blob: ff39b4f6433ab764224fb5f676aabbc5099e8976 [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 Houchardbc89ad82019-07-09 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 Houchardbc89ad82019-07-09 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 Tarreau25e2ab52013-12-04 11:17:05 +0100794 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
795 if (s->proxy->timeout.check) {
796 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
797 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200798 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200799 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100800
Willy Tarreau83749182007-04-15 20:56:27 +0200801 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200802 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200803 out:
804 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200805}
806
Willy Tarreaubaaee002006-06-26 02:48:02 +0200807/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200808 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200809 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900810 * set_server_check_status() to update check->status, check->duration
811 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200812
813 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
814 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
815 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
816 * response to an SSL HELLO (the principle is that this is enough to
817 * distinguish between an SSL server and a pure TCP relay). All other cases will
818 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
819 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100820 *
821 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200822 * via the out label.
823 *
824 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200825 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200826static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200827{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200828 struct connection *conn = cs->conn;
829 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900830 struct server *s = check->server;
831 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200832 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100833 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200834 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200835
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100836 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200837 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200838
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200839 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100840 if (!(conn->flags & CO_FL_ERROR))
841 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200842 goto out;
843 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200844
Willy Tarreauc09572f2017-10-04 11:58:22 +0200845 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100846 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200847 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200848
Willy Tarreau83749182007-04-15 20:56:27 +0200849 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
850 * but the connection was closed on the remote end. Fortunately, recv still
851 * works correctly and we don't need to do the getsockopt() on linux.
852 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000853
854 /* Set buffer to point to the end of the data already read, and check
855 * that there is free space remaining. If the buffer is full, proceed
856 * with running the checks without attempting another socket read.
857 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000858
Willy Tarreau03938182010-03-17 21:52:07 +0100859 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000860
Olivier Houchard511efea2018-08-16 15:30:32 +0200861 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200862 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100863 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200864 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200865 /* Report network errors only if we got no other data. Otherwise
866 * we'll let the upper layers decide whether the response is OK
867 * or not. It is very common that an RST sent by the server is
868 * reported as an error just after the last data chunk.
869 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200870 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100871 goto out_wakeup;
872 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200873 }
874
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100875
Willy Tarreau03938182010-03-17 21:52:07 +0100876 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200877 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100878 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 if (b_data(&check->bi) < b_size(&check->bi))
880 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100881 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200882 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100883 done = 1; /* buffer full, don't wait for more data */
884 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200885
Nick Chalk57b1bf72010-03-16 15:50:46 +0000886 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900887 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200888 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200889 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100890 goto wait_more_data;
891
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100892 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200893 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
894 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
895 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
896 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
897 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
898 cut_crlf(b_head(&check->bi));
899 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200900
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100901 goto out_wakeup;
902 }
903
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200904 check->code = str2uic(b_head(&check->bi) + 9);
905 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200906
Willy Tarreaubd741542010-03-16 18:46:54 +0100907 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200908 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000909 /* 404 may be accepted as "stopping" only if the server was up */
910 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900911 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000912 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100913 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
914 /* Run content verification check... We know we have at least 13 chars */
915 if (!httpchk_expect(s, done))
916 goto wait_more_data;
917 }
918 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200919 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100920 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900921 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100922 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000923 else {
924 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900925 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000926 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200927 break;
928
929 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200930 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100931 goto wait_more_data;
932
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100933 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200934 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900935 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200936 else
Simon Horman4a741432013-02-23 15:35:38 +0900937 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200938 break;
939
940 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200941 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100942 goto wait_more_data;
943
Willy Tarreau5488a622019-08-06 16:26:31 +0200944 /* do not reset when closing, servers don't like this */
945 if (conn_ctrl_ready(cs->conn))
946 fdtab[cs->conn->handle.fd].linger_risk = 0;
947
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200948 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200949 if ((b_data(&check->bi) < strlen("000\r")) ||
950 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
951 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
952 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
953 cut_crlf(b_head(&check->bi));
954 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200955 goto out_wakeup;
956 }
957
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200958 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200959
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200960 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200961 cut_crlf(desc);
962
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100963 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200964 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900965 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200966 else
Simon Horman4a741432013-02-23 15:35:38 +0900967 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200968 break;
969
Simon Hormana2b9dad2013-02-12 10:45:54 +0900970 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100971 int status = HCHK_STATUS_CHECKED;
972 const char *hs = NULL; /* health status */
973 const char *as = NULL; /* admin status */
974 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200975 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100976 const char *err = NULL; /* first error to report */
977 const char *wrn = NULL; /* first warning to report */
978 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900979
Willy Tarreau81f5d942013-12-09 20:51:51 +0100980 /* We're getting an agent check response. The agent could
981 * have been disabled in the mean time with a long check
982 * still pending. It is important that we ignore the whole
983 * response.
984 */
985 if (!(check->server->agent.state & CHK_ST_ENABLED))
986 break;
987
988 /* The agent supports strings made of a single line ended by the
989 * first CR ('\r') or LF ('\n'). This line is composed of words
990 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
991 * line may optionally contained a description of a state change
992 * after a sharp ('#'), which is only considered if a health state
993 * is announced.
994 *
995 * Words may be composed of :
996 * - a numeric weight suffixed by the percent character ('%').
997 * - a health status among "up", "down", "stopped", and "fail".
998 * - an admin status among "ready", "drain", "maint".
999 *
1000 * These words may appear in any order. If multiple words of the
1001 * same category appear, the last one wins.
1002 */
1003
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001004 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001005 while (*p && *p != '\n' && *p != '\r')
1006 p++;
1007
1008 if (!*p) {
1009 if (!done)
1010 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001011
Willy Tarreau9809b782013-12-11 21:40:11 +01001012 /* at least inform the admin that the agent is mis-behaving */
1013 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1014 break;
1015 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001016
Willy Tarreau9809b782013-12-11 21:40:11 +01001017 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001018 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001019
Willy Tarreau81f5d942013-12-09 20:51:51 +01001020 while (*cmd) {
1021 /* look for next word */
1022 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1023 cmd++;
1024 continue;
1025 }
Simon Horman671b6f02013-11-25 10:46:39 +09001026
Willy Tarreau81f5d942013-12-09 20:51:51 +01001027 if (*cmd == '#') {
1028 /* this is the beginning of a health status description,
1029 * skip the sharp and blanks.
1030 */
1031 cmd++;
1032 while (*cmd == '\t' || *cmd == ' ')
1033 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001034 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001035 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001036
1037 /* find the end of the word so that we have a null-terminated
1038 * word between <cmd> and <p>.
1039 */
1040 p = cmd + 1;
1041 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1042 p++;
1043 if (*p)
1044 *p++ = 0;
1045
1046 /* first, health statuses */
1047 if (strcasecmp(cmd, "up") == 0) {
1048 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001049 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001050 hs = cmd;
1051 }
1052 else if (strcasecmp(cmd, "down") == 0) {
1053 check->health = 0;
1054 status = HCHK_STATUS_L7STS;
1055 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001056 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001057 else if (strcasecmp(cmd, "stopped") == 0) {
1058 check->health = 0;
1059 status = HCHK_STATUS_L7STS;
1060 hs = cmd;
1061 }
1062 else if (strcasecmp(cmd, "fail") == 0) {
1063 check->health = 0;
1064 status = HCHK_STATUS_L7STS;
1065 hs = cmd;
1066 }
1067 /* admin statuses */
1068 else if (strcasecmp(cmd, "ready") == 0) {
1069 as = cmd;
1070 }
1071 else if (strcasecmp(cmd, "drain") == 0) {
1072 as = cmd;
1073 }
1074 else if (strcasecmp(cmd, "maint") == 0) {
1075 as = cmd;
1076 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001077 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001078 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1079 ps = cmd;
1080 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001081 /* try to parse a maxconn here */
1082 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1083 cs = cmd;
1084 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001085 else {
1086 /* keep a copy of the first error */
1087 if (!err)
1088 err = cmd;
1089 }
1090 /* skip to next word */
1091 cmd = p;
1092 }
1093 /* here, cmd points either to \0 or to the beginning of a
1094 * description. Skip possible leading spaces.
1095 */
1096 while (*cmd == ' ' || *cmd == '\n')
1097 cmd++;
1098
1099 /* First, update the admin status so that we avoid sending other
1100 * possibly useless warnings and can also update the health if
1101 * present after going back up.
1102 */
1103 if (as) {
1104 if (strcasecmp(as, "drain") == 0)
1105 srv_adm_set_drain(check->server);
1106 else if (strcasecmp(as, "maint") == 0)
1107 srv_adm_set_maint(check->server);
1108 else
1109 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001110 }
1111
Willy Tarreau81f5d942013-12-09 20:51:51 +01001112 /* now change weights */
1113 if (ps) {
1114 const char *msg;
1115
1116 msg = server_parse_weight_change_request(s, ps);
1117 if (!wrn || !*wrn)
1118 wrn = msg;
1119 }
1120
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001121 if (cs) {
1122 const char *msg;
1123
1124 cs += strlen("maxconn:");
1125
1126 msg = server_parse_maxconn_change_request(s, cs);
1127 if (!wrn || !*wrn)
1128 wrn = msg;
1129 }
1130
Willy Tarreau81f5d942013-12-09 20:51:51 +01001131 /* and finally health status */
1132 if (hs) {
1133 /* We'll report some of the warnings and errors we have
1134 * here. Down reports are critical, we leave them untouched.
1135 * Lack of report, or report of 'UP' leaves the room for
1136 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001137 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001138 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001139 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001140
1141 if (!*msg || status == HCHK_STATUS_L7OKD) {
1142 if (err && *err)
1143 msg = err;
1144 else if (wrn && *wrn)
1145 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001146 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001147
1148 t = get_trash_chunk();
1149 chunk_printf(t, "via agent : %s%s%s%s",
1150 hs, *msg ? " (" : "",
1151 msg, *msg ? ")" : "");
1152
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001153 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001154 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001155 else if (err && *err) {
1156 /* No status change but we'd like to report something odd.
1157 * Just report the current state and copy the message.
1158 */
1159 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001160 set_server_check_status(check, status/*check->status*/,
1161 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001162
Willy Tarreau81f5d942013-12-09 20:51:51 +01001163 }
1164 else if (wrn && *wrn) {
1165 /* No status change but we'd like to report something odd.
1166 * Just report the current state and copy the message.
1167 */
1168 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001169 set_server_check_status(check, status/*check->status*/,
1170 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001171 }
1172 else
1173 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001174 break;
1175 }
1176
Willy Tarreau1620ec32011-08-06 17:05:02 +02001177 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001178 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001179 goto wait_more_data;
1180
Willy Tarreau5488a622019-08-06 16:26:31 +02001181 /* do not reset when closing, servers don't like this */
1182 if (conn_ctrl_ready(cs->conn))
1183 fdtab[cs->conn->handle.fd].linger_risk = 0;
1184
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001185 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001186 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001187 }
1188 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001189 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1190 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001191 else
1192 desc = "PostgreSQL unknown error";
1193
Simon Horman4a741432013-02-23 15:35:38 +09001194 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001195 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001196 break;
1197
1198 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001199 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001200 goto wait_more_data;
1201
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001202 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001203 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001204 }
1205 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001206 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001207 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001208 break;
1209
1210 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001211 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001212 goto wait_more_data;
1213
Willy Tarreau5488a622019-08-06 16:26:31 +02001214 /* do not reset when closing, servers don't like this */
1215 if (conn_ctrl_ready(cs->conn))
1216 fdtab[cs->conn->handle.fd].linger_risk = 0;
1217
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001218 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001219 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001220 /* We set the MySQL Version in description for information purpose
1221 * FIXME : it can be cool to use MySQL Version for other purpose,
1222 * like mark as down old MySQL server.
1223 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001224 if (b_data(&check->bi) > 51) {
1225 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001226 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001227 }
1228 else {
1229 if (!done)
1230 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001231
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001232 /* it seems we have a OK packet but without a valid length,
1233 * it must be a protocol error
1234 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001235 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001236 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001237 }
1238 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001240 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001241 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001242 }
1243 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001244 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1245 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1246 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001247
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001248 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001250 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001251 /* We have only one MySQL packet and it is a Handshake Initialization packet
1252 * but we need to have a second packet to know if it is alright
1253 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001254 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001255 goto wait_more_data;
1256 }
1257 else {
1258 /* We have only one packet and it is an Error packet,
1259 * an error message is attached, so we can display it
1260 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001261 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001262 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001263 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001264 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001265 } else if (b_data(&check->bi) > first_packet_len + 4) {
1266 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1267 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1268 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001269
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001270 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001271 /* We have 2 packets and that's good */
1272 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001273 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001274 /* No error packet */
1275 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001276 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001277 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001278 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001279 }
1280 else {
1281 /* An error message is attached in the Error packet
1282 * so we can display it ! :)
1283 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001284 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001285 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001286 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001287 }
1288 }
1289 }
1290 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001291 if (!done)
1292 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001293
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001294 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001295 * it must be a protocol error
1296 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001297 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001298 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001299 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001300 }
1301 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001302 break;
1303
1304 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001305 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001306 goto wait_more_data;
1307
1308 /* Check if the server speaks LDAP (ASN.1/BER)
1309 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1310 * http://tools.ietf.org/html/rfc4511
1311 */
1312
1313 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1314 * LDAPMessage: 0x30: SEQUENCE
1315 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001316 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001317 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001318 }
1319 else {
1320 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001321 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001322
1323 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1324 * messageID: 0x02 0x01 0x01: INTEGER 1
1325 * protocolOp: 0x61: bindResponse
1326 */
1327 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001328 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001329 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001330 goto out_wakeup;
1331 }
1332
1333 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001334 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001335
1336 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1337 * ldapResult: 0x0a 0x01: ENUMERATION
1338 */
1339 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001340 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001341 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001342 goto out_wakeup;
1343 }
1344
1345 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1346 * resultCode
1347 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001348 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001349 if (check->code) {
1350 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 +02001351 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001352 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001353 }
1354 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001355 break;
1356
Christopher Fauletba7bc162016-11-07 21:07:38 +01001357 case PR_O2_SPOP_CHK: {
1358 unsigned int framesz;
1359 char err[HCHK_DESC_LEN];
1360
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001361 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001362 goto wait_more_data;
1363
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001364 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001365 framesz = ntohl(framesz);
1366
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001367 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001368 goto wait_more_data;
1369
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001370 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001371 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1372 else
1373 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1374 break;
1375 }
1376
Willy Tarreau1620ec32011-08-06 17:05:02 +02001377 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001378 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001379 break;
1380 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001381
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001382 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001383 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001384 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001385 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001386
Nick Chalk57b1bf72010-03-16 15:50:46 +00001387 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001388 *b_head(&check->bi) = '\0';
1389 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001390
Steven Davidovitz544d4812017-03-08 11:06:20 -08001391 /* Close the connection... We still attempt to nicely close if,
1392 * for instance, SSL needs to send a "close notify." Later, we perform
1393 * a hard close and reset the connection if some data are pending,
1394 * otherwise we end up with many TIME_WAITs and eat all the source port
1395 * range quickly. To avoid sending RSTs all the time, we first try to
1396 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001397 */
Olivier Houchard6c7e96a2019-07-02 16:35:18 +02001398 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1399 * connection, to make sure cs_shutw() will not lead to a shutdown()
1400 * that would provoke TIME_WAITs.
1401 */
1402 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001403 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001404
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001405 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001406 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001407 conn->flags |= CO_FL_ERROR;
1408
Willy Tarreaufdccded2008-08-29 18:19:04 +02001409 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001410out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001411 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001412
1413 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001414 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001415 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001416}
1417
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001418/*
1419 * This function is used only for server health-checks. It handles connection
1420 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001421 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1422 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001423 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001424static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001425{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001426 struct connection *conn = cs->conn;
1427 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001428 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001429 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001430
Olivier Houchard0923fa42019-01-11 18:43:04 +01001431 if (check->server)
1432 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1433 else
1434 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001435
Willy Tarreauc09572f2017-10-04 11:58:22 +02001436 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001437 if (check->type == PR_O2_TCPCHK_CHK) {
1438 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001439 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001440 conn = cs->conn;
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001441 } else if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001442 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001443
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001444 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001445 /* We may get error reports bypassing the I/O handlers, typically
1446 * the case when sending a pure TCP check which fails, then the I/O
1447 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001448 * main processing task so let's simply wake it up. If we get here,
1449 * we expect errno to still be valid.
1450 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001451 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001452 task_wakeup(check->task, TASK_WOKEN_IO);
1453 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001454 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001455 /* we may get here if only a connection probe was required : we
1456 * don't have any data to send nor anything expected in response,
1457 * so the completion of the connection establishment is enough.
1458 */
1459 task_wakeup(check->task, TASK_WOKEN_IO);
1460 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001461
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001462 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001463 /* Check complete or aborted. If connection not yet closed do it
1464 * now and wake the check task up to be sure the result is
1465 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001466 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001467 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001468 ret = -1;
Olivier Houchardf4949572019-07-02 17:42:22 +02001469 /* We may have been scheduled to run, and the
1470 * I/O handler expects to have a cs, so remove
1471 * the tasklet
1472 */
1473 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001474 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001475 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001476
Olivier Houchard0923fa42019-01-11 18:43:04 +01001477 if (check->server)
1478 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1479 else
1480 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001481
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001482 /* if a connection got replaced, we must absolutely prevent the connection
1483 * handler from touching its fd, and perform the FD polling updates ourselves
1484 */
1485 if (ret < 0)
1486 conn_cond_update_polling(conn);
1487
1488 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001489}
1490
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001491struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001492 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001493 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001494};
1495
Willy Tarreaubaaee002006-06-26 02:48:02 +02001496/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001497 * updates the server's weight during a warmup stage. Once the final weight is
1498 * reached, the task automatically stops. Note that any server status change
1499 * must have updated s->last_change accordingly.
1500 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001501static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001502{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001503 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001504
1505 /* by default, plan on stopping the task */
1506 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001507 if ((s->next_admin & SRV_ADMF_MAINT) ||
1508 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001509 return t;
1510
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001511 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1512
Willy Tarreau892337c2014-05-13 23:41:20 +02001513 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001514 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001515
1516 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001517 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001518
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001519 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1520
Willy Tarreau2e993902011-10-31 11:53:20 +01001521 /* get back there in 1 second or 1/20th of the slowstart interval,
1522 * whichever is greater, resulting in small 5% steps.
1523 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001524 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001525 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1526 return t;
1527}
1528
Willy Tarreau894c6422017-10-04 15:58:52 +02001529/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1530 * none was found.
1531 */
1532static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1533{
1534 struct tcpcheck_rule *r;
1535
1536 list_for_each_entry(r, list, list) {
1537 if (r->action != TCPCHK_ACT_COMMENT)
1538 return r;
1539 }
1540 return NULL;
1541}
1542
Willy Tarreau2e993902011-10-31 11:53:20 +01001543/*
Simon Horman98637e52014-06-20 12:30:16 +09001544 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001545 *
1546 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001547 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1548 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1549 * - SF_ERR_SRVTO if there are no more servers
1550 * - SF_ERR_SRVCL if the connection was refused by the server
1551 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1552 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1553 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001554 * - 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 +01001555 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001556 * Note that we try to prevent the network stack from sending the ACK during the
1557 * connect() when a pure TCP check is used (without PROXY protocol).
1558 */
Simon Horman98637e52014-06-20 12:30:16 +09001559static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001560{
1561 struct check *check = t->context;
1562 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001563 struct conn_stream *cs = check->cs;
1564 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001565 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001566 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001567 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001568 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001569
Willy Tarreau00149122017-10-04 18:05:01 +02001570 /* we cannot have a connection here */
1571 if (conn)
1572 return SF_ERR_INTERNAL;
1573
Simon Hormanb00d17a2014-06-13 16:18:16 +09001574 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001575 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001576 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001577 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1578 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001579
1580 /* prepare the check buffer.
1581 * This should not be used if check is the secondary agent check
1582 * of a server as s->proxy->check_req will relate to the
1583 * configuration of the primary check. Similarly, tcp-check uses
1584 * its own strings.
1585 */
1586 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001587 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001588
1589 /* we want to check if this host replies to HTTP or SSLv3 requests
1590 * so we'll send the request, and won't wake the checker up now.
1591 */
1592 if ((check->type) == PR_O2_SSL3_CHK) {
1593 /* SSL requires that we put Unix time in the request */
1594 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001595 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001596 }
1597 else if ((check->type) == PR_O2_HTTP_CHK) {
1598 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001599 b_putblk(&check->bo, trash.area,
1600 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001601 /* prevent HTTP keep-alive when "http-check expect" is used */
1602 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001603 b_putist(&check->bo, ist("Connection: close\r\n"));
1604 b_putist(&check->bo, ist("\r\n"));
1605 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001606 }
1607 }
1608
James Brown55f9ff12015-10-21 18:19:05 -07001609 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001610 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001611 }
1612
Willy Tarreauf411cce2017-10-04 16:21:19 +02001613 /* for tcp-checks, the initial connection setup is handled separately as
1614 * it may be sent to a specific port and not to the server's.
1615 */
1616 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1617 tcpcheck_main(check);
1618 return SF_ERR_UP;
1619 }
1620
Simon Hormanb00d17a2014-06-13 16:18:16 +09001621 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001622 cs = check->cs = cs_new(NULL);
1623 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001624 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001625 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001626 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001627 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001628
Willy Tarreauca79f592019-07-17 19:04:47 +02001629 if (!sockaddr_alloc(&conn->dst))
1630 return SF_ERR_RESOURCE;
1631
Simon Horman41f58762015-01-30 11:22:56 +09001632 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001633 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001634 *conn->dst = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001635 }
1636 else {
1637 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001638 *conn->dst = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001639 }
1640
Alexander Liu2a54bb72019-05-22 19:44:48 +08001641 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1642 conn->send_proxy_ofs = 1;
1643 conn->flags |= CO_FL_SOCKS4;
1644 }
1645
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001646 proto = protocol_by_family(conn->dst->ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001647 conn->target = &s->obj_type;
1648
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001649 if ((conn->dst->ss_family == AF_INET) || (conn->dst->ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001650 int i = 0;
1651
1652 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001653 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001654 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001655
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001656 set_host_port(conn->dst, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001657 }
1658
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001659 /* no client address */
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001660
Willy Tarreaube373152018-09-06 11:45:30 +02001661 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001662 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1663 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001664 cs_attach(cs, check, &check_conn_cb);
1665
Willy Tarreauf3d34822014-12-08 12:11:28 +01001666 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001667 if (check->type != 0)
1668 connflags |= CONNECT_HAS_DATA;
1669 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1670 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1671 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001672
Willy Tarreaue7dff022015-04-03 01:14:29 +02001673 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001674 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001675 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001676
Willy Tarreau16257f62017-11-02 15:45:00 +01001677
Olivier Houchard9130a962017-10-17 17:33:43 +02001678#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001679 if (ret == SF_ERR_NONE) {
1680 if (s->check.sni)
1681 ssl_sock_set_servername(conn, s->check.sni);
1682 if (s->check.alpn_str)
1683 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1684 s->check.alpn_len);
1685 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001686#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001687 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001688 conn->send_proxy_ofs = 1;
1689 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001690 if (xprt_add_hs(conn) < 0)
1691 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001692 }
1693
1694 return ret;
1695}
1696
Simon Horman98637e52014-06-20 12:30:16 +09001697static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001698static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001699__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001700
1701void block_sigchld(void)
1702{
1703 sigset_t set;
1704 sigemptyset(&set);
1705 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001706 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001707}
1708
1709void unblock_sigchld(void)
1710{
1711 sigset_t set;
1712 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001713 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001714 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001715}
1716
Simon Horman98637e52014-06-20 12:30:16 +09001717static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1718{
1719 struct pid_list *elem;
1720 struct check *check = t->context;
1721
Willy Tarreaubafbe012017-11-24 17:34:44 +01001722 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001723 if (!elem)
1724 return NULL;
1725 elem->pid = pid;
1726 elem->t = t;
1727 elem->exited = 0;
1728 check->curpid = elem;
1729 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001730
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001731 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001732 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001733 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001734
Simon Horman98637e52014-06-20 12:30:16 +09001735 return elem;
1736}
1737
Simon Horman98637e52014-06-20 12:30:16 +09001738static void pid_list_del(struct pid_list *elem)
1739{
1740 struct check *check;
1741
1742 if (!elem)
1743 return;
1744
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_DEL(&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 if (!elem->exited)
1750 kill(elem->pid, SIGTERM);
1751
1752 check = elem->t->context;
1753 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001754 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001755}
1756
1757/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1758static void pid_list_expire(pid_t pid, int status)
1759{
1760 struct pid_list *elem;
1761
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001762 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001763 list_for_each_entry(elem, &pid_list, list) {
1764 if (elem->pid == pid) {
1765 elem->t->expire = now_ms;
1766 elem->status = status;
1767 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001768 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001769 break;
Simon Horman98637e52014-06-20 12:30:16 +09001770 }
1771 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001772 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001773}
1774
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001775static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001776{
1777 pid_t pid;
1778 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001779
Simon Horman98637e52014-06-20 12:30:16 +09001780 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1781 pid_list_expire(pid, status);
1782}
1783
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001784static int init_pid_list(void)
1785{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001786 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001787 /* Nothing to do */
1788 return 0;
1789
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001790 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001791 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1792 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001793 return 1;
1794 }
1795
Willy Tarreaubafbe012017-11-24 17:34:44 +01001796 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1797 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001798 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1799 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001800 return 1;
1801 }
1802
1803 return 0;
1804}
1805
Cyril Bontéac92a062014-12-27 22:28:38 +01001806/* helper macro to set an environment variable and jump to a specific label on failure. */
1807#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001808
1809/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001810 * helper function to allocate enough memory to store an environment variable.
1811 * It will also check that the environment variable is updatable, and silently
1812 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001813 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001814static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001815{
1816 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001817 char *envname;
1818 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001819
Cyril Bontéac92a062014-12-27 22:28:38 +01001820 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001821 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001822 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001823 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001824
1825 envname = extcheck_envs[idx].name;
1826 vmaxlen = extcheck_envs[idx].vmaxlen;
1827
1828 /* Check if the environment variable is already set, and silently reject
1829 * the update if this one is not updatable. */
1830 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1831 return 0;
1832
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001833 /* Instead of sending NOT_USED, sending an empty value is preferable */
1834 if (strcmp(value, "NOT_USED") == 0) {
1835 value = "";
1836 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001837
1838 len = strlen(envname) + 1;
1839 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1840 len += strlen(value);
1841 else
1842 len += vmaxlen;
1843
1844 if (!check->envp[idx])
1845 check->envp[idx] = malloc(len + 1);
1846
1847 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001848 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001849 return 1;
1850 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001851 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001852 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001853 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001854 return 1;
1855 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001856 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001857 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001858 return 1;
1859 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001860 return 0;
1861}
Simon Horman98637e52014-06-20 12:30:16 +09001862
1863static int prepare_external_check(struct check *check)
1864{
1865 struct server *s = check->server;
1866 struct proxy *px = s->proxy;
1867 struct listener *listener = NULL, *l;
1868 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001869 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001870 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001871
1872 list_for_each_entry(l, &px->conf.listeners, by_fe)
1873 /* Use the first INET, INET6 or UNIX listener */
1874 if (l->addr.ss_family == AF_INET ||
1875 l->addr.ss_family == AF_INET6 ||
1876 l->addr.ss_family == AF_UNIX) {
1877 listener = l;
1878 break;
1879 }
1880
Simon Horman98637e52014-06-20 12:30:16 +09001881 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001882 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1883 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001884 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001885 goto err;
1886 }
Simon Horman98637e52014-06-20 12:30:16 +09001887
Cyril Bontéac92a062014-12-27 22:28:38 +01001888 check->argv = calloc(6, sizeof(char *));
1889 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001890 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001891 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001892 }
Simon Horman98637e52014-06-20 12:30:16 +09001893
1894 check->argv[0] = px->check_command;
1895
Cyril Bonté777be862014-12-02 21:21:35 +01001896 if (!listener) {
1897 check->argv[1] = strdup("NOT_USED");
1898 check->argv[2] = strdup("NOT_USED");
1899 }
1900 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001901 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001902 addr_to_str(&listener->addr, buf, sizeof(buf));
1903 check->argv[1] = strdup(buf);
1904 port_to_str(&listener->addr, buf, sizeof(buf));
1905 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001906 }
1907 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001908 const struct sockaddr_un *un;
1909
1910 un = (struct sockaddr_un *)&listener->addr;
1911 check->argv[1] = strdup(un->sun_path);
1912 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001913 }
1914 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001915 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001916 goto err;
1917 }
1918
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001919 addr_to_str(&s->addr, buf, sizeof(buf));
1920 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001921
1922 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1923 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1924 else
1925 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001926 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001927
Cyril Bontéac92a062014-12-27 22:28:38 +01001928 for (i = 0; i < 5; i++) {
1929 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001930 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001931 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001932 }
1933 }
Simon Horman98637e52014-06-20 12:30:16 +09001934
Cyril Bontéac92a062014-12-27 22:28:38 +01001935 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001936 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001937 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1938 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1939 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1940 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001941 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001942 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1943 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1944 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1945 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1946 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1947 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1948
1949 /* Ensure that we don't leave any hole in check->envp */
1950 for (i = 0; i < EXTCHK_SIZE; i++)
1951 if (!check->envp[i])
1952 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001953
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001954 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001955err:
1956 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001957 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001958 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001959 free(check->envp);
1960 check->envp = NULL;
1961 }
1962
1963 if (check->argv) {
1964 for (i = 1; i < 5; i++)
1965 free(check->argv[i]);
1966 free(check->argv);
1967 check->argv = NULL;
1968 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001969 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001970}
1971
Simon Hormanb00d17a2014-06-13 16:18:16 +09001972/*
Simon Horman98637e52014-06-20 12:30:16 +09001973 * establish a server health-check that makes use of a process.
1974 *
1975 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001976 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001977 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001978 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001979 *
1980 * Blocks and then unblocks SIGCHLD
1981 */
1982static int connect_proc_chk(struct task *t)
1983{
Cyril Bontéac92a062014-12-27 22:28:38 +01001984 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001985 struct check *check = t->context;
1986 struct server *s = check->server;
1987 struct proxy *px = s->proxy;
1988 int status;
1989 pid_t pid;
1990
Willy Tarreaue7dff022015-04-03 01:14:29 +02001991 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001992
1993 block_sigchld();
1994
1995 pid = fork();
1996 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001997 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1998 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001999 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2000 goto out;
2001 }
2002 if (pid == 0) {
2003 /* Child */
2004 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002005 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002006 int fd;
2007
2008 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2009 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2010
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002011 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002012
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002013 /* restore the initial FD limits */
2014 limit.rlim_cur = rlim_fd_cur_at_boot;
2015 limit.rlim_max = rlim_fd_max_at_boot;
2016 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2017 getrlimit(RLIMIT_NOFILE, &limit);
2018 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2019 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2020 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2021 }
2022
Simon Horman98637e52014-06-20 12:30:16 +09002023 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01002024 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Willy Tarreau2df8cad2019-07-01 07:51:29 +02002025 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002026 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002027 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2028 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002029 exit(-1);
2030 }
2031
2032 /* Parent */
2033 if (check->result == CHK_RES_UNKNOWN) {
2034 if (pid_list_add(pid, t) != NULL) {
2035 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2036
2037 if (px->timeout.check && px->timeout.connect) {
2038 int t_con = tick_add(now_ms, px->timeout.connect);
2039 t->expire = tick_first(t->expire, t_con);
2040 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002041 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002042 goto out;
2043 }
2044 else {
2045 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2046 }
2047 kill(pid, SIGTERM); /* process creation error */
2048 }
2049 else
2050 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2051
2052out:
2053 unblock_sigchld();
2054 return status;
2055}
2056
2057/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002058 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002059 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002060 *
2061 * Please do NOT place any return statement in this function and only leave
2062 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002063 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002064static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002065{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002066 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002067 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002068 int rv;
2069 int ret;
2070 int expired = tick_is_expired(t->expire, now_ms);
2071
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002072 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002073 if (!(check->state & CHK_ST_INPROGRESS)) {
2074 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002075 if (!expired) /* woke up too early */
2076 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002077
2078 /* we don't send any health-checks when the proxy is
2079 * stopped, the server should not be checked or the check
2080 * is disabled.
2081 */
2082 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2083 s->proxy->state == PR_STSTOPPED)
2084 goto reschedule;
2085
2086 /* we'll initiate a new check */
2087 set_server_check_status(check, HCHK_STATUS_START, NULL);
2088
2089 check->state |= CHK_ST_INPROGRESS;
2090
Simon Hormandbf70192015-01-30 11:22:53 +09002091 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002092 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002093 /* the process was forked, we allow up to min(inter,
2094 * timeout.connect) for it to report its status, but
2095 * only when timeout.check is set as it may be to short
2096 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002097 */
2098 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2099
2100 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2101 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2102 t->expire = tick_first(t->expire, t_con);
2103 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002104 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002105 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002106 }
2107
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002108 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002109
2110 check->state &= ~CHK_ST_INPROGRESS;
2111 check_notify_failure(check);
2112
2113 /* we allow up to min(inter, timeout.connect) for a connection
2114 * to establish but only when timeout.check is set
2115 * as it may be to short for a full check otherwise
2116 */
2117 while (tick_is_expired(t->expire, now_ms)) {
2118 int t_con;
2119
2120 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2121 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2122
2123 if (s->proxy->timeout.check)
2124 t->expire = tick_first(t->expire, t_con);
2125 }
2126 }
2127 else {
2128 /* there was a test running.
2129 * First, let's check whether there was an uncaught error,
2130 * which can happen on connect timeout or error.
2131 */
2132 if (check->result == CHK_RES_UNKNOWN) {
2133 /* good connection is enough for pure TCP check */
2134 struct pid_list *elem = check->curpid;
2135 int status = HCHK_STATUS_UNKNOWN;
2136
2137 if (elem->exited) {
2138 status = elem->status; /* Save in case the process exits between use below */
2139 if (!WIFEXITED(status))
2140 check->code = -1;
2141 else
2142 check->code = WEXITSTATUS(status);
2143 if (!WIFEXITED(status) || WEXITSTATUS(status))
2144 status = HCHK_STATUS_PROCERR;
2145 else
2146 status = HCHK_STATUS_PROCOK;
2147 } else if (expired) {
2148 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002149 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002150 kill(elem->pid, SIGTERM);
2151 }
2152 set_server_check_status(check, status, NULL);
2153 }
2154
2155 if (check->result == CHK_RES_FAILED) {
2156 /* a failure or timeout detected */
2157 check_notify_failure(check);
2158 }
2159 else if (check->result == CHK_RES_CONDPASS) {
2160 /* check is OK but asks for stopping mode */
2161 check_notify_stopping(check);
2162 }
2163 else if (check->result == CHK_RES_PASSED) {
2164 /* a success was detected */
2165 check_notify_success(check);
2166 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002167 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002168 check->state &= ~CHK_ST_INPROGRESS;
2169
2170 pid_list_del(check->curpid);
2171
2172 rv = 0;
2173 if (global.spread_checks > 0) {
2174 rv = srv_getinter(check) * global.spread_checks / 100;
2175 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2176 }
2177 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2178 }
2179
2180 reschedule:
2181 while (tick_is_expired(t->expire, now_ms))
2182 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002183
2184 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002185 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002186 return t;
2187}
2188
2189/*
2190 * manages a server health-check that uses a connection. Returns
2191 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002192 *
2193 * Please do NOT place any return statement in this function and only leave
2194 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002195 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002196static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002197{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002198 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002199 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002200 struct conn_stream *cs = check->cs;
2201 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002202 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002203 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002204 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002205
Olivier Houchard0923fa42019-01-11 18:43:04 +01002206 if (check->server)
2207 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002208 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002209 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002210 if (!expired) /* woke up too early */
2211 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002212
Simon Horman671b6f02013-11-25 10:46:39 +09002213 /* we don't send any health-checks when the proxy is
2214 * stopped, the server should not be checked or the check
2215 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002216 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002217 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002218 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002219 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002220
2221 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002222 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002223
Willy Tarreau2c115e52013-12-11 19:41:16 +01002224 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002225 b_reset(&check->bi);
2226 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002227
Simon Hormandbf70192015-01-30 11:22:53 +09002228 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002229 cs = check->cs;
2230 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002231
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002232 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002233 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002234 goto out_unlock;
2235
Willy Tarreaue7dff022015-04-03 01:14:29 +02002236 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002237 /* we allow up to min(inter, timeout.connect) for a connection
2238 * to establish but only when timeout.check is set
2239 * as it may be to short for a full check otherwise
2240 */
Simon Horman4a741432013-02-23 15:35:38 +09002241 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002242
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002243 if (proxy->timeout.check && proxy->timeout.connect) {
2244 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002245 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002246 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002247
Olivier Houchard53216e72018-10-10 15:46:36 +02002248 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002249 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002250
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002251 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002252 goto reschedule;
2253
Willy Tarreaue7dff022015-04-03 01:14:29 +02002254 case SF_ERR_SRVTO: /* ETIMEDOUT */
2255 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002256 if (conn)
2257 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002258 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002259 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002260 /* should share same code than cases below */
2261 case SF_ERR_CHK_PORT:
2262 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002263 case SF_ERR_PRXCOND:
2264 case SF_ERR_RESOURCE:
2265 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002266 if (conn)
2267 conn->flags |= CO_FL_ERROR;
2268 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002269 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002270 }
2271
Willy Tarreau5a78f362012-11-23 12:47:05 +01002272 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002273 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002274 if (check->wait_list.events)
2275 cs->conn->xprt->unsubscribe(cs->conn,
2276 cs->conn->xprt_ctx,
2277 check->wait_list.events,
2278 &check->wait_list);
2279 /* We may have been scheduled to run, and the
2280 * I/O handler expects to have a cs, so remove
2281 * the tasklet
2282 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002283 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002284 cs_destroy(cs);
2285 cs = check->cs = NULL;
2286 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002287 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002288
Willy Tarreau2c115e52013-12-11 19:41:16 +01002289 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002290 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002291
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002292 /* we allow up to min(inter, timeout.connect) for a connection
2293 * to establish but only when timeout.check is set
2294 * as it may be to short for a full check otherwise
2295 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002296 while (tick_is_expired(t->expire, now_ms)) {
2297 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002298
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002299 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002300 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002301
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002302 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002303 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002304 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002305 }
2306 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002307 /* there was a test running.
2308 * First, let's check whether there was an uncaught error,
2309 * which can happen on connect timeout or error.
2310 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002311 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002312 /* good connection is enough for pure TCP check */
2313 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002314 if (check->use_ssl)
2315 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002316 else
Simon Horman4a741432013-02-23 15:35:38 +09002317 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002318 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002319 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002320 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002321 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002322 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002323 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002324 }
2325
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002326 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002327 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002328 /* The check was aborted and the connection was not yet closed.
2329 * This can happen upon timeout, or when an external event such
2330 * as a failed response coupled with "observe layer7" caused the
2331 * server state to be suddenly changed.
2332 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002333 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002334 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002335 }
2336
Willy Tarreauac59f362017-10-08 11:10:19 +02002337 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002338 if (check->wait_list.events)
2339 cs->conn->xprt->unsubscribe(cs->conn,
2340 cs->conn->xprt_ctx,
2341 check->wait_list.events,
2342 &check->wait_list);
2343 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002344 * I/O handler expects to have a cs, so remove
2345 * the tasklet
2346 */
2347 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002348 cs_destroy(cs);
2349 cs = check->cs = NULL;
2350 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002351 }
2352
Olivier Houchard0923fa42019-01-11 18:43:04 +01002353 if (check->server) {
2354 if (check->result == CHK_RES_FAILED) {
2355 /* a failure or timeout detected */
2356 check_notify_failure(check);
2357 }
2358 else if (check->result == CHK_RES_CONDPASS) {
2359 /* check is OK but asks for stopping mode */
2360 check_notify_stopping(check);
2361 }
2362 else if (check->result == CHK_RES_PASSED) {
2363 /* a success was detected */
2364 check_notify_success(check);
2365 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002366 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002367 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002368 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002369
Olivier Houchard0923fa42019-01-11 18:43:04 +01002370 if (check->server) {
2371 rv = 0;
2372 if (global.spread_checks > 0) {
2373 rv = srv_getinter(check) * global.spread_checks / 100;
2374 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2375 }
2376 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002377 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002378 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002379
2380 reschedule:
2381 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002382 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002383 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002384 if (check->server)
2385 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002386 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002387}
2388
Simon Horman98637e52014-06-20 12:30:16 +09002389/*
2390 * manages a server health-check. Returns
2391 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2392 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002393static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002394{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002395 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002396
2397 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002398 return process_chk_proc(t, context, state);
2399 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002400
Simon Horman98637e52014-06-20 12:30:16 +09002401}
2402
Simon Horman5c942422013-11-25 10:46:32 +09002403static int start_check_task(struct check *check, int mininter,
2404 int nbcheck, int srvpos)
2405{
2406 struct task *t;
2407 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002408 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002409 ha_alert("Starting [%s:%s] check: out of memory.\n",
2410 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002411 return 0;
2412 }
2413
2414 check->task = t;
2415 t->process = process_chk;
2416 t->context = check;
2417
Willy Tarreau1746eec2014-04-25 10:46:47 +02002418 if (mininter < srv_getinter(check))
2419 mininter = srv_getinter(check);
2420
2421 if (global.max_spread_checks && mininter > global.max_spread_checks)
2422 mininter = global.max_spread_checks;
2423
Simon Horman5c942422013-11-25 10:46:32 +09002424 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002425 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002426 check->start = now;
2427 task_queue(t);
2428
2429 return 1;
2430}
2431
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002432/*
2433 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002434 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002435 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002436static int start_checks()
2437{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002438
2439 struct proxy *px;
2440 struct server *s;
2441 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002442 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002443
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002444 /* 1- count the checkers to run simultaneously.
2445 * We also determine the minimum interval among all of those which
2446 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2447 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002448 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002449 * too short an interval for all others.
2450 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002451 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002452 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002453 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002454 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002455 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002456 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002457 }
2458 /* We need a warmup task that will be called when the server
2459 * state switches from down to up.
2460 */
2461 s->warmup = t;
2462 t->process = server_warmup;
2463 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002464 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002465 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002466 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 +01002467 }
2468
Willy Tarreaud8514a22013-12-11 21:10:14 +01002469 if (s->check.state & CHK_ST_CONFIGURED) {
2470 nbcheck++;
2471 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2472 (!mininter || mininter > srv_getinter(&s->check)))
2473 mininter = srv_getinter(&s->check);
2474 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002475
Willy Tarreaud8514a22013-12-11 21:10:14 +01002476 if (s->agent.state & CHK_ST_CONFIGURED) {
2477 nbcheck++;
2478 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2479 (!mininter || mininter > srv_getinter(&s->agent)))
2480 mininter = srv_getinter(&s->agent);
2481 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002482 }
2483 }
2484
Simon Horman4a741432013-02-23 15:35:38 +09002485 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002486 return 0;
2487
2488 srand((unsigned)time(NULL));
2489
2490 /*
2491 * 2- start them as far as possible from each others. For this, we will
2492 * start them after their interval set to the min interval divided by
2493 * the number of servers, weighted by the server's position in the list.
2494 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002495 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002496 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2497 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002498 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002499 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002500 }
2501 }
2502
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002503 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002504 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002505 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002506 if (s->check.type == PR_O2_EXT_CHK) {
2507 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002508 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002509 }
Simon Hormand60d6912013-11-25 10:46:36 +09002510 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002511 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002512 srvpos++;
2513 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002514
Simon Hormand60d6912013-11-25 10:46:36 +09002515 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002516 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002517 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002518 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002519 }
2520 srvpos++;
2521 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002522 }
2523 }
2524 return 0;
2525}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002526
2527/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002528 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002529 * The buffer MUST be terminated by a null byte before calling this function.
2530 * Sets server status appropriately. The caller is responsible for ensuring
2531 * that the buffer contains at least 13 characters. If <done> is zero, we may
2532 * return 0 to indicate that data is required to decide of a match.
2533 */
2534static int httpchk_expect(struct server *s, int done)
2535{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002536 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002537 char status_code[] = "000";
2538 char *contentptr;
2539 int crlf;
2540 int ret;
2541
2542 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2543 case PR_O2_EXP_STS:
2544 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002545 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2546 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002547
2548 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2549 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2550 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002551 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002552
2553 /* we necessarily have the response, so there are no partial failures */
2554 if (s->proxy->options2 & PR_O2_EXP_INV)
2555 ret = !ret;
2556
Simon Horman4a741432013-02-23 15:35:38 +09002557 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002558 break;
2559
2560 case PR_O2_EXP_STR:
2561 case PR_O2_EXP_RSTR:
2562 /* very simple response parser: ignore CR and only count consecutive LFs,
2563 * stop with contentptr pointing to first char after the double CRLF or
2564 * to '\0' if crlf < 2.
2565 */
2566 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002567 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002568 if (crlf >= 2)
2569 break;
2570 if (*contentptr == '\r')
2571 continue;
2572 else if (*contentptr == '\n')
2573 crlf++;
2574 else
2575 crlf = 0;
2576 }
2577
2578 /* Check that response contains a body... */
2579 if (crlf < 2) {
2580 if (!done)
2581 return 0;
2582
Simon Horman4a741432013-02-23 15:35:38 +09002583 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002584 "HTTP content check could not find a response body");
2585 return 1;
2586 }
2587
2588 /* Check that response body is not empty... */
2589 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002590 if (!done)
2591 return 0;
2592
Simon Horman4a741432013-02-23 15:35:38 +09002593 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002594 "HTTP content check found empty response body");
2595 return 1;
2596 }
2597
2598 /* Check the response content against the supplied string
2599 * or regex... */
2600 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2601 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2602 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002603 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002604
2605 /* if we don't match, we may need to wait more */
2606 if (!ret && !done)
2607 return 0;
2608
2609 if (ret) {
2610 /* content matched */
2611 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002612 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002613 "HTTP check matched unwanted content");
2614 else
Simon Horman4a741432013-02-23 15:35:38 +09002615 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002616 "HTTP content check matched");
2617 }
2618 else {
2619 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002620 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002621 "HTTP check did not match unwanted content");
2622 else
Simon Horman4a741432013-02-23 15:35:38 +09002623 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002624 "HTTP content check did not match");
2625 }
2626 break;
2627 }
2628 return 1;
2629}
2630
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002631/*
2632 * return the id of a step in a send/expect session
2633 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002634static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002635{
2636 struct tcpcheck_rule *cur = NULL, *next = NULL;
2637 int i = 0;
2638
Willy Tarreau213c6782014-10-02 14:51:02 +02002639 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002640 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002641 return 0;
2642
Simon Hormane16c1b32015-01-30 11:22:57 +09002643 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002644
2645 /* no step => first step */
2646 if (cur == NULL)
2647 return 1;
2648
2649 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002650 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002651 if (next->list.p == &cur->list)
2652 break;
2653 ++i;
2654 }
2655
2656 return i;
2657}
2658
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002659/*
2660 * return the latest known comment before (including) the given stepid
2661 * returns NULL if no comment found
2662 */
2663static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2664{
2665 struct tcpcheck_rule *cur = NULL;
2666 char *ret = NULL;
2667 int i = 0;
2668
2669 /* not even started anything yet, return latest comment found before any action */
2670 if (!check->current_step) {
2671 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2672 if (cur->action == TCPCHK_ACT_COMMENT)
2673 ret = cur->comment;
2674 else
2675 goto return_comment;
2676 }
2677 }
2678
2679 i = 1;
2680 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2681 if (cur->comment)
2682 ret = cur->comment;
2683
2684 if (i >= stepid)
2685 goto return_comment;
2686
2687 ++i;
2688 }
2689
2690 return_comment:
2691 return ret;
2692}
2693
Willy Tarreaube74b882017-10-04 16:22:49 +02002694/* proceed with next steps for the TCP checks <check>. Note that this is called
2695 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002696 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2697 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002698 *
2699 * Please do NOT place any return statement in this function and only leave
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002700 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002701 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002702static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002703{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002704 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002705 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002706 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002707 struct conn_stream *cs = check->cs;
2708 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002709 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002710 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002711 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002712 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002713 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002714
Willy Tarreauef953952014-10-02 14:30:14 +02002715 /* here, we know that the check is complete or that it failed */
2716 if (check->result != CHK_RES_UNKNOWN)
2717 goto out_end_tcpcheck;
2718
2719 /* We have 4 possibilities here :
2720 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002721 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002722 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2723 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002724 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002725 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002726 * be waiting for a connection attempt to complete. conn!=NULL.
2727 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002728 *
2729 * #2 and #3 are quite similar, we want both the connection and the
2730 * handshake to complete before going any further. Thus we must always
2731 * wait for a connection to complete unless we're before and existing
2732 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002733 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002734
2735 /* find first rule and skip comments */
2736 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2737 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2738 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2739
Willy Tarreau00149122017-10-04 18:05:01 +02002740 if ((check->current_step || &next->list == head) &&
2741 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002742 /* we allow up to min(inter, timeout.connect) for a connection
2743 * to establish but only when timeout.check is set
2744 * as it may be to short for a full check otherwise
2745 */
2746 while (tick_is_expired(t->expire, now_ms)) {
2747 int t_con;
2748
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002749 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002750 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2751
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002752 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002753 t->expire = tick_first(t->expire, t_con);
2754 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002755 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002756 }
2757
Willy Tarreauef953952014-10-02 14:30:14 +02002758 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002759 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002760 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002761 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002762 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002763
Willy Tarreau213c6782014-10-02 14:51:02 +02002764 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002765 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002766 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002767 b_reset(&check->bo);
2768 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002769 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002770 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002771 if (proxy->timeout.check)
2772 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002773 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002774
Willy Tarreauabca5b62013-12-06 14:19:25 +01002775 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002776 /* We have to try to flush the output buffer before reading, at
2777 * the end, or if we're about to send a string that does not fit
2778 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002779 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002780 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002781 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002782 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002783 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002784 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002785 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002786
Olivier Houcharded0f2072018-08-16 15:41:52 +02002787 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002788 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002789
2790 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002791 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002792 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002793 goto out_end_tcpcheck;
2794 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002795 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002796 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002797 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002798 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002799 goto out;
2800 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002801 }
2802
Willy Tarreau263013d2015-05-13 11:59:14 +02002803 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002804 break;
2805
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002806 /* have 'next' point to the next rule or NULL if we're on the
2807 * last one, connect() needs this.
2808 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002809 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002810
2811 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002812 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002813 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002814
2815 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002816 if (&next->list == head)
2817 next = NULL;
2818
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002819 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2820 struct protocol *proto;
2821 struct xprt_ops *xprt;
2822
Willy Tarreau00149122017-10-04 18:05:01 +02002823 /* For a connect action we'll create a new connection.
2824 * We may also have to kill a previous one. But we don't
2825 * want to leave *without* a connection if we came here
2826 * from the connection layer, hence with a connection.
2827 * Thus we'll proceed in the following order :
2828 * 1: close but not release previous connection
2829 * 2: try to get a new connection
2830 * 3: release and replace the old one on success
2831 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002832 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002833 cs_close(check->cs);
2834 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002835 }
2836
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002837 /* mark the step as started */
2838 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002839
2840 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002841 cs = cs_new(NULL);
2842 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002843 step = tcpcheck_get_step_id(check);
2844 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2845 comment = tcpcheck_get_step_comment(check, step);
2846 if (comment)
2847 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002848 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2849 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002850 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002851 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002852 }
2853
Olivier Houchard49065542019-05-31 19:20:36 +02002854 if (check->cs) {
2855 if (check->wait_list.events)
2856 cs->conn->xprt->unsubscribe(cs->conn,
2857 cs->conn->xprt_ctx,
2858 check->wait_list.events,
2859 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002860 /* We may have been scheduled to run, and the
2861 * I/O handler expects to have a cs, so remove
2862 * the tasklet
2863 */
2864 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002865 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002866 }
Willy Tarreau00149122017-10-04 18:05:01 +02002867
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002868 check->cs = cs;
2869 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002870 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002871 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002872 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002873
2874 /* no client address */
Willy Tarreauca79f592019-07-17 19:04:47 +02002875
2876 if (!sockaddr_alloc(&conn->dst)) {
2877 ret = SF_ERR_RESOURCE;
2878 goto fail_check;
2879 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002880
Simon Horman41f58762015-01-30 11:22:56 +09002881 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002882 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002883 *conn->dst = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002884 }
2885 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002886 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002887 *conn->dst = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002888 }
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002889 proto = protocol_by_family(conn->dst->ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002890
2891 /* port */
2892 if (check->current_step->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002893 set_host_port(conn->dst, check->current_step->port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002894 else if (check->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002895 set_host_port(conn->dst, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002896 else if (s->svc_port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002897 set_host_port(conn->dst, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002898
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002899 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002900 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002901 }
2902 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002903 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002904 }
Willy Tarreaube373152018-09-06 11:45:30 +02002905
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002906 conn_prepare(conn, proto, xprt);
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002907
2908 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2909 ret = SF_ERR_RESOURCE;
2910 goto fail_check;
2911 }
2912
Willy Tarreaube373152018-09-06 11:45:30 +02002913 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002914
Willy Tarreaue7dff022015-04-03 01:14:29 +02002915 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002916 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002917 ret = proto->connect(conn,
Olivier Houchardfdcb0072019-05-06 18:32:29 +02002918 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 +01002919 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2920 conn->send_proxy_ofs = 1;
2921 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002922 if (xprt_add_hs(conn) < 0)
2923 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002924 }
2925
2926 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002927 * - SF_ERR_NONE if everything's OK
2928 * - SF_ERR_SRVTO if there are no more servers
2929 * - SF_ERR_SRVCL if the connection was refused by the server
2930 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2931 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2932 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002933 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002934 * Note that we try to prevent the network stack from sending the ACK during the
2935 * connect() when a pure TCP check is used (without PROXY protocol).
2936 */
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002937 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002938 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002939 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002940 /* we allow up to min(inter, timeout.connect) for a connection
2941 * to establish but only when timeout.check is set
2942 * as it may be to short for a full check otherwise
2943 */
2944 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2945
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002946 if (proxy->timeout.check && proxy->timeout.connect) {
2947 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002948 t->expire = tick_first(t->expire, t_con);
2949 }
2950 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002951 case SF_ERR_SRVTO: /* ETIMEDOUT */
2952 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002953 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002954 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002955 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002956 comment = tcpcheck_get_step_comment(check, step);
2957 if (comment)
2958 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002959 set_server_check_status(check, HCHK_STATUS_L4CON,
2960 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002961 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002962 case SF_ERR_PRXCOND:
2963 case SF_ERR_RESOURCE:
2964 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002965 step = tcpcheck_get_step_id(check);
2966 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002967 comment = tcpcheck_get_step_comment(check, step);
2968 if (comment)
2969 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002970 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2971 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002972 goto out_end_tcpcheck;
2973 }
2974
2975 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002976 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002977
2978 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002979 while (&check->current_step->list != head &&
2980 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002981 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002982
Willy Tarreauf2c87352015-05-13 12:08:21 +02002983 if (&check->current_step->list == head)
2984 break;
2985
Willy Tarreau7df8ca62019-07-15 10:57:51 +02002986 /* don't do anything until the connection is established */
2987 if (!(conn->flags & CO_FL_CONNECTED))
2988 break;
2989
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002990 } /* end 'connect' */
2991 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2992 /* mark the step as started */
2993 check->last_started_step = check->current_step;
2994
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002995 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002996 if (*b_head(&check->bi) != '\0') {
2997 *b_head(&check->bi) = '\0';
2998 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002999 }
3000
Willy Tarreaubbae3f02017-08-30 09:59:52 +02003001 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003002 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003003 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003004 goto out_end_tcpcheck;
3005 }
3006
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003007 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003008 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003009 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003010 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003011 set_server_check_status(check, HCHK_STATUS_L7RSP,
3012 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003013 goto out_end_tcpcheck;
3014 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003015
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003016 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003017 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003018 continue;
3019
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003020 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3021 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003022
Willy Tarreauabca5b62013-12-06 14:19:25 +01003023 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003024 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003025
3026 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003027 while (&check->current_step->list != head &&
3028 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003029 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003030
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003031 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003032 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003033 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003034 goto out_end_tcpcheck;
3035
Olivier Houchard0ba6c852019-07-18 15:09:08 +02003036 /* If we already subscribed, then we tried to received
3037 * and failed, so there's no point trying again.
3038 */
3039 if (check->wait_list.events & SUB_RETRY_RECV)
3040 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003041 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003042 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003043 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003044 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003045 /* Report network errors only if we got no other data. Otherwise
3046 * we'll let the upper layers decide whether the response is OK
3047 * or not. It is very common that an RST sent by the server is
3048 * reported as an error just after the last data chunk.
3049 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003050 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003051 goto out_end_tcpcheck;
3052 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003053 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003054 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003055 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003056 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003057 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003058 }
3059
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003060 /* mark the step as started */
3061 check->last_started_step = check->current_step;
3062
3063
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003064 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003065 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003066 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003067 if (b_data(&check->bi) < b_size(&check->bi)) {
3068 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003069 }
3070 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003071 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003072 done = 1; /* buffer full, don't wait for more data */
3073 }
3074
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003075 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003076
3077 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003078 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003079 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003080 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003081
3082 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003083 step = tcpcheck_get_step_id(check);
3084 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003085 comment = tcpcheck_get_step_comment(check, step);
3086 if (comment)
3087 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003088 set_server_check_status(check, HCHK_STATUS_L7RSP,
3089 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003090
3091 goto out_end_tcpcheck;
3092 }
3093
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003094 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003095 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003096
Willy Tarreaua970c282013-12-06 12:47:19 +01003097 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003098 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003099 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 +02003100 else if (check->current_step->expect_regex != NULL)
3101 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003102
3103 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003104 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003105
3106 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003107 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003108 if (ret) {
3109 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003110 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003111 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003112 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003113 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003114 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003115 }
3116 else {
3117 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003118 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003119 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003120 comment = tcpcheck_get_step_comment(check, step);
3121 if (comment)
3122 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003123 set_server_check_status(check, HCHK_STATUS_L7RSP,
3124 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003125 goto out_end_tcpcheck;
3126 }
3127 /* matched and was supposed to => OK, next step */
3128 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003129 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003130 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003131
3132 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003133 while (&check->current_step->list != head &&
3134 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003135 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003136
Willy Tarreauf2c87352015-05-13 12:08:21 +02003137 if (&check->current_step->list == head)
3138 break;
3139
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003140 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003141 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003142 }
3143 }
3144 else {
3145 /* not matched */
3146 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003147 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003148 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003149 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003150
3151 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003152 while (&check->current_step->list != head &&
3153 check->current_step->action == TCPCHK_ACT_COMMENT)
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
Willy Tarreauf2c87352015-05-13 12:08:21 +02003156 if (&check->current_step->list == head)
3157 break;
3158
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003159 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003160 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003161 }
3162 /* not matched but was supposed to => ERROR */
3163 else {
3164 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003165 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003166 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003167 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003168 }
3169 else {
3170 /* we were looking for a regex */
3171 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003172 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003173 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003174 comment = tcpcheck_get_step_comment(check, step);
3175 if (comment)
3176 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003177 set_server_check_status(check, HCHK_STATUS_L7RSP,
3178 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003179 goto out_end_tcpcheck;
3180 }
3181 }
3182 } /* end expect */
3183 } /* end loop over double chained step list */
3184
Baptiste Assmann248f1172018-03-01 21:49:01 +01003185 /* don't do anything until the connection is established */
3186 if (!(conn->flags & CO_FL_CONNECTED)) {
3187 /* update expire time, should be done by process_chk */
3188 /* we allow up to min(inter, timeout.connect) for a connection
3189 * to establish but only when timeout.check is set
3190 * as it may be to short for a full check otherwise
3191 */
3192 while (tick_is_expired(t->expire, now_ms)) {
3193 int t_con;
3194
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003195 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003196 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3197
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003198 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003199 t->expire = tick_first(t->expire, t_con);
3200 }
3201 goto out;
3202 }
3203
Willy Tarreau263013d2015-05-13 11:59:14 +02003204 /* We're waiting for some I/O to complete, we've reached the end of the
3205 * rules, or both. Do what we have to do, otherwise we're done.
3206 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003207 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003208 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3209 goto out_end_tcpcheck;
3210 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003211
Willy Tarreau53c5a042015-05-13 11:38:17 +02003212 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003213 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003214 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003215 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003216
3217 out_end_tcpcheck:
3218 /* collect possible new errors */
Willy Tarreauef91c932019-07-23 14:37:47 +02003219 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003220 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003221
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003222 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003223 check->current_step = NULL;
3224
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003225 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003226 conn->flags |= CO_FL_ERROR;
3227
Christopher Fauletb6102852017-11-28 10:06:29 +01003228 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003229 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003230}
3231
Simon Hormanb1900d52015-01-30 11:22:54 +09003232const char *init_check(struct check *check, int type)
3233{
3234 check->type = type;
3235
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003236 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3237 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003238
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003239 check->bi.area = calloc(check->bi.size, sizeof(char));
3240 check->bo.area = calloc(check->bo.size, sizeof(char));
3241
3242 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003243 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003244
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003245 check->wait_list.tasklet = tasklet_new();
3246 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003247 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003248 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003249 check->wait_list.tasklet->process = event_srv_chk_io;
3250 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003251 return NULL;
3252}
3253
Simon Hormanbfb5d332015-01-30 11:22:55 +09003254void free_check(struct check *check)
3255{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003256 free(check->bi.area);
3257 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003258 if (check->cs) {
3259 free(check->cs->conn);
3260 check->cs->conn = NULL;
3261 cs_free(check->cs);
3262 check->cs = NULL;
3263 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003264}
3265
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003266void email_alert_free(struct email_alert *alert)
3267{
3268 struct tcpcheck_rule *rule, *back;
3269
3270 if (!alert)
3271 return;
3272
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003273 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3274 LIST_DEL(&rule->list);
3275 free(rule->comment);
3276 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003277 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003278 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003279 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003280 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003281}
3282
Olivier Houchard9f6af332018-05-25 14:04:04 +02003283static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003284{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003285 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003286 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003287 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003288
3289 q = container_of(check, typeof(*q), check);
3290
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003291 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003292 while (1) {
3293 if (!(check->state & CHK_ST_ENABLED)) {
3294 if (LIST_ISEMPTY(&q->email_alerts)) {
3295 /* All alerts processed, queue the task */
3296 t->expire = TICK_ETERNITY;
3297 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003298 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003299 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003300
3301 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003302 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003303 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003304 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003305 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003306 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003307 }
3308
Olivier Houchard9f6af332018-05-25 14:04:04 +02003309 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003310 if (check->state & CHK_ST_INPROGRESS)
3311 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003312
3313 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3314 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003315 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003316 check->server = NULL;
3317 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003318 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003319 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003320 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003321 return t;
3322}
3323
Christopher Faulet0108bb32017-10-20 21:34:32 +02003324/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3325 *
3326 * The function returns 1 in success case, otherwise, it returns 0 and err is
3327 * filled.
3328 */
3329int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003330{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003331 struct mailer *mailer;
3332 struct email_alertq *queues;
3333 const char *err_str;
3334 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003335
Christopher Faulet0108bb32017-10-20 21:34:32 +02003336 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3337 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003338 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003339 }
3340
Christopher Faulet0108bb32017-10-20 21:34:32 +02003341 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3342 struct email_alertq *q = &queues[i];
3343 struct check *check = &q->check;
3344 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003345
3346 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003347 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003348 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003349 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003350 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003351 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003352 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3353 memprintf(err, "%s", err_str);
3354 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003355 }
3356
3357 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003358 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003359 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003360
Emeric Brunc60def82017-09-27 14:59:38 +02003361 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003362 memprintf(err, "out of memory while allocating mailer alerts task");
3363 goto error;
3364 }
3365
3366 check->task = t;
3367 t->process = process_email_alert;
3368 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003369
Christopher Faulet0108bb32017-10-20 21:34:32 +02003370 /* check this in one ms */
3371 t->expire = TICK_ETERNITY;
3372 check->start = now;
3373 task_queue(t);
3374 }
3375
3376 mls->users++;
3377 free(p->email_alert.mailers.name);
3378 p->email_alert.mailers.m = mls;
3379 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003381
3382 error:
3383 for (i = 0; i < mls->count; i++) {
3384 struct email_alertq *q = &queues[i];
3385 struct check *check = &q->check;
3386
Willy Tarreauf6562792019-05-07 19:05:35 +02003387 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003388 free_check(check);
3389 }
3390 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003391 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003392 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003393}
3394
3395
3396static int add_tcpcheck_expect_str(struct list *list, const char *str)
3397{
3398 struct tcpcheck_rule *tcpcheck;
3399
Willy Tarreaubafbe012017-11-24 17:34:44 +01003400 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003401 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003402 memset(tcpcheck, 0, sizeof(*tcpcheck));
3403 tcpcheck->action = TCPCHK_ACT_EXPECT;
3404 tcpcheck->string = strdup(str);
3405 tcpcheck->expect_regex = NULL;
3406 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003407 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003408 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003409 return 0;
3410 }
3411
3412 LIST_ADDQ(list, &tcpcheck->list);
3413 return 1;
3414}
3415
3416static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3417{
3418 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003419 const char *in;
3420 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003421 int i;
3422
Willy Tarreaubafbe012017-11-24 17:34:44 +01003423 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003424 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003425 memset(tcpcheck, 0, sizeof(*tcpcheck));
3426 tcpcheck->action = TCPCHK_ACT_SEND;
3427 tcpcheck->expect_regex = NULL;
3428 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003429 tcpcheck->string_len = 0;
3430 for (i = 0; strs[i]; i++)
3431 tcpcheck->string_len += strlen(strs[i]);
3432
3433 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3434 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003435 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003436 return 0;
3437 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003438
Willy Tarreau64345aa2016-08-10 19:29:09 +02003439 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003440 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003441 for (in = strs[i]; (*dst = *in++); dst++);
3442 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003443
3444 LIST_ADDQ(list, &tcpcheck->list);
3445 return 1;
3446}
3447
Christopher Faulet0108bb32017-10-20 21:34:32 +02003448static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3449 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003450{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003451 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003452 struct tcpcheck_rule *tcpcheck;
3453 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003454
Willy Tarreaubafbe012017-11-24 17:34:44 +01003455 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003456 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003457 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003458 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003459 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003460
Willy Tarreaubafbe012017-11-24 17:34:44 +01003461 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003462 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003463 memset(tcpcheck, 0, sizeof(*tcpcheck));
3464 tcpcheck->action = TCPCHK_ACT_CONNECT;
3465 tcpcheck->comment = NULL;
3466 tcpcheck->string = NULL;
3467 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003468 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3469
3470 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3471 goto error;
3472
3473 {
3474 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3475 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3476 goto error;
3477 }
3478
3479 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3480 goto error;
3481
3482 {
3483 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3484 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3485 goto error;
3486 }
3487
3488 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3489 goto error;
3490
3491 {
3492 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3493 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3494 goto error;
3495 }
3496
3497 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3498 goto error;
3499
3500 {
3501 const char * const strs[2] = { "DATA\r\n" };
3502 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3503 goto error;
3504 }
3505
3506 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3507 goto error;
3508
3509 {
3510 struct tm tm;
3511 char datestr[48];
3512 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003513 "From: ", p->email_alert.from, "\r\n",
3514 "To: ", p->email_alert.to, "\r\n",
3515 "Date: ", datestr, "\r\n",
3516 "Subject: [HAproxy Alert] ", msg, "\r\n",
3517 "\r\n",
3518 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003519 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003520 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003521 NULL
3522 };
3523
3524 get_localtime(date.tv_sec, &tm);
3525
3526 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3527 goto error;
3528 }
3529
3530 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3531 goto error;
3532 }
3533
3534 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3535 goto error;
3536
3537 {
3538 const char * const strs[2] = { "QUIT\r\n" };
3539 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3540 goto error;
3541 }
3542
3543 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3544 goto error;
3545
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003546 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003547 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003548 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003549 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003550 return 1;
3551
3552error:
3553 email_alert_free(alert);
3554 return 0;
3555}
3556
Christopher Faulet0108bb32017-10-20 21:34:32 +02003557static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003558{
3559 int i;
3560 struct mailer *mailer;
3561
3562 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3563 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003564 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003565 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003566 return;
3567 }
3568 }
3569
3570 return;
3571}
3572
3573/*
3574 * Send email alert if configured.
3575 */
Simon Horman64e34162015-02-06 11:11:57 +09003576void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003577{
3578 va_list argp;
3579 char buf[1024];
3580 int len;
3581 struct proxy *p = s->proxy;
3582
Christopher Faulet0108bb32017-10-20 21:34:32 +02003583 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003584 return;
3585
3586 va_start(argp, format);
3587 len = vsnprintf(buf, sizeof(buf), format, argp);
3588 va_end(argp);
3589
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003590 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003591 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003592 return;
3593 }
3594
Christopher Faulet0108bb32017-10-20 21:34:32 +02003595 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003596}
3597
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003598/*
3599 * Return value:
3600 * the port to be used for the health check
3601 * 0 in case no port could be found for the check
3602 */
3603int srv_check_healthcheck_port(struct check *chk)
3604{
3605 int i = 0;
3606 struct server *srv = NULL;
3607
3608 srv = chk->server;
3609
3610 /* If neither a port nor an addr was specified and no check transport
3611 * layer is forced, then the transport layer used by the checks is the
3612 * same as for the production traffic. Otherwise we use raw_sock by
3613 * default, unless one is specified.
3614 */
3615 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003616 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003617 chk->send_proxy |= (srv->pp_opts);
3618 }
3619
3620 /* by default, we use the health check port ocnfigured */
3621 if (chk->port > 0)
3622 return chk->port;
3623
3624 /* try to get the port from check_core.addr if check.port not set */
3625 i = get_host_port(&chk->addr);
3626 if (i > 0)
3627 return i;
3628
3629 /* try to get the port from server address */
3630 /* prevent MAPPORTS from working at this point, since checks could
3631 * not be performed in such case (MAPPORTS impose a relative ports
3632 * based on live traffic)
3633 */
3634 if (srv->flags & SRV_F_MAPPORTS)
3635 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003636
3637 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003638 if (i > 0)
3639 return i;
3640
3641 return 0;
3642}
3643
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003644REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003645
Willy Tarreaubd741542010-03-16 18:46:54 +01003646/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003647 * Local variables:
3648 * c-indent-level: 8
3649 * c-basic-offset: 8
3650 * End:
3651 */