blob: 96924dd126192f0c85e763d9eb7efc7d3398faf9 [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 Tarreau489bbd32019-09-24 10:43:03 +0200878 /* the rest of the code below expects the connection to be ready! */
879 if (!(conn->flags & CO_FL_CONNECTED) && !done)
880 goto wait_more_data;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100881
Willy Tarreau03938182010-03-17 21:52:07 +0100882 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200883 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100884 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200885 if (b_data(&check->bi) < b_size(&check->bi))
886 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100887 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200888 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100889 done = 1; /* buffer full, don't wait for more data */
890 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200891
Nick Chalk57b1bf72010-03-16 15:50:46 +0000892 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900893 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200894 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200895 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100896 goto wait_more_data;
897
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100898 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200899 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
900 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
901 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
902 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
903 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
904 cut_crlf(b_head(&check->bi));
905 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200906
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100907 goto out_wakeup;
908 }
909
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200910 check->code = str2uic(b_head(&check->bi) + 9);
911 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200912
Willy Tarreaubd741542010-03-16 18:46:54 +0100913 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200914 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000915 /* 404 may be accepted as "stopping" only if the server was up */
916 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900917 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000918 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100919 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
920 /* Run content verification check... We know we have at least 13 chars */
921 if (!httpchk_expect(s, done))
922 goto wait_more_data;
923 }
924 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200925 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100926 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100928 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000929 else {
930 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900931 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000932 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200933 break;
934
935 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200936 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100937 goto wait_more_data;
938
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100939 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200940 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900941 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200942 else
Simon Horman4a741432013-02-23 15:35:38 +0900943 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200944 break;
945
946 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200947 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100948 goto wait_more_data;
949
Willy Tarreau55706782019-08-06 16:26:31 +0200950 /* do not reset when closing, servers don't like this */
951 if (conn_ctrl_ready(cs->conn))
952 fdtab[cs->conn->handle.fd].linger_risk = 0;
953
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200954 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200955 if ((b_data(&check->bi) < strlen("000\r")) ||
956 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
957 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
958 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
959 cut_crlf(b_head(&check->bi));
960 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200961 goto out_wakeup;
962 }
963
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200964 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200965
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200966 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200967 cut_crlf(desc);
968
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100969 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200970 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900971 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200972 else
Simon Horman4a741432013-02-23 15:35:38 +0900973 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200974 break;
975
Simon Hormana2b9dad2013-02-12 10:45:54 +0900976 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100977 int status = HCHK_STATUS_CHECKED;
978 const char *hs = NULL; /* health status */
979 const char *as = NULL; /* admin status */
980 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200981 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100982 const char *err = NULL; /* first error to report */
983 const char *wrn = NULL; /* first warning to report */
984 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900985
Willy Tarreau81f5d942013-12-09 20:51:51 +0100986 /* We're getting an agent check response. The agent could
987 * have been disabled in the mean time with a long check
988 * still pending. It is important that we ignore the whole
989 * response.
990 */
991 if (!(check->server->agent.state & CHK_ST_ENABLED))
992 break;
993
994 /* The agent supports strings made of a single line ended by the
995 * first CR ('\r') or LF ('\n'). This line is composed of words
996 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
997 * line may optionally contained a description of a state change
998 * after a sharp ('#'), which is only considered if a health state
999 * is announced.
1000 *
1001 * Words may be composed of :
1002 * - a numeric weight suffixed by the percent character ('%').
1003 * - a health status among "up", "down", "stopped", and "fail".
1004 * - an admin status among "ready", "drain", "maint".
1005 *
1006 * These words may appear in any order. If multiple words of the
1007 * same category appear, the last one wins.
1008 */
1009
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001010 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001011 while (*p && *p != '\n' && *p != '\r')
1012 p++;
1013
1014 if (!*p) {
1015 if (!done)
1016 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001017
Willy Tarreau9809b782013-12-11 21:40:11 +01001018 /* at least inform the admin that the agent is mis-behaving */
1019 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1020 break;
1021 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001022
Willy Tarreau9809b782013-12-11 21:40:11 +01001023 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001024 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001025
Willy Tarreau81f5d942013-12-09 20:51:51 +01001026 while (*cmd) {
1027 /* look for next word */
1028 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1029 cmd++;
1030 continue;
1031 }
Simon Horman671b6f02013-11-25 10:46:39 +09001032
Willy Tarreau81f5d942013-12-09 20:51:51 +01001033 if (*cmd == '#') {
1034 /* this is the beginning of a health status description,
1035 * skip the sharp and blanks.
1036 */
1037 cmd++;
1038 while (*cmd == '\t' || *cmd == ' ')
1039 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001040 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001041 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001042
1043 /* find the end of the word so that we have a null-terminated
1044 * word between <cmd> and <p>.
1045 */
1046 p = cmd + 1;
1047 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1048 p++;
1049 if (*p)
1050 *p++ = 0;
1051
1052 /* first, health statuses */
1053 if (strcasecmp(cmd, "up") == 0) {
1054 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001055 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001056 hs = cmd;
1057 }
1058 else if (strcasecmp(cmd, "down") == 0) {
1059 check->health = 0;
1060 status = HCHK_STATUS_L7STS;
1061 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001062 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001063 else if (strcasecmp(cmd, "stopped") == 0) {
1064 check->health = 0;
1065 status = HCHK_STATUS_L7STS;
1066 hs = cmd;
1067 }
1068 else if (strcasecmp(cmd, "fail") == 0) {
1069 check->health = 0;
1070 status = HCHK_STATUS_L7STS;
1071 hs = cmd;
1072 }
1073 /* admin statuses */
1074 else if (strcasecmp(cmd, "ready") == 0) {
1075 as = cmd;
1076 }
1077 else if (strcasecmp(cmd, "drain") == 0) {
1078 as = cmd;
1079 }
1080 else if (strcasecmp(cmd, "maint") == 0) {
1081 as = cmd;
1082 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001083 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001084 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1085 ps = cmd;
1086 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001087 /* try to parse a maxconn here */
1088 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1089 cs = cmd;
1090 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001091 else {
1092 /* keep a copy of the first error */
1093 if (!err)
1094 err = cmd;
1095 }
1096 /* skip to next word */
1097 cmd = p;
1098 }
1099 /* here, cmd points either to \0 or to the beginning of a
1100 * description. Skip possible leading spaces.
1101 */
1102 while (*cmd == ' ' || *cmd == '\n')
1103 cmd++;
1104
1105 /* First, update the admin status so that we avoid sending other
1106 * possibly useless warnings and can also update the health if
1107 * present after going back up.
1108 */
1109 if (as) {
1110 if (strcasecmp(as, "drain") == 0)
1111 srv_adm_set_drain(check->server);
1112 else if (strcasecmp(as, "maint") == 0)
1113 srv_adm_set_maint(check->server);
1114 else
1115 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001116 }
1117
Willy Tarreau81f5d942013-12-09 20:51:51 +01001118 /* now change weights */
1119 if (ps) {
1120 const char *msg;
1121
1122 msg = server_parse_weight_change_request(s, ps);
1123 if (!wrn || !*wrn)
1124 wrn = msg;
1125 }
1126
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001127 if (cs) {
1128 const char *msg;
1129
1130 cs += strlen("maxconn:");
1131
1132 msg = server_parse_maxconn_change_request(s, cs);
1133 if (!wrn || !*wrn)
1134 wrn = msg;
1135 }
1136
Willy Tarreau81f5d942013-12-09 20:51:51 +01001137 /* and finally health status */
1138 if (hs) {
1139 /* We'll report some of the warnings and errors we have
1140 * here. Down reports are critical, we leave them untouched.
1141 * Lack of report, or report of 'UP' leaves the room for
1142 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001143 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001144 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001145 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001146
1147 if (!*msg || status == HCHK_STATUS_L7OKD) {
1148 if (err && *err)
1149 msg = err;
1150 else if (wrn && *wrn)
1151 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001152 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001153
1154 t = get_trash_chunk();
1155 chunk_printf(t, "via agent : %s%s%s%s",
1156 hs, *msg ? " (" : "",
1157 msg, *msg ? ")" : "");
1158
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001159 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001160 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001161 else if (err && *err) {
1162 /* No status change but we'd like to report something odd.
1163 * Just report the current state and copy the message.
1164 */
1165 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001166 set_server_check_status(check, status/*check->status*/,
1167 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001168
Willy Tarreau81f5d942013-12-09 20:51:51 +01001169 }
1170 else if (wrn && *wrn) {
1171 /* No status change but we'd like to report something odd.
1172 * Just report the current state and copy the message.
1173 */
1174 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001175 set_server_check_status(check, status/*check->status*/,
1176 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001177 }
1178 else
1179 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001180 break;
1181 }
1182
Willy Tarreau1620ec32011-08-06 17:05:02 +02001183 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001184 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001185 goto wait_more_data;
1186
Willy Tarreau55706782019-08-06 16:26:31 +02001187 /* do not reset when closing, servers don't like this */
1188 if (conn_ctrl_ready(cs->conn))
1189 fdtab[cs->conn->handle.fd].linger_risk = 0;
1190
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001191 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001192 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001193 }
1194 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001195 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1196 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001197 else
1198 desc = "PostgreSQL unknown error";
1199
Simon Horman4a741432013-02-23 15:35:38 +09001200 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001201 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001202 break;
1203
1204 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001206 goto wait_more_data;
1207
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001208 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001209 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001210 }
1211 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001212 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001213 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001214 break;
1215
1216 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001217 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001218 goto wait_more_data;
1219
Willy Tarreau55706782019-08-06 16:26:31 +02001220 /* do not reset when closing, servers don't like this */
1221 if (conn_ctrl_ready(cs->conn))
1222 fdtab[cs->conn->handle.fd].linger_risk = 0;
1223
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001224 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001225 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001226 /* We set the MySQL Version in description for information purpose
1227 * FIXME : it can be cool to use MySQL Version for other purpose,
1228 * like mark as down old MySQL server.
1229 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001230 if (b_data(&check->bi) > 51) {
1231 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001232 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001233 }
1234 else {
1235 if (!done)
1236 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001237
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001238 /* it seems we have a OK packet but without a valid length,
1239 * it must be a protocol error
1240 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001241 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001242 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001243 }
1244 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001246 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001247 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001248 }
1249 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001250 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1251 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1252 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001253
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001254 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001255 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001256 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001257 /* We have only one MySQL packet and it is a Handshake Initialization packet
1258 * but we need to have a second packet to know if it is alright
1259 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001260 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001261 goto wait_more_data;
1262 }
1263 else {
1264 /* We have only one packet and it is an Error packet,
1265 * an error message is attached, so we can display it
1266 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001267 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001268 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001269 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001270 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001271 } else if (b_data(&check->bi) > first_packet_len + 4) {
1272 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1273 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1274 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001275
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001276 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001277 /* We have 2 packets and that's good */
1278 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001279 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001280 /* No error packet */
1281 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001282 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001283 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001284 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001285 }
1286 else {
1287 /* An error message is attached in the Error packet
1288 * so we can display it ! :)
1289 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001290 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001291 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001292 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001293 }
1294 }
1295 }
1296 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001297 if (!done)
1298 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001299
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001300 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001301 * it must be a protocol error
1302 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001303 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001304 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001305 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001306 }
1307 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001308 break;
1309
1310 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001311 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001312 goto wait_more_data;
1313
1314 /* Check if the server speaks LDAP (ASN.1/BER)
1315 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1316 * http://tools.ietf.org/html/rfc4511
1317 */
1318
1319 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1320 * LDAPMessage: 0x30: SEQUENCE
1321 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001322 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001323 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001324 }
1325 else {
1326 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001327 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001328
1329 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1330 * messageID: 0x02 0x01 0x01: INTEGER 1
1331 * protocolOp: 0x61: bindResponse
1332 */
1333 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001334 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001335 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001336 goto out_wakeup;
1337 }
1338
1339 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001340 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001341
1342 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1343 * ldapResult: 0x0a 0x01: ENUMERATION
1344 */
1345 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001346 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001347 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001348 goto out_wakeup;
1349 }
1350
1351 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1352 * resultCode
1353 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001354 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001355 if (check->code) {
1356 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 +02001357 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001358 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001359 }
1360 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001361 break;
1362
Christopher Fauletba7bc162016-11-07 21:07:38 +01001363 case PR_O2_SPOP_CHK: {
1364 unsigned int framesz;
1365 char err[HCHK_DESC_LEN];
1366
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001367 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001368 goto wait_more_data;
1369
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001370 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001371 framesz = ntohl(framesz);
1372
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001373 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001374 goto wait_more_data;
1375
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001376 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001377 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1378 else
1379 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1380 break;
1381 }
1382
Willy Tarreau1620ec32011-08-06 17:05:02 +02001383 default:
Willy Tarreaufbeee552019-09-05 18:43:22 +02001384 /* good connection is enough for pure TCP check */
1385 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
1386 if (check->use_ssl)
1387 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
1388 else
1389 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
1390 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001391 break;
1392 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001393
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001394 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001395 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001396 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001397 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001398
Nick Chalk57b1bf72010-03-16 15:50:46 +00001399 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001400 *b_head(&check->bi) = '\0';
1401 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001402
Steven Davidovitz544d4812017-03-08 11:06:20 -08001403 /* Close the connection... We still attempt to nicely close if,
1404 * for instance, SSL needs to send a "close notify." Later, we perform
1405 * a hard close and reset the connection if some data are pending,
1406 * otherwise we end up with many TIME_WAITs and eat all the source port
1407 * range quickly. To avoid sending RSTs all the time, we first try to
1408 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001409 */
Olivier Houchard8aa445d2019-07-02 16:35:18 +02001410 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1411 * connection, to make sure cs_shutw() will not lead to a shutdown()
1412 * that would provoke TIME_WAITs.
1413 */
1414 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001415 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001416
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001417 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001418 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001419 conn->flags |= CO_FL_ERROR;
1420
Willy Tarreaufdccded2008-08-29 18:19:04 +02001421 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001422out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001423 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001424
1425 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001426 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001427 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001428}
1429
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001430/*
1431 * This function is used only for server health-checks. It handles connection
1432 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001433 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1434 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001435 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001436static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001437{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001438 struct connection *conn = cs->conn;
1439 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001440 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001441 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001442
Olivier Houchard0923fa42019-01-11 18:43:04 +01001443 if (check->server)
1444 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1445 else
1446 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001447
Willy Tarreauc09572f2017-10-04 11:58:22 +02001448 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001449 if (check->type == PR_O2_TCPCHK_CHK) {
1450 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001451 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001452 conn = cs->conn;
Willy Tarreauab160a42019-09-05 17:38:40 +02001453 } else {
1454 if (!(check->wait_list.events & SUB_RETRY_SEND))
1455 __event_srv_chk_w(cs);
1456 if (!(check->wait_list.events & SUB_RETRY_RECV))
1457 __event_srv_chk_r(cs);
1458 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001459
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001460 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001461 /* We may get error reports bypassing the I/O handlers, typically
1462 * the case when sending a pure TCP check which fails, then the I/O
1463 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001464 * main processing task so let's simply wake it up. If we get here,
1465 * we expect errno to still be valid.
1466 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001467 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001468 task_wakeup(check->task, TASK_WOKEN_IO);
1469 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001470 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001471 /* we may get here if only a connection probe was required : we
1472 * don't have any data to send nor anything expected in response,
1473 * so the completion of the connection establishment is enough.
1474 */
1475 task_wakeup(check->task, TASK_WOKEN_IO);
1476 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001477
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001478 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001479 /* Check complete or aborted. If connection not yet closed do it
1480 * now and wake the check task up to be sure the result is
1481 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001482 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001483 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001484 ret = -1;
Olivier Houchard865ed9b2019-07-02 17:42:22 +02001485 /* We may have been scheduled to run, and the
1486 * I/O handler expects to have a cs, so remove
1487 * the tasklet
1488 */
1489 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001490 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001491 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001492
Olivier Houchard0923fa42019-01-11 18:43:04 +01001493 if (check->server)
1494 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1495 else
1496 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001497
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001498 /* if a connection got replaced, we must absolutely prevent the connection
1499 * handler from touching its fd, and perform the FD polling updates ourselves
1500 */
1501 if (ret < 0)
1502 conn_cond_update_polling(conn);
1503
1504 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001505}
1506
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001507struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001508 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001509 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001510};
1511
Willy Tarreaubaaee002006-06-26 02:48:02 +02001512/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001513 * updates the server's weight during a warmup stage. Once the final weight is
1514 * reached, the task automatically stops. Note that any server status change
1515 * must have updated s->last_change accordingly.
1516 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001517static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001518{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001519 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001520
1521 /* by default, plan on stopping the task */
1522 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001523 if ((s->next_admin & SRV_ADMF_MAINT) ||
1524 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001525 return t;
1526
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001527 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1528
Willy Tarreau892337c2014-05-13 23:41:20 +02001529 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001530 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001531
1532 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001533 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001534
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001535 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1536
Willy Tarreau2e993902011-10-31 11:53:20 +01001537 /* get back there in 1 second or 1/20th of the slowstart interval,
1538 * whichever is greater, resulting in small 5% steps.
1539 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001540 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001541 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1542 return t;
1543}
1544
Willy Tarreau894c6422017-10-04 15:58:52 +02001545/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1546 * none was found.
1547 */
1548static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1549{
1550 struct tcpcheck_rule *r;
1551
1552 list_for_each_entry(r, list, list) {
1553 if (r->action != TCPCHK_ACT_COMMENT)
1554 return r;
1555 }
1556 return NULL;
1557}
1558
Willy Tarreau2e993902011-10-31 11:53:20 +01001559/*
Simon Horman98637e52014-06-20 12:30:16 +09001560 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001561 *
1562 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001563 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1564 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1565 * - SF_ERR_SRVTO if there are no more servers
1566 * - SF_ERR_SRVCL if the connection was refused by the server
1567 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1568 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1569 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001570 * - 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 +01001571 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001572 * Note that we try to prevent the network stack from sending the ACK during the
1573 * connect() when a pure TCP check is used (without PROXY protocol).
1574 */
Simon Horman98637e52014-06-20 12:30:16 +09001575static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001576{
1577 struct check *check = t->context;
1578 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001579 struct conn_stream *cs = check->cs;
1580 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001581 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001582 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001583 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001584 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001585
Willy Tarreau00149122017-10-04 18:05:01 +02001586 /* we cannot have a connection here */
1587 if (conn)
1588 return SF_ERR_INTERNAL;
1589
Simon Hormanb00d17a2014-06-13 16:18:16 +09001590 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001591 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001592 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001593 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1594 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001595
1596 /* prepare the check buffer.
1597 * This should not be used if check is the secondary agent check
1598 * of a server as s->proxy->check_req will relate to the
1599 * configuration of the primary check. Similarly, tcp-check uses
1600 * its own strings.
1601 */
1602 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001603 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001604
1605 /* we want to check if this host replies to HTTP or SSLv3 requests
1606 * so we'll send the request, and won't wake the checker up now.
1607 */
1608 if ((check->type) == PR_O2_SSL3_CHK) {
1609 /* SSL requires that we put Unix time in the request */
1610 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001611 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001612 }
1613 else if ((check->type) == PR_O2_HTTP_CHK) {
1614 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001615 b_putblk(&check->bo, trash.area,
1616 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001617 /* prevent HTTP keep-alive when "http-check expect" is used */
1618 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001619 b_putist(&check->bo, ist("Connection: close\r\n"));
1620 b_putist(&check->bo, ist("\r\n"));
1621 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001622 }
1623 }
1624
James Brown55f9ff12015-10-21 18:19:05 -07001625 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001626 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001627 }
1628
Willy Tarreauf411cce2017-10-04 16:21:19 +02001629 /* for tcp-checks, the initial connection setup is handled separately as
1630 * it may be sent to a specific port and not to the server's.
1631 */
1632 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1633 tcpcheck_main(check);
1634 return SF_ERR_UP;
1635 }
1636
Simon Hormanb00d17a2014-06-13 16:18:16 +09001637 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001638 cs = check->cs = cs_new(NULL);
1639 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001640 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001641 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001642 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001643 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001644
Simon Horman41f58762015-01-30 11:22:56 +09001645 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001646 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001647 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001648 }
1649 else {
1650 /* we'll connect to the addr on the server */
1651 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001652 }
1653
Alexander Liu2a54bb72019-05-22 19:44:48 +08001654 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1655 conn->send_proxy_ofs = 1;
1656 conn->flags |= CO_FL_SOCKS4;
1657 }
1658
Olivier Houchard6377a002017-12-01 22:04:05 +01001659 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001660 conn->target = &s->obj_type;
1661
Willy Tarreaube373152018-09-06 11:45:30 +02001662 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001663 int i = 0;
1664
1665 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001666 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001667 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001668
1669 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001670 }
1671
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001672 /* no client address */
1673 clear_addr(&conn->addr.from);
1674
Willy Tarreaube373152018-09-06 11:45:30 +02001675 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001676 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1677 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001678 cs_attach(cs, check, &check_conn_cb);
1679
Willy Tarreauf3d34822014-12-08 12:11:28 +01001680 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001681 if (check->type != 0)
1682 connflags |= CONNECT_HAS_DATA;
1683 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1684 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1685 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001686
Willy Tarreaue7dff022015-04-03 01:14:29 +02001687 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001688 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001689 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001690
Willy Tarreau16257f62017-11-02 15:45:00 +01001691
Olivier Houchard9130a962017-10-17 17:33:43 +02001692#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001693 if (ret == SF_ERR_NONE) {
1694 if (s->check.sni)
1695 ssl_sock_set_servername(conn, s->check.sni);
1696 if (s->check.alpn_str)
1697 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1698 s->check.alpn_len);
1699 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001700#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001701 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001702 conn->send_proxy_ofs = 1;
1703 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001704 if (xprt_add_hs(conn) < 0)
1705 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001706 }
1707
1708 return ret;
1709}
1710
Simon Horman98637e52014-06-20 12:30:16 +09001711static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001712static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001713__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001714
1715void block_sigchld(void)
1716{
1717 sigset_t set;
1718 sigemptyset(&set);
1719 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001720 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001721}
1722
1723void unblock_sigchld(void)
1724{
1725 sigset_t set;
1726 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001727 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001728 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001729}
1730
Simon Horman98637e52014-06-20 12:30:16 +09001731static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1732{
1733 struct pid_list *elem;
1734 struct check *check = t->context;
1735
Willy Tarreaubafbe012017-11-24 17:34:44 +01001736 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001737 if (!elem)
1738 return NULL;
1739 elem->pid = pid;
1740 elem->t = t;
1741 elem->exited = 0;
1742 check->curpid = elem;
1743 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001744
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001745 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001746 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001747 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001748
Simon Horman98637e52014-06-20 12:30:16 +09001749 return elem;
1750}
1751
Simon Horman98637e52014-06-20 12:30:16 +09001752static void pid_list_del(struct pid_list *elem)
1753{
1754 struct check *check;
1755
1756 if (!elem)
1757 return;
1758
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001759 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001760 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001761 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001762
Simon Horman98637e52014-06-20 12:30:16 +09001763 if (!elem->exited)
1764 kill(elem->pid, SIGTERM);
1765
1766 check = elem->t->context;
1767 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001768 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001769}
1770
1771/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1772static void pid_list_expire(pid_t pid, int status)
1773{
1774 struct pid_list *elem;
1775
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001776 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001777 list_for_each_entry(elem, &pid_list, list) {
1778 if (elem->pid == pid) {
1779 elem->t->expire = now_ms;
1780 elem->status = status;
1781 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001782 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001783 break;
Simon Horman98637e52014-06-20 12:30:16 +09001784 }
1785 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001786 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001787}
1788
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001789static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001790{
1791 pid_t pid;
1792 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001793
Simon Horman98637e52014-06-20 12:30:16 +09001794 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1795 pid_list_expire(pid, status);
1796}
1797
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001798static int init_pid_list(void)
1799{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001800 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001801 /* Nothing to do */
1802 return 0;
1803
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001804 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001805 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1806 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001807 return 1;
1808 }
1809
Willy Tarreaubafbe012017-11-24 17:34:44 +01001810 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1811 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001812 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1813 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001814 return 1;
1815 }
1816
1817 return 0;
1818}
1819
Cyril Bontéac92a062014-12-27 22:28:38 +01001820/* helper macro to set an environment variable and jump to a specific label on failure. */
1821#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001822
1823/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001824 * helper function to allocate enough memory to store an environment variable.
1825 * It will also check that the environment variable is updatable, and silently
1826 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001827 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001828static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001829{
1830 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001831 char *envname;
1832 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001833
Cyril Bontéac92a062014-12-27 22:28:38 +01001834 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001835 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001836 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001837 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001838
1839 envname = extcheck_envs[idx].name;
1840 vmaxlen = extcheck_envs[idx].vmaxlen;
1841
1842 /* Check if the environment variable is already set, and silently reject
1843 * the update if this one is not updatable. */
1844 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1845 return 0;
1846
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001847 /* Instead of sending NOT_USED, sending an empty value is preferable */
1848 if (strcmp(value, "NOT_USED") == 0) {
1849 value = "";
1850 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001851
1852 len = strlen(envname) + 1;
1853 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1854 len += strlen(value);
1855 else
1856 len += vmaxlen;
1857
1858 if (!check->envp[idx])
1859 check->envp[idx] = malloc(len + 1);
1860
1861 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001862 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001863 return 1;
1864 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001865 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001866 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001867 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001868 return 1;
1869 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001870 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001871 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001872 return 1;
1873 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001874 return 0;
1875}
Simon Horman98637e52014-06-20 12:30:16 +09001876
1877static int prepare_external_check(struct check *check)
1878{
1879 struct server *s = check->server;
1880 struct proxy *px = s->proxy;
1881 struct listener *listener = NULL, *l;
1882 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001883 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001884 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001885
1886 list_for_each_entry(l, &px->conf.listeners, by_fe)
1887 /* Use the first INET, INET6 or UNIX listener */
1888 if (l->addr.ss_family == AF_INET ||
1889 l->addr.ss_family == AF_INET6 ||
1890 l->addr.ss_family == AF_UNIX) {
1891 listener = l;
1892 break;
1893 }
1894
Simon Horman98637e52014-06-20 12:30:16 +09001895 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001896 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1897 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001898 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001899 goto err;
1900 }
Simon Horman98637e52014-06-20 12:30:16 +09001901
Cyril Bontéac92a062014-12-27 22:28:38 +01001902 check->argv = calloc(6, sizeof(char *));
1903 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001904 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001905 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001906 }
Simon Horman98637e52014-06-20 12:30:16 +09001907
1908 check->argv[0] = px->check_command;
1909
Cyril Bonté777be862014-12-02 21:21:35 +01001910 if (!listener) {
1911 check->argv[1] = strdup("NOT_USED");
1912 check->argv[2] = strdup("NOT_USED");
1913 }
1914 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001915 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001916 addr_to_str(&listener->addr, buf, sizeof(buf));
1917 check->argv[1] = strdup(buf);
1918 port_to_str(&listener->addr, buf, sizeof(buf));
1919 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001920 }
1921 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001922 const struct sockaddr_un *un;
1923
1924 un = (struct sockaddr_un *)&listener->addr;
1925 check->argv[1] = strdup(un->sun_path);
1926 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001927 }
1928 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001929 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001930 goto err;
1931 }
1932
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001933 addr_to_str(&s->addr, buf, sizeof(buf));
1934 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001935
1936 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1937 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1938 else
1939 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001940 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001941
Cyril Bontéac92a062014-12-27 22:28:38 +01001942 for (i = 0; i < 5; i++) {
1943 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001944 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001945 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001946 }
1947 }
Simon Horman98637e52014-06-20 12:30:16 +09001948
Cyril Bontéac92a062014-12-27 22:28:38 +01001949 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001950 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001951 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1952 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1953 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1954 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001955 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001956 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1957 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1958 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1959 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1960 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1961 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1962
1963 /* Ensure that we don't leave any hole in check->envp */
1964 for (i = 0; i < EXTCHK_SIZE; i++)
1965 if (!check->envp[i])
1966 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001967
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001968 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001969err:
1970 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001971 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001972 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001973 free(check->envp);
1974 check->envp = NULL;
1975 }
1976
1977 if (check->argv) {
1978 for (i = 1; i < 5; i++)
1979 free(check->argv[i]);
1980 free(check->argv);
1981 check->argv = NULL;
1982 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001983 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001984}
1985
Simon Hormanb00d17a2014-06-13 16:18:16 +09001986/*
Simon Horman98637e52014-06-20 12:30:16 +09001987 * establish a server health-check that makes use of a process.
1988 *
1989 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001990 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001991 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001992 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001993 *
1994 * Blocks and then unblocks SIGCHLD
1995 */
1996static int connect_proc_chk(struct task *t)
1997{
Cyril Bontéac92a062014-12-27 22:28:38 +01001998 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001999 struct check *check = t->context;
2000 struct server *s = check->server;
2001 struct proxy *px = s->proxy;
2002 int status;
2003 pid_t pid;
2004
Willy Tarreaue7dff022015-04-03 01:14:29 +02002005 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002006
2007 block_sigchld();
2008
2009 pid = fork();
2010 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002011 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
2012 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002013 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2014 goto out;
2015 }
2016 if (pid == 0) {
2017 /* Child */
2018 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002019 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002020 int fd;
2021
2022 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2023 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2024
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002025 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002026
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002027 /* restore the initial FD limits */
2028 limit.rlim_cur = rlim_fd_cur_at_boot;
2029 limit.rlim_max = rlim_fd_max_at_boot;
2030 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2031 getrlimit(RLIMIT_NOFILE, &limit);
2032 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2033 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2034 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2035 }
2036
Simon Horman98637e52014-06-20 12:30:16 +09002037 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01002038 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Willy Tarreauafc313e2019-07-01 07:51:29 +02002039 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002040 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002041 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2042 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002043 exit(-1);
2044 }
2045
2046 /* Parent */
2047 if (check->result == CHK_RES_UNKNOWN) {
2048 if (pid_list_add(pid, t) != NULL) {
2049 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2050
2051 if (px->timeout.check && px->timeout.connect) {
2052 int t_con = tick_add(now_ms, px->timeout.connect);
2053 t->expire = tick_first(t->expire, t_con);
2054 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002055 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002056 goto out;
2057 }
2058 else {
2059 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2060 }
2061 kill(pid, SIGTERM); /* process creation error */
2062 }
2063 else
2064 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2065
2066out:
2067 unblock_sigchld();
2068 return status;
2069}
2070
2071/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002072 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002073 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002074 *
2075 * Please do NOT place any return statement in this function and only leave
2076 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002077 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002078static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002079{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002080 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002081 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002082 int rv;
2083 int ret;
2084 int expired = tick_is_expired(t->expire, now_ms);
2085
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002086 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002087 if (!(check->state & CHK_ST_INPROGRESS)) {
2088 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002089 if (!expired) /* woke up too early */
2090 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002091
2092 /* we don't send any health-checks when the proxy is
2093 * stopped, the server should not be checked or the check
2094 * is disabled.
2095 */
2096 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2097 s->proxy->state == PR_STSTOPPED)
2098 goto reschedule;
2099
2100 /* we'll initiate a new check */
2101 set_server_check_status(check, HCHK_STATUS_START, NULL);
2102
2103 check->state |= CHK_ST_INPROGRESS;
2104
Simon Hormandbf70192015-01-30 11:22:53 +09002105 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002106 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002107 /* the process was forked, we allow up to min(inter,
2108 * timeout.connect) for it to report its status, but
2109 * only when timeout.check is set as it may be to short
2110 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002111 */
2112 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2113
2114 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2115 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2116 t->expire = tick_first(t->expire, t_con);
2117 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002118 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002119 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002120 }
2121
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002122 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002123
2124 check->state &= ~CHK_ST_INPROGRESS;
2125 check_notify_failure(check);
2126
2127 /* we allow up to min(inter, timeout.connect) for a connection
2128 * to establish but only when timeout.check is set
2129 * as it may be to short for a full check otherwise
2130 */
2131 while (tick_is_expired(t->expire, now_ms)) {
2132 int t_con;
2133
2134 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2135 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2136
2137 if (s->proxy->timeout.check)
2138 t->expire = tick_first(t->expire, t_con);
2139 }
2140 }
2141 else {
2142 /* there was a test running.
2143 * First, let's check whether there was an uncaught error,
2144 * which can happen on connect timeout or error.
2145 */
2146 if (check->result == CHK_RES_UNKNOWN) {
2147 /* good connection is enough for pure TCP check */
2148 struct pid_list *elem = check->curpid;
2149 int status = HCHK_STATUS_UNKNOWN;
2150
2151 if (elem->exited) {
2152 status = elem->status; /* Save in case the process exits between use below */
2153 if (!WIFEXITED(status))
2154 check->code = -1;
2155 else
2156 check->code = WEXITSTATUS(status);
2157 if (!WIFEXITED(status) || WEXITSTATUS(status))
2158 status = HCHK_STATUS_PROCERR;
2159 else
2160 status = HCHK_STATUS_PROCOK;
2161 } else if (expired) {
2162 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002163 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002164 kill(elem->pid, SIGTERM);
2165 }
2166 set_server_check_status(check, status, NULL);
2167 }
2168
2169 if (check->result == CHK_RES_FAILED) {
2170 /* a failure or timeout detected */
2171 check_notify_failure(check);
2172 }
2173 else if (check->result == CHK_RES_CONDPASS) {
2174 /* check is OK but asks for stopping mode */
2175 check_notify_stopping(check);
2176 }
2177 else if (check->result == CHK_RES_PASSED) {
2178 /* a success was detected */
2179 check_notify_success(check);
2180 }
Willy Tarreaub1437112019-09-03 18:55:02 +02002181 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002182 check->state &= ~CHK_ST_INPROGRESS;
2183
2184 pid_list_del(check->curpid);
2185
2186 rv = 0;
2187 if (global.spread_checks > 0) {
2188 rv = srv_getinter(check) * global.spread_checks / 100;
2189 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2190 }
2191 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2192 }
2193
2194 reschedule:
2195 while (tick_is_expired(t->expire, now_ms))
2196 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002197
2198 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002199 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002200 return t;
2201}
2202
2203/*
2204 * manages a server health-check that uses a connection. Returns
2205 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002206 *
2207 * Please do NOT place any return statement in this function and only leave
2208 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002209 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002210static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002211{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002212 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002213 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002214 struct conn_stream *cs = check->cs;
2215 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002216 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002217 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002218 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002219
Olivier Houchard0923fa42019-01-11 18:43:04 +01002220 if (check->server)
2221 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002222 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002223 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002224 if (!expired) /* woke up too early */
2225 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002226
Simon Horman671b6f02013-11-25 10:46:39 +09002227 /* we don't send any health-checks when the proxy is
2228 * stopped, the server should not be checked or the check
2229 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002230 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002231 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002232 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002233 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002234
2235 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002236 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002237
Willy Tarreau2c115e52013-12-11 19:41:16 +01002238 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002239 b_reset(&check->bi);
2240 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002241
Olivier Houchard68123bc2019-11-29 16:18:51 +01002242 task_set_affinity(t, tid_bit);
Simon Hormandbf70192015-01-30 11:22:53 +09002243 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002244 cs = check->cs;
2245 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002246
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002247 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002248 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002249 goto out_unlock;
2250
Willy Tarreaue7dff022015-04-03 01:14:29 +02002251 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002252 /* we allow up to min(inter, timeout.connect) for a connection
2253 * to establish but only when timeout.check is set
2254 * as it may be to short for a full check otherwise
2255 */
Simon Horman4a741432013-02-23 15:35:38 +09002256 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002257
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002258 if (proxy->timeout.check && proxy->timeout.connect) {
2259 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002260 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002261 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002262
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002263 if (check->type) {
2264 /* send the request if we have one. We avoid receiving
2265 * if not connected, unless we didn't subscribe for
2266 * sending since otherwise we won't be woken up.
2267 */
2268 __event_srv_chk_w(cs);
Willy Tarreauab160a42019-09-05 17:38:40 +02002269 if (!(conn->flags & CO_FL_WAIT_L4_CONN) ||
2270 !(check->wait_list.events & SUB_RETRY_SEND))
2271 __event_srv_chk_r(cs);
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002272 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002273
Willy Tarreau5a78f362012-11-23 12:47:05 +01002274 goto reschedule;
2275
Willy Tarreaue7dff022015-04-03 01:14:29 +02002276 case SF_ERR_SRVTO: /* ETIMEDOUT */
2277 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002278 if (conn)
2279 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002280 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002281 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002282 /* should share same code than cases below */
2283 case SF_ERR_CHK_PORT:
2284 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002285 case SF_ERR_PRXCOND:
2286 case SF_ERR_RESOURCE:
2287 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002288 if (conn)
2289 conn->flags |= CO_FL_ERROR;
2290 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002291 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002292 }
2293
Willy Tarreau5a78f362012-11-23 12:47:05 +01002294 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard68123bc2019-11-29 16:18:51 +01002295 task_set_affinity(t, MAX_THREADS_MASK);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002296 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002297 if (check->wait_list.events)
2298 cs->conn->xprt->unsubscribe(cs->conn,
2299 cs->conn->xprt_ctx,
2300 check->wait_list.events,
2301 &check->wait_list);
2302 /* We may have been scheduled to run, and the
2303 * I/O handler expects to have a cs, so remove
2304 * the tasklet
2305 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002306 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002307 cs_destroy(cs);
2308 cs = check->cs = NULL;
2309 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002310 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002311
Willy Tarreau2c115e52013-12-11 19:41:16 +01002312 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002313 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002314
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002315 /* we allow up to min(inter, timeout.connect) for a connection
2316 * to establish but only when timeout.check is set
2317 * as it may be to short for a full check otherwise
2318 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002319 while (tick_is_expired(t->expire, now_ms)) {
2320 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002321
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002322 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002323 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002324
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002325 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002326 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002327 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002328 }
2329 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002330 /* there was a test running.
2331 * First, let's check whether there was an uncaught error,
2332 * which can happen on connect timeout or error.
2333 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002334 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002335 /* good connection is enough for pure TCP check */
2336 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002337 if (check->use_ssl)
2338 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002339 else
Simon Horman4a741432013-02-23 15:35:38 +09002340 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002341 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002342 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002343 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002344 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002345 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002346 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002347 }
2348
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002349 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002350 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002351 /* The check was aborted and the connection was not yet closed.
2352 * This can happen upon timeout, or when an external event such
2353 * as a failed response coupled with "observe layer7" caused the
2354 * server state to be suddenly changed.
2355 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002356 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002357 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002358 }
2359
Willy Tarreauac59f362017-10-08 11:10:19 +02002360 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002361 if (check->wait_list.events)
2362 cs->conn->xprt->unsubscribe(cs->conn,
2363 cs->conn->xprt_ctx,
2364 check->wait_list.events,
2365 &check->wait_list);
2366 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002367 * I/O handler expects to have a cs, so remove
2368 * the tasklet
2369 */
2370 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002371 cs_destroy(cs);
2372 cs = check->cs = NULL;
2373 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002374 }
2375
Olivier Houchard0923fa42019-01-11 18:43:04 +01002376 if (check->server) {
2377 if (check->result == CHK_RES_FAILED) {
2378 /* a failure or timeout detected */
2379 check_notify_failure(check);
2380 }
2381 else if (check->result == CHK_RES_CONDPASS) {
2382 /* check is OK but asks for stopping mode */
2383 check_notify_stopping(check);
2384 }
2385 else if (check->result == CHK_RES_PASSED) {
2386 /* a success was detected */
2387 check_notify_success(check);
2388 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002389 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002390 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002391 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002392
Olivier Houchard0923fa42019-01-11 18:43:04 +01002393 if (check->server) {
2394 rv = 0;
2395 if (global.spread_checks > 0) {
2396 rv = srv_getinter(check) * global.spread_checks / 100;
2397 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2398 }
2399 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002400 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002401 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002402
2403 reschedule:
2404 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002405 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002406 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002407 if (check->server)
2408 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002409 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002410}
2411
Simon Horman98637e52014-06-20 12:30:16 +09002412/*
2413 * manages a server health-check. Returns
2414 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2415 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002416static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002417{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002418 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002419
2420 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002421 return process_chk_proc(t, context, state);
2422 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002423
Simon Horman98637e52014-06-20 12:30:16 +09002424}
2425
Simon Horman5c942422013-11-25 10:46:32 +09002426static int start_check_task(struct check *check, int mininter,
2427 int nbcheck, int srvpos)
2428{
2429 struct task *t;
Willy Tarreaub1437112019-09-03 18:55:02 +02002430 unsigned long thread_mask = MAX_THREADS_MASK;
2431
2432 if (check->type == PR_O2_EXT_CHK)
2433 thread_mask = 1;
2434
Simon Horman5c942422013-11-25 10:46:32 +09002435 /* task for the check */
Willy Tarreaub1437112019-09-03 18:55:02 +02002436 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002437 ha_alert("Starting [%s:%s] check: out of memory.\n",
2438 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002439 return 0;
2440 }
2441
2442 check->task = t;
2443 t->process = process_chk;
2444 t->context = check;
2445
Willy Tarreau1746eec2014-04-25 10:46:47 +02002446 if (mininter < srv_getinter(check))
2447 mininter = srv_getinter(check);
2448
2449 if (global.max_spread_checks && mininter > global.max_spread_checks)
2450 mininter = global.max_spread_checks;
2451
Simon Horman5c942422013-11-25 10:46:32 +09002452 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002453 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002454 check->start = now;
2455 task_queue(t);
2456
2457 return 1;
2458}
2459
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002460/*
2461 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002462 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002463 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002464static int start_checks()
2465{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002466
2467 struct proxy *px;
2468 struct server *s;
2469 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002470 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002471
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002472 /* 1- count the checkers to run simultaneously.
2473 * We also determine the minimum interval among all of those which
2474 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2475 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002476 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002477 * too short an interval for all others.
2478 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002479 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002480 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002481 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002482 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002483 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002484 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002485 }
2486 /* We need a warmup task that will be called when the server
2487 * state switches from down to up.
2488 */
2489 s->warmup = t;
2490 t->process = server_warmup;
2491 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002492 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002493 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002494 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 +01002495 }
2496
Willy Tarreaud8514a22013-12-11 21:10:14 +01002497 if (s->check.state & CHK_ST_CONFIGURED) {
2498 nbcheck++;
2499 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2500 (!mininter || mininter > srv_getinter(&s->check)))
2501 mininter = srv_getinter(&s->check);
2502 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002503
Willy Tarreaud8514a22013-12-11 21:10:14 +01002504 if (s->agent.state & CHK_ST_CONFIGURED) {
2505 nbcheck++;
2506 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2507 (!mininter || mininter > srv_getinter(&s->agent)))
2508 mininter = srv_getinter(&s->agent);
2509 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002510 }
2511 }
2512
Simon Horman4a741432013-02-23 15:35:38 +09002513 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002514 return 0;
2515
2516 srand((unsigned)time(NULL));
2517
2518 /*
2519 * 2- start them as far as possible from each others. For this, we will
2520 * start them after their interval set to the min interval divided by
2521 * the number of servers, weighted by the server's position in the list.
2522 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002523 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002524 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2525 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002526 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002527 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002528 }
2529 }
2530
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002531 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002532 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002533 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002534 if (s->check.type == PR_O2_EXT_CHK) {
2535 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002536 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002537 }
Simon Hormand60d6912013-11-25 10:46:36 +09002538 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002539 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002540 srvpos++;
2541 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002542
Simon Hormand60d6912013-11-25 10:46:36 +09002543 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002544 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002545 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002546 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002547 }
2548 srvpos++;
2549 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002550 }
2551 }
2552 return 0;
2553}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002554
2555/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002556 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002557 * The buffer MUST be terminated by a null byte before calling this function.
2558 * Sets server status appropriately. The caller is responsible for ensuring
2559 * that the buffer contains at least 13 characters. If <done> is zero, we may
2560 * return 0 to indicate that data is required to decide of a match.
2561 */
2562static int httpchk_expect(struct server *s, int done)
2563{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002564 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002565 char status_code[] = "000";
2566 char *contentptr;
2567 int crlf;
2568 int ret;
2569
2570 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2571 case PR_O2_EXP_STS:
2572 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002573 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2574 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002575
2576 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2577 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2578 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002579 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002580
2581 /* we necessarily have the response, so there are no partial failures */
2582 if (s->proxy->options2 & PR_O2_EXP_INV)
2583 ret = !ret;
2584
Simon Horman4a741432013-02-23 15:35:38 +09002585 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002586 break;
2587
2588 case PR_O2_EXP_STR:
2589 case PR_O2_EXP_RSTR:
2590 /* very simple response parser: ignore CR and only count consecutive LFs,
2591 * stop with contentptr pointing to first char after the double CRLF or
2592 * to '\0' if crlf < 2.
2593 */
2594 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002595 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002596 if (crlf >= 2)
2597 break;
2598 if (*contentptr == '\r')
2599 continue;
2600 else if (*contentptr == '\n')
2601 crlf++;
2602 else
2603 crlf = 0;
2604 }
2605
2606 /* Check that response contains a body... */
2607 if (crlf < 2) {
2608 if (!done)
2609 return 0;
2610
Simon Horman4a741432013-02-23 15:35:38 +09002611 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002612 "HTTP content check could not find a response body");
2613 return 1;
2614 }
2615
2616 /* Check that response body is not empty... */
2617 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002618 if (!done)
2619 return 0;
2620
Simon Horman4a741432013-02-23 15:35:38 +09002621 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002622 "HTTP content check found empty response body");
2623 return 1;
2624 }
2625
2626 /* Check the response content against the supplied string
2627 * or regex... */
2628 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2629 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2630 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002631 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002632
2633 /* if we don't match, we may need to wait more */
2634 if (!ret && !done)
2635 return 0;
2636
2637 if (ret) {
2638 /* content matched */
2639 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002640 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002641 "HTTP check matched unwanted content");
2642 else
Simon Horman4a741432013-02-23 15:35:38 +09002643 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002644 "HTTP content check matched");
2645 }
2646 else {
2647 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002648 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002649 "HTTP check did not match unwanted content");
2650 else
Simon Horman4a741432013-02-23 15:35:38 +09002651 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002652 "HTTP content check did not match");
2653 }
2654 break;
2655 }
2656 return 1;
2657}
2658
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002659/*
2660 * return the id of a step in a send/expect session
2661 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002662static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663{
2664 struct tcpcheck_rule *cur = NULL, *next = NULL;
2665 int i = 0;
2666
Willy Tarreau213c6782014-10-02 14:51:02 +02002667 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002668 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002669 return 0;
2670
Simon Hormane16c1b32015-01-30 11:22:57 +09002671 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002672
2673 /* no step => first step */
2674 if (cur == NULL)
2675 return 1;
2676
2677 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002678 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002679 if (next->list.p == &cur->list)
2680 break;
2681 ++i;
2682 }
2683
2684 return i;
2685}
2686
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002687/*
2688 * return the latest known comment before (including) the given stepid
2689 * returns NULL if no comment found
2690 */
2691static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2692{
2693 struct tcpcheck_rule *cur = NULL;
2694 char *ret = NULL;
2695 int i = 0;
2696
2697 /* not even started anything yet, return latest comment found before any action */
2698 if (!check->current_step) {
2699 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2700 if (cur->action == TCPCHK_ACT_COMMENT)
2701 ret = cur->comment;
2702 else
2703 goto return_comment;
2704 }
2705 }
2706
2707 i = 1;
2708 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2709 if (cur->comment)
2710 ret = cur->comment;
2711
2712 if (i >= stepid)
2713 goto return_comment;
2714
2715 ++i;
2716 }
2717
2718 return_comment:
2719 return ret;
2720}
2721
Willy Tarreaube74b882017-10-04 16:22:49 +02002722/* proceed with next steps for the TCP checks <check>. Note that this is called
2723 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002724 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2725 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002726 *
2727 * Please do NOT place any return statement in this function and only leave
Willy Tarreau495c3a42019-07-17 18:48:07 +02002728 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002729 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002730static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002731{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002732 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002733 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002734 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002735 struct conn_stream *cs = check->cs;
2736 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002737 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002738 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002739 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002740 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002741 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002742
Willy Tarreauef953952014-10-02 14:30:14 +02002743 /* here, we know that the check is complete or that it failed */
2744 if (check->result != CHK_RES_UNKNOWN)
2745 goto out_end_tcpcheck;
2746
2747 /* We have 4 possibilities here :
2748 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002749 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002750 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2751 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002752 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002753 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002754 * be waiting for a connection attempt to complete. conn!=NULL.
2755 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002756 *
2757 * #2 and #3 are quite similar, we want both the connection and the
2758 * handshake to complete before going any further. Thus we must always
2759 * wait for a connection to complete unless we're before and existing
2760 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002761 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002762
2763 /* find first rule and skip comments */
2764 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2765 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2766 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2767
Willy Tarreau00149122017-10-04 18:05:01 +02002768 if ((check->current_step || &next->list == head) &&
2769 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002770 /* we allow up to min(inter, timeout.connect) for a connection
2771 * to establish but only when timeout.check is set
2772 * as it may be to short for a full check otherwise
2773 */
2774 while (tick_is_expired(t->expire, now_ms)) {
2775 int t_con;
2776
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002777 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002778 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2779
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002780 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002781 t->expire = tick_first(t->expire, t_con);
2782 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002783 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002784 }
2785
Willy Tarreauef953952014-10-02 14:30:14 +02002786 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002787 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002788 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002789 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002790 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002791
Willy Tarreau213c6782014-10-02 14:51:02 +02002792 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002793 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002794 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002795 b_reset(&check->bo);
2796 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002797 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002798 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002799 if (proxy->timeout.check)
2800 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002801 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002802
Willy Tarreauabca5b62013-12-06 14:19:25 +01002803 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002804 /* We have to try to flush the output buffer before reading, at
2805 * the end, or if we're about to send a string that does not fit
2806 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002807 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002808 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002809 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002810 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002811 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002812 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002813 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002814
Olivier Houcharded0f2072018-08-16 15:41:52 +02002815 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002816 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002817
2818 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002819 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002820 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002821 goto out_end_tcpcheck;
2822 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002823 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002824 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002825 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002826 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002827 goto out;
2828 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002829 }
2830
Willy Tarreau263013d2015-05-13 11:59:14 +02002831 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002832 break;
2833
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002834 /* have 'next' point to the next rule or NULL if we're on the
2835 * last one, connect() needs this.
2836 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002837 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002838
2839 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002840 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002841 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002842
2843 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002844 if (&next->list == head)
2845 next = NULL;
2846
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002847 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2848 struct protocol *proto;
2849 struct xprt_ops *xprt;
2850
Willy Tarreau00149122017-10-04 18:05:01 +02002851 /* For a connect action we'll create a new connection.
2852 * We may also have to kill a previous one. But we don't
2853 * want to leave *without* a connection if we came here
2854 * from the connection layer, hence with a connection.
2855 * Thus we'll proceed in the following order :
2856 * 1: close but not release previous connection
2857 * 2: try to get a new connection
2858 * 3: release and replace the old one on success
2859 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002860 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002861 cs_close(check->cs);
2862 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002863 }
2864
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002865 /* mark the step as started */
2866 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002867
2868 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002869 cs = cs_new(NULL);
2870 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002871 step = tcpcheck_get_step_id(check);
2872 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2873 comment = tcpcheck_get_step_comment(check, step);
2874 if (comment)
2875 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002876 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2877 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002878 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002879 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002880 }
2881
Olivier Houchard49065542019-05-31 19:20:36 +02002882 if (check->cs) {
2883 if (check->wait_list.events)
2884 cs->conn->xprt->unsubscribe(cs->conn,
2885 cs->conn->xprt_ctx,
2886 check->wait_list.events,
2887 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002888 /* We may have been scheduled to run, and the
2889 * I/O handler expects to have a cs, so remove
2890 * the tasklet
2891 */
2892 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002893 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002894 }
Willy Tarreau00149122017-10-04 18:05:01 +02002895
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002896 check->cs = cs;
2897 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002898 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002899 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002900 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002901
2902 /* no client address */
2903 clear_addr(&conn->addr.from);
2904
Simon Horman41f58762015-01-30 11:22:56 +09002905 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002906 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002907 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002908 }
2909 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002910 /* we'll connect to the addr on the server */
2911 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002912 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002913 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002914
2915 /* port */
2916 if (check->current_step->port)
2917 set_host_port(&conn->addr.to, check->current_step->port);
2918 else if (check->port)
2919 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002920 else if (s->svc_port)
2921 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002922
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002923 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002924 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002925 }
2926 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002927 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002928 }
Willy Tarreaube373152018-09-06 11:45:30 +02002929
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002930 conn_prepare(conn, proto, xprt);
Willy Tarreau495c3a42019-07-17 18:48:07 +02002931
2932 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2933 ret = SF_ERR_RESOURCE;
2934 goto fail_check;
2935 }
2936
Willy Tarreaube373152018-09-06 11:45:30 +02002937 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002938
Willy Tarreaue7dff022015-04-03 01:14:29 +02002939 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002940 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002941 ret = proto->connect(conn,
Olivier Houchardfdcb0072019-05-06 18:32:29 +02002942 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 +01002943 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2944 conn->send_proxy_ofs = 1;
2945 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002946 if (xprt_add_hs(conn) < 0)
2947 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002948 }
2949
2950 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002951 * - SF_ERR_NONE if everything's OK
2952 * - SF_ERR_SRVTO if there are no more servers
2953 * - SF_ERR_SRVCL if the connection was refused by the server
2954 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2955 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2956 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002957 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002958 * Note that we try to prevent the network stack from sending the ACK during the
2959 * connect() when a pure TCP check is used (without PROXY protocol).
2960 */
Willy Tarreau495c3a42019-07-17 18:48:07 +02002961 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002962 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002963 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002964 /* we allow up to min(inter, timeout.connect) for a connection
2965 * to establish but only when timeout.check is set
2966 * as it may be to short for a full check otherwise
2967 */
2968 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2969
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002970 if (proxy->timeout.check && proxy->timeout.connect) {
2971 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002972 t->expire = tick_first(t->expire, t_con);
2973 }
2974 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002975 case SF_ERR_SRVTO: /* ETIMEDOUT */
2976 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002977 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002978 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002979 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002980 comment = tcpcheck_get_step_comment(check, step);
2981 if (comment)
2982 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002983 set_server_check_status(check, HCHK_STATUS_L4CON,
2984 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002985 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002986 case SF_ERR_PRXCOND:
2987 case SF_ERR_RESOURCE:
2988 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002989 step = tcpcheck_get_step_id(check);
2990 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002991 comment = tcpcheck_get_step_comment(check, step);
2992 if (comment)
2993 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002994 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2995 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002996 goto out_end_tcpcheck;
2997 }
2998
2999 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003000 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003001
3002 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003003 while (&check->current_step->list != head &&
3004 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003005 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003006
Willy Tarreauf2c87352015-05-13 12:08:21 +02003007 if (&check->current_step->list == head)
3008 break;
3009
Willy Tarreau7199e962019-07-15 10:57:51 +02003010 /* don't do anything until the connection is established */
3011 if (!(conn->flags & CO_FL_CONNECTED))
3012 break;
3013
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003014 } /* end 'connect' */
3015 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3016 /* mark the step as started */
3017 check->last_started_step = check->current_step;
3018
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003019 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003020 if (*b_head(&check->bi) != '\0') {
3021 *b_head(&check->bi) = '\0';
3022 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003023 }
3024
Willy Tarreaubbae3f02017-08-30 09:59:52 +02003025 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003026 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003027 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003028 goto out_end_tcpcheck;
3029 }
3030
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003031 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003032 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003033 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003034 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003035 set_server_check_status(check, HCHK_STATUS_L7RSP,
3036 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003037 goto out_end_tcpcheck;
3038 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003039
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003040 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003041 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003042 continue;
3043
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003044 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3045 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003046
Willy Tarreauabca5b62013-12-06 14:19:25 +01003047 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003048 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003049
3050 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003051 while (&check->current_step->list != head &&
3052 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003053 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003054
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003055 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003056 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003057 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003058 goto out_end_tcpcheck;
3059
Olivier Houchard48bf0572019-07-18 15:09:08 +02003060 /* If we already subscribed, then we tried to received
3061 * and failed, so there's no point trying again.
3062 */
3063 if (check->wait_list.events & SUB_RETRY_RECV)
3064 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003065 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003066 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003067 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003068 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003069 /* Report network errors only if we got no other data. Otherwise
3070 * we'll let the upper layers decide whether the response is OK
3071 * or not. It is very common that an RST sent by the server is
3072 * reported as an error just after the last data chunk.
3073 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003074 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003075 goto out_end_tcpcheck;
3076 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003077 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003078 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003079 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003080 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003081 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003082 }
3083
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003084 /* mark the step as started */
3085 check->last_started_step = check->current_step;
3086
3087
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003088 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003089 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003090 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003091 if (b_data(&check->bi) < b_size(&check->bi)) {
3092 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003093 }
3094 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003095 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003096 done = 1; /* buffer full, don't wait for more data */
3097 }
3098
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003099 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003100
3101 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003102 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003103 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003104 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003105
3106 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003107 step = tcpcheck_get_step_id(check);
3108 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003109 comment = tcpcheck_get_step_comment(check, step);
3110 if (comment)
3111 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003112 set_server_check_status(check, HCHK_STATUS_L7RSP,
3113 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003114
3115 goto out_end_tcpcheck;
3116 }
3117
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003118 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003119 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003120
Willy Tarreaua970c282013-12-06 12:47:19 +01003121 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003122 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003123 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 +02003124 else if (check->current_step->expect_regex != NULL)
3125 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003126
3127 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003128 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003129
3130 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003131 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003132 if (ret) {
3133 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003134 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003135 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003136 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003137 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003138 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003139 }
3140 else {
3141 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003142 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003143 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003144 comment = tcpcheck_get_step_comment(check, step);
3145 if (comment)
3146 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003147 set_server_check_status(check, HCHK_STATUS_L7RSP,
3148 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003149 goto out_end_tcpcheck;
3150 }
3151 /* matched and was supposed to => OK, next step */
3152 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003153 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003154 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003155
3156 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003157 while (&check->current_step->list != head &&
3158 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003159 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003160
Willy Tarreauf2c87352015-05-13 12:08:21 +02003161 if (&check->current_step->list == head)
3162 break;
3163
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003164 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003165 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003166 }
3167 }
3168 else {
3169 /* not matched */
3170 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003171 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003172 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003173 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003174
3175 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003176 while (&check->current_step->list != head &&
3177 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003178 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003179
Willy Tarreauf2c87352015-05-13 12:08:21 +02003180 if (&check->current_step->list == head)
3181 break;
3182
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003183 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003184 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003185 }
3186 /* not matched but was supposed to => ERROR */
3187 else {
3188 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003189 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003190 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003191 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003192 }
3193 else {
3194 /* we were looking for a regex */
3195 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003196 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003197 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003198 comment = tcpcheck_get_step_comment(check, step);
3199 if (comment)
3200 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003201 set_server_check_status(check, HCHK_STATUS_L7RSP,
3202 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003203 goto out_end_tcpcheck;
3204 }
3205 }
3206 } /* end expect */
3207 } /* end loop over double chained step list */
3208
Baptiste Assmann248f1172018-03-01 21:49:01 +01003209 /* don't do anything until the connection is established */
3210 if (!(conn->flags & CO_FL_CONNECTED)) {
3211 /* update expire time, should be done by process_chk */
3212 /* we allow up to min(inter, timeout.connect) for a connection
3213 * to establish but only when timeout.check is set
3214 * as it may be to short for a full check otherwise
3215 */
3216 while (tick_is_expired(t->expire, now_ms)) {
3217 int t_con;
3218
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003219 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003220 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3221
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003222 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003223 t->expire = tick_first(t->expire, t_con);
3224 }
3225 goto out;
3226 }
3227
Willy Tarreau263013d2015-05-13 11:59:14 +02003228 /* We're waiting for some I/O to complete, we've reached the end of the
3229 * rules, or both. Do what we have to do, otherwise we're done.
3230 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003231 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003232 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3233 goto out_end_tcpcheck;
3234 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003235
Willy Tarreau53c5a042015-05-13 11:38:17 +02003236 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003237 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003238 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003239 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003240
3241 out_end_tcpcheck:
3242 /* collect possible new errors */
Willy Tarreau3d84b1c2019-07-23 14:37:47 +02003243 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003244 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003245
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003246 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003247 check->current_step = NULL;
3248
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003249 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003250 conn->flags |= CO_FL_ERROR;
3251
Christopher Fauletb6102852017-11-28 10:06:29 +01003252 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003253 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003254}
3255
Simon Hormanb1900d52015-01-30 11:22:54 +09003256const char *init_check(struct check *check, int type)
3257{
3258 check->type = type;
3259
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003260 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3261 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003262
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003263 check->bi.area = calloc(check->bi.size, sizeof(char));
3264 check->bo.area = calloc(check->bo.size, sizeof(char));
3265
3266 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003267 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003268
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003269 check->wait_list.tasklet = tasklet_new();
3270 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003271 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003272 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003273 check->wait_list.tasklet->process = event_srv_chk_io;
3274 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003275 return NULL;
3276}
3277
Simon Hormanbfb5d332015-01-30 11:22:55 +09003278void free_check(struct check *check)
3279{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003280 free(check->bi.area);
3281 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003282 if (check->cs) {
3283 free(check->cs->conn);
3284 check->cs->conn = NULL;
3285 cs_free(check->cs);
3286 check->cs = NULL;
3287 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003288}
3289
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003290void email_alert_free(struct email_alert *alert)
3291{
3292 struct tcpcheck_rule *rule, *back;
3293
3294 if (!alert)
3295 return;
3296
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003297 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3298 LIST_DEL(&rule->list);
3299 free(rule->comment);
3300 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003301 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003302 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003303 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003304 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003305}
3306
Olivier Houchard9f6af332018-05-25 14:04:04 +02003307static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003308{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003309 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003310 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003311 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003312
3313 q = container_of(check, typeof(*q), check);
3314
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003315 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003316 while (1) {
3317 if (!(check->state & CHK_ST_ENABLED)) {
3318 if (LIST_ISEMPTY(&q->email_alerts)) {
3319 /* All alerts processed, queue the task */
3320 t->expire = TICK_ETERNITY;
3321 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003322 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003323 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003324
3325 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003326 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003327 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003328 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003329 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003330 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003331 }
3332
Olivier Houchard9f6af332018-05-25 14:04:04 +02003333 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003334 if (check->state & CHK_ST_INPROGRESS)
3335 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003336
3337 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3338 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003339 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003340 check->server = NULL;
3341 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003342 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003343 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003344 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003345 return t;
3346}
3347
Christopher Faulet0108bb32017-10-20 21:34:32 +02003348/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3349 *
3350 * The function returns 1 in success case, otherwise, it returns 0 and err is
3351 * filled.
3352 */
3353int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003354{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003355 struct mailer *mailer;
3356 struct email_alertq *queues;
3357 const char *err_str;
3358 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003359
Christopher Faulet0108bb32017-10-20 21:34:32 +02003360 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3361 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003362 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003363 }
3364
Christopher Faulet0108bb32017-10-20 21:34:32 +02003365 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3366 struct email_alertq *q = &queues[i];
3367 struct check *check = &q->check;
3368 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003369
3370 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003371 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003372 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003373 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003374 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003375 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003376 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3377 memprintf(err, "%s", err_str);
3378 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003379 }
3380
3381 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003382 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003383 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003384
Emeric Brunc60def82017-09-27 14:59:38 +02003385 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003386 memprintf(err, "out of memory while allocating mailer alerts task");
3387 goto error;
3388 }
3389
3390 check->task = t;
3391 t->process = process_email_alert;
3392 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003393
Christopher Faulet0108bb32017-10-20 21:34:32 +02003394 /* check this in one ms */
3395 t->expire = TICK_ETERNITY;
3396 check->start = now;
3397 task_queue(t);
3398 }
3399
3400 mls->users++;
3401 free(p->email_alert.mailers.name);
3402 p->email_alert.mailers.m = mls;
3403 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003404 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003405
3406 error:
3407 for (i = 0; i < mls->count; i++) {
3408 struct email_alertq *q = &queues[i];
3409 struct check *check = &q->check;
3410
Willy Tarreauf6562792019-05-07 19:05:35 +02003411 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003412 free_check(check);
3413 }
3414 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003415 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003416 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003417}
3418
3419
3420static int add_tcpcheck_expect_str(struct list *list, const char *str)
3421{
3422 struct tcpcheck_rule *tcpcheck;
3423
Willy Tarreaubafbe012017-11-24 17:34:44 +01003424 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003425 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003426 memset(tcpcheck, 0, sizeof(*tcpcheck));
3427 tcpcheck->action = TCPCHK_ACT_EXPECT;
3428 tcpcheck->string = strdup(str);
3429 tcpcheck->expect_regex = NULL;
3430 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003431 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003432 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003433 return 0;
3434 }
3435
3436 LIST_ADDQ(list, &tcpcheck->list);
3437 return 1;
3438}
3439
3440static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3441{
3442 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003443 const char *in;
3444 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003445 int i;
3446
Willy Tarreaubafbe012017-11-24 17:34:44 +01003447 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003448 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003449 memset(tcpcheck, 0, sizeof(*tcpcheck));
3450 tcpcheck->action = TCPCHK_ACT_SEND;
3451 tcpcheck->expect_regex = NULL;
3452 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003453 tcpcheck->string_len = 0;
3454 for (i = 0; strs[i]; i++)
3455 tcpcheck->string_len += strlen(strs[i]);
3456
3457 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3458 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003459 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003460 return 0;
3461 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003462
Willy Tarreau64345aa2016-08-10 19:29:09 +02003463 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003464 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003465 for (in = strs[i]; (*dst = *in++); dst++);
3466 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003467
3468 LIST_ADDQ(list, &tcpcheck->list);
3469 return 1;
3470}
3471
Christopher Faulet0108bb32017-10-20 21:34:32 +02003472static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3473 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003474{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003475 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003476 struct tcpcheck_rule *tcpcheck;
3477 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478
Willy Tarreaubafbe012017-11-24 17:34:44 +01003479 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003480 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003481 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003482 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003483 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003484
Willy Tarreaubafbe012017-11-24 17:34:44 +01003485 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003486 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003487 memset(tcpcheck, 0, sizeof(*tcpcheck));
3488 tcpcheck->action = TCPCHK_ACT_CONNECT;
3489 tcpcheck->comment = NULL;
3490 tcpcheck->string = NULL;
3491 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003492 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3493
3494 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3495 goto error;
3496
3497 {
3498 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3499 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3500 goto error;
3501 }
3502
3503 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3504 goto error;
3505
3506 {
3507 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3508 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3509 goto error;
3510 }
3511
3512 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3513 goto error;
3514
3515 {
3516 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3517 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3518 goto error;
3519 }
3520
3521 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3522 goto error;
3523
3524 {
3525 const char * const strs[2] = { "DATA\r\n" };
3526 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3527 goto error;
3528 }
3529
3530 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3531 goto error;
3532
3533 {
3534 struct tm tm;
3535 char datestr[48];
3536 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003537 "From: ", p->email_alert.from, "\r\n",
3538 "To: ", p->email_alert.to, "\r\n",
3539 "Date: ", datestr, "\r\n",
3540 "Subject: [HAproxy Alert] ", msg, "\r\n",
3541 "\r\n",
3542 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003543 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003544 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003545 NULL
3546 };
3547
3548 get_localtime(date.tv_sec, &tm);
3549
3550 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3551 goto error;
3552 }
3553
3554 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3555 goto error;
3556 }
3557
3558 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3559 goto error;
3560
3561 {
3562 const char * const strs[2] = { "QUIT\r\n" };
3563 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3564 goto error;
3565 }
3566
3567 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3568 goto error;
3569
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003570 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003571 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003572 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003573 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003574 return 1;
3575
3576error:
3577 email_alert_free(alert);
3578 return 0;
3579}
3580
Christopher Faulet0108bb32017-10-20 21:34:32 +02003581static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003582{
3583 int i;
3584 struct mailer *mailer;
3585
3586 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3587 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003588 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003589 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003590 return;
3591 }
3592 }
3593
3594 return;
3595}
3596
3597/*
3598 * Send email alert if configured.
3599 */
Simon Horman64e34162015-02-06 11:11:57 +09003600void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003601{
3602 va_list argp;
3603 char buf[1024];
3604 int len;
3605 struct proxy *p = s->proxy;
3606
Christopher Faulet0108bb32017-10-20 21:34:32 +02003607 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003608 return;
3609
3610 va_start(argp, format);
3611 len = vsnprintf(buf, sizeof(buf), format, argp);
3612 va_end(argp);
3613
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003614 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003615 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003616 return;
3617 }
3618
Christopher Faulet0108bb32017-10-20 21:34:32 +02003619 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003620}
3621
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003622/*
3623 * Return value:
3624 * the port to be used for the health check
3625 * 0 in case no port could be found for the check
3626 */
3627int srv_check_healthcheck_port(struct check *chk)
3628{
3629 int i = 0;
3630 struct server *srv = NULL;
3631
3632 srv = chk->server;
3633
3634 /* If neither a port nor an addr was specified and no check transport
3635 * layer is forced, then the transport layer used by the checks is the
3636 * same as for the production traffic. Otherwise we use raw_sock by
3637 * default, unless one is specified.
3638 */
3639 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003640 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003641 chk->send_proxy |= (srv->pp_opts);
3642 }
3643
3644 /* by default, we use the health check port ocnfigured */
3645 if (chk->port > 0)
3646 return chk->port;
3647
3648 /* try to get the port from check_core.addr if check.port not set */
3649 i = get_host_port(&chk->addr);
3650 if (i > 0)
3651 return i;
3652
3653 /* try to get the port from server address */
3654 /* prevent MAPPORTS from working at this point, since checks could
3655 * not be performed in such case (MAPPORTS impose a relative ports
3656 * based on live traffic)
3657 */
3658 if (srv->flags & SRV_F_MAPPORTS)
3659 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003660
3661 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003662 if (i > 0)
3663 return i;
3664
3665 return 0;
3666}
3667
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003668REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003669
Willy Tarreaubd741542010-03-16 18:46:54 +01003670/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003671 * Local variables:
3672 * c-indent-level: 8
3673 * c-basic-offset: 8
3674 * End:
3675 */