blob: 6d7b6aa271df89e2751c865346d10dadce6316d1 [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>
25#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040026#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090027#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020028#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020029#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020030#include <arpa/inet.h>
31
Willy Tarreauc7e42382012-08-24 19:22:53 +020032#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020033#include <common/compat.h>
34#include <common/config.h>
35#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020036#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020037#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020038#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020039
40#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020041#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010042#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043
44#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020045#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010046#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020047#include <proto/fd.h>
48#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020049#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020050#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020051#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010052#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010053#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010054#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020055#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020056#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010057#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020058#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020059#include <proto/log.h>
60#include <proto/dns.h>
61#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020062
Olivier Houchard9130a962017-10-17 17:33:43 +020063#ifdef USE_OPENSSL
64#include <proto/ssl_sock.h>
65#endif /* USE_OPENSSL */
66
Willy Tarreaubd741542010-03-16 18:46:54 +010067static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090068static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020069static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020070static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020071static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020072static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020073static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010074
Willy Tarreau8ceae722018-11-26 11:58:30 +010075DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
76DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020077
78
Simon Horman63a4a822012-03-19 07:24:41 +090079static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010080 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
81 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020082 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020083
Willy Tarreau23964182014-05-20 20:56:30 +020084 /* Below we have finished checks */
85 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020089
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010090 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
91 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
92 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020093
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010094 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
95 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
96 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020097
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010098 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
99 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200101 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200102
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100103 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
104 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
105 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900106
107 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
108 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200109 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200110};
111
Cyril Bontéac92a062014-12-27 22:28:38 +0100112const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
113 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
122 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
123 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
124};
125
Simon Horman63a4a822012-03-19 07:24:41 +0900126static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100127 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
128
129 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
130 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
131
132 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
133 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
134 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
135 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
136
137 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
138 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
139 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
140};
141
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200142/*
143 * Convert check_status code to description
144 */
145const char *get_check_status_description(short check_status) {
146
147 const char *desc;
148
149 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200150 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200151 else
152 desc = NULL;
153
154 if (desc && *desc)
155 return desc;
156 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200157 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200158}
159
160/*
161 * Convert check_status code to short info
162 */
163const char *get_check_status_info(short check_status) {
164
165 const char *info;
166
167 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200168 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200169 else
170 info = NULL;
171
172 if (info && *info)
173 return info;
174 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200175 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200176}
177
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100178const char *get_analyze_status(short analyze_status) {
179
180 const char *desc;
181
182 if (analyze_status < HANA_STATUS_SIZE)
183 desc = analyze_statuses[analyze_status].desc;
184 else
185 desc = NULL;
186
187 if (desc && *desc)
188 return desc;
189 else
190 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
191}
192
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200193/*
Simon Horman4a741432013-02-23 15:35:38 +0900194 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200195 * an adequate CHK_RES_* value. The new check->health is computed based
196 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200197 *
198 * Show information in logs about failed health check if server is UP
199 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200200 */
Simon Horman4a741432013-02-23 15:35:38 +0900201static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100202{
Simon Horman4a741432013-02-23 15:35:38 +0900203 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200204 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200205 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900206
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200207 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100208 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900209 check->desc[0] = '\0';
210 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200211 return;
212 }
213
Simon Horman4a741432013-02-23 15:35:38 +0900214 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200215 return;
216
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200217 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900218 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
219 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200220 } else
Simon Horman4a741432013-02-23 15:35:38 +0900221 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200222
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200224 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900225 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200226
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100227 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900228 check->duration = -1;
229 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200230 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900231 check->duration = tv_ms_elapsed(&check->start, &now);
232 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200233 }
234
Willy Tarreau23964182014-05-20 20:56:30 +0200235 /* no change is expected if no state change occurred */
236 if (check->result == CHK_RES_NEUTRAL)
237 return;
238
Olivier Houchard0923fa42019-01-11 18:43:04 +0100239 /* If the check was really just sending a mail, it won't have an
240 * associated server, so we're done now.
241 */
242 if (!s)
243 return;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200244 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200245
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200246 switch (check->result) {
247 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200248 /* Failure to connect to the agent as a secondary check should not
249 * cause the server to be marked down.
250 */
251 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900252 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200253 (check->health > 0)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200254 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200255 report = 1;
256 check->health--;
257 if (check->health < check->rise)
258 check->health = 0;
259 }
260 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200261
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200262 case CHK_RES_PASSED:
263 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
264 if ((check->health < check->rise + check->fall - 1) &&
265 (check->result == CHK_RES_PASSED || check->health > 0)) {
266 report = 1;
267 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200268
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200269 if (check->health >= check->rise)
270 check->health = check->rise + check->fall - 1; /* OK now */
271 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200272
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200273 /* clear consecutive_errors if observing is enabled */
274 if (s->onerror)
275 s->consecutive_errors = 0;
276 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100277
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200278 default:
279 break;
280 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200281
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200282 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
283 (status != prev_status || report)) {
284 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200285 "%s check for %sserver %s/%s %s%s",
286 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200287 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100288 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100289 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200290 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200291
Emeric Brun5a133512017-10-19 14:42:30 +0200292 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200293
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100294 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200295 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
296 (check->health >= check->rise) ? check->fall : check->rise,
297 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200298
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200299 ha_warning("%s.\n", trash.area);
300 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
301 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200302 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200303}
304
Willy Tarreau4eec5472014-05-20 22:32:27 +0200305/* Marks the check <check>'s server down if the current check is already failed
306 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200307 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200308static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200309{
Simon Horman4a741432013-02-23 15:35:38 +0900310 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900311
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200312 /* The agent secondary check should only cause a server to be marked
313 * as down if check->status is HCHK_STATUS_L7STS, which indicates
314 * that the agent returned "fail", "stopped" or "down".
315 * The implication here is that failure to connect to the agent
316 * as a secondary check should not cause the server to be marked
317 * down. */
318 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
319 return;
320
Willy Tarreau4eec5472014-05-20 22:32:27 +0200321 if (check->health > 0)
322 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100323
Willy Tarreau4eec5472014-05-20 22:32:27 +0200324 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200325 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200326}
327
Willy Tarreauaf549582014-05-16 17:37:50 +0200328/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200329 * it isn't in maintenance, it is not tracking a down server and other checks
330 * comply. The rule is simple : by default, a server is up, unless any of the
331 * following conditions is true :
332 * - health check failed (check->health < rise)
333 * - agent check failed (agent->health < rise)
334 * - the server tracks a down server (track && track->state == STOPPED)
335 * Note that if the server has a slowstart, it will switch to STARTING instead
336 * of RUNNING. Also, only the health checks support the nolb mode, so the
337 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200338 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200339static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200340{
Simon Horman4a741432013-02-23 15:35:38 +0900341 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100342
Emeric Brun52a91d32017-08-31 14:41:55 +0200343 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200344 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100345
Emeric Brun52a91d32017-08-31 14:41:55 +0200346 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200347 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100348
Willy Tarreau3e048382014-05-21 10:30:54 +0200349 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
350 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100351
Willy Tarreau3e048382014-05-21 10:30:54 +0200352 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
353 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200354
Emeric Brun52a91d32017-08-31 14:41:55 +0200355 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200356 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100357
Emeric Brun5a133512017-10-19 14:42:30 +0200358 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100359}
360
Willy Tarreaudb58b792014-05-21 13:57:23 +0200361/* Marks the check <check> as valid and tries to set its server into stopping mode
362 * if it was running or starting, and provided it isn't in maintenance and other
363 * checks comply. The conditions for the server to be marked in stopping mode are
364 * the same as for it to be turned up. Also, only the health checks support the
365 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200366 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200367static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200368{
Simon Horman4a741432013-02-23 15:35:38 +0900369 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100370
Emeric Brun52a91d32017-08-31 14:41:55 +0200371 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200372 return;
373
Willy Tarreaudb58b792014-05-21 13:57:23 +0200374 if (check->state & CHK_ST_AGENT)
375 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100376
Emeric Brun52a91d32017-08-31 14:41:55 +0200377 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200378 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100379
Willy Tarreaudb58b792014-05-21 13:57:23 +0200380 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
381 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100382
Willy Tarreaudb58b792014-05-21 13:57:23 +0200383 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
384 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100385
Willy Tarreaub26881a2017-12-23 11:16:49 +0100386 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100387}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200388
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100389/* note: use health_adjust() only, which first checks that the observe mode is
390 * enabled.
391 */
392void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100393{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100394 int failed;
395 int expire;
396
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100397 if (s->observe >= HANA_OBS_SIZE)
398 return;
399
Willy Tarreaubb956662013-01-24 00:37:39 +0100400 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100401 return;
402
403 switch (analyze_statuses[status].lr[s->observe - 1]) {
404 case 1:
405 failed = 1;
406 break;
407
408 case 2:
409 failed = 0;
410 break;
411
412 default:
413 return;
414 }
415
416 if (!failed) {
417 /* good: clear consecutive_errors */
418 s->consecutive_errors = 0;
419 return;
420 }
421
Christopher Faulet29f77e82017-06-08 14:04:45 +0200422 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100423
424 if (s->consecutive_errors < s->consecutive_errors_limit)
425 return;
426
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100427 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
428 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100429
430 switch (s->onerror) {
431 case HANA_ONERR_FASTINTER:
432 /* force fastinter - nothing to do here as all modes force it */
433 break;
434
435 case HANA_ONERR_SUDDTH:
436 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900437 if (s->check.health > s->check.rise)
438 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100439
440 /* no break - fall through */
441
442 case HANA_ONERR_FAILCHK:
443 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200444 set_server_check_status(&s->check, HCHK_STATUS_HANA,
445 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200446 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100447 break;
448
449 case HANA_ONERR_MARKDWN:
450 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900451 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200452 set_server_check_status(&s->check, HCHK_STATUS_HANA,
453 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200454 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100455 break;
456
457 default:
458 /* write a warning? */
459 break;
460 }
461
462 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200463 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100464
Simon Horman66183002013-02-23 10:16:43 +0900465 if (s->check.fastinter) {
466 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300467 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200468 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300469 /* requeue check task with new expire */
470 task_queue(s->check.task);
471 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100472 }
473}
474
Willy Tarreaua1dab552014-04-14 15:04:54 +0200475static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100476{
477 int sv_state;
478 int ratio;
479 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800480 char addr[46];
481 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100482 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
483 "UP %d/%d", "UP",
484 "NOLB %d/%d", "NOLB",
485 "no check" };
486
487 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
488 hlen += 24;
489
Willy Tarreauff5ae352013-12-11 20:36:34 +0100490 if (!(s->check.state & CHK_ST_ENABLED))
491 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200492 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900493 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100494 sv_state = 3; /* UP */
495 else
496 sv_state = 2; /* going down */
497
Emeric Brun52a91d32017-08-31 14:41:55 +0200498 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100499 sv_state += 2;
500 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900501 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100502 sv_state = 1; /* going up */
503 else
504 sv_state = 0; /* DOWN */
505 }
506
Willy Tarreaua1dab552014-04-14 15:04:54 +0200507 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100508 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200509 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
510 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100511
Joseph Lynch514061c2015-01-15 17:52:59 -0800512 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100513 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
514 snprintf(port, sizeof(port), "%u", s->svc_port);
515 else
516 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800517
518 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
519 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100520 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200521 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100522 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
523 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
524 s->nbpend);
525
Emeric Brun52a91d32017-08-31 14:41:55 +0200526 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100527 now.tv_sec < s->last_change + s->slowstart &&
528 now.tv_sec >= s->last_change) {
529 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200530 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100531 }
532
533 buffer[hlen++] = '\r';
534 buffer[hlen++] = '\n';
535
536 return hlen;
537}
538
Willy Tarreau20a18342013-12-05 00:31:46 +0100539/* Check the connection. If an error has already been reported or the socket is
540 * closed, keep errno intact as it is supposed to contain the valid error code.
541 * If no error is reported, check the socket's error queue using getsockopt().
542 * Warning, this must be done only once when returning from poll, and never
543 * after an I/O error was attempted, otherwise the error queue might contain
544 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
545 * socket. Returns non-zero if an error was reported, zero if everything is
546 * clean (including a properly closed socket).
547 */
548static int retrieve_errno_from_socket(struct connection *conn)
549{
550 int skerr;
551 socklen_t lskerr = sizeof(skerr);
552
553 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
554 return 1;
555
Willy Tarreau3c728722014-01-23 13:50:42 +0100556 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100557 return 0;
558
Willy Tarreau585744b2017-08-24 14:31:19 +0200559 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100560 errno = skerr;
561
562 if (errno == EAGAIN)
563 errno = 0;
564
565 if (!errno) {
566 /* we could not retrieve an error, that does not mean there is
567 * none. Just don't change anything and only report the prior
568 * error if any.
569 */
570 if (conn->flags & CO_FL_ERROR)
571 return 1;
572 else
573 return 0;
574 }
575
576 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
577 return 1;
578}
579
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100580/* Try to collect as much information as possible on the connection status,
581 * and adjust the server status accordingly. It may make use of <errno_bck>
582 * if non-null when the caller is absolutely certain of its validity (eg:
583 * checked just after a syscall). If the caller doesn't have a valid errno,
584 * it can pass zero, and retrieve_errno_from_socket() will be called to try
585 * to extract errno from the socket. If no error is reported, it will consider
586 * the <expired> flag. This is intended to be used when a connection error was
587 * reported in conn->flags or when a timeout was reported in <expired>. The
588 * function takes care of not updating a server status which was already set.
589 * All situations where at least one of <expired> or CO_FL_ERROR are set
590 * produce a status.
591 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200592static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100593{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200594 struct conn_stream *cs = check->cs;
595 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100596 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200597 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200598 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200599 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100600
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100601 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100602 return;
603
604 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200605 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100606 retrieve_errno_from_socket(conn);
607
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200608 if (conn && !(conn->flags & CO_FL_ERROR) &&
609 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100610 return;
611
612 /* we'll try to build a meaningful error message depending on the
613 * context of the error possibly present in conn->err_code, and the
614 * socket error possibly collected above. This is useful to know the
615 * exact step of the L6 layer (eg: SSL handshake).
616 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200617 chk = get_trash_chunk();
618
619 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900620 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200621 if (!step)
622 chunk_printf(chk, " at initial connection step of tcp-check");
623 else {
624 chunk_printf(chk, " at step %d of tcp-check", step);
625 /* we were looking for a string */
626 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
627 if (check->last_started_step->port)
628 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
629 else
630 chunk_appendf(chk, " (connect)");
631 }
632 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
633 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200634 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200635 else if (check->last_started_step->expect_regex)
636 chunk_appendf(chk, " (expect regex)");
637 }
638 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
639 chunk_appendf(chk, " (send)");
640 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200641
642 comment = tcpcheck_get_step_comment(check, step);
643 if (comment)
644 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200645 }
646 }
647
Willy Tarreau00149122017-10-04 18:05:01 +0200648 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100649 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200650 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
651 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100652 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200653 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
654 chk->area);
655 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100656 }
657 else {
658 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200659 chunk_printf(&trash, "%s%s", strerror(errno),
660 chk->area);
661 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100662 }
663 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200664 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100665 }
666 }
667
Willy Tarreau00149122017-10-04 18:05:01 +0200668 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200669 /* NOTE: this is reported after <fall> tries */
670 chunk_printf(chk, "No port available for the TCP connection");
671 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
672 }
673
Willy Tarreau00149122017-10-04 18:05:01 +0200674 if (!conn) {
675 /* connection allocation error before the connection was established */
676 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
677 }
678 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100679 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200680 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100681 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
682 else if (expired)
683 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200684
685 /*
686 * might be due to a server IP change.
687 * Let's trigger a DNS resolution if none are currently running.
688 */
Olivier Houchard0923fa42019-01-11 18:43:04 +0100689 if (check->server)
690 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200691
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100692 }
693 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
694 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200695 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100696 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
697 else if (expired)
698 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
699 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200700 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100701 /* I/O error after connection was established and before we could diagnose */
702 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
703 }
704 else if (expired) {
705 /* connection established but expired check */
706 if (check->type == PR_O2_SSL3_CHK)
707 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
708 else /* HTTP, SMTP, ... */
709 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
710 }
711
712 return;
713}
714
Olivier Houchard5c110b92018-08-14 17:04:58 +0200715/* This function checks if any I/O is wanted, and if so, attempts to do so */
716static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200717{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200718 struct check *check = ctx;
719 struct conn_stream *cs = check->cs;
Olivier Houchard0923fa42019-01-11 18:43:04 +0100720 struct email_alertq *q = container_of(check, typeof(*q), check);
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200721
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100722 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200723 wake_srv_chk(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100724 if (!(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100725 if (check->server)
726 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
727 else
728 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200729 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100730 if (check->server)
731 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
732 else
733 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200734 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200735 return NULL;
736}
737
738/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100739 *
740 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200741 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
742 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200743 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200744static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200745{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200746 struct connection *conn = cs->conn;
747 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900748 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900749 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200750
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100751 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100752 goto out_wakeup;
753
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200754 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100755 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200756 goto out;
757 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100758
Willy Tarreau20a18342013-12-05 00:31:46 +0100759 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200760 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100761 goto out_wakeup;
762 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100763
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200764 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100765 /* if the output is closed, we can't do anything */
766 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200767 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100768 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200769 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200770
Willy Tarreau06559ac2013-12-05 01:53:08 +0100771 /* here, we know that the connection is established. That's enough for
772 * a pure TCP check.
773 */
774 if (!check->type)
775 goto out_wakeup;
776
Willy Tarreauc09572f2017-10-04 11:58:22 +0200777 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100778 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200779 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200780
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200781 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200782 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200783 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200784 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200785 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100786 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200787 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200788 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100789 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200790 goto out;
791 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100792 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200793
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100794 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
795 if (s->proxy->timeout.check) {
796 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
797 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200798 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200799 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100800
Willy Tarreau83749182007-04-15 20:56:27 +0200801 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200802 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200803 out:
804 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200805}
806
Willy Tarreaubaaee002006-06-26 02:48:02 +0200807/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200808 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200809 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900810 * set_server_check_status() to update check->status, check->duration
811 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200812
813 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
814 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
815 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
816 * response to an SSL HELLO (the principle is that this is enough to
817 * distinguish between an SSL server and a pure TCP relay). All other cases will
818 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
819 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100820 *
821 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200822 * via the out label.
823 *
824 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200825 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200826static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200827{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200828 struct connection *conn = cs->conn;
829 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900830 struct server *s = check->server;
831 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200832 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100833 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200834 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200835
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100836 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200837 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200838
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200839 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100840 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 Tarreau892337c2014-05-13 23:41:20 +02001488 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001489 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001490
1491 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001492 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001493
1494 /* get back there in 1 second or 1/20th of the slowstart interval,
1495 * whichever is greater, resulting in small 5% steps.
1496 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001497 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001498 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1499 return t;
1500}
1501
Willy Tarreau894c6422017-10-04 15:58:52 +02001502/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1503 * none was found.
1504 */
1505static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1506{
1507 struct tcpcheck_rule *r;
1508
1509 list_for_each_entry(r, list, list) {
1510 if (r->action != TCPCHK_ACT_COMMENT)
1511 return r;
1512 }
1513 return NULL;
1514}
1515
Willy Tarreau2e993902011-10-31 11:53:20 +01001516/*
Simon Horman98637e52014-06-20 12:30:16 +09001517 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001518 *
1519 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001520 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1521 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1522 * - SF_ERR_SRVTO if there are no more servers
1523 * - SF_ERR_SRVCL if the connection was refused by the server
1524 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1525 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1526 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001527 * - 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 +01001528 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001529 * Note that we try to prevent the network stack from sending the ACK during the
1530 * connect() when a pure TCP check is used (without PROXY protocol).
1531 */
Simon Horman98637e52014-06-20 12:30:16 +09001532static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001533{
1534 struct check *check = t->context;
1535 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001536 struct conn_stream *cs = check->cs;
1537 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001538 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001539 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001540 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001541 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001542
Willy Tarreau00149122017-10-04 18:05:01 +02001543 /* we cannot have a connection here */
1544 if (conn)
1545 return SF_ERR_INTERNAL;
1546
Simon Hormanb00d17a2014-06-13 16:18:16 +09001547 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001548 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001549 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001550 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1551 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001552
1553 /* prepare the check buffer.
1554 * This should not be used if check is the secondary agent check
1555 * of a server as s->proxy->check_req will relate to the
1556 * configuration of the primary check. Similarly, tcp-check uses
1557 * its own strings.
1558 */
1559 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001560 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001561
1562 /* we want to check if this host replies to HTTP or SSLv3 requests
1563 * so we'll send the request, and won't wake the checker up now.
1564 */
1565 if ((check->type) == PR_O2_SSL3_CHK) {
1566 /* SSL requires that we put Unix time in the request */
1567 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001568 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001569 }
1570 else if ((check->type) == PR_O2_HTTP_CHK) {
1571 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001572 b_putblk(&check->bo, trash.area,
1573 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001574 /* prevent HTTP keep-alive when "http-check expect" is used */
1575 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001576 b_putist(&check->bo, ist("Connection: close\r\n"));
1577 b_putist(&check->bo, ist("\r\n"));
1578 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001579 }
1580 }
1581
James Brown55f9ff12015-10-21 18:19:05 -07001582 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001583 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001584 }
1585
Willy Tarreauf411cce2017-10-04 16:21:19 +02001586 /* for tcp-checks, the initial connection setup is handled separately as
1587 * it may be sent to a specific port and not to the server's.
1588 */
1589 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1590 tcpcheck_main(check);
1591 return SF_ERR_UP;
1592 }
1593
Simon Hormanb00d17a2014-06-13 16:18:16 +09001594 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001595 cs = check->cs = cs_new(NULL);
1596 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001597 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001598 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001599 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001600 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001601
Simon Horman41f58762015-01-30 11:22:56 +09001602 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001603 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001604 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001605 }
1606 else {
1607 /* we'll connect to the addr on the server */
1608 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001609 }
1610
Olivier Houchard6377a002017-12-01 22:04:05 +01001611 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001612 conn->target = &s->obj_type;
1613
Willy Tarreaube373152018-09-06 11:45:30 +02001614 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001615 int i = 0;
1616
1617 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001618 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001619 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001620
1621 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001622 }
1623
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001624 /* no client address */
1625 clear_addr(&conn->addr.from);
1626
Willy Tarreaube373152018-09-06 11:45:30 +02001627 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001628 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1629 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001630 cs_attach(cs, check, &check_conn_cb);
1631
Willy Tarreauf3d34822014-12-08 12:11:28 +01001632 /* only plain tcp-check supports quick ACK */
1633 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1634
Willy Tarreauf411cce2017-10-04 16:21:19 +02001635 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1636 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001637
Willy Tarreaue7dff022015-04-03 01:14:29 +02001638 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001639 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001640 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001641
Willy Tarreau16257f62017-11-02 15:45:00 +01001642
Olivier Houchard9130a962017-10-17 17:33:43 +02001643#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001644 if (ret == SF_ERR_NONE) {
1645 if (s->check.sni)
1646 ssl_sock_set_servername(conn, s->check.sni);
1647 if (s->check.alpn_str)
1648 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1649 s->check.alpn_len);
1650 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001651#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001652 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001653 conn->send_proxy_ofs = 1;
1654 conn->flags |= CO_FL_SEND_PROXY;
1655 }
1656
1657 return ret;
1658}
1659
Simon Horman98637e52014-06-20 12:30:16 +09001660static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001661static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001662__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001663
1664void block_sigchld(void)
1665{
1666 sigset_t set;
1667 sigemptyset(&set);
1668 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001669 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001670}
1671
1672void unblock_sigchld(void)
1673{
1674 sigset_t set;
1675 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001676 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001677 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001678}
1679
Simon Horman98637e52014-06-20 12:30:16 +09001680static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1681{
1682 struct pid_list *elem;
1683 struct check *check = t->context;
1684
Willy Tarreaubafbe012017-11-24 17:34:44 +01001685 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001686 if (!elem)
1687 return NULL;
1688 elem->pid = pid;
1689 elem->t = t;
1690 elem->exited = 0;
1691 check->curpid = elem;
1692 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001693
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001694 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001695 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001696 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001697
Simon Horman98637e52014-06-20 12:30:16 +09001698 return elem;
1699}
1700
Simon Horman98637e52014-06-20 12:30:16 +09001701static void pid_list_del(struct pid_list *elem)
1702{
1703 struct check *check;
1704
1705 if (!elem)
1706 return;
1707
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001708 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001709 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001710 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001711
Simon Horman98637e52014-06-20 12:30:16 +09001712 if (!elem->exited)
1713 kill(elem->pid, SIGTERM);
1714
1715 check = elem->t->context;
1716 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001717 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001718}
1719
1720/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1721static void pid_list_expire(pid_t pid, int status)
1722{
1723 struct pid_list *elem;
1724
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001725 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001726 list_for_each_entry(elem, &pid_list, list) {
1727 if (elem->pid == pid) {
1728 elem->t->expire = now_ms;
1729 elem->status = status;
1730 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001731 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001732 break;
Simon Horman98637e52014-06-20 12:30:16 +09001733 }
1734 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001735 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001736}
1737
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001738static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001739{
1740 pid_t pid;
1741 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001742
Simon Horman98637e52014-06-20 12:30:16 +09001743 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1744 pid_list_expire(pid, status);
1745}
1746
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001747static int init_pid_list(void)
1748{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001749 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001750 /* Nothing to do */
1751 return 0;
1752
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001753 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001754 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1755 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001756 return 1;
1757 }
1758
Willy Tarreaubafbe012017-11-24 17:34:44 +01001759 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1760 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001761 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1762 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001763 return 1;
1764 }
1765
1766 return 0;
1767}
1768
Cyril Bontéac92a062014-12-27 22:28:38 +01001769/* helper macro to set an environment variable and jump to a specific label on failure. */
1770#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001771
1772/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001773 * helper function to allocate enough memory to store an environment variable.
1774 * It will also check that the environment variable is updatable, and silently
1775 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001776 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001777static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001778{
1779 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001780 char *envname;
1781 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001782
Cyril Bontéac92a062014-12-27 22:28:38 +01001783 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001784 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001785 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001786 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001787
1788 envname = extcheck_envs[idx].name;
1789 vmaxlen = extcheck_envs[idx].vmaxlen;
1790
1791 /* Check if the environment variable is already set, and silently reject
1792 * the update if this one is not updatable. */
1793 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1794 return 0;
1795
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001796 /* Instead of sending NOT_USED, sending an empty value is preferable */
1797 if (strcmp(value, "NOT_USED") == 0) {
1798 value = "";
1799 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001800
1801 len = strlen(envname) + 1;
1802 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1803 len += strlen(value);
1804 else
1805 len += vmaxlen;
1806
1807 if (!check->envp[idx])
1808 check->envp[idx] = malloc(len + 1);
1809
1810 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001811 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001812 return 1;
1813 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001814 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001815 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001816 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001817 return 1;
1818 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001819 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001820 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001821 return 1;
1822 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001823 return 0;
1824}
Simon Horman98637e52014-06-20 12:30:16 +09001825
1826static int prepare_external_check(struct check *check)
1827{
1828 struct server *s = check->server;
1829 struct proxy *px = s->proxy;
1830 struct listener *listener = NULL, *l;
1831 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001832 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001833 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001834
1835 list_for_each_entry(l, &px->conf.listeners, by_fe)
1836 /* Use the first INET, INET6 or UNIX listener */
1837 if (l->addr.ss_family == AF_INET ||
1838 l->addr.ss_family == AF_INET6 ||
1839 l->addr.ss_family == AF_UNIX) {
1840 listener = l;
1841 break;
1842 }
1843
Simon Horman98637e52014-06-20 12:30:16 +09001844 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001845 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1846 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001847 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001848 goto err;
1849 }
Simon Horman98637e52014-06-20 12:30:16 +09001850
Cyril Bontéac92a062014-12-27 22:28:38 +01001851 check->argv = calloc(6, sizeof(char *));
1852 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001853 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001854 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001855 }
Simon Horman98637e52014-06-20 12:30:16 +09001856
1857 check->argv[0] = px->check_command;
1858
Cyril Bonté777be862014-12-02 21:21:35 +01001859 if (!listener) {
1860 check->argv[1] = strdup("NOT_USED");
1861 check->argv[2] = strdup("NOT_USED");
1862 }
1863 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001864 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001865 addr_to_str(&listener->addr, buf, sizeof(buf));
1866 check->argv[1] = strdup(buf);
1867 port_to_str(&listener->addr, buf, sizeof(buf));
1868 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001869 }
1870 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001871 const struct sockaddr_un *un;
1872
1873 un = (struct sockaddr_un *)&listener->addr;
1874 check->argv[1] = strdup(un->sun_path);
1875 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001876 }
1877 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001878 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001879 goto err;
1880 }
1881
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001882 addr_to_str(&s->addr, buf, sizeof(buf));
1883 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001884
1885 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1886 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1887 else
1888 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001889 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001890
Cyril Bontéac92a062014-12-27 22:28:38 +01001891 for (i = 0; i < 5; i++) {
1892 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001893 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001894 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001895 }
1896 }
Simon Horman98637e52014-06-20 12:30:16 +09001897
Cyril Bontéac92a062014-12-27 22:28:38 +01001898 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001899 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001900 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1901 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1902 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1903 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001904 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001905 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1906 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1907 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1908 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1909 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1910 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1911
1912 /* Ensure that we don't leave any hole in check->envp */
1913 for (i = 0; i < EXTCHK_SIZE; i++)
1914 if (!check->envp[i])
1915 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001916
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001917 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001918err:
1919 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001920 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001921 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001922 free(check->envp);
1923 check->envp = NULL;
1924 }
1925
1926 if (check->argv) {
1927 for (i = 1; i < 5; i++)
1928 free(check->argv[i]);
1929 free(check->argv);
1930 check->argv = NULL;
1931 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001932 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001933}
1934
Simon Hormanb00d17a2014-06-13 16:18:16 +09001935/*
Simon Horman98637e52014-06-20 12:30:16 +09001936 * establish a server health-check that makes use of a process.
1937 *
1938 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001939 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001940 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001941 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001942 *
1943 * Blocks and then unblocks SIGCHLD
1944 */
1945static int connect_proc_chk(struct task *t)
1946{
Cyril Bontéac92a062014-12-27 22:28:38 +01001947 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001948 struct check *check = t->context;
1949 struct server *s = check->server;
1950 struct proxy *px = s->proxy;
1951 int status;
1952 pid_t pid;
1953
Willy Tarreaue7dff022015-04-03 01:14:29 +02001954 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001955
1956 block_sigchld();
1957
1958 pid = fork();
1959 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001960 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1961 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001962 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1963 goto out;
1964 }
1965 if (pid == 0) {
1966 /* Child */
1967 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001968 int fd;
1969
1970 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1971 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1972
Willy Tarreau2555ccf2019-02-21 22:22:06 +01001973 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02001974
Simon Horman98637e52014-06-20 12:30:16 +09001975 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001976 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001977 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001978 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1979 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001980 exit(-1);
1981 }
1982
1983 /* Parent */
1984 if (check->result == CHK_RES_UNKNOWN) {
1985 if (pid_list_add(pid, t) != NULL) {
1986 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1987
1988 if (px->timeout.check && px->timeout.connect) {
1989 int t_con = tick_add(now_ms, px->timeout.connect);
1990 t->expire = tick_first(t->expire, t_con);
1991 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001992 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001993 goto out;
1994 }
1995 else {
1996 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1997 }
1998 kill(pid, SIGTERM); /* process creation error */
1999 }
2000 else
2001 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2002
2003out:
2004 unblock_sigchld();
2005 return status;
2006}
2007
2008/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002009 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002010 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002011 *
2012 * Please do NOT place any return statement in this function and only leave
2013 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002014 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002015static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002016{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002017 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002018 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002019 int rv;
2020 int ret;
2021 int expired = tick_is_expired(t->expire, now_ms);
2022
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002023 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002024 if (!(check->state & CHK_ST_INPROGRESS)) {
2025 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002026 if (!expired) /* woke up too early */
2027 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002028
2029 /* we don't send any health-checks when the proxy is
2030 * stopped, the server should not be checked or the check
2031 * is disabled.
2032 */
2033 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2034 s->proxy->state == PR_STSTOPPED)
2035 goto reschedule;
2036
2037 /* we'll initiate a new check */
2038 set_server_check_status(check, HCHK_STATUS_START, NULL);
2039
2040 check->state |= CHK_ST_INPROGRESS;
2041
Simon Hormandbf70192015-01-30 11:22:53 +09002042 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002043 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002044 /* the process was forked, we allow up to min(inter,
2045 * timeout.connect) for it to report its status, but
2046 * only when timeout.check is set as it may be to short
2047 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002048 */
2049 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2050
2051 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2052 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2053 t->expire = tick_first(t->expire, t_con);
2054 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002055 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002056 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002057 }
2058
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002059 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002060
2061 check->state &= ~CHK_ST_INPROGRESS;
2062 check_notify_failure(check);
2063
2064 /* we allow up to min(inter, timeout.connect) for a connection
2065 * to establish but only when timeout.check is set
2066 * as it may be to short for a full check otherwise
2067 */
2068 while (tick_is_expired(t->expire, now_ms)) {
2069 int t_con;
2070
2071 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2072 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2073
2074 if (s->proxy->timeout.check)
2075 t->expire = tick_first(t->expire, t_con);
2076 }
2077 }
2078 else {
2079 /* there was a test running.
2080 * First, let's check whether there was an uncaught error,
2081 * which can happen on connect timeout or error.
2082 */
2083 if (check->result == CHK_RES_UNKNOWN) {
2084 /* good connection is enough for pure TCP check */
2085 struct pid_list *elem = check->curpid;
2086 int status = HCHK_STATUS_UNKNOWN;
2087
2088 if (elem->exited) {
2089 status = elem->status; /* Save in case the process exits between use below */
2090 if (!WIFEXITED(status))
2091 check->code = -1;
2092 else
2093 check->code = WEXITSTATUS(status);
2094 if (!WIFEXITED(status) || WEXITSTATUS(status))
2095 status = HCHK_STATUS_PROCERR;
2096 else
2097 status = HCHK_STATUS_PROCOK;
2098 } else if (expired) {
2099 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002100 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002101 kill(elem->pid, SIGTERM);
2102 }
2103 set_server_check_status(check, status, NULL);
2104 }
2105
2106 if (check->result == CHK_RES_FAILED) {
2107 /* a failure or timeout detected */
2108 check_notify_failure(check);
2109 }
2110 else if (check->result == CHK_RES_CONDPASS) {
2111 /* check is OK but asks for stopping mode */
2112 check_notify_stopping(check);
2113 }
2114 else if (check->result == CHK_RES_PASSED) {
2115 /* a success was detected */
2116 check_notify_success(check);
2117 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002118 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002119 check->state &= ~CHK_ST_INPROGRESS;
2120
2121 pid_list_del(check->curpid);
2122
2123 rv = 0;
2124 if (global.spread_checks > 0) {
2125 rv = srv_getinter(check) * global.spread_checks / 100;
2126 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2127 }
2128 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2129 }
2130
2131 reschedule:
2132 while (tick_is_expired(t->expire, now_ms))
2133 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002134
2135 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002136 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002137 return t;
2138}
2139
2140/*
2141 * manages a server health-check that uses a connection. Returns
2142 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002143 *
2144 * Please do NOT place any return statement in this function and only leave
2145 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002146 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002147static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002148{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002149 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002150 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002151 struct conn_stream *cs = check->cs;
2152 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002153 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002154 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002155 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002156
Olivier Houchard0923fa42019-01-11 18:43:04 +01002157 if (check->server)
2158 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002159 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002160 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002161 if (!expired) /* woke up too early */
2162 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002163
Simon Horman671b6f02013-11-25 10:46:39 +09002164 /* we don't send any health-checks when the proxy is
2165 * stopped, the server should not be checked or the check
2166 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002167 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002168 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002169 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002170 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002171
2172 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002173 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002174
Willy Tarreau2c115e52013-12-11 19:41:16 +01002175 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002176 b_reset(&check->bi);
2177 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002178
Simon Hormandbf70192015-01-30 11:22:53 +09002179 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002180 cs = check->cs;
2181 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002182
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002183 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002184 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002185 goto out_unlock;
2186
Willy Tarreaue7dff022015-04-03 01:14:29 +02002187 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002188 /* we allow up to min(inter, timeout.connect) for a connection
2189 * to establish but only when timeout.check is set
2190 * as it may be to short for a full check otherwise
2191 */
Simon Horman4a741432013-02-23 15:35:38 +09002192 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002193
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002194 if (proxy->timeout.check && proxy->timeout.connect) {
2195 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002196 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002197 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002198
Olivier Houchard53216e72018-10-10 15:46:36 +02002199 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002200 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002201
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002202 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002203 goto reschedule;
2204
Willy Tarreaue7dff022015-04-03 01:14:29 +02002205 case SF_ERR_SRVTO: /* ETIMEDOUT */
2206 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002207 if (conn)
2208 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002209 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002210 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002211 /* should share same code than cases below */
2212 case SF_ERR_CHK_PORT:
2213 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002214 case SF_ERR_PRXCOND:
2215 case SF_ERR_RESOURCE:
2216 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002217 if (conn)
2218 conn->flags |= CO_FL_ERROR;
2219 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002220 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002221 }
2222
Willy Tarreau5a78f362012-11-23 12:47:05 +01002223 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002224 if (cs) {
2225 cs_destroy(cs);
2226 cs = check->cs = NULL;
2227 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002228 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002229
Willy Tarreau2c115e52013-12-11 19:41:16 +01002230 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002231 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002232
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002233 /* we allow up to min(inter, timeout.connect) for a connection
2234 * to establish but only when timeout.check is set
2235 * as it may be to short for a full check otherwise
2236 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002237 while (tick_is_expired(t->expire, now_ms)) {
2238 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002239
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002240 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002241 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002242
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002243 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002244 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002245 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002246 }
2247 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002248 /* there was a test running.
2249 * First, let's check whether there was an uncaught error,
2250 * which can happen on connect timeout or error.
2251 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002252 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002253 /* good connection is enough for pure TCP check */
2254 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002255 if (check->use_ssl)
2256 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002257 else
Simon Horman4a741432013-02-23 15:35:38 +09002258 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002259 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002260 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002261 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002262 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002263 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002264 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002265 }
2266
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002267 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002268 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002269 /* The check was aborted and the connection was not yet closed.
2270 * This can happen upon timeout, or when an external event such
2271 * as a failed response coupled with "observe layer7" caused the
2272 * server state to be suddenly changed.
2273 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002274 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002275 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002276 }
2277
Willy Tarreauac59f362017-10-08 11:10:19 +02002278 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002279 cs_destroy(cs);
2280 cs = check->cs = NULL;
2281 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002282 }
2283
Olivier Houchard0923fa42019-01-11 18:43:04 +01002284 if (check->server) {
2285 if (check->result == CHK_RES_FAILED) {
2286 /* a failure or timeout detected */
2287 check_notify_failure(check);
2288 }
2289 else if (check->result == CHK_RES_CONDPASS) {
2290 /* check is OK but asks for stopping mode */
2291 check_notify_stopping(check);
2292 }
2293 else if (check->result == CHK_RES_PASSED) {
2294 /* a success was detected */
2295 check_notify_success(check);
2296 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002297 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002298 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002299 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002300
Olivier Houchard0923fa42019-01-11 18:43:04 +01002301 if (check->server) {
2302 rv = 0;
2303 if (global.spread_checks > 0) {
2304 rv = srv_getinter(check) * global.spread_checks / 100;
2305 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2306 }
2307 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002308 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002309 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002310
2311 reschedule:
2312 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002313 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002314 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002315 if (check->server)
2316 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002317 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002318}
2319
Simon Horman98637e52014-06-20 12:30:16 +09002320/*
2321 * manages a server health-check. Returns
2322 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2323 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002324static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002325{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002326 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002327
2328 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002329 return process_chk_proc(t, context, state);
2330 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002331
Simon Horman98637e52014-06-20 12:30:16 +09002332}
2333
Simon Horman5c942422013-11-25 10:46:32 +09002334static int start_check_task(struct check *check, int mininter,
2335 int nbcheck, int srvpos)
2336{
2337 struct task *t;
2338 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002339 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002340 ha_alert("Starting [%s:%s] check: out of memory.\n",
2341 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002342 return 0;
2343 }
2344
2345 check->task = t;
2346 t->process = process_chk;
2347 t->context = check;
2348
Willy Tarreau1746eec2014-04-25 10:46:47 +02002349 if (mininter < srv_getinter(check))
2350 mininter = srv_getinter(check);
2351
2352 if (global.max_spread_checks && mininter > global.max_spread_checks)
2353 mininter = global.max_spread_checks;
2354
Simon Horman5c942422013-11-25 10:46:32 +09002355 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002356 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002357 check->start = now;
2358 task_queue(t);
2359
2360 return 1;
2361}
2362
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002363/*
2364 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002365 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002366 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002367static int start_checks()
2368{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002369
2370 struct proxy *px;
2371 struct server *s;
2372 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002373 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002374
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002375 /* 1- count the checkers to run simultaneously.
2376 * We also determine the minimum interval among all of those which
2377 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2378 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002379 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002380 * too short an interval for all others.
2381 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002382 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002383 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002384 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002385 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002386 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002387 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002388 }
2389 /* We need a warmup task that will be called when the server
2390 * state switches from down to up.
2391 */
2392 s->warmup = t;
2393 t->process = server_warmup;
2394 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002395 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002396 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002397 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 +01002398 }
2399
Willy Tarreaud8514a22013-12-11 21:10:14 +01002400 if (s->check.state & CHK_ST_CONFIGURED) {
2401 nbcheck++;
2402 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2403 (!mininter || mininter > srv_getinter(&s->check)))
2404 mininter = srv_getinter(&s->check);
2405 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002406
Willy Tarreaud8514a22013-12-11 21:10:14 +01002407 if (s->agent.state & CHK_ST_CONFIGURED) {
2408 nbcheck++;
2409 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2410 (!mininter || mininter > srv_getinter(&s->agent)))
2411 mininter = srv_getinter(&s->agent);
2412 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002413 }
2414 }
2415
Simon Horman4a741432013-02-23 15:35:38 +09002416 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002417 return 0;
2418
2419 srand((unsigned)time(NULL));
2420
2421 /*
2422 * 2- start them as far as possible from each others. For this, we will
2423 * start them after their interval set to the min interval divided by
2424 * the number of servers, weighted by the server's position in the list.
2425 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002426 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002427 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2428 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002429 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002430 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002431 }
2432 }
2433
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002434 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002435 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002436 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002437 if (s->check.type == PR_O2_EXT_CHK) {
2438 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002439 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002440 }
Simon Hormand60d6912013-11-25 10:46:36 +09002441 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002442 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002443 srvpos++;
2444 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002445
Simon Hormand60d6912013-11-25 10:46:36 +09002446 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002447 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002448 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002449 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002450 }
2451 srvpos++;
2452 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002453 }
2454 }
2455 return 0;
2456}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002457
2458/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002459 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002460 * The buffer MUST be terminated by a null byte before calling this function.
2461 * Sets server status appropriately. The caller is responsible for ensuring
2462 * that the buffer contains at least 13 characters. If <done> is zero, we may
2463 * return 0 to indicate that data is required to decide of a match.
2464 */
2465static int httpchk_expect(struct server *s, int done)
2466{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002467 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002468 char status_code[] = "000";
2469 char *contentptr;
2470 int crlf;
2471 int ret;
2472
2473 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2474 case PR_O2_EXP_STS:
2475 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002476 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2477 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002478
2479 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2480 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2481 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002482 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002483
2484 /* we necessarily have the response, so there are no partial failures */
2485 if (s->proxy->options2 & PR_O2_EXP_INV)
2486 ret = !ret;
2487
Simon Horman4a741432013-02-23 15:35:38 +09002488 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002489 break;
2490
2491 case PR_O2_EXP_STR:
2492 case PR_O2_EXP_RSTR:
2493 /* very simple response parser: ignore CR and only count consecutive LFs,
2494 * stop with contentptr pointing to first char after the double CRLF or
2495 * to '\0' if crlf < 2.
2496 */
2497 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002498 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002499 if (crlf >= 2)
2500 break;
2501 if (*contentptr == '\r')
2502 continue;
2503 else if (*contentptr == '\n')
2504 crlf++;
2505 else
2506 crlf = 0;
2507 }
2508
2509 /* Check that response contains a body... */
2510 if (crlf < 2) {
2511 if (!done)
2512 return 0;
2513
Simon Horman4a741432013-02-23 15:35:38 +09002514 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002515 "HTTP content check could not find a response body");
2516 return 1;
2517 }
2518
2519 /* Check that response body is not empty... */
2520 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002521 if (!done)
2522 return 0;
2523
Simon Horman4a741432013-02-23 15:35:38 +09002524 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002525 "HTTP content check found empty response body");
2526 return 1;
2527 }
2528
2529 /* Check the response content against the supplied string
2530 * or regex... */
2531 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2532 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2533 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002534 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002535
2536 /* if we don't match, we may need to wait more */
2537 if (!ret && !done)
2538 return 0;
2539
2540 if (ret) {
2541 /* content matched */
2542 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002543 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002544 "HTTP check matched unwanted content");
2545 else
Simon Horman4a741432013-02-23 15:35:38 +09002546 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002547 "HTTP content check matched");
2548 }
2549 else {
2550 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002551 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002552 "HTTP check did not match unwanted content");
2553 else
Simon Horman4a741432013-02-23 15:35:38 +09002554 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002555 "HTTP content check did not match");
2556 }
2557 break;
2558 }
2559 return 1;
2560}
2561
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002562/*
2563 * return the id of a step in a send/expect session
2564 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002565static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002566{
2567 struct tcpcheck_rule *cur = NULL, *next = NULL;
2568 int i = 0;
2569
Willy Tarreau213c6782014-10-02 14:51:02 +02002570 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002571 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002572 return 0;
2573
Simon Hormane16c1b32015-01-30 11:22:57 +09002574 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002575
2576 /* no step => first step */
2577 if (cur == NULL)
2578 return 1;
2579
2580 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002581 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002582 if (next->list.p == &cur->list)
2583 break;
2584 ++i;
2585 }
2586
2587 return i;
2588}
2589
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002590/*
2591 * return the latest known comment before (including) the given stepid
2592 * returns NULL if no comment found
2593 */
2594static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2595{
2596 struct tcpcheck_rule *cur = NULL;
2597 char *ret = NULL;
2598 int i = 0;
2599
2600 /* not even started anything yet, return latest comment found before any action */
2601 if (!check->current_step) {
2602 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2603 if (cur->action == TCPCHK_ACT_COMMENT)
2604 ret = cur->comment;
2605 else
2606 goto return_comment;
2607 }
2608 }
2609
2610 i = 1;
2611 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2612 if (cur->comment)
2613 ret = cur->comment;
2614
2615 if (i >= stepid)
2616 goto return_comment;
2617
2618 ++i;
2619 }
2620
2621 return_comment:
2622 return ret;
2623}
2624
Willy Tarreaube74b882017-10-04 16:22:49 +02002625/* proceed with next steps for the TCP checks <check>. Note that this is called
2626 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002627 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2628 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002629 *
2630 * Please do NOT place any return statement in this function and only leave
2631 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002632 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002633static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002634{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002635 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002636 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002637 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002638 struct conn_stream *cs = check->cs;
2639 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002640 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002641 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002642 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002643 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002644 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002645
Willy Tarreauef953952014-10-02 14:30:14 +02002646 /* here, we know that the check is complete or that it failed */
2647 if (check->result != CHK_RES_UNKNOWN)
2648 goto out_end_tcpcheck;
2649
2650 /* We have 4 possibilities here :
2651 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002652 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002653 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2654 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002655 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002656 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002657 * be waiting for a connection attempt to complete. conn!=NULL.
2658 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002659 *
2660 * #2 and #3 are quite similar, we want both the connection and the
2661 * handshake to complete before going any further. Thus we must always
2662 * wait for a connection to complete unless we're before and existing
2663 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002664 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002665
2666 /* find first rule and skip comments */
2667 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2668 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2669 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2670
Willy Tarreau00149122017-10-04 18:05:01 +02002671 if ((check->current_step || &next->list == head) &&
2672 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002673 /* we allow up to min(inter, timeout.connect) for a connection
2674 * to establish but only when timeout.check is set
2675 * as it may be to short for a full check otherwise
2676 */
2677 while (tick_is_expired(t->expire, now_ms)) {
2678 int t_con;
2679
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002680 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002681 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2682
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002683 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002684 t->expire = tick_first(t->expire, t_con);
2685 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002686 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002687 }
2688
Willy Tarreauef953952014-10-02 14:30:14 +02002689 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002690 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002691 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002692 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002693 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002694
Willy Tarreau213c6782014-10-02 14:51:02 +02002695 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002696 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002697 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002698 b_reset(&check->bo);
2699 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002700 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002701 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002702 if (proxy->timeout.check)
2703 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002704 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002705
Willy Tarreauabca5b62013-12-06 14:19:25 +01002706 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002707 /* We have to try to flush the output buffer before reading, at
2708 * the end, or if we're about to send a string that does not fit
2709 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002710 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002711 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002712 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002713 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002714 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002715 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002716 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002717
Olivier Houcharded0f2072018-08-16 15:41:52 +02002718 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002719 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002720
2721 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002722 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002723 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002724 goto out_end_tcpcheck;
2725 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002726 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002727 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002728 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002729 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002730 goto out;
2731 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002732 }
2733
Willy Tarreau263013d2015-05-13 11:59:14 +02002734 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002735 break;
2736
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002737 /* have 'next' point to the next rule or NULL if we're on the
2738 * last one, connect() needs this.
2739 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002740 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002741
2742 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002743 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002744 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002745
2746 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002747 if (&next->list == head)
2748 next = NULL;
2749
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002750 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2751 struct protocol *proto;
2752 struct xprt_ops *xprt;
2753
Willy Tarreau00149122017-10-04 18:05:01 +02002754 /* For a connect action we'll create a new connection.
2755 * We may also have to kill a previous one. But we don't
2756 * want to leave *without* a connection if we came here
2757 * from the connection layer, hence with a connection.
2758 * Thus we'll proceed in the following order :
2759 * 1: close but not release previous connection
2760 * 2: try to get a new connection
2761 * 3: release and replace the old one on success
2762 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002763 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002764 cs_close(check->cs);
2765 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002766 }
2767
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002768 /* mark the step as started */
2769 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002770
2771 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002772 cs = cs_new(NULL);
2773 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002774 step = tcpcheck_get_step_id(check);
2775 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2776 comment = tcpcheck_get_step_comment(check, step);
2777 if (comment)
2778 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002779 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2780 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002781 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002782 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002783 }
2784
Willy Tarreauac59f362017-10-08 11:10:19 +02002785 if (check->cs)
2786 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002787
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002788 check->cs = cs;
2789 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002790 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002791 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002792 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002793
2794 /* no client address */
2795 clear_addr(&conn->addr.from);
2796
Simon Horman41f58762015-01-30 11:22:56 +09002797 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002798 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002799 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002800 }
2801 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002802 /* we'll connect to the addr on the server */
2803 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002804 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002805 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002806
2807 /* port */
2808 if (check->current_step->port)
2809 set_host_port(&conn->addr.to, check->current_step->port);
2810 else if (check->port)
2811 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002812 else if (s->svc_port)
2813 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002814
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002815 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002816 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002817 }
2818 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002819 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002820 }
Willy Tarreaube373152018-09-06 11:45:30 +02002821
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002822 conn_prepare(conn, proto, xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01002823 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0)
2824 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02002825 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002826
Willy Tarreaue7dff022015-04-03 01:14:29 +02002827 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002828 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002829 ret = proto->connect(conn,
2830 1 /* I/O polling is always needed */,
2831 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002832 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2833 conn->send_proxy_ofs = 1;
2834 conn->flags |= CO_FL_SEND_PROXY;
2835 }
2836
2837 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002838 * - SF_ERR_NONE if everything's OK
2839 * - SF_ERR_SRVTO if there are no more servers
2840 * - SF_ERR_SRVCL if the connection was refused by the server
2841 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2842 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2843 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002844 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002845 * Note that we try to prevent the network stack from sending the ACK during the
2846 * connect() when a pure TCP check is used (without PROXY protocol).
2847 */
2848 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002849 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002850 /* we allow up to min(inter, timeout.connect) for a connection
2851 * to establish but only when timeout.check is set
2852 * as it may be to short for a full check otherwise
2853 */
2854 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2855
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002856 if (proxy->timeout.check && proxy->timeout.connect) {
2857 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002858 t->expire = tick_first(t->expire, t_con);
2859 }
2860 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002861 case SF_ERR_SRVTO: /* ETIMEDOUT */
2862 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002863 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002864 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002865 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002866 comment = tcpcheck_get_step_comment(check, step);
2867 if (comment)
2868 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002869 set_server_check_status(check, HCHK_STATUS_L4CON,
2870 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002871 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002872 case SF_ERR_PRXCOND:
2873 case SF_ERR_RESOURCE:
2874 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002875 step = tcpcheck_get_step_id(check);
2876 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002877 comment = tcpcheck_get_step_comment(check, step);
2878 if (comment)
2879 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002880 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2881 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002882 goto out_end_tcpcheck;
2883 }
2884
2885 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002886 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002887
2888 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002889 while (&check->current_step->list != head &&
2890 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002891 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002892
Willy Tarreauf2c87352015-05-13 12:08:21 +02002893 if (&check->current_step->list == head)
2894 break;
2895
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002896 } /* end 'connect' */
2897 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2898 /* mark the step as started */
2899 check->last_started_step = check->current_step;
2900
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002901 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002902 if (*b_head(&check->bi) != '\0') {
2903 *b_head(&check->bi) = '\0';
2904 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002905 }
2906
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002907 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002908 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002909 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002910 goto out_end_tcpcheck;
2911 }
2912
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002913 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002914 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002915 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002916 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002917 set_server_check_status(check, HCHK_STATUS_L7RSP,
2918 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002919 goto out_end_tcpcheck;
2920 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002921
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002922 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002923 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002924 continue;
2925
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002926 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2927 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002928
Willy Tarreauabca5b62013-12-06 14:19:25 +01002929 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002930 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002931
2932 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002933 while (&check->current_step->list != head &&
2934 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002935 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002936
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002937 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002938 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002939 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002940 goto out_end_tcpcheck;
2941
Olivier Houchard511efea2018-08-16 15:30:32 +02002942 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002943 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002944 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002945 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002946 /* Report network errors only if we got no other data. Otherwise
2947 * we'll let the upper layers decide whether the response is OK
2948 * or not. It is very common that an RST sent by the server is
2949 * reported as an error just after the last data chunk.
2950 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002951 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002952 goto out_end_tcpcheck;
2953 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002954 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002955 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002956 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002957 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002958 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002959 }
2960
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002961 /* mark the step as started */
2962 check->last_started_step = check->current_step;
2963
2964
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002965 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002966 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002967 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002968 if (b_data(&check->bi) < b_size(&check->bi)) {
2969 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002970 }
2971 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002972 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002973 done = 1; /* buffer full, don't wait for more data */
2974 }
2975
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002976 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977
2978 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002979 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002981 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002982
2983 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002984 step = tcpcheck_get_step_id(check);
2985 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002986 comment = tcpcheck_get_step_comment(check, step);
2987 if (comment)
2988 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002989 set_server_check_status(check, HCHK_STATUS_L7RSP,
2990 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002991
2992 goto out_end_tcpcheck;
2993 }
2994
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002995 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002996 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002997
Willy Tarreaua970c282013-12-06 12:47:19 +01002998 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002999 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003000 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 +02003001 else if (check->current_step->expect_regex != NULL)
3002 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003003
3004 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003005 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003006
3007 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003008 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003009 if (ret) {
3010 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003011 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003012 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003013 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003014 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003015 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003016 }
3017 else {
3018 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003019 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003020 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003021 comment = tcpcheck_get_step_comment(check, step);
3022 if (comment)
3023 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003024 set_server_check_status(check, HCHK_STATUS_L7RSP,
3025 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003026 goto out_end_tcpcheck;
3027 }
3028 /* matched and was supposed to => OK, next step */
3029 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003030 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003031 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003032
3033 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003034 while (&check->current_step->list != head &&
3035 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003036 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003037
Willy Tarreauf2c87352015-05-13 12:08:21 +02003038 if (&check->current_step->list == head)
3039 break;
3040
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003041 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003042 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003043 }
3044 }
3045 else {
3046 /* not matched */
3047 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003048 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003049 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003050 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003051
3052 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003053 while (&check->current_step->list != head &&
3054 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003055 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003056
Willy Tarreauf2c87352015-05-13 12:08:21 +02003057 if (&check->current_step->list == head)
3058 break;
3059
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003060 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003061 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003062 }
3063 /* not matched but was supposed to => ERROR */
3064 else {
3065 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003066 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003067 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003068 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003069 }
3070 else {
3071 /* we were looking for a regex */
3072 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003073 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003074 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003075 comment = tcpcheck_get_step_comment(check, step);
3076 if (comment)
3077 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003078 set_server_check_status(check, HCHK_STATUS_L7RSP,
3079 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003080 goto out_end_tcpcheck;
3081 }
3082 }
3083 } /* end expect */
3084 } /* end loop over double chained step list */
3085
Baptiste Assmann248f1172018-03-01 21:49:01 +01003086 /* don't do anything until the connection is established */
3087 if (!(conn->flags & CO_FL_CONNECTED)) {
3088 /* update expire time, should be done by process_chk */
3089 /* we allow up to min(inter, timeout.connect) for a connection
3090 * to establish but only when timeout.check is set
3091 * as it may be to short for a full check otherwise
3092 */
3093 while (tick_is_expired(t->expire, now_ms)) {
3094 int t_con;
3095
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003096 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003097 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3098
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003099 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003100 t->expire = tick_first(t->expire, t_con);
3101 }
3102 goto out;
3103 }
3104
Willy Tarreau263013d2015-05-13 11:59:14 +02003105 /* We're waiting for some I/O to complete, we've reached the end of the
3106 * rules, or both. Do what we have to do, otherwise we're done.
3107 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003108 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003109 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3110 goto out_end_tcpcheck;
3111 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003112
Willy Tarreau53c5a042015-05-13 11:38:17 +02003113 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003114 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003115 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003116 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003117
3118 out_end_tcpcheck:
3119 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003120 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003121 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003122
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003123 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003124 check->current_step = NULL;
3125
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003126 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003127 conn->flags |= CO_FL_ERROR;
3128
Christopher Fauletb6102852017-11-28 10:06:29 +01003129 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003130 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003131}
3132
Simon Hormanb1900d52015-01-30 11:22:54 +09003133const char *init_check(struct check *check, int type)
3134{
3135 check->type = type;
3136
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003137 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3138 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003139
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003140 check->bi.area = calloc(check->bi.size, sizeof(char));
3141 check->bo.area = calloc(check->bo.size, sizeof(char));
3142
3143 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003144 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003145
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003146 check->wait_list.task = tasklet_new();
3147 if (!check->wait_list.task)
3148 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003149 check->wait_list.events = 0;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003150 check->wait_list.task->process = event_srv_chk_io;
3151 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003152 return NULL;
3153}
3154
Simon Hormanbfb5d332015-01-30 11:22:55 +09003155void free_check(struct check *check)
3156{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003157 free(check->bi.area);
3158 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003159 if (check->cs) {
3160 free(check->cs->conn);
3161 check->cs->conn = NULL;
3162 cs_free(check->cs);
3163 check->cs = NULL;
3164 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003165}
3166
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003167void email_alert_free(struct email_alert *alert)
3168{
3169 struct tcpcheck_rule *rule, *back;
3170
3171 if (!alert)
3172 return;
3173
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003174 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3175 LIST_DEL(&rule->list);
3176 free(rule->comment);
3177 free(rule->string);
3178 if (rule->expect_regex)
3179 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003180 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003181 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003182 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003183}
3184
Olivier Houchard9f6af332018-05-25 14:04:04 +02003185static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003186{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003187 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003188 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003189 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003190
3191 q = container_of(check, typeof(*q), check);
3192
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003193 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003194 while (1) {
3195 if (!(check->state & CHK_ST_ENABLED)) {
3196 if (LIST_ISEMPTY(&q->email_alerts)) {
3197 /* All alerts processed, queue the task */
3198 t->expire = TICK_ETERNITY;
3199 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003200 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003201 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003202
3203 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003204 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003205 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003206 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003207 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003208 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003209 }
3210
Olivier Houchard9f6af332018-05-25 14:04:04 +02003211 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003212 if (check->state & CHK_ST_INPROGRESS)
3213 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003214
3215 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3216 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003217 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003218 check->server = NULL;
3219 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003220 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003221 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003222 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003223 return t;
3224}
3225
Christopher Faulet0108bb32017-10-20 21:34:32 +02003226/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3227 *
3228 * The function returns 1 in success case, otherwise, it returns 0 and err is
3229 * filled.
3230 */
3231int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003232{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003233 struct mailer *mailer;
3234 struct email_alertq *queues;
3235 const char *err_str;
3236 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003237
Christopher Faulet0108bb32017-10-20 21:34:32 +02003238 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3239 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003240 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003241 }
3242
Christopher Faulet0108bb32017-10-20 21:34:32 +02003243 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3244 struct email_alertq *q = &queues[i];
3245 struct check *check = &q->check;
3246 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003247
3248 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003249 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003250 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003251 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003252 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003253 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003254 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3255 memprintf(err, "%s", err_str);
3256 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003257 }
3258
3259 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003260 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003261 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003262
Emeric Brunc60def82017-09-27 14:59:38 +02003263 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003264 memprintf(err, "out of memory while allocating mailer alerts task");
3265 goto error;
3266 }
3267
3268 check->task = t;
3269 t->process = process_email_alert;
3270 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003271
Christopher Faulet0108bb32017-10-20 21:34:32 +02003272 /* check this in one ms */
3273 t->expire = TICK_ETERNITY;
3274 check->start = now;
3275 task_queue(t);
3276 }
3277
3278 mls->users++;
3279 free(p->email_alert.mailers.name);
3280 p->email_alert.mailers.m = mls;
3281 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003282 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003283
3284 error:
3285 for (i = 0; i < mls->count; i++) {
3286 struct email_alertq *q = &queues[i];
3287 struct check *check = &q->check;
3288
3289 if (check->task) {
3290 task_delete(check->task);
3291 task_free(check->task);
3292 check->task = NULL;
3293 }
3294 free_check(check);
3295 }
3296 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003297 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003298 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003299}
3300
3301
3302static int add_tcpcheck_expect_str(struct list *list, const char *str)
3303{
3304 struct tcpcheck_rule *tcpcheck;
3305
Willy Tarreaubafbe012017-11-24 17:34:44 +01003306 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003307 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003308 memset(tcpcheck, 0, sizeof(*tcpcheck));
3309 tcpcheck->action = TCPCHK_ACT_EXPECT;
3310 tcpcheck->string = strdup(str);
3311 tcpcheck->expect_regex = NULL;
3312 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003313 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003314 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003315 return 0;
3316 }
3317
3318 LIST_ADDQ(list, &tcpcheck->list);
3319 return 1;
3320}
3321
3322static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3323{
3324 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003325 const char *in;
3326 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327 int i;
3328
Willy Tarreaubafbe012017-11-24 17:34:44 +01003329 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003330 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003331 memset(tcpcheck, 0, sizeof(*tcpcheck));
3332 tcpcheck->action = TCPCHK_ACT_SEND;
3333 tcpcheck->expect_regex = NULL;
3334 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003335 tcpcheck->string_len = 0;
3336 for (i = 0; strs[i]; i++)
3337 tcpcheck->string_len += strlen(strs[i]);
3338
3339 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3340 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003341 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003342 return 0;
3343 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003344
Willy Tarreau64345aa2016-08-10 19:29:09 +02003345 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003346 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003347 for (in = strs[i]; (*dst = *in++); dst++);
3348 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003349
3350 LIST_ADDQ(list, &tcpcheck->list);
3351 return 1;
3352}
3353
Christopher Faulet0108bb32017-10-20 21:34:32 +02003354static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3355 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003356{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003357 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003358 struct tcpcheck_rule *tcpcheck;
3359 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003360
Willy Tarreaubafbe012017-11-24 17:34:44 +01003361 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003362 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003363 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003364 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003365 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003366
Willy Tarreaubafbe012017-11-24 17:34:44 +01003367 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003368 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003369 memset(tcpcheck, 0, sizeof(*tcpcheck));
3370 tcpcheck->action = TCPCHK_ACT_CONNECT;
3371 tcpcheck->comment = NULL;
3372 tcpcheck->string = NULL;
3373 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003374 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3375
3376 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3377 goto error;
3378
3379 {
3380 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3381 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3382 goto error;
3383 }
3384
3385 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3386 goto error;
3387
3388 {
3389 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3390 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3391 goto error;
3392 }
3393
3394 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3395 goto error;
3396
3397 {
3398 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3399 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3400 goto error;
3401 }
3402
3403 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3404 goto error;
3405
3406 {
3407 const char * const strs[2] = { "DATA\r\n" };
3408 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3409 goto error;
3410 }
3411
3412 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3413 goto error;
3414
3415 {
3416 struct tm tm;
3417 char datestr[48];
3418 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003419 "From: ", p->email_alert.from, "\r\n",
3420 "To: ", p->email_alert.to, "\r\n",
3421 "Date: ", datestr, "\r\n",
3422 "Subject: [HAproxy Alert] ", msg, "\r\n",
3423 "\r\n",
3424 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003425 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003426 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003427 NULL
3428 };
3429
3430 get_localtime(date.tv_sec, &tm);
3431
3432 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3433 goto error;
3434 }
3435
3436 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3437 goto error;
3438 }
3439
3440 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3441 goto error;
3442
3443 {
3444 const char * const strs[2] = { "QUIT\r\n" };
3445 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3446 goto error;
3447 }
3448
3449 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3450 goto error;
3451
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003452 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003453 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003454 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003455 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003456 return 1;
3457
3458error:
3459 email_alert_free(alert);
3460 return 0;
3461}
3462
Christopher Faulet0108bb32017-10-20 21:34:32 +02003463static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003464{
3465 int i;
3466 struct mailer *mailer;
3467
3468 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3469 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003470 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003471 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003472 return;
3473 }
3474 }
3475
3476 return;
3477}
3478
3479/*
3480 * Send email alert if configured.
3481 */
Simon Horman64e34162015-02-06 11:11:57 +09003482void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003483{
3484 va_list argp;
3485 char buf[1024];
3486 int len;
3487 struct proxy *p = s->proxy;
3488
Christopher Faulet0108bb32017-10-20 21:34:32 +02003489 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003490 return;
3491
3492 va_start(argp, format);
3493 len = vsnprintf(buf, sizeof(buf), format, argp);
3494 va_end(argp);
3495
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003496 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003497 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003498 return;
3499 }
3500
Christopher Faulet0108bb32017-10-20 21:34:32 +02003501 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003502}
3503
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003504/*
3505 * Return value:
3506 * the port to be used for the health check
3507 * 0 in case no port could be found for the check
3508 */
3509int srv_check_healthcheck_port(struct check *chk)
3510{
3511 int i = 0;
3512 struct server *srv = NULL;
3513
3514 srv = chk->server;
3515
3516 /* If neither a port nor an addr was specified and no check transport
3517 * layer is forced, then the transport layer used by the checks is the
3518 * same as for the production traffic. Otherwise we use raw_sock by
3519 * default, unless one is specified.
3520 */
3521 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003522 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003523 chk->send_proxy |= (srv->pp_opts);
3524 }
3525
3526 /* by default, we use the health check port ocnfigured */
3527 if (chk->port > 0)
3528 return chk->port;
3529
3530 /* try to get the port from check_core.addr if check.port not set */
3531 i = get_host_port(&chk->addr);
3532 if (i > 0)
3533 return i;
3534
3535 /* try to get the port from server address */
3536 /* prevent MAPPORTS from working at this point, since checks could
3537 * not be performed in such case (MAPPORTS impose a relative ports
3538 * based on live traffic)
3539 */
3540 if (srv->flags & SRV_F_MAPPORTS)
3541 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003542
3543 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003544 if (i > 0)
3545 return i;
3546
3547 return 0;
3548}
3549
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003550REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003551
Willy Tarreaubd741542010-03-16 18:46:54 +01003552/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003553 * Local variables:
3554 * c-indent-level: 8
3555 * c-basic-offset: 8
3556 * End:
3557 */