blob: 2359176429542d3e24ab284cca18ba105aca51dd [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
Willy Tarreau9f6dc722019-03-01 11:15:10 +010025#include <sys/resource.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020026#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090028#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020029#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020030#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020031#include <arpa/inet.h>
32
Willy Tarreauc7e42382012-08-24 19:22:53 +020033#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/compat.h>
35#include <common/config.h>
36#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020037#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020038#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020039#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040
41#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020042#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010043#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020044
45#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020046#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010047#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020048#include <proto/fd.h>
49#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020050#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020052#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
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);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020073static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020074static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010075
Willy Tarreau8ceae722018-11-26 11:58:30 +010076DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
77DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020078
79
Simon Horman63a4a822012-03-19 07:24:41 +090080static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010081 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
82 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020083 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020084
Willy Tarreau23964182014-05-20 20:56:30 +020085 /* Below we have finished checks */
86 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010087 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010088
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010089 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020090
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010091 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
92 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
93 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020094
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010095 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
96 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
97 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010099 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
100 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200101
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200102 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200103
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100104 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
105 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
106 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900107
108 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
109 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200110 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200111};
112
Cyril Bontéac92a062014-12-27 22:28:38 +0100113const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
114 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
122 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
123 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
124 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
125};
126
Simon Horman63a4a822012-03-19 07:24:41 +0900127static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100128 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
129
130 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
131 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
132
133 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
134 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
135 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
136 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
137
138 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
139 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
140 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
141};
142
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200143/*
144 * Convert check_status code to description
145 */
146const char *get_check_status_description(short check_status) {
147
148 const char *desc;
149
150 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200151 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200152 else
153 desc = NULL;
154
155 if (desc && *desc)
156 return desc;
157 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200158 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200159}
160
161/*
162 * Convert check_status code to short info
163 */
164const char *get_check_status_info(short check_status) {
165
166 const char *info;
167
168 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200169 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200170 else
171 info = NULL;
172
173 if (info && *info)
174 return info;
175 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200176 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200177}
178
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100179const char *get_analyze_status(short analyze_status) {
180
181 const char *desc;
182
183 if (analyze_status < HANA_STATUS_SIZE)
184 desc = analyze_statuses[analyze_status].desc;
185 else
186 desc = NULL;
187
188 if (desc && *desc)
189 return desc;
190 else
191 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
192}
193
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200194/*
Simon Horman4a741432013-02-23 15:35:38 +0900195 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200196 * an adequate CHK_RES_* value. The new check->health is computed based
197 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200198 *
199 * Show information in logs about failed health check if server is UP
200 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200201 */
Simon Horman4a741432013-02-23 15:35:38 +0900202static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100203{
Simon Horman4a741432013-02-23 15:35:38 +0900204 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200205 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200206 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900207
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200208 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100209 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900210 check->desc[0] = '\0';
211 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200212 return;
213 }
214
Simon Horman4a741432013-02-23 15:35:38 +0900215 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200216 return;
217
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200218 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900219 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
220 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200221 } else
Simon Horman4a741432013-02-23 15:35:38 +0900222 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200223
Simon Horman4a741432013-02-23 15:35:38 +0900224 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200225 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900226 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200227
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100228 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900229 check->duration = -1;
230 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200231 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900232 check->duration = tv_ms_elapsed(&check->start, &now);
233 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200234 }
235
Willy Tarreau23964182014-05-20 20:56:30 +0200236 /* no change is expected if no state change occurred */
237 if (check->result == CHK_RES_NEUTRAL)
238 return;
239
Olivier Houchard0923fa42019-01-11 18:43:04 +0100240 /* If the check was really just sending a mail, it won't have an
241 * associated server, so we're done now.
242 */
243 if (!s)
244 return;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200245 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200246
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200247 switch (check->result) {
248 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200249 /* Failure to connect to the agent as a secondary check should not
250 * cause the server to be marked down.
251 */
252 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900253 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200254 (check->health > 0)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200255 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200256 report = 1;
257 check->health--;
258 if (check->health < check->rise)
259 check->health = 0;
260 }
261 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200262
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200263 case CHK_RES_PASSED:
264 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
265 if ((check->health < check->rise + check->fall - 1) &&
266 (check->result == CHK_RES_PASSED || check->health > 0)) {
267 report = 1;
268 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200269
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200270 if (check->health >= check->rise)
271 check->health = check->rise + check->fall - 1; /* OK now */
272 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200273
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200274 /* clear consecutive_errors if observing is enabled */
275 if (s->onerror)
276 s->consecutive_errors = 0;
277 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100278
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200279 default:
280 break;
281 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200282
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200283 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
284 (status != prev_status || report)) {
285 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200286 "%s check for %sserver %s/%s %s%s",
287 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200288 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100289 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100290 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200291 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200292
Emeric Brun5a133512017-10-19 14:42:30 +0200293 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200294
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100295 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200296 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
297 (check->health >= check->rise) ? check->fall : check->rise,
298 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200299
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200300 ha_warning("%s.\n", trash.area);
301 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
302 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200303 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200304}
305
Willy Tarreau4eec5472014-05-20 22:32:27 +0200306/* Marks the check <check>'s server down if the current check is already failed
307 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200308 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200309static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200310{
Simon Horman4a741432013-02-23 15:35:38 +0900311 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900312
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200313 /* The agent secondary check should only cause a server to be marked
314 * as down if check->status is HCHK_STATUS_L7STS, which indicates
315 * that the agent returned "fail", "stopped" or "down".
316 * The implication here is that failure to connect to the agent
317 * as a secondary check should not cause the server to be marked
318 * down. */
319 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
320 return;
321
Willy Tarreau4eec5472014-05-20 22:32:27 +0200322 if (check->health > 0)
323 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100324
Willy Tarreau4eec5472014-05-20 22:32:27 +0200325 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200326 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200327}
328
Willy Tarreauaf549582014-05-16 17:37:50 +0200329/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200330 * it isn't in maintenance, it is not tracking a down server and other checks
331 * comply. The rule is simple : by default, a server is up, unless any of the
332 * following conditions is true :
333 * - health check failed (check->health < rise)
334 * - agent check failed (agent->health < rise)
335 * - the server tracks a down server (track && track->state == STOPPED)
336 * Note that if the server has a slowstart, it will switch to STARTING instead
337 * of RUNNING. Also, only the health checks support the nolb mode, so the
338 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200339 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200340static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200341{
Simon Horman4a741432013-02-23 15:35:38 +0900342 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100343
Emeric Brun52a91d32017-08-31 14:41:55 +0200344 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200345 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100346
Emeric Brun52a91d32017-08-31 14:41:55 +0200347 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200348 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100349
Willy Tarreau3e048382014-05-21 10:30:54 +0200350 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
351 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100352
Willy Tarreau3e048382014-05-21 10:30:54 +0200353 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
354 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200355
Emeric Brun52a91d32017-08-31 14:41:55 +0200356 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200357 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100358
Emeric Brun5a133512017-10-19 14:42:30 +0200359 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100360}
361
Willy Tarreaudb58b792014-05-21 13:57:23 +0200362/* Marks the check <check> as valid and tries to set its server into stopping mode
363 * if it was running or starting, and provided it isn't in maintenance and other
364 * checks comply. The conditions for the server to be marked in stopping mode are
365 * the same as for it to be turned up. Also, only the health checks support the
366 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200367 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200368static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200369{
Simon Horman4a741432013-02-23 15:35:38 +0900370 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100371
Emeric Brun52a91d32017-08-31 14:41:55 +0200372 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200373 return;
374
Willy Tarreaudb58b792014-05-21 13:57:23 +0200375 if (check->state & CHK_ST_AGENT)
376 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100377
Emeric Brun52a91d32017-08-31 14:41:55 +0200378 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200379 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100380
Willy Tarreaudb58b792014-05-21 13:57:23 +0200381 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
382 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100383
Willy Tarreaudb58b792014-05-21 13:57:23 +0200384 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
385 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100386
Willy Tarreaub26881a2017-12-23 11:16:49 +0100387 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100388}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200389
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100390/* note: use health_adjust() only, which first checks that the observe mode is
391 * enabled.
392 */
393void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100394{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100395 int failed;
396 int expire;
397
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100398 if (s->observe >= HANA_OBS_SIZE)
399 return;
400
Willy Tarreaubb956662013-01-24 00:37:39 +0100401 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100402 return;
403
404 switch (analyze_statuses[status].lr[s->observe - 1]) {
405 case 1:
406 failed = 1;
407 break;
408
409 case 2:
410 failed = 0;
411 break;
412
413 default:
414 return;
415 }
416
417 if (!failed) {
418 /* good: clear consecutive_errors */
419 s->consecutive_errors = 0;
420 return;
421 }
422
Christopher Faulet29f77e82017-06-08 14:04:45 +0200423 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100424
425 if (s->consecutive_errors < s->consecutive_errors_limit)
426 return;
427
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100428 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
429 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100430
431 switch (s->onerror) {
432 case HANA_ONERR_FASTINTER:
433 /* force fastinter - nothing to do here as all modes force it */
434 break;
435
436 case HANA_ONERR_SUDDTH:
437 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900438 if (s->check.health > s->check.rise)
439 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100440
441 /* no break - fall through */
442
443 case HANA_ONERR_FAILCHK:
444 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200445 set_server_check_status(&s->check, HCHK_STATUS_HANA,
446 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200447 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100448 break;
449
450 case HANA_ONERR_MARKDWN:
451 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900452 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200453 set_server_check_status(&s->check, HCHK_STATUS_HANA,
454 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200455 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100456 break;
457
458 default:
459 /* write a warning? */
460 break;
461 }
462
463 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200464 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100465
Simon Horman66183002013-02-23 10:16:43 +0900466 if (s->check.fastinter) {
467 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300468 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200469 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300470 /* requeue check task with new expire */
471 task_queue(s->check.task);
472 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100473 }
474}
475
Willy Tarreaua1dab552014-04-14 15:04:54 +0200476static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100477{
478 int sv_state;
479 int ratio;
480 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800481 char addr[46];
482 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100483 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
484 "UP %d/%d", "UP",
485 "NOLB %d/%d", "NOLB",
486 "no check" };
487
488 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
489 hlen += 24;
490
Willy Tarreauff5ae352013-12-11 20:36:34 +0100491 if (!(s->check.state & CHK_ST_ENABLED))
492 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200493 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900494 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100495 sv_state = 3; /* UP */
496 else
497 sv_state = 2; /* going down */
498
Emeric Brun52a91d32017-08-31 14:41:55 +0200499 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100500 sv_state += 2;
501 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900502 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100503 sv_state = 1; /* going up */
504 else
505 sv_state = 0; /* DOWN */
506 }
507
Willy Tarreaua1dab552014-04-14 15:04:54 +0200508 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100509 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200510 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
511 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100512
Joseph Lynch514061c2015-01-15 17:52:59 -0800513 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100514 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
515 snprintf(port, sizeof(port), "%u", s->svc_port);
516 else
517 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800518
519 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
520 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100521 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200522 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100523 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
524 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
525 s->nbpend);
526
Emeric Brun52a91d32017-08-31 14:41:55 +0200527 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100528 now.tv_sec < s->last_change + s->slowstart &&
529 now.tv_sec >= s->last_change) {
530 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200531 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100532 }
533
534 buffer[hlen++] = '\r';
535 buffer[hlen++] = '\n';
536
537 return hlen;
538}
539
Willy Tarreau20a18342013-12-05 00:31:46 +0100540/* Check the connection. If an error has already been reported or the socket is
541 * closed, keep errno intact as it is supposed to contain the valid error code.
542 * If no error is reported, check the socket's error queue using getsockopt().
543 * Warning, this must be done only once when returning from poll, and never
544 * after an I/O error was attempted, otherwise the error queue might contain
545 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
546 * socket. Returns non-zero if an error was reported, zero if everything is
547 * clean (including a properly closed socket).
548 */
549static int retrieve_errno_from_socket(struct connection *conn)
550{
551 int skerr;
552 socklen_t lskerr = sizeof(skerr);
553
554 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
555 return 1;
556
Willy Tarreau3c728722014-01-23 13:50:42 +0100557 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100558 return 0;
559
Willy Tarreau585744b2017-08-24 14:31:19 +0200560 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100561 errno = skerr;
562
563 if (errno == EAGAIN)
564 errno = 0;
565
566 if (!errno) {
567 /* we could not retrieve an error, that does not mean there is
568 * none. Just don't change anything and only report the prior
569 * error if any.
570 */
571 if (conn->flags & CO_FL_ERROR)
572 return 1;
573 else
574 return 0;
575 }
576
577 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
578 return 1;
579}
580
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100581/* Try to collect as much information as possible on the connection status,
582 * and adjust the server status accordingly. It may make use of <errno_bck>
583 * if non-null when the caller is absolutely certain of its validity (eg:
584 * checked just after a syscall). If the caller doesn't have a valid errno,
585 * it can pass zero, and retrieve_errno_from_socket() will be called to try
586 * to extract errno from the socket. If no error is reported, it will consider
587 * the <expired> flag. This is intended to be used when a connection error was
588 * reported in conn->flags or when a timeout was reported in <expired>. The
589 * function takes care of not updating a server status which was already set.
590 * All situations where at least one of <expired> or CO_FL_ERROR are set
591 * produce a status.
592 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200593static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100594{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200595 struct conn_stream *cs = check->cs;
596 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100597 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200598 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200599 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200600 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100602 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100603 return;
604
605 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200606 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100607 retrieve_errno_from_socket(conn);
608
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200609 if (conn && !(conn->flags & CO_FL_ERROR) &&
610 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100611 return;
612
613 /* we'll try to build a meaningful error message depending on the
614 * context of the error possibly present in conn->err_code, and the
615 * socket error possibly collected above. This is useful to know the
616 * exact step of the L6 layer (eg: SSL handshake).
617 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200618 chk = get_trash_chunk();
619
620 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900621 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200622 if (!step)
623 chunk_printf(chk, " at initial connection step of tcp-check");
624 else {
625 chunk_printf(chk, " at step %d of tcp-check", step);
626 /* we were looking for a string */
627 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
628 if (check->last_started_step->port)
629 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
630 else
631 chunk_appendf(chk, " (connect)");
632 }
633 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
634 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200635 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200636 else if (check->last_started_step->expect_regex)
637 chunk_appendf(chk, " (expect regex)");
638 }
639 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
640 chunk_appendf(chk, " (send)");
641 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200642
643 comment = tcpcheck_get_step_comment(check, step);
644 if (comment)
645 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200646 }
647 }
648
Willy Tarreau00149122017-10-04 18:05:01 +0200649 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100650 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200651 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
652 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100653 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200654 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
655 chk->area);
656 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100657 }
658 else {
659 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200660 chunk_printf(&trash, "%s%s", strerror(errno),
661 chk->area);
662 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100663 }
664 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200665 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100666 }
667 }
668
Willy Tarreau00149122017-10-04 18:05:01 +0200669 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200670 /* NOTE: this is reported after <fall> tries */
671 chunk_printf(chk, "No port available for the TCP connection");
672 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
673 }
674
Willy Tarreau00149122017-10-04 18:05:01 +0200675 if (!conn) {
676 /* connection allocation error before the connection was established */
677 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
678 }
679 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100680 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200681 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100682 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
683 else if (expired)
684 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200685
686 /*
687 * might be due to a server IP change.
688 * Let's trigger a DNS resolution if none are currently running.
689 */
Olivier Houchard0923fa42019-01-11 18:43:04 +0100690 if (check->server)
691 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200692
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100693 }
694 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
695 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200696 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100697 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
698 else if (expired)
699 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
700 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200701 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100702 /* I/O error after connection was established and before we could diagnose */
703 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
704 }
705 else if (expired) {
706 /* connection established but expired check */
707 if (check->type == PR_O2_SSL3_CHK)
708 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
709 else /* HTTP, SMTP, ... */
710 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
711 }
712
713 return;
714}
715
Olivier Houchard5c110b92018-08-14 17:04:58 +0200716/* This function checks if any I/O is wanted, and if so, attempts to do so */
717static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200718{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200719 struct check *check = ctx;
720 struct conn_stream *cs = check->cs;
Olivier Houchard0923fa42019-01-11 18:43:04 +0100721 struct email_alertq *q = container_of(check, typeof(*q), check);
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200722
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100723 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200724 wake_srv_chk(cs);
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100725 if (!(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100726 if (check->server)
727 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
728 else
729 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200730 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100731 if (check->server)
732 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
733 else
734 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200735 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200736 return NULL;
737}
738
739/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100740 *
741 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200742 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
743 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200744 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200745static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200746{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200747 struct connection *conn = cs->conn;
748 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900749 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900750 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200751
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100752 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100753 goto out_wakeup;
754
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200755 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100756 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200757 goto out;
758 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100759
Willy Tarreau20a18342013-12-05 00:31:46 +0100760 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200761 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100762 goto out_wakeup;
763 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100764
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200765 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100766 /* if the output is closed, we can't do anything */
767 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200768 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100769 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200770 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200771
Willy Tarreau06559ac2013-12-05 01:53:08 +0100772 /* here, we know that the connection is established. That's enough for
773 * a pure TCP check.
774 */
775 if (!check->type)
776 goto out_wakeup;
777
Willy Tarreauc09572f2017-10-04 11:58:22 +0200778 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100779 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200780 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200781
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200782 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200783 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200784 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200785 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200786 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100787 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200788 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200789 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100790 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200791 goto out;
792 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100793 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200794
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100795 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
796 if (s->proxy->timeout.check) {
797 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
798 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200799 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200800 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100801
Willy Tarreau83749182007-04-15 20:56:27 +0200802 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200803 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200804 out:
805 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200806}
807
Willy Tarreaubaaee002006-06-26 02:48:02 +0200808/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200809 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200810 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900811 * set_server_check_status() to update check->status, check->duration
812 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200813
814 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
815 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
816 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
817 * response to an SSL HELLO (the principle is that this is enough to
818 * distinguish between an SSL server and a pure TCP relay). All other cases will
819 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
820 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100821 *
822 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200823 * via the out label.
824 *
825 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200826 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200827static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200828{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200829 struct connection *conn = cs->conn;
830 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900831 struct server *s = check->server;
832 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200833 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100834 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200835 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200836
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100837 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200838 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200839
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200840 if (conn->flags & CO_FL_HANDSHAKE) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100841 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200842 goto out;
843 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200844
Willy Tarreauc09572f2017-10-04 11:58:22 +0200845 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100846 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200847 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200848
Willy Tarreau83749182007-04-15 20:56:27 +0200849 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
850 * but the connection was closed on the remote end. Fortunately, recv still
851 * works correctly and we don't need to do the getsockopt() on linux.
852 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000853
854 /* Set buffer to point to the end of the data already read, and check
855 * that there is free space remaining. If the buffer is full, proceed
856 * with running the checks without attempting another socket read.
857 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000858
Willy Tarreau03938182010-03-17 21:52:07 +0100859 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000860
Olivier Houchard511efea2018-08-16 15:30:32 +0200861 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200862 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100863 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200864 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200865 /* Report network errors only if we got no other data. Otherwise
866 * we'll let the upper layers decide whether the response is OK
867 * or not. It is very common that an RST sent by the server is
868 * reported as an error just after the last data chunk.
869 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200870 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100871 goto out_wakeup;
872 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200873 }
874
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100875
Willy Tarreau03938182010-03-17 21:52:07 +0100876 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200877 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100878 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 if (b_data(&check->bi) < b_size(&check->bi))
880 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100881 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200882 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100883 done = 1; /* buffer full, don't wait for more data */
884 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200885
Nick Chalk57b1bf72010-03-16 15:50:46 +0000886 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900887 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200888 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200889 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100890 goto wait_more_data;
891
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100892 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200893 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
894 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
895 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
896 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
897 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
898 cut_crlf(b_head(&check->bi));
899 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200900
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100901 goto out_wakeup;
902 }
903
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200904 check->code = str2uic(b_head(&check->bi) + 9);
905 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200906
Willy Tarreaubd741542010-03-16 18:46:54 +0100907 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200908 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000909 /* 404 may be accepted as "stopping" only if the server was up */
910 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900911 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000912 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100913 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
914 /* Run content verification check... We know we have at least 13 chars */
915 if (!httpchk_expect(s, done))
916 goto wait_more_data;
917 }
918 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200919 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100920 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900921 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100922 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000923 else {
924 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900925 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000926 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200927 break;
928
929 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200930 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100931 goto wait_more_data;
932
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100933 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200934 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900935 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200936 else
Simon Horman4a741432013-02-23 15:35:38 +0900937 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200938 break;
939
940 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200941 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100942 goto wait_more_data;
943
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200944 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200945 if ((b_data(&check->bi) < strlen("000\r")) ||
946 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
947 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
948 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
949 cut_crlf(b_head(&check->bi));
950 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200951 goto out_wakeup;
952 }
953
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200954 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200955
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200956 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200957 cut_crlf(desc);
958
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100959 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200960 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900961 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200962 else
Simon Horman4a741432013-02-23 15:35:38 +0900963 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200964 break;
965
Simon Hormana2b9dad2013-02-12 10:45:54 +0900966 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100967 int status = HCHK_STATUS_CHECKED;
968 const char *hs = NULL; /* health status */
969 const char *as = NULL; /* admin status */
970 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200971 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100972 const char *err = NULL; /* first error to report */
973 const char *wrn = NULL; /* first warning to report */
974 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900975
Willy Tarreau81f5d942013-12-09 20:51:51 +0100976 /* We're getting an agent check response. The agent could
977 * have been disabled in the mean time with a long check
978 * still pending. It is important that we ignore the whole
979 * response.
980 */
981 if (!(check->server->agent.state & CHK_ST_ENABLED))
982 break;
983
984 /* The agent supports strings made of a single line ended by the
985 * first CR ('\r') or LF ('\n'). This line is composed of words
986 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
987 * line may optionally contained a description of a state change
988 * after a sharp ('#'), which is only considered if a health state
989 * is announced.
990 *
991 * Words may be composed of :
992 * - a numeric weight suffixed by the percent character ('%').
993 * - a health status among "up", "down", "stopped", and "fail".
994 * - an admin status among "ready", "drain", "maint".
995 *
996 * These words may appear in any order. If multiple words of the
997 * same category appear, the last one wins.
998 */
999
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001000 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001001 while (*p && *p != '\n' && *p != '\r')
1002 p++;
1003
1004 if (!*p) {
1005 if (!done)
1006 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001007
Willy Tarreau9809b782013-12-11 21:40:11 +01001008 /* at least inform the admin that the agent is mis-behaving */
1009 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1010 break;
1011 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001012
Willy Tarreau9809b782013-12-11 21:40:11 +01001013 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001014 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001015
Willy Tarreau81f5d942013-12-09 20:51:51 +01001016 while (*cmd) {
1017 /* look for next word */
1018 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1019 cmd++;
1020 continue;
1021 }
Simon Horman671b6f02013-11-25 10:46:39 +09001022
Willy Tarreau81f5d942013-12-09 20:51:51 +01001023 if (*cmd == '#') {
1024 /* this is the beginning of a health status description,
1025 * skip the sharp and blanks.
1026 */
1027 cmd++;
1028 while (*cmd == '\t' || *cmd == ' ')
1029 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001030 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001031 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001032
1033 /* find the end of the word so that we have a null-terminated
1034 * word between <cmd> and <p>.
1035 */
1036 p = cmd + 1;
1037 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1038 p++;
1039 if (*p)
1040 *p++ = 0;
1041
1042 /* first, health statuses */
1043 if (strcasecmp(cmd, "up") == 0) {
1044 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001045 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001046 hs = cmd;
1047 }
1048 else if (strcasecmp(cmd, "down") == 0) {
1049 check->health = 0;
1050 status = HCHK_STATUS_L7STS;
1051 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001052 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001053 else if (strcasecmp(cmd, "stopped") == 0) {
1054 check->health = 0;
1055 status = HCHK_STATUS_L7STS;
1056 hs = cmd;
1057 }
1058 else if (strcasecmp(cmd, "fail") == 0) {
1059 check->health = 0;
1060 status = HCHK_STATUS_L7STS;
1061 hs = cmd;
1062 }
1063 /* admin statuses */
1064 else if (strcasecmp(cmd, "ready") == 0) {
1065 as = cmd;
1066 }
1067 else if (strcasecmp(cmd, "drain") == 0) {
1068 as = cmd;
1069 }
1070 else if (strcasecmp(cmd, "maint") == 0) {
1071 as = cmd;
1072 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001073 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001074 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1075 ps = cmd;
1076 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001077 /* try to parse a maxconn here */
1078 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1079 cs = cmd;
1080 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001081 else {
1082 /* keep a copy of the first error */
1083 if (!err)
1084 err = cmd;
1085 }
1086 /* skip to next word */
1087 cmd = p;
1088 }
1089 /* here, cmd points either to \0 or to the beginning of a
1090 * description. Skip possible leading spaces.
1091 */
1092 while (*cmd == ' ' || *cmd == '\n')
1093 cmd++;
1094
1095 /* First, update the admin status so that we avoid sending other
1096 * possibly useless warnings and can also update the health if
1097 * present after going back up.
1098 */
1099 if (as) {
1100 if (strcasecmp(as, "drain") == 0)
1101 srv_adm_set_drain(check->server);
1102 else if (strcasecmp(as, "maint") == 0)
1103 srv_adm_set_maint(check->server);
1104 else
1105 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001106 }
1107
Willy Tarreau81f5d942013-12-09 20:51:51 +01001108 /* now change weights */
1109 if (ps) {
1110 const char *msg;
1111
1112 msg = server_parse_weight_change_request(s, ps);
1113 if (!wrn || !*wrn)
1114 wrn = msg;
1115 }
1116
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001117 if (cs) {
1118 const char *msg;
1119
1120 cs += strlen("maxconn:");
1121
1122 msg = server_parse_maxconn_change_request(s, cs);
1123 if (!wrn || !*wrn)
1124 wrn = msg;
1125 }
1126
Willy Tarreau81f5d942013-12-09 20:51:51 +01001127 /* and finally health status */
1128 if (hs) {
1129 /* We'll report some of the warnings and errors we have
1130 * here. Down reports are critical, we leave them untouched.
1131 * Lack of report, or report of 'UP' leaves the room for
1132 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001133 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001134 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001135 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001136
1137 if (!*msg || status == HCHK_STATUS_L7OKD) {
1138 if (err && *err)
1139 msg = err;
1140 else if (wrn && *wrn)
1141 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001142 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001143
1144 t = get_trash_chunk();
1145 chunk_printf(t, "via agent : %s%s%s%s",
1146 hs, *msg ? " (" : "",
1147 msg, *msg ? ")" : "");
1148
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001149 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001150 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001151 else if (err && *err) {
1152 /* No status change but we'd like to report something odd.
1153 * Just report the current state and copy the message.
1154 */
1155 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001156 set_server_check_status(check, status/*check->status*/,
1157 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001158
Willy Tarreau81f5d942013-12-09 20:51:51 +01001159 }
1160 else if (wrn && *wrn) {
1161 /* No status change but we'd like to report something odd.
1162 * Just report the current state and copy the message.
1163 */
1164 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001165 set_server_check_status(check, status/*check->status*/,
1166 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001167 }
1168 else
1169 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001170 break;
1171 }
1172
Willy Tarreau1620ec32011-08-06 17:05:02 +02001173 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001174 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001175 goto wait_more_data;
1176
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001177 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001178 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001179 }
1180 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001181 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1182 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001183 else
1184 desc = "PostgreSQL unknown error";
1185
Simon Horman4a741432013-02-23 15:35:38 +09001186 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001187 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001188 break;
1189
1190 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001191 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001192 goto wait_more_data;
1193
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001194 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001195 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001196 }
1197 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001198 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001199 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001200 break;
1201
1202 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001203 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001204 goto wait_more_data;
1205
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001206 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001207 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001208 /* We set the MySQL Version in description for information purpose
1209 * FIXME : it can be cool to use MySQL Version for other purpose,
1210 * like mark as down old MySQL server.
1211 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001212 if (b_data(&check->bi) > 51) {
1213 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001214 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001215 }
1216 else {
1217 if (!done)
1218 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001219
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001220 /* it seems we have a OK packet but without a valid length,
1221 * it must be a protocol error
1222 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001223 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001224 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001225 }
1226 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001227 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001228 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001229 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001230 }
1231 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001232 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1233 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1234 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001235
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001236 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001237 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001238 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 /* We have only one MySQL packet and it is a Handshake Initialization packet
1240 * but we need to have a second packet to know if it is alright
1241 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001242 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001243 goto wait_more_data;
1244 }
1245 else {
1246 /* We have only one packet and it is an Error packet,
1247 * an error message is attached, so we can display it
1248 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001249 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001250 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001251 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001252 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001253 } else if (b_data(&check->bi) > first_packet_len + 4) {
1254 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1255 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1256 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001257
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001258 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001259 /* We have 2 packets and that's good */
1260 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001261 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001262 /* No error packet */
1263 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001264 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001265 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001266 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001267 }
1268 else {
1269 /* An error message is attached in the Error packet
1270 * so we can display it ! :)
1271 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001272 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001273 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001274 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001275 }
1276 }
1277 }
1278 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001279 if (!done)
1280 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001281
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001282 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001283 * it must be a protocol error
1284 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001285 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001286 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001287 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001288 }
1289 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001290 break;
1291
1292 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001293 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001294 goto wait_more_data;
1295
1296 /* Check if the server speaks LDAP (ASN.1/BER)
1297 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1298 * http://tools.ietf.org/html/rfc4511
1299 */
1300
1301 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1302 * LDAPMessage: 0x30: SEQUENCE
1303 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001304 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001305 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001306 }
1307 else {
1308 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001309 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001310
1311 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1312 * messageID: 0x02 0x01 0x01: INTEGER 1
1313 * protocolOp: 0x61: bindResponse
1314 */
1315 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001316 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001317 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001318 goto out_wakeup;
1319 }
1320
1321 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001322 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001323
1324 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1325 * ldapResult: 0x0a 0x01: ENUMERATION
1326 */
1327 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001328 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001329 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001330 goto out_wakeup;
1331 }
1332
1333 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1334 * resultCode
1335 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001336 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001337 if (check->code) {
1338 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 +02001339 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001340 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001341 }
1342 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001343 break;
1344
Christopher Fauletba7bc162016-11-07 21:07:38 +01001345 case PR_O2_SPOP_CHK: {
1346 unsigned int framesz;
1347 char err[HCHK_DESC_LEN];
1348
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001349 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001350 goto wait_more_data;
1351
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001352 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001353 framesz = ntohl(framesz);
1354
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001355 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001356 goto wait_more_data;
1357
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001358 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001359 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1360 else
1361 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1362 break;
1363 }
1364
Willy Tarreau1620ec32011-08-06 17:05:02 +02001365 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001366 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001367 break;
1368 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001369
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001370 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001371 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001372 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001373 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001374
Nick Chalk57b1bf72010-03-16 15:50:46 +00001375 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001376 *b_head(&check->bi) = '\0';
1377 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001378
Steven Davidovitz544d4812017-03-08 11:06:20 -08001379 /* Close the connection... We still attempt to nicely close if,
1380 * for instance, SSL needs to send a "close notify." Later, we perform
1381 * a hard close and reset the connection if some data are pending,
1382 * otherwise we end up with many TIME_WAITs and eat all the source port
1383 * range quickly. To avoid sending RSTs all the time, we first try to
1384 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001385 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001386 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001387
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001388 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001389 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001390 conn->flags |= CO_FL_ERROR;
1391
Willy Tarreaufdccded2008-08-29 18:19:04 +02001392 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001393out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001394 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001395
1396 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001397 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001398 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001399}
1400
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001401/*
1402 * This function is used only for server health-checks. It handles connection
1403 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001404 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1405 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001406 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001407static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001408{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001409 struct connection *conn = cs->conn;
1410 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001411 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001412 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001413
Olivier Houchard0923fa42019-01-11 18:43:04 +01001414 if (check->server)
1415 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1416 else
1417 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001418
Willy Tarreauc09572f2017-10-04 11:58:22 +02001419 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001420 if (check->type == PR_O2_TCPCHK_CHK) {
1421 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001422 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001423 conn = cs->conn;
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001424 } else if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001425 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001426
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001427 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001428 /* We may get error reports bypassing the I/O handlers, typically
1429 * the case when sending a pure TCP check which fails, then the I/O
1430 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001431 * main processing task so let's simply wake it up. If we get here,
1432 * we expect errno to still be valid.
1433 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001434 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001435 task_wakeup(check->task, TASK_WOKEN_IO);
1436 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001437 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001438 /* we may get here if only a connection probe was required : we
1439 * don't have any data to send nor anything expected in response,
1440 * so the completion of the connection establishment is enough.
1441 */
1442 task_wakeup(check->task, TASK_WOKEN_IO);
1443 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001444
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001445 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001446 /* Check complete or aborted. If connection not yet closed do it
1447 * now and wake the check task up to be sure the result is
1448 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001449 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001450 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001451 ret = -1;
Christopher Faulet774c4862019-01-21 14:15:50 +01001452 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001453 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001454
Olivier Houchard0923fa42019-01-11 18:43:04 +01001455 if (check->server)
1456 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1457 else
1458 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001459
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001460 /* if a connection got replaced, we must absolutely prevent the connection
1461 * handler from touching its fd, and perform the FD polling updates ourselves
1462 */
1463 if (ret < 0)
1464 conn_cond_update_polling(conn);
1465
1466 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001467}
1468
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001469struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001470 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001471 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001472};
1473
Willy Tarreaubaaee002006-06-26 02:48:02 +02001474/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001475 * updates the server's weight during a warmup stage. Once the final weight is
1476 * reached, the task automatically stops. Note that any server status change
1477 * must have updated s->last_change accordingly.
1478 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001479static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001480{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001481 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001482
1483 /* by default, plan on stopping the task */
1484 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001485 if ((s->next_admin & SRV_ADMF_MAINT) ||
1486 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001487 return t;
1488
Willy Tarreau892337c2014-05-13 23:41:20 +02001489 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001490 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001491
1492 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001493 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001494
1495 /* get back there in 1 second or 1/20th of the slowstart interval,
1496 * whichever is greater, resulting in small 5% steps.
1497 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001498 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001499 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1500 return t;
1501}
1502
Willy Tarreau894c6422017-10-04 15:58:52 +02001503/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1504 * none was found.
1505 */
1506static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1507{
1508 struct tcpcheck_rule *r;
1509
1510 list_for_each_entry(r, list, list) {
1511 if (r->action != TCPCHK_ACT_COMMENT)
1512 return r;
1513 }
1514 return NULL;
1515}
1516
Willy Tarreau2e993902011-10-31 11:53:20 +01001517/*
Simon Horman98637e52014-06-20 12:30:16 +09001518 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001519 *
1520 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001521 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1522 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1523 * - SF_ERR_SRVTO if there are no more servers
1524 * - SF_ERR_SRVCL if the connection was refused by the server
1525 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1526 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1527 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001528 * - 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 +01001529 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001530 * Note that we try to prevent the network stack from sending the ACK during the
1531 * connect() when a pure TCP check is used (without PROXY protocol).
1532 */
Simon Horman98637e52014-06-20 12:30:16 +09001533static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001534{
1535 struct check *check = t->context;
1536 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001537 struct conn_stream *cs = check->cs;
1538 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001539 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001540 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001541 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001542 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001543
Willy Tarreau00149122017-10-04 18:05:01 +02001544 /* we cannot have a connection here */
1545 if (conn)
1546 return SF_ERR_INTERNAL;
1547
Simon Hormanb00d17a2014-06-13 16:18:16 +09001548 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001549 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001550 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001551 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1552 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001553
1554 /* prepare the check buffer.
1555 * This should not be used if check is the secondary agent check
1556 * of a server as s->proxy->check_req will relate to the
1557 * configuration of the primary check. Similarly, tcp-check uses
1558 * its own strings.
1559 */
1560 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001561 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001562
1563 /* we want to check if this host replies to HTTP or SSLv3 requests
1564 * so we'll send the request, and won't wake the checker up now.
1565 */
1566 if ((check->type) == PR_O2_SSL3_CHK) {
1567 /* SSL requires that we put Unix time in the request */
1568 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001569 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001570 }
1571 else if ((check->type) == PR_O2_HTTP_CHK) {
1572 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001573 b_putblk(&check->bo, trash.area,
1574 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001575 /* prevent HTTP keep-alive when "http-check expect" is used */
1576 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001577 b_putist(&check->bo, ist("Connection: close\r\n"));
1578 b_putist(&check->bo, ist("\r\n"));
1579 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580 }
1581 }
1582
James Brown55f9ff12015-10-21 18:19:05 -07001583 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001584 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001585 }
1586
Willy Tarreauf411cce2017-10-04 16:21:19 +02001587 /* for tcp-checks, the initial connection setup is handled separately as
1588 * it may be sent to a specific port and not to the server's.
1589 */
1590 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1591 tcpcheck_main(check);
1592 return SF_ERR_UP;
1593 }
1594
Simon Hormanb00d17a2014-06-13 16:18:16 +09001595 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001596 cs = check->cs = cs_new(NULL);
1597 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001598 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001599 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001600 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001601 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001602
Simon Horman41f58762015-01-30 11:22:56 +09001603 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001604 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001605 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001606 }
1607 else {
1608 /* we'll connect to the addr on the server */
1609 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001610 }
1611
Olivier Houchard6377a002017-12-01 22:04:05 +01001612 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001613 conn->target = &s->obj_type;
1614
Willy Tarreaube373152018-09-06 11:45:30 +02001615 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001616 int i = 0;
1617
1618 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001619 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001620 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001621
1622 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001623 }
1624
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001625 /* no client address */
1626 clear_addr(&conn->addr.from);
1627
Willy Tarreaube373152018-09-06 11:45:30 +02001628 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001629 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1630 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001631 cs_attach(cs, check, &check_conn_cb);
1632
Willy Tarreauf3d34822014-12-08 12:11:28 +01001633 /* only plain tcp-check supports quick ACK */
1634 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1635
Willy Tarreauf411cce2017-10-04 16:21:19 +02001636 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1637 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001638
Willy Tarreaue7dff022015-04-03 01:14:29 +02001639 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001640 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001641 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001642
Willy Tarreau16257f62017-11-02 15:45:00 +01001643
Olivier Houchard9130a962017-10-17 17:33:43 +02001644#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001645 if (ret == SF_ERR_NONE) {
1646 if (s->check.sni)
1647 ssl_sock_set_servername(conn, s->check.sni);
1648 if (s->check.alpn_str)
1649 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1650 s->check.alpn_len);
1651 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001652#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001653 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001654 conn->send_proxy_ofs = 1;
1655 conn->flags |= CO_FL_SEND_PROXY;
1656 }
1657
1658 return ret;
1659}
1660
Simon Horman98637e52014-06-20 12:30:16 +09001661static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001662static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001663__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001664
1665void block_sigchld(void)
1666{
1667 sigset_t set;
1668 sigemptyset(&set);
1669 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001670 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001671}
1672
1673void unblock_sigchld(void)
1674{
1675 sigset_t set;
1676 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001677 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001678 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001679}
1680
Simon Horman98637e52014-06-20 12:30:16 +09001681static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1682{
1683 struct pid_list *elem;
1684 struct check *check = t->context;
1685
Willy Tarreaubafbe012017-11-24 17:34:44 +01001686 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001687 if (!elem)
1688 return NULL;
1689 elem->pid = pid;
1690 elem->t = t;
1691 elem->exited = 0;
1692 check->curpid = elem;
1693 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001694
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001695 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001696 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001697 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001698
Simon Horman98637e52014-06-20 12:30:16 +09001699 return elem;
1700}
1701
Simon Horman98637e52014-06-20 12:30:16 +09001702static void pid_list_del(struct pid_list *elem)
1703{
1704 struct check *check;
1705
1706 if (!elem)
1707 return;
1708
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001709 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001710 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001711 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001712
Simon Horman98637e52014-06-20 12:30:16 +09001713 if (!elem->exited)
1714 kill(elem->pid, SIGTERM);
1715
1716 check = elem->t->context;
1717 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001718 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001719}
1720
1721/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1722static void pid_list_expire(pid_t pid, int status)
1723{
1724 struct pid_list *elem;
1725
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001726 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001727 list_for_each_entry(elem, &pid_list, list) {
1728 if (elem->pid == pid) {
1729 elem->t->expire = now_ms;
1730 elem->status = status;
1731 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001732 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001733 break;
Simon Horman98637e52014-06-20 12:30:16 +09001734 }
1735 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001736 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001737}
1738
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001739static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001740{
1741 pid_t pid;
1742 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001743
Simon Horman98637e52014-06-20 12:30:16 +09001744 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1745 pid_list_expire(pid, status);
1746}
1747
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001748static int init_pid_list(void)
1749{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001750 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001751 /* Nothing to do */
1752 return 0;
1753
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001754 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001755 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1756 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001757 return 1;
1758 }
1759
Willy Tarreaubafbe012017-11-24 17:34:44 +01001760 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1761 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001762 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1763 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001764 return 1;
1765 }
1766
1767 return 0;
1768}
1769
Cyril Bontéac92a062014-12-27 22:28:38 +01001770/* helper macro to set an environment variable and jump to a specific label on failure. */
1771#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001772
1773/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001774 * helper function to allocate enough memory to store an environment variable.
1775 * It will also check that the environment variable is updatable, and silently
1776 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001777 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001778static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001779{
1780 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001781 char *envname;
1782 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001783
Cyril Bontéac92a062014-12-27 22:28:38 +01001784 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001785 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001786 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001787 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001788
1789 envname = extcheck_envs[idx].name;
1790 vmaxlen = extcheck_envs[idx].vmaxlen;
1791
1792 /* Check if the environment variable is already set, and silently reject
1793 * the update if this one is not updatable. */
1794 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1795 return 0;
1796
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001797 /* Instead of sending NOT_USED, sending an empty value is preferable */
1798 if (strcmp(value, "NOT_USED") == 0) {
1799 value = "";
1800 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001801
1802 len = strlen(envname) + 1;
1803 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1804 len += strlen(value);
1805 else
1806 len += vmaxlen;
1807
1808 if (!check->envp[idx])
1809 check->envp[idx] = malloc(len + 1);
1810
1811 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001812 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001813 return 1;
1814 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001815 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001816 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001817 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001818 return 1;
1819 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001820 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001821 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001822 return 1;
1823 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001824 return 0;
1825}
Simon Horman98637e52014-06-20 12:30:16 +09001826
1827static int prepare_external_check(struct check *check)
1828{
1829 struct server *s = check->server;
1830 struct proxy *px = s->proxy;
1831 struct listener *listener = NULL, *l;
1832 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001833 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001834 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001835
1836 list_for_each_entry(l, &px->conf.listeners, by_fe)
1837 /* Use the first INET, INET6 or UNIX listener */
1838 if (l->addr.ss_family == AF_INET ||
1839 l->addr.ss_family == AF_INET6 ||
1840 l->addr.ss_family == AF_UNIX) {
1841 listener = l;
1842 break;
1843 }
1844
Simon Horman98637e52014-06-20 12:30:16 +09001845 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001846 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1847 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001848 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001849 goto err;
1850 }
Simon Horman98637e52014-06-20 12:30:16 +09001851
Cyril Bontéac92a062014-12-27 22:28:38 +01001852 check->argv = calloc(6, sizeof(char *));
1853 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001854 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001855 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001856 }
Simon Horman98637e52014-06-20 12:30:16 +09001857
1858 check->argv[0] = px->check_command;
1859
Cyril Bonté777be862014-12-02 21:21:35 +01001860 if (!listener) {
1861 check->argv[1] = strdup("NOT_USED");
1862 check->argv[2] = strdup("NOT_USED");
1863 }
1864 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001865 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001866 addr_to_str(&listener->addr, buf, sizeof(buf));
1867 check->argv[1] = strdup(buf);
1868 port_to_str(&listener->addr, buf, sizeof(buf));
1869 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001870 }
1871 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001872 const struct sockaddr_un *un;
1873
1874 un = (struct sockaddr_un *)&listener->addr;
1875 check->argv[1] = strdup(un->sun_path);
1876 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001877 }
1878 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001879 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001880 goto err;
1881 }
1882
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001883 addr_to_str(&s->addr, buf, sizeof(buf));
1884 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001885
1886 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1887 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1888 else
1889 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001890 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001891
Cyril Bontéac92a062014-12-27 22:28:38 +01001892 for (i = 0; i < 5; i++) {
1893 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001894 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001895 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001896 }
1897 }
Simon Horman98637e52014-06-20 12:30:16 +09001898
Cyril Bontéac92a062014-12-27 22:28:38 +01001899 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001900 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001901 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1902 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1903 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1904 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001905 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001906 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1907 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1908 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1909 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1910 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1911 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1912
1913 /* Ensure that we don't leave any hole in check->envp */
1914 for (i = 0; i < EXTCHK_SIZE; i++)
1915 if (!check->envp[i])
1916 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001917
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001918 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001919err:
1920 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001921 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001922 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001923 free(check->envp);
1924 check->envp = NULL;
1925 }
1926
1927 if (check->argv) {
1928 for (i = 1; i < 5; i++)
1929 free(check->argv[i]);
1930 free(check->argv);
1931 check->argv = NULL;
1932 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001933 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001934}
1935
Simon Hormanb00d17a2014-06-13 16:18:16 +09001936/*
Simon Horman98637e52014-06-20 12:30:16 +09001937 * establish a server health-check that makes use of a process.
1938 *
1939 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001940 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001941 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001942 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001943 *
1944 * Blocks and then unblocks SIGCHLD
1945 */
1946static int connect_proc_chk(struct task *t)
1947{
Cyril Bontéac92a062014-12-27 22:28:38 +01001948 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001949 struct check *check = t->context;
1950 struct server *s = check->server;
1951 struct proxy *px = s->proxy;
1952 int status;
1953 pid_t pid;
1954
Willy Tarreaue7dff022015-04-03 01:14:29 +02001955 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001956
1957 block_sigchld();
1958
1959 pid = fork();
1960 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001961 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1962 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001963 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1964 goto out;
1965 }
1966 if (pid == 0) {
1967 /* Child */
1968 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001969 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001970 int fd;
1971
1972 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1973 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1974
Willy Tarreau2555ccf2019-02-21 22:22:06 +01001975 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02001976
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001977 /* restore the initial FD limits */
1978 limit.rlim_cur = rlim_fd_cur_at_boot;
1979 limit.rlim_max = rlim_fd_max_at_boot;
1980 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
1981 getrlimit(RLIMIT_NOFILE, &limit);
1982 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
1983 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
1984 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
1985 }
1986
Simon Horman98637e52014-06-20 12:30:16 +09001987 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001988 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001989 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001990 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1991 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001992 exit(-1);
1993 }
1994
1995 /* Parent */
1996 if (check->result == CHK_RES_UNKNOWN) {
1997 if (pid_list_add(pid, t) != NULL) {
1998 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1999
2000 if (px->timeout.check && px->timeout.connect) {
2001 int t_con = tick_add(now_ms, px->timeout.connect);
2002 t->expire = tick_first(t->expire, t_con);
2003 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002004 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002005 goto out;
2006 }
2007 else {
2008 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2009 }
2010 kill(pid, SIGTERM); /* process creation error */
2011 }
2012 else
2013 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2014
2015out:
2016 unblock_sigchld();
2017 return status;
2018}
2019
2020/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002021 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002022 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002023 *
2024 * Please do NOT place any return statement in this function and only leave
2025 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002026 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002027static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002028{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002029 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002030 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002031 int rv;
2032 int ret;
2033 int expired = tick_is_expired(t->expire, now_ms);
2034
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002035 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002036 if (!(check->state & CHK_ST_INPROGRESS)) {
2037 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002038 if (!expired) /* woke up too early */
2039 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002040
2041 /* we don't send any health-checks when the proxy is
2042 * stopped, the server should not be checked or the check
2043 * is disabled.
2044 */
2045 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2046 s->proxy->state == PR_STSTOPPED)
2047 goto reschedule;
2048
2049 /* we'll initiate a new check */
2050 set_server_check_status(check, HCHK_STATUS_START, NULL);
2051
2052 check->state |= CHK_ST_INPROGRESS;
2053
Simon Hormandbf70192015-01-30 11:22:53 +09002054 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002055 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002056 /* the process was forked, we allow up to min(inter,
2057 * timeout.connect) for it to report its status, but
2058 * only when timeout.check is set as it may be to short
2059 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002060 */
2061 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2062
2063 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2064 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2065 t->expire = tick_first(t->expire, t_con);
2066 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002067 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002068 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002069 }
2070
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002071 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002072
2073 check->state &= ~CHK_ST_INPROGRESS;
2074 check_notify_failure(check);
2075
2076 /* we allow up to min(inter, timeout.connect) for a connection
2077 * to establish but only when timeout.check is set
2078 * as it may be to short for a full check otherwise
2079 */
2080 while (tick_is_expired(t->expire, now_ms)) {
2081 int t_con;
2082
2083 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2084 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2085
2086 if (s->proxy->timeout.check)
2087 t->expire = tick_first(t->expire, t_con);
2088 }
2089 }
2090 else {
2091 /* there was a test running.
2092 * First, let's check whether there was an uncaught error,
2093 * which can happen on connect timeout or error.
2094 */
2095 if (check->result == CHK_RES_UNKNOWN) {
2096 /* good connection is enough for pure TCP check */
2097 struct pid_list *elem = check->curpid;
2098 int status = HCHK_STATUS_UNKNOWN;
2099
2100 if (elem->exited) {
2101 status = elem->status; /* Save in case the process exits between use below */
2102 if (!WIFEXITED(status))
2103 check->code = -1;
2104 else
2105 check->code = WEXITSTATUS(status);
2106 if (!WIFEXITED(status) || WEXITSTATUS(status))
2107 status = HCHK_STATUS_PROCERR;
2108 else
2109 status = HCHK_STATUS_PROCOK;
2110 } else if (expired) {
2111 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002112 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002113 kill(elem->pid, SIGTERM);
2114 }
2115 set_server_check_status(check, status, NULL);
2116 }
2117
2118 if (check->result == CHK_RES_FAILED) {
2119 /* a failure or timeout detected */
2120 check_notify_failure(check);
2121 }
2122 else if (check->result == CHK_RES_CONDPASS) {
2123 /* check is OK but asks for stopping mode */
2124 check_notify_stopping(check);
2125 }
2126 else if (check->result == CHK_RES_PASSED) {
2127 /* a success was detected */
2128 check_notify_success(check);
2129 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002130 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002131 check->state &= ~CHK_ST_INPROGRESS;
2132
2133 pid_list_del(check->curpid);
2134
2135 rv = 0;
2136 if (global.spread_checks > 0) {
2137 rv = srv_getinter(check) * global.spread_checks / 100;
2138 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2139 }
2140 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2141 }
2142
2143 reschedule:
2144 while (tick_is_expired(t->expire, now_ms))
2145 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002146
2147 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002148 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002149 return t;
2150}
2151
2152/*
2153 * manages a server health-check that uses a connection. Returns
2154 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002155 *
2156 * Please do NOT place any return statement in this function and only leave
2157 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002158 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002159static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002160{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002161 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002162 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002163 struct conn_stream *cs = check->cs;
2164 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002165 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002166 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002167 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002168
Olivier Houchard0923fa42019-01-11 18:43:04 +01002169 if (check->server)
2170 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002171 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002172 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002173 if (!expired) /* woke up too early */
2174 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002175
Simon Horman671b6f02013-11-25 10:46:39 +09002176 /* we don't send any health-checks when the proxy is
2177 * stopped, the server should not be checked or the check
2178 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002179 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002180 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002181 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002182 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002183
2184 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002185 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002186
Willy Tarreau2c115e52013-12-11 19:41:16 +01002187 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002188 b_reset(&check->bi);
2189 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002190
Simon Hormandbf70192015-01-30 11:22:53 +09002191 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002192 cs = check->cs;
2193 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002194
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002195 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002196 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002197 goto out_unlock;
2198
Willy Tarreaue7dff022015-04-03 01:14:29 +02002199 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002200 /* we allow up to min(inter, timeout.connect) for a connection
2201 * to establish but only when timeout.check is set
2202 * as it may be to short for a full check otherwise
2203 */
Simon Horman4a741432013-02-23 15:35:38 +09002204 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002205
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002206 if (proxy->timeout.check && proxy->timeout.connect) {
2207 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002208 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002209 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002210
Olivier Houchard53216e72018-10-10 15:46:36 +02002211 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002212 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002213
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002214 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002215 goto reschedule;
2216
Willy Tarreaue7dff022015-04-03 01:14:29 +02002217 case SF_ERR_SRVTO: /* ETIMEDOUT */
2218 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002219 if (conn)
2220 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002221 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002222 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002223 /* should share same code than cases below */
2224 case SF_ERR_CHK_PORT:
2225 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002226 case SF_ERR_PRXCOND:
2227 case SF_ERR_RESOURCE:
2228 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002229 if (conn)
2230 conn->flags |= CO_FL_ERROR;
2231 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002232 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002233 }
2234
Willy Tarreau5a78f362012-11-23 12:47:05 +01002235 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002236 if (cs) {
2237 cs_destroy(cs);
2238 cs = check->cs = NULL;
2239 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002240 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002241
Willy Tarreau2c115e52013-12-11 19:41:16 +01002242 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002243 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002244
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002245 /* we allow up to min(inter, timeout.connect) for a connection
2246 * to establish but only when timeout.check is set
2247 * as it may be to short for a full check otherwise
2248 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002249 while (tick_is_expired(t->expire, now_ms)) {
2250 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002251
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002252 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002253 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002254
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002255 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002256 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002257 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002258 }
2259 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002260 /* there was a test running.
2261 * First, let's check whether there was an uncaught error,
2262 * which can happen on connect timeout or error.
2263 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002264 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002265 /* good connection is enough for pure TCP check */
2266 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002267 if (check->use_ssl)
2268 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002269 else
Simon Horman4a741432013-02-23 15:35:38 +09002270 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002271 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002272 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002273 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002274 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002275 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002276 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002277 }
2278
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002279 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002280 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002281 /* The check was aborted and the connection was not yet closed.
2282 * This can happen upon timeout, or when an external event such
2283 * as a failed response coupled with "observe layer7" caused the
2284 * server state to be suddenly changed.
2285 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002286 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002287 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002288 }
2289
Willy Tarreauac59f362017-10-08 11:10:19 +02002290 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002291 cs_destroy(cs);
2292 cs = check->cs = NULL;
2293 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002294 }
2295
Olivier Houchard0923fa42019-01-11 18:43:04 +01002296 if (check->server) {
2297 if (check->result == CHK_RES_FAILED) {
2298 /* a failure or timeout detected */
2299 check_notify_failure(check);
2300 }
2301 else if (check->result == CHK_RES_CONDPASS) {
2302 /* check is OK but asks for stopping mode */
2303 check_notify_stopping(check);
2304 }
2305 else if (check->result == CHK_RES_PASSED) {
2306 /* a success was detected */
2307 check_notify_success(check);
2308 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002309 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002310 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002311 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002312
Olivier Houchard0923fa42019-01-11 18:43:04 +01002313 if (check->server) {
2314 rv = 0;
2315 if (global.spread_checks > 0) {
2316 rv = srv_getinter(check) * global.spread_checks / 100;
2317 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2318 }
2319 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002320 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002321 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002322
2323 reschedule:
2324 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002325 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002326 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002327 if (check->server)
2328 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002329 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002330}
2331
Simon Horman98637e52014-06-20 12:30:16 +09002332/*
2333 * manages a server health-check. Returns
2334 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2335 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002336static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002337{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002338 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002339
2340 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002341 return process_chk_proc(t, context, state);
2342 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002343
Simon Horman98637e52014-06-20 12:30:16 +09002344}
2345
Simon Horman5c942422013-11-25 10:46:32 +09002346static int start_check_task(struct check *check, int mininter,
2347 int nbcheck, int srvpos)
2348{
2349 struct task *t;
2350 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002351 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002352 ha_alert("Starting [%s:%s] check: out of memory.\n",
2353 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002354 return 0;
2355 }
2356
2357 check->task = t;
2358 t->process = process_chk;
2359 t->context = check;
2360
Willy Tarreau1746eec2014-04-25 10:46:47 +02002361 if (mininter < srv_getinter(check))
2362 mininter = srv_getinter(check);
2363
2364 if (global.max_spread_checks && mininter > global.max_spread_checks)
2365 mininter = global.max_spread_checks;
2366
Simon Horman5c942422013-11-25 10:46:32 +09002367 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002368 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002369 check->start = now;
2370 task_queue(t);
2371
2372 return 1;
2373}
2374
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002375/*
2376 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002377 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002378 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002379static int start_checks()
2380{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002381
2382 struct proxy *px;
2383 struct server *s;
2384 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002385 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002386
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002387 /* 1- count the checkers to run simultaneously.
2388 * We also determine the minimum interval among all of those which
2389 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2390 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002391 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002392 * too short an interval for all others.
2393 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002394 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002395 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002396 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002397 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002398 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002399 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002400 }
2401 /* We need a warmup task that will be called when the server
2402 * state switches from down to up.
2403 */
2404 s->warmup = t;
2405 t->process = server_warmup;
2406 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002407 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002408 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002409 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 +01002410 }
2411
Willy Tarreaud8514a22013-12-11 21:10:14 +01002412 if (s->check.state & CHK_ST_CONFIGURED) {
2413 nbcheck++;
2414 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2415 (!mininter || mininter > srv_getinter(&s->check)))
2416 mininter = srv_getinter(&s->check);
2417 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002418
Willy Tarreaud8514a22013-12-11 21:10:14 +01002419 if (s->agent.state & CHK_ST_CONFIGURED) {
2420 nbcheck++;
2421 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2422 (!mininter || mininter > srv_getinter(&s->agent)))
2423 mininter = srv_getinter(&s->agent);
2424 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002425 }
2426 }
2427
Simon Horman4a741432013-02-23 15:35:38 +09002428 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002429 return 0;
2430
2431 srand((unsigned)time(NULL));
2432
2433 /*
2434 * 2- start them as far as possible from each others. For this, we will
2435 * start them after their interval set to the min interval divided by
2436 * the number of servers, weighted by the server's position in the list.
2437 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002438 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002439 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2440 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002441 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002442 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002443 }
2444 }
2445
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002446 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002447 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002448 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002449 if (s->check.type == PR_O2_EXT_CHK) {
2450 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002451 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002452 }
Simon Hormand60d6912013-11-25 10:46:36 +09002453 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002454 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002455 srvpos++;
2456 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002457
Simon Hormand60d6912013-11-25 10:46:36 +09002458 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002459 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002460 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002461 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002462 }
2463 srvpos++;
2464 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002465 }
2466 }
2467 return 0;
2468}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002469
2470/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002471 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002472 * The buffer MUST be terminated by a null byte before calling this function.
2473 * Sets server status appropriately. The caller is responsible for ensuring
2474 * that the buffer contains at least 13 characters. If <done> is zero, we may
2475 * return 0 to indicate that data is required to decide of a match.
2476 */
2477static int httpchk_expect(struct server *s, int done)
2478{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002479 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002480 char status_code[] = "000";
2481 char *contentptr;
2482 int crlf;
2483 int ret;
2484
2485 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2486 case PR_O2_EXP_STS:
2487 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002488 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2489 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002490
2491 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2492 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2493 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002494 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002495
2496 /* we necessarily have the response, so there are no partial failures */
2497 if (s->proxy->options2 & PR_O2_EXP_INV)
2498 ret = !ret;
2499
Simon Horman4a741432013-02-23 15:35:38 +09002500 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002501 break;
2502
2503 case PR_O2_EXP_STR:
2504 case PR_O2_EXP_RSTR:
2505 /* very simple response parser: ignore CR and only count consecutive LFs,
2506 * stop with contentptr pointing to first char after the double CRLF or
2507 * to '\0' if crlf < 2.
2508 */
2509 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002510 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002511 if (crlf >= 2)
2512 break;
2513 if (*contentptr == '\r')
2514 continue;
2515 else if (*contentptr == '\n')
2516 crlf++;
2517 else
2518 crlf = 0;
2519 }
2520
2521 /* Check that response contains a body... */
2522 if (crlf < 2) {
2523 if (!done)
2524 return 0;
2525
Simon Horman4a741432013-02-23 15:35:38 +09002526 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002527 "HTTP content check could not find a response body");
2528 return 1;
2529 }
2530
2531 /* Check that response body is not empty... */
2532 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002533 if (!done)
2534 return 0;
2535
Simon Horman4a741432013-02-23 15:35:38 +09002536 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002537 "HTTP content check found empty response body");
2538 return 1;
2539 }
2540
2541 /* Check the response content against the supplied string
2542 * or regex... */
2543 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2544 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2545 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002546 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002547
2548 /* if we don't match, we may need to wait more */
2549 if (!ret && !done)
2550 return 0;
2551
2552 if (ret) {
2553 /* content matched */
2554 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002555 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002556 "HTTP check matched unwanted content");
2557 else
Simon Horman4a741432013-02-23 15:35:38 +09002558 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002559 "HTTP content check matched");
2560 }
2561 else {
2562 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002563 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002564 "HTTP check did not match unwanted content");
2565 else
Simon Horman4a741432013-02-23 15:35:38 +09002566 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002567 "HTTP content check did not match");
2568 }
2569 break;
2570 }
2571 return 1;
2572}
2573
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002574/*
2575 * return the id of a step in a send/expect session
2576 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002577static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002578{
2579 struct tcpcheck_rule *cur = NULL, *next = NULL;
2580 int i = 0;
2581
Willy Tarreau213c6782014-10-02 14:51:02 +02002582 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002583 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002584 return 0;
2585
Simon Hormane16c1b32015-01-30 11:22:57 +09002586 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002587
2588 /* no step => first step */
2589 if (cur == NULL)
2590 return 1;
2591
2592 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002593 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002594 if (next->list.p == &cur->list)
2595 break;
2596 ++i;
2597 }
2598
2599 return i;
2600}
2601
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002602/*
2603 * return the latest known comment before (including) the given stepid
2604 * returns NULL if no comment found
2605 */
2606static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2607{
2608 struct tcpcheck_rule *cur = NULL;
2609 char *ret = NULL;
2610 int i = 0;
2611
2612 /* not even started anything yet, return latest comment found before any action */
2613 if (!check->current_step) {
2614 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2615 if (cur->action == TCPCHK_ACT_COMMENT)
2616 ret = cur->comment;
2617 else
2618 goto return_comment;
2619 }
2620 }
2621
2622 i = 1;
2623 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2624 if (cur->comment)
2625 ret = cur->comment;
2626
2627 if (i >= stepid)
2628 goto return_comment;
2629
2630 ++i;
2631 }
2632
2633 return_comment:
2634 return ret;
2635}
2636
Willy Tarreaube74b882017-10-04 16:22:49 +02002637/* proceed with next steps for the TCP checks <check>. Note that this is called
2638 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002639 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2640 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002641 *
2642 * Please do NOT place any return statement in this function and only leave
2643 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002644 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002645static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002646{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002647 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002648 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002649 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002650 struct conn_stream *cs = check->cs;
2651 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002652 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002653 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002654 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002655 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002656 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002657
Willy Tarreauef953952014-10-02 14:30:14 +02002658 /* here, we know that the check is complete or that it failed */
2659 if (check->result != CHK_RES_UNKNOWN)
2660 goto out_end_tcpcheck;
2661
2662 /* We have 4 possibilities here :
2663 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002664 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002665 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2666 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002667 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002668 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002669 * be waiting for a connection attempt to complete. conn!=NULL.
2670 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002671 *
2672 * #2 and #3 are quite similar, we want both the connection and the
2673 * handshake to complete before going any further. Thus we must always
2674 * wait for a connection to complete unless we're before and existing
2675 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002676 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002677
2678 /* find first rule and skip comments */
2679 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2680 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2681 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2682
Willy Tarreau00149122017-10-04 18:05:01 +02002683 if ((check->current_step || &next->list == head) &&
2684 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002685 /* we allow up to min(inter, timeout.connect) for a connection
2686 * to establish but only when timeout.check is set
2687 * as it may be to short for a full check otherwise
2688 */
2689 while (tick_is_expired(t->expire, now_ms)) {
2690 int t_con;
2691
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002692 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002693 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2694
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002695 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002696 t->expire = tick_first(t->expire, t_con);
2697 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002698 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002699 }
2700
Willy Tarreauef953952014-10-02 14:30:14 +02002701 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002702 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002703 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002704 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002705 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002706
Willy Tarreau213c6782014-10-02 14:51:02 +02002707 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002708 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002709 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002710 b_reset(&check->bo);
2711 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002712 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002713 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002714 if (proxy->timeout.check)
2715 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002716 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002717
Willy Tarreauabca5b62013-12-06 14:19:25 +01002718 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002719 /* We have to try to flush the output buffer before reading, at
2720 * the end, or if we're about to send a string that does not fit
2721 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002722 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002723 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002724 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002725 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002726 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002727 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002728 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002729
Olivier Houcharded0f2072018-08-16 15:41:52 +02002730 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002731 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002732
2733 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002734 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002735 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002736 goto out_end_tcpcheck;
2737 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002738 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002739 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002740 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002741 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002742 goto out;
2743 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002744 }
2745
Willy Tarreau263013d2015-05-13 11:59:14 +02002746 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002747 break;
2748
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002749 /* have 'next' point to the next rule or NULL if we're on the
2750 * last one, connect() needs this.
2751 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002752 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002753
2754 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002755 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002756 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002757
2758 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002759 if (&next->list == head)
2760 next = NULL;
2761
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002762 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2763 struct protocol *proto;
2764 struct xprt_ops *xprt;
2765
Willy Tarreau00149122017-10-04 18:05:01 +02002766 /* For a connect action we'll create a new connection.
2767 * We may also have to kill a previous one. But we don't
2768 * want to leave *without* a connection if we came here
2769 * from the connection layer, hence with a connection.
2770 * Thus we'll proceed in the following order :
2771 * 1: close but not release previous connection
2772 * 2: try to get a new connection
2773 * 3: release and replace the old one on success
2774 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002775 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002776 cs_close(check->cs);
2777 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002778 }
2779
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002780 /* mark the step as started */
2781 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002782
2783 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002784 cs = cs_new(NULL);
2785 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002786 step = tcpcheck_get_step_id(check);
2787 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2788 comment = tcpcheck_get_step_comment(check, step);
2789 if (comment)
2790 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002791 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2792 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002793 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002794 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002795 }
2796
Willy Tarreauac59f362017-10-08 11:10:19 +02002797 if (check->cs)
2798 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002799
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002800 check->cs = cs;
2801 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002802 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002803 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002804 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002805
2806 /* no client address */
2807 clear_addr(&conn->addr.from);
2808
Simon Horman41f58762015-01-30 11:22:56 +09002809 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002810 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002811 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002812 }
2813 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002814 /* we'll connect to the addr on the server */
2815 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002816 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002817 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002818
2819 /* port */
2820 if (check->current_step->port)
2821 set_host_port(&conn->addr.to, check->current_step->port);
2822 else if (check->port)
2823 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002824 else if (s->svc_port)
2825 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002826
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002827 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002828 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002829 }
2830 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002831 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002832 }
Willy Tarreaube373152018-09-06 11:45:30 +02002833
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002834 conn_prepare(conn, proto, xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01002835 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0)
2836 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02002837 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002838
Willy Tarreaue7dff022015-04-03 01:14:29 +02002839 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002840 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002841 ret = proto->connect(conn,
2842 1 /* I/O polling is always needed */,
2843 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002844 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2845 conn->send_proxy_ofs = 1;
2846 conn->flags |= CO_FL_SEND_PROXY;
2847 }
2848
2849 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002850 * - SF_ERR_NONE if everything's OK
2851 * - SF_ERR_SRVTO if there are no more servers
2852 * - SF_ERR_SRVCL if the connection was refused by the server
2853 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2854 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2855 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002856 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002857 * Note that we try to prevent the network stack from sending the ACK during the
2858 * connect() when a pure TCP check is used (without PROXY protocol).
2859 */
2860 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002861 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002862 /* we allow up to min(inter, timeout.connect) for a connection
2863 * to establish but only when timeout.check is set
2864 * as it may be to short for a full check otherwise
2865 */
2866 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2867
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002868 if (proxy->timeout.check && proxy->timeout.connect) {
2869 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002870 t->expire = tick_first(t->expire, t_con);
2871 }
2872 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002873 case SF_ERR_SRVTO: /* ETIMEDOUT */
2874 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002875 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002876 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002877 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002878 comment = tcpcheck_get_step_comment(check, step);
2879 if (comment)
2880 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002881 set_server_check_status(check, HCHK_STATUS_L4CON,
2882 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002883 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002884 case SF_ERR_PRXCOND:
2885 case SF_ERR_RESOURCE:
2886 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002887 step = tcpcheck_get_step_id(check);
2888 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002889 comment = tcpcheck_get_step_comment(check, step);
2890 if (comment)
2891 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002892 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2893 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002894 goto out_end_tcpcheck;
2895 }
2896
2897 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002898 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002899
2900 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002901 while (&check->current_step->list != head &&
2902 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002903 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002904
Willy Tarreauf2c87352015-05-13 12:08:21 +02002905 if (&check->current_step->list == head)
2906 break;
2907
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002908 } /* end 'connect' */
2909 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2910 /* mark the step as started */
2911 check->last_started_step = check->current_step;
2912
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002913 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002914 if (*b_head(&check->bi) != '\0') {
2915 *b_head(&check->bi) = '\0';
2916 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002917 }
2918
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002919 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002920 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002921 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002922 goto out_end_tcpcheck;
2923 }
2924
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002925 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002926 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002927 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002928 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002929 set_server_check_status(check, HCHK_STATUS_L7RSP,
2930 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002931 goto out_end_tcpcheck;
2932 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002933
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002934 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002935 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002936 continue;
2937
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002938 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2939 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002940
Willy Tarreauabca5b62013-12-06 14:19:25 +01002941 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002942 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002943
2944 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002945 while (&check->current_step->list != head &&
2946 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002947 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002948
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002949 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002950 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002951 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002952 goto out_end_tcpcheck;
2953
Olivier Houchard511efea2018-08-16 15:30:32 +02002954 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002955 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002956 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002957 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002958 /* Report network errors only if we got no other data. Otherwise
2959 * we'll let the upper layers decide whether the response is OK
2960 * or not. It is very common that an RST sent by the server is
2961 * reported as an error just after the last data chunk.
2962 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002963 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002964 goto out_end_tcpcheck;
2965 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002966 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002967 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002968 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002969 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002970 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002971 }
2972
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002973 /* mark the step as started */
2974 check->last_started_step = check->current_step;
2975
2976
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002978 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002979 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002980 if (b_data(&check->bi) < b_size(&check->bi)) {
2981 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002982 }
2983 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002984 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002985 done = 1; /* buffer full, don't wait for more data */
2986 }
2987
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002988 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002989
2990 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002991 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002992 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002993 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002994
2995 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002996 step = tcpcheck_get_step_id(check);
2997 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
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
3004 goto out_end_tcpcheck;
3005 }
3006
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003007 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003008 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003009
Willy Tarreaua970c282013-12-06 12:47:19 +01003010 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003011 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003012 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 +02003013 else if (check->current_step->expect_regex != NULL)
3014 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003015
3016 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003017 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003018
3019 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003020 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003021 if (ret) {
3022 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003023 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003024 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003025 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003026 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003027 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003028 }
3029 else {
3030 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003031 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003032 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003033 comment = tcpcheck_get_step_comment(check, step);
3034 if (comment)
3035 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003036 set_server_check_status(check, HCHK_STATUS_L7RSP,
3037 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003038 goto out_end_tcpcheck;
3039 }
3040 /* matched and was supposed to => OK, next step */
3041 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003042 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003043 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003044
3045 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003046 while (&check->current_step->list != head &&
3047 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003048 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003049
Willy Tarreauf2c87352015-05-13 12:08:21 +02003050 if (&check->current_step->list == head)
3051 break;
3052
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003053 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003054 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003055 }
3056 }
3057 else {
3058 /* not matched */
3059 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003060 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003061 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003062 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003063
3064 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003065 while (&check->current_step->list != head &&
3066 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003067 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003068
Willy Tarreauf2c87352015-05-13 12:08:21 +02003069 if (&check->current_step->list == head)
3070 break;
3071
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003072 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003073 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003074 }
3075 /* not matched but was supposed to => ERROR */
3076 else {
3077 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003078 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003079 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003080 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003081 }
3082 else {
3083 /* we were looking for a regex */
3084 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003085 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003086 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003087 comment = tcpcheck_get_step_comment(check, step);
3088 if (comment)
3089 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003090 set_server_check_status(check, HCHK_STATUS_L7RSP,
3091 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003092 goto out_end_tcpcheck;
3093 }
3094 }
3095 } /* end expect */
3096 } /* end loop over double chained step list */
3097
Baptiste Assmann248f1172018-03-01 21:49:01 +01003098 /* don't do anything until the connection is established */
3099 if (!(conn->flags & CO_FL_CONNECTED)) {
3100 /* update expire time, should be done by process_chk */
3101 /* we allow up to min(inter, timeout.connect) for a connection
3102 * to establish but only when timeout.check is set
3103 * as it may be to short for a full check otherwise
3104 */
3105 while (tick_is_expired(t->expire, now_ms)) {
3106 int t_con;
3107
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003108 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003109 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3110
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003111 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003112 t->expire = tick_first(t->expire, t_con);
3113 }
3114 goto out;
3115 }
3116
Willy Tarreau263013d2015-05-13 11:59:14 +02003117 /* We're waiting for some I/O to complete, we've reached the end of the
3118 * rules, or both. Do what we have to do, otherwise we're done.
3119 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003120 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003121 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3122 goto out_end_tcpcheck;
3123 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003124
Willy Tarreau53c5a042015-05-13 11:38:17 +02003125 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003126 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003127 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003128 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003129
3130 out_end_tcpcheck:
3131 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003132 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003133 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003134
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003135 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003136 check->current_step = NULL;
3137
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003138 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003139 conn->flags |= CO_FL_ERROR;
3140
Christopher Fauletb6102852017-11-28 10:06:29 +01003141 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003142 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003143}
3144
Simon Hormanb1900d52015-01-30 11:22:54 +09003145const char *init_check(struct check *check, int type)
3146{
3147 check->type = type;
3148
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003149 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3150 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003151
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003152 check->bi.area = calloc(check->bi.size, sizeof(char));
3153 check->bo.area = calloc(check->bo.size, sizeof(char));
3154
3155 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003156 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003157
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003158 check->wait_list.task = tasklet_new();
3159 if (!check->wait_list.task)
3160 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003161 check->wait_list.events = 0;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003162 check->wait_list.task->process = event_srv_chk_io;
3163 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003164 return NULL;
3165}
3166
Simon Hormanbfb5d332015-01-30 11:22:55 +09003167void free_check(struct check *check)
3168{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003169 free(check->bi.area);
3170 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003171 if (check->cs) {
3172 free(check->cs->conn);
3173 check->cs->conn = NULL;
3174 cs_free(check->cs);
3175 check->cs = NULL;
3176 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003177}
3178
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003179void email_alert_free(struct email_alert *alert)
3180{
3181 struct tcpcheck_rule *rule, *back;
3182
3183 if (!alert)
3184 return;
3185
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003186 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3187 LIST_DEL(&rule->list);
3188 free(rule->comment);
3189 free(rule->string);
3190 if (rule->expect_regex)
3191 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003192 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003193 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003194 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003195}
3196
Olivier Houchard9f6af332018-05-25 14:04:04 +02003197static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003198{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003199 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003200 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003201 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003202
3203 q = container_of(check, typeof(*q), check);
3204
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003205 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003206 while (1) {
3207 if (!(check->state & CHK_ST_ENABLED)) {
3208 if (LIST_ISEMPTY(&q->email_alerts)) {
3209 /* All alerts processed, queue the task */
3210 t->expire = TICK_ETERNITY;
3211 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003212 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003213 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003214
3215 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003216 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003217 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003218 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003219 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003220 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003221 }
3222
Olivier Houchard9f6af332018-05-25 14:04:04 +02003223 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003224 if (check->state & CHK_ST_INPROGRESS)
3225 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003226
3227 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3228 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003229 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003230 check->server = NULL;
3231 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003232 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003233 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003234 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003235 return t;
3236}
3237
Christopher Faulet0108bb32017-10-20 21:34:32 +02003238/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3239 *
3240 * The function returns 1 in success case, otherwise, it returns 0 and err is
3241 * filled.
3242 */
3243int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003244{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003245 struct mailer *mailer;
3246 struct email_alertq *queues;
3247 const char *err_str;
3248 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003249
Christopher Faulet0108bb32017-10-20 21:34:32 +02003250 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3251 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003252 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003253 }
3254
Christopher Faulet0108bb32017-10-20 21:34:32 +02003255 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3256 struct email_alertq *q = &queues[i];
3257 struct check *check = &q->check;
3258 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003259
3260 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003261 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003262 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003263 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003264 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003265 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003266 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3267 memprintf(err, "%s", err_str);
3268 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003269 }
3270
3271 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003272 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003273 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003274
Emeric Brunc60def82017-09-27 14:59:38 +02003275 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003276 memprintf(err, "out of memory while allocating mailer alerts task");
3277 goto error;
3278 }
3279
3280 check->task = t;
3281 t->process = process_email_alert;
3282 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003283
Christopher Faulet0108bb32017-10-20 21:34:32 +02003284 /* check this in one ms */
3285 t->expire = TICK_ETERNITY;
3286 check->start = now;
3287 task_queue(t);
3288 }
3289
3290 mls->users++;
3291 free(p->email_alert.mailers.name);
3292 p->email_alert.mailers.m = mls;
3293 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003294 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003295
3296 error:
3297 for (i = 0; i < mls->count; i++) {
3298 struct email_alertq *q = &queues[i];
3299 struct check *check = &q->check;
3300
3301 if (check->task) {
3302 task_delete(check->task);
3303 task_free(check->task);
3304 check->task = NULL;
3305 }
3306 free_check(check);
3307 }
3308 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003309 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003310 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003311}
3312
3313
3314static int add_tcpcheck_expect_str(struct list *list, const char *str)
3315{
3316 struct tcpcheck_rule *tcpcheck;
3317
Willy Tarreaubafbe012017-11-24 17:34:44 +01003318 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003319 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003320 memset(tcpcheck, 0, sizeof(*tcpcheck));
3321 tcpcheck->action = TCPCHK_ACT_EXPECT;
3322 tcpcheck->string = strdup(str);
3323 tcpcheck->expect_regex = NULL;
3324 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003325 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003326 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327 return 0;
3328 }
3329
3330 LIST_ADDQ(list, &tcpcheck->list);
3331 return 1;
3332}
3333
3334static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3335{
3336 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003337 const char *in;
3338 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003339 int i;
3340
Willy Tarreaubafbe012017-11-24 17:34:44 +01003341 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003342 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003343 memset(tcpcheck, 0, sizeof(*tcpcheck));
3344 tcpcheck->action = TCPCHK_ACT_SEND;
3345 tcpcheck->expect_regex = NULL;
3346 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003347 tcpcheck->string_len = 0;
3348 for (i = 0; strs[i]; i++)
3349 tcpcheck->string_len += strlen(strs[i]);
3350
3351 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3352 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003353 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003354 return 0;
3355 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003356
Willy Tarreau64345aa2016-08-10 19:29:09 +02003357 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003358 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003359 for (in = strs[i]; (*dst = *in++); dst++);
3360 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003361
3362 LIST_ADDQ(list, &tcpcheck->list);
3363 return 1;
3364}
3365
Christopher Faulet0108bb32017-10-20 21:34:32 +02003366static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3367 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003368{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003369 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003370 struct tcpcheck_rule *tcpcheck;
3371 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003372
Willy Tarreaubafbe012017-11-24 17:34:44 +01003373 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003374 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003375 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003376 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003377 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003378
Willy Tarreaubafbe012017-11-24 17:34:44 +01003379 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003381 memset(tcpcheck, 0, sizeof(*tcpcheck));
3382 tcpcheck->action = TCPCHK_ACT_CONNECT;
3383 tcpcheck->comment = NULL;
3384 tcpcheck->string = NULL;
3385 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003386 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3387
3388 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3389 goto error;
3390
3391 {
3392 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3393 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3394 goto error;
3395 }
3396
3397 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3398 goto error;
3399
3400 {
3401 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3402 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3403 goto error;
3404 }
3405
3406 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3407 goto error;
3408
3409 {
3410 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3411 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3412 goto error;
3413 }
3414
3415 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3416 goto error;
3417
3418 {
3419 const char * const strs[2] = { "DATA\r\n" };
3420 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3421 goto error;
3422 }
3423
3424 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3425 goto error;
3426
3427 {
3428 struct tm tm;
3429 char datestr[48];
3430 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003431 "From: ", p->email_alert.from, "\r\n",
3432 "To: ", p->email_alert.to, "\r\n",
3433 "Date: ", datestr, "\r\n",
3434 "Subject: [HAproxy Alert] ", msg, "\r\n",
3435 "\r\n",
3436 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003437 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003438 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003439 NULL
3440 };
3441
3442 get_localtime(date.tv_sec, &tm);
3443
3444 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3445 goto error;
3446 }
3447
3448 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3449 goto error;
3450 }
3451
3452 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3453 goto error;
3454
3455 {
3456 const char * const strs[2] = { "QUIT\r\n" };
3457 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3458 goto error;
3459 }
3460
3461 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3462 goto error;
3463
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003464 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003465 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003466 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003467 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003468 return 1;
3469
3470error:
3471 email_alert_free(alert);
3472 return 0;
3473}
3474
Christopher Faulet0108bb32017-10-20 21:34:32 +02003475static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003476{
3477 int i;
3478 struct mailer *mailer;
3479
3480 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3481 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003482 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003483 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003484 return;
3485 }
3486 }
3487
3488 return;
3489}
3490
3491/*
3492 * Send email alert if configured.
3493 */
Simon Horman64e34162015-02-06 11:11:57 +09003494void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003495{
3496 va_list argp;
3497 char buf[1024];
3498 int len;
3499 struct proxy *p = s->proxy;
3500
Christopher Faulet0108bb32017-10-20 21:34:32 +02003501 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003502 return;
3503
3504 va_start(argp, format);
3505 len = vsnprintf(buf, sizeof(buf), format, argp);
3506 va_end(argp);
3507
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003508 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003509 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003510 return;
3511 }
3512
Christopher Faulet0108bb32017-10-20 21:34:32 +02003513 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003514}
3515
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003516/*
3517 * Return value:
3518 * the port to be used for the health check
3519 * 0 in case no port could be found for the check
3520 */
3521int srv_check_healthcheck_port(struct check *chk)
3522{
3523 int i = 0;
3524 struct server *srv = NULL;
3525
3526 srv = chk->server;
3527
3528 /* If neither a port nor an addr was specified and no check transport
3529 * layer is forced, then the transport layer used by the checks is the
3530 * same as for the production traffic. Otherwise we use raw_sock by
3531 * default, unless one is specified.
3532 */
3533 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003534 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003535 chk->send_proxy |= (srv->pp_opts);
3536 }
3537
3538 /* by default, we use the health check port ocnfigured */
3539 if (chk->port > 0)
3540 return chk->port;
3541
3542 /* try to get the port from check_core.addr if check.port not set */
3543 i = get_host_port(&chk->addr);
3544 if (i > 0)
3545 return i;
3546
3547 /* try to get the port from server address */
3548 /* prevent MAPPORTS from working at this point, since checks could
3549 * not be performed in such case (MAPPORTS impose a relative ports
3550 * based on live traffic)
3551 */
3552 if (srv->flags & SRV_F_MAPPORTS)
3553 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003554
3555 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003556 if (i > 0)
3557 return i;
3558
3559 return 0;
3560}
3561
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003562REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003563
Willy Tarreaubd741542010-03-16 18:46:54 +01003564/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003565 * Local variables:
3566 * c-indent-level: 8
3567 * c-basic-offset: 8
3568 * End:
3569 */