blob: ad25ac094f989fd8face582ce082f83d502ca48a [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 Tarreau3d300592007-03-18 18:34:41 +010052#include <proto/proto_http.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020063
Olivier Houchard9130a962017-10-17 17:33:43 +020064#ifdef USE_OPENSSL
65#include <proto/ssl_sock.h>
66#endif /* USE_OPENSSL */
67
Willy Tarreaubd741542010-03-16 18:46:54 +010068static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090069static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020070static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020071static int tcpcheck_main(struct check *);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Willy Tarreaubafbe012017-11-24 17:34:44 +010073static struct pool_head *pool_head_email_alert = NULL;
74static struct pool_head *pool_head_tcpcheck_rule = NULL;
Christopher Faulet31dff9b2017-10-23 15:45:20 +020075
76
Simon Horman63a4a822012-03-19 07:24:41 +090077static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010078 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
79 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020080 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020081
Willy Tarreau23964182014-05-20 20:56:30 +020082 /* Below we have finished checks */
83 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010085
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
89 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
90 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
93 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
94 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020095
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010096 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
97 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020099 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100101 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
102 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
103 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900104
105 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
106 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200107 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200108};
109
Cyril Bontéac92a062014-12-27 22:28:38 +0100110const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
111 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
122};
123
Simon Horman63a4a822012-03-19 07:24:41 +0900124static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100125 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
126
127 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
128 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
129
130 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
131 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
132 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
133 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
134
135 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
136 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
138};
139
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200140/*
141 * Convert check_status code to description
142 */
143const char *get_check_status_description(short check_status) {
144
145 const char *desc;
146
147 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200148 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200149 else
150 desc = NULL;
151
152 if (desc && *desc)
153 return desc;
154 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200155 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200156}
157
158/*
159 * Convert check_status code to short info
160 */
161const char *get_check_status_info(short check_status) {
162
163 const char *info;
164
165 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200166 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200167 else
168 info = NULL;
169
170 if (info && *info)
171 return info;
172 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200173 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200174}
175
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100176const char *get_analyze_status(short analyze_status) {
177
178 const char *desc;
179
180 if (analyze_status < HANA_STATUS_SIZE)
181 desc = analyze_statuses[analyze_status].desc;
182 else
183 desc = NULL;
184
185 if (desc && *desc)
186 return desc;
187 else
188 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
189}
190
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200191/*
Simon Horman4a741432013-02-23 15:35:38 +0900192 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200193 * an adequate CHK_RES_* value. The new check->health is computed based
194 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200195 *
196 * Show information in logs about failed health check if server is UP
197 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200198 */
Simon Horman4a741432013-02-23 15:35:38 +0900199static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100200{
Simon Horman4a741432013-02-23 15:35:38 +0900201 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200202 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200203 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900204
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200205 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100206 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900207 check->desc[0] = '\0';
208 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200209 return;
210 }
211
Simon Horman4a741432013-02-23 15:35:38 +0900212 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200213 return;
214
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200215 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900216 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
217 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200218 } else
Simon Horman4a741432013-02-23 15:35:38 +0900219 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200220
Simon Horman4a741432013-02-23 15:35:38 +0900221 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200222 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200224
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100225 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900226 check->duration = -1;
227 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200228 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900229 check->duration = tv_ms_elapsed(&check->start, &now);
230 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200231 }
232
Willy Tarreau23964182014-05-20 20:56:30 +0200233 /* no change is expected if no state change occurred */
234 if (check->result == CHK_RES_NEUTRAL)
235 return;
236
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200237 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200238
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200239 switch (check->result) {
240 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200241 /* Failure to connect to the agent as a secondary check should not
242 * cause the server to be marked down.
243 */
244 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900245 (check->status >= HCHK_STATUS_L57DATA)) &&
Willy Tarreau12634e12014-05-23 11:32:36 +0200246 (check->health >= check->rise)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200247 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200248 report = 1;
249 check->health--;
250 if (check->health < check->rise)
251 check->health = 0;
252 }
253 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200254
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200255 case CHK_RES_PASSED:
256 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
257 if ((check->health < check->rise + check->fall - 1) &&
258 (check->result == CHK_RES_PASSED || check->health > 0)) {
259 report = 1;
260 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200261
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200262 if (check->health >= check->rise)
263 check->health = check->rise + check->fall - 1; /* OK now */
264 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200265
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200266 /* clear consecutive_errors if observing is enabled */
267 if (s->onerror)
268 s->consecutive_errors = 0;
269 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100270
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200271 default:
272 break;
273 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200274
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200275 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
276 (status != prev_status || report)) {
277 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200278 "%s check for %sserver %s/%s %s%s",
279 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200280 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100281 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100282 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200283 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200284
Emeric Brun5a133512017-10-19 14:42:30 +0200285 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200286
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100287 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200288 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
289 (check->health >= check->rise) ? check->fall : check->rise,
290 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200291
Christopher Faulet767a84b2017-11-24 16:50:31 +0100292 ha_warning("%s.\n", trash.str);
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100293 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
Simon Horman7ea9be02015-04-30 13:10:33 +0900294 send_email_alert(s, LOG_INFO, "%s", trash.str);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200295 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200296}
297
Willy Tarreau4eec5472014-05-20 22:32:27 +0200298/* Marks the check <check>'s server down if the current check is already failed
299 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200300 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200301static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200302{
Simon Horman4a741432013-02-23 15:35:38 +0900303 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900304
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200305 /* The agent secondary check should only cause a server to be marked
306 * as down if check->status is HCHK_STATUS_L7STS, which indicates
307 * that the agent returned "fail", "stopped" or "down".
308 * The implication here is that failure to connect to the agent
309 * as a secondary check should not cause the server to be marked
310 * down. */
311 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
312 return;
313
Willy Tarreau4eec5472014-05-20 22:32:27 +0200314 if (check->health > 0)
315 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100316
Willy Tarreau4eec5472014-05-20 22:32:27 +0200317 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200318 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200319}
320
Willy Tarreauaf549582014-05-16 17:37:50 +0200321/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200322 * it isn't in maintenance, it is not tracking a down server and other checks
323 * comply. The rule is simple : by default, a server is up, unless any of the
324 * following conditions is true :
325 * - health check failed (check->health < rise)
326 * - agent check failed (agent->health < rise)
327 * - the server tracks a down server (track && track->state == STOPPED)
328 * Note that if the server has a slowstart, it will switch to STARTING instead
329 * of RUNNING. Also, only the health checks support the nolb mode, so the
330 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200331 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200332static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200333{
Simon Horman4a741432013-02-23 15:35:38 +0900334 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100335
Emeric Brun52a91d32017-08-31 14:41:55 +0200336 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200337 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100338
Emeric Brun52a91d32017-08-31 14:41:55 +0200339 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200340 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100341
Willy Tarreau3e048382014-05-21 10:30:54 +0200342 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
343 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100344
Willy Tarreau3e048382014-05-21 10:30:54 +0200345 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
346 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200347
Emeric Brun52a91d32017-08-31 14:41:55 +0200348 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200349 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100350
Emeric Brun5a133512017-10-19 14:42:30 +0200351 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100352}
353
Willy Tarreaudb58b792014-05-21 13:57:23 +0200354/* Marks the check <check> as valid and tries to set its server into stopping mode
355 * if it was running or starting, and provided it isn't in maintenance and other
356 * checks comply. The conditions for the server to be marked in stopping mode are
357 * the same as for it to be turned up. Also, only the health checks support the
358 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200359 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200360static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200361{
Simon Horman4a741432013-02-23 15:35:38 +0900362 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100363
Emeric Brun52a91d32017-08-31 14:41:55 +0200364 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200365 return;
366
Willy Tarreaudb58b792014-05-21 13:57:23 +0200367 if (check->state & CHK_ST_AGENT)
368 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100369
Emeric Brun52a91d32017-08-31 14:41:55 +0200370 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200371 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100372
Willy Tarreaudb58b792014-05-21 13:57:23 +0200373 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
374 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100375
Willy Tarreaudb58b792014-05-21 13:57:23 +0200376 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
377 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100378
Emeric Brun5a133512017-10-19 14:42:30 +0200379 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100380}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200381
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100382/* note: use health_adjust() only, which first checks that the observe mode is
383 * enabled.
384 */
385void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100386{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100387 int failed;
388 int expire;
389
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100390 if (s->observe >= HANA_OBS_SIZE)
391 return;
392
Willy Tarreaubb956662013-01-24 00:37:39 +0100393 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100394 return;
395
396 switch (analyze_statuses[status].lr[s->observe - 1]) {
397 case 1:
398 failed = 1;
399 break;
400
401 case 2:
402 failed = 0;
403 break;
404
405 default:
406 return;
407 }
408
409 if (!failed) {
410 /* good: clear consecutive_errors */
411 s->consecutive_errors = 0;
412 return;
413 }
414
Christopher Faulet29f77e82017-06-08 14:04:45 +0200415 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100416
417 if (s->consecutive_errors < s->consecutive_errors_limit)
418 return;
419
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100420 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
421 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100422
423 switch (s->onerror) {
424 case HANA_ONERR_FASTINTER:
425 /* force fastinter - nothing to do here as all modes force it */
426 break;
427
428 case HANA_ONERR_SUDDTH:
429 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900430 if (s->check.health > s->check.rise)
431 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100432
433 /* no break - fall through */
434
435 case HANA_ONERR_FAILCHK:
436 /* simulate a failed health check */
Simon Horman4a741432013-02-23 15:35:38 +0900437 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200438 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100439 break;
440
441 case HANA_ONERR_MARKDWN:
442 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900443 s->check.health = s->check.rise;
Simon Horman4a741432013-02-23 15:35:38 +0900444 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200445 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100446 break;
447
448 default:
449 /* write a warning? */
450 break;
451 }
452
453 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200454 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100455
Simon Horman66183002013-02-23 10:16:43 +0900456 if (s->check.fastinter) {
457 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300458 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200459 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300460 /* requeue check task with new expire */
461 task_queue(s->check.task);
462 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100463 }
464}
465
Willy Tarreaua1dab552014-04-14 15:04:54 +0200466static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100467{
468 int sv_state;
469 int ratio;
470 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800471 char addr[46];
472 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100473 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
474 "UP %d/%d", "UP",
475 "NOLB %d/%d", "NOLB",
476 "no check" };
477
478 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
479 hlen += 24;
480
Willy Tarreauff5ae352013-12-11 20:36:34 +0100481 if (!(s->check.state & CHK_ST_ENABLED))
482 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200483 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900484 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100485 sv_state = 3; /* UP */
486 else
487 sv_state = 2; /* going down */
488
Emeric Brun52a91d32017-08-31 14:41:55 +0200489 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100490 sv_state += 2;
491 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900492 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100493 sv_state = 1; /* going up */
494 else
495 sv_state = 0; /* DOWN */
496 }
497
Willy Tarreaua1dab552014-04-14 15:04:54 +0200498 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100499 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200500 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
501 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100502
Joseph Lynch514061c2015-01-15 17:52:59 -0800503 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100504 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
505 snprintf(port, sizeof(port), "%u", s->svc_port);
506 else
507 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800508
509 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
510 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100511 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200512 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100513 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
514 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
515 s->nbpend);
516
Emeric Brun52a91d32017-08-31 14:41:55 +0200517 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100518 now.tv_sec < s->last_change + s->slowstart &&
519 now.tv_sec >= s->last_change) {
520 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200521 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100522 }
523
524 buffer[hlen++] = '\r';
525 buffer[hlen++] = '\n';
526
527 return hlen;
528}
529
Willy Tarreau20a18342013-12-05 00:31:46 +0100530/* Check the connection. If an error has already been reported or the socket is
531 * closed, keep errno intact as it is supposed to contain the valid error code.
532 * If no error is reported, check the socket's error queue using getsockopt().
533 * Warning, this must be done only once when returning from poll, and never
534 * after an I/O error was attempted, otherwise the error queue might contain
535 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
536 * socket. Returns non-zero if an error was reported, zero if everything is
537 * clean (including a properly closed socket).
538 */
539static int retrieve_errno_from_socket(struct connection *conn)
540{
541 int skerr;
542 socklen_t lskerr = sizeof(skerr);
543
544 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
545 return 1;
546
Willy Tarreau3c728722014-01-23 13:50:42 +0100547 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100548 return 0;
549
Willy Tarreau585744b2017-08-24 14:31:19 +0200550 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100551 errno = skerr;
552
553 if (errno == EAGAIN)
554 errno = 0;
555
556 if (!errno) {
557 /* we could not retrieve an error, that does not mean there is
558 * none. Just don't change anything and only report the prior
559 * error if any.
560 */
561 if (conn->flags & CO_FL_ERROR)
562 return 1;
563 else
564 return 0;
565 }
566
567 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
568 return 1;
569}
570
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100571/* Try to collect as much information as possible on the connection status,
572 * and adjust the server status accordingly. It may make use of <errno_bck>
573 * if non-null when the caller is absolutely certain of its validity (eg:
574 * checked just after a syscall). If the caller doesn't have a valid errno,
575 * it can pass zero, and retrieve_errno_from_socket() will be called to try
576 * to extract errno from the socket. If no error is reported, it will consider
577 * the <expired> flag. This is intended to be used when a connection error was
578 * reported in conn->flags or when a timeout was reported in <expired>. The
579 * function takes care of not updating a server status which was already set.
580 * All situations where at least one of <expired> or CO_FL_ERROR are set
581 * produce a status.
582 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200583static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100584{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200585 struct conn_stream *cs = check->cs;
586 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100587 const char *err_msg;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200588 struct chunk *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200589 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200590 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100591
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100592 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100593 return;
594
595 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200596 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100597 retrieve_errno_from_socket(conn);
598
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200599 if (conn && !(conn->flags & CO_FL_ERROR) &&
600 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601 return;
602
603 /* we'll try to build a meaningful error message depending on the
604 * context of the error possibly present in conn->err_code, and the
605 * socket error possibly collected above. This is useful to know the
606 * exact step of the L6 layer (eg: SSL handshake).
607 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200608 chk = get_trash_chunk();
609
610 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900611 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200612 if (!step)
613 chunk_printf(chk, " at initial connection step of tcp-check");
614 else {
615 chunk_printf(chk, " at step %d of tcp-check", step);
616 /* we were looking for a string */
617 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
618 if (check->last_started_step->port)
619 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
620 else
621 chunk_appendf(chk, " (connect)");
622 }
623 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
624 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200625 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200626 else if (check->last_started_step->expect_regex)
627 chunk_appendf(chk, " (expect regex)");
628 }
629 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
630 chunk_appendf(chk, " (send)");
631 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200632
633 comment = tcpcheck_get_step_comment(check, step);
634 if (comment)
635 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200636 }
637 }
638
Willy Tarreau00149122017-10-04 18:05:01 +0200639 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100640 if (errno && errno != EAGAIN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200641 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100642 else
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200643 chunk_printf(&trash, "%s%s", conn_err_code_str(conn), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100644 err_msg = trash.str;
645 }
646 else {
647 if (errno && errno != EAGAIN) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200648 chunk_printf(&trash, "%s%s", strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100649 err_msg = trash.str;
650 }
651 else {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200652 err_msg = chk->str;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100653 }
654 }
655
Willy Tarreau00149122017-10-04 18:05:01 +0200656 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200657 /* NOTE: this is reported after <fall> tries */
658 chunk_printf(chk, "No port available for the TCP connection");
659 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
660 }
661
Willy Tarreau00149122017-10-04 18:05:01 +0200662 if (!conn) {
663 /* connection allocation error before the connection was established */
664 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
665 }
666 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100667 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200668 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100669 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
670 else if (expired)
671 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200672
673 /*
674 * might be due to a server IP change.
675 * Let's trigger a DNS resolution if none are currently running.
676 */
Christopher Faulet67957bd2017-09-27 11:00:59 +0200677 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200678
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100679 }
680 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
681 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200682 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100683 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
684 else if (expired)
685 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
686 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200687 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100688 /* I/O error after connection was established and before we could diagnose */
689 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
690 }
691 else if (expired) {
692 /* connection established but expired check */
693 if (check->type == PR_O2_SSL3_CHK)
694 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
695 else /* HTTP, SMTP, ... */
696 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
697 }
698
699 return;
700}
701
Willy Tarreaubaaee002006-06-26 02:48:02 +0200702/*
703 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200704 * the connection acknowledgement. If the proxy requires L7 health-checks,
705 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900706 * to set check->status, check->duration and check->result.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100707 *
708 * Please do NOT place any return statement in this function and only leave
709 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200710 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200711static void event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200712{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200713 struct connection *conn = cs->conn;
714 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900715 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900716 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200717
Christopher Faulet2a944ee2017-11-07 10:42:54 +0100718 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100719 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100720 goto out_wakeup;
721
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100722 if (conn->flags & CO_FL_HANDSHAKE)
723 goto out_unlock;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100724
Willy Tarreau20a18342013-12-05 00:31:46 +0100725 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200726 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200727 __cs_stop_both(cs);
Willy Tarreau20a18342013-12-05 00:31:46 +0100728 goto out_wakeup;
729 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100730
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200731 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100732 /* if the output is closed, we can't do anything */
733 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200734 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100735 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200736 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200737
Willy Tarreau06559ac2013-12-05 01:53:08 +0100738 /* here, we know that the connection is established. That's enough for
739 * a pure TCP check.
740 */
741 if (!check->type)
742 goto out_wakeup;
743
Willy Tarreauc09572f2017-10-04 11:58:22 +0200744 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100745 if (check->type == PR_O2_TCPCHK_CHK)
746 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200747
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100748 if (check->bo->o) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200749 conn->mux->snd_buf(cs, check->bo, 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200750 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200751 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200752 __cs_stop_both(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100753 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200754 }
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100755 if (check->bo->o)
756 goto out_unlock;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100757 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200758
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100759 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
760 if (s->proxy->timeout.check) {
761 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
762 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200763 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100764 goto out_nowake;
765
Willy Tarreau83749182007-04-15 20:56:27 +0200766 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200767 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200768 out_nowake:
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200769 __cs_stop_send(cs); /* nothing more to write */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100770 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +0100771 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200772}
773
Willy Tarreaubaaee002006-06-26 02:48:02 +0200774/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200775 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200776 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900777 * set_server_check_status() to update check->status, check->duration
778 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200779
780 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
781 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
782 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
783 * response to an SSL HELLO (the principle is that this is enough to
784 * distinguish between an SSL server and a pure TCP relay). All other cases will
785 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
786 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100787 *
788 * Please do NOT place any return statement in this function and only leave
789 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200790 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200791static void event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200792{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200793 struct connection *conn = cs->conn;
794 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900795 struct server *s = check->server;
796 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200797 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100798 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200799 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200800
Christopher Faulet2a944ee2017-11-07 10:42:54 +0100801 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brun9f0b4582017-10-23 14:39:51 +0200802
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100803 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200804 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200805
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100806 if (conn->flags & CO_FL_HANDSHAKE)
807 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200808
Willy Tarreauc09572f2017-10-04 11:58:22 +0200809 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100810 if (check->type == PR_O2_TCPCHK_CHK)
811 goto out_unlock;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200812
Willy Tarreau83749182007-04-15 20:56:27 +0200813 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
814 * but the connection was closed on the remote end. Fortunately, recv still
815 * works correctly and we don't need to do the getsockopt() on linux.
816 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000817
818 /* Set buffer to point to the end of the data already read, and check
819 * that there is free space remaining. If the buffer is full, proceed
820 * with running the checks without attempting another socket read.
821 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000822
Willy Tarreau03938182010-03-17 21:52:07 +0100823 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000824
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200825 conn->mux->rcv_buf(cs, check->bi, check->bi->size);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200826 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100827 done = 1;
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200828 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !check->bi->i) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200829 /* Report network errors only if we got no other data. Otherwise
830 * we'll let the upper layers decide whether the response is OK
831 * or not. It is very common that an RST sent by the server is
832 * reported as an error just after the last data chunk.
833 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200834 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100835 goto out_wakeup;
836 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200837 }
838
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100839
Willy Tarreau03938182010-03-17 21:52:07 +0100840 /* Intermediate or complete response received.
Simon Horman4a741432013-02-23 15:35:38 +0900841 * Terminate string in check->bi->data buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100842 */
Simon Horman4a741432013-02-23 15:35:38 +0900843 if (check->bi->i < check->bi->size)
844 check->bi->data[check->bi->i] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100845 else {
Simon Horman4a741432013-02-23 15:35:38 +0900846 check->bi->data[check->bi->i - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100847 done = 1; /* buffer full, don't wait for more data */
848 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200849
Nick Chalk57b1bf72010-03-16 15:50:46 +0000850 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900851 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200852 case PR_O2_HTTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900853 if (!done && check->bi->i < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100854 goto wait_more_data;
855
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100856 /* Check if the server speaks HTTP 1.X */
Simon Horman4a741432013-02-23 15:35:38 +0900857 if ((check->bi->i < strlen("HTTP/1.0 000\r")) ||
858 (memcmp(check->bi->data, "HTTP/1.", 7) != 0 ||
859 (*(check->bi->data + 12) != ' ' && *(check->bi->data + 12) != '\r')) ||
860 !isdigit((unsigned char) *(check->bi->data + 9)) || !isdigit((unsigned char) *(check->bi->data + 10)) ||
861 !isdigit((unsigned char) *(check->bi->data + 11))) {
862 cut_crlf(check->bi->data);
863 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200864
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100865 goto out_wakeup;
866 }
867
Simon Horman4a741432013-02-23 15:35:38 +0900868 check->code = str2uic(check->bi->data + 9);
869 desc = ltrim(check->bi->data + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200870
Willy Tarreaubd741542010-03-16 18:46:54 +0100871 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200872 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000873 /* 404 may be accepted as "stopping" only if the server was up */
874 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900875 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000876 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100877 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
878 /* Run content verification check... We know we have at least 13 chars */
879 if (!httpchk_expect(s, done))
880 goto wait_more_data;
881 }
882 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Simon Horman4a741432013-02-23 15:35:38 +0900883 else if (*(check->bi->data + 9) == '2' || *(check->bi->data + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100884 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900885 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100886 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000887 else {
888 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900889 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000890 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200891 break;
892
893 case PR_O2_SSL3_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900894 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100895 goto wait_more_data;
896
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100897 /* Check for SSLv3 alert or handshake */
Simon Horman4a741432013-02-23 15:35:38 +0900898 if ((check->bi->i >= 5) && (*check->bi->data == 0x15 || *check->bi->data == 0x16))
899 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200900 else
Simon Horman4a741432013-02-23 15:35:38 +0900901 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200902 break;
903
904 case PR_O2_SMTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900905 if (!done && check->bi->i < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100906 goto wait_more_data;
907
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200908 /* Check if the server speaks SMTP */
Simon Horman4a741432013-02-23 15:35:38 +0900909 if ((check->bi->i < strlen("000\r")) ||
910 (*(check->bi->data + 3) != ' ' && *(check->bi->data + 3) != '\r') ||
911 !isdigit((unsigned char) *check->bi->data) || !isdigit((unsigned char) *(check->bi->data + 1)) ||
912 !isdigit((unsigned char) *(check->bi->data + 2))) {
913 cut_crlf(check->bi->data);
914 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200915 goto out_wakeup;
916 }
917
Simon Horman4a741432013-02-23 15:35:38 +0900918 check->code = str2uic(check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200919
Simon Horman4a741432013-02-23 15:35:38 +0900920 desc = ltrim(check->bi->data + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200921 cut_crlf(desc);
922
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100923 /* Check for SMTP code 2xx (should be 250) */
Simon Horman4a741432013-02-23 15:35:38 +0900924 if (*check->bi->data == '2')
925 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200926 else
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200928 break;
929
Simon Hormana2b9dad2013-02-12 10:45:54 +0900930 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100931 int status = HCHK_STATUS_CHECKED;
932 const char *hs = NULL; /* health status */
933 const char *as = NULL; /* admin status */
934 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200935 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100936 const char *err = NULL; /* first error to report */
937 const char *wrn = NULL; /* first warning to report */
938 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900939
Willy Tarreau81f5d942013-12-09 20:51:51 +0100940 /* We're getting an agent check response. The agent could
941 * have been disabled in the mean time with a long check
942 * still pending. It is important that we ignore the whole
943 * response.
944 */
945 if (!(check->server->agent.state & CHK_ST_ENABLED))
946 break;
947
948 /* The agent supports strings made of a single line ended by the
949 * first CR ('\r') or LF ('\n'). This line is composed of words
950 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
951 * line may optionally contained a description of a state change
952 * after a sharp ('#'), which is only considered if a health state
953 * is announced.
954 *
955 * Words may be composed of :
956 * - a numeric weight suffixed by the percent character ('%').
957 * - a health status among "up", "down", "stopped", and "fail".
958 * - an admin status among "ready", "drain", "maint".
959 *
960 * These words may appear in any order. If multiple words of the
961 * same category appear, the last one wins.
962 */
963
Willy Tarreau9809b782013-12-11 21:40:11 +0100964 p = check->bi->data;
965 while (*p && *p != '\n' && *p != '\r')
966 p++;
967
968 if (!*p) {
969 if (!done)
970 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900971
Willy Tarreau9809b782013-12-11 21:40:11 +0100972 /* at least inform the admin that the agent is mis-behaving */
973 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
974 break;
975 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100976
Willy Tarreau9809b782013-12-11 21:40:11 +0100977 *p = 0;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100978 cmd = check->bi->data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900979
Willy Tarreau81f5d942013-12-09 20:51:51 +0100980 while (*cmd) {
981 /* look for next word */
982 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
983 cmd++;
984 continue;
985 }
Simon Horman671b6f02013-11-25 10:46:39 +0900986
Willy Tarreau81f5d942013-12-09 20:51:51 +0100987 if (*cmd == '#') {
988 /* this is the beginning of a health status description,
989 * skip the sharp and blanks.
990 */
991 cmd++;
992 while (*cmd == '\t' || *cmd == ' ')
993 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +0900994 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900995 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100996
997 /* find the end of the word so that we have a null-terminated
998 * word between <cmd> and <p>.
999 */
1000 p = cmd + 1;
1001 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1002 p++;
1003 if (*p)
1004 *p++ = 0;
1005
1006 /* first, health statuses */
1007 if (strcasecmp(cmd, "up") == 0) {
1008 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001009 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001010 hs = cmd;
1011 }
1012 else if (strcasecmp(cmd, "down") == 0) {
1013 check->health = 0;
1014 status = HCHK_STATUS_L7STS;
1015 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001016 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001017 else if (strcasecmp(cmd, "stopped") == 0) {
1018 check->health = 0;
1019 status = HCHK_STATUS_L7STS;
1020 hs = cmd;
1021 }
1022 else if (strcasecmp(cmd, "fail") == 0) {
1023 check->health = 0;
1024 status = HCHK_STATUS_L7STS;
1025 hs = cmd;
1026 }
1027 /* admin statuses */
1028 else if (strcasecmp(cmd, "ready") == 0) {
1029 as = cmd;
1030 }
1031 else if (strcasecmp(cmd, "drain") == 0) {
1032 as = cmd;
1033 }
1034 else if (strcasecmp(cmd, "maint") == 0) {
1035 as = cmd;
1036 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001037 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001038 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1039 ps = cmd;
1040 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001041 /* try to parse a maxconn here */
1042 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1043 cs = cmd;
1044 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001045 else {
1046 /* keep a copy of the first error */
1047 if (!err)
1048 err = cmd;
1049 }
1050 /* skip to next word */
1051 cmd = p;
1052 }
1053 /* here, cmd points either to \0 or to the beginning of a
1054 * description. Skip possible leading spaces.
1055 */
1056 while (*cmd == ' ' || *cmd == '\n')
1057 cmd++;
1058
1059 /* First, update the admin status so that we avoid sending other
1060 * possibly useless warnings and can also update the health if
1061 * present after going back up.
1062 */
1063 if (as) {
1064 if (strcasecmp(as, "drain") == 0)
1065 srv_adm_set_drain(check->server);
1066 else if (strcasecmp(as, "maint") == 0)
1067 srv_adm_set_maint(check->server);
1068 else
1069 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001070 }
1071
Willy Tarreau81f5d942013-12-09 20:51:51 +01001072 /* now change weights */
1073 if (ps) {
1074 const char *msg;
1075
1076 msg = server_parse_weight_change_request(s, ps);
1077 if (!wrn || !*wrn)
1078 wrn = msg;
1079 }
1080
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001081 if (cs) {
1082 const char *msg;
1083
1084 cs += strlen("maxconn:");
1085
1086 msg = server_parse_maxconn_change_request(s, cs);
1087 if (!wrn || !*wrn)
1088 wrn = msg;
1089 }
1090
Willy Tarreau81f5d942013-12-09 20:51:51 +01001091 /* and finally health status */
1092 if (hs) {
1093 /* We'll report some of the warnings and errors we have
1094 * here. Down reports are critical, we leave them untouched.
1095 * Lack of report, or report of 'UP' leaves the room for
1096 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001097 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001098 const char *msg = cmd;
1099 struct chunk *t;
1100
1101 if (!*msg || status == HCHK_STATUS_L7OKD) {
1102 if (err && *err)
1103 msg = err;
1104 else if (wrn && *wrn)
1105 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001106 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001107
1108 t = get_trash_chunk();
1109 chunk_printf(t, "via agent : %s%s%s%s",
1110 hs, *msg ? " (" : "",
1111 msg, *msg ? ")" : "");
1112
1113 set_server_check_status(check, status, t->str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001114 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001115 else if (err && *err) {
1116 /* No status change but we'd like to report something odd.
1117 * Just report the current state and copy the message.
1118 */
1119 chunk_printf(&trash, "agent reports an error : %s", err);
1120 set_server_check_status(check, status/*check->status*/, trash.str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001121
Willy Tarreau81f5d942013-12-09 20:51:51 +01001122 }
1123 else if (wrn && *wrn) {
1124 /* No status change but we'd like to report something odd.
1125 * Just report the current state and copy the message.
1126 */
1127 chunk_printf(&trash, "agent warns : %s", wrn);
1128 set_server_check_status(check, status/*check->status*/, trash.str);
1129 }
1130 else
1131 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001132 break;
1133 }
1134
Willy Tarreau1620ec32011-08-06 17:05:02 +02001135 case PR_O2_PGSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001136 if (!done && check->bi->i < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001137 goto wait_more_data;
1138
Simon Horman4a741432013-02-23 15:35:38 +09001139 if (check->bi->data[0] == 'R') {
1140 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001141 }
1142 else {
Simon Horman4a741432013-02-23 15:35:38 +09001143 if ((check->bi->data[0] == 'E') && (check->bi->data[5]!=0) && (check->bi->data[6]!=0))
1144 desc = &check->bi->data[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001145 else
1146 desc = "PostgreSQL unknown error";
1147
Simon Horman4a741432013-02-23 15:35:38 +09001148 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001149 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001150 break;
1151
1152 case PR_O2_REDIS_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001153 if (!done && check->bi->i < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001154 goto wait_more_data;
1155
Simon Horman4a741432013-02-23 15:35:38 +09001156 if (strcmp(check->bi->data, "+PONG\r\n") == 0) {
1157 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001158 }
1159 else {
Simon Horman4a741432013-02-23 15:35:38 +09001160 set_server_check_status(check, HCHK_STATUS_L7STS, check->bi->data);
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001161 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001162 break;
1163
1164 case PR_O2_MYSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001165 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001166 goto wait_more_data;
1167
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001168 if (s->proxy->check_len == 0) { // old mode
Simon Horman4a741432013-02-23 15:35:38 +09001169 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001170 /* We set the MySQL Version in description for information purpose
1171 * FIXME : it can be cool to use MySQL Version for other purpose,
1172 * like mark as down old MySQL server.
1173 */
Simon Horman4a741432013-02-23 15:35:38 +09001174 if (check->bi->i > 51) {
1175 desc = ltrim(check->bi->data + 5, ' ');
1176 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001177 }
1178 else {
1179 if (!done)
1180 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001181
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001182 /* it seems we have a OK packet but without a valid length,
1183 * it must be a protocol error
1184 */
Simon Horman4a741432013-02-23 15:35:38 +09001185 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001186 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001187 }
1188 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001189 /* An error message is attached in the Error packet */
Simon Horman4a741432013-02-23 15:35:38 +09001190 desc = ltrim(check->bi->data + 7, ' ');
1191 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001192 }
1193 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001194 unsigned int first_packet_len = ((unsigned int) *check->bi->data) +
1195 (((unsigned int) *(check->bi->data + 1)) << 8) +
1196 (((unsigned int) *(check->bi->data + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001197
Simon Horman4a741432013-02-23 15:35:38 +09001198 if (check->bi->i == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001199 /* MySQL Error packet always begin with field_count = 0xff */
Simon Horman4a741432013-02-23 15:35:38 +09001200 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001201 /* We have only one MySQL packet and it is a Handshake Initialization packet
1202 * but we need to have a second packet to know if it is alright
1203 */
Simon Horman4a741432013-02-23 15:35:38 +09001204 if (!done && check->bi->i < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001205 goto wait_more_data;
1206 }
1207 else {
1208 /* We have only one packet and it is an Error packet,
1209 * an error message is attached, so we can display it
1210 */
Simon Horman4a741432013-02-23 15:35:38 +09001211 desc = &check->bi->data[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001212 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001213 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001214 }
Simon Horman4a741432013-02-23 15:35:38 +09001215 } else if (check->bi->i > first_packet_len + 4) {
1216 unsigned int second_packet_len = ((unsigned int) *(check->bi->data + first_packet_len + 4)) +
1217 (((unsigned int) *(check->bi->data + first_packet_len + 5)) << 8) +
1218 (((unsigned int) *(check->bi->data + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001219
Simon Horman4a741432013-02-23 15:35:38 +09001220 if (check->bi->i == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001221 /* We have 2 packets and that's good */
1222 /* Check if the second packet is a MySQL Error packet or not */
Simon Horman4a741432013-02-23 15:35:38 +09001223 if (*(check->bi->data + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001224 /* No error packet */
1225 /* We set the MySQL Version in description for information purpose */
Simon Horman4a741432013-02-23 15:35:38 +09001226 desc = &check->bi->data[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001227 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001228 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 }
1230 else {
1231 /* An error message is attached in the Error packet
1232 * so we can display it ! :)
1233 */
Simon Horman4a741432013-02-23 15:35:38 +09001234 desc = &check->bi->data[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001235 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001236 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001237 }
1238 }
1239 }
1240 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001241 if (!done)
1242 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001243
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001244 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001245 * it must be a protocol error
1246 */
Simon Horman4a741432013-02-23 15:35:38 +09001247 desc = &check->bi->data[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001248 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001249 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001250 }
1251 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001252 break;
1253
1254 case PR_O2_LDAP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001255 if (!done && check->bi->i < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001256 goto wait_more_data;
1257
1258 /* Check if the server speaks LDAP (ASN.1/BER)
1259 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1260 * http://tools.ietf.org/html/rfc4511
1261 */
1262
1263 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1264 * LDAPMessage: 0x30: SEQUENCE
1265 */
Simon Horman4a741432013-02-23 15:35:38 +09001266 if ((check->bi->i < 14) || (*(check->bi->data) != '\x30')) {
1267 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001268 }
1269 else {
1270 /* size of LDAPMessage */
Simon Horman4a741432013-02-23 15:35:38 +09001271 msglen = (*(check->bi->data + 1) & 0x80) ? (*(check->bi->data + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001272
1273 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1274 * messageID: 0x02 0x01 0x01: INTEGER 1
1275 * protocolOp: 0x61: bindResponse
1276 */
1277 if ((msglen > 2) ||
Simon Horman4a741432013-02-23 15:35:38 +09001278 (memcmp(check->bi->data + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
1279 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001280 goto out_wakeup;
1281 }
1282
1283 /* size of bindResponse */
Simon Horman4a741432013-02-23 15:35:38 +09001284 msglen += (*(check->bi->data + msglen + 6) & 0x80) ? (*(check->bi->data + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001285
1286 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1287 * ldapResult: 0x0a 0x01: ENUMERATION
1288 */
1289 if ((msglen > 4) ||
Simon Horman4a741432013-02-23 15:35:38 +09001290 (memcmp(check->bi->data + 7 + msglen, "\x0a\x01", 2) != 0)) {
1291 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001292 goto out_wakeup;
1293 }
1294
1295 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1296 * resultCode
1297 */
Simon Horman4a741432013-02-23 15:35:38 +09001298 check->code = *(check->bi->data + msglen + 9);
1299 if (check->code) {
1300 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 +02001301 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001302 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001303 }
1304 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001305 break;
1306
Christopher Fauletba7bc162016-11-07 21:07:38 +01001307 case PR_O2_SPOP_CHK: {
1308 unsigned int framesz;
1309 char err[HCHK_DESC_LEN];
1310
1311 if (!done && check->bi->i < 4)
1312 goto wait_more_data;
1313
1314 memcpy(&framesz, check->bi->data, 4);
1315 framesz = ntohl(framesz);
1316
1317 if (!done && check->bi->i < (4+framesz))
1318 goto wait_more_data;
1319
Christopher Faulet8ef75252017-02-20 22:56:03 +01001320 if (!spoe_handle_healthcheck_response(check->bi->data+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001321 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1322 else
1323 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1324 break;
1325 }
1326
Willy Tarreau1620ec32011-08-06 17:05:02 +02001327 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001328 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001329 break;
1330 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001331
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001332 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001333 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001334 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001335 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001336
Nick Chalk57b1bf72010-03-16 15:50:46 +00001337 /* Reset the check buffer... */
Simon Horman4a741432013-02-23 15:35:38 +09001338 *check->bi->data = '\0';
1339 check->bi->i = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +00001340
Steven Davidovitz544d4812017-03-08 11:06:20 -08001341 /* Close the connection... We still attempt to nicely close if,
1342 * for instance, SSL needs to send a "close notify." Later, we perform
1343 * a hard close and reset the connection if some data are pending,
1344 * otherwise we end up with many TIME_WAITs and eat all the source port
1345 * range quickly. To avoid sending RSTs all the time, we first try to
1346 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001347 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001348 __cs_stop_both(cs);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001349 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001350
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001351 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001352 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001353 conn->flags |= CO_FL_ERROR;
1354
Willy Tarreaufdccded2008-08-29 18:19:04 +02001355 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001356 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001357 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau3267d362012-08-17 23:53:56 +02001358 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001359
1360 wait_more_data:
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001361 __cs_want_recv(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001362 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001363}
1364
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001365/*
1366 * This function is used only for server health-checks. It handles connection
1367 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001368 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1369 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001370 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001371static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001372{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001373 struct connection *conn = cs->conn;
1374 struct check *check = cs->data;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001375 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001376
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001377 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001378
Willy Tarreauc09572f2017-10-04 11:58:22 +02001379 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001380 if (check->type == PR_O2_TCPCHK_CHK) {
1381 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001382 cs = check->cs;
1383 conn = cs_conn(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001384 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001385
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001386 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001387 /* We may get error reports bypassing the I/O handlers, typically
1388 * the case when sending a pure TCP check which fails, then the I/O
1389 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001390 * main processing task so let's simply wake it up. If we get here,
1391 * we expect errno to still be valid.
1392 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001393 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001394 __cs_stop_both(cs);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001395 task_wakeup(check->task, TASK_WOKEN_IO);
1396 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +02001397 else if (!(conn->flags & (CO_FL_XPRT_RD_ENA|CO_FL_XPRT_WR_ENA|CO_FL_HANDSHAKE))) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001398 /* we may get here if only a connection probe was required : we
1399 * don't have any data to send nor anything expected in response,
1400 * so the completion of the connection establishment is enough.
1401 */
1402 task_wakeup(check->task, TASK_WOKEN_IO);
1403 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001404
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001405 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001406 /* We're here because nobody wants to handle the error, so we
1407 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001408 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001409 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001410 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001411 ret = -1;
Willy Tarreau2d351b62013-12-05 02:36:25 +01001412 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001413
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001414 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001415
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001416 /* if a connection got replaced, we must absolutely prevent the connection
1417 * handler from touching its fd, and perform the FD polling updates ourselves
1418 */
1419 if (ret < 0)
1420 conn_cond_update_polling(conn);
1421
1422 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001423}
1424
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001425struct data_cb check_conn_cb = {
1426 .recv = event_srv_chk_r,
1427 .send = event_srv_chk_w,
1428 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001429 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001430};
1431
Willy Tarreaubaaee002006-06-26 02:48:02 +02001432/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001433 * updates the server's weight during a warmup stage. Once the final weight is
1434 * reached, the task automatically stops. Note that any server status change
1435 * must have updated s->last_change accordingly.
1436 */
1437static struct task *server_warmup(struct task *t)
1438{
1439 struct server *s = t->context;
1440
1441 /* by default, plan on stopping the task */
1442 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001443 if ((s->next_admin & SRV_ADMF_MAINT) ||
1444 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001445 return t;
1446
Willy Tarreau892337c2014-05-13 23:41:20 +02001447 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001448 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001449
1450 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001451 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001452
1453 /* get back there in 1 second or 1/20th of the slowstart interval,
1454 * whichever is greater, resulting in small 5% steps.
1455 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001456 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001457 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1458 return t;
1459}
1460
Willy Tarreau894c6422017-10-04 15:58:52 +02001461/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1462 * none was found.
1463 */
1464static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1465{
1466 struct tcpcheck_rule *r;
1467
1468 list_for_each_entry(r, list, list) {
1469 if (r->action != TCPCHK_ACT_COMMENT)
1470 return r;
1471 }
1472 return NULL;
1473}
1474
Willy Tarreau2e993902011-10-31 11:53:20 +01001475/*
Simon Horman98637e52014-06-20 12:30:16 +09001476 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001477 *
1478 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001479 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1480 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1481 * - SF_ERR_SRVTO if there are no more servers
1482 * - SF_ERR_SRVCL if the connection was refused by the server
1483 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1484 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1485 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001486 * - 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 +01001487 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001488 * Note that we try to prevent the network stack from sending the ACK during the
1489 * connect() when a pure TCP check is used (without PROXY protocol).
1490 */
Simon Horman98637e52014-06-20 12:30:16 +09001491static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001492{
1493 struct check *check = t->context;
1494 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001495 struct conn_stream *cs = check->cs;
1496 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001497 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001498 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001499 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001500 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001501
Willy Tarreau00149122017-10-04 18:05:01 +02001502 /* we cannot have a connection here */
1503 if (conn)
1504 return SF_ERR_INTERNAL;
1505
Simon Hormanb00d17a2014-06-13 16:18:16 +09001506 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001507 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001508 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001509 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1510 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001511
1512 /* prepare the check buffer.
1513 * This should not be used if check is the secondary agent check
1514 * of a server as s->proxy->check_req will relate to the
1515 * configuration of the primary check. Similarly, tcp-check uses
1516 * its own strings.
1517 */
1518 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
1519 bo_putblk(check->bo, s->proxy->check_req, s->proxy->check_len);
1520
1521 /* we want to check if this host replies to HTTP or SSLv3 requests
1522 * so we'll send the request, and won't wake the checker up now.
1523 */
1524 if ((check->type) == PR_O2_SSL3_CHK) {
1525 /* SSL requires that we put Unix time in the request */
1526 int gmt_time = htonl(date.tv_sec);
1527 memcpy(check->bo->data + 11, &gmt_time, 4);
1528 }
1529 else if ((check->type) == PR_O2_HTTP_CHK) {
1530 if (s->proxy->options2 & PR_O2_CHK_SNDST)
1531 bo_putblk(check->bo, trash.str, httpchk_build_status_header(s, trash.str, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001532 /* prevent HTTP keep-alive when "http-check expect" is used */
1533 if (s->proxy->options2 & PR_O2_EXP_TYPE)
1534 bo_putstr(check->bo, "Connection: close\r\n");
Simon Hormanb00d17a2014-06-13 16:18:16 +09001535 bo_putstr(check->bo, "\r\n");
1536 *check->bo->p = '\0'; /* to make gdb output easier to read */
1537 }
1538 }
1539
James Brown55f9ff12015-10-21 18:19:05 -07001540 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
1541 bo_putblk(check->bo, check->send_string, check->send_string_len);
1542 }
1543
Willy Tarreauf411cce2017-10-04 16:21:19 +02001544 /* for tcp-checks, the initial connection setup is handled separately as
1545 * it may be sent to a specific port and not to the server's.
1546 */
1547 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1548 tcpcheck_main(check);
1549 return SF_ERR_UP;
1550 }
1551
Simon Hormanb00d17a2014-06-13 16:18:16 +09001552 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001553 cs = check->cs = cs_new(NULL);
1554 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001555 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001556 conn = cs->conn;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001557
Simon Horman41f58762015-01-30 11:22:56 +09001558 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001559 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001560 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001561 }
1562 else {
1563 /* we'll connect to the addr on the server */
1564 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001565 }
1566
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001567 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
1568 int i = 0;
1569
1570 i = srv_check_healthcheck_port(check);
1571 if (i == 0) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001572 cs->data = check;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001573 return SF_ERR_CHK_PORT;
1574 }
1575
1576 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001577 }
1578
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001579 proto = protocol_by_family(conn->addr.to.ss_family);
1580
1581 conn_prepare(conn, proto, check->xprt);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001582 conn_install_mux(conn, &mux_pt_ops, cs);
1583 cs_attach(cs, check, &check_conn_cb);
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001584 conn->target = &s->obj_type;
1585
1586 /* no client address */
1587 clear_addr(&conn->addr.from);
1588
Willy Tarreauf3d34822014-12-08 12:11:28 +01001589 /* only plain tcp-check supports quick ACK */
1590 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1591
Willy Tarreauf411cce2017-10-04 16:21:19 +02001592 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1593 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001594
Willy Tarreaue7dff022015-04-03 01:14:29 +02001595 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001596 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001597 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001598
1599 if (check->type)
1600 cs_want_send(cs);
1601
Olivier Houchard9130a962017-10-17 17:33:43 +02001602#ifdef USE_OPENSSL
1603 if (s->check.sni)
1604 ssl_sock_set_servername(conn, s->check.sni);
1605#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001606 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001607 conn->send_proxy_ofs = 1;
1608 conn->flags |= CO_FL_SEND_PROXY;
1609 }
1610
1611 return ret;
1612}
1613
Simon Horman98637e52014-06-20 12:30:16 +09001614static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001615static struct pool_head *pool_head_pid_list;
Christopher Faulet9dcf9b62017-11-13 10:34:01 +01001616__decl_hathreads(HA_SPINLOCK_T pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001617
1618void block_sigchld(void)
1619{
1620 sigset_t set;
1621 sigemptyset(&set);
1622 sigaddset(&set, SIGCHLD);
Willy Tarreauebc92442016-06-21 17:29:46 +02001623 assert(sigprocmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001624}
1625
1626void unblock_sigchld(void)
1627{
1628 sigset_t set;
1629 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001630 sigaddset(&set, SIGCHLD);
1631 assert(sigprocmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001632}
1633
Simon Horman98637e52014-06-20 12:30:16 +09001634static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1635{
1636 struct pid_list *elem;
1637 struct check *check = t->context;
1638
Willy Tarreaubafbe012017-11-24 17:34:44 +01001639 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001640 if (!elem)
1641 return NULL;
1642 elem->pid = pid;
1643 elem->t = t;
1644 elem->exited = 0;
1645 check->curpid = elem;
1646 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001647
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001648 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001649 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001650 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001651
Simon Horman98637e52014-06-20 12:30:16 +09001652 return elem;
1653}
1654
Simon Horman98637e52014-06-20 12:30:16 +09001655static void pid_list_del(struct pid_list *elem)
1656{
1657 struct check *check;
1658
1659 if (!elem)
1660 return;
1661
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001662 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001663 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001664 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001665
Simon Horman98637e52014-06-20 12:30:16 +09001666 if (!elem->exited)
1667 kill(elem->pid, SIGTERM);
1668
1669 check = elem->t->context;
1670 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001671 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001672}
1673
1674/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1675static void pid_list_expire(pid_t pid, int status)
1676{
1677 struct pid_list *elem;
1678
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001679 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001680 list_for_each_entry(elem, &pid_list, list) {
1681 if (elem->pid == pid) {
1682 elem->t->expire = now_ms;
1683 elem->status = status;
1684 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001685 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001686 break;
Simon Horman98637e52014-06-20 12:30:16 +09001687 }
1688 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001689 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001690}
1691
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001692static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001693{
1694 pid_t pid;
1695 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001696
Simon Horman98637e52014-06-20 12:30:16 +09001697 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1698 pid_list_expire(pid, status);
1699}
1700
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001701static int init_pid_list(void)
1702{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001703 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001704 /* Nothing to do */
1705 return 0;
1706
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001707 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001708 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1709 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001710 return 1;
1711 }
1712
Willy Tarreaubafbe012017-11-24 17:34:44 +01001713 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1714 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001715 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1716 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001717 return 1;
1718 }
1719
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001720 HA_SPIN_INIT(&pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001721
Simon Horman98637e52014-06-20 12:30:16 +09001722 return 0;
1723}
1724
Cyril Bontéac92a062014-12-27 22:28:38 +01001725/* helper macro to set an environment variable and jump to a specific label on failure. */
1726#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001727
1728/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001729 * helper function to allocate enough memory to store an environment variable.
1730 * It will also check that the environment variable is updatable, and silently
1731 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001732 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001733static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001734{
1735 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001736 char *envname;
1737 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001738
Cyril Bontéac92a062014-12-27 22:28:38 +01001739 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001740 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001741 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001742 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001743
1744 envname = extcheck_envs[idx].name;
1745 vmaxlen = extcheck_envs[idx].vmaxlen;
1746
1747 /* Check if the environment variable is already set, and silently reject
1748 * the update if this one is not updatable. */
1749 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1750 return 0;
1751
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001752 /* Instead of sending NOT_USED, sending an empty value is preferable */
1753 if (strcmp(value, "NOT_USED") == 0) {
1754 value = "";
1755 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001756
1757 len = strlen(envname) + 1;
1758 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1759 len += strlen(value);
1760 else
1761 len += vmaxlen;
1762
1763 if (!check->envp[idx])
1764 check->envp[idx] = malloc(len + 1);
1765
1766 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001767 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001768 return 1;
1769 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001770 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001771 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001772 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001773 return 1;
1774 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001775 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001776 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001777 return 1;
1778 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001779 return 0;
1780}
Simon Horman98637e52014-06-20 12:30:16 +09001781
1782static int prepare_external_check(struct check *check)
1783{
1784 struct server *s = check->server;
1785 struct proxy *px = s->proxy;
1786 struct listener *listener = NULL, *l;
1787 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001788 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001789 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001790
1791 list_for_each_entry(l, &px->conf.listeners, by_fe)
1792 /* Use the first INET, INET6 or UNIX listener */
1793 if (l->addr.ss_family == AF_INET ||
1794 l->addr.ss_family == AF_INET6 ||
1795 l->addr.ss_family == AF_UNIX) {
1796 listener = l;
1797 break;
1798 }
1799
Simon Horman98637e52014-06-20 12:30:16 +09001800 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001801 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1802 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001803 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001804 goto err;
1805 }
Simon Horman98637e52014-06-20 12:30:16 +09001806
Cyril Bontéac92a062014-12-27 22:28:38 +01001807 check->argv = calloc(6, sizeof(char *));
1808 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001809 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001810 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001811 }
Simon Horman98637e52014-06-20 12:30:16 +09001812
1813 check->argv[0] = px->check_command;
1814
Cyril Bonté777be862014-12-02 21:21:35 +01001815 if (!listener) {
1816 check->argv[1] = strdup("NOT_USED");
1817 check->argv[2] = strdup("NOT_USED");
1818 }
1819 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001820 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001821 addr_to_str(&listener->addr, buf, sizeof(buf));
1822 check->argv[1] = strdup(buf);
1823 port_to_str(&listener->addr, buf, sizeof(buf));
1824 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001825 }
1826 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001827 const struct sockaddr_un *un;
1828
1829 un = (struct sockaddr_un *)&listener->addr;
1830 check->argv[1] = strdup(un->sun_path);
1831 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001832 }
1833 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001834 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001835 goto err;
1836 }
1837
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001838 addr_to_str(&s->addr, buf, sizeof(buf));
1839 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001840
1841 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1842 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1843 else
1844 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001845 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001846
Cyril Bontéac92a062014-12-27 22:28:38 +01001847 for (i = 0; i < 5; i++) {
1848 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001849 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001850 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001851 }
1852 }
Simon Horman98637e52014-06-20 12:30:16 +09001853
Cyril Bontéac92a062014-12-27 22:28:38 +01001854 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001855 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001856 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1857 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1858 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1859 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001860 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001861 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1862 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1863 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1864 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1865 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1866 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1867
1868 /* Ensure that we don't leave any hole in check->envp */
1869 for (i = 0; i < EXTCHK_SIZE; i++)
1870 if (!check->envp[i])
1871 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001872
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001873 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001874err:
1875 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001876 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001877 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001878 free(check->envp);
1879 check->envp = NULL;
1880 }
1881
1882 if (check->argv) {
1883 for (i = 1; i < 5; i++)
1884 free(check->argv[i]);
1885 free(check->argv);
1886 check->argv = NULL;
1887 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001888 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001889}
1890
Simon Hormanb00d17a2014-06-13 16:18:16 +09001891/*
Simon Horman98637e52014-06-20 12:30:16 +09001892 * establish a server health-check that makes use of a process.
1893 *
1894 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001895 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001896 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001897 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001898 *
1899 * Blocks and then unblocks SIGCHLD
1900 */
1901static int connect_proc_chk(struct task *t)
1902{
Cyril Bontéac92a062014-12-27 22:28:38 +01001903 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001904 struct check *check = t->context;
1905 struct server *s = check->server;
1906 struct proxy *px = s->proxy;
1907 int status;
1908 pid_t pid;
1909
Willy Tarreaue7dff022015-04-03 01:14:29 +02001910 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001911
1912 block_sigchld();
1913
1914 pid = fork();
1915 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001916 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1917 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001918 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1919 goto out;
1920 }
1921 if (pid == 0) {
1922 /* Child */
1923 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001924 int fd;
1925
1926 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1927 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1928
1929 while (fd < global.rlimit_nofile)
1930 close(fd++);
1931
Simon Horman98637e52014-06-20 12:30:16 +09001932 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001933 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001934 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001935 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1936 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001937 exit(-1);
1938 }
1939
1940 /* Parent */
1941 if (check->result == CHK_RES_UNKNOWN) {
1942 if (pid_list_add(pid, t) != NULL) {
1943 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1944
1945 if (px->timeout.check && px->timeout.connect) {
1946 int t_con = tick_add(now_ms, px->timeout.connect);
1947 t->expire = tick_first(t->expire, t_con);
1948 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001949 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001950 goto out;
1951 }
1952 else {
1953 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1954 }
1955 kill(pid, SIGTERM); /* process creation error */
1956 }
1957 else
1958 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1959
1960out:
1961 unblock_sigchld();
1962 return status;
1963}
1964
1965/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02001966 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001967 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001968 *
1969 * Please do NOT place any return statement in this function and only leave
1970 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001971 */
Simon Horman98637e52014-06-20 12:30:16 +09001972static struct task *process_chk_proc(struct task *t)
1973{
1974 struct check *check = t->context;
1975 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09001976 int rv;
1977 int ret;
1978 int expired = tick_is_expired(t->expire, now_ms);
1979
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001980 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09001981 if (!(check->state & CHK_ST_INPROGRESS)) {
1982 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001983 if (!expired) /* woke up too early */
1984 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09001985
1986 /* we don't send any health-checks when the proxy is
1987 * stopped, the server should not be checked or the check
1988 * is disabled.
1989 */
1990 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
1991 s->proxy->state == PR_STSTOPPED)
1992 goto reschedule;
1993
1994 /* we'll initiate a new check */
1995 set_server_check_status(check, HCHK_STATUS_START, NULL);
1996
1997 check->state |= CHK_ST_INPROGRESS;
1998
Simon Hormandbf70192015-01-30 11:22:53 +09001999 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002000 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002001 /* the process was forked, we allow up to min(inter,
2002 * timeout.connect) for it to report its status, but
2003 * only when timeout.check is set as it may be to short
2004 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002005 */
2006 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2007
2008 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2009 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2010 t->expire = tick_first(t->expire, t_con);
2011 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002012 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002013 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002014 }
2015
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002016 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002017
2018 check->state &= ~CHK_ST_INPROGRESS;
2019 check_notify_failure(check);
2020
2021 /* we allow up to min(inter, timeout.connect) for a connection
2022 * to establish but only when timeout.check is set
2023 * as it may be to short for a full check otherwise
2024 */
2025 while (tick_is_expired(t->expire, now_ms)) {
2026 int t_con;
2027
2028 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2029 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2030
2031 if (s->proxy->timeout.check)
2032 t->expire = tick_first(t->expire, t_con);
2033 }
2034 }
2035 else {
2036 /* there was a test running.
2037 * First, let's check whether there was an uncaught error,
2038 * which can happen on connect timeout or error.
2039 */
2040 if (check->result == CHK_RES_UNKNOWN) {
2041 /* good connection is enough for pure TCP check */
2042 struct pid_list *elem = check->curpid;
2043 int status = HCHK_STATUS_UNKNOWN;
2044
2045 if (elem->exited) {
2046 status = elem->status; /* Save in case the process exits between use below */
2047 if (!WIFEXITED(status))
2048 check->code = -1;
2049 else
2050 check->code = WEXITSTATUS(status);
2051 if (!WIFEXITED(status) || WEXITSTATUS(status))
2052 status = HCHK_STATUS_PROCERR;
2053 else
2054 status = HCHK_STATUS_PROCOK;
2055 } else if (expired) {
2056 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002057 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002058 kill(elem->pid, SIGTERM);
2059 }
2060 set_server_check_status(check, status, NULL);
2061 }
2062
2063 if (check->result == CHK_RES_FAILED) {
2064 /* a failure or timeout detected */
2065 check_notify_failure(check);
2066 }
2067 else if (check->result == CHK_RES_CONDPASS) {
2068 /* check is OK but asks for stopping mode */
2069 check_notify_stopping(check);
2070 }
2071 else if (check->result == CHK_RES_PASSED) {
2072 /* a success was detected */
2073 check_notify_success(check);
2074 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002075 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002076 check->state &= ~CHK_ST_INPROGRESS;
2077
2078 pid_list_del(check->curpid);
2079
2080 rv = 0;
2081 if (global.spread_checks > 0) {
2082 rv = srv_getinter(check) * global.spread_checks / 100;
2083 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2084 }
2085 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2086 }
2087
2088 reschedule:
2089 while (tick_is_expired(t->expire, now_ms))
2090 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002091
2092 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002093 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002094 return t;
2095}
2096
2097/*
2098 * manages a server health-check that uses a connection. Returns
2099 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002100 *
2101 * Please do NOT place any return statement in this function and only leave
2102 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002103 */
2104static struct task *process_chk_conn(struct task *t)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002105{
Simon Horman4a741432013-02-23 15:35:38 +09002106 struct check *check = t->context;
2107 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002108 struct conn_stream *cs = check->cs;
2109 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002110 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002111 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002112 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002113
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002114 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002115 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002116 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002117 if (!expired) /* woke up too early */
2118 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002119
Simon Horman671b6f02013-11-25 10:46:39 +09002120 /* we don't send any health-checks when the proxy is
2121 * stopped, the server should not be checked or the check
2122 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002123 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002124 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002125 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002126 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002127
2128 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002129 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002130
Willy Tarreau2c115e52013-12-11 19:41:16 +01002131 check->state |= CHK_ST_INPROGRESS;
Simon Horman4a741432013-02-23 15:35:38 +09002132 check->bi->p = check->bi->data;
2133 check->bi->i = 0;
2134 check->bo->p = check->bo->data;
2135 check->bo->o = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002136
Simon Hormandbf70192015-01-30 11:22:53 +09002137 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002138 cs = check->cs;
2139 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002140
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002141 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002142 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002143 goto out_unlock;
2144
Willy Tarreaue7dff022015-04-03 01:14:29 +02002145 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002146 /* we allow up to min(inter, timeout.connect) for a connection
2147 * to establish but only when timeout.check is set
2148 * as it may be to short for a full check otherwise
2149 */
Simon Horman4a741432013-02-23 15:35:38 +09002150 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002151
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002152 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2153 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2154 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002155 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002156
2157 if (check->type)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002158 cs_want_recv(cs); /* prepare for reading a possible reply */
Willy Tarreau06559ac2013-12-05 01:53:08 +01002159
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002160 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002161 goto reschedule;
2162
Willy Tarreaue7dff022015-04-03 01:14:29 +02002163 case SF_ERR_SRVTO: /* ETIMEDOUT */
2164 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002165 if (conn)
2166 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002167 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002168 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002169 /* should share same code than cases below */
2170 case SF_ERR_CHK_PORT:
2171 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002172 case SF_ERR_PRXCOND:
2173 case SF_ERR_RESOURCE:
2174 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002175 if (conn)
2176 conn->flags |= CO_FL_ERROR;
2177 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002178 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002179 }
2180
Willy Tarreau5a78f362012-11-23 12:47:05 +01002181 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002182 if (cs) {
2183 cs_destroy(cs);
2184 cs = check->cs = NULL;
2185 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002186 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002187
Willy Tarreau2c115e52013-12-11 19:41:16 +01002188 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002189 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002190
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002191 /* we allow up to min(inter, timeout.connect) for a connection
2192 * to establish but only when timeout.check is set
2193 * as it may be to short for a full check otherwise
2194 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002195 while (tick_is_expired(t->expire, now_ms)) {
2196 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002197
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002198 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002199 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002200
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002201 if (s->proxy->timeout.check)
2202 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002203 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002204 }
2205 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002206 /* there was a test running.
2207 * First, let's check whether there was an uncaught error,
2208 * which can happen on connect timeout or error.
2209 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002210 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002211 /* good connection is enough for pure TCP check */
2212 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002213 if (check->use_ssl)
2214 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002215 else
Simon Horman4a741432013-02-23 15:35:38 +09002216 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002217 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002218 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002219 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002220 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002221 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002222 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002223 }
2224
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002225 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002226 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002227 /* The check was aborted and the connection was not yet closed.
2228 * This can happen upon timeout, or when an external event such
2229 * as a failed response coupled with "observe layer7" caused the
2230 * server state to be suddenly changed.
2231 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002232 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002233 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002234 }
2235
Willy Tarreauac59f362017-10-08 11:10:19 +02002236 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002237 cs_destroy(cs);
2238 cs = check->cs = NULL;
2239 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002240 }
2241
Willy Tarreauaf549582014-05-16 17:37:50 +02002242 if (check->result == CHK_RES_FAILED) {
2243 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002244 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002245 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002246 else if (check->result == CHK_RES_CONDPASS) {
2247 /* check is OK but asks for stopping mode */
2248 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002249 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002250 else if (check->result == CHK_RES_PASSED) {
2251 /* a success was detected */
2252 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002253 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002254 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002255 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002256
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002257 rv = 0;
2258 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002259 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002260 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002261 }
Simon Horman4a741432013-02-23 15:35:38 +09002262 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002263 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002264
2265 reschedule:
2266 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002267 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002268 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002269 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002270 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002271}
2272
Simon Horman98637e52014-06-20 12:30:16 +09002273/*
2274 * manages a server health-check. Returns
2275 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2276 */
2277static struct task *process_chk(struct task *t)
2278{
2279 struct check *check = t->context;
2280
2281 if (check->type == PR_O2_EXT_CHK)
2282 return process_chk_proc(t);
2283 return process_chk_conn(t);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002284
Simon Horman98637e52014-06-20 12:30:16 +09002285}
2286
Simon Horman5c942422013-11-25 10:46:32 +09002287static int start_check_task(struct check *check, int mininter,
2288 int nbcheck, int srvpos)
2289{
2290 struct task *t;
2291 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002292 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002293 ha_alert("Starting [%s:%s] check: out of memory.\n",
2294 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002295 return 0;
2296 }
2297
2298 check->task = t;
2299 t->process = process_chk;
2300 t->context = check;
2301
Willy Tarreau1746eec2014-04-25 10:46:47 +02002302 if (mininter < srv_getinter(check))
2303 mininter = srv_getinter(check);
2304
2305 if (global.max_spread_checks && mininter > global.max_spread_checks)
2306 mininter = global.max_spread_checks;
2307
Simon Horman5c942422013-11-25 10:46:32 +09002308 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002309 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002310 check->start = now;
2311 task_queue(t);
2312
2313 return 1;
2314}
2315
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002316/*
2317 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002318 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002319 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002320static int start_checks()
2321{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002322
2323 struct proxy *px;
2324 struct server *s;
2325 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002326 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002327
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002328 /* 1- count the checkers to run simultaneously.
2329 * We also determine the minimum interval among all of those which
2330 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2331 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002332 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002333 * too short an interval for all others.
2334 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002335 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002336 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002337 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002338 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002339 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002340 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002341 }
2342 /* We need a warmup task that will be called when the server
2343 * state switches from down to up.
2344 */
2345 s->warmup = t;
2346 t->process = server_warmup;
2347 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002348 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002349 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002350 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 +01002351 }
2352
Willy Tarreaud8514a22013-12-11 21:10:14 +01002353 if (s->check.state & CHK_ST_CONFIGURED) {
2354 nbcheck++;
2355 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2356 (!mininter || mininter > srv_getinter(&s->check)))
2357 mininter = srv_getinter(&s->check);
2358 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002359
Willy Tarreaud8514a22013-12-11 21:10:14 +01002360 if (s->agent.state & CHK_ST_CONFIGURED) {
2361 nbcheck++;
2362 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2363 (!mininter || mininter > srv_getinter(&s->agent)))
2364 mininter = srv_getinter(&s->agent);
2365 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002366 }
2367 }
2368
Simon Horman4a741432013-02-23 15:35:38 +09002369 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002370 return 0;
2371
2372 srand((unsigned)time(NULL));
2373
2374 /*
2375 * 2- start them as far as possible from each others. For this, we will
2376 * start them after their interval set to the min interval divided by
2377 * the number of servers, weighted by the server's position in the list.
2378 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002379 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002380 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2381 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002382 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002383 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002384 }
2385 }
2386
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002387 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002388 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002389 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002390 if (s->check.type == PR_O2_EXT_CHK) {
2391 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002392 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002393 }
Simon Hormand60d6912013-11-25 10:46:36 +09002394 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002395 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002396 srvpos++;
2397 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002398
Simon Hormand60d6912013-11-25 10:46:36 +09002399 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002400 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002401 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002402 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002403 }
2404 srvpos++;
2405 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002406 }
2407 }
2408 return 0;
2409}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002410
2411/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002412 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002413 * The buffer MUST be terminated by a null byte before calling this function.
2414 * Sets server status appropriately. The caller is responsible for ensuring
2415 * that the buffer contains at least 13 characters. If <done> is zero, we may
2416 * return 0 to indicate that data is required to decide of a match.
2417 */
2418static int httpchk_expect(struct server *s, int done)
2419{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002420 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002421 char status_code[] = "000";
2422 char *contentptr;
2423 int crlf;
2424 int ret;
2425
2426 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2427 case PR_O2_EXP_STS:
2428 case PR_O2_EXP_RSTS:
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002429 memcpy(status_code, s->check.bi->data + 9, 3);
2430 memcpy(status_msg + strlen(status_msg) - 4, s->check.bi->data + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002431
2432 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2433 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2434 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002435 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002436
2437 /* we necessarily have the response, so there are no partial failures */
2438 if (s->proxy->options2 & PR_O2_EXP_INV)
2439 ret = !ret;
2440
Simon Horman4a741432013-02-23 15:35:38 +09002441 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002442 break;
2443
2444 case PR_O2_EXP_STR:
2445 case PR_O2_EXP_RSTR:
2446 /* very simple response parser: ignore CR and only count consecutive LFs,
2447 * stop with contentptr pointing to first char after the double CRLF or
2448 * to '\0' if crlf < 2.
2449 */
2450 crlf = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002451 for (contentptr = s->check.bi->data; *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002452 if (crlf >= 2)
2453 break;
2454 if (*contentptr == '\r')
2455 continue;
2456 else if (*contentptr == '\n')
2457 crlf++;
2458 else
2459 crlf = 0;
2460 }
2461
2462 /* Check that response contains a body... */
2463 if (crlf < 2) {
2464 if (!done)
2465 return 0;
2466
Simon Horman4a741432013-02-23 15:35:38 +09002467 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002468 "HTTP content check could not find a response body");
2469 return 1;
2470 }
2471
2472 /* Check that response body is not empty... */
2473 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002474 if (!done)
2475 return 0;
2476
Simon Horman4a741432013-02-23 15:35:38 +09002477 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002478 "HTTP content check found empty response body");
2479 return 1;
2480 }
2481
2482 /* Check the response content against the supplied string
2483 * or regex... */
2484 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2485 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2486 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002487 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002488
2489 /* if we don't match, we may need to wait more */
2490 if (!ret && !done)
2491 return 0;
2492
2493 if (ret) {
2494 /* content matched */
2495 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002496 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002497 "HTTP check matched unwanted content");
2498 else
Simon Horman4a741432013-02-23 15:35:38 +09002499 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002500 "HTTP content check matched");
2501 }
2502 else {
2503 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002504 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002505 "HTTP check did not match unwanted content");
2506 else
Simon Horman4a741432013-02-23 15:35:38 +09002507 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002508 "HTTP content check did not match");
2509 }
2510 break;
2511 }
2512 return 1;
2513}
2514
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002515/*
2516 * return the id of a step in a send/expect session
2517 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002518static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002519{
2520 struct tcpcheck_rule *cur = NULL, *next = NULL;
2521 int i = 0;
2522
Willy Tarreau213c6782014-10-02 14:51:02 +02002523 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002524 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002525 return 0;
2526
Simon Hormane16c1b32015-01-30 11:22:57 +09002527 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002528
2529 /* no step => first step */
2530 if (cur == NULL)
2531 return 1;
2532
2533 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002534 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002535 if (next->list.p == &cur->list)
2536 break;
2537 ++i;
2538 }
2539
2540 return i;
2541}
2542
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002543/*
2544 * return the latest known comment before (including) the given stepid
2545 * returns NULL if no comment found
2546 */
2547static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2548{
2549 struct tcpcheck_rule *cur = NULL;
2550 char *ret = NULL;
2551 int i = 0;
2552
2553 /* not even started anything yet, return latest comment found before any action */
2554 if (!check->current_step) {
2555 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2556 if (cur->action == TCPCHK_ACT_COMMENT)
2557 ret = cur->comment;
2558 else
2559 goto return_comment;
2560 }
2561 }
2562
2563 i = 1;
2564 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2565 if (cur->comment)
2566 ret = cur->comment;
2567
2568 if (i >= stepid)
2569 goto return_comment;
2570
2571 ++i;
2572 }
2573
2574 return_comment:
2575 return ret;
2576}
2577
Willy Tarreaube74b882017-10-04 16:22:49 +02002578/* proceed with next steps for the TCP checks <check>. Note that this is called
2579 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002580 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2581 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002582 *
2583 * Please do NOT place any return statement in this function and only leave
2584 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002585 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002586static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002587{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002588 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002589 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002590 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002591 struct conn_stream *cs = check->cs;
2592 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002593 struct server *s = check->server;
2594 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002595 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002596 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002597
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002598 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01002599
Willy Tarreauef953952014-10-02 14:30:14 +02002600 /* here, we know that the check is complete or that it failed */
2601 if (check->result != CHK_RES_UNKNOWN)
2602 goto out_end_tcpcheck;
2603
2604 /* We have 4 possibilities here :
2605 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002606 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002607 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2608 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002609 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002610 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002611 * be waiting for a connection attempt to complete. conn!=NULL.
2612 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002613 *
2614 * #2 and #3 are quite similar, we want both the connection and the
2615 * handshake to complete before going any further. Thus we must always
2616 * wait for a connection to complete unless we're before and existing
2617 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002618 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002619
2620 /* find first rule and skip comments */
2621 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2622 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2623 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2624
Willy Tarreau00149122017-10-04 18:05:01 +02002625 if ((check->current_step || &next->list == head) &&
2626 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002627 /* we allow up to min(inter, timeout.connect) for a connection
2628 * to establish but only when timeout.check is set
2629 * as it may be to short for a full check otherwise
2630 */
2631 while (tick_is_expired(t->expire, now_ms)) {
2632 int t_con;
2633
2634 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2635 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2636
2637 if (s->proxy->timeout.check)
2638 t->expire = tick_first(t->expire, t_con);
2639 }
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002640 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002641 }
2642
Willy Tarreauef953952014-10-02 14:30:14 +02002643 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002644 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002645 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002646 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002647 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002648
Willy Tarreau213c6782014-10-02 14:51:02 +02002649 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002650 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002651 check->last_started_step = NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002652 check->bo->p = check->bo->data;
2653 check->bo->o = 0;
2654 check->bi->p = check->bi->data;
2655 check->bi->i = 0;
Willy Tarreau449f9522015-05-13 15:39:48 +02002656 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002657 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2658 if (s->proxy->timeout.check)
2659 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2660 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002661
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002662 /* It's only the rules which will enable send/recv */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002663 if (cs)
2664 cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002665
Willy Tarreauabca5b62013-12-06 14:19:25 +01002666 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002667 /* We have to try to flush the output buffer before reading, at
2668 * the end, or if we're about to send a string that does not fit
2669 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002670 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002671 */
2672 if (check->bo->o &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002673 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002674 check->current_step->action != TCPCHK_ACT_SEND ||
2675 check->current_step->string_len >= buffer_total_space(check->bo))) {
2676
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002677 __cs_want_send(cs);
2678 if (conn->mux->snd_buf(cs, check->bo, 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002679 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002680 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002681 __cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002682 goto out_end_tcpcheck;
2683 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002684 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002685 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002686 }
2687
Willy Tarreau263013d2015-05-13 11:59:14 +02002688 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002689 break;
2690
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002691 /* have 'next' point to the next rule or NULL if we're on the
2692 * last one, connect() needs this.
2693 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002694 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002695
2696 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002697 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002698 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002699
2700 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002701 if (&next->list == head)
2702 next = NULL;
2703
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002704 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2705 struct protocol *proto;
2706 struct xprt_ops *xprt;
2707
Willy Tarreau00149122017-10-04 18:05:01 +02002708 /* For a connect action we'll create a new connection.
2709 * We may also have to kill a previous one. But we don't
2710 * want to leave *without* a connection if we came here
2711 * from the connection layer, hence with a connection.
2712 * Thus we'll proceed in the following order :
2713 * 1: close but not release previous connection
2714 * 2: try to get a new connection
2715 * 3: release and replace the old one on success
2716 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002717 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002718 cs_close(check->cs);
2719 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002720 }
2721
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002722 /* mark the step as started */
2723 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002724
2725 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002726 cs = cs_new(NULL);
2727 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002728 step = tcpcheck_get_step_id(check);
2729 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2730 comment = tcpcheck_get_step_comment(check, step);
2731 if (comment)
2732 chunk_appendf(&trash, " comment: '%s'", comment);
2733 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2734 check->current_step = NULL;
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002735 goto out_unlock;
Willy Tarreau00149122017-10-04 18:05:01 +02002736 }
2737
Willy Tarreauac59f362017-10-08 11:10:19 +02002738 if (check->cs)
2739 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002740
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002741 check->cs = cs;
2742 conn = cs->conn;
2743 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002744 conn->target = &s->obj_type;
2745
2746 /* no client address */
2747 clear_addr(&conn->addr.from);
2748
Simon Horman41f58762015-01-30 11:22:56 +09002749 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002750 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002751 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002752 }
2753 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002754 /* we'll connect to the addr on the server */
2755 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002756 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002757 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002758
2759 /* port */
2760 if (check->current_step->port)
2761 set_host_port(&conn->addr.to, check->current_step->port);
2762 else if (check->port)
2763 set_host_port(&conn->addr.to, check->port);
2764
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002765 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002766 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002767 }
2768 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002769 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002770 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002771 conn_prepare(conn, proto, xprt);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002772 conn_install_mux(conn, &mux_pt_ops, cs);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002773
Willy Tarreaue7dff022015-04-03 01:14:29 +02002774 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002775 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002776 ret = proto->connect(conn,
2777 1 /* I/O polling is always needed */,
2778 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002779 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2780 conn->send_proxy_ofs = 1;
2781 conn->flags |= CO_FL_SEND_PROXY;
2782 }
2783
2784 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002785 * - SF_ERR_NONE if everything's OK
2786 * - SF_ERR_SRVTO if there are no more servers
2787 * - SF_ERR_SRVCL if the connection was refused by the server
2788 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2789 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2790 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002791 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002792 * Note that we try to prevent the network stack from sending the ACK during the
2793 * connect() when a pure TCP check is used (without PROXY protocol).
2794 */
2795 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002796 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002797 /* we allow up to min(inter, timeout.connect) for a connection
2798 * to establish but only when timeout.check is set
2799 * as it may be to short for a full check otherwise
2800 */
2801 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2802
2803 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2804 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2805 t->expire = tick_first(t->expire, t_con);
2806 }
2807 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002808 case SF_ERR_SRVTO: /* ETIMEDOUT */
2809 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002810 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002811 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002812 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002813 comment = tcpcheck_get_step_comment(check, step);
2814 if (comment)
2815 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002816 set_server_check_status(check, HCHK_STATUS_L4CON, trash.str);
2817 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002818 case SF_ERR_PRXCOND:
2819 case SF_ERR_RESOURCE:
2820 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002821 step = tcpcheck_get_step_id(check);
2822 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002823 comment = tcpcheck_get_step_comment(check, step);
2824 if (comment)
2825 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002826 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2827 goto out_end_tcpcheck;
2828 }
2829
2830 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002831 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002832
2833 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002834 while (&check->current_step->list != head &&
2835 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002836 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002837
Willy Tarreauf2c87352015-05-13 12:08:21 +02002838 if (&check->current_step->list == head)
2839 break;
2840
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002841 /* don't do anything until the connection is established */
2842 if (!(conn->flags & CO_FL_CONNECTED)) {
2843 /* update expire time, should be done by process_chk */
2844 /* we allow up to min(inter, timeout.connect) for a connection
2845 * to establish but only when timeout.check is set
2846 * as it may be to short for a full check otherwise
2847 */
2848 while (tick_is_expired(t->expire, now_ms)) {
2849 int t_con;
2850
2851 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2852 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2853
2854 if (s->proxy->timeout.check)
2855 t->expire = tick_first(t->expire, t_con);
2856 }
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002857 goto out_unlock;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002858 }
2859
2860 } /* end 'connect' */
2861 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2862 /* mark the step as started */
2863 check->last_started_step = check->current_step;
2864
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002865 /* reset the read buffer */
2866 if (*check->bi->data != '\0') {
2867 *check->bi->data = '\0';
2868 check->bi->i = 0;
2869 }
2870
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002871 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002872 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002873 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002874 goto out_end_tcpcheck;
2875 }
2876
Willy Tarreauabca5b62013-12-06 14:19:25 +01002877 if (check->current_step->string_len >= check->bo->size) {
2878 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%d) at step %d",
2879 check->current_step->string_len, check->bo->size,
Simon Hormane16c1b32015-01-30 11:22:57 +09002880 tcpcheck_get_step_id(check));
Willy Tarreauabca5b62013-12-06 14:19:25 +01002881 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2882 goto out_end_tcpcheck;
2883 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002884
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002885 /* do not try to send if there is no space */
2886 if (check->current_step->string_len >= buffer_total_space(check->bo))
2887 continue;
2888
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002889 bo_putblk(check->bo, check->current_step->string, check->current_step->string_len);
2890 *check->bo->p = '\0'; /* to make gdb output easier to read */
2891
Willy Tarreauabca5b62013-12-06 14:19:25 +01002892 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002893 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002894
2895 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002896 while (&check->current_step->list != head &&
2897 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002898 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002899
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002900 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002901 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002902 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002903 goto out_end_tcpcheck;
2904
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002905 __cs_want_recv(cs);
2906 if (conn->mux->rcv_buf(cs, check->bi, check->bi->size) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002907 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002908 done = 1;
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002909 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !check->bi->i) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002910 /* Report network errors only if we got no other data. Otherwise
2911 * we'll let the upper layers decide whether the response is OK
2912 * or not. It is very common that an RST sent by the server is
2913 * reported as an error just after the last data chunk.
2914 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002915 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002916 goto out_end_tcpcheck;
2917 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002918 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002919 else
Willy Tarreau263013d2015-05-13 11:59:14 +02002920 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002921 }
2922
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002923 /* mark the step as started */
2924 check->last_started_step = check->current_step;
2925
2926
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002927 /* Intermediate or complete response received.
2928 * Terminate string in check->bi->data buffer.
2929 */
2930 if (check->bi->i < check->bi->size) {
2931 check->bi->data[check->bi->i] = '\0';
2932 }
2933 else {
2934 check->bi->data[check->bi->i - 1] = '\0';
2935 done = 1; /* buffer full, don't wait for more data */
2936 }
2937
2938 contentptr = check->bi->data;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002939
2940 /* Check that response body is not empty... */
Willy Tarreauec6b0122014-05-13 17:57:29 +02002941 if (!check->bi->i) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002942 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002943 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002944
2945 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002946 step = tcpcheck_get_step_id(check);
2947 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002948 comment = tcpcheck_get_step_comment(check, step);
2949 if (comment)
2950 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002951 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2952
2953 goto out_end_tcpcheck;
2954 }
2955
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002956 if (!done && (check->current_step->string != NULL) && (check->bi->i < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002957 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002958
Willy Tarreaua970c282013-12-06 12:47:19 +01002959 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002960 if (check->current_step->string != NULL)
2961 ret = my_memmem(contentptr, check->bi->i, check->current_step->string, check->current_step->string_len) != NULL;
2962 else if (check->current_step->expect_regex != NULL)
2963 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002964
2965 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002966 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002967
2968 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002969 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002970 if (ret) {
2971 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002972 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002973 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002974 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002975 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002976 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977 }
2978 else {
2979 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002980 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002981 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002982 comment = tcpcheck_get_step_comment(check, step);
2983 if (comment)
2984 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002985 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2986 goto out_end_tcpcheck;
2987 }
2988 /* matched and was supposed to => OK, next step */
2989 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002990 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002991 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002992
2993 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002994 while (&check->current_step->list != head &&
2995 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002996 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002997
Willy Tarreauf2c87352015-05-13 12:08:21 +02002998 if (&check->current_step->list == head)
2999 break;
3000
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003001 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003002 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003003 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003004 }
3005 }
3006 else {
3007 /* not matched */
3008 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003009 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003010 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003011 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003012
3013 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003014 while (&check->current_step->list != head &&
3015 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003016 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003017
Willy Tarreauf2c87352015-05-13 12:08:21 +02003018 if (&check->current_step->list == head)
3019 break;
3020
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003021 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003022 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003023 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003024 }
3025 /* not matched but was supposed to => ERROR */
3026 else {
3027 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003028 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003029 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003030 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003031 }
3032 else {
3033 /* we were looking for a regex */
3034 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003035 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003036 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003037 comment = tcpcheck_get_step_comment(check, step);
3038 if (comment)
3039 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003040 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
3041 goto out_end_tcpcheck;
3042 }
3043 }
3044 } /* end expect */
3045 } /* end loop over double chained step list */
3046
Willy Tarreau263013d2015-05-13 11:59:14 +02003047 /* We're waiting for some I/O to complete, we've reached the end of the
3048 * rules, or both. Do what we have to do, otherwise we're done.
3049 */
3050 if (&check->current_step->list == head && !check->bo->o) {
3051 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3052 goto out_end_tcpcheck;
3053 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003054
Willy Tarreau53c5a042015-05-13 11:38:17 +02003055 /* warning, current_step may now point to the head */
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003056 if (check->bo->o)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003057 __cs_want_send(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003058
Willy Tarreau53c5a042015-05-13 11:38:17 +02003059 if (&check->current_step->list != head &&
3060 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003061 __cs_want_recv(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01003062 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003063
3064 out_end_tcpcheck:
3065 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003066 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003067 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003068
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003069 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003070 check->current_step = NULL;
3071
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003072 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003073 conn->flags |= CO_FL_ERROR;
3074
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003075 __cs_stop_both(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01003076
3077 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003078 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003079 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003080}
3081
Simon Hormanb1900d52015-01-30 11:22:54 +09003082const char *init_check(struct check *check, int type)
3083{
3084 check->type = type;
3085
3086 /* Allocate buffer for requests... */
3087 if ((check->bi = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3088 return "out of memory while allocating check buffer";
3089 }
3090 check->bi->size = global.tune.chksize;
3091
3092 /* Allocate buffer for responses... */
3093 if ((check->bo = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3094 return "out of memory while allocating check buffer";
3095 }
3096 check->bo->size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003097 return NULL;
3098}
3099
Simon Hormanbfb5d332015-01-30 11:22:55 +09003100void free_check(struct check *check)
3101{
3102 free(check->bi);
Frédéric Lécaille6e0843c2017-03-21 16:39:15 +01003103 check->bi = NULL;
Simon Hormanbfb5d332015-01-30 11:22:55 +09003104 free(check->bo);
Frédéric Lécaille6e0843c2017-03-21 16:39:15 +01003105 check->bo = NULL;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003106 free(check->cs->conn);
3107 check->cs->conn = NULL;
3108 cs_free(check->cs);
3109 check->cs = NULL;
Simon Hormanbfb5d332015-01-30 11:22:55 +09003110}
3111
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003112void email_alert_free(struct email_alert *alert)
3113{
3114 struct tcpcheck_rule *rule, *back;
3115
3116 if (!alert)
3117 return;
3118
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003119 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3120 LIST_DEL(&rule->list);
3121 free(rule->comment);
3122 free(rule->string);
3123 if (rule->expect_regex)
3124 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003125 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003126 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003127 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003128}
3129
3130static struct task *process_email_alert(struct task *t)
3131{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003132 struct check *check = t->context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003133 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003134 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003135
3136 q = container_of(check, typeof(*q), check);
3137
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003138 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003139 while (1) {
3140 if (!(check->state & CHK_ST_ENABLED)) {
3141 if (LIST_ISEMPTY(&q->email_alerts)) {
3142 /* All alerts processed, queue the task */
3143 t->expire = TICK_ETERNITY;
3144 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003145 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003146 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003147
3148 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003149 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003150 t->expire = now_ms;
3151 check->server = alert->srv;
3152 check->tcpcheck_rules = &alert->tcpcheck_rules;
3153 check->status = HCHK_STATUS_INI;
3154 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003155 }
3156
Christopher Faulet0108bb32017-10-20 21:34:32 +02003157 process_chk(t);
3158 if (check->state & CHK_ST_INPROGRESS)
3159 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003160
3161 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3162 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003163 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003164 check->server = NULL;
3165 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003166 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003167 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003168 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003169 return t;
3170}
3171
Christopher Faulet0108bb32017-10-20 21:34:32 +02003172/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3173 *
3174 * The function returns 1 in success case, otherwise, it returns 0 and err is
3175 * filled.
3176 */
3177int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003178{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003179 struct mailer *mailer;
3180 struct email_alertq *queues;
3181 const char *err_str;
3182 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003183
Christopher Faulet0108bb32017-10-20 21:34:32 +02003184 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3185 memprintf(err, "out of memory while allocating mailer alerts queues");
3186 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003187 }
3188
Christopher Faulet0108bb32017-10-20 21:34:32 +02003189 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3190 struct email_alertq *q = &queues[i];
3191 struct check *check = &q->check;
3192 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003193
3194 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003195 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003196 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003197 check->rise = DEF_AGENT_RISETIME;
3198 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003199 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3200 memprintf(err, "%s", err_str);
3201 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003202 }
3203
3204 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003205 check->addr = mailer->addr;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003206 if (!get_host_port(&mailer->addr))
3207 /* Default to submission port */
3208 check->port = 587;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003209 //check->server = s;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003210
Emeric Brunc60def82017-09-27 14:59:38 +02003211 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003212 memprintf(err, "out of memory while allocating mailer alerts task");
3213 goto error;
3214 }
3215
3216 check->task = t;
3217 t->process = process_email_alert;
3218 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003219
Christopher Faulet0108bb32017-10-20 21:34:32 +02003220 /* check this in one ms */
3221 t->expire = TICK_ETERNITY;
3222 check->start = now;
3223 task_queue(t);
3224 }
3225
3226 mls->users++;
3227 free(p->email_alert.mailers.name);
3228 p->email_alert.mailers.m = mls;
3229 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003230 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003231
3232 error:
3233 for (i = 0; i < mls->count; i++) {
3234 struct email_alertq *q = &queues[i];
3235 struct check *check = &q->check;
3236
3237 if (check->task) {
3238 task_delete(check->task);
3239 task_free(check->task);
3240 check->task = NULL;
3241 }
3242 free_check(check);
3243 }
3244 free(queues);
3245 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003246}
3247
3248
3249static int add_tcpcheck_expect_str(struct list *list, const char *str)
3250{
3251 struct tcpcheck_rule *tcpcheck;
3252
Willy Tarreaubafbe012017-11-24 17:34:44 +01003253 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003254 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003255 memset(tcpcheck, 0, sizeof(*tcpcheck));
3256 tcpcheck->action = TCPCHK_ACT_EXPECT;
3257 tcpcheck->string = strdup(str);
3258 tcpcheck->expect_regex = NULL;
3259 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003260 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003261 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003262 return 0;
3263 }
3264
3265 LIST_ADDQ(list, &tcpcheck->list);
3266 return 1;
3267}
3268
3269static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3270{
3271 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003272 const char *in;
3273 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003274 int i;
3275
Willy Tarreaubafbe012017-11-24 17:34:44 +01003276 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003277 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003278 memset(tcpcheck, 0, sizeof(*tcpcheck));
3279 tcpcheck->action = TCPCHK_ACT_SEND;
3280 tcpcheck->expect_regex = NULL;
3281 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003282 tcpcheck->string_len = 0;
3283 for (i = 0; strs[i]; i++)
3284 tcpcheck->string_len += strlen(strs[i]);
3285
3286 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3287 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003288 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003289 return 0;
3290 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003291
Willy Tarreau64345aa2016-08-10 19:29:09 +02003292 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003293 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003294 for (in = strs[i]; (*dst = *in++); dst++);
3295 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003296
3297 LIST_ADDQ(list, &tcpcheck->list);
3298 return 1;
3299}
3300
Christopher Faulet0108bb32017-10-20 21:34:32 +02003301static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3302 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003303{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003304 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003305 struct tcpcheck_rule *tcpcheck;
3306 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003307
Willy Tarreaubafbe012017-11-24 17:34:44 +01003308 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003309 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003310 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003311 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003312 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003313
Willy Tarreaubafbe012017-11-24 17:34:44 +01003314 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003315 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003316 memset(tcpcheck, 0, sizeof(*tcpcheck));
3317 tcpcheck->action = TCPCHK_ACT_CONNECT;
3318 tcpcheck->comment = NULL;
3319 tcpcheck->string = NULL;
3320 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003321 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3322
3323 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3324 goto error;
3325
3326 {
3327 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3328 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3329 goto error;
3330 }
3331
3332 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3333 goto error;
3334
3335 {
3336 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3337 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3338 goto error;
3339 }
3340
3341 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3342 goto error;
3343
3344 {
3345 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3346 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3347 goto error;
3348 }
3349
3350 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3351 goto error;
3352
3353 {
3354 const char * const strs[2] = { "DATA\r\n" };
3355 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3356 goto error;
3357 }
3358
3359 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3360 goto error;
3361
3362 {
3363 struct tm tm;
3364 char datestr[48];
3365 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003366 "From: ", p->email_alert.from, "\r\n",
3367 "To: ", p->email_alert.to, "\r\n",
3368 "Date: ", datestr, "\r\n",
3369 "Subject: [HAproxy Alert] ", msg, "\r\n",
3370 "\r\n",
3371 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003372 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003373 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003374 NULL
3375 };
3376
3377 get_localtime(date.tv_sec, &tm);
3378
3379 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3380 goto error;
3381 }
3382
3383 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3384 goto error;
3385 }
3386
3387 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3388 goto error;
3389
3390 {
3391 const char * const strs[2] = { "QUIT\r\n" };
3392 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3393 goto error;
3394 }
3395
3396 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3397 goto error;
3398
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003399 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003400 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003401 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003402 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003403 return 1;
3404
3405error:
3406 email_alert_free(alert);
3407 return 0;
3408}
3409
Christopher Faulet0108bb32017-10-20 21:34:32 +02003410static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003411{
3412 int i;
3413 struct mailer *mailer;
3414
3415 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3416 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003417 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003418 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003419 return;
3420 }
3421 }
3422
3423 return;
3424}
3425
3426/*
3427 * Send email alert if configured.
3428 */
Simon Horman64e34162015-02-06 11:11:57 +09003429void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003430{
3431 va_list argp;
3432 char buf[1024];
3433 int len;
3434 struct proxy *p = s->proxy;
3435
Christopher Faulet0108bb32017-10-20 21:34:32 +02003436 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003437 return;
3438
3439 va_start(argp, format);
3440 len = vsnprintf(buf, sizeof(buf), format, argp);
3441 va_end(argp);
3442
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003443 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003444 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003445 return;
3446 }
3447
Christopher Faulet0108bb32017-10-20 21:34:32 +02003448 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003449}
3450
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003451/*
3452 * Return value:
3453 * the port to be used for the health check
3454 * 0 in case no port could be found for the check
3455 */
3456int srv_check_healthcheck_port(struct check *chk)
3457{
3458 int i = 0;
3459 struct server *srv = NULL;
3460
3461 srv = chk->server;
3462
3463 /* If neither a port nor an addr was specified and no check transport
3464 * layer is forced, then the transport layer used by the checks is the
3465 * same as for the production traffic. Otherwise we use raw_sock by
3466 * default, unless one is specified.
3467 */
3468 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003469 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003470 chk->send_proxy |= (srv->pp_opts);
3471 }
3472
3473 /* by default, we use the health check port ocnfigured */
3474 if (chk->port > 0)
3475 return chk->port;
3476
3477 /* try to get the port from check_core.addr if check.port not set */
3478 i = get_host_port(&chk->addr);
3479 if (i > 0)
3480 return i;
3481
3482 /* try to get the port from server address */
3483 /* prevent MAPPORTS from working at this point, since checks could
3484 * not be performed in such case (MAPPORTS impose a relative ports
3485 * based on live traffic)
3486 */
3487 if (srv->flags & SRV_F_MAPPORTS)
3488 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003489
3490 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003491 if (i > 0)
3492 return i;
3493
3494 return 0;
3495}
3496
Willy Tarreau865c5142016-12-21 20:04:48 +01003497__attribute__((constructor))
3498static void __check_init(void)
3499{
3500 hap_register_post_check(start_checks);
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003501
Willy Tarreaubafbe012017-11-24 17:34:44 +01003502 pool_head_email_alert = create_pool("email_alert", sizeof(struct email_alert), MEM_F_SHARED);
3503 pool_head_tcpcheck_rule = create_pool("tcpcheck_rule", sizeof(struct tcpcheck_rule), MEM_F_SHARED);
Willy Tarreau865c5142016-12-21 20:04:48 +01003504}
3505
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003506
Willy Tarreaubd741542010-03-16 18:46:54 +01003507/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003508 * Local variables:
3509 * c-indent-level: 8
3510 * c-basic-offset: 8
3511 * End:
3512 */