blob: af6ddd453a1af265e8ba879c9af5fcc09bc2b9c2 [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 *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020072static void __event_srv_chk_w(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010073
Willy Tarreaubafbe012017-11-24 17:34:44 +010074static struct pool_head *pool_head_email_alert = NULL;
75static struct pool_head *pool_head_tcpcheck_rule = NULL;
Christopher Faulet31dff9b2017-10-23 15:45:20 +020076
77
Simon Horman63a4a822012-03-19 07:24:41 +090078static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010079 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
80 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020081 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020082
Willy Tarreau23964182014-05-20 20:56:30 +020083 /* Below we have finished checks */
84 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010085 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010086
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010087 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020088
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010089 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
90 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
91 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020092
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010093 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
94 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
95 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020096
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010097 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
98 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020099
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200100 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200101
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100102 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
103 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
104 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900105
106 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
107 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200108 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200109};
110
Cyril Bontéac92a062014-12-27 22:28:38 +0100111const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
112 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
122 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
123};
124
Simon Horman63a4a822012-03-19 07:24:41 +0900125static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100126 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
127
128 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
129 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
130
131 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
132 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
133 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
134 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
135
136 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
138 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
139};
140
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200141/*
142 * Convert check_status code to description
143 */
144const char *get_check_status_description(short check_status) {
145
146 const char *desc;
147
148 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200149 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200150 else
151 desc = NULL;
152
153 if (desc && *desc)
154 return desc;
155 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200156 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200157}
158
159/*
160 * Convert check_status code to short info
161 */
162const char *get_check_status_info(short check_status) {
163
164 const char *info;
165
166 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200167 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200168 else
169 info = NULL;
170
171 if (info && *info)
172 return info;
173 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200174 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200175}
176
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100177const char *get_analyze_status(short analyze_status) {
178
179 const char *desc;
180
181 if (analyze_status < HANA_STATUS_SIZE)
182 desc = analyze_statuses[analyze_status].desc;
183 else
184 desc = NULL;
185
186 if (desc && *desc)
187 return desc;
188 else
189 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
190}
191
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200192/*
Simon Horman4a741432013-02-23 15:35:38 +0900193 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200194 * an adequate CHK_RES_* value. The new check->health is computed based
195 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200196 *
197 * Show information in logs about failed health check if server is UP
198 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200199 */
Simon Horman4a741432013-02-23 15:35:38 +0900200static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100201{
Simon Horman4a741432013-02-23 15:35:38 +0900202 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200203 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200204 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900205
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200206 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100207 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900208 check->desc[0] = '\0';
209 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200210 return;
211 }
212
Simon Horman4a741432013-02-23 15:35:38 +0900213 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200214 return;
215
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200216 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900217 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
218 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200219 } else
Simon Horman4a741432013-02-23 15:35:38 +0900220 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200221
Simon Horman4a741432013-02-23 15:35:38 +0900222 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200223 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900224 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200225
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100226 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900227 check->duration = -1;
228 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200229 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900230 check->duration = tv_ms_elapsed(&check->start, &now);
231 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200232 }
233
Willy Tarreau23964182014-05-20 20:56:30 +0200234 /* no change is expected if no state change occurred */
235 if (check->result == CHK_RES_NEUTRAL)
236 return;
237
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200238 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200239
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200240 switch (check->result) {
241 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200242 /* Failure to connect to the agent as a secondary check should not
243 * cause the server to be marked down.
244 */
245 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900246 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200247 (check->health > 0)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200248 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200249 report = 1;
250 check->health--;
251 if (check->health < check->rise)
252 check->health = 0;
253 }
254 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200255
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200256 case CHK_RES_PASSED:
257 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
258 if ((check->health < check->rise + check->fall - 1) &&
259 (check->result == CHK_RES_PASSED || check->health > 0)) {
260 report = 1;
261 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200262
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200263 if (check->health >= check->rise)
264 check->health = check->rise + check->fall - 1; /* OK now */
265 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200266
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200267 /* clear consecutive_errors if observing is enabled */
268 if (s->onerror)
269 s->consecutive_errors = 0;
270 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100271
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200272 default:
273 break;
274 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200275
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200276 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
277 (status != prev_status || report)) {
278 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200279 "%s check for %sserver %s/%s %s%s",
280 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200281 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100282 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100283 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200284 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200285
Emeric Brun5a133512017-10-19 14:42:30 +0200286 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200287
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100288 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200289 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
290 (check->health >= check->rise) ? check->fall : check->rise,
291 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200292
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200293 ha_warning("%s.\n", trash.area);
294 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
295 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200296 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200297}
298
Willy Tarreau4eec5472014-05-20 22:32:27 +0200299/* Marks the check <check>'s server down if the current check is already failed
300 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200301 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200302static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200303{
Simon Horman4a741432013-02-23 15:35:38 +0900304 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900305
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200306 /* The agent secondary check should only cause a server to be marked
307 * as down if check->status is HCHK_STATUS_L7STS, which indicates
308 * that the agent returned "fail", "stopped" or "down".
309 * The implication here is that failure to connect to the agent
310 * as a secondary check should not cause the server to be marked
311 * down. */
312 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
313 return;
314
Willy Tarreau4eec5472014-05-20 22:32:27 +0200315 if (check->health > 0)
316 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100317
Willy Tarreau4eec5472014-05-20 22:32:27 +0200318 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200319 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200320}
321
Willy Tarreauaf549582014-05-16 17:37:50 +0200322/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200323 * it isn't in maintenance, it is not tracking a down server and other checks
324 * comply. The rule is simple : by default, a server is up, unless any of the
325 * following conditions is true :
326 * - health check failed (check->health < rise)
327 * - agent check failed (agent->health < rise)
328 * - the server tracks a down server (track && track->state == STOPPED)
329 * Note that if the server has a slowstart, it will switch to STARTING instead
330 * of RUNNING. Also, only the health checks support the nolb mode, so the
331 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200332 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200333static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200334{
Simon Horman4a741432013-02-23 15:35:38 +0900335 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100336
Emeric Brun52a91d32017-08-31 14:41:55 +0200337 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200338 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100339
Emeric Brun52a91d32017-08-31 14:41:55 +0200340 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200341 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100342
Willy Tarreau3e048382014-05-21 10:30:54 +0200343 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
344 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100345
Willy Tarreau3e048382014-05-21 10:30:54 +0200346 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
347 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200348
Emeric Brun52a91d32017-08-31 14:41:55 +0200349 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200350 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100351
Emeric Brun5a133512017-10-19 14:42:30 +0200352 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100353}
354
Willy Tarreaudb58b792014-05-21 13:57:23 +0200355/* Marks the check <check> as valid and tries to set its server into stopping mode
356 * if it was running or starting, and provided it isn't in maintenance and other
357 * checks comply. The conditions for the server to be marked in stopping mode are
358 * the same as for it to be turned up. Also, only the health checks support the
359 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200360 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200361static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200362{
Simon Horman4a741432013-02-23 15:35:38 +0900363 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100364
Emeric Brun52a91d32017-08-31 14:41:55 +0200365 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200366 return;
367
Willy Tarreaudb58b792014-05-21 13:57:23 +0200368 if (check->state & CHK_ST_AGENT)
369 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100370
Emeric Brun52a91d32017-08-31 14:41:55 +0200371 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200372 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100373
Willy Tarreaudb58b792014-05-21 13:57:23 +0200374 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
375 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100376
Willy Tarreaudb58b792014-05-21 13:57:23 +0200377 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
378 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100379
Willy Tarreaub26881a2017-12-23 11:16:49 +0100380 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100381}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200382
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100383/* note: use health_adjust() only, which first checks that the observe mode is
384 * enabled.
385 */
386void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100387{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100388 int failed;
389 int expire;
390
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100391 if (s->observe >= HANA_OBS_SIZE)
392 return;
393
Willy Tarreaubb956662013-01-24 00:37:39 +0100394 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100395 return;
396
397 switch (analyze_statuses[status].lr[s->observe - 1]) {
398 case 1:
399 failed = 1;
400 break;
401
402 case 2:
403 failed = 0;
404 break;
405
406 default:
407 return;
408 }
409
410 if (!failed) {
411 /* good: clear consecutive_errors */
412 s->consecutive_errors = 0;
413 return;
414 }
415
Christopher Faulet29f77e82017-06-08 14:04:45 +0200416 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100417
418 if (s->consecutive_errors < s->consecutive_errors_limit)
419 return;
420
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100421 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
422 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100423
424 switch (s->onerror) {
425 case HANA_ONERR_FASTINTER:
426 /* force fastinter - nothing to do here as all modes force it */
427 break;
428
429 case HANA_ONERR_SUDDTH:
430 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900431 if (s->check.health > s->check.rise)
432 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100433
434 /* no break - fall through */
435
436 case HANA_ONERR_FAILCHK:
437 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200438 set_server_check_status(&s->check, HCHK_STATUS_HANA,
439 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200440 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100441 break;
442
443 case HANA_ONERR_MARKDWN:
444 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900445 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200446 set_server_check_status(&s->check, HCHK_STATUS_HANA,
447 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200448 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100449 break;
450
451 default:
452 /* write a warning? */
453 break;
454 }
455
456 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200457 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100458
Simon Horman66183002013-02-23 10:16:43 +0900459 if (s->check.fastinter) {
460 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300461 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200462 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300463 /* requeue check task with new expire */
464 task_queue(s->check.task);
465 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100466 }
467}
468
Willy Tarreaua1dab552014-04-14 15:04:54 +0200469static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100470{
471 int sv_state;
472 int ratio;
473 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800474 char addr[46];
475 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100476 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
477 "UP %d/%d", "UP",
478 "NOLB %d/%d", "NOLB",
479 "no check" };
480
481 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
482 hlen += 24;
483
Willy Tarreauff5ae352013-12-11 20:36:34 +0100484 if (!(s->check.state & CHK_ST_ENABLED))
485 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200486 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900487 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100488 sv_state = 3; /* UP */
489 else
490 sv_state = 2; /* going down */
491
Emeric Brun52a91d32017-08-31 14:41:55 +0200492 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100493 sv_state += 2;
494 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900495 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100496 sv_state = 1; /* going up */
497 else
498 sv_state = 0; /* DOWN */
499 }
500
Willy Tarreaua1dab552014-04-14 15:04:54 +0200501 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100502 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200503 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
504 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100505
Joseph Lynch514061c2015-01-15 17:52:59 -0800506 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100507 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
508 snprintf(port, sizeof(port), "%u", s->svc_port);
509 else
510 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800511
512 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
513 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100514 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200515 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100516 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
517 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
518 s->nbpend);
519
Emeric Brun52a91d32017-08-31 14:41:55 +0200520 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100521 now.tv_sec < s->last_change + s->slowstart &&
522 now.tv_sec >= s->last_change) {
523 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200524 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100525 }
526
527 buffer[hlen++] = '\r';
528 buffer[hlen++] = '\n';
529
530 return hlen;
531}
532
Willy Tarreau20a18342013-12-05 00:31:46 +0100533/* Check the connection. If an error has already been reported or the socket is
534 * closed, keep errno intact as it is supposed to contain the valid error code.
535 * If no error is reported, check the socket's error queue using getsockopt().
536 * Warning, this must be done only once when returning from poll, and never
537 * after an I/O error was attempted, otherwise the error queue might contain
538 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
539 * socket. Returns non-zero if an error was reported, zero if everything is
540 * clean (including a properly closed socket).
541 */
542static int retrieve_errno_from_socket(struct connection *conn)
543{
544 int skerr;
545 socklen_t lskerr = sizeof(skerr);
546
547 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
548 return 1;
549
Willy Tarreau3c728722014-01-23 13:50:42 +0100550 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100551 return 0;
552
Willy Tarreau585744b2017-08-24 14:31:19 +0200553 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100554 errno = skerr;
555
556 if (errno == EAGAIN)
557 errno = 0;
558
559 if (!errno) {
560 /* we could not retrieve an error, that does not mean there is
561 * none. Just don't change anything and only report the prior
562 * error if any.
563 */
564 if (conn->flags & CO_FL_ERROR)
565 return 1;
566 else
567 return 0;
568 }
569
570 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
571 return 1;
572}
573
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100574/* Try to collect as much information as possible on the connection status,
575 * and adjust the server status accordingly. It may make use of <errno_bck>
576 * if non-null when the caller is absolutely certain of its validity (eg:
577 * checked just after a syscall). If the caller doesn't have a valid errno,
578 * it can pass zero, and retrieve_errno_from_socket() will be called to try
579 * to extract errno from the socket. If no error is reported, it will consider
580 * the <expired> flag. This is intended to be used when a connection error was
581 * reported in conn->flags or when a timeout was reported in <expired>. The
582 * function takes care of not updating a server status which was already set.
583 * All situations where at least one of <expired> or CO_FL_ERROR are set
584 * produce a status.
585 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200586static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100587{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200588 struct conn_stream *cs = check->cs;
589 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100590 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200591 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200592 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200593 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100594
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100595 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100596 return;
597
598 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200599 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100600 retrieve_errno_from_socket(conn);
601
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200602 if (conn && !(conn->flags & CO_FL_ERROR) &&
603 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100604 return;
605
606 /* we'll try to build a meaningful error message depending on the
607 * context of the error possibly present in conn->err_code, and the
608 * socket error possibly collected above. This is useful to know the
609 * exact step of the L6 layer (eg: SSL handshake).
610 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200611 chk = get_trash_chunk();
612
613 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900614 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200615 if (!step)
616 chunk_printf(chk, " at initial connection step of tcp-check");
617 else {
618 chunk_printf(chk, " at step %d of tcp-check", step);
619 /* we were looking for a string */
620 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
621 if (check->last_started_step->port)
622 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
623 else
624 chunk_appendf(chk, " (connect)");
625 }
626 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
627 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200628 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200629 else if (check->last_started_step->expect_regex)
630 chunk_appendf(chk, " (expect regex)");
631 }
632 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
633 chunk_appendf(chk, " (send)");
634 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200635
636 comment = tcpcheck_get_step_comment(check, step);
637 if (comment)
638 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200639 }
640 }
641
Willy Tarreau00149122017-10-04 18:05:01 +0200642 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100643 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200644 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
645 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100646 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200647 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
648 chk->area);
649 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100650 }
651 else {
652 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200653 chunk_printf(&trash, "%s%s", strerror(errno),
654 chk->area);
655 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100656 }
657 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200658 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100659 }
660 }
661
Willy Tarreau00149122017-10-04 18:05:01 +0200662 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200663 /* NOTE: this is reported after <fall> tries */
664 chunk_printf(chk, "No port available for the TCP connection");
665 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
666 }
667
Willy Tarreau00149122017-10-04 18:05:01 +0200668 if (!conn) {
669 /* connection allocation error before the connection was established */
670 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
671 }
672 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100673 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200674 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100675 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
676 else if (expired)
677 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200678
679 /*
680 * might be due to a server IP change.
681 * Let's trigger a DNS resolution if none are currently running.
682 */
Christopher Faulet67957bd2017-09-27 11:00:59 +0200683 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200684
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100685 }
686 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
687 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200688 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100689 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
690 else if (expired)
691 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
692 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200693 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100694 /* I/O error after connection was established and before we could diagnose */
695 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
696 }
697 else if (expired) {
698 /* connection established but expired check */
699 if (check->type == PR_O2_SSL3_CHK)
700 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
701 else /* HTTP, SMTP, ... */
702 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
703 }
704
705 return;
706}
707
Willy Tarreaubaaee002006-06-26 02:48:02 +0200708/*
709 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200710 * the connection acknowledgement. If the proxy requires L7 health-checks,
711 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900712 * to set check->status, check->duration and check->result.
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200713 */
714static struct task *event_srv_chk_w(struct task *task, void *ctx, unsigned short state)
715{
716 struct conn_stream *cs = ctx;
717 struct check __maybe_unused *check = cs->data;
718
719 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
720 __event_srv_chk_w(cs);
721 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
722 return NULL;
723}
724
725/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100726 *
727 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200728 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
729 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200731static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200732{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200733 struct connection *conn = cs->conn;
734 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900735 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900736 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200737
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100738 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100739 goto out_wakeup;
740
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200741 if (conn->flags & CO_FL_HANDSHAKE) {
742 if (cs->wait_list.task->process != event_srv_chk_w) {
743 cs->wait_list.task->process = event_srv_chk_w;
744 cs->wait_list.task->context = cs;
745 }
746 LIST_ADDQ(&conn->send_wait_list, &cs->wait_list.list);
747 goto out;
748 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100749
Willy Tarreau20a18342013-12-05 00:31:46 +0100750 if (retrieve_errno_from_socket(conn)) {
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 Tarreau20a18342013-12-05 00:31:46 +0100753 goto out_wakeup;
754 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100755
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200756 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100757 /* if the output is closed, we can't do anything */
758 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200759 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100760 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200761 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200762
Willy Tarreau06559ac2013-12-05 01:53:08 +0100763 /* here, we know that the connection is established. That's enough for
764 * a pure TCP check.
765 */
766 if (!check->type)
767 goto out_wakeup;
768
Willy Tarreauc09572f2017-10-04 11:58:22 +0200769 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100770 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200771 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200772
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200773 if (b_data(&check->bo)) {
774 b_del(&check->bo, conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0));
775 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200776 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200777 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200778 __cs_stop_both(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100779 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200780 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200781 if (b_data(&check->bo)) {
Willy Tarreau83188852018-07-18 08:18:20 +0200782 conn->mux->subscribe(cs, SUB_CAN_SEND, wl_set_waitcb(&cs->wait_list, event_srv_chk_w, cs));
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200783 goto out;
784 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100785 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200786
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100787 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
788 if (s->proxy->timeout.check) {
789 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
790 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200791 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100792 goto out_nowake;
793
Willy Tarreau83749182007-04-15 20:56:27 +0200794 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200795 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200796 out_nowake:
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200797 __cs_stop_send(cs); /* nothing more to write */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200798 out:
799 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200800}
801
Willy Tarreaubaaee002006-06-26 02:48:02 +0200802/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200803 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200804 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900805 * set_server_check_status() to update check->status, check->duration
806 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200807
808 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
809 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
810 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
811 * response to an SSL HELLO (the principle is that this is enough to
812 * distinguish between an SSL server and a pure TCP relay). All other cases will
813 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
814 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100815 *
816 * Please do NOT place any return statement in this function and only leave
817 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200818 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200819static void event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200820{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200821 struct connection *conn = cs->conn;
822 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900823 struct server *s = check->server;
824 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200825 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100826 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200827 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200828
Christopher Faulet2a944ee2017-11-07 10:42:54 +0100829 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brun9f0b4582017-10-23 14:39:51 +0200830
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100831 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200832 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200833
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100834 if (conn->flags & CO_FL_HANDSHAKE)
835 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200836
Willy Tarreauc09572f2017-10-04 11:58:22 +0200837 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100838 if (check->type == PR_O2_TCPCHK_CHK)
839 goto out_unlock;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200840
Willy Tarreau83749182007-04-15 20:56:27 +0200841 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
842 * but the connection was closed on the remote end. Fortunately, recv still
843 * works correctly and we don't need to do the getsockopt() on linux.
844 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000845
846 /* Set buffer to point to the end of the data already read, and check
847 * that there is free space remaining. If the buffer is full, proceed
848 * with running the checks without attempting another socket read.
849 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000850
Willy Tarreau03938182010-03-17 21:52:07 +0100851 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000852
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200853 conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200854 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100855 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200856 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200857 /* Report network errors only if we got no other data. Otherwise
858 * we'll let the upper layers decide whether the response is OK
859 * or not. It is very common that an RST sent by the server is
860 * reported as an error just after the last data chunk.
861 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200862 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100863 goto out_wakeup;
864 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200865 }
866
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100867
Willy Tarreau03938182010-03-17 21:52:07 +0100868 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200869 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100870 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200871 if (b_data(&check->bi) < b_size(&check->bi))
872 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100873 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200874 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100875 done = 1; /* buffer full, don't wait for more data */
876 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200877
Nick Chalk57b1bf72010-03-16 15:50:46 +0000878 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900879 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200880 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200881 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100882 goto wait_more_data;
883
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100884 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200885 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
886 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
887 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
888 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
889 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
890 cut_crlf(b_head(&check->bi));
891 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200892
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100893 goto out_wakeup;
894 }
895
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200896 check->code = str2uic(b_head(&check->bi) + 9);
897 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200898
Willy Tarreaubd741542010-03-16 18:46:54 +0100899 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200900 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000901 /* 404 may be accepted as "stopping" only if the server was up */
902 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900903 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000904 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100905 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
906 /* Run content verification check... We know we have at least 13 chars */
907 if (!httpchk_expect(s, done))
908 goto wait_more_data;
909 }
910 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200911 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100912 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900913 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100914 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000915 else {
916 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900917 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000918 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200919 break;
920
921 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200922 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100923 goto wait_more_data;
924
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100925 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200926 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200928 else
Simon Horman4a741432013-02-23 15:35:38 +0900929 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200930 break;
931
932 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200933 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100934 goto wait_more_data;
935
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200936 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200937 if ((b_data(&check->bi) < strlen("000\r")) ||
938 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
939 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
940 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
941 cut_crlf(b_head(&check->bi));
942 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200943 goto out_wakeup;
944 }
945
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200946 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200947
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200948 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200949 cut_crlf(desc);
950
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100951 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200952 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900953 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200954 else
Simon Horman4a741432013-02-23 15:35:38 +0900955 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200956 break;
957
Simon Hormana2b9dad2013-02-12 10:45:54 +0900958 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100959 int status = HCHK_STATUS_CHECKED;
960 const char *hs = NULL; /* health status */
961 const char *as = NULL; /* admin status */
962 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200963 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100964 const char *err = NULL; /* first error to report */
965 const char *wrn = NULL; /* first warning to report */
966 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900967
Willy Tarreau81f5d942013-12-09 20:51:51 +0100968 /* We're getting an agent check response. The agent could
969 * have been disabled in the mean time with a long check
970 * still pending. It is important that we ignore the whole
971 * response.
972 */
973 if (!(check->server->agent.state & CHK_ST_ENABLED))
974 break;
975
976 /* The agent supports strings made of a single line ended by the
977 * first CR ('\r') or LF ('\n'). This line is composed of words
978 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
979 * line may optionally contained a description of a state change
980 * after a sharp ('#'), which is only considered if a health state
981 * is announced.
982 *
983 * Words may be composed of :
984 * - a numeric weight suffixed by the percent character ('%').
985 * - a health status among "up", "down", "stopped", and "fail".
986 * - an admin status among "ready", "drain", "maint".
987 *
988 * These words may appear in any order. If multiple words of the
989 * same category appear, the last one wins.
990 */
991
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200992 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +0100993 while (*p && *p != '\n' && *p != '\r')
994 p++;
995
996 if (!*p) {
997 if (!done)
998 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900999
Willy Tarreau9809b782013-12-11 21:40:11 +01001000 /* at least inform the admin that the agent is mis-behaving */
1001 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1002 break;
1003 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001004
Willy Tarreau9809b782013-12-11 21:40:11 +01001005 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001006 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001007
Willy Tarreau81f5d942013-12-09 20:51:51 +01001008 while (*cmd) {
1009 /* look for next word */
1010 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1011 cmd++;
1012 continue;
1013 }
Simon Horman671b6f02013-11-25 10:46:39 +09001014
Willy Tarreau81f5d942013-12-09 20:51:51 +01001015 if (*cmd == '#') {
1016 /* this is the beginning of a health status description,
1017 * skip the sharp and blanks.
1018 */
1019 cmd++;
1020 while (*cmd == '\t' || *cmd == ' ')
1021 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001022 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001023 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001024
1025 /* find the end of the word so that we have a null-terminated
1026 * word between <cmd> and <p>.
1027 */
1028 p = cmd + 1;
1029 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1030 p++;
1031 if (*p)
1032 *p++ = 0;
1033
1034 /* first, health statuses */
1035 if (strcasecmp(cmd, "up") == 0) {
1036 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001037 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001038 hs = cmd;
1039 }
1040 else if (strcasecmp(cmd, "down") == 0) {
1041 check->health = 0;
1042 status = HCHK_STATUS_L7STS;
1043 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001044 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001045 else if (strcasecmp(cmd, "stopped") == 0) {
1046 check->health = 0;
1047 status = HCHK_STATUS_L7STS;
1048 hs = cmd;
1049 }
1050 else if (strcasecmp(cmd, "fail") == 0) {
1051 check->health = 0;
1052 status = HCHK_STATUS_L7STS;
1053 hs = cmd;
1054 }
1055 /* admin statuses */
1056 else if (strcasecmp(cmd, "ready") == 0) {
1057 as = cmd;
1058 }
1059 else if (strcasecmp(cmd, "drain") == 0) {
1060 as = cmd;
1061 }
1062 else if (strcasecmp(cmd, "maint") == 0) {
1063 as = cmd;
1064 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001065 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001066 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1067 ps = cmd;
1068 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001069 /* try to parse a maxconn here */
1070 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1071 cs = cmd;
1072 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001073 else {
1074 /* keep a copy of the first error */
1075 if (!err)
1076 err = cmd;
1077 }
1078 /* skip to next word */
1079 cmd = p;
1080 }
1081 /* here, cmd points either to \0 or to the beginning of a
1082 * description. Skip possible leading spaces.
1083 */
1084 while (*cmd == ' ' || *cmd == '\n')
1085 cmd++;
1086
1087 /* First, update the admin status so that we avoid sending other
1088 * possibly useless warnings and can also update the health if
1089 * present after going back up.
1090 */
1091 if (as) {
1092 if (strcasecmp(as, "drain") == 0)
1093 srv_adm_set_drain(check->server);
1094 else if (strcasecmp(as, "maint") == 0)
1095 srv_adm_set_maint(check->server);
1096 else
1097 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001098 }
1099
Willy Tarreau81f5d942013-12-09 20:51:51 +01001100 /* now change weights */
1101 if (ps) {
1102 const char *msg;
1103
1104 msg = server_parse_weight_change_request(s, ps);
1105 if (!wrn || !*wrn)
1106 wrn = msg;
1107 }
1108
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001109 if (cs) {
1110 const char *msg;
1111
1112 cs += strlen("maxconn:");
1113
1114 msg = server_parse_maxconn_change_request(s, cs);
1115 if (!wrn || !*wrn)
1116 wrn = msg;
1117 }
1118
Willy Tarreau81f5d942013-12-09 20:51:51 +01001119 /* and finally health status */
1120 if (hs) {
1121 /* We'll report some of the warnings and errors we have
1122 * here. Down reports are critical, we leave them untouched.
1123 * Lack of report, or report of 'UP' leaves the room for
1124 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001125 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001126 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001127 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001128
1129 if (!*msg || status == HCHK_STATUS_L7OKD) {
1130 if (err && *err)
1131 msg = err;
1132 else if (wrn && *wrn)
1133 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001134 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001135
1136 t = get_trash_chunk();
1137 chunk_printf(t, "via agent : %s%s%s%s",
1138 hs, *msg ? " (" : "",
1139 msg, *msg ? ")" : "");
1140
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001141 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001142 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001143 else if (err && *err) {
1144 /* No status change but we'd like to report something odd.
1145 * Just report the current state and copy the message.
1146 */
1147 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001148 set_server_check_status(check, status/*check->status*/,
1149 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001150
Willy Tarreau81f5d942013-12-09 20:51:51 +01001151 }
1152 else if (wrn && *wrn) {
1153 /* No status change but we'd like to report something odd.
1154 * Just report the current state and copy the message.
1155 */
1156 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001157 set_server_check_status(check, status/*check->status*/,
1158 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001159 }
1160 else
1161 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001162 break;
1163 }
1164
Willy Tarreau1620ec32011-08-06 17:05:02 +02001165 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001166 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001167 goto wait_more_data;
1168
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001169 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001170 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001171 }
1172 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001173 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1174 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001175 else
1176 desc = "PostgreSQL unknown error";
1177
Simon Horman4a741432013-02-23 15:35:38 +09001178 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001179 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001180 break;
1181
1182 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001183 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001184 goto wait_more_data;
1185
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001186 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001187 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001188 }
1189 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001190 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001191 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001192 break;
1193
1194 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001195 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001196 goto wait_more_data;
1197
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001198 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001199 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001200 /* We set the MySQL Version in description for information purpose
1201 * FIXME : it can be cool to use MySQL Version for other purpose,
1202 * like mark as down old MySQL server.
1203 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001204 if (b_data(&check->bi) > 51) {
1205 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001206 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001207 }
1208 else {
1209 if (!done)
1210 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001211
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001212 /* it seems we have a OK packet but without a valid length,
1213 * it must be a protocol error
1214 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001215 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001216 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001217 }
1218 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001219 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001220 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001221 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001222 }
1223 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001224 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1225 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1226 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001227
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001228 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001230 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001231 /* We have only one MySQL packet and it is a Handshake Initialization packet
1232 * but we need to have a second packet to know if it is alright
1233 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001234 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001235 goto wait_more_data;
1236 }
1237 else {
1238 /* We have only one packet and it is an Error packet,
1239 * an error message is attached, so we can display it
1240 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001241 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001242 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001243 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001244 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001245 } else if (b_data(&check->bi) > first_packet_len + 4) {
1246 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1247 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1248 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001250 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001251 /* We have 2 packets and that's good */
1252 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001253 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001254 /* No error packet */
1255 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001256 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001257 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001258 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001259 }
1260 else {
1261 /* An error message is attached in the Error packet
1262 * so we can display it ! :)
1263 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001264 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001265 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001266 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001267 }
1268 }
1269 }
1270 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001271 if (!done)
1272 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001273
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001274 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001275 * it must be a protocol error
1276 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001277 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001278 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001279 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001280 }
1281 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001282 break;
1283
1284 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001285 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001286 goto wait_more_data;
1287
1288 /* Check if the server speaks LDAP (ASN.1/BER)
1289 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1290 * http://tools.ietf.org/html/rfc4511
1291 */
1292
1293 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1294 * LDAPMessage: 0x30: SEQUENCE
1295 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001296 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001297 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001298 }
1299 else {
1300 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001301 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001302
1303 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1304 * messageID: 0x02 0x01 0x01: INTEGER 1
1305 * protocolOp: 0x61: bindResponse
1306 */
1307 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001308 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001309 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001310 goto out_wakeup;
1311 }
1312
1313 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001314 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001315
1316 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1317 * ldapResult: 0x0a 0x01: ENUMERATION
1318 */
1319 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001320 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001321 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001322 goto out_wakeup;
1323 }
1324
1325 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1326 * resultCode
1327 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001328 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001329 if (check->code) {
1330 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 +02001331 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001332 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001333 }
1334 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001335 break;
1336
Christopher Fauletba7bc162016-11-07 21:07:38 +01001337 case PR_O2_SPOP_CHK: {
1338 unsigned int framesz;
1339 char err[HCHK_DESC_LEN];
1340
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001341 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001342 goto wait_more_data;
1343
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001344 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001345 framesz = ntohl(framesz);
1346
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001347 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001348 goto wait_more_data;
1349
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001350 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001351 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1352 else
1353 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1354 break;
1355 }
1356
Willy Tarreau1620ec32011-08-06 17:05:02 +02001357 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001358 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001359 break;
1360 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001361
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001362 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001363 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001364 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001365 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001366
Nick Chalk57b1bf72010-03-16 15:50:46 +00001367 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001368 *b_head(&check->bi) = '\0';
1369 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001370
Steven Davidovitz544d4812017-03-08 11:06:20 -08001371 /* Close the connection... We still attempt to nicely close if,
1372 * for instance, SSL needs to send a "close notify." Later, we perform
1373 * a hard close and reset the connection if some data are pending,
1374 * otherwise we end up with many TIME_WAITs and eat all the source port
1375 * range quickly. To avoid sending RSTs all the time, we first try to
1376 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001377 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001378 __cs_stop_both(cs);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001379 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001380
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001381 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001382 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001383 conn->flags |= CO_FL_ERROR;
1384
Willy Tarreaufdccded2008-08-29 18:19:04 +02001385 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001386 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001387 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau3267d362012-08-17 23:53:56 +02001388 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001389
1390 wait_more_data:
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001391 __cs_want_recv(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001392 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001393}
1394
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001395/*
1396 * This function is used only for server health-checks. It handles connection
1397 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001398 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1399 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001400 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001401static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001402{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001403 struct connection *conn = cs->conn;
1404 struct check *check = cs->data;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001405 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001406
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001407 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001408
Willy Tarreauc09572f2017-10-04 11:58:22 +02001409 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001410 if (check->type == PR_O2_TCPCHK_CHK) {
1411 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001412 cs = check->cs;
1413 conn = cs_conn(cs);
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001414 } else if (LIST_ISEMPTY(&cs->wait_list.list))
1415 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001416
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001417 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001418 /* We may get error reports bypassing the I/O handlers, typically
1419 * the case when sending a pure TCP check which fails, then the I/O
1420 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001421 * main processing task so let's simply wake it up. If we get here,
1422 * we expect errno to still be valid.
1423 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001424 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001425 __cs_stop_both(cs);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001426 task_wakeup(check->task, TASK_WOKEN_IO);
1427 }
Willy Tarreau2500fc22018-04-03 19:31:38 +02001428 else if (!(conn->flags & CO_FL_HANDSHAKE) && !(cs->flags & (CS_FL_DATA_RD_ENA|CS_FL_DATA_WR_ENA))) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001429 /* we may get here if only a connection probe was required : we
1430 * don't have any data to send nor anything expected in response,
1431 * so the completion of the connection establishment is enough.
1432 */
1433 task_wakeup(check->task, TASK_WOKEN_IO);
1434 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001435
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001436 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001437 /* We're here because nobody wants to handle the error, so we
1438 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001439 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001440 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001441 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001442 ret = -1;
Willy Tarreau2d351b62013-12-05 02:36:25 +01001443 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001444
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001445 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001446
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001447 /* if a connection got replaced, we must absolutely prevent the connection
1448 * handler from touching its fd, and perform the FD polling updates ourselves
1449 */
1450 if (ret < 0)
1451 conn_cond_update_polling(conn);
1452
1453 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001454}
1455
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001456struct data_cb check_conn_cb = {
1457 .recv = event_srv_chk_r,
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001458 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001459 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001460};
1461
Willy Tarreaubaaee002006-06-26 02:48:02 +02001462/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001463 * updates the server's weight during a warmup stage. Once the final weight is
1464 * reached, the task automatically stops. Note that any server status change
1465 * must have updated s->last_change accordingly.
1466 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001467static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001468{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001469 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001470
1471 /* by default, plan on stopping the task */
1472 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001473 if ((s->next_admin & SRV_ADMF_MAINT) ||
1474 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001475 return t;
1476
Willy Tarreau892337c2014-05-13 23:41:20 +02001477 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001478 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001479
1480 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001481 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001482
1483 /* get back there in 1 second or 1/20th of the slowstart interval,
1484 * whichever is greater, resulting in small 5% steps.
1485 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001486 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001487 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1488 return t;
1489}
1490
Willy Tarreau894c6422017-10-04 15:58:52 +02001491/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1492 * none was found.
1493 */
1494static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1495{
1496 struct tcpcheck_rule *r;
1497
1498 list_for_each_entry(r, list, list) {
1499 if (r->action != TCPCHK_ACT_COMMENT)
1500 return r;
1501 }
1502 return NULL;
1503}
1504
Willy Tarreau2e993902011-10-31 11:53:20 +01001505/*
Simon Horman98637e52014-06-20 12:30:16 +09001506 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001507 *
1508 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001509 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1510 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1511 * - SF_ERR_SRVTO if there are no more servers
1512 * - SF_ERR_SRVCL if the connection was refused by the server
1513 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1514 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1515 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001516 * - 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 +01001517 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001518 * Note that we try to prevent the network stack from sending the ACK during the
1519 * connect() when a pure TCP check is used (without PROXY protocol).
1520 */
Simon Horman98637e52014-06-20 12:30:16 +09001521static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001522{
1523 struct check *check = t->context;
1524 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001525 struct conn_stream *cs = check->cs;
1526 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001527 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001528 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001529 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001530 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001531
Willy Tarreau00149122017-10-04 18:05:01 +02001532 /* we cannot have a connection here */
1533 if (conn)
1534 return SF_ERR_INTERNAL;
1535
Simon Hormanb00d17a2014-06-13 16:18:16 +09001536 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001537 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001538 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001539 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1540 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001541
1542 /* prepare the check buffer.
1543 * This should not be used if check is the secondary agent check
1544 * of a server as s->proxy->check_req will relate to the
1545 * configuration of the primary check. Similarly, tcp-check uses
1546 * its own strings.
1547 */
1548 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001549 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001550
1551 /* we want to check if this host replies to HTTP or SSLv3 requests
1552 * so we'll send the request, and won't wake the checker up now.
1553 */
1554 if ((check->type) == PR_O2_SSL3_CHK) {
1555 /* SSL requires that we put Unix time in the request */
1556 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001557 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001558 }
1559 else if ((check->type) == PR_O2_HTTP_CHK) {
1560 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001561 b_putblk(&check->bo, trash.area,
1562 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001563 /* prevent HTTP keep-alive when "http-check expect" is used */
1564 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001565 b_putist(&check->bo, ist("Connection: close\r\n"));
1566 b_putist(&check->bo, ist("\r\n"));
1567 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001568 }
1569 }
1570
James Brown55f9ff12015-10-21 18:19:05 -07001571 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001572 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001573 }
1574
Willy Tarreauf411cce2017-10-04 16:21:19 +02001575 /* for tcp-checks, the initial connection setup is handled separately as
1576 * it may be sent to a specific port and not to the server's.
1577 */
1578 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1579 tcpcheck_main(check);
1580 return SF_ERR_UP;
1581 }
1582
Simon Hormanb00d17a2014-06-13 16:18:16 +09001583 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001584 cs = check->cs = cs_new(NULL);
1585 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001586 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001587 conn = cs->conn;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001588
Simon Horman41f58762015-01-30 11:22:56 +09001589 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001590 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001591 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001592 }
1593 else {
1594 /* we'll connect to the addr on the server */
1595 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001596 }
1597
Olivier Houchard6377a002017-12-01 22:04:05 +01001598 proto = protocol_by_family(conn->addr.to.ss_family);
1599
1600 conn_prepare(conn, proto, check->xprt);
1601 conn_install_mux(conn, &mux_pt_ops, cs);
1602 cs_attach(cs, check, &check_conn_cb);
1603 conn->target = &s->obj_type;
1604
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001605 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
1606 int i = 0;
1607
1608 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001609 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001610 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001611
1612 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001613 }
1614
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001615 /* no client address */
1616 clear_addr(&conn->addr.from);
1617
Willy Tarreauf3d34822014-12-08 12:11:28 +01001618 /* only plain tcp-check supports quick ACK */
1619 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1620
Willy Tarreauf411cce2017-10-04 16:21:19 +02001621 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1622 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001623
Willy Tarreaue7dff022015-04-03 01:14:29 +02001624 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001625 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001626 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001627
1628 if (check->type)
1629 cs_want_send(cs);
1630
Olivier Houchard9130a962017-10-17 17:33:43 +02001631#ifdef USE_OPENSSL
1632 if (s->check.sni)
1633 ssl_sock_set_servername(conn, s->check.sni);
1634#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001635 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001636 conn->send_proxy_ofs = 1;
1637 conn->flags |= CO_FL_SEND_PROXY;
1638 }
1639
1640 return ret;
1641}
1642
Simon Horman98637e52014-06-20 12:30:16 +09001643static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001644static struct pool_head *pool_head_pid_list;
Christopher Faulet9dcf9b62017-11-13 10:34:01 +01001645__decl_hathreads(HA_SPINLOCK_T pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001646
1647void block_sigchld(void)
1648{
1649 sigset_t set;
1650 sigemptyset(&set);
1651 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001652 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001653}
1654
1655void unblock_sigchld(void)
1656{
1657 sigset_t set;
1658 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001659 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001660 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001661}
1662
Simon Horman98637e52014-06-20 12:30:16 +09001663static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1664{
1665 struct pid_list *elem;
1666 struct check *check = t->context;
1667
Willy Tarreaubafbe012017-11-24 17:34:44 +01001668 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001669 if (!elem)
1670 return NULL;
1671 elem->pid = pid;
1672 elem->t = t;
1673 elem->exited = 0;
1674 check->curpid = elem;
1675 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001676
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001677 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001678 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001679 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001680
Simon Horman98637e52014-06-20 12:30:16 +09001681 return elem;
1682}
1683
Simon Horman98637e52014-06-20 12:30:16 +09001684static void pid_list_del(struct pid_list *elem)
1685{
1686 struct check *check;
1687
1688 if (!elem)
1689 return;
1690
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001691 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001692 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001693 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001694
Simon Horman98637e52014-06-20 12:30:16 +09001695 if (!elem->exited)
1696 kill(elem->pid, SIGTERM);
1697
1698 check = elem->t->context;
1699 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001700 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001701}
1702
1703/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1704static void pid_list_expire(pid_t pid, int status)
1705{
1706 struct pid_list *elem;
1707
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001708 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001709 list_for_each_entry(elem, &pid_list, list) {
1710 if (elem->pid == pid) {
1711 elem->t->expire = now_ms;
1712 elem->status = status;
1713 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001714 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001715 break;
Simon Horman98637e52014-06-20 12:30:16 +09001716 }
1717 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001718 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001719}
1720
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001721static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001722{
1723 pid_t pid;
1724 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001725
Simon Horman98637e52014-06-20 12:30:16 +09001726 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1727 pid_list_expire(pid, status);
1728}
1729
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001730static int init_pid_list(void)
1731{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001732 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001733 /* Nothing to do */
1734 return 0;
1735
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001736 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001737 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1738 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001739 return 1;
1740 }
1741
Willy Tarreaubafbe012017-11-24 17:34:44 +01001742 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1743 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001744 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1745 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001746 return 1;
1747 }
1748
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001749 HA_SPIN_INIT(&pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001750
Simon Horman98637e52014-06-20 12:30:16 +09001751 return 0;
1752}
1753
Cyril Bontéac92a062014-12-27 22:28:38 +01001754/* helper macro to set an environment variable and jump to a specific label on failure. */
1755#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001756
1757/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001758 * helper function to allocate enough memory to store an environment variable.
1759 * It will also check that the environment variable is updatable, and silently
1760 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001761 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001762static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001763{
1764 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001765 char *envname;
1766 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001767
Cyril Bontéac92a062014-12-27 22:28:38 +01001768 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001769 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001770 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001771 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001772
1773 envname = extcheck_envs[idx].name;
1774 vmaxlen = extcheck_envs[idx].vmaxlen;
1775
1776 /* Check if the environment variable is already set, and silently reject
1777 * the update if this one is not updatable. */
1778 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1779 return 0;
1780
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001781 /* Instead of sending NOT_USED, sending an empty value is preferable */
1782 if (strcmp(value, "NOT_USED") == 0) {
1783 value = "";
1784 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001785
1786 len = strlen(envname) + 1;
1787 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1788 len += strlen(value);
1789 else
1790 len += vmaxlen;
1791
1792 if (!check->envp[idx])
1793 check->envp[idx] = malloc(len + 1);
1794
1795 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001796 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001797 return 1;
1798 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001799 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001800 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001801 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001802 return 1;
1803 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001804 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001805 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001806 return 1;
1807 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001808 return 0;
1809}
Simon Horman98637e52014-06-20 12:30:16 +09001810
1811static int prepare_external_check(struct check *check)
1812{
1813 struct server *s = check->server;
1814 struct proxy *px = s->proxy;
1815 struct listener *listener = NULL, *l;
1816 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001817 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001818 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001819
1820 list_for_each_entry(l, &px->conf.listeners, by_fe)
1821 /* Use the first INET, INET6 or UNIX listener */
1822 if (l->addr.ss_family == AF_INET ||
1823 l->addr.ss_family == AF_INET6 ||
1824 l->addr.ss_family == AF_UNIX) {
1825 listener = l;
1826 break;
1827 }
1828
Simon Horman98637e52014-06-20 12:30:16 +09001829 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001830 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1831 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001832 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001833 goto err;
1834 }
Simon Horman98637e52014-06-20 12:30:16 +09001835
Cyril Bontéac92a062014-12-27 22:28:38 +01001836 check->argv = calloc(6, sizeof(char *));
1837 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001838 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001839 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001840 }
Simon Horman98637e52014-06-20 12:30:16 +09001841
1842 check->argv[0] = px->check_command;
1843
Cyril Bonté777be862014-12-02 21:21:35 +01001844 if (!listener) {
1845 check->argv[1] = strdup("NOT_USED");
1846 check->argv[2] = strdup("NOT_USED");
1847 }
1848 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001849 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001850 addr_to_str(&listener->addr, buf, sizeof(buf));
1851 check->argv[1] = strdup(buf);
1852 port_to_str(&listener->addr, buf, sizeof(buf));
1853 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001854 }
1855 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001856 const struct sockaddr_un *un;
1857
1858 un = (struct sockaddr_un *)&listener->addr;
1859 check->argv[1] = strdup(un->sun_path);
1860 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001861 }
1862 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001863 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001864 goto err;
1865 }
1866
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001867 addr_to_str(&s->addr, buf, sizeof(buf));
1868 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001869
1870 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1871 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1872 else
1873 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001874 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001875
Cyril Bontéac92a062014-12-27 22:28:38 +01001876 for (i = 0; i < 5; i++) {
1877 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001878 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001879 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001880 }
1881 }
Simon Horman98637e52014-06-20 12:30:16 +09001882
Cyril Bontéac92a062014-12-27 22:28:38 +01001883 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001884 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001885 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1886 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1887 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1888 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001889 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001890 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1891 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1892 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1893 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1894 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1895 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1896
1897 /* Ensure that we don't leave any hole in check->envp */
1898 for (i = 0; i < EXTCHK_SIZE; i++)
1899 if (!check->envp[i])
1900 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001901
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001902 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001903err:
1904 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001905 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001906 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001907 free(check->envp);
1908 check->envp = NULL;
1909 }
1910
1911 if (check->argv) {
1912 for (i = 1; i < 5; i++)
1913 free(check->argv[i]);
1914 free(check->argv);
1915 check->argv = NULL;
1916 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001917 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001918}
1919
Simon Hormanb00d17a2014-06-13 16:18:16 +09001920/*
Simon Horman98637e52014-06-20 12:30:16 +09001921 * establish a server health-check that makes use of a process.
1922 *
1923 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001924 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001925 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001926 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001927 *
1928 * Blocks and then unblocks SIGCHLD
1929 */
1930static int connect_proc_chk(struct task *t)
1931{
Cyril Bontéac92a062014-12-27 22:28:38 +01001932 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001933 struct check *check = t->context;
1934 struct server *s = check->server;
1935 struct proxy *px = s->proxy;
1936 int status;
1937 pid_t pid;
1938
Willy Tarreaue7dff022015-04-03 01:14:29 +02001939 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001940
1941 block_sigchld();
1942
1943 pid = fork();
1944 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001945 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1946 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001947 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1948 goto out;
1949 }
1950 if (pid == 0) {
1951 /* Child */
1952 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001953 int fd;
1954
1955 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1956 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1957
1958 while (fd < global.rlimit_nofile)
1959 close(fd++);
1960
Simon Horman98637e52014-06-20 12:30:16 +09001961 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001962 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001963 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001964 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1965 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001966 exit(-1);
1967 }
1968
1969 /* Parent */
1970 if (check->result == CHK_RES_UNKNOWN) {
1971 if (pid_list_add(pid, t) != NULL) {
1972 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1973
1974 if (px->timeout.check && px->timeout.connect) {
1975 int t_con = tick_add(now_ms, px->timeout.connect);
1976 t->expire = tick_first(t->expire, t_con);
1977 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001978 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001979 goto out;
1980 }
1981 else {
1982 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1983 }
1984 kill(pid, SIGTERM); /* process creation error */
1985 }
1986 else
1987 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1988
1989out:
1990 unblock_sigchld();
1991 return status;
1992}
1993
1994/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02001995 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001996 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001997 *
1998 * Please do NOT place any return statement in this function and only leave
1999 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002000 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002001static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002002{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002003 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002004 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002005 int rv;
2006 int ret;
2007 int expired = tick_is_expired(t->expire, now_ms);
2008
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002009 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002010 if (!(check->state & CHK_ST_INPROGRESS)) {
2011 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002012 if (!expired) /* woke up too early */
2013 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002014
2015 /* we don't send any health-checks when the proxy is
2016 * stopped, the server should not be checked or the check
2017 * is disabled.
2018 */
2019 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2020 s->proxy->state == PR_STSTOPPED)
2021 goto reschedule;
2022
2023 /* we'll initiate a new check */
2024 set_server_check_status(check, HCHK_STATUS_START, NULL);
2025
2026 check->state |= CHK_ST_INPROGRESS;
2027
Simon Hormandbf70192015-01-30 11:22:53 +09002028 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002029 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002030 /* the process was forked, we allow up to min(inter,
2031 * timeout.connect) for it to report its status, but
2032 * only when timeout.check is set as it may be to short
2033 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002034 */
2035 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2036
2037 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2038 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2039 t->expire = tick_first(t->expire, t_con);
2040 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002041 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002042 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002043 }
2044
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002045 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002046
2047 check->state &= ~CHK_ST_INPROGRESS;
2048 check_notify_failure(check);
2049
2050 /* we allow up to min(inter, timeout.connect) for a connection
2051 * to establish but only when timeout.check is set
2052 * as it may be to short for a full check otherwise
2053 */
2054 while (tick_is_expired(t->expire, now_ms)) {
2055 int t_con;
2056
2057 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2058 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2059
2060 if (s->proxy->timeout.check)
2061 t->expire = tick_first(t->expire, t_con);
2062 }
2063 }
2064 else {
2065 /* there was a test running.
2066 * First, let's check whether there was an uncaught error,
2067 * which can happen on connect timeout or error.
2068 */
2069 if (check->result == CHK_RES_UNKNOWN) {
2070 /* good connection is enough for pure TCP check */
2071 struct pid_list *elem = check->curpid;
2072 int status = HCHK_STATUS_UNKNOWN;
2073
2074 if (elem->exited) {
2075 status = elem->status; /* Save in case the process exits between use below */
2076 if (!WIFEXITED(status))
2077 check->code = -1;
2078 else
2079 check->code = WEXITSTATUS(status);
2080 if (!WIFEXITED(status) || WEXITSTATUS(status))
2081 status = HCHK_STATUS_PROCERR;
2082 else
2083 status = HCHK_STATUS_PROCOK;
2084 } else if (expired) {
2085 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002086 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002087 kill(elem->pid, SIGTERM);
2088 }
2089 set_server_check_status(check, status, NULL);
2090 }
2091
2092 if (check->result == CHK_RES_FAILED) {
2093 /* a failure or timeout detected */
2094 check_notify_failure(check);
2095 }
2096 else if (check->result == CHK_RES_CONDPASS) {
2097 /* check is OK but asks for stopping mode */
2098 check_notify_stopping(check);
2099 }
2100 else if (check->result == CHK_RES_PASSED) {
2101 /* a success was detected */
2102 check_notify_success(check);
2103 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002104 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002105 check->state &= ~CHK_ST_INPROGRESS;
2106
2107 pid_list_del(check->curpid);
2108
2109 rv = 0;
2110 if (global.spread_checks > 0) {
2111 rv = srv_getinter(check) * global.spread_checks / 100;
2112 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2113 }
2114 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2115 }
2116
2117 reschedule:
2118 while (tick_is_expired(t->expire, now_ms))
2119 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002120
2121 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002122 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002123 return t;
2124}
2125
2126/*
2127 * manages a server health-check that uses a connection. Returns
2128 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002129 *
2130 * Please do NOT place any return statement in this function and only leave
2131 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002132 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002133static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002134{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002135 struct check *check = context;
Simon Horman4a741432013-02-23 15:35:38 +09002136 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002137 struct conn_stream *cs = check->cs;
2138 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002139 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002140 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002141 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002142
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002143 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002144 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002145 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002146 if (!expired) /* woke up too early */
2147 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002148
Simon Horman671b6f02013-11-25 10:46:39 +09002149 /* we don't send any health-checks when the proxy is
2150 * stopped, the server should not be checked or the check
2151 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002152 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002153 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002154 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002155 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002156
2157 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002158 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002159
Willy Tarreau2c115e52013-12-11 19:41:16 +01002160 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002161 b_reset(&check->bi);
2162 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002163
Simon Hormandbf70192015-01-30 11:22:53 +09002164 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002165 cs = check->cs;
2166 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002167
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002168 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002169 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002170 goto out_unlock;
2171
Willy Tarreaue7dff022015-04-03 01:14:29 +02002172 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002173 /* we allow up to min(inter, timeout.connect) for a connection
2174 * to establish but only when timeout.check is set
2175 * as it may be to short for a full check otherwise
2176 */
Simon Horman4a741432013-02-23 15:35:38 +09002177 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002178
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002179 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2180 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2181 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002182 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002183
2184 if (check->type)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002185 cs_want_recv(cs); /* prepare for reading a possible reply */
Willy Tarreau06559ac2013-12-05 01:53:08 +01002186
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002187 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002188 goto reschedule;
2189
Willy Tarreaue7dff022015-04-03 01:14:29 +02002190 case SF_ERR_SRVTO: /* ETIMEDOUT */
2191 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002192 if (conn)
2193 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002194 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002195 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002196 /* should share same code than cases below */
2197 case SF_ERR_CHK_PORT:
2198 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002199 case SF_ERR_PRXCOND:
2200 case SF_ERR_RESOURCE:
2201 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002202 if (conn)
2203 conn->flags |= CO_FL_ERROR;
2204 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002205 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002206 }
2207
Willy Tarreau5a78f362012-11-23 12:47:05 +01002208 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002209 if (cs) {
2210 cs_destroy(cs);
2211 cs = check->cs = NULL;
2212 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002213 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002214
Willy Tarreau2c115e52013-12-11 19:41:16 +01002215 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002216 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002217
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002218 /* we allow up to min(inter, timeout.connect) for a connection
2219 * to establish but only when timeout.check is set
2220 * as it may be to short for a full check otherwise
2221 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002222 while (tick_is_expired(t->expire, now_ms)) {
2223 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002224
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002225 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002226 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002227
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002228 if (s->proxy->timeout.check)
2229 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002230 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002231 }
2232 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002233 /* there was a test running.
2234 * First, let's check whether there was an uncaught error,
2235 * which can happen on connect timeout or error.
2236 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002237 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002238 /* good connection is enough for pure TCP check */
2239 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002240 if (check->use_ssl)
2241 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002242 else
Simon Horman4a741432013-02-23 15:35:38 +09002243 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002244 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002245 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002246 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002247 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002248 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002249 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002250 }
2251
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002252 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002253 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002254 /* The check was aborted and the connection was not yet closed.
2255 * This can happen upon timeout, or when an external event such
2256 * as a failed response coupled with "observe layer7" caused the
2257 * server state to be suddenly changed.
2258 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002259 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002260 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002261 }
2262
Willy Tarreauac59f362017-10-08 11:10:19 +02002263 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002264 cs_destroy(cs);
2265 cs = check->cs = NULL;
2266 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002267 }
2268
Willy Tarreauaf549582014-05-16 17:37:50 +02002269 if (check->result == CHK_RES_FAILED) {
2270 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002271 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002272 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002273 else if (check->result == CHK_RES_CONDPASS) {
2274 /* check is OK but asks for stopping mode */
2275 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002276 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002277 else if (check->result == CHK_RES_PASSED) {
2278 /* a success was detected */
2279 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002280 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002281 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002282 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002283
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002284 rv = 0;
2285 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002286 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002287 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002288 }
Simon Horman4a741432013-02-23 15:35:38 +09002289 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002290 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002291
2292 reschedule:
2293 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002294 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002295 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002296 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002297 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002298}
2299
Simon Horman98637e52014-06-20 12:30:16 +09002300/*
2301 * manages a server health-check. Returns
2302 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2303 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002304static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002305{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002306 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002307
2308 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002309 return process_chk_proc(t, context, state);
2310 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002311
Simon Horman98637e52014-06-20 12:30:16 +09002312}
2313
Simon Horman5c942422013-11-25 10:46:32 +09002314static int start_check_task(struct check *check, int mininter,
2315 int nbcheck, int srvpos)
2316{
2317 struct task *t;
2318 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002319 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002320 ha_alert("Starting [%s:%s] check: out of memory.\n",
2321 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002322 return 0;
2323 }
2324
2325 check->task = t;
2326 t->process = process_chk;
2327 t->context = check;
2328
Willy Tarreau1746eec2014-04-25 10:46:47 +02002329 if (mininter < srv_getinter(check))
2330 mininter = srv_getinter(check);
2331
2332 if (global.max_spread_checks && mininter > global.max_spread_checks)
2333 mininter = global.max_spread_checks;
2334
Simon Horman5c942422013-11-25 10:46:32 +09002335 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002336 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002337 check->start = now;
2338 task_queue(t);
2339
2340 return 1;
2341}
2342
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002343/*
2344 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002345 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002346 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002347static int start_checks()
2348{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002349
2350 struct proxy *px;
2351 struct server *s;
2352 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002353 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002354
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002355 /* 1- count the checkers to run simultaneously.
2356 * We also determine the minimum interval among all of those which
2357 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2358 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002359 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002360 * too short an interval for all others.
2361 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002362 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002363 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002364 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002365 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002366 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002367 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002368 }
2369 /* We need a warmup task that will be called when the server
2370 * state switches from down to up.
2371 */
2372 s->warmup = t;
2373 t->process = server_warmup;
2374 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002375 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002376 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002377 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 +01002378 }
2379
Willy Tarreaud8514a22013-12-11 21:10:14 +01002380 if (s->check.state & CHK_ST_CONFIGURED) {
2381 nbcheck++;
2382 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2383 (!mininter || mininter > srv_getinter(&s->check)))
2384 mininter = srv_getinter(&s->check);
2385 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002386
Willy Tarreaud8514a22013-12-11 21:10:14 +01002387 if (s->agent.state & CHK_ST_CONFIGURED) {
2388 nbcheck++;
2389 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2390 (!mininter || mininter > srv_getinter(&s->agent)))
2391 mininter = srv_getinter(&s->agent);
2392 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002393 }
2394 }
2395
Simon Horman4a741432013-02-23 15:35:38 +09002396 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002397 return 0;
2398
2399 srand((unsigned)time(NULL));
2400
2401 /*
2402 * 2- start them as far as possible from each others. For this, we will
2403 * start them after their interval set to the min interval divided by
2404 * the number of servers, weighted by the server's position in the list.
2405 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002406 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002407 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2408 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002409 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002410 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002411 }
2412 }
2413
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002414 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002415 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002416 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002417 if (s->check.type == PR_O2_EXT_CHK) {
2418 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002419 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002420 }
Simon Hormand60d6912013-11-25 10:46:36 +09002421 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002422 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002423 srvpos++;
2424 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002425
Simon Hormand60d6912013-11-25 10:46:36 +09002426 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002427 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002428 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002429 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002430 }
2431 srvpos++;
2432 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002433 }
2434 }
2435 return 0;
2436}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002437
2438/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002439 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002440 * The buffer MUST be terminated by a null byte before calling this function.
2441 * Sets server status appropriately. The caller is responsible for ensuring
2442 * that the buffer contains at least 13 characters. If <done> is zero, we may
2443 * return 0 to indicate that data is required to decide of a match.
2444 */
2445static int httpchk_expect(struct server *s, int done)
2446{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002447 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002448 char status_code[] = "000";
2449 char *contentptr;
2450 int crlf;
2451 int ret;
2452
2453 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2454 case PR_O2_EXP_STS:
2455 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002456 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2457 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002458
2459 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2460 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2461 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002462 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002463
2464 /* we necessarily have the response, so there are no partial failures */
2465 if (s->proxy->options2 & PR_O2_EXP_INV)
2466 ret = !ret;
2467
Simon Horman4a741432013-02-23 15:35:38 +09002468 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002469 break;
2470
2471 case PR_O2_EXP_STR:
2472 case PR_O2_EXP_RSTR:
2473 /* very simple response parser: ignore CR and only count consecutive LFs,
2474 * stop with contentptr pointing to first char after the double CRLF or
2475 * to '\0' if crlf < 2.
2476 */
2477 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002478 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002479 if (crlf >= 2)
2480 break;
2481 if (*contentptr == '\r')
2482 continue;
2483 else if (*contentptr == '\n')
2484 crlf++;
2485 else
2486 crlf = 0;
2487 }
2488
2489 /* Check that response contains a body... */
2490 if (crlf < 2) {
2491 if (!done)
2492 return 0;
2493
Simon Horman4a741432013-02-23 15:35:38 +09002494 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002495 "HTTP content check could not find a response body");
2496 return 1;
2497 }
2498
2499 /* Check that response body is not empty... */
2500 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002501 if (!done)
2502 return 0;
2503
Simon Horman4a741432013-02-23 15:35:38 +09002504 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002505 "HTTP content check found empty response body");
2506 return 1;
2507 }
2508
2509 /* Check the response content against the supplied string
2510 * or regex... */
2511 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2512 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2513 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002514 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002515
2516 /* if we don't match, we may need to wait more */
2517 if (!ret && !done)
2518 return 0;
2519
2520 if (ret) {
2521 /* content matched */
2522 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002523 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002524 "HTTP check matched unwanted content");
2525 else
Simon Horman4a741432013-02-23 15:35:38 +09002526 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002527 "HTTP content check matched");
2528 }
2529 else {
2530 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002531 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002532 "HTTP check did not match unwanted content");
2533 else
Simon Horman4a741432013-02-23 15:35:38 +09002534 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002535 "HTTP content check did not match");
2536 }
2537 break;
2538 }
2539 return 1;
2540}
2541
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002542/*
2543 * return the id of a step in a send/expect session
2544 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002545static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002546{
2547 struct tcpcheck_rule *cur = NULL, *next = NULL;
2548 int i = 0;
2549
Willy Tarreau213c6782014-10-02 14:51:02 +02002550 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002551 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002552 return 0;
2553
Simon Hormane16c1b32015-01-30 11:22:57 +09002554 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002555
2556 /* no step => first step */
2557 if (cur == NULL)
2558 return 1;
2559
2560 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002561 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002562 if (next->list.p == &cur->list)
2563 break;
2564 ++i;
2565 }
2566
2567 return i;
2568}
2569
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002570/*
2571 * return the latest known comment before (including) the given stepid
2572 * returns NULL if no comment found
2573 */
2574static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2575{
2576 struct tcpcheck_rule *cur = NULL;
2577 char *ret = NULL;
2578 int i = 0;
2579
2580 /* not even started anything yet, return latest comment found before any action */
2581 if (!check->current_step) {
2582 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2583 if (cur->action == TCPCHK_ACT_COMMENT)
2584 ret = cur->comment;
2585 else
2586 goto return_comment;
2587 }
2588 }
2589
2590 i = 1;
2591 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2592 if (cur->comment)
2593 ret = cur->comment;
2594
2595 if (i >= stepid)
2596 goto return_comment;
2597
2598 ++i;
2599 }
2600
2601 return_comment:
2602 return ret;
2603}
2604
Willy Tarreaube74b882017-10-04 16:22:49 +02002605/* proceed with next steps for the TCP checks <check>. Note that this is called
2606 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002607 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2608 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002609 *
2610 * Please do NOT place any return statement in this function and only leave
2611 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002612 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002613static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002614{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002615 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002616 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002617 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002618 struct conn_stream *cs = check->cs;
2619 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002620 struct server *s = check->server;
2621 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002622 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002623 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002624
Willy Tarreauef953952014-10-02 14:30:14 +02002625 /* here, we know that the check is complete or that it failed */
2626 if (check->result != CHK_RES_UNKNOWN)
2627 goto out_end_tcpcheck;
2628
2629 /* We have 4 possibilities here :
2630 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002631 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002632 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2633 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002634 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002635 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002636 * be waiting for a connection attempt to complete. conn!=NULL.
2637 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002638 *
2639 * #2 and #3 are quite similar, we want both the connection and the
2640 * handshake to complete before going any further. Thus we must always
2641 * wait for a connection to complete unless we're before and existing
2642 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002643 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002644
2645 /* find first rule and skip comments */
2646 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2647 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2648 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2649
Willy Tarreau00149122017-10-04 18:05:01 +02002650 if ((check->current_step || &next->list == head) &&
2651 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002652 /* we allow up to min(inter, timeout.connect) for a connection
2653 * to establish but only when timeout.check is set
2654 * as it may be to short for a full check otherwise
2655 */
2656 while (tick_is_expired(t->expire, now_ms)) {
2657 int t_con;
2658
2659 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2660 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2661
2662 if (s->proxy->timeout.check)
2663 t->expire = tick_first(t->expire, t_con);
2664 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002665 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002666 }
2667
Willy Tarreauef953952014-10-02 14:30:14 +02002668 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002669 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002670 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002671 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002672 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002673
Willy Tarreau213c6782014-10-02 14:51:02 +02002674 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002675 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002676 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002677 b_reset(&check->bo);
2678 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002679 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002680 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2681 if (s->proxy->timeout.check)
2682 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2683 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002684
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002685 /* It's only the rules which will enable send/recv */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002686 if (cs)
2687 cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002688
Willy Tarreauabca5b62013-12-06 14:19:25 +01002689 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002690 /* We have to try to flush the output buffer before reading, at
2691 * the end, or if we're about to send a string that does not fit
2692 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002693 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002694 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002695 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002696 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002697 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002698 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002699 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002700
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002701 __cs_want_send(cs);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002702 ret = conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
2703 b_del(&check->bo, ret);
2704 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002705
2706 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002707 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002708 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002709 __cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002710 goto out_end_tcpcheck;
2711 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002712 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002713 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002714 }
2715
Willy Tarreau263013d2015-05-13 11:59:14 +02002716 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002717 break;
2718
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002719 /* have 'next' point to the next rule or NULL if we're on the
2720 * last one, connect() needs this.
2721 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002722 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002723
2724 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002725 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002726 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002727
2728 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002729 if (&next->list == head)
2730 next = NULL;
2731
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002732 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2733 struct protocol *proto;
2734 struct xprt_ops *xprt;
2735
Willy Tarreau00149122017-10-04 18:05:01 +02002736 /* For a connect action we'll create a new connection.
2737 * We may also have to kill a previous one. But we don't
2738 * want to leave *without* a connection if we came here
2739 * from the connection layer, hence with a connection.
2740 * Thus we'll proceed in the following order :
2741 * 1: close but not release previous connection
2742 * 2: try to get a new connection
2743 * 3: release and replace the old one on success
2744 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002745 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002746 cs_close(check->cs);
2747 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002748 }
2749
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002750 /* mark the step as started */
2751 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002752
2753 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002754 cs = cs_new(NULL);
2755 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002756 step = tcpcheck_get_step_id(check);
2757 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2758 comment = tcpcheck_get_step_comment(check, step);
2759 if (comment)
2760 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002761 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2762 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002763 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002764 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002765 }
2766
Willy Tarreauac59f362017-10-08 11:10:19 +02002767 if (check->cs)
2768 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002769
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002770 check->cs = cs;
2771 conn = cs->conn;
2772 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002773 conn->target = &s->obj_type;
2774
2775 /* no client address */
2776 clear_addr(&conn->addr.from);
2777
Simon Horman41f58762015-01-30 11:22:56 +09002778 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002779 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002780 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002781 }
2782 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002783 /* we'll connect to the addr on the server */
2784 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002785 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002786 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002787
2788 /* port */
2789 if (check->current_step->port)
2790 set_host_port(&conn->addr.to, check->current_step->port);
2791 else if (check->port)
2792 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002793 else if (s->svc_port)
2794 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002795
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002796 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002797 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002798 }
2799 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002800 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002801 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002802 conn_prepare(conn, proto, xprt);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002803 conn_install_mux(conn, &mux_pt_ops, cs);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002804
Willy Tarreaue7dff022015-04-03 01:14:29 +02002805 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002806 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002807 ret = proto->connect(conn,
2808 1 /* I/O polling is always needed */,
2809 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002810 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2811 conn->send_proxy_ofs = 1;
2812 conn->flags |= CO_FL_SEND_PROXY;
2813 }
2814
2815 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002816 * - SF_ERR_NONE if everything's OK
2817 * - SF_ERR_SRVTO if there are no more servers
2818 * - SF_ERR_SRVCL if the connection was refused by the server
2819 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2820 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2821 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002822 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002823 * Note that we try to prevent the network stack from sending the ACK during the
2824 * connect() when a pure TCP check is used (without PROXY protocol).
2825 */
2826 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002827 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002828 /* we allow up to min(inter, timeout.connect) for a connection
2829 * to establish but only when timeout.check is set
2830 * as it may be to short for a full check otherwise
2831 */
2832 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2833
2834 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2835 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2836 t->expire = tick_first(t->expire, t_con);
2837 }
2838 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002839 case SF_ERR_SRVTO: /* ETIMEDOUT */
2840 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002841 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002842 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002843 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002844 comment = tcpcheck_get_step_comment(check, step);
2845 if (comment)
2846 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002847 set_server_check_status(check, HCHK_STATUS_L4CON,
2848 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002849 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002850 case SF_ERR_PRXCOND:
2851 case SF_ERR_RESOURCE:
2852 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002853 step = tcpcheck_get_step_id(check);
2854 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002855 comment = tcpcheck_get_step_comment(check, step);
2856 if (comment)
2857 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002858 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2859 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002860 goto out_end_tcpcheck;
2861 }
2862
2863 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002864 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002865
2866 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002867 while (&check->current_step->list != head &&
2868 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002869 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002870
Willy Tarreauf2c87352015-05-13 12:08:21 +02002871 if (&check->current_step->list == head)
2872 break;
2873
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002874 } /* end 'connect' */
2875 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2876 /* mark the step as started */
2877 check->last_started_step = check->current_step;
2878
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002879 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002880 if (*b_head(&check->bi) != '\0') {
2881 *b_head(&check->bi) = '\0';
2882 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002883 }
2884
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002885 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002886 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002887 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002888 goto out_end_tcpcheck;
2889 }
2890
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002891 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002892 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002893 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002894 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002895 set_server_check_status(check, HCHK_STATUS_L7RSP,
2896 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002897 goto out_end_tcpcheck;
2898 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002899
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002900 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002901 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002902 continue;
2903
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002904 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2905 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002906
Willy Tarreauabca5b62013-12-06 14:19:25 +01002907 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002908 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002909
2910 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002911 while (&check->current_step->list != head &&
2912 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002913 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002914
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002915 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002916 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002917 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002918 goto out_end_tcpcheck;
2919
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002920 __cs_want_recv(cs);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002921 if (conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002922 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002923 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002924 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002925 /* Report network errors only if we got no other data. Otherwise
2926 * we'll let the upper layers decide whether the response is OK
2927 * or not. It is very common that an RST sent by the server is
2928 * reported as an error just after the last data chunk.
2929 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002930 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002931 goto out_end_tcpcheck;
2932 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002933 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002934 else
Willy Tarreau263013d2015-05-13 11:59:14 +02002935 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002936 }
2937
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002938 /* mark the step as started */
2939 check->last_started_step = check->current_step;
2940
2941
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002942 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002943 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002944 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002945 if (b_data(&check->bi) < b_size(&check->bi)) {
2946 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002947 }
2948 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002949 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002950 done = 1; /* buffer full, don't wait for more data */
2951 }
2952
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002953 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002954
2955 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002956 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002957 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002958 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002959
2960 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002961 step = tcpcheck_get_step_id(check);
2962 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002963 comment = tcpcheck_get_step_comment(check, step);
2964 if (comment)
2965 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002966 set_server_check_status(check, HCHK_STATUS_L7RSP,
2967 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002968
2969 goto out_end_tcpcheck;
2970 }
2971
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002972 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002973 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002974
Willy Tarreaua970c282013-12-06 12:47:19 +01002975 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002976 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002977 ret = my_memmem(contentptr, b_data(&check->bi), check->current_step->string, check->current_step->string_len) != NULL;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002978 else if (check->current_step->expect_regex != NULL)
2979 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980
2981 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002982 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002983
2984 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002985 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002986 if (ret) {
2987 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002988 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002989 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002990 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002991 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002992 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002993 }
2994 else {
2995 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002996 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002997 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002998 comment = tcpcheck_get_step_comment(check, step);
2999 if (comment)
3000 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003001 set_server_check_status(check, HCHK_STATUS_L7RSP,
3002 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003003 goto out_end_tcpcheck;
3004 }
3005 /* matched and was supposed to => OK, next step */
3006 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003007 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003008 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003009
3010 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003011 while (&check->current_step->list != head &&
3012 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003013 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003014
Willy Tarreauf2c87352015-05-13 12:08:21 +02003015 if (&check->current_step->list == head)
3016 break;
3017
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003018 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003019 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003020 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003021 }
3022 }
3023 else {
3024 /* not matched */
3025 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003026 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003027 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003028 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003029
3030 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003031 while (&check->current_step->list != head &&
3032 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003033 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003034
Willy Tarreauf2c87352015-05-13 12:08:21 +02003035 if (&check->current_step->list == head)
3036 break;
3037
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003038 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003039 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003040 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003041 }
3042 /* not matched but was supposed to => ERROR */
3043 else {
3044 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003045 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003046 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003047 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003048 }
3049 else {
3050 /* we were looking for a regex */
3051 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003052 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003053 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003054 comment = tcpcheck_get_step_comment(check, step);
3055 if (comment)
3056 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003057 set_server_check_status(check, HCHK_STATUS_L7RSP,
3058 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003059 goto out_end_tcpcheck;
3060 }
3061 }
3062 } /* end expect */
3063 } /* end loop over double chained step list */
3064
Baptiste Assmann248f1172018-03-01 21:49:01 +01003065 /* don't do anything until the connection is established */
3066 if (!(conn->flags & CO_FL_CONNECTED)) {
3067 /* update expire time, should be done by process_chk */
3068 /* we allow up to min(inter, timeout.connect) for a connection
3069 * to establish but only when timeout.check is set
3070 * as it may be to short for a full check otherwise
3071 */
3072 while (tick_is_expired(t->expire, now_ms)) {
3073 int t_con;
3074
3075 t_con = tick_add(t->expire, s->proxy->timeout.connect);
3076 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3077
3078 if (s->proxy->timeout.check)
3079 t->expire = tick_first(t->expire, t_con);
3080 }
3081 goto out;
3082 }
3083
Willy Tarreau263013d2015-05-13 11:59:14 +02003084 /* We're waiting for some I/O to complete, we've reached the end of the
3085 * rules, or both. Do what we have to do, otherwise we're done.
3086 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003087 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003088 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3089 goto out_end_tcpcheck;
3090 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003091
Willy Tarreau53c5a042015-05-13 11:38:17 +02003092 /* warning, current_step may now point to the head */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003093 if (b_data(&check->bo))
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003094 __cs_want_send(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003095
Willy Tarreau53c5a042015-05-13 11:38:17 +02003096 if (&check->current_step->list != head &&
3097 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003098 __cs_want_recv(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003099 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003100
3101 out_end_tcpcheck:
3102 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003103 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003104 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003105
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003106 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003107 check->current_step = NULL;
3108
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003109 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003110 conn->flags |= CO_FL_ERROR;
3111
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003112 __cs_stop_both(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01003113
Christopher Fauletb6102852017-11-28 10:06:29 +01003114 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003115 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003116}
3117
Simon Hormanb1900d52015-01-30 11:22:54 +09003118const char *init_check(struct check *check, int type)
3119{
3120 check->type = type;
3121
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003122 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3123 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003124
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003125 check->bi.area = calloc(check->bi.size, sizeof(char));
3126 check->bo.area = calloc(check->bo.size, sizeof(char));
3127
3128 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003129 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003130
Simon Hormanb1900d52015-01-30 11:22:54 +09003131 return NULL;
3132}
3133
Simon Hormanbfb5d332015-01-30 11:22:55 +09003134void free_check(struct check *check)
3135{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003136 free(check->bi.area);
3137 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003138 if (check->cs) {
3139 free(check->cs->conn);
3140 check->cs->conn = NULL;
3141 cs_free(check->cs);
3142 check->cs = NULL;
3143 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003144}
3145
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003146void email_alert_free(struct email_alert *alert)
3147{
3148 struct tcpcheck_rule *rule, *back;
3149
3150 if (!alert)
3151 return;
3152
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003153 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3154 LIST_DEL(&rule->list);
3155 free(rule->comment);
3156 free(rule->string);
3157 if (rule->expect_regex)
3158 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003159 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003160 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003161 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003162}
3163
Olivier Houchard9f6af332018-05-25 14:04:04 +02003164static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003165{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003166 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003167 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003168 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003169
3170 q = container_of(check, typeof(*q), check);
3171
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003172 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003173 while (1) {
3174 if (!(check->state & CHK_ST_ENABLED)) {
3175 if (LIST_ISEMPTY(&q->email_alerts)) {
3176 /* All alerts processed, queue the task */
3177 t->expire = TICK_ETERNITY;
3178 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003179 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003180 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003181
3182 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003183 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003184 t->expire = now_ms;
3185 check->server = alert->srv;
3186 check->tcpcheck_rules = &alert->tcpcheck_rules;
PiBa-NL1714b9f2017-12-06 01:35:43 +01003187 check->status = HCHK_STATUS_UNKNOWN; // the UNKNOWN status is used to exit set_server_check_status(.) early
Christopher Faulet0108bb32017-10-20 21:34:32 +02003188 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003189 }
3190
Olivier Houchard9f6af332018-05-25 14:04:04 +02003191 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003192 if (check->state & CHK_ST_INPROGRESS)
3193 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003194
3195 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3196 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003197 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003198 check->server = NULL;
3199 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003200 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003201 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003202 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003203 return t;
3204}
3205
Christopher Faulet0108bb32017-10-20 21:34:32 +02003206/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3207 *
3208 * The function returns 1 in success case, otherwise, it returns 0 and err is
3209 * filled.
3210 */
3211int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003212{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003213 struct mailer *mailer;
3214 struct email_alertq *queues;
3215 const char *err_str;
3216 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003217
Christopher Faulet0108bb32017-10-20 21:34:32 +02003218 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3219 memprintf(err, "out of memory while allocating mailer alerts queues");
3220 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003221 }
3222
Christopher Faulet0108bb32017-10-20 21:34:32 +02003223 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3224 struct email_alertq *q = &queues[i];
3225 struct check *check = &q->check;
3226 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003227
3228 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003229 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003230 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003231 check->rise = DEF_AGENT_RISETIME;
3232 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003233 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3234 memprintf(err, "%s", err_str);
3235 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003236 }
3237
3238 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003239 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003240 check->port = get_host_port(&mailer->addr);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003241 //check->server = s;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003242
Emeric Brunc60def82017-09-27 14:59:38 +02003243 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003244 memprintf(err, "out of memory while allocating mailer alerts task");
3245 goto error;
3246 }
3247
3248 check->task = t;
3249 t->process = process_email_alert;
3250 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003251
Christopher Faulet0108bb32017-10-20 21:34:32 +02003252 /* check this in one ms */
3253 t->expire = TICK_ETERNITY;
3254 check->start = now;
3255 task_queue(t);
3256 }
3257
3258 mls->users++;
3259 free(p->email_alert.mailers.name);
3260 p->email_alert.mailers.m = mls;
3261 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003262 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003263
3264 error:
3265 for (i = 0; i < mls->count; i++) {
3266 struct email_alertq *q = &queues[i];
3267 struct check *check = &q->check;
3268
3269 if (check->task) {
3270 task_delete(check->task);
3271 task_free(check->task);
3272 check->task = NULL;
3273 }
3274 free_check(check);
3275 }
3276 free(queues);
3277 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003278}
3279
3280
3281static int add_tcpcheck_expect_str(struct list *list, const char *str)
3282{
3283 struct tcpcheck_rule *tcpcheck;
3284
Willy Tarreaubafbe012017-11-24 17:34:44 +01003285 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003286 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003287 memset(tcpcheck, 0, sizeof(*tcpcheck));
3288 tcpcheck->action = TCPCHK_ACT_EXPECT;
3289 tcpcheck->string = strdup(str);
3290 tcpcheck->expect_regex = NULL;
3291 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003292 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003293 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003294 return 0;
3295 }
3296
3297 LIST_ADDQ(list, &tcpcheck->list);
3298 return 1;
3299}
3300
3301static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3302{
3303 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003304 const char *in;
3305 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306 int i;
3307
Willy Tarreaubafbe012017-11-24 17:34:44 +01003308 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003309 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003310 memset(tcpcheck, 0, sizeof(*tcpcheck));
3311 tcpcheck->action = TCPCHK_ACT_SEND;
3312 tcpcheck->expect_regex = NULL;
3313 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003314 tcpcheck->string_len = 0;
3315 for (i = 0; strs[i]; i++)
3316 tcpcheck->string_len += strlen(strs[i]);
3317
3318 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3319 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003320 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003321 return 0;
3322 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003323
Willy Tarreau64345aa2016-08-10 19:29:09 +02003324 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003325 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003326 for (in = strs[i]; (*dst = *in++); dst++);
3327 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003328
3329 LIST_ADDQ(list, &tcpcheck->list);
3330 return 1;
3331}
3332
Christopher Faulet0108bb32017-10-20 21:34:32 +02003333static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3334 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003335{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003336 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003337 struct tcpcheck_rule *tcpcheck;
3338 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003339
Willy Tarreaubafbe012017-11-24 17:34:44 +01003340 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003341 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003342 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003343 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003344 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003345
Willy Tarreaubafbe012017-11-24 17:34:44 +01003346 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003347 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003348 memset(tcpcheck, 0, sizeof(*tcpcheck));
3349 tcpcheck->action = TCPCHK_ACT_CONNECT;
3350 tcpcheck->comment = NULL;
3351 tcpcheck->string = NULL;
3352 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003353 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3354
3355 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3356 goto error;
3357
3358 {
3359 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3360 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3361 goto error;
3362 }
3363
3364 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3365 goto error;
3366
3367 {
3368 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3369 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3370 goto error;
3371 }
3372
3373 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3374 goto error;
3375
3376 {
3377 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3378 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3379 goto error;
3380 }
3381
3382 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3383 goto error;
3384
3385 {
3386 const char * const strs[2] = { "DATA\r\n" };
3387 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3388 goto error;
3389 }
3390
3391 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3392 goto error;
3393
3394 {
3395 struct tm tm;
3396 char datestr[48];
3397 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003398 "From: ", p->email_alert.from, "\r\n",
3399 "To: ", p->email_alert.to, "\r\n",
3400 "Date: ", datestr, "\r\n",
3401 "Subject: [HAproxy Alert] ", msg, "\r\n",
3402 "\r\n",
3403 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003404 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003405 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003406 NULL
3407 };
3408
3409 get_localtime(date.tv_sec, &tm);
3410
3411 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3412 goto error;
3413 }
3414
3415 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3416 goto error;
3417 }
3418
3419 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3420 goto error;
3421
3422 {
3423 const char * const strs[2] = { "QUIT\r\n" };
3424 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3425 goto error;
3426 }
3427
3428 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3429 goto error;
3430
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003431 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003432 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003433 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003434 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003435 return 1;
3436
3437error:
3438 email_alert_free(alert);
3439 return 0;
3440}
3441
Christopher Faulet0108bb32017-10-20 21:34:32 +02003442static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003443{
3444 int i;
3445 struct mailer *mailer;
3446
3447 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3448 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003449 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003450 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003451 return;
3452 }
3453 }
3454
3455 return;
3456}
3457
3458/*
3459 * Send email alert if configured.
3460 */
Simon Horman64e34162015-02-06 11:11:57 +09003461void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003462{
3463 va_list argp;
3464 char buf[1024];
3465 int len;
3466 struct proxy *p = s->proxy;
3467
Christopher Faulet0108bb32017-10-20 21:34:32 +02003468 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003469 return;
3470
3471 va_start(argp, format);
3472 len = vsnprintf(buf, sizeof(buf), format, argp);
3473 va_end(argp);
3474
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003475 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003476 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003477 return;
3478 }
3479
Christopher Faulet0108bb32017-10-20 21:34:32 +02003480 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003481}
3482
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003483/*
3484 * Return value:
3485 * the port to be used for the health check
3486 * 0 in case no port could be found for the check
3487 */
3488int srv_check_healthcheck_port(struct check *chk)
3489{
3490 int i = 0;
3491 struct server *srv = NULL;
3492
3493 srv = chk->server;
3494
3495 /* If neither a port nor an addr was specified and no check transport
3496 * layer is forced, then the transport layer used by the checks is the
3497 * same as for the production traffic. Otherwise we use raw_sock by
3498 * default, unless one is specified.
3499 */
3500 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003501 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003502 chk->send_proxy |= (srv->pp_opts);
3503 }
3504
3505 /* by default, we use the health check port ocnfigured */
3506 if (chk->port > 0)
3507 return chk->port;
3508
3509 /* try to get the port from check_core.addr if check.port not set */
3510 i = get_host_port(&chk->addr);
3511 if (i > 0)
3512 return i;
3513
3514 /* try to get the port from server address */
3515 /* prevent MAPPORTS from working at this point, since checks could
3516 * not be performed in such case (MAPPORTS impose a relative ports
3517 * based on live traffic)
3518 */
3519 if (srv->flags & SRV_F_MAPPORTS)
3520 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003521
3522 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003523 if (i > 0)
3524 return i;
3525
3526 return 0;
3527}
3528
Willy Tarreau865c5142016-12-21 20:04:48 +01003529__attribute__((constructor))
3530static void __check_init(void)
3531{
3532 hap_register_post_check(start_checks);
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003533
Willy Tarreaubafbe012017-11-24 17:34:44 +01003534 pool_head_email_alert = create_pool("email_alert", sizeof(struct email_alert), MEM_F_SHARED);
3535 pool_head_tcpcheck_rule = create_pool("tcpcheck_rule", sizeof(struct tcpcheck_rule), MEM_F_SHARED);
Willy Tarreau865c5142016-12-21 20:04:48 +01003536}
3537
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003538
Willy Tarreaubd741542010-03-16 18:46:54 +01003539/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003540 * Local variables:
3541 * c-indent-level: 8
3542 * c-basic-offset: 8
3543 * End:
3544 */