if (!(conn->flags & CO_FL_WAIT_L4_CONN) ||
2266 !(check->wait_list.events & SUB_RETRY_SEND))
2267 __event_srv_chk_r(cs);
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002268 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002269
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002270 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002271 goto reschedule;
2272
Willy Tarreaue7dff022015-04-03 01:14:29 +02002273 case SF_ERR_SRVTO: /* ETIMEDOUT */
2274 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002275 if (conn)
2276 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002277 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002278 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002279 /* should share same code than cases below */
2280 case SF_ERR_CHK_PORT:
2281 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002282 case SF_ERR_PRXCOND:
2283 case SF_ERR_RESOURCE:
2284 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002285 if (conn)
2286 conn->flags |= CO_FL_ERROR;
2287 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002288 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002289 }
2290
Willy Tarreau5a78f362012-11-23 12:47:05 +01002291 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002292 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002293 if (check->wait_list.events)
2294 cs->conn->xprt->unsubscribe(cs->conn,
2295 cs->conn->xprt_ctx,
2296 check->wait_list.events,
2297 &check->wait_list);
2298 /* We may have been scheduled to run, and the
2299 * I/O handler expects to have a cs, so remove
2300 * the tasklet
2301 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002302 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002303 cs_destroy(cs);
2304 cs = check->cs = NULL;
2305 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002306 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002307
Willy Tarreau2c115e52013-12-11 19:41:16 +01002308 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002309 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002310
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002311 /* we allow up to min(inter, timeout.connect) for a connection
2312 * to establish but only when timeout.check is set
2313 * as it may be to short for a full check otherwise
2314 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002315 while (tick_is_expired(t->expire, now_ms)) {
2316 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002317
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002318 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002319 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002320
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002321 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002322 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002323 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002324 }
2325 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002326 /* there was a test running.
2327 * First, let's check whether there was an uncaught error,
2328 * which can happen on connect timeout or error.
2329 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002330 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002331 /* good connection is enough for pure TCP check */
2332 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002333 if (check->use_ssl)
2334 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002335 else
Simon Horman4a741432013-02-23 15:35:38 +09002336 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002337 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002338 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002339 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002340 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002341 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002342 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002343 }
2344
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002345 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002346 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002347 /* The check was aborted and the connection was not yet closed.
2348 * This can happen upon timeout, or when an external event such
2349 * as a failed response coupled with "observe layer7" caused the
2350 * server state to be suddenly changed.
2351 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002352 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002353 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002354 }
2355
Willy Tarreauac59f362017-10-08 11:10:19 +02002356 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002357 if (check->wait_list.events)
2358 cs->conn->xprt->unsubscribe(cs->conn,
2359 cs->conn->xprt_ctx,
2360 check->wait_list.events,
2361 &check->wait_list);
2362 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002363 * I/O handler expects to have a cs, so remove
2364 * the tasklet
2365 */
2366 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002367 cs_destroy(cs);
2368 cs = check->cs = NULL;
2369 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002370 }
2371
Olivier Houchard0923fa42019-01-11 18:43:04 +01002372 if (check->server) {
2373 if (check->result == CHK_RES_FAILED) {
2374 /* a failure or timeout detected */
2375 check_notify_failure(check);
2376 }
2377 else if (check->result == CHK_RES_CONDPASS) {
2378 /* check is OK but asks for stopping mode */
2379 check_notify_stopping(check);
2380 }
2381 else if (check->result == CHK_RES_PASSED) {
2382 /* a success was detected */
2383 check_notify_success(check);
2384 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002385 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002386 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002387 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002388
Olivier Houchard0923fa42019-01-11 18:43:04 +01002389 if (check->server) {
2390 rv = 0;
2391 if (global.spread_checks > 0) {
2392 rv = srv_getinter(check) * global.spread_checks / 100;
2393 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2394 }
2395 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002396 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002397 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002398
2399 reschedule:
2400 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002401 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002402 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002403 if (check->server)
2404 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002405 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002406}
2407
Simon Horman98637e52014-06-20 12:30:16 +09002408/*
2409 * manages a server health-check. Returns
2410 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2411 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002412static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002413{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002414 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002415
2416 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002417 return process_chk_proc(t, context, state);
2418 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002419
Simon Horman98637e52014-06-20 12:30:16 +09002420}
2421
Simon Horman5c942422013-11-25 10:46:32 +09002422static int start_check_task(struct check *check, int mininter,
2423 int nbcheck, int srvpos)
2424{
2425 struct task *t;
Willy Tarreaub1437112019-09-03 18:55:02 +02002426 unsigned long thread_mask = MAX_THREADS_MASK;
2427
2428 if (check->type == PR_O2_EXT_CHK)
2429 thread_mask = 1;
2430
Simon Horman5c942422013-11-25 10:46:32 +09002431 /* task for the check */
Willy Tarreaub1437112019-09-03 18:55:02 +02002432 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002433 ha_alert("Starting [%s:%s] check: out of memory.\n",
2434 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002435 return 0;
2436 }
2437
2438 check->task = t;
2439 t->process = process_chk;
2440 t->context = check;
2441
Willy Tarreau1746eec2014-04-25 10:46:47 +02002442 if (mininter < srv_getinter(check))
2443 mininter = srv_getinter(check);
2444
2445 if (global.max_spread_checks && mininter > global.max_spread_checks)
2446 mininter = global.max_spread_checks;
2447
Simon Horman5c942422013-11-25 10:46:32 +09002448 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002449 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002450 check->start = now;
2451 task_queue(t);
2452
2453 return 1;
2454}
2455
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002456/*
2457 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002458 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002459 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002460static int start_checks()
2461{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002462
2463 struct proxy *px;
2464 struct server *s;
2465 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002466 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002467
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002468 /* 1- count the checkers to run simultaneously.
2469 * We also determine the minimum interval among all of those which
2470 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2471 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002472 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002473 * too short an interval for all others.
2474 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002475 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002476 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002477 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002478 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002479 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002480 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002481 }
2482 /* We need a warmup task that will be called when the server
2483 * state switches from down to up.
2484 */
2485 s->warmup = t;
2486 t->process = server_warmup;
2487 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002488 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002489 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002490 task_schedule(s->warmup, tick_add(now_ms, MS_TO_TICKS(MAX(1000, (now.tv_sec - s->last_change)) / 20)));
Willy Tarreaue7b73482013-11-21 11:50:50 +01002491 }
2492
Willy Tarreaud8514a22013-12-11 21:10:14 +01002493 if (s->check.state & CHK_ST_CONFIGURED) {
2494 nbcheck++;
2495 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2496 (!mininter || mininter > srv_getinter(&s->check)))
2497 mininter = srv_getinter(&s->check);
2498 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002499
Willy Tarreaud8514a22013-12-11 21:10:14 +01002500 if (s->agent.state & CHK_ST_CONFIGURED) {
2501 nbcheck++;
2502 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2503 (!mininter || mininter > srv_getinter(&s->agent)))
2504 mininter = srv_getinter(&s->agent);
2505 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002506 }
2507 }
2508
Simon Horman4a741432013-02-23 15:35:38 +09002509 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002510 return 0;
2511
2512 srand((unsigned)time(NULL));
2513
2514 /*
2515 * 2- start them as far as possible from each others. For this, we will
2516 * start them after their interval set to the min interval divided by
2517 * the number of servers, weighted by the server's position in the list.
2518 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002519 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002520 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2521 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002522 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002523 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002524 }
2525 }
2526
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002527 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002528 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002529 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002530 if (s->check.type == PR_O2_EXT_CHK) {
2531 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002532 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002533 }
Simon Hormand60d6912013-11-25 10:46:36 +09002534 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002535 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002536 srvpos++;
2537 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002538
Simon Hormand60d6912013-11-25 10:46:36 +09002539 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002540 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002541 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002542 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002543 }
2544 srvpos++;
2545 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002546 }
2547 }
2548 return 0;
2549}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002550
2551/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002552 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002553 * The buffer MUST be terminated by a null byte before calling this function.
2554 * Sets server status appropriately. The caller is responsible for ensuring
2555 * that the buffer contains at least 13 characters. If <done> is zero, we may
2556 * return 0 to indicate that data is required to decide of a match.
2557 */
2558static int httpchk_expect(struct server *s, int done)
2559{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002560 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002561 char status_code[] = "000";
2562 char *contentptr;
2563 int crlf;
2564 int ret;
2565
2566 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2567 case PR_O2_EXP_STS:
2568 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002569 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2570 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002571
2572 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2573 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2574 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002575 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002576
2577 /* we necessarily have the response, so there are no partial failures */
2578 if (s->proxy->options2 & PR_O2_EXP_INV)
2579 ret = !ret;
2580
Simon Horman4a741432013-02-23 15:35:38 +09002581 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002582 break;
2583
2584 case PR_O2_EXP_STR:
2585 case PR_O2_EXP_RSTR:
2586 /* very simple response parser: ignore CR and only count consecutive LFs,
2587 * stop with contentptr pointing to first char after the double CRLF or
2588 * to '\0' if crlf < 2.
2589 */
2590 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002591 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002592 if (crlf >= 2)
2593 break;
2594 if (*contentptr == '\r')
2595 continue;
2596 else if (*contentptr == '\n')
2597 crlf++;
2598 else
2599 crlf = 0;
2600 }
2601
2602 /* Check that response contains a body... */
2603 if (crlf < 2) {
2604 if (!done)
2605 return 0;
2606
Simon Horman4a741432013-02-23 15:35:38 +09002607 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002608 "HTTP content check could not find a response body");
2609 return 1;
2610 }
2611
2612 /* Check that response body is not empty... */
2613 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002614 if (!done)
2615 return 0;
2616
Simon Horman4a741432013-02-23 15:35:38 +09002617 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002618 "HTTP content check found empty response body");
2619 return 1;
2620 }
2621
2622 /* Check the response content against the supplied string
2623 * or regex... */
2624 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2625 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2626 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002627 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002628
2629 /* if we don't match, we may need to wait more */
2630 if (!ret && !done)
2631 return 0;
2632
2633 if (ret) {
2634 /* content matched */
2635 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002636 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002637 "HTTP check matched unwanted content");
2638 else
Simon Horman4a741432013-02-23 15:35:38 +09002639 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002640 "HTTP content check matched");
2641 }
2642 else {
2643 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002644 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002645 "HTTP check did not match unwanted content");
2646 else
Simon Horman4a741432013-02-23 15:35:38 +09002647 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002648 "HTTP content check did not match");
2649 }
2650 break;
2651 }
2652 return 1;
2653}
2654
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002655/*
2656 * return the id of a step in a send/expect session
2657 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002658static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002659{
2660 struct tcpcheck_rule *cur = NULL, *next = NULL;
2661 int i = 0;
2662
Willy Tarreau213c6782014-10-02 14:51:02 +02002663 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002664 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002665 return 0;
2666
Simon Hormane16c1b32015-01-30 11:22:57 +09002667 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002668
2669 /* no step => first step */
2670 if (cur == NULL)
2671 return 1;
2672
2673 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002674 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002675 if (next->list.p == &cur->list)
2676 break;
2677 ++i;
2678 }
2679
2680 return i;
2681}
2682
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002683/*
2684 * return the latest known comment before (including) the given stepid
2685 * returns NULL if no comment found
2686 */
2687static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2688{
2689 struct tcpcheck_rule *cur = NULL;
2690 char *ret = NULL;
2691 int i = 0;
2692
2693 /* not even started anything yet, return latest comment found before any action */
2694 if (!check->current_step) {
2695 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2696 if (cur->action == TCPCHK_ACT_COMMENT)
2697 ret = cur->comment;
2698 else
2699 goto return_comment;
2700 }
2701 }
2702
2703 i = 1;
2704 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2705 if (cur->comment)
2706 ret = cur->comment;
2707
2708 if (i >= stepid)
2709 goto return_comment;
2710
2711 ++i;
2712 }
2713
2714 return_comment:
2715 return ret;
2716}
2717
Willy Tarreaube74b882017-10-04 16:22:49 +02002718/* proceed with next steps for the TCP checks <check>. Note that this is called
2719 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002720 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2721 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002722 *
2723 * Please do NOT place any return statement in this function and only leave
Willy Tarreau495c3a42019-07-17 18:48:07 +02002724 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002725 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002726static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002727{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002728 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002729 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002730 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002731 struct conn_stream *cs = check->cs;
2732 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002733 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002734 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002735 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002736 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002737 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002738
Willy Tarreauef953952014-10-02 14:30:14 +02002739 /* here, we know that the check is complete or that it failed */
2740 if (check->result != CHK_RES_UNKNOWN)
2741 goto out_end_tcpcheck;
2742
2743 /* We have 4 possibilities here :
2744 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002745 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002746 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2747 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002748 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002749 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002750 * be waiting for a connection attempt to complete. conn!=NULL.
2751 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002752 *
2753 * #2 and #3 are quite similar, we want both the connection and the
2754 * handshake to complete before going any further. Thus we must always
2755 * wait for a connection to complete unless we're before and existing
2756 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002757 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002758
2759 /* find first rule and skip comments */
2760 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2761 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2762 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2763
Willy Tarreau00149122017-10-04 18:05:01 +02002764 if ((check->current_step || &next->list == head) &&
2765 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002766 /* we allow up to min(inter, timeout.connect) for a connection
2767 * to establish but only when timeout.check is set
2768 * as it may be to short for a full check otherwise
2769 */
2770 while (tick_is_expired(t->expire, now_ms)) {
2771 int t_con;
2772
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002773 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002774 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2775
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002776 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002777 t->expire = tick_first(t->expire, t_con);
2778 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002779 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002780 }
2781
Willy Tarreauef953952014-10-02 14:30:14 +02002782 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002783 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002784 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002785 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002786 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002787
Willy Tarreau213c6782014-10-02 14:51:02 +02002788 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002789 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002790 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002791 b_reset(&check->bo);
2792 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002793 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002794 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002795 if (proxy->timeout.check)
2796 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002797 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002798
Willy Tarreauabca5b62013-12-06 14:19:25 +01002799 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002800 /* We have to try to flush the output buffer before reading, at
2801 * the end, or if we're about to send a string that does not fit
2802 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002803 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002804 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002805 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002806 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002807 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002808 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002809 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002810
Olivier Houcharded0f2072018-08-16 15:41:52 +02002811 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002812 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002813
2814 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002815 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002816 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002817 goto out_end_tcpcheck;
2818 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002819 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002820 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002821 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002822 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002823 goto out;
2824 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002825 }
2826
Willy Tarreau263013d2015-05-13 11:59:14 +02002827 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002828 break;
2829
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002830 /* have 'next' point to the next rule or NULL if we're on the
2831 * last one, connect() needs this.
2832 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002833 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002834
2835 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002836 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002837 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002838
2839 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002840 if (&next->list == head)
2841 next = NULL;
2842
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002843 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2844 struct protocol *proto;
2845 struct xprt_ops *xprt;
2846
Willy Tarreau00149122017-10-04 18:05:01 +02002847 /* For a connect action we'll create a new connection.
2848 * We may also have to kill a previous one. But we don't
2849 * want to leave *without* a connection if we came here
2850 * from the connection layer, hence with a connection.
2851 * Thus we'll proceed in the following order :
2852 * 1: close but not release previous connection
2853 * 2: try to get a new connection
2854 * 3: release and replace the old one on success
2855 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002856 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002857 cs_close(check->cs);
2858 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002859 }
2860
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002861 /* mark the step as started */
2862 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002863
2864 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002865 cs = cs_new(NULL);
2866 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002867 step = tcpcheck_get_step_id(check);
2868 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2869 comment = tcpcheck_get_step_comment(check, step);
2870 if (comment)
2871 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002872 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2873 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002874 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002875 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002876 }
2877
Olivier Houchard49065542019-05-31 19:20:36 +02002878 if (check->cs) {
2879 if (check->wait_list.events)
2880 cs->conn->xprt->unsubscribe(cs->conn,
2881 cs->conn->xprt_ctx,
2882 check->wait_list.events,
2883 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002884 /* We may have been scheduled to run, and the
2885 * I/O handler expects to have a cs, so remove
2886 * the tasklet
2887 */
2888 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002889 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002890 }
Willy Tarreau00149122017-10-04 18:05:01 +02002891
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002892 check->cs = cs;
2893 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002894 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002895 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002896 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002897
2898 /* no client address */
2899 clear_addr(&conn->addr.from);
2900
Simon Horman41f58762015-01-30 11:22:56 +09002901 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002902 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002903 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002904 }
2905 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002906 /* we'll connect to the addr on the server */
2907 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002908 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002909 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002910
2911 /* port */
2912 if (check->current_step->port)
2913 set_host_port(&conn->addr.to, check->current_step->port);
2914 else if (check->port)
2915 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002916 else if (s->svc_port)
2917 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002918
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002919 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002920 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002921 }
2922 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002923 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002924 }
Willy Tarreaube373152018-09-06 11:45:30 +02002925
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002926 conn_prepare(conn, proto, xprt);
Willy Tarreau495c3a42019-07-17 18:48:07 +02002927
2928 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2929 ret = SF_ERR_RESOURCE;
2930 goto fail_check;
2931 }
2932
Willy Tarreaube373152018-09-06 11:45:30 +02002933 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002934
Willy Tarreaue7dff022015-04-03 01:14:29 +02002935 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002936 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002937 ret = proto->connect(conn,
Olivier Houchardfdcb0072019-05-06 18:32:29 +02002938 CONNECT_HAS_DATA /* I/O polling is always needed */ | (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : CONNECT_DELACK_ALWAYS);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002939 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2940 conn->send_proxy_ofs = 1;
2941 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002942 if (xprt_add_hs(conn) < 0)
2943 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002944 }
2945
2946 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002947 * - SF_ERR_NONE if everything's OK
2948 * - SF_ERR_SRVTO if there are no more servers
2949 * - SF_ERR_SRVCL if the connection was refused by the server
2950 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2951 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2952 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002953 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002954 * Note that we try to prevent the network stack from sending the ACK during the
2955 * connect() when a pure TCP check is used (without PROXY protocol).
2956 */
Willy Tarreau495c3a42019-07-17 18:48:07 +02002957 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002958 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002959 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002960 /* we allow up to min(inter, timeout.connect) for a connection
2961 * to establish but only when timeout.check is set
2962 * as it may be to short for a full check otherwise
2963 */
2964 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2965
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002966 if (proxy->timeout.check && proxy->timeout.connect) {
2967 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002968 t->expire = tick_first(t->expire, t_con);
2969 }
2970 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002971 case SF_ERR_SRVTO: /* ETIMEDOUT */
2972 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002973 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002974 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002975 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002976 comment = tcpcheck_get_step_comment(check, step);
2977 if (comment)
2978 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002979 set_server_check_status(check, HCHK_STATUS_L4CON,
2980 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002981 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002982 case SF_ERR_PRXCOND:
2983 case SF_ERR_RESOURCE:
2984 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002985 step = tcpcheck_get_step_id(check);
2986 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002987 comment = tcpcheck_get_step_comment(check, step);
2988 if (comment)
2989 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002990 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2991 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002992 goto out_end_tcpcheck;
2993 }
2994
2995 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002996 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002997
2998 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002999 while (&check->current_step->list != head &&
3000 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003001 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003002
Willy Tarreauf2c87352015-05-13 12:08:21 +02003003 if (&check->current_step->list == head)
3004 break;
3005
Willy Tarreau7199e962019-07-15 10:57:51 +02003006 /* don't do anything until the connection is established */
3007 if (!(conn->flags & CO_FL_CONNECTED))
3008 break;
3009
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003010 } /* end 'connect' */
3011 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3012 /* mark the step as started */
3013 check->last_started_step = check->current_step;
3014
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003015 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003016 if (*b_head(&check->bi) != '\0') {
3017 *b_head(&check->bi) = '\0';
3018 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003019 }
3020
Willy Tarreaubbae3f02017-08-30 09:59:52 +02003021 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003022 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003023 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003024 goto out_end_tcpcheck;
3025 }
3026
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003027 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003028 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003029 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003030 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003031 set_server_check_status(check, HCHK_STATUS_L7RSP,
3032 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003033 goto out_end_tcpcheck;
3034 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003035
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003036 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003037 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003038 continue;
3039
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003040 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3041 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003042
Willy Tarreauabca5b62013-12-06 14:19:25 +01003043 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003044 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003045
3046 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003047 while (&check->current_step->list != head &&
3048 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003049 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003050
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003051 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003052 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003053 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003054 goto out_end_tcpcheck;
3055
Olivier Houchard48bf0572019-07-18 15:09:08 +02003056 /* If we already subscribed, then we tried to received
3057 * and failed, so there's no point trying again.
3058 */
3059 if (check->wait_list.events & SUB_RETRY_RECV)
3060 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003061 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003062 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003063 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003064 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003065 /* Report network errors only if we got no other data. Otherwise
3066 * we'll let the upper layers decide whether the response is OK
3067 * or not. It is very common that an RST sent by the server is
3068 * reported as an error just after the last data chunk.
3069 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003070 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003071 goto out_end_tcpcheck;
3072 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003073 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003074 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003075 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003076 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003077 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003078 }
3079
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003080 /* mark the step as started */
3081 check->last_started_step = check->current_step;
3082
3083
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003084 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003085 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003086 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003087 if (b_data(&check->bi) < b_size(&check->bi)) {
3088 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003089 }
3090 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003091 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003092 done = 1; /* buffer full, don't wait for more data */
3093 }
3094
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003095 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003096
3097 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003098 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003099 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003100 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003101
3102 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003103 step = tcpcheck_get_step_id(check);
3104 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003105 comment = tcpcheck_get_step_comment(check, step);
3106 if (comment)
3107 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003108 set_server_check_status(check, HCHK_STATUS_L7RSP,
3109 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003110
3111 goto out_end_tcpcheck;
3112 }
3113
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003114 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003115 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003116
Willy Tarreaua970c282013-12-06 12:47:19 +01003117 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003118 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003119 ret = my_memmem(contentptr, b_data(&check->bi), check->current_step->string, check->current_step->string_len) != NULL;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003120 else if (check->current_step->expect_regex != NULL)
3121 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003122
3123 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003124 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003125
3126 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003127 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003128 if (ret) {
3129 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003130 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003131 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003132 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003133 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003134 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003135 }
3136 else {
3137 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003138 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003139 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003140 comment = tcpcheck_get_step_comment(check, step);
3141 if (comment)
3142 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003143 set_server_check_status(check, HCHK_STATUS_L7RSP,
3144 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003145 goto out_end_tcpcheck;
3146 }
3147 /* matched and was supposed to => OK, next step */
3148 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003149 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003150 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003151
3152 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003153 while (&check->current_step->list != head &&
3154 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003155 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003156
Willy Tarreauf2c87352015-05-13 12:08:21 +02003157 if (&check->current_step->list == head)
3158 break;
3159
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003160 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003161 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003162 }
3163 }
3164 else {
3165 /* not matched */
3166 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003167 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003168 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003169 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003170
3171 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003172 while (&check->current_step->list != head &&
3173 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003174 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003175
Willy Tarreauf2c87352015-05-13 12:08:21 +02003176 if (&check->current_step->list == head)
3177 break;
3178
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003179 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003180 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003181 }
3182 /* not matched but was supposed to => ERROR */
3183 else {
3184 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003185 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003186 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003187 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003188 }
3189 else {
3190 /* we were looking for a regex */
3191 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003192 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003193 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003194 comment = tcpcheck_get_step_comment(check, step);
3195 if (comment)
3196 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003197 set_server_check_status(check, HCHK_STATUS_L7RSP,
3198 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003199 goto out_end_tcpcheck;
3200 }
3201 }
3202 } /* end expect */
3203 } /* end loop over double chained step list */
3204
Baptiste Assmann248f1172018-03-01 21:49:01 +01003205 /* don't do anything until the connection is established */
3206 if (!(conn->flags & CO_FL_CONNECTED)) {
3207 /* update expire time, should be done by process_chk */
3208 /* we allow up to min(inter, timeout.connect) for a connection
3209 * to establish but only when timeout.check is set
3210 * as it may be to short for a full check otherwise
3211 */
3212 while (tick_is_expired(t->expire, now_ms)) {
3213 int t_con;
3214
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003215 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003216 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3217
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003218 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003219 t->expire = tick_first(t->expire, t_con);
3220 }
3221 goto out;
3222 }
3223
Willy Tarreau263013d2015-05-13 11:59:14 +02003224 /* We're waiting for some I/O to complete, we've reached the end of the
3225 * rules, or both. Do what we have to do, otherwise we're done.
3226 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003227 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003228 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3229 goto out_end_tcpcheck;
3230 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003231
Willy Tarreau53c5a042015-05-13 11:38:17 +02003232 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003233 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003234 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003235 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003236
3237 out_end_tcpcheck:
3238 /* collect possible new errors */
Willy Tarreau3d84b1c2019-07-23 14:37:47 +02003239 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003240 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003241
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003242 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003243 check->current_step = NULL;
3244
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003245 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003246 conn->flags |= CO_FL_ERROR;
3247
Christopher Fauletb6102852017-11-28 10:06:29 +01003248 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003249 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003250}
3251
Simon Hormanb1900d52015-01-30 11:22:54 +09003252const char *init_check(struct check *check, int type)
3253{
3254 check->type = type;
3255
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003256 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3257 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003258
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003259 check->bi.area = calloc(check->bi.size, sizeof(char));
3260 check->bo.area = calloc(check->bo.size, sizeof(char));
3261
3262 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003263 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003264
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003265 check->wait_list.tasklet = tasklet_new();
3266 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003267 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003268 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003269 check->wait_list.tasklet->process = event_srv_chk_io;
3270 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003271 return NULL;
3272}
3273
Simon Hormanbfb5d332015-01-30 11:22:55 +09003274void free_check(struct check *check)
3275{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003276 free(check->bi.area);
3277 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003278 if (check->cs) {
3279 free(check->cs->conn);
3280 check->cs->conn = NULL;
3281 cs_free(check->cs);
3282 check->cs = NULL;
3283 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003284}
3285
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003286void email_alert_free(struct email_alert *alert)
3287{
3288 struct tcpcheck_rule *rule, *back;
3289
3290 if (!alert)
3291 return;
3292
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003293 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3294 LIST_DEL(&rule->list);
3295 free(rule->comment);
3296 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003297 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003298 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003299 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003300 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003301}
3302
Olivier Houchard9f6af332018-05-25 14:04:04 +02003303static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003304{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003305 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003307 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003308
3309 q = container_of(check, typeof(*q), check);
3310
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003311 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003312 while (1) {
3313 if (!(check->state & CHK_ST_ENABLED)) {
3314 if (LIST_ISEMPTY(&q->email_alerts)) {
3315 /* All alerts processed, queue the task */
3316 t->expire = TICK_ETERNITY;
3317 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003318 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003319 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003320
3321 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003322 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003323 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003324 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003325 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003326 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327 }
3328
Olivier Houchard9f6af332018-05-25 14:04:04 +02003329 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003330 if (check->state & CHK_ST_INPROGRESS)
3331 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003332
3333 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3334 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003335 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003336 check->server = NULL;
3337 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003338 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003339 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003340 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003341 return t;
3342}
3343
Christopher Faulet0108bb32017-10-20 21:34:32 +02003344/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3345 *
3346 * The function returns 1 in success case, otherwise, it returns 0 and err is
3347 * filled.
3348 */
3349int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003350{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003351 struct mailer *mailer;
3352 struct email_alertq *queues;
3353 const char *err_str;
3354 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003355
Christopher Faulet0108bb32017-10-20 21:34:32 +02003356 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3357 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003358 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003359 }
3360
Christopher Faulet0108bb32017-10-20 21:34:32 +02003361 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3362 struct email_alertq *q = &queues[i];
3363 struct check *check = &q->check;
3364 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003365
3366 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003367 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003368 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003369 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003370 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003371 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003372 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3373 memprintf(err, "%s", err_str);
3374 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003375 }
3376
3377 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003378 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003379 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380
Emeric Brunc60def82017-09-27 14:59:38 +02003381 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003382 memprintf(err, "out of memory while allocating mailer alerts task");
3383 goto error;
3384 }
3385
3386 check->task = t;
3387 t->process = process_email_alert;
3388 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003389
Christopher Faulet0108bb32017-10-20 21:34:32 +02003390 /* check this in one ms */
3391 t->expire = TICK_ETERNITY;
3392 check->start = now;
3393 task_queue(t);
3394 }
3395
3396 mls->users++;
3397 free(p->email_alert.mailers.name);
3398 p->email_alert.mailers.m = mls;
3399 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003400 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003401
3402 error:
3403 for (i = 0; i < mls->count; i++) {
3404 struct email_alertq *q = &queues[i];
3405 struct check *check = &q->check;
3406
Willy Tarreauf6562792019-05-07 19:05:35 +02003407 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003408 free_check(check);
3409 }
3410 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003411 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003412 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003413}
3414
3415
3416static int add_tcpcheck_expect_str(struct list *list, const char *str)
3417{
3418 struct tcpcheck_rule *tcpcheck;
3419
Willy Tarreaubafbe012017-11-24 17:34:44 +01003420 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003421 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003422 memset(tcpcheck, 0, sizeof(*tcpcheck));
3423 tcpcheck->action = TCPCHK_ACT_EXPECT;
3424 tcpcheck->string = strdup(str);
3425 tcpcheck->expect_regex = NULL;
3426 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003427 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003428 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003429 return 0;
3430 }
3431
3432 LIST_ADDQ(list, &tcpcheck->list);
3433 return 1;
3434}
3435
3436static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3437{
3438 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003439 const char *in;
3440 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003441 int i;
3442
Willy Tarreaubafbe012017-11-24 17:34:44 +01003443 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003444 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003445 memset(tcpcheck, 0, sizeof(*tcpcheck));
3446 tcpcheck->action = TCPCHK_ACT_SEND;
3447 tcpcheck->expect_regex = NULL;
3448 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003449 tcpcheck->string_len = 0;
3450 for (i = 0; strs[i]; i++)
3451 tcpcheck->string_len += strlen(strs[i]);
3452
3453 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3454 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003455 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003456 return 0;
3457 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003458
Willy Tarreau64345aa2016-08-10 19:29:09 +02003459 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003460 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003461 for (in = strs[i]; (*dst = *in++); dst++);
3462 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003463
3464 LIST_ADDQ(list, &tcpcheck->list);
3465 return 1;
3466}
3467
Christopher Faulet0108bb32017-10-20 21:34:32 +02003468static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3469 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003470{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003471 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003472 struct tcpcheck_rule *tcpcheck;
3473 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003474
Willy Tarreaubafbe012017-11-24 17:34:44 +01003475 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003476 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003477 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003479 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003480
Willy Tarreaubafbe012017-11-24 17:34:44 +01003481 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003482 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003483 memset(tcpcheck, 0, sizeof(*tcpcheck));
3484 tcpcheck->action = TCPCHK_ACT_CONNECT;
3485 tcpcheck->comment = NULL;
3486 tcpcheck->string = NULL;
3487 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003488 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3489
3490 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3491 goto error;
3492
3493 {
3494 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3495 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3496 goto error;
3497 }
3498
3499 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3500 goto error;
3501
3502 {
3503 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3504 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3505 goto error;
3506 }
3507
3508 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3509 goto error;
3510
3511 {
3512 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3513 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3514 goto error;
3515 }
3516
3517 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3518 goto error;
3519
3520 {
3521 const char * const strs[2] = { "DATA\r\n" };
3522 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3523 goto error;
3524 }
3525
3526 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3527 goto error;
3528
3529 {
3530 struct tm tm;
3531 char datestr[48];
3532 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003533 "From: ", p->email_alert.from, "\r\n",
3534 "To: ", p->email_alert.to, "\r\n",
3535 "Date: ", datestr, "\r\n",
3536 "Subject: [HAproxy Alert] ", msg, "\r\n",
3537 "\r\n",
3538 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003539 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003540 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003541 NULL
3542 };
3543
3544 get_localtime(date.tv_sec, &tm);
3545
3546 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3547 goto error;
3548 }
3549
3550 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3551 goto error;
3552 }
3553
3554 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3555 goto error;
3556
3557 {
3558 const char * const strs[2] = { "QUIT\r\n" };
3559 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3560 goto error;
3561 }
3562
3563 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3564 goto error;
3565
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003566 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003567 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003568 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003569 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003570 return 1;
3571
3572error:
3573 email_alert_free(alert);
3574 return 0;
3575}
3576
Christopher Faulet0108bb32017-10-20 21:34:32 +02003577static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003578{
3579 int i;
3580 struct mailer *mailer;
3581
3582 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3583 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003584 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003585 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003586 return;
3587 }
3588 }
3589
3590 return;
3591}
3592
3593/*
3594 * Send email alert if configured.
3595 */
Simon Horman64e34162015-02-06 11:11:57 +09003596void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003597{
3598 va_list argp;
3599 char buf[1024];
3600 int len;
3601 struct proxy *p = s->proxy;
3602
Christopher Faulet0108bb32017-10-20 21:34:32 +02003603 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003604 return;
3605
3606 va_start(argp, format);
3607 len = vsnprintf(buf, sizeof(buf), format, argp);
3608 va_end(argp);
3609
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003610 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003611 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003612 return;
3613 }
3614
Christopher Faulet0108bb32017-10-20 21:34:32 +02003615 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003616}
3617
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003618/*
3619 * Return value:
3620 * the port to be used for the health check
3621 * 0 in case no port could be found for the check
3622 */
3623int srv_check_healthcheck_port(struct check *chk)
3624{
3625 int i = 0;
3626 struct server *srv = NULL;
3627
3628 srv = chk->server;
3629
3630 /* If neither a port nor an addr was specified and no check transport
3631 * layer is forced, then the transport layer used by the checks is the
3632 * same as for the production traffic. Otherwise we use raw_sock by
3633 * default, unless one is specified.
3634 */
3635 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003636 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003637 chk->send_proxy |= (srv->pp_opts);
3638 }
3639
3640 /* by default, we use the health check port ocnfigured */
3641 if (chk->port > 0)
3642 return chk->port;
3643
3644 /* try to get the port from check_core.addr if check.port not set */
3645 i = get_host_port(&chk->addr);
3646 if (i > 0)
3647 return i;
3648
3649 /* try to get the port from server address */
3650 /* prevent MAPPORTS from working at this point, since checks could
3651 * not be performed in such case (MAPPORTS impose a relative ports
3652 * based on live traffic)
3653 */
3654 if (srv->flags & SRV_F_MAPPORTS)
3655 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003656
3657 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003658 if (i > 0)
3659 return i;
3660
3661 return 0;
3662}
3663
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003664REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003665
Willy Tarreaubd741542010-03-16 18:46:54 +01003666/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003667 * Local variables:
3668 * c-indent-level: 8
3669 * c-basic-offset: 8
3670 * End:
3671 */