blob: 90f46147717dd708368f3f2862ec7a20b2b898f5 [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 Houchard4501c3e2018-08-28 19:36:18 +0200719
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100720 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200721 wake_srv_chk(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100722 if (!(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100723 if (check->server)
724 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
725 else
726 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200727 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100728 if (check->server)
729 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
730 else
731 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200732 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200733 return NULL;
734}
735
736/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100737 *
738 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200739 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
740 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200741 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200742static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200743{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200744 struct connection *conn = cs->conn;
745 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900746 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900747 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200748
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100749 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100750 goto out_wakeup;
751
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200752 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100753 if (!(conn->flags & CO_FL_ERROR))
754 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200755 goto out;
756 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100757
Willy Tarreau20a18342013-12-05 00:31:46 +0100758 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200759 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100760 goto out_wakeup;
761 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100762
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200763 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100764 /* if the output is closed, we can't do anything */
765 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200766 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100767 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200768 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200769
Willy Tarreau06559ac2013-12-05 01:53:08 +0100770 /* here, we know that the connection is established. That's enough for
771 * a pure TCP check.
772 */
773 if (!check->type)
774 goto out_wakeup;
775
Willy Tarreauc09572f2017-10-04 11:58:22 +0200776 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100777 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200778 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200779
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200780 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200781 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200782 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200783 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200784 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100785 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200786 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200787 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100788 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200789 goto out;
790 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100791 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200792
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100793 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
794 if (s->proxy->timeout.check) {
795 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
796 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200797 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200798 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100799
Willy Tarreau83749182007-04-15 20:56:27 +0200800 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200801 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200802 out:
803 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200804}
805
Willy Tarreaubaaee002006-06-26 02:48:02 +0200806/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200807 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200808 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900809 * set_server_check_status() to update check->status, check->duration
810 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200811
812 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
813 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
814 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
815 * response to an SSL HELLO (the principle is that this is enough to
816 * distinguish between an SSL server and a pure TCP relay). All other cases will
817 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
818 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100819 *
820 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200821 * via the out label.
822 *
823 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200824 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200825static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200826{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200827 struct connection *conn = cs->conn;
828 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900829 struct server *s = check->server;
830 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200831 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100832 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200833 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200834
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100835 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200836 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200837
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200838 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100839 if (!(conn->flags & CO_FL_ERROR))
840 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200841 goto out;
842 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200843
Willy Tarreauc09572f2017-10-04 11:58:22 +0200844 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100845 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200846 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200847
Willy Tarreau83749182007-04-15 20:56:27 +0200848 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
849 * but the connection was closed on the remote end. Fortunately, recv still
850 * works correctly and we don't need to do the getsockopt() on linux.
851 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000852
853 /* Set buffer to point to the end of the data already read, and check
854 * that there is free space remaining. If the buffer is full, proceed
855 * with running the checks without attempting another socket read.
856 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000857
Willy Tarreau03938182010-03-17 21:52:07 +0100858 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000859
Olivier Houchard511efea2018-08-16 15:30:32 +0200860 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200861 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100862 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200863 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200864 /* Report network errors only if we got no other data. Otherwise
865 * we'll let the upper layers decide whether the response is OK
866 * or not. It is very common that an RST sent by the server is
867 * reported as an error just after the last data chunk.
868 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200869 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100870 goto out_wakeup;
871 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200872 }
873
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100874
Willy Tarreau03938182010-03-17 21:52:07 +0100875 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200876 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100877 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200878 if (b_data(&check->bi) < b_size(&check->bi))
879 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100880 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200881 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100882 done = 1; /* buffer full, don't wait for more data */
883 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200884
Nick Chalk57b1bf72010-03-16 15:50:46 +0000885 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900886 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200887 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200888 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100889 goto wait_more_data;
890
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100891 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200892 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
893 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
894 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
895 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
896 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
897 cut_crlf(b_head(&check->bi));
898 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200899
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100900 goto out_wakeup;
901 }
902
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200903 check->code = str2uic(b_head(&check->bi) + 9);
904 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200905
Willy Tarreaubd741542010-03-16 18:46:54 +0100906 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200907 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000908 /* 404 may be accepted as "stopping" only if the server was up */
909 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900910 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000911 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100912 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
913 /* Run content verification check... We know we have at least 13 chars */
914 if (!httpchk_expect(s, done))
915 goto wait_more_data;
916 }
917 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200918 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100919 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900920 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100921 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000922 else {
923 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900924 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000925 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200926 break;
927
928 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200929 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100930 goto wait_more_data;
931
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100932 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200933 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900934 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200935 else
Simon Horman4a741432013-02-23 15:35:38 +0900936 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200937 break;
938
939 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200940 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100941 goto wait_more_data;
942
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200943 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200944 if ((b_data(&check->bi) < strlen("000\r")) ||
945 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
946 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
947 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
948 cut_crlf(b_head(&check->bi));
949 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200950 goto out_wakeup;
951 }
952
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200953 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200954
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200955 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200956 cut_crlf(desc);
957
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100958 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200959 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900960 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200961 else
Simon Horman4a741432013-02-23 15:35:38 +0900962 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200963 break;
964
Simon Hormana2b9dad2013-02-12 10:45:54 +0900965 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100966 int status = HCHK_STATUS_CHECKED;
967 const char *hs = NULL; /* health status */
968 const char *as = NULL; /* admin status */
969 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200970 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100971 const char *err = NULL; /* first error to report */
972 const char *wrn = NULL; /* first warning to report */
973 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900974
Willy Tarreau81f5d942013-12-09 20:51:51 +0100975 /* We're getting an agent check response. The agent could
976 * have been disabled in the mean time with a long check
977 * still pending. It is important that we ignore the whole
978 * response.
979 */
980 if (!(check->server->agent.state & CHK_ST_ENABLED))
981 break;
982
983 /* The agent supports strings made of a single line ended by the
984 * first CR ('\r') or LF ('\n'). This line is composed of words
985 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
986 * line may optionally contained a description of a state change
987 * after a sharp ('#'), which is only considered if a health state
988 * is announced.
989 *
990 * Words may be composed of :
991 * - a numeric weight suffixed by the percent character ('%').
992 * - a health status among "up", "down", "stopped", and "fail".
993 * - an admin status among "ready", "drain", "maint".
994 *
995 * These words may appear in any order. If multiple words of the
996 * same category appear, the last one wins.
997 */
998
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200999 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001000 while (*p && *p != '\n' && *p != '\r')
1001 p++;
1002
1003 if (!*p) {
1004 if (!done)
1005 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001006
Willy Tarreau9809b782013-12-11 21:40:11 +01001007 /* at least inform the admin that the agent is mis-behaving */
1008 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1009 break;
1010 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001011
Willy Tarreau9809b782013-12-11 21:40:11 +01001012 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001013 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001014
Willy Tarreau81f5d942013-12-09 20:51:51 +01001015 while (*cmd) {
1016 /* look for next word */
1017 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1018 cmd++;
1019 continue;
1020 }
Simon Horman671b6f02013-11-25 10:46:39 +09001021
Willy Tarreau81f5d942013-12-09 20:51:51 +01001022 if (*cmd == '#') {
1023 /* this is the beginning of a health status description,
1024 * skip the sharp and blanks.
1025 */
1026 cmd++;
1027 while (*cmd == '\t' || *cmd == ' ')
1028 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001029 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001030 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001031
1032 /* find the end of the word so that we have a null-terminated
1033 * word between <cmd> and <p>.
1034 */
1035 p = cmd + 1;
1036 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1037 p++;
1038 if (*p)
1039 *p++ = 0;
1040
1041 /* first, health statuses */
1042 if (strcasecmp(cmd, "up") == 0) {
1043 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001044 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001045 hs = cmd;
1046 }
1047 else if (strcasecmp(cmd, "down") == 0) {
1048 check->health = 0;
1049 status = HCHK_STATUS_L7STS;
1050 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001051 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001052 else if (strcasecmp(cmd, "stopped") == 0) {
1053 check->health = 0;
1054 status = HCHK_STATUS_L7STS;
1055 hs = cmd;
1056 }
1057 else if (strcasecmp(cmd, "fail") == 0) {
1058 check->health = 0;
1059 status = HCHK_STATUS_L7STS;
1060 hs = cmd;
1061 }
1062 /* admin statuses */
1063 else if (strcasecmp(cmd, "ready") == 0) {
1064 as = cmd;
1065 }
1066 else if (strcasecmp(cmd, "drain") == 0) {
1067 as = cmd;
1068 }
1069 else if (strcasecmp(cmd, "maint") == 0) {
1070 as = cmd;
1071 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001072 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001073 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1074 ps = cmd;
1075 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001076 /* try to parse a maxconn here */
1077 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1078 cs = cmd;
1079 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001080 else {
1081 /* keep a copy of the first error */
1082 if (!err)
1083 err = cmd;
1084 }
1085 /* skip to next word */
1086 cmd = p;
1087 }
1088 /* here, cmd points either to \0 or to the beginning of a
1089 * description. Skip possible leading spaces.
1090 */
1091 while (*cmd == ' ' || *cmd == '\n')
1092 cmd++;
1093
1094 /* First, update the admin status so that we avoid sending other
1095 * possibly useless warnings and can also update the health if
1096 * present after going back up.
1097 */
1098 if (as) {
1099 if (strcasecmp(as, "drain") == 0)
1100 srv_adm_set_drain(check->server);
1101 else if (strcasecmp(as, "maint") == 0)
1102 srv_adm_set_maint(check->server);
1103 else
1104 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001105 }
1106
Willy Tarreau81f5d942013-12-09 20:51:51 +01001107 /* now change weights */
1108 if (ps) {
1109 const char *msg;
1110
1111 msg = server_parse_weight_change_request(s, ps);
1112 if (!wrn || !*wrn)
1113 wrn = msg;
1114 }
1115
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001116 if (cs) {
1117 const char *msg;
1118
1119 cs += strlen("maxconn:");
1120
1121 msg = server_parse_maxconn_change_request(s, cs);
1122 if (!wrn || !*wrn)
1123 wrn = msg;
1124 }
1125
Willy Tarreau81f5d942013-12-09 20:51:51 +01001126 /* and finally health status */
1127 if (hs) {
1128 /* We'll report some of the warnings and errors we have
1129 * here. Down reports are critical, we leave them untouched.
1130 * Lack of report, or report of 'UP' leaves the room for
1131 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001132 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001133 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001134 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001135
1136 if (!*msg || status == HCHK_STATUS_L7OKD) {
1137 if (err && *err)
1138 msg = err;
1139 else if (wrn && *wrn)
1140 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001141 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001142
1143 t = get_trash_chunk();
1144 chunk_printf(t, "via agent : %s%s%s%s",
1145 hs, *msg ? " (" : "",
1146 msg, *msg ? ")" : "");
1147
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001148 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001149 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001150 else if (err && *err) {
1151 /* No status change but we'd like to report something odd.
1152 * Just report the current state and copy the message.
1153 */
1154 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001155 set_server_check_status(check, status/*check->status*/,
1156 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001157
Willy Tarreau81f5d942013-12-09 20:51:51 +01001158 }
1159 else if (wrn && *wrn) {
1160 /* No status change but we'd like to report something odd.
1161 * Just report the current state and copy the message.
1162 */
1163 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001164 set_server_check_status(check, status/*check->status*/,
1165 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001166 }
1167 else
1168 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001169 break;
1170 }
1171
Willy Tarreau1620ec32011-08-06 17:05:02 +02001172 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001173 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001174 goto wait_more_data;
1175
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001176 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001177 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001178 }
1179 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001180 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1181 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001182 else
1183 desc = "PostgreSQL unknown error";
1184
Simon Horman4a741432013-02-23 15:35:38 +09001185 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001186 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001187 break;
1188
1189 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001190 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001191 goto wait_more_data;
1192
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001193 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001194 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001195 }
1196 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001197 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001198 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001199 break;
1200
1201 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001202 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001203 goto wait_more_data;
1204
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001205 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001206 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001207 /* We set the MySQL Version in description for information purpose
1208 * FIXME : it can be cool to use MySQL Version for other purpose,
1209 * like mark as down old MySQL server.
1210 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001211 if (b_data(&check->bi) > 51) {
1212 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001213 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001214 }
1215 else {
1216 if (!done)
1217 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001218
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001219 /* it seems we have a OK packet but without a valid length,
1220 * it must be a protocol error
1221 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001222 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001223 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001224 }
1225 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001226 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001227 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001228 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 }
1230 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001231 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1232 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1233 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001234
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001235 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001236 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001237 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001238 /* We have only one MySQL packet and it is a Handshake Initialization packet
1239 * but we need to have a second packet to know if it is alright
1240 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001241 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001242 goto wait_more_data;
1243 }
1244 else {
1245 /* We have only one packet and it is an Error packet,
1246 * an error message is attached, so we can display it
1247 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001248 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001249 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001250 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001251 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001252 } else if (b_data(&check->bi) > first_packet_len + 4) {
1253 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1254 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1255 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001256
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001257 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001258 /* We have 2 packets and that's good */
1259 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001260 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001261 /* No error packet */
1262 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001263 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001264 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001265 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001266 }
1267 else {
1268 /* An error message is attached in the Error packet
1269 * so we can display it ! :)
1270 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001271 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001272 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001273 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001274 }
1275 }
1276 }
1277 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001278 if (!done)
1279 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001280
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001281 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001282 * it must be a protocol error
1283 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001284 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001285 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001286 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001287 }
1288 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001289 break;
1290
1291 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001292 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001293 goto wait_more_data;
1294
1295 /* Check if the server speaks LDAP (ASN.1/BER)
1296 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1297 * http://tools.ietf.org/html/rfc4511
1298 */
1299
1300 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1301 * LDAPMessage: 0x30: SEQUENCE
1302 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001303 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001304 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001305 }
1306 else {
1307 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001308 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001309
1310 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1311 * messageID: 0x02 0x01 0x01: INTEGER 1
1312 * protocolOp: 0x61: bindResponse
1313 */
1314 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001315 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001316 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001317 goto out_wakeup;
1318 }
1319
1320 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001321 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001322
1323 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1324 * ldapResult: 0x0a 0x01: ENUMERATION
1325 */
1326 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001327 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001328 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001329 goto out_wakeup;
1330 }
1331
1332 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1333 * resultCode
1334 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001335 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001336 if (check->code) {
1337 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 +02001338 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001339 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001340 }
1341 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001342 break;
1343
Christopher Fauletba7bc162016-11-07 21:07:38 +01001344 case PR_O2_SPOP_CHK: {
1345 unsigned int framesz;
1346 char err[HCHK_DESC_LEN];
1347
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001348 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001349 goto wait_more_data;
1350
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001351 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001352 framesz = ntohl(framesz);
1353
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001354 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001355 goto wait_more_data;
1356
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001357 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001358 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1359 else
1360 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1361 break;
1362 }
1363
Willy Tarreau1620ec32011-08-06 17:05:02 +02001364 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001365 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001366 break;
1367 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001368
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001369 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001370 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001371 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001372 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001373
Nick Chalk57b1bf72010-03-16 15:50:46 +00001374 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001375 *b_head(&check->bi) = '\0';
1376 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001377
Steven Davidovitz544d4812017-03-08 11:06:20 -08001378 /* Close the connection... We still attempt to nicely close if,
1379 * for instance, SSL needs to send a "close notify." Later, we perform
1380 * a hard close and reset the connection if some data are pending,
1381 * otherwise we end up with many TIME_WAITs and eat all the source port
1382 * range quickly. To avoid sending RSTs all the time, we first try to
1383 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001384 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001385 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001386
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001387 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001388 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001389 conn->flags |= CO_FL_ERROR;
1390
Willy Tarreaufdccded2008-08-29 18:19:04 +02001391 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001392out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001393 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001394
1395 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001396 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001397 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001398}
1399
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001400/*
1401 * This function is used only for server health-checks. It handles connection
1402 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001403 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1404 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001405 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001406static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001407{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001408 struct connection *conn = cs->conn;
1409 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001410 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001411 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001412
Olivier Houchard0923fa42019-01-11 18:43:04 +01001413 if (check->server)
1414 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1415 else
1416 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001417
Willy Tarreauc09572f2017-10-04 11:58:22 +02001418 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001419 if (check->type == PR_O2_TCPCHK_CHK) {
1420 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001421 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001422 conn = cs->conn;
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001423 } else if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001424 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001425
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001426 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001427 /* We may get error reports bypassing the I/O handlers, typically
1428 * the case when sending a pure TCP check which fails, then the I/O
1429 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001430 * main processing task so let's simply wake it up. If we get here,
1431 * we expect errno to still be valid.
1432 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001433 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001434 task_wakeup(check->task, TASK_WOKEN_IO);
1435 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001436 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001437 /* we may get here if only a connection probe was required : we
1438 * don't have any data to send nor anything expected in response,
1439 * so the completion of the connection establishment is enough.
1440 */
1441 task_wakeup(check->task, TASK_WOKEN_IO);
1442 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001443
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001444 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001445 /* Check complete or aborted. If connection not yet closed do it
1446 * now and wake the check task up to be sure the result is
1447 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001448 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001449 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001450 ret = -1;
Christopher Faulet774c4862019-01-21 14:15:50 +01001451 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001452 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001453
Olivier Houchard0923fa42019-01-11 18:43:04 +01001454 if (check->server)
1455 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1456 else
1457 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001458
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001459 /* if a connection got replaced, we must absolutely prevent the connection
1460 * handler from touching its fd, and perform the FD polling updates ourselves
1461 */
1462 if (ret < 0)
1463 conn_cond_update_polling(conn);
1464
1465 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001466}
1467
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001468struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001469 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001470 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001471};
1472
Willy Tarreaubaaee002006-06-26 02:48:02 +02001473/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001474 * updates the server's weight during a warmup stage. Once the final weight is
1475 * reached, the task automatically stops. Note that any server status change
1476 * must have updated s->last_change accordingly.
1477 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001478static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001479{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001480 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001481
1482 /* by default, plan on stopping the task */
1483 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001484 if ((s->next_admin & SRV_ADMF_MAINT) ||
1485 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001486 return t;
1487
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001488 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1489
Willy Tarreau892337c2014-05-13 23:41:20 +02001490 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001491 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001492
1493 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001494 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001495
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001496 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1497
Willy Tarreau2e993902011-10-31 11:53:20 +01001498 /* get back there in 1 second or 1/20th of the slowstart interval,
1499 * whichever is greater, resulting in small 5% steps.
1500 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001501 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001502 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1503 return t;
1504}
1505
Willy Tarreau894c6422017-10-04 15:58:52 +02001506/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1507 * none was found.
1508 */
1509static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1510{
1511 struct tcpcheck_rule *r;
1512
1513 list_for_each_entry(r, list, list) {
1514 if (r->action != TCPCHK_ACT_COMMENT)
1515 return r;
1516 }
1517 return NULL;
1518}
1519
Willy Tarreau2e993902011-10-31 11:53:20 +01001520/*
Simon Horman98637e52014-06-20 12:30:16 +09001521 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001522 *
1523 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001524 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1525 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1526 * - SF_ERR_SRVTO if there are no more servers
1527 * - SF_ERR_SRVCL if the connection was refused by the server
1528 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1529 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1530 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001531 * - 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 +01001532 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001533 * Note that we try to prevent the network stack from sending the ACK during the
1534 * connect() when a pure TCP check is used (without PROXY protocol).
1535 */
Simon Horman98637e52014-06-20 12:30:16 +09001536static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001537{
1538 struct check *check = t->context;
1539 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001540 struct conn_stream *cs = check->cs;
1541 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001542 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001543 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001544 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001545 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001546
Willy Tarreau00149122017-10-04 18:05:01 +02001547 /* we cannot have a connection here */
1548 if (conn)
1549 return SF_ERR_INTERNAL;
1550
Simon Hormanb00d17a2014-06-13 16:18:16 +09001551 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001552 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001553 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001554 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1555 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001556
1557 /* prepare the check buffer.
1558 * This should not be used if check is the secondary agent check
1559 * of a server as s->proxy->check_req will relate to the
1560 * configuration of the primary check. Similarly, tcp-check uses
1561 * its own strings.
1562 */
1563 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001564 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001565
1566 /* we want to check if this host replies to HTTP or SSLv3 requests
1567 * so we'll send the request, and won't wake the checker up now.
1568 */
1569 if ((check->type) == PR_O2_SSL3_CHK) {
1570 /* SSL requires that we put Unix time in the request */
1571 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001572 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001573 }
1574 else if ((check->type) == PR_O2_HTTP_CHK) {
1575 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001576 b_putblk(&check->bo, trash.area,
1577 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001578 /* prevent HTTP keep-alive when "http-check expect" is used */
1579 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001580 b_putist(&check->bo, ist("Connection: close\r\n"));
1581 b_putist(&check->bo, ist("\r\n"));
1582 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001583 }
1584 }
1585
James Brown55f9ff12015-10-21 18:19:05 -07001586 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001587 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001588 }
1589
Willy Tarreauf411cce2017-10-04 16:21:19 +02001590 /* for tcp-checks, the initial connection setup is handled separately as
1591 * it may be sent to a specific port and not to the server's.
1592 */
1593 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1594 tcpcheck_main(check);
1595 return SF_ERR_UP;
1596 }
1597
Simon Hormanb00d17a2014-06-13 16:18:16 +09001598 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001599 cs = check->cs = cs_new(NULL);
1600 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001601 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001602 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001603 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001604 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001605
Simon Horman41f58762015-01-30 11:22:56 +09001606 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001607 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001608 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001609 }
1610 else {
1611 /* we'll connect to the addr on the server */
1612 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001613 }
1614
Alexander Liu2a54bb72019-05-22 19:44:48 +08001615 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1616 conn->send_proxy_ofs = 1;
1617 conn->flags |= CO_FL_SOCKS4;
1618 }
1619
Olivier Houchard6377a002017-12-01 22:04:05 +01001620 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001621 conn->target = &s->obj_type;
1622
Willy Tarreaube373152018-09-06 11:45:30 +02001623 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001624 int i = 0;
1625
1626 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001627 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001628 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001629
1630 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001631 }
1632
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001633 /* no client address */
1634 clear_addr(&conn->addr.from);
1635
Willy Tarreaube373152018-09-06 11:45:30 +02001636 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001637 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1638 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001639 cs_attach(cs, check, &check_conn_cb);
1640
Willy Tarreauf3d34822014-12-08 12:11:28 +01001641 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001642 if (check->type != 0)
1643 connflags |= CONNECT_HAS_DATA;
1644 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1645 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1646 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001647
Willy Tarreaue7dff022015-04-03 01:14:29 +02001648 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001649 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001650 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001651
Willy Tarreau16257f62017-11-02 15:45:00 +01001652
Olivier Houchard9130a962017-10-17 17:33:43 +02001653#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001654 if (ret == SF_ERR_NONE) {
1655 if (s->check.sni)
1656 ssl_sock_set_servername(conn, s->check.sni);
1657 if (s->check.alpn_str)
1658 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1659 s->check.alpn_len);
1660 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001661#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001662 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001663 conn->send_proxy_ofs = 1;
1664 conn->flags |= CO_FL_SEND_PROXY;
1665 }
1666
1667 return ret;
1668}
1669
Simon Horman98637e52014-06-20 12:30:16 +09001670static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001671static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001672__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001673
1674void block_sigchld(void)
1675{
1676 sigset_t set;
1677 sigemptyset(&set);
1678 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001679 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001680}
1681
1682void unblock_sigchld(void)
1683{
1684 sigset_t set;
1685 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001686 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001687 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001688}
1689
Simon Horman98637e52014-06-20 12:30:16 +09001690static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1691{
1692 struct pid_list *elem;
1693 struct check *check = t->context;
1694
Willy Tarreaubafbe012017-11-24 17:34:44 +01001695 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001696 if (!elem)
1697 return NULL;
1698 elem->pid = pid;
1699 elem->t = t;
1700 elem->exited = 0;
1701 check->curpid = elem;
1702 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001703
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001704 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001705 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001706 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001707
Simon Horman98637e52014-06-20 12:30:16 +09001708 return elem;
1709}
1710
Simon Horman98637e52014-06-20 12:30:16 +09001711static void pid_list_del(struct pid_list *elem)
1712{
1713 struct check *check;
1714
1715 if (!elem)
1716 return;
1717
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001718 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001719 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001720 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001721
Simon Horman98637e52014-06-20 12:30:16 +09001722 if (!elem->exited)
1723 kill(elem->pid, SIGTERM);
1724
1725 check = elem->t->context;
1726 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001727 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001728}
1729
1730/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1731static void pid_list_expire(pid_t pid, int status)
1732{
1733 struct pid_list *elem;
1734
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001735 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001736 list_for_each_entry(elem, &pid_list, list) {
1737 if (elem->pid == pid) {
1738 elem->t->expire = now_ms;
1739 elem->status = status;
1740 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001741 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001742 break;
Simon Horman98637e52014-06-20 12:30:16 +09001743 }
1744 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001745 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001746}
1747
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001748static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001749{
1750 pid_t pid;
1751 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001752
Simon Horman98637e52014-06-20 12:30:16 +09001753 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1754 pid_list_expire(pid, status);
1755}
1756
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001757static int init_pid_list(void)
1758{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001759 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001760 /* Nothing to do */
1761 return 0;
1762
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001763 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001764 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1765 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001766 return 1;
1767 }
1768
Willy Tarreaubafbe012017-11-24 17:34:44 +01001769 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1770 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001771 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1772 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001773 return 1;
1774 }
1775
1776 return 0;
1777}
1778
Cyril Bontéac92a062014-12-27 22:28:38 +01001779/* helper macro to set an environment variable and jump to a specific label on failure. */
1780#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001781
1782/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001783 * helper function to allocate enough memory to store an environment variable.
1784 * It will also check that the environment variable is updatable, and silently
1785 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001786 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001787static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001788{
1789 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001790 char *envname;
1791 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001792
Cyril Bontéac92a062014-12-27 22:28:38 +01001793 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001794 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001795 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001796 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001797
1798 envname = extcheck_envs[idx].name;
1799 vmaxlen = extcheck_envs[idx].vmaxlen;
1800
1801 /* Check if the environment variable is already set, and silently reject
1802 * the update if this one is not updatable. */
1803 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1804 return 0;
1805
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001806 /* Instead of sending NOT_USED, sending an empty value is preferable */
1807 if (strcmp(value, "NOT_USED") == 0) {
1808 value = "";
1809 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001810
1811 len = strlen(envname) + 1;
1812 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1813 len += strlen(value);
1814 else
1815 len += vmaxlen;
1816
1817 if (!check->envp[idx])
1818 check->envp[idx] = malloc(len + 1);
1819
1820 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001821 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001822 return 1;
1823 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001824 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001825 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001826 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001827 return 1;
1828 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001829 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001830 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001831 return 1;
1832 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001833 return 0;
1834}
Simon Horman98637e52014-06-20 12:30:16 +09001835
1836static int prepare_external_check(struct check *check)
1837{
1838 struct server *s = check->server;
1839 struct proxy *px = s->proxy;
1840 struct listener *listener = NULL, *l;
1841 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001842 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001843 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001844
1845 list_for_each_entry(l, &px->conf.listeners, by_fe)
1846 /* Use the first INET, INET6 or UNIX listener */
1847 if (l->addr.ss_family == AF_INET ||
1848 l->addr.ss_family == AF_INET6 ||
1849 l->addr.ss_family == AF_UNIX) {
1850 listener = l;
1851 break;
1852 }
1853
Simon Horman98637e52014-06-20 12:30:16 +09001854 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001855 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1856 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001857 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001858 goto err;
1859 }
Simon Horman98637e52014-06-20 12:30:16 +09001860
Cyril Bontéac92a062014-12-27 22:28:38 +01001861 check->argv = calloc(6, sizeof(char *));
1862 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001863 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001864 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001865 }
Simon Horman98637e52014-06-20 12:30:16 +09001866
1867 check->argv[0] = px->check_command;
1868
Cyril Bonté777be862014-12-02 21:21:35 +01001869 if (!listener) {
1870 check->argv[1] = strdup("NOT_USED");
1871 check->argv[2] = strdup("NOT_USED");
1872 }
1873 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001874 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001875 addr_to_str(&listener->addr, buf, sizeof(buf));
1876 check->argv[1] = strdup(buf);
1877 port_to_str(&listener->addr, buf, sizeof(buf));
1878 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001879 }
1880 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001881 const struct sockaddr_un *un;
1882
1883 un = (struct sockaddr_un *)&listener->addr;
1884 check->argv[1] = strdup(un->sun_path);
1885 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001886 }
1887 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001888 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001889 goto err;
1890 }
1891
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001892 addr_to_str(&s->addr, buf, sizeof(buf));
1893 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001894
1895 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1896 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1897 else
1898 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001899 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001900
Cyril Bontéac92a062014-12-27 22:28:38 +01001901 for (i = 0; i < 5; i++) {
1902 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001903 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001904 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001905 }
1906 }
Simon Horman98637e52014-06-20 12:30:16 +09001907
Cyril Bontéac92a062014-12-27 22:28:38 +01001908 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001909 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001910 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1911 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1912 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1913 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001914 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001915 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1916 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1917 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1918 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1919 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1920 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1921
1922 /* Ensure that we don't leave any hole in check->envp */
1923 for (i = 0; i < EXTCHK_SIZE; i++)
1924 if (!check->envp[i])
1925 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001926
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001927 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001928err:
1929 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001930 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001931 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001932 free(check->envp);
1933 check->envp = NULL;
1934 }
1935
1936 if (check->argv) {
1937 for (i = 1; i < 5; i++)
1938 free(check->argv[i]);
1939 free(check->argv);
1940 check->argv = NULL;
1941 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001942 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001943}
1944
Simon Hormanb00d17a2014-06-13 16:18:16 +09001945/*
Simon Horman98637e52014-06-20 12:30:16 +09001946 * establish a server health-check that makes use of a process.
1947 *
1948 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001949 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001950 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001951 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001952 *
1953 * Blocks and then unblocks SIGCHLD
1954 */
1955static int connect_proc_chk(struct task *t)
1956{
Cyril Bontéac92a062014-12-27 22:28:38 +01001957 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001958 struct check *check = t->context;
1959 struct server *s = check->server;
1960 struct proxy *px = s->proxy;
1961 int status;
1962 pid_t pid;
1963
Willy Tarreaue7dff022015-04-03 01:14:29 +02001964 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001965
1966 block_sigchld();
1967
1968 pid = fork();
1969 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001970 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1971 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001972 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1973 goto out;
1974 }
1975 if (pid == 0) {
1976 /* Child */
1977 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001978 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001979 int fd;
1980
1981 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1982 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1983
Willy Tarreau2555ccf2019-02-21 22:22:06 +01001984 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02001985
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001986 /* restore the initial FD limits */
1987 limit.rlim_cur = rlim_fd_cur_at_boot;
1988 limit.rlim_max = rlim_fd_max_at_boot;
1989 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
1990 getrlimit(RLIMIT_NOFILE, &limit);
1991 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
1992 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
1993 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
1994 }
1995
Simon Horman98637e52014-06-20 12:30:16 +09001996 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001997 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001998 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001999 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2000 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002001 exit(-1);
2002 }
2003
2004 /* Parent */
2005 if (check->result == CHK_RES_UNKNOWN) {
2006 if (pid_list_add(pid, t) != NULL) {
2007 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2008
2009 if (px->timeout.check && px->timeout.connect) {
2010 int t_con = tick_add(now_ms, px->timeout.connect);
2011 t->expire = tick_first(t->expire, t_con);
2012 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002013 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002014 goto out;
2015 }
2016 else {
2017 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2018 }
2019 kill(pid, SIGTERM); /* process creation error */
2020 }
2021 else
2022 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2023
2024out:
2025 unblock_sigchld();
2026 return status;
2027}
2028
2029/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002030 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002031 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002032 *
2033 * Please do NOT place any return statement in this function and only leave
2034 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002035 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002036static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002037{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002038 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002039 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002040 int rv;
2041 int ret;
2042 int expired = tick_is_expired(t->expire, now_ms);
2043
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002044 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002045 if (!(check->state & CHK_ST_INPROGRESS)) {
2046 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002047 if (!expired) /* woke up too early */
2048 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002049
2050 /* we don't send any health-checks when the proxy is
2051 * stopped, the server should not be checked or the check
2052 * is disabled.
2053 */
2054 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2055 s->proxy->state == PR_STSTOPPED)
2056 goto reschedule;
2057
2058 /* we'll initiate a new check */
2059 set_server_check_status(check, HCHK_STATUS_START, NULL);
2060
2061 check->state |= CHK_ST_INPROGRESS;
2062
Simon Hormandbf70192015-01-30 11:22:53 +09002063 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002064 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002065 /* the process was forked, we allow up to min(inter,
2066 * timeout.connect) for it to report its status, but
2067 * only when timeout.check is set as it may be to short
2068 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002069 */
2070 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2071
2072 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2073 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2074 t->expire = tick_first(t->expire, t_con);
2075 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002076 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002077 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002078 }
2079
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002080 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002081
2082 check->state &= ~CHK_ST_INPROGRESS;
2083 check_notify_failure(check);
2084
2085 /* we allow up to min(inter, timeout.connect) for a connection
2086 * to establish but only when timeout.check is set
2087 * as it may be to short for a full check otherwise
2088 */
2089 while (tick_is_expired(t->expire, now_ms)) {
2090 int t_con;
2091
2092 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2093 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2094
2095 if (s->proxy->timeout.check)
2096 t->expire = tick_first(t->expire, t_con);
2097 }
2098 }
2099 else {
2100 /* there was a test running.
2101 * First, let's check whether there was an uncaught error,
2102 * which can happen on connect timeout or error.
2103 */
2104 if (check->result == CHK_RES_UNKNOWN) {
2105 /* good connection is enough for pure TCP check */
2106 struct pid_list *elem = check->curpid;
2107 int status = HCHK_STATUS_UNKNOWN;
2108
2109 if (elem->exited) {
2110 status = elem->status; /* Save in case the process exits between use below */
2111 if (!WIFEXITED(status))
2112 check->code = -1;
2113 else
2114 check->code = WEXITSTATUS(status);
2115 if (!WIFEXITED(status) || WEXITSTATUS(status))
2116 status = HCHK_STATUS_PROCERR;
2117 else
2118 status = HCHK_STATUS_PROCOK;
2119 } else if (expired) {
2120 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002121 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002122 kill(elem->pid, SIGTERM);
2123 }
2124 set_server_check_status(check, status, NULL);
2125 }
2126
2127 if (check->result == CHK_RES_FAILED) {
2128 /* a failure or timeout detected */
2129 check_notify_failure(check);
2130 }
2131 else if (check->result == CHK_RES_CONDPASS) {
2132 /* check is OK but asks for stopping mode */
2133 check_notify_stopping(check);
2134 }
2135 else if (check->result == CHK_RES_PASSED) {
2136 /* a success was detected */
2137 check_notify_success(check);
2138 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002139 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002140 check->state &= ~CHK_ST_INPROGRESS;
2141
2142 pid_list_del(check->curpid);
2143
2144 rv = 0;
2145 if (global.spread_checks > 0) {
2146 rv = srv_getinter(check) * global.spread_checks / 100;
2147 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2148 }
2149 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2150 }
2151
2152 reschedule:
2153 while (tick_is_expired(t->expire, now_ms))
2154 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002155
2156 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002157 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002158 return t;
2159}
2160
2161/*
2162 * manages a server health-check that uses a connection. Returns
2163 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002164 *
2165 * Please do NOT place any return statement in this function and only leave
2166 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002167 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002168static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002169{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002170 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002171 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002172 struct conn_stream *cs = check->cs;
2173 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002174 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002175 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002176 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002177
Olivier Houchard0923fa42019-01-11 18:43:04 +01002178 if (check->server)
2179 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002180 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002181 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002182 if (!expired) /* woke up too early */
2183 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002184
Simon Horman671b6f02013-11-25 10:46:39 +09002185 /* we don't send any health-checks when the proxy is
2186 * stopped, the server should not be checked or the check
2187 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002188 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002189 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002190 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002191 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002192
2193 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002194 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002195
Willy Tarreau2c115e52013-12-11 19:41:16 +01002196 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002197 b_reset(&check->bi);
2198 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002199
Simon Hormandbf70192015-01-30 11:22:53 +09002200 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002201 cs = check->cs;
2202 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002203
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002204 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002205 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002206 goto out_unlock;
2207
Willy Tarreaue7dff022015-04-03 01:14:29 +02002208 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002209 /* we allow up to min(inter, timeout.connect) for a connection
2210 * to establish but only when timeout.check is set
2211 * as it may be to short for a full check otherwise
2212 */
Simon Horman4a741432013-02-23 15:35:38 +09002213 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002214
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002215 if (proxy->timeout.check && proxy->timeout.connect) {
2216 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002217 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002218 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002219
Olivier Houchard53216e72018-10-10 15:46:36 +02002220 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002221 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002222
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002223 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002224 goto reschedule;
2225
Willy Tarreaue7dff022015-04-03 01:14:29 +02002226 case SF_ERR_SRVTO: /* ETIMEDOUT */
2227 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002228 if (conn)
2229 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002230 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002231 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002232 /* should share same code than cases below */
2233 case SF_ERR_CHK_PORT:
2234 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002235 case SF_ERR_PRXCOND:
2236 case SF_ERR_RESOURCE:
2237 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002238 if (conn)
2239 conn->flags |= CO_FL_ERROR;
2240 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002241 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002242 }
2243
Willy Tarreau5a78f362012-11-23 12:47:05 +01002244 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002245 if (cs) {
2246 cs_destroy(cs);
2247 cs = check->cs = NULL;
2248 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002249 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002250
Willy Tarreau2c115e52013-12-11 19:41:16 +01002251 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002252 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002253
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002254 /* we allow up to min(inter, timeout.connect) for a connection
2255 * to establish but only when timeout.check is set
2256 * as it may be to short for a full check otherwise
2257 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002258 while (tick_is_expired(t->expire, now_ms)) {
2259 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002260
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002261 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002262 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002263
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002264 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002265 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002266 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002267 }
2268 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002269 /* there was a test running.
2270 * First, let's check whether there was an uncaught error,
2271 * which can happen on connect timeout or error.
2272 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002273 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002274 /* good connection is enough for pure TCP check */
2275 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002276 if (check->use_ssl)
2277 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002278 else
Simon Horman4a741432013-02-23 15:35:38 +09002279 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002280 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002281 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002282 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002283 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002284 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002285 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002286 }
2287
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002288 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002289 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002290 /* The check was aborted and the connection was not yet closed.
2291 * This can happen upon timeout, or when an external event such
2292 * as a failed response coupled with "observe layer7" caused the
2293 * server state to be suddenly changed.
2294 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002295 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002296 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002297 }
2298
Willy Tarreauac59f362017-10-08 11:10:19 +02002299 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002300 cs_destroy(cs);
2301 cs = check->cs = NULL;
2302 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002303 }
2304
Olivier Houchard0923fa42019-01-11 18:43:04 +01002305 if (check->server) {
2306 if (check->result == CHK_RES_FAILED) {
2307 /* a failure or timeout detected */
2308 check_notify_failure(check);
2309 }
2310 else if (check->result == CHK_RES_CONDPASS) {
2311 /* check is OK but asks for stopping mode */
2312 check_notify_stopping(check);
2313 }
2314 else if (check->result == CHK_RES_PASSED) {
2315 /* a success was detected */
2316 check_notify_success(check);
2317 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002318 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002319 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002320 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002321
Olivier Houchard0923fa42019-01-11 18:43:04 +01002322 if (check->server) {
2323 rv = 0;
2324 if (global.spread_checks > 0) {
2325 rv = srv_getinter(check) * global.spread_checks / 100;
2326 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2327 }
2328 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002329 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002330 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002331
2332 reschedule:
2333 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002334 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002335 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002336 if (check->server)
2337 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002338 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002339}
2340
Simon Horman98637e52014-06-20 12:30:16 +09002341/*
2342 * manages a server health-check. Returns
2343 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2344 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002345static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002346{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002347 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002348
2349 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002350 return process_chk_proc(t, context, state);
2351 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002352
Simon Horman98637e52014-06-20 12:30:16 +09002353}
2354
Simon Horman5c942422013-11-25 10:46:32 +09002355static int start_check_task(struct check *check, int mininter,
2356 int nbcheck, int srvpos)
2357{
2358 struct task *t;
2359 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002360 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002361 ha_alert("Starting [%s:%s] check: out of memory.\n",
2362 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002363 return 0;
2364 }
2365
2366 check->task = t;
2367 t->process = process_chk;
2368 t->context = check;
2369
Willy Tarreau1746eec2014-04-25 10:46:47 +02002370 if (mininter < srv_getinter(check))
2371 mininter = srv_getinter(check);
2372
2373 if (global.max_spread_checks && mininter > global.max_spread_checks)
2374 mininter = global.max_spread_checks;
2375
Simon Horman5c942422013-11-25 10:46:32 +09002376 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002377 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002378 check->start = now;
2379 task_queue(t);
2380
2381 return 1;
2382}
2383
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002384/*
2385 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002386 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002387 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002388static int start_checks()
2389{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002390
2391 struct proxy *px;
2392 struct server *s;
2393 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002394 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002395
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002396 /* 1- count the checkers to run simultaneously.
2397 * We also determine the minimum interval among all of those which
2398 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2399 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002400 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002401 * too short an interval for all others.
2402 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002403 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002404 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002405 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002406 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002407 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002408 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002409 }
2410 /* We need a warmup task that will be called when the server
2411 * state switches from down to up.
2412 */
2413 s->warmup = t;
2414 t->process = server_warmup;
2415 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002416 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002417 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002418 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 +01002419 }
2420
Willy Tarreaud8514a22013-12-11 21:10:14 +01002421 if (s->check.state & CHK_ST_CONFIGURED) {
2422 nbcheck++;
2423 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2424 (!mininter || mininter > srv_getinter(&s->check)))
2425 mininter = srv_getinter(&s->check);
2426 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002427
Willy Tarreaud8514a22013-12-11 21:10:14 +01002428 if (s->agent.state & CHK_ST_CONFIGURED) {
2429 nbcheck++;
2430 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2431 (!mininter || mininter > srv_getinter(&s->agent)))
2432 mininter = srv_getinter(&s->agent);
2433 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002434 }
2435 }
2436
Simon Horman4a741432013-02-23 15:35:38 +09002437 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002438 return 0;
2439
2440 srand((unsigned)time(NULL));
2441
2442 /*
2443 * 2- start them as far as possible from each others. For this, we will
2444 * start them after their interval set to the min interval divided by
2445 * the number of servers, weighted by the server's position in the list.
2446 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002447 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002448 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2449 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002450 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002451 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002452 }
2453 }
2454
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002455 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002456 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002457 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002458 if (s->check.type == PR_O2_EXT_CHK) {
2459 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002460 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002461 }
Simon Hormand60d6912013-11-25 10:46:36 +09002462 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002463 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002464 srvpos++;
2465 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002466
Simon Hormand60d6912013-11-25 10:46:36 +09002467 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002468 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002469 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002470 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002471 }
2472 srvpos++;
2473 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002474 }
2475 }
2476 return 0;
2477}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002478
2479/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002480 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002481 * The buffer MUST be terminated by a null byte before calling this function.
2482 * Sets server status appropriately. The caller is responsible for ensuring
2483 * that the buffer contains at least 13 characters. If <done> is zero, we may
2484 * return 0 to indicate that data is required to decide of a match.
2485 */
2486static int httpchk_expect(struct server *s, int done)
2487{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002488 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002489 char status_code[] = "000";
2490 char *contentptr;
2491 int crlf;
2492 int ret;
2493
2494 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2495 case PR_O2_EXP_STS:
2496 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002497 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2498 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002499
2500 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2501 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2502 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002503 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002504
2505 /* we necessarily have the response, so there are no partial failures */
2506 if (s->proxy->options2 & PR_O2_EXP_INV)
2507 ret = !ret;
2508
Simon Horman4a741432013-02-23 15:35:38 +09002509 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002510 break;
2511
2512 case PR_O2_EXP_STR:
2513 case PR_O2_EXP_RSTR:
2514 /* very simple response parser: ignore CR and only count consecutive LFs,
2515 * stop with contentptr pointing to first char after the double CRLF or
2516 * to '\0' if crlf < 2.
2517 */
2518 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002519 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002520 if (crlf >= 2)
2521 break;
2522 if (*contentptr == '\r')
2523 continue;
2524 else if (*contentptr == '\n')
2525 crlf++;
2526 else
2527 crlf = 0;
2528 }
2529
2530 /* Check that response contains a body... */
2531 if (crlf < 2) {
2532 if (!done)
2533 return 0;
2534
Simon Horman4a741432013-02-23 15:35:38 +09002535 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002536 "HTTP content check could not find a response body");
2537 return 1;
2538 }
2539
2540 /* Check that response body is not empty... */
2541 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002542 if (!done)
2543 return 0;
2544
Simon Horman4a741432013-02-23 15:35:38 +09002545 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002546 "HTTP content check found empty response body");
2547 return 1;
2548 }
2549
2550 /* Check the response content against the supplied string
2551 * or regex... */
2552 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2553 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2554 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002555 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002556
2557 /* if we don't match, we may need to wait more */
2558 if (!ret && !done)
2559 return 0;
2560
2561 if (ret) {
2562 /* content matched */
2563 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002564 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002565 "HTTP check matched unwanted content");
2566 else
Simon Horman4a741432013-02-23 15:35:38 +09002567 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002568 "HTTP content check matched");
2569 }
2570 else {
2571 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002572 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002573 "HTTP check did not match unwanted content");
2574 else
Simon Horman4a741432013-02-23 15:35:38 +09002575 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002576 "HTTP content check did not match");
2577 }
2578 break;
2579 }
2580 return 1;
2581}
2582
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002583/*
2584 * return the id of a step in a send/expect session
2585 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002586static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002587{
2588 struct tcpcheck_rule *cur = NULL, *next = NULL;
2589 int i = 0;
2590
Willy Tarreau213c6782014-10-02 14:51:02 +02002591 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002592 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002593 return 0;
2594
Simon Hormane16c1b32015-01-30 11:22:57 +09002595 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002596
2597 /* no step => first step */
2598 if (cur == NULL)
2599 return 1;
2600
2601 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002602 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002603 if (next->list.p == &cur->list)
2604 break;
2605 ++i;
2606 }
2607
2608 return i;
2609}
2610
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002611/*
2612 * return the latest known comment before (including) the given stepid
2613 * returns NULL if no comment found
2614 */
2615static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2616{
2617 struct tcpcheck_rule *cur = NULL;
2618 char *ret = NULL;
2619 int i = 0;
2620
2621 /* not even started anything yet, return latest comment found before any action */
2622 if (!check->current_step) {
2623 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2624 if (cur->action == TCPCHK_ACT_COMMENT)
2625 ret = cur->comment;
2626 else
2627 goto return_comment;
2628 }
2629 }
2630
2631 i = 1;
2632 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2633 if (cur->comment)
2634 ret = cur->comment;
2635
2636 if (i >= stepid)
2637 goto return_comment;
2638
2639 ++i;
2640 }
2641
2642 return_comment:
2643 return ret;
2644}
2645
Willy Tarreaube74b882017-10-04 16:22:49 +02002646/* proceed with next steps for the TCP checks <check>. Note that this is called
2647 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002648 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2649 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002650 *
2651 * Please do NOT place any return statement in this function and only leave
2652 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002653 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002654static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002655{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002656 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002657 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002658 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002659 struct conn_stream *cs = check->cs;
2660 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002661 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002662 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002664 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002665 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002666
Willy Tarreauef953952014-10-02 14:30:14 +02002667 /* here, we know that the check is complete or that it failed */
2668 if (check->result != CHK_RES_UNKNOWN)
2669 goto out_end_tcpcheck;
2670
2671 /* We have 4 possibilities here :
2672 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002673 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002674 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2675 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002676 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002677 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002678 * be waiting for a connection attempt to complete. conn!=NULL.
2679 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002680 *
2681 * #2 and #3 are quite similar, we want both the connection and the
2682 * handshake to complete before going any further. Thus we must always
2683 * wait for a connection to complete unless we're before and existing
2684 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002685 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002686
2687 /* find first rule and skip comments */
2688 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2689 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2690 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2691
Willy Tarreau00149122017-10-04 18:05:01 +02002692 if ((check->current_step || &next->list == head) &&
2693 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002694 /* we allow up to min(inter, timeout.connect) for a connection
2695 * to establish but only when timeout.check is set
2696 * as it may be to short for a full check otherwise
2697 */
2698 while (tick_is_expired(t->expire, now_ms)) {
2699 int t_con;
2700
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002701 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002702 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2703
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002704 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002705 t->expire = tick_first(t->expire, t_con);
2706 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002707 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002708 }
2709
Willy Tarreauef953952014-10-02 14:30:14 +02002710 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002711 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002712 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002713 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002714 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002715
Willy Tarreau213c6782014-10-02 14:51:02 +02002716 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002717 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002718 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002719 b_reset(&check->bo);
2720 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002721 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002722 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002723 if (proxy->timeout.check)
2724 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002725 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002726
Willy Tarreauabca5b62013-12-06 14:19:25 +01002727 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002728 /* We have to try to flush the output buffer before reading, at
2729 * the end, or if we're about to send a string that does not fit
2730 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002731 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002732 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002733 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002734 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002735 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002736 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002737 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002738
Olivier Houcharded0f2072018-08-16 15:41:52 +02002739 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002740 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002741
2742 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002743 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002744 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002745 goto out_end_tcpcheck;
2746 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002747 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002748 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002749 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002750 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002751 goto out;
2752 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002753 }
2754
Willy Tarreau263013d2015-05-13 11:59:14 +02002755 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002756 break;
2757
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002758 /* have 'next' point to the next rule or NULL if we're on the
2759 * last one, connect() needs this.
2760 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002761 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002762
2763 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002764 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002765 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002766
2767 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002768 if (&next->list == head)
2769 next = NULL;
2770
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002771 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2772 struct protocol *proto;
2773 struct xprt_ops *xprt;
2774
Willy Tarreau00149122017-10-04 18:05:01 +02002775 /* For a connect action we'll create a new connection.
2776 * We may also have to kill a previous one. But we don't
2777 * want to leave *without* a connection if we came here
2778 * from the connection layer, hence with a connection.
2779 * Thus we'll proceed in the following order :
2780 * 1: close but not release previous connection
2781 * 2: try to get a new connection
2782 * 3: release and replace the old one on success
2783 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002784 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002785 cs_close(check->cs);
2786 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002787 }
2788
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002789 /* mark the step as started */
2790 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002791
2792 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002793 cs = cs_new(NULL);
2794 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002795 step = tcpcheck_get_step_id(check);
2796 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2797 comment = tcpcheck_get_step_comment(check, step);
2798 if (comment)
2799 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002800 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2801 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002802 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002803 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002804 }
2805
Willy Tarreauac59f362017-10-08 11:10:19 +02002806 if (check->cs)
2807 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002808
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002809 check->cs = cs;
2810 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002811 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002812 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002813 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002814
2815 /* no client address */
2816 clear_addr(&conn->addr.from);
2817
Simon Horman41f58762015-01-30 11:22:56 +09002818 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002819 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002820 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002821 }
2822 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002823 /* we'll connect to the addr on the server */
2824 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002825 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002826 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002827
2828 /* port */
2829 if (check->current_step->port)
2830 set_host_port(&conn->addr.to, check->current_step->port);
2831 else if (check->port)
2832 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002833 else if (s->svc_port)
2834 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002835
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002836 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002837 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002838 }
2839 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002840 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002841 }
Willy Tarreaube373152018-09-06 11:45:30 +02002842
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002843 conn_prepare(conn, proto, xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01002844 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0)
2845 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02002846 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002847
Willy Tarreaue7dff022015-04-03 01:14:29 +02002848 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002849 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002850 ret = proto->connect(conn,
Olivier Houchardfdcb0072019-05-06 18:32:29 +02002851 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 +01002852 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2853 conn->send_proxy_ofs = 1;
2854 conn->flags |= CO_FL_SEND_PROXY;
2855 }
2856
2857 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002858 * - SF_ERR_NONE if everything's OK
2859 * - SF_ERR_SRVTO if there are no more servers
2860 * - SF_ERR_SRVCL if the connection was refused by the server
2861 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2862 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2863 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002864 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002865 * Note that we try to prevent the network stack from sending the ACK during the
2866 * connect() when a pure TCP check is used (without PROXY protocol).
2867 */
2868 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002869 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002870 /* we allow up to min(inter, timeout.connect) for a connection
2871 * to establish but only when timeout.check is set
2872 * as it may be to short for a full check otherwise
2873 */
2874 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2875
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002876 if (proxy->timeout.check && proxy->timeout.connect) {
2877 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002878 t->expire = tick_first(t->expire, t_con);
2879 }
2880 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002881 case SF_ERR_SRVTO: /* ETIMEDOUT */
2882 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002883 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002884 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002885 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002886 comment = tcpcheck_get_step_comment(check, step);
2887 if (comment)
2888 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002889 set_server_check_status(check, HCHK_STATUS_L4CON,
2890 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002891 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002892 case SF_ERR_PRXCOND:
2893 case SF_ERR_RESOURCE:
2894 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002895 step = tcpcheck_get_step_id(check);
2896 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002897 comment = tcpcheck_get_step_comment(check, step);
2898 if (comment)
2899 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002900 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2901 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002902 goto out_end_tcpcheck;
2903 }
2904
2905 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002906 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002907
2908 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002909 while (&check->current_step->list != head &&
2910 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002911 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002912
Willy Tarreauf2c87352015-05-13 12:08:21 +02002913 if (&check->current_step->list == head)
2914 break;
2915
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002916 } /* end 'connect' */
2917 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2918 /* mark the step as started */
2919 check->last_started_step = check->current_step;
2920
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002921 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002922 if (*b_head(&check->bi) != '\0') {
2923 *b_head(&check->bi) = '\0';
2924 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002925 }
2926
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002927 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002928 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002929 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002930 goto out_end_tcpcheck;
2931 }
2932
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002933 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002934 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002935 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002936 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002937 set_server_check_status(check, HCHK_STATUS_L7RSP,
2938 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002939 goto out_end_tcpcheck;
2940 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002941
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002942 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002943 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002944 continue;
2945
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002946 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2947 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002948
Willy Tarreauabca5b62013-12-06 14:19:25 +01002949 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002950 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002951
2952 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002953 while (&check->current_step->list != head &&
2954 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002955 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002956
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002957 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002958 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002959 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002960 goto out_end_tcpcheck;
2961
Olivier Houchard511efea2018-08-16 15:30:32 +02002962 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002963 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002964 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002965 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002966 /* Report network errors only if we got no other data. Otherwise
2967 * we'll let the upper layers decide whether the response is OK
2968 * or not. It is very common that an RST sent by the server is
2969 * reported as an error just after the last data chunk.
2970 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002971 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002972 goto out_end_tcpcheck;
2973 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002974 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002975 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002976 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002977 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002978 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002979 }
2980
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002981 /* mark the step as started */
2982 check->last_started_step = check->current_step;
2983
2984
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002985 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002986 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002987 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002988 if (b_data(&check->bi) < b_size(&check->bi)) {
2989 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002990 }
2991 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002992 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002993 done = 1; /* buffer full, don't wait for more data */
2994 }
2995
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002996 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002997
2998 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002999 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003000 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003001 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003002
3003 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003004 step = tcpcheck_get_step_id(check);
3005 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003006 comment = tcpcheck_get_step_comment(check, step);
3007 if (comment)
3008 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003009 set_server_check_status(check, HCHK_STATUS_L7RSP,
3010 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003011
3012 goto out_end_tcpcheck;
3013 }
3014
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003015 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003016 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003017
Willy Tarreaua970c282013-12-06 12:47:19 +01003018 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003019 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003020 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 +02003021 else if (check->current_step->expect_regex != NULL)
3022 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003023
3024 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003025 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003026
3027 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003028 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003029 if (ret) {
3030 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003031 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003032 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003033 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003034 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003035 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003036 }
3037 else {
3038 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003039 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003040 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003041 comment = tcpcheck_get_step_comment(check, step);
3042 if (comment)
3043 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003044 set_server_check_status(check, HCHK_STATUS_L7RSP,
3045 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003046 goto out_end_tcpcheck;
3047 }
3048 /* matched and was supposed to => OK, next step */
3049 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003050 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003051 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003052
3053 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003054 while (&check->current_step->list != head &&
3055 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003056 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003057
Willy Tarreauf2c87352015-05-13 12:08:21 +02003058 if (&check->current_step->list == head)
3059 break;
3060
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003061 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003062 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003063 }
3064 }
3065 else {
3066 /* not matched */
3067 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003068 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003069 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003070 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003071
3072 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003073 while (&check->current_step->list != head &&
3074 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003075 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003076
Willy Tarreauf2c87352015-05-13 12:08:21 +02003077 if (&check->current_step->list == head)
3078 break;
3079
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003080 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003081 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003082 }
3083 /* not matched but was supposed to => ERROR */
3084 else {
3085 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003086 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003087 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003088 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003089 }
3090 else {
3091 /* we were looking for a regex */
3092 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003093 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003094 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003095 comment = tcpcheck_get_step_comment(check, step);
3096 if (comment)
3097 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003098 set_server_check_status(check, HCHK_STATUS_L7RSP,
3099 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003100 goto out_end_tcpcheck;
3101 }
3102 }
3103 } /* end expect */
3104 } /* end loop over double chained step list */
3105
Baptiste Assmann248f1172018-03-01 21:49:01 +01003106 /* don't do anything until the connection is established */
3107 if (!(conn->flags & CO_FL_CONNECTED)) {
3108 /* update expire time, should be done by process_chk */
3109 /* we allow up to min(inter, timeout.connect) for a connection
3110 * to establish but only when timeout.check is set
3111 * as it may be to short for a full check otherwise
3112 */
3113 while (tick_is_expired(t->expire, now_ms)) {
3114 int t_con;
3115
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003116 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003117 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3118
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003119 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003120 t->expire = tick_first(t->expire, t_con);
3121 }
3122 goto out;
3123 }
3124
Willy Tarreau263013d2015-05-13 11:59:14 +02003125 /* We're waiting for some I/O to complete, we've reached the end of the
3126 * rules, or both. Do what we have to do, otherwise we're done.
3127 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003128 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003129 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3130 goto out_end_tcpcheck;
3131 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003132
Willy Tarreau53c5a042015-05-13 11:38:17 +02003133 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003134 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003135 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003136 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003137
3138 out_end_tcpcheck:
3139 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003140 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003141 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003142
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003143 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003144 check->current_step = NULL;
3145
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003146 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003147 conn->flags |= CO_FL_ERROR;
3148
Christopher Fauletb6102852017-11-28 10:06:29 +01003149 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003150 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003151}
3152
Simon Hormanb1900d52015-01-30 11:22:54 +09003153const char *init_check(struct check *check, int type)
3154{
3155 check->type = type;
3156
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003157 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3158 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003159
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003160 check->bi.area = calloc(check->bi.size, sizeof(char));
3161 check->bo.area = calloc(check->bo.size, sizeof(char));
3162
3163 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003164 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003165
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003166 check->wait_list.task = tasklet_new();
3167 if (!check->wait_list.task)
3168 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003169 check->wait_list.events = 0;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003170 check->wait_list.task->process = event_srv_chk_io;
3171 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003172 return NULL;
3173}
3174
Simon Hormanbfb5d332015-01-30 11:22:55 +09003175void free_check(struct check *check)
3176{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003177 free(check->bi.area);
3178 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003179 if (check->cs) {
3180 free(check->cs->conn);
3181 check->cs->conn = NULL;
3182 cs_free(check->cs);
3183 check->cs = NULL;
3184 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003185}
3186
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003187void email_alert_free(struct email_alert *alert)
3188{
3189 struct tcpcheck_rule *rule, *back;
3190
3191 if (!alert)
3192 return;
3193
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003194 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3195 LIST_DEL(&rule->list);
3196 free(rule->comment);
3197 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003198 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003199 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003200 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003201 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003202}
3203
Olivier Houchard9f6af332018-05-25 14:04:04 +02003204static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003205{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003206 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003207 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003208 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003209
3210 q = container_of(check, typeof(*q), check);
3211
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003212 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003213 while (1) {
3214 if (!(check->state & CHK_ST_ENABLED)) {
3215 if (LIST_ISEMPTY(&q->email_alerts)) {
3216 /* All alerts processed, queue the task */
3217 t->expire = TICK_ETERNITY;
3218 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003219 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003220 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003221
3222 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003223 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003224 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003225 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003226 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003227 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003228 }
3229
Olivier Houchard9f6af332018-05-25 14:04:04 +02003230 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003231 if (check->state & CHK_ST_INPROGRESS)
3232 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003233
3234 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3235 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003236 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003237 check->server = NULL;
3238 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003239 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003240 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003241 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003242 return t;
3243}
3244
Christopher Faulet0108bb32017-10-20 21:34:32 +02003245/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3246 *
3247 * The function returns 1 in success case, otherwise, it returns 0 and err is
3248 * filled.
3249 */
3250int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003251{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003252 struct mailer *mailer;
3253 struct email_alertq *queues;
3254 const char *err_str;
3255 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003256
Christopher Faulet0108bb32017-10-20 21:34:32 +02003257 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3258 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003259 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003260 }
3261
Christopher Faulet0108bb32017-10-20 21:34:32 +02003262 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3263 struct email_alertq *q = &queues[i];
3264 struct check *check = &q->check;
3265 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003266
3267 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003268 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003269 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003270 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003271 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003272 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003273 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3274 memprintf(err, "%s", err_str);
3275 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003276 }
3277
3278 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003279 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003280 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003281
Emeric Brunc60def82017-09-27 14:59:38 +02003282 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003283 memprintf(err, "out of memory while allocating mailer alerts task");
3284 goto error;
3285 }
3286
3287 check->task = t;
3288 t->process = process_email_alert;
3289 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003290
Christopher Faulet0108bb32017-10-20 21:34:32 +02003291 /* check this in one ms */
3292 t->expire = TICK_ETERNITY;
3293 check->start = now;
3294 task_queue(t);
3295 }
3296
3297 mls->users++;
3298 free(p->email_alert.mailers.name);
3299 p->email_alert.mailers.m = mls;
3300 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003301 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003302
3303 error:
3304 for (i = 0; i < mls->count; i++) {
3305 struct email_alertq *q = &queues[i];
3306 struct check *check = &q->check;
3307
Willy Tarreauf6562792019-05-07 19:05:35 +02003308 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003309 free_check(check);
3310 }
3311 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003312 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003313 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003314}
3315
3316
3317static int add_tcpcheck_expect_str(struct list *list, const char *str)
3318{
3319 struct tcpcheck_rule *tcpcheck;
3320
Willy Tarreaubafbe012017-11-24 17:34:44 +01003321 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003322 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003323 memset(tcpcheck, 0, sizeof(*tcpcheck));
3324 tcpcheck->action = TCPCHK_ACT_EXPECT;
3325 tcpcheck->string = strdup(str);
3326 tcpcheck->expect_regex = NULL;
3327 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003328 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003329 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003330 return 0;
3331 }
3332
3333 LIST_ADDQ(list, &tcpcheck->list);
3334 return 1;
3335}
3336
3337static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3338{
3339 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003340 const char *in;
3341 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003342 int i;
3343
Willy Tarreaubafbe012017-11-24 17:34:44 +01003344 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003345 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003346 memset(tcpcheck, 0, sizeof(*tcpcheck));
3347 tcpcheck->action = TCPCHK_ACT_SEND;
3348 tcpcheck->expect_regex = NULL;
3349 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003350 tcpcheck->string_len = 0;
3351 for (i = 0; strs[i]; i++)
3352 tcpcheck->string_len += strlen(strs[i]);
3353
3354 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3355 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003356 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003357 return 0;
3358 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003359
Willy Tarreau64345aa2016-08-10 19:29:09 +02003360 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003361 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003362 for (in = strs[i]; (*dst = *in++); dst++);
3363 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003364
3365 LIST_ADDQ(list, &tcpcheck->list);
3366 return 1;
3367}
3368
Christopher Faulet0108bb32017-10-20 21:34:32 +02003369static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3370 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003371{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003372 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003373 struct tcpcheck_rule *tcpcheck;
3374 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003375
Willy Tarreaubafbe012017-11-24 17:34:44 +01003376 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003377 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003378 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003379 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003380 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003381
Willy Tarreaubafbe012017-11-24 17:34:44 +01003382 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003383 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003384 memset(tcpcheck, 0, sizeof(*tcpcheck));
3385 tcpcheck->action = TCPCHK_ACT_CONNECT;
3386 tcpcheck->comment = NULL;
3387 tcpcheck->string = NULL;
3388 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003389 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3390
3391 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3392 goto error;
3393
3394 {
3395 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3396 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3397 goto error;
3398 }
3399
3400 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3401 goto error;
3402
3403 {
3404 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3405 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3406 goto error;
3407 }
3408
3409 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3410 goto error;
3411
3412 {
3413 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3414 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3415 goto error;
3416 }
3417
3418 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3419 goto error;
3420
3421 {
3422 const char * const strs[2] = { "DATA\r\n" };
3423 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3424 goto error;
3425 }
3426
3427 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3428 goto error;
3429
3430 {
3431 struct tm tm;
3432 char datestr[48];
3433 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003434 "From: ", p->email_alert.from, "\r\n",
3435 "To: ", p->email_alert.to, "\r\n",
3436 "Date: ", datestr, "\r\n",
3437 "Subject: [HAproxy Alert] ", msg, "\r\n",
3438 "\r\n",
3439 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003440 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003441 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003442 NULL
3443 };
3444
3445 get_localtime(date.tv_sec, &tm);
3446
3447 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3448 goto error;
3449 }
3450
3451 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3452 goto error;
3453 }
3454
3455 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3456 goto error;
3457
3458 {
3459 const char * const strs[2] = { "QUIT\r\n" };
3460 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3461 goto error;
3462 }
3463
3464 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3465 goto error;
3466
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003467 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003468 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003469 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003470 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003471 return 1;
3472
3473error:
3474 email_alert_free(alert);
3475 return 0;
3476}
3477
Christopher Faulet0108bb32017-10-20 21:34:32 +02003478static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003479{
3480 int i;
3481 struct mailer *mailer;
3482
3483 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3484 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003485 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003486 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003487 return;
3488 }
3489 }
3490
3491 return;
3492}
3493
3494/*
3495 * Send email alert if configured.
3496 */
Simon Horman64e34162015-02-06 11:11:57 +09003497void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003498{
3499 va_list argp;
3500 char buf[1024];
3501 int len;
3502 struct proxy *p = s->proxy;
3503
Christopher Faulet0108bb32017-10-20 21:34:32 +02003504 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003505 return;
3506
3507 va_start(argp, format);
3508 len = vsnprintf(buf, sizeof(buf), format, argp);
3509 va_end(argp);
3510
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003511 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003512 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003513 return;
3514 }
3515
Christopher Faulet0108bb32017-10-20 21:34:32 +02003516 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003517}
3518
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003519/*
3520 * Return value:
3521 * the port to be used for the health check
3522 * 0 in case no port could be found for the check
3523 */
3524int srv_check_healthcheck_port(struct check *chk)
3525{
3526 int i = 0;
3527 struct server *srv = NULL;
3528
3529 srv = chk->server;
3530
3531 /* If neither a port nor an addr was specified and no check transport
3532 * layer is forced, then the transport layer used by the checks is the
3533 * same as for the production traffic. Otherwise we use raw_sock by
3534 * default, unless one is specified.
3535 */
3536 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003537 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003538 chk->send_proxy |= (srv->pp_opts);
3539 }
3540
3541 /* by default, we use the health check port ocnfigured */
3542 if (chk->port > 0)
3543 return chk->port;
3544
3545 /* try to get the port from check_core.addr if check.port not set */
3546 i = get_host_port(&chk->addr);
3547 if (i > 0)
3548 return i;
3549
3550 /* try to get the port from server address */
3551 /* prevent MAPPORTS from working at this point, since checks could
3552 * not be performed in such case (MAPPORTS impose a relative ports
3553 * based on live traffic)
3554 */
3555 if (srv->flags & SRV_F_MAPPORTS)
3556 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003557
3558 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003559 if (i > 0)
3560 return i;
3561
3562 return 0;
3563}
3564
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003565REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003566
Willy Tarreaubd741542010-03-16 18:46:54 +01003567/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003568 * Local variables:
3569 * c-indent-level: 8
3570 * c-basic-offset: 8
3571 * End:
3572 */