blob: e6b0bd1d063f37e84143a21c9c3c5cc9ff7208c3 [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)) {
Olivier Houchard7059c552019-03-08 18:49:32 +0100255 _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
Olivier Houchard7059c552019-03-08 18:49:32 +0100423 _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;
Olivier Houchard7059c552019-03-08 18:49:32 +0100464 _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) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100756 if (!(conn->flags & CO_FL_ERROR))
757 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200758 goto out;
759 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100760
Willy Tarreau20a18342013-12-05 00:31:46 +0100761 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200762 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100763 goto out_wakeup;
764 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100765
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200766 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100767 /* if the output is closed, we can't do anything */
768 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200769 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100770 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200771 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200772
Willy Tarreau06559ac2013-12-05 01:53:08 +0100773 /* here, we know that the connection is established. That's enough for
774 * a pure TCP check.
775 */
776 if (!check->type)
777 goto out_wakeup;
778
Willy Tarreauc09572f2017-10-04 11:58:22 +0200779 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100780 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200781 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200782
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200783 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200784 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200785 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200786 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200787 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100788 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200789 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200790 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100791 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200792 goto out;
793 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100794 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200795
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100796 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
797 if (s->proxy->timeout.check) {
798 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
799 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200800 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200801 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100802
Willy Tarreau83749182007-04-15 20:56:27 +0200803 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200804 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200805 out:
806 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200807}
808
Willy Tarreaubaaee002006-06-26 02:48:02 +0200809/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200810 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200811 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900812 * set_server_check_status() to update check->status, check->duration
813 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200814
815 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
816 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
817 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
818 * response to an SSL HELLO (the principle is that this is enough to
819 * distinguish between an SSL server and a pure TCP relay). All other cases will
820 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
821 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100822 *
823 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200824 * via the out label.
825 *
826 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200827 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200828static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200829{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200830 struct connection *conn = cs->conn;
831 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900832 struct server *s = check->server;
833 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200834 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100835 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200836 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200837
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100838 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200839 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200840
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200841 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100842 if (!(conn->flags & CO_FL_ERROR))
843 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200844 goto out;
845 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200846
Willy Tarreauc09572f2017-10-04 11:58:22 +0200847 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100848 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200849 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200850
Willy Tarreau83749182007-04-15 20:56:27 +0200851 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
852 * but the connection was closed on the remote end. Fortunately, recv still
853 * works correctly and we don't need to do the getsockopt() on linux.
854 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000855
856 /* Set buffer to point to the end of the data already read, and check
857 * that there is free space remaining. If the buffer is full, proceed
858 * with running the checks without attempting another socket read.
859 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000860
Willy Tarreau03938182010-03-17 21:52:07 +0100861 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000862
Olivier Houchard511efea2018-08-16 15:30:32 +0200863 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200864 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100865 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200866 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200867 /* Report network errors only if we got no other data. Otherwise
868 * we'll let the upper layers decide whether the response is OK
869 * or not. It is very common that an RST sent by the server is
870 * reported as an error just after the last data chunk.
871 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200872 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100873 goto out_wakeup;
874 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200875 }
876
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100877
Willy Tarreau03938182010-03-17 21:52:07 +0100878 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100880 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200881 if (b_data(&check->bi) < b_size(&check->bi))
882 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100883 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200884 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100885 done = 1; /* buffer full, don't wait for more data */
886 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200887
Nick Chalk57b1bf72010-03-16 15:50:46 +0000888 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900889 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200890 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200891 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100892 goto wait_more_data;
893
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100894 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200895 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
896 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
897 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
898 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
899 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
900 cut_crlf(b_head(&check->bi));
901 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200902
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100903 goto out_wakeup;
904 }
905
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200906 check->code = str2uic(b_head(&check->bi) + 9);
907 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200908
Willy Tarreaubd741542010-03-16 18:46:54 +0100909 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200910 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000911 /* 404 may be accepted as "stopping" only if the server was up */
912 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900913 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000914 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100915 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
916 /* Run content verification check... We know we have at least 13 chars */
917 if (!httpchk_expect(s, done))
918 goto wait_more_data;
919 }
920 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200921 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100922 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900923 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100924 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000925 else {
926 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000928 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200929 break;
930
931 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200932 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100933 goto wait_more_data;
934
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100935 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200936 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900937 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200938 else
Simon Horman4a741432013-02-23 15:35:38 +0900939 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200940 break;
941
942 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200943 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100944 goto wait_more_data;
945
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200946 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200947 if ((b_data(&check->bi) < strlen("000\r")) ||
948 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
949 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
950 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
951 cut_crlf(b_head(&check->bi));
952 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200953 goto out_wakeup;
954 }
955
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200956 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200957
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200958 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200959 cut_crlf(desc);
960
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100961 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200962 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900963 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200964 else
Simon Horman4a741432013-02-23 15:35:38 +0900965 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200966 break;
967
Simon Hormana2b9dad2013-02-12 10:45:54 +0900968 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100969 int status = HCHK_STATUS_CHECKED;
970 const char *hs = NULL; /* health status */
971 const char *as = NULL; /* admin status */
972 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200973 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100974 const char *err = NULL; /* first error to report */
975 const char *wrn = NULL; /* first warning to report */
976 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900977
Willy Tarreau81f5d942013-12-09 20:51:51 +0100978 /* We're getting an agent check response. The agent could
979 * have been disabled in the mean time with a long check
980 * still pending. It is important that we ignore the whole
981 * response.
982 */
983 if (!(check->server->agent.state & CHK_ST_ENABLED))
984 break;
985
986 /* The agent supports strings made of a single line ended by the
987 * first CR ('\r') or LF ('\n'). This line is composed of words
988 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
989 * line may optionally contained a description of a state change
990 * after a sharp ('#'), which is only considered if a health state
991 * is announced.
992 *
993 * Words may be composed of :
994 * - a numeric weight suffixed by the percent character ('%').
995 * - a health status among "up", "down", "stopped", and "fail".
996 * - an admin status among "ready", "drain", "maint".
997 *
998 * These words may appear in any order. If multiple words of the
999 * same category appear, the last one wins.
1000 */
1001
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001002 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001003 while (*p && *p != '\n' && *p != '\r')
1004 p++;
1005
1006 if (!*p) {
1007 if (!done)
1008 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001009
Willy Tarreau9809b782013-12-11 21:40:11 +01001010 /* at least inform the admin that the agent is mis-behaving */
1011 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1012 break;
1013 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001014
Willy Tarreau9809b782013-12-11 21:40:11 +01001015 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001016 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001017
Willy Tarreau81f5d942013-12-09 20:51:51 +01001018 while (*cmd) {
1019 /* look for next word */
1020 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1021 cmd++;
1022 continue;
1023 }
Simon Horman671b6f02013-11-25 10:46:39 +09001024
Willy Tarreau81f5d942013-12-09 20:51:51 +01001025 if (*cmd == '#') {
1026 /* this is the beginning of a health status description,
1027 * skip the sharp and blanks.
1028 */
1029 cmd++;
1030 while (*cmd == '\t' || *cmd == ' ')
1031 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001032 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001033 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001034
1035 /* find the end of the word so that we have a null-terminated
1036 * word between <cmd> and <p>.
1037 */
1038 p = cmd + 1;
1039 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1040 p++;
1041 if (*p)
1042 *p++ = 0;
1043
1044 /* first, health statuses */
1045 if (strcasecmp(cmd, "up") == 0) {
1046 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001047 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001048 hs = cmd;
1049 }
1050 else if (strcasecmp(cmd, "down") == 0) {
1051 check->health = 0;
1052 status = HCHK_STATUS_L7STS;
1053 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001054 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001055 else if (strcasecmp(cmd, "stopped") == 0) {
1056 check->health = 0;
1057 status = HCHK_STATUS_L7STS;
1058 hs = cmd;
1059 }
1060 else if (strcasecmp(cmd, "fail") == 0) {
1061 check->health = 0;
1062 status = HCHK_STATUS_L7STS;
1063 hs = cmd;
1064 }
1065 /* admin statuses */
1066 else if (strcasecmp(cmd, "ready") == 0) {
1067 as = cmd;
1068 }
1069 else if (strcasecmp(cmd, "drain") == 0) {
1070 as = cmd;
1071 }
1072 else if (strcasecmp(cmd, "maint") == 0) {
1073 as = cmd;
1074 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001075 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001076 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1077 ps = cmd;
1078 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001079 /* try to parse a maxconn here */
1080 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1081 cs = cmd;
1082 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001083 else {
1084 /* keep a copy of the first error */
1085 if (!err)
1086 err = cmd;
1087 }
1088 /* skip to next word */
1089 cmd = p;
1090 }
1091 /* here, cmd points either to \0 or to the beginning of a
1092 * description. Skip possible leading spaces.
1093 */
1094 while (*cmd == ' ' || *cmd == '\n')
1095 cmd++;
1096
1097 /* First, update the admin status so that we avoid sending other
1098 * possibly useless warnings and can also update the health if
1099 * present after going back up.
1100 */
1101 if (as) {
1102 if (strcasecmp(as, "drain") == 0)
1103 srv_adm_set_drain(check->server);
1104 else if (strcasecmp(as, "maint") == 0)
1105 srv_adm_set_maint(check->server);
1106 else
1107 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001108 }
1109
Willy Tarreau81f5d942013-12-09 20:51:51 +01001110 /* now change weights */
1111 if (ps) {
1112 const char *msg;
1113
1114 msg = server_parse_weight_change_request(s, ps);
1115 if (!wrn || !*wrn)
1116 wrn = msg;
1117 }
1118
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001119 if (cs) {
1120 const char *msg;
1121
1122 cs += strlen("maxconn:");
1123
1124 msg = server_parse_maxconn_change_request(s, cs);
1125 if (!wrn || !*wrn)
1126 wrn = msg;
1127 }
1128
Willy Tarreau81f5d942013-12-09 20:51:51 +01001129 /* and finally health status */
1130 if (hs) {
1131 /* We'll report some of the warnings and errors we have
1132 * here. Down reports are critical, we leave them untouched.
1133 * Lack of report, or report of 'UP' leaves the room for
1134 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001135 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001136 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001137 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001138
1139 if (!*msg || status == HCHK_STATUS_L7OKD) {
1140 if (err && *err)
1141 msg = err;
1142 else if (wrn && *wrn)
1143 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001144 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001145
1146 t = get_trash_chunk();
1147 chunk_printf(t, "via agent : %s%s%s%s",
1148 hs, *msg ? " (" : "",
1149 msg, *msg ? ")" : "");
1150
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001151 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001152 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001153 else if (err && *err) {
1154 /* No status change but we'd like to report something odd.
1155 * Just report the current state and copy the message.
1156 */
1157 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001158 set_server_check_status(check, status/*check->status*/,
1159 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001160
Willy Tarreau81f5d942013-12-09 20:51:51 +01001161 }
1162 else if (wrn && *wrn) {
1163 /* No status change but we'd like to report something odd.
1164 * Just report the current state and copy the message.
1165 */
1166 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001167 set_server_check_status(check, status/*check->status*/,
1168 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001169 }
1170 else
1171 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001172 break;
1173 }
1174
Willy Tarreau1620ec32011-08-06 17:05:02 +02001175 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001176 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001177 goto wait_more_data;
1178
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001179 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001180 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001181 }
1182 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001183 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1184 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001185 else
1186 desc = "PostgreSQL unknown error";
1187
Simon Horman4a741432013-02-23 15:35:38 +09001188 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001189 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001190 break;
1191
1192 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001193 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001194 goto wait_more_data;
1195
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001196 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001197 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001198 }
1199 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001200 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001201 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001202 break;
1203
1204 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001206 goto wait_more_data;
1207
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001208 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001209 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001210 /* We set the MySQL Version in description for information purpose
1211 * FIXME : it can be cool to use MySQL Version for other purpose,
1212 * like mark as down old MySQL server.
1213 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001214 if (b_data(&check->bi) > 51) {
1215 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001216 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001217 }
1218 else {
1219 if (!done)
1220 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001221
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001222 /* it seems we have a OK packet but without a valid length,
1223 * it must be a protocol error
1224 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001225 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001226 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001227 }
1228 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001230 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001231 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001232 }
1233 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001234 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1235 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1236 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001237
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001238 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001240 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001241 /* We have only one MySQL packet and it is a Handshake Initialization packet
1242 * but we need to have a second packet to know if it is alright
1243 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001244 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 goto wait_more_data;
1246 }
1247 else {
1248 /* We have only one packet and it is an Error packet,
1249 * an error message is attached, so we can display it
1250 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001251 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001252 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001253 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001254 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001255 } else if (b_data(&check->bi) > first_packet_len + 4) {
1256 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1257 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1258 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001259
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001260 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001261 /* We have 2 packets and that's good */
1262 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001263 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001264 /* No error packet */
1265 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001266 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001267 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001268 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001269 }
1270 else {
1271 /* An error message is attached in the Error packet
1272 * so we can display it ! :)
1273 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001274 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001275 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001276 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001277 }
1278 }
1279 }
1280 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001281 if (!done)
1282 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001283
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001284 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001285 * it must be a protocol error
1286 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001287 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001288 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001289 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001290 }
1291 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001292 break;
1293
1294 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001295 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001296 goto wait_more_data;
1297
1298 /* Check if the server speaks LDAP (ASN.1/BER)
1299 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1300 * http://tools.ietf.org/html/rfc4511
1301 */
1302
1303 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1304 * LDAPMessage: 0x30: SEQUENCE
1305 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001306 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001307 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001308 }
1309 else {
1310 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001311 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001312
1313 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1314 * messageID: 0x02 0x01 0x01: INTEGER 1
1315 * protocolOp: 0x61: bindResponse
1316 */
1317 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001318 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001319 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001320 goto out_wakeup;
1321 }
1322
1323 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001324 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001325
1326 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1327 * ldapResult: 0x0a 0x01: ENUMERATION
1328 */
1329 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001330 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001331 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001332 goto out_wakeup;
1333 }
1334
1335 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1336 * resultCode
1337 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001338 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001339 if (check->code) {
1340 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 +02001341 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001342 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001343 }
1344 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001345 break;
1346
Christopher Fauletba7bc162016-11-07 21:07:38 +01001347 case PR_O2_SPOP_CHK: {
1348 unsigned int framesz;
1349 char err[HCHK_DESC_LEN];
1350
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001351 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001352 goto wait_more_data;
1353
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001354 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001355 framesz = ntohl(framesz);
1356
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001357 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001358 goto wait_more_data;
1359
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001360 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001361 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1362 else
1363 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1364 break;
1365 }
1366
Willy Tarreau1620ec32011-08-06 17:05:02 +02001367 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001368 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001369 break;
1370 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001371
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001372 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001373 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001374 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001375 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001376
Nick Chalk57b1bf72010-03-16 15:50:46 +00001377 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001378 *b_head(&check->bi) = '\0';
1379 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001380
Steven Davidovitz544d4812017-03-08 11:06:20 -08001381 /* Close the connection... We still attempt to nicely close if,
1382 * for instance, SSL needs to send a "close notify." Later, we perform
1383 * a hard close and reset the connection if some data are pending,
1384 * otherwise we end up with many TIME_WAITs and eat all the source port
1385 * range quickly. To avoid sending RSTs all the time, we first try to
1386 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001387 */
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001388 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001389
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001390 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001391 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001392 conn->flags |= CO_FL_ERROR;
1393
Willy Tarreaufdccded2008-08-29 18:19:04 +02001394 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001395out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001396 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001397
1398 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001399 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001400 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001401}
1402
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001403/*
1404 * This function is used only for server health-checks. It handles connection
1405 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001406 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1407 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001408 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001409static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001410{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001411 struct connection *conn = cs->conn;
1412 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001413 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001414 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001415
Olivier Houchard0923fa42019-01-11 18:43:04 +01001416 if (check->server)
1417 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1418 else
1419 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001420
Willy Tarreauc09572f2017-10-04 11:58:22 +02001421 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001422 if (check->type == PR_O2_TCPCHK_CHK) {
1423 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001424 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001425 conn = cs->conn;
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001426 } else if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001427 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001428
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001429 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001430 /* We may get error reports bypassing the I/O handlers, typically
1431 * the case when sending a pure TCP check which fails, then the I/O
1432 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001433 * main processing task so let's simply wake it up. If we get here,
1434 * we expect errno to still be valid.
1435 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001436 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001437 task_wakeup(check->task, TASK_WOKEN_IO);
1438 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001439 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001440 /* we may get here if only a connection probe was required : we
1441 * don't have any data to send nor anything expected in response,
1442 * so the completion of the connection establishment is enough.
1443 */
1444 task_wakeup(check->task, TASK_WOKEN_IO);
1445 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001446
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001447 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001448 /* Check complete or aborted. If connection not yet closed do it
1449 * now and wake the check task up to be sure the result is
1450 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001451 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001452 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001453 ret = -1;
Christopher Faulet774c4862019-01-21 14:15:50 +01001454 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001455 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001456
Olivier Houchard0923fa42019-01-11 18:43:04 +01001457 if (check->server)
1458 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1459 else
1460 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001461
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001462 /* if a connection got replaced, we must absolutely prevent the connection
1463 * handler from touching its fd, and perform the FD polling updates ourselves
1464 */
1465 if (ret < 0)
1466 conn_cond_update_polling(conn);
1467
1468 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001469}
1470
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001471struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001472 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001473 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001474};
1475
Willy Tarreaubaaee002006-06-26 02:48:02 +02001476/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001477 * updates the server's weight during a warmup stage. Once the final weight is
1478 * reached, the task automatically stops. Note that any server status change
1479 * must have updated s->last_change accordingly.
1480 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001481static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001482{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001483 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001484
1485 /* by default, plan on stopping the task */
1486 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001487 if ((s->next_admin & SRV_ADMF_MAINT) ||
1488 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001489 return t;
1490
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001491 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1492
Willy Tarreau892337c2014-05-13 23:41:20 +02001493 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001494 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001495
1496 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001497 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001498
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001499 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1500
Willy Tarreau2e993902011-10-31 11:53:20 +01001501 /* get back there in 1 second or 1/20th of the slowstart interval,
1502 * whichever is greater, resulting in small 5% steps.
1503 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001504 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001505 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1506 return t;
1507}
1508
Willy Tarreau894c6422017-10-04 15:58:52 +02001509/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1510 * none was found.
1511 */
1512static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1513{
1514 struct tcpcheck_rule *r;
1515
1516 list_for_each_entry(r, list, list) {
1517 if (r->action != TCPCHK_ACT_COMMENT)
1518 return r;
1519 }
1520 return NULL;
1521}
1522
Willy Tarreau2e993902011-10-31 11:53:20 +01001523/*
Simon Horman98637e52014-06-20 12:30:16 +09001524 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001525 *
1526 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001527 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1528 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1529 * - SF_ERR_SRVTO if there are no more servers
1530 * - SF_ERR_SRVCL if the connection was refused by the server
1531 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1532 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1533 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001534 * - 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 +01001535 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001536 * Note that we try to prevent the network stack from sending the ACK during the
1537 * connect() when a pure TCP check is used (without PROXY protocol).
1538 */
Simon Horman98637e52014-06-20 12:30:16 +09001539static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001540{
1541 struct check *check = t->context;
1542 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001543 struct conn_stream *cs = check->cs;
1544 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001545 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001546 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001547 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001548 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001549
Willy Tarreau00149122017-10-04 18:05:01 +02001550 /* we cannot have a connection here */
1551 if (conn)
1552 return SF_ERR_INTERNAL;
1553
Simon Hormanb00d17a2014-06-13 16:18:16 +09001554 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001555 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001556 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001557 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1558 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001559
1560 /* prepare the check buffer.
1561 * This should not be used if check is the secondary agent check
1562 * of a server as s->proxy->check_req will relate to the
1563 * configuration of the primary check. Similarly, tcp-check uses
1564 * its own strings.
1565 */
1566 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001567 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001568
1569 /* we want to check if this host replies to HTTP or SSLv3 requests
1570 * so we'll send the request, and won't wake the checker up now.
1571 */
1572 if ((check->type) == PR_O2_SSL3_CHK) {
1573 /* SSL requires that we put Unix time in the request */
1574 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001575 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001576 }
1577 else if ((check->type) == PR_O2_HTTP_CHK) {
1578 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001579 b_putblk(&check->bo, trash.area,
1580 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001581 /* prevent HTTP keep-alive when "http-check expect" is used */
1582 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001583 b_putist(&check->bo, ist("Connection: close\r\n"));
1584 b_putist(&check->bo, ist("\r\n"));
1585 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001586 }
1587 }
1588
James Brown55f9ff12015-10-21 18:19:05 -07001589 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001590 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001591 }
1592
Willy Tarreauf411cce2017-10-04 16:21:19 +02001593 /* for tcp-checks, the initial connection setup is handled separately as
1594 * it may be sent to a specific port and not to the server's.
1595 */
1596 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1597 tcpcheck_main(check);
1598 return SF_ERR_UP;
1599 }
1600
Simon Hormanb00d17a2014-06-13 16:18:16 +09001601 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001602 cs = check->cs = cs_new(NULL);
1603 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001604 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001605 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001606 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001607 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001608
Simon Horman41f58762015-01-30 11:22:56 +09001609 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001610 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001611 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001612 }
1613 else {
1614 /* we'll connect to the addr on the server */
1615 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001616 }
1617
Olivier Houchard6377a002017-12-01 22:04:05 +01001618 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001619 conn->target = &s->obj_type;
1620
Willy Tarreaube373152018-09-06 11:45:30 +02001621 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001622 int i = 0;
1623
1624 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001625 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001626 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001627
1628 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001629 }
1630
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001631 /* no client address */
1632 clear_addr(&conn->addr.from);
1633
Willy Tarreaube373152018-09-06 11:45:30 +02001634 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001635 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1636 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001637 cs_attach(cs, check, &check_conn_cb);
1638
Willy Tarreauf3d34822014-12-08 12:11:28 +01001639 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001640 if (check->type != 0)
1641 connflags |= CONNECT_HAS_DATA;
1642 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1643 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1644 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001645
Willy Tarreaue7dff022015-04-03 01:14:29 +02001646 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001647 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001648 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001649
Willy Tarreau16257f62017-11-02 15:45:00 +01001650
Olivier Houchard9130a962017-10-17 17:33:43 +02001651#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001652 if (ret == SF_ERR_NONE) {
1653 if (s->check.sni)
1654 ssl_sock_set_servername(conn, s->check.sni);
1655 if (s->check.alpn_str)
1656 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1657 s->check.alpn_len);
1658 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001659#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001660 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001661 conn->send_proxy_ofs = 1;
1662 conn->flags |= CO_FL_SEND_PROXY;
1663 }
1664
1665 return ret;
1666}
1667
Simon Horman98637e52014-06-20 12:30:16 +09001668static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001669static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001670__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001671
1672void block_sigchld(void)
1673{
1674 sigset_t set;
1675 sigemptyset(&set);
1676 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001677 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001678}
1679
1680void unblock_sigchld(void)
1681{
1682 sigset_t set;
1683 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001684 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001685 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001686}
1687
Simon Horman98637e52014-06-20 12:30:16 +09001688static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1689{
1690 struct pid_list *elem;
1691 struct check *check = t->context;
1692
Willy Tarreaubafbe012017-11-24 17:34:44 +01001693 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001694 if (!elem)
1695 return NULL;
1696 elem->pid = pid;
1697 elem->t = t;
1698 elem->exited = 0;
1699 check->curpid = elem;
1700 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001701
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001702 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001703 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001704 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001705
Simon Horman98637e52014-06-20 12:30:16 +09001706 return elem;
1707}
1708
Simon Horman98637e52014-06-20 12:30:16 +09001709static void pid_list_del(struct pid_list *elem)
1710{
1711 struct check *check;
1712
1713 if (!elem)
1714 return;
1715
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001716 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001717 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001718 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001719
Simon Horman98637e52014-06-20 12:30:16 +09001720 if (!elem->exited)
1721 kill(elem->pid, SIGTERM);
1722
1723 check = elem->t->context;
1724 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001725 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001726}
1727
1728/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1729static void pid_list_expire(pid_t pid, int status)
1730{
1731 struct pid_list *elem;
1732
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001733 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001734 list_for_each_entry(elem, &pid_list, list) {
1735 if (elem->pid == pid) {
1736 elem->t->expire = now_ms;
1737 elem->status = status;
1738 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001739 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001740 break;
Simon Horman98637e52014-06-20 12:30:16 +09001741 }
1742 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001743 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001744}
1745
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001746static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001747{
1748 pid_t pid;
1749 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001750
Simon Horman98637e52014-06-20 12:30:16 +09001751 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1752 pid_list_expire(pid, status);
1753}
1754
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001755static int init_pid_list(void)
1756{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001757 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001758 /* Nothing to do */
1759 return 0;
1760
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001761 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001762 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1763 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001764 return 1;
1765 }
1766
Willy Tarreaubafbe012017-11-24 17:34:44 +01001767 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1768 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001769 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1770 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001771 return 1;
1772 }
1773
1774 return 0;
1775}
1776
Cyril Bontéac92a062014-12-27 22:28:38 +01001777/* helper macro to set an environment variable and jump to a specific label on failure. */
1778#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001779
1780/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001781 * helper function to allocate enough memory to store an environment variable.
1782 * It will also check that the environment variable is updatable, and silently
1783 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001784 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001785static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001786{
1787 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001788 char *envname;
1789 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001790
Cyril Bontéac92a062014-12-27 22:28:38 +01001791 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001792 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001793 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001794 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001795
1796 envname = extcheck_envs[idx].name;
1797 vmaxlen = extcheck_envs[idx].vmaxlen;
1798
1799 /* Check if the environment variable is already set, and silently reject
1800 * the update if this one is not updatable. */
1801 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1802 return 0;
1803
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001804 /* Instead of sending NOT_USED, sending an empty value is preferable */
1805 if (strcmp(value, "NOT_USED") == 0) {
1806 value = "";
1807 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001808
1809 len = strlen(envname) + 1;
1810 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1811 len += strlen(value);
1812 else
1813 len += vmaxlen;
1814
1815 if (!check->envp[idx])
1816 check->envp[idx] = malloc(len + 1);
1817
1818 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001819 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001820 return 1;
1821 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001822 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001823 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001824 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001825 return 1;
1826 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001827 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001828 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001829 return 1;
1830 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001831 return 0;
1832}
Simon Horman98637e52014-06-20 12:30:16 +09001833
1834static int prepare_external_check(struct check *check)
1835{
1836 struct server *s = check->server;
1837 struct proxy *px = s->proxy;
1838 struct listener *listener = NULL, *l;
1839 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001840 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001841 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001842
1843 list_for_each_entry(l, &px->conf.listeners, by_fe)
1844 /* Use the first INET, INET6 or UNIX listener */
1845 if (l->addr.ss_family == AF_INET ||
1846 l->addr.ss_family == AF_INET6 ||
1847 l->addr.ss_family == AF_UNIX) {
1848 listener = l;
1849 break;
1850 }
1851
Simon Horman98637e52014-06-20 12:30:16 +09001852 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001853 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1854 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001855 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001856 goto err;
1857 }
Simon Horman98637e52014-06-20 12:30:16 +09001858
Cyril Bontéac92a062014-12-27 22:28:38 +01001859 check->argv = calloc(6, sizeof(char *));
1860 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001861 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001862 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001863 }
Simon Horman98637e52014-06-20 12:30:16 +09001864
1865 check->argv[0] = px->check_command;
1866
Cyril Bonté777be862014-12-02 21:21:35 +01001867 if (!listener) {
1868 check->argv[1] = strdup("NOT_USED");
1869 check->argv[2] = strdup("NOT_USED");
1870 }
1871 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001872 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001873 addr_to_str(&listener->addr, buf, sizeof(buf));
1874 check->argv[1] = strdup(buf);
1875 port_to_str(&listener->addr, buf, sizeof(buf));
1876 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001877 }
1878 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001879 const struct sockaddr_un *un;
1880
1881 un = (struct sockaddr_un *)&listener->addr;
1882 check->argv[1] = strdup(un->sun_path);
1883 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001884 }
1885 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001886 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001887 goto err;
1888 }
1889
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001890 addr_to_str(&s->addr, buf, sizeof(buf));
1891 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001892
1893 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1894 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1895 else
1896 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001897 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001898
Cyril Bontéac92a062014-12-27 22:28:38 +01001899 for (i = 0; i < 5; i++) {
1900 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001901 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001902 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001903 }
1904 }
Simon Horman98637e52014-06-20 12:30:16 +09001905
Cyril Bontéac92a062014-12-27 22:28:38 +01001906 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001907 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001908 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1909 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1910 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1911 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001912 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001913 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1914 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1915 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1916 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1917 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1918 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1919
1920 /* Ensure that we don't leave any hole in check->envp */
1921 for (i = 0; i < EXTCHK_SIZE; i++)
1922 if (!check->envp[i])
1923 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001924
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001925 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001926err:
1927 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001928 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001929 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001930 free(check->envp);
1931 check->envp = NULL;
1932 }
1933
1934 if (check->argv) {
1935 for (i = 1; i < 5; i++)
1936 free(check->argv[i]);
1937 free(check->argv);
1938 check->argv = NULL;
1939 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001940 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001941}
1942
Simon Hormanb00d17a2014-06-13 16:18:16 +09001943/*
Simon Horman98637e52014-06-20 12:30:16 +09001944 * establish a server health-check that makes use of a process.
1945 *
1946 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001947 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001948 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001949 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001950 *
1951 * Blocks and then unblocks SIGCHLD
1952 */
1953static int connect_proc_chk(struct task *t)
1954{
Cyril Bontéac92a062014-12-27 22:28:38 +01001955 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001956 struct check *check = t->context;
1957 struct server *s = check->server;
1958 struct proxy *px = s->proxy;
1959 int status;
1960 pid_t pid;
1961
Willy Tarreaue7dff022015-04-03 01:14:29 +02001962 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001963
1964 block_sigchld();
1965
1966 pid = fork();
1967 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001968 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1969 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001970 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1971 goto out;
1972 }
1973 if (pid == 0) {
1974 /* Child */
1975 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001976 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001977 int fd;
1978
1979 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1980 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1981
Willy Tarreau2555ccf2019-02-21 22:22:06 +01001982 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02001983
Willy Tarreau9f6dc722019-03-01 11:15:10 +01001984 /* restore the initial FD limits */
1985 limit.rlim_cur = rlim_fd_cur_at_boot;
1986 limit.rlim_max = rlim_fd_max_at_boot;
1987 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
1988 getrlimit(RLIMIT_NOFILE, &limit);
1989 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
1990 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
1991 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
1992 }
1993
Simon Horman98637e52014-06-20 12:30:16 +09001994 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001995 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001996 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001997 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1998 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001999 exit(-1);
2000 }
2001
2002 /* Parent */
2003 if (check->result == CHK_RES_UNKNOWN) {
2004 if (pid_list_add(pid, t) != NULL) {
2005 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2006
2007 if (px->timeout.check && px->timeout.connect) {
2008 int t_con = tick_add(now_ms, px->timeout.connect);
2009 t->expire = tick_first(t->expire, t_con);
2010 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002011 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002012 goto out;
2013 }
2014 else {
2015 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2016 }
2017 kill(pid, SIGTERM); /* process creation error */
2018 }
2019 else
2020 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2021
2022out:
2023 unblock_sigchld();
2024 return status;
2025}
2026
2027/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002028 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002029 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002030 *
2031 * Please do NOT place any return statement in this function and only leave
2032 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002033 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002034static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002035{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002036 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002037 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002038 int rv;
2039 int ret;
2040 int expired = tick_is_expired(t->expire, now_ms);
2041
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002042 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002043 if (!(check->state & CHK_ST_INPROGRESS)) {
2044 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002045 if (!expired) /* woke up too early */
2046 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002047
2048 /* we don't send any health-checks when the proxy is
2049 * stopped, the server should not be checked or the check
2050 * is disabled.
2051 */
2052 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2053 s->proxy->state == PR_STSTOPPED)
2054 goto reschedule;
2055
2056 /* we'll initiate a new check */
2057 set_server_check_status(check, HCHK_STATUS_START, NULL);
2058
2059 check->state |= CHK_ST_INPROGRESS;
2060
Simon Hormandbf70192015-01-30 11:22:53 +09002061 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002062 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002063 /* the process was forked, we allow up to min(inter,
2064 * timeout.connect) for it to report its status, but
2065 * only when timeout.check is set as it may be to short
2066 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002067 */
2068 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2069
2070 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2071 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2072 t->expire = tick_first(t->expire, t_con);
2073 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002074 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002075 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002076 }
2077
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002078 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002079
2080 check->state &= ~CHK_ST_INPROGRESS;
2081 check_notify_failure(check);
2082
2083 /* we allow up to min(inter, timeout.connect) for a connection
2084 * to establish but only when timeout.check is set
2085 * as it may be to short for a full check otherwise
2086 */
2087 while (tick_is_expired(t->expire, now_ms)) {
2088 int t_con;
2089
2090 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2091 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2092
2093 if (s->proxy->timeout.check)
2094 t->expire = tick_first(t->expire, t_con);
2095 }
2096 }
2097 else {
2098 /* there was a test running.
2099 * First, let's check whether there was an uncaught error,
2100 * which can happen on connect timeout or error.
2101 */
2102 if (check->result == CHK_RES_UNKNOWN) {
2103 /* good connection is enough for pure TCP check */
2104 struct pid_list *elem = check->curpid;
2105 int status = HCHK_STATUS_UNKNOWN;
2106
2107 if (elem->exited) {
2108 status = elem->status; /* Save in case the process exits between use below */
2109 if (!WIFEXITED(status))
2110 check->code = -1;
2111 else
2112 check->code = WEXITSTATUS(status);
2113 if (!WIFEXITED(status) || WEXITSTATUS(status))
2114 status = HCHK_STATUS_PROCERR;
2115 else
2116 status = HCHK_STATUS_PROCOK;
2117 } else if (expired) {
2118 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002119 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002120 kill(elem->pid, SIGTERM);
2121 }
2122 set_server_check_status(check, status, NULL);
2123 }
2124
2125 if (check->result == CHK_RES_FAILED) {
2126 /* a failure or timeout detected */
2127 check_notify_failure(check);
2128 }
2129 else if (check->result == CHK_RES_CONDPASS) {
2130 /* check is OK but asks for stopping mode */
2131 check_notify_stopping(check);
2132 }
2133 else if (check->result == CHK_RES_PASSED) {
2134 /* a success was detected */
2135 check_notify_success(check);
2136 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002137 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002138 check->state &= ~CHK_ST_INPROGRESS;
2139
2140 pid_list_del(check->curpid);
2141
2142 rv = 0;
2143 if (global.spread_checks > 0) {
2144 rv = srv_getinter(check) * global.spread_checks / 100;
2145 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2146 }
2147 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2148 }
2149
2150 reschedule:
2151 while (tick_is_expired(t->expire, now_ms))
2152 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002153
2154 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002155 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002156 return t;
2157}
2158
2159/*
2160 * manages a server health-check that uses a connection. Returns
2161 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002162 *
2163 * Please do NOT place any return statement in this function and only leave
2164 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002165 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002166static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002167{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002168 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002169 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002170 struct conn_stream *cs = check->cs;
2171 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002172 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002173 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002174 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002175
Olivier Houchard0923fa42019-01-11 18:43:04 +01002176 if (check->server)
2177 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002178 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002179 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002180 if (!expired) /* woke up too early */
2181 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002182
Simon Horman671b6f02013-11-25 10:46:39 +09002183 /* we don't send any health-checks when the proxy is
2184 * stopped, the server should not be checked or the check
2185 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002186 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002187 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002188 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002189 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002190
2191 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002192 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002193
Willy Tarreau2c115e52013-12-11 19:41:16 +01002194 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002195 b_reset(&check->bi);
2196 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002197
Simon Hormandbf70192015-01-30 11:22:53 +09002198 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002199 cs = check->cs;
2200 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002201
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002202 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002203 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002204 goto out_unlock;
2205
Willy Tarreaue7dff022015-04-03 01:14:29 +02002206 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002207 /* we allow up to min(inter, timeout.connect) for a connection
2208 * to establish but only when timeout.check is set
2209 * as it may be to short for a full check otherwise
2210 */
Simon Horman4a741432013-02-23 15:35:38 +09002211 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002212
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002213 if (proxy->timeout.check && proxy->timeout.connect) {
2214 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002215 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002216 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002217
Olivier Houchard53216e72018-10-10 15:46:36 +02002218 if (check->type)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002219 __event_srv_chk_r(cs);
Willy Tarreau06559ac2013-12-05 01:53:08 +01002220
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002221 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002222 goto reschedule;
2223
Willy Tarreaue7dff022015-04-03 01:14:29 +02002224 case SF_ERR_SRVTO: /* ETIMEDOUT */
2225 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002226 if (conn)
2227 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002228 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002229 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002230 /* should share same code than cases below */
2231 case SF_ERR_CHK_PORT:
2232 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002233 case SF_ERR_PRXCOND:
2234 case SF_ERR_RESOURCE:
2235 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002236 if (conn)
2237 conn->flags |= CO_FL_ERROR;
2238 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002239 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002240 }
2241
Willy Tarreau5a78f362012-11-23 12:47:05 +01002242 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002243 if (cs) {
2244 cs_destroy(cs);
2245 cs = check->cs = NULL;
2246 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002247 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002248
Willy Tarreau2c115e52013-12-11 19:41:16 +01002249 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002250 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002251
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002252 /* we allow up to min(inter, timeout.connect) for a connection
2253 * to establish but only when timeout.check is set
2254 * as it may be to short for a full check otherwise
2255 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002256 while (tick_is_expired(t->expire, now_ms)) {
2257 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002258
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002259 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002260 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002261
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002262 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002263 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002264 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002265 }
2266 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002267 /* there was a test running.
2268 * First, let's check whether there was an uncaught error,
2269 * which can happen on connect timeout or error.
2270 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002271 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002272 /* good connection is enough for pure TCP check */
2273 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002274 if (check->use_ssl)
2275 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002276 else
Simon Horman4a741432013-02-23 15:35:38 +09002277 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002278 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002279 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002280 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002281 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002282 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002283 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002284 }
2285
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002286 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002287 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002288 /* The check was aborted and the connection was not yet closed.
2289 * This can happen upon timeout, or when an external event such
2290 * as a failed response coupled with "observe layer7" caused the
2291 * server state to be suddenly changed.
2292 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002293 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002294 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002295 }
2296
Willy Tarreauac59f362017-10-08 11:10:19 +02002297 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002298 cs_destroy(cs);
2299 cs = check->cs = NULL;
2300 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002301 }
2302
Olivier Houchard0923fa42019-01-11 18:43:04 +01002303 if (check->server) {
2304 if (check->result == CHK_RES_FAILED) {
2305 /* a failure or timeout detected */
2306 check_notify_failure(check);
2307 }
2308 else if (check->result == CHK_RES_CONDPASS) {
2309 /* check is OK but asks for stopping mode */
2310 check_notify_stopping(check);
2311 }
2312 else if (check->result == CHK_RES_PASSED) {
2313 /* a success was detected */
2314 check_notify_success(check);
2315 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002316 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002317 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002318 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002319
Olivier Houchard0923fa42019-01-11 18:43:04 +01002320 if (check->server) {
2321 rv = 0;
2322 if (global.spread_checks > 0) {
2323 rv = srv_getinter(check) * global.spread_checks / 100;
2324 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2325 }
2326 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002327 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002328 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002329
2330 reschedule:
2331 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002332 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002333 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002334 if (check->server)
2335 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002336 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002337}
2338
Simon Horman98637e52014-06-20 12:30:16 +09002339/*
2340 * manages a server health-check. Returns
2341 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2342 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002343static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002344{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002345 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002346
2347 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002348 return process_chk_proc(t, context, state);
2349 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002350
Simon Horman98637e52014-06-20 12:30:16 +09002351}
2352
Simon Horman5c942422013-11-25 10:46:32 +09002353static int start_check_task(struct check *check, int mininter,
2354 int nbcheck, int srvpos)
2355{
2356 struct task *t;
2357 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002358 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002359 ha_alert("Starting [%s:%s] check: out of memory.\n",
2360 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002361 return 0;
2362 }
2363
2364 check->task = t;
2365 t->process = process_chk;
2366 t->context = check;
2367
Willy Tarreau1746eec2014-04-25 10:46:47 +02002368 if (mininter < srv_getinter(check))
2369 mininter = srv_getinter(check);
2370
2371 if (global.max_spread_checks && mininter > global.max_spread_checks)
2372 mininter = global.max_spread_checks;
2373
Simon Horman5c942422013-11-25 10:46:32 +09002374 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002375 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002376 check->start = now;
2377 task_queue(t);
2378
2379 return 1;
2380}
2381
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002382/*
2383 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002384 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002385 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002386static int start_checks()
2387{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002388
2389 struct proxy *px;
2390 struct server *s;
2391 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002392 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002393
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002394 /* 1- count the checkers to run simultaneously.
2395 * We also determine the minimum interval among all of those which
2396 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2397 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002398 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002399 * too short an interval for all others.
2400 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002401 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002402 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002403 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002404 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002405 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002406 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002407 }
2408 /* We need a warmup task that will be called when the server
2409 * state switches from down to up.
2410 */
2411 s->warmup = t;
2412 t->process = server_warmup;
2413 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002414 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002415 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002416 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 +01002417 }
2418
Willy Tarreaud8514a22013-12-11 21:10:14 +01002419 if (s->check.state & CHK_ST_CONFIGURED) {
2420 nbcheck++;
2421 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2422 (!mininter || mininter > srv_getinter(&s->check)))
2423 mininter = srv_getinter(&s->check);
2424 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002425
Willy Tarreaud8514a22013-12-11 21:10:14 +01002426 if (s->agent.state & CHK_ST_CONFIGURED) {
2427 nbcheck++;
2428 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2429 (!mininter || mininter > srv_getinter(&s->agent)))
2430 mininter = srv_getinter(&s->agent);
2431 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002432 }
2433 }
2434
Simon Horman4a741432013-02-23 15:35:38 +09002435 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002436 return 0;
2437
2438 srand((unsigned)time(NULL));
2439
2440 /*
2441 * 2- start them as far as possible from each others. For this, we will
2442 * start them after their interval set to the min interval divided by
2443 * the number of servers, weighted by the server's position in the list.
2444 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002445 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002446 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2447 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002448 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002449 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002450 }
2451 }
2452
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002453 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002454 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002455 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002456 if (s->check.type == PR_O2_EXT_CHK) {
2457 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002458 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002459 }
Simon Hormand60d6912013-11-25 10:46:36 +09002460 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002461 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002462 srvpos++;
2463 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002464
Simon Hormand60d6912013-11-25 10:46:36 +09002465 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002466 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002467 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002468 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002469 }
2470 srvpos++;
2471 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002472 }
2473 }
2474 return 0;
2475}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002476
2477/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002478 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002479 * The buffer MUST be terminated by a null byte before calling this function.
2480 * Sets server status appropriately. The caller is responsible for ensuring
2481 * that the buffer contains at least 13 characters. If <done> is zero, we may
2482 * return 0 to indicate that data is required to decide of a match.
2483 */
2484static int httpchk_expect(struct server *s, int done)
2485{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002486 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002487 char status_code[] = "000";
2488 char *contentptr;
2489 int crlf;
2490 int ret;
2491
2492 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2493 case PR_O2_EXP_STS:
2494 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002495 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2496 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002497
2498 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2499 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2500 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002501 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002502
2503 /* we necessarily have the response, so there are no partial failures */
2504 if (s->proxy->options2 & PR_O2_EXP_INV)
2505 ret = !ret;
2506
Simon Horman4a741432013-02-23 15:35:38 +09002507 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002508 break;
2509
2510 case PR_O2_EXP_STR:
2511 case PR_O2_EXP_RSTR:
2512 /* very simple response parser: ignore CR and only count consecutive LFs,
2513 * stop with contentptr pointing to first char after the double CRLF or
2514 * to '\0' if crlf < 2.
2515 */
2516 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002517 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002518 if (crlf >= 2)
2519 break;
2520 if (*contentptr == '\r')
2521 continue;
2522 else if (*contentptr == '\n')
2523 crlf++;
2524 else
2525 crlf = 0;
2526 }
2527
2528 /* Check that response contains a body... */
2529 if (crlf < 2) {
2530 if (!done)
2531 return 0;
2532
Simon Horman4a741432013-02-23 15:35:38 +09002533 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002534 "HTTP content check could not find a response body");
2535 return 1;
2536 }
2537
2538 /* Check that response body is not empty... */
2539 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002540 if (!done)
2541 return 0;
2542
Simon Horman4a741432013-02-23 15:35:38 +09002543 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002544 "HTTP content check found empty response body");
2545 return 1;
2546 }
2547
2548 /* Check the response content against the supplied string
2549 * or regex... */
2550 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2551 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2552 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002553 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002554
2555 /* if we don't match, we may need to wait more */
2556 if (!ret && !done)
2557 return 0;
2558
2559 if (ret) {
2560 /* content matched */
2561 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002562 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002563 "HTTP check matched unwanted content");
2564 else
Simon Horman4a741432013-02-23 15:35:38 +09002565 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002566 "HTTP content check matched");
2567 }
2568 else {
2569 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002570 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002571 "HTTP check did not match unwanted content");
2572 else
Simon Horman4a741432013-02-23 15:35:38 +09002573 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002574 "HTTP content check did not match");
2575 }
2576 break;
2577 }
2578 return 1;
2579}
2580
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002581/*
2582 * return the id of a step in a send/expect session
2583 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002584static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002585{
2586 struct tcpcheck_rule *cur = NULL, *next = NULL;
2587 int i = 0;
2588
Willy Tarreau213c6782014-10-02 14:51:02 +02002589 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002590 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002591 return 0;
2592
Simon Hormane16c1b32015-01-30 11:22:57 +09002593 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002594
2595 /* no step => first step */
2596 if (cur == NULL)
2597 return 1;
2598
2599 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002600 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002601 if (next->list.p == &cur->list)
2602 break;
2603 ++i;
2604 }
2605
2606 return i;
2607}
2608
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002609/*
2610 * return the latest known comment before (including) the given stepid
2611 * returns NULL if no comment found
2612 */
2613static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2614{
2615 struct tcpcheck_rule *cur = NULL;
2616 char *ret = NULL;
2617 int i = 0;
2618
2619 /* not even started anything yet, return latest comment found before any action */
2620 if (!check->current_step) {
2621 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2622 if (cur->action == TCPCHK_ACT_COMMENT)
2623 ret = cur->comment;
2624 else
2625 goto return_comment;
2626 }
2627 }
2628
2629 i = 1;
2630 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2631 if (cur->comment)
2632 ret = cur->comment;
2633
2634 if (i >= stepid)
2635 goto return_comment;
2636
2637 ++i;
2638 }
2639
2640 return_comment:
2641 return ret;
2642}
2643
Willy Tarreaube74b882017-10-04 16:22:49 +02002644/* proceed with next steps for the TCP checks <check>. Note that this is called
2645 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002646 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2647 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002648 *
2649 * Please do NOT place any return statement in this function and only leave
2650 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002651 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002652static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002653{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002654 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002655 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002656 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002657 struct conn_stream *cs = check->cs;
2658 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002659 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002660 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002661 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002662 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002663 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002664
Willy Tarreauef953952014-10-02 14:30:14 +02002665 /* here, we know that the check is complete or that it failed */
2666 if (check->result != CHK_RES_UNKNOWN)
2667 goto out_end_tcpcheck;
2668
2669 /* We have 4 possibilities here :
2670 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002671 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002672 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2673 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002674 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002675 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002676 * be waiting for a connection attempt to complete. conn!=NULL.
2677 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002678 *
2679 * #2 and #3 are quite similar, we want both the connection and the
2680 * handshake to complete before going any further. Thus we must always
2681 * wait for a connection to complete unless we're before and existing
2682 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002683 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002684
2685 /* find first rule and skip comments */
2686 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2687 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2688 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2689
Willy Tarreau00149122017-10-04 18:05:01 +02002690 if ((check->current_step || &next->list == head) &&
2691 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002692 /* we allow up to min(inter, timeout.connect) for a connection
2693 * to establish but only when timeout.check is set
2694 * as it may be to short for a full check otherwise
2695 */
2696 while (tick_is_expired(t->expire, now_ms)) {
2697 int t_con;
2698
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002699 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002700 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2701
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002702 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002703 t->expire = tick_first(t->expire, t_con);
2704 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002705 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002706 }
2707
Willy Tarreauef953952014-10-02 14:30:14 +02002708 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002709 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002710 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002711 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002712 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002713
Willy Tarreau213c6782014-10-02 14:51:02 +02002714 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002715 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002716 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002717 b_reset(&check->bo);
2718 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002719 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002720 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002721 if (proxy->timeout.check)
2722 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002723 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002724
Willy Tarreauabca5b62013-12-06 14:19:25 +01002725 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002726 /* We have to try to flush the output buffer before reading, at
2727 * the end, or if we're about to send a string that does not fit
2728 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002729 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002730 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002731 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002732 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002733 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002734 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002735 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002736
Olivier Houcharded0f2072018-08-16 15:41:52 +02002737 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002738 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002739
2740 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002741 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002742 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002743 goto out_end_tcpcheck;
2744 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002745 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002746 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002747 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002748 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002749 goto out;
2750 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002751 }
2752
Willy Tarreau263013d2015-05-13 11:59:14 +02002753 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002754 break;
2755
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002756 /* have 'next' point to the next rule or NULL if we're on the
2757 * last one, connect() needs this.
2758 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002759 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002760
2761 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002762 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002763 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002764
2765 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002766 if (&next->list == head)
2767 next = NULL;
2768
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002769 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2770 struct protocol *proto;
2771 struct xprt_ops *xprt;
2772
Willy Tarreau00149122017-10-04 18:05:01 +02002773 /* For a connect action we'll create a new connection.
2774 * We may also have to kill a previous one. But we don't
2775 * want to leave *without* a connection if we came here
2776 * from the connection layer, hence with a connection.
2777 * Thus we'll proceed in the following order :
2778 * 1: close but not release previous connection
2779 * 2: try to get a new connection
2780 * 3: release and replace the old one on success
2781 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002782 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002783 cs_close(check->cs);
2784 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002785 }
2786
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002787 /* mark the step as started */
2788 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002789
2790 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002791 cs = cs_new(NULL);
2792 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002793 step = tcpcheck_get_step_id(check);
2794 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2795 comment = tcpcheck_get_step_comment(check, step);
2796 if (comment)
2797 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002798 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2799 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002800 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002801 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002802 }
2803
Willy Tarreauac59f362017-10-08 11:10:19 +02002804 if (check->cs)
2805 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002806
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002807 check->cs = cs;
2808 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002809 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002810 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002811 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002812
2813 /* no client address */
2814 clear_addr(&conn->addr.from);
2815
Simon Horman41f58762015-01-30 11:22:56 +09002816 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002817 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002818 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002819 }
2820 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002821 /* we'll connect to the addr on the server */
2822 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002823 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002824 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002825
2826 /* port */
2827 if (check->current_step->port)
2828 set_host_port(&conn->addr.to, check->current_step->port);
2829 else if (check->port)
2830 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002831 else if (s->svc_port)
2832 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002833
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002834 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002835 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002836 }
2837 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002838 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002839 }
Willy Tarreaube373152018-09-06 11:45:30 +02002840
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002841 conn_prepare(conn, proto, xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01002842 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0)
2843 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02002844 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002845
Willy Tarreaue7dff022015-04-03 01:14:29 +02002846 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002847 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002848 ret = proto->connect(conn,
Olivier Houchardfdcb0072019-05-06 18:32:29 +02002849 CONNECT_HAS_DATA /* I/O polling is always needed */ | (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : CONNECT_DELACK_ALWAYS);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002850 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2851 conn->send_proxy_ofs = 1;
2852 conn->flags |= CO_FL_SEND_PROXY;
2853 }
2854
2855 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002856 * - SF_ERR_NONE if everything's OK
2857 * - SF_ERR_SRVTO if there are no more servers
2858 * - SF_ERR_SRVCL if the connection was refused by the server
2859 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2860 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2861 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002862 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002863 * Note that we try to prevent the network stack from sending the ACK during the
2864 * connect() when a pure TCP check is used (without PROXY protocol).
2865 */
2866 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002867 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002868 /* we allow up to min(inter, timeout.connect) for a connection
2869 * to establish but only when timeout.check is set
2870 * as it may be to short for a full check otherwise
2871 */
2872 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2873
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002874 if (proxy->timeout.check && proxy->timeout.connect) {
2875 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002876 t->expire = tick_first(t->expire, t_con);
2877 }
2878 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002879 case SF_ERR_SRVTO: /* ETIMEDOUT */
2880 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002881 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002882 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002883 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002884 comment = tcpcheck_get_step_comment(check, step);
2885 if (comment)
2886 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002887 set_server_check_status(check, HCHK_STATUS_L4CON,
2888 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002889 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002890 case SF_ERR_PRXCOND:
2891 case SF_ERR_RESOURCE:
2892 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002893 step = tcpcheck_get_step_id(check);
2894 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002895 comment = tcpcheck_get_step_comment(check, step);
2896 if (comment)
2897 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002898 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2899 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002900 goto out_end_tcpcheck;
2901 }
2902
2903 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002904 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002905
2906 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002907 while (&check->current_step->list != head &&
2908 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002909 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002910
Willy Tarreauf2c87352015-05-13 12:08:21 +02002911 if (&check->current_step->list == head)
2912 break;
2913
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002914 } /* end 'connect' */
2915 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2916 /* mark the step as started */
2917 check->last_started_step = check->current_step;
2918
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002919 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002920 if (*b_head(&check->bi) != '\0') {
2921 *b_head(&check->bi) = '\0';
2922 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002923 }
2924
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002925 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002926 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002927 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002928 goto out_end_tcpcheck;
2929 }
2930
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002931 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002932 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002933 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002934 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002935 set_server_check_status(check, HCHK_STATUS_L7RSP,
2936 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002937 goto out_end_tcpcheck;
2938 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002939
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002940 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002941 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002942 continue;
2943
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002944 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2945 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002946
Willy Tarreauabca5b62013-12-06 14:19:25 +01002947 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002948 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002949
2950 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002951 while (&check->current_step->list != head &&
2952 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002953 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002954
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002955 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002956 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002957 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002958 goto out_end_tcpcheck;
2959
Olivier Houchard511efea2018-08-16 15:30:32 +02002960 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002961 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002962 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002963 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002964 /* Report network errors only if we got no other data. Otherwise
2965 * we'll let the upper layers decide whether the response is OK
2966 * or not. It is very common that an RST sent by the server is
2967 * reported as an error just after the last data chunk.
2968 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002969 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002970 goto out_end_tcpcheck;
2971 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002972 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002973 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002974 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002975 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002976 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977 }
2978
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002979 /* mark the step as started */
2980 check->last_started_step = check->current_step;
2981
2982
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002983 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002984 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002985 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002986 if (b_data(&check->bi) < b_size(&check->bi)) {
2987 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002988 }
2989 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002990 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002991 done = 1; /* buffer full, don't wait for more data */
2992 }
2993
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002994 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002995
2996 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002997 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002998 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002999 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003000
3001 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003002 step = tcpcheck_get_step_id(check);
3003 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003004 comment = tcpcheck_get_step_comment(check, step);
3005 if (comment)
3006 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003007 set_server_check_status(check, HCHK_STATUS_L7RSP,
3008 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003009
3010 goto out_end_tcpcheck;
3011 }
3012
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003013 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003014 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003015
Willy Tarreaua970c282013-12-06 12:47:19 +01003016 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003017 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003018 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 +02003019 else if (check->current_step->expect_regex != NULL)
3020 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003021
3022 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003023 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003024
3025 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003026 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003027 if (ret) {
3028 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003029 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003030 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003031 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003032 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003033 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003034 }
3035 else {
3036 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003037 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003038 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003039 comment = tcpcheck_get_step_comment(check, step);
3040 if (comment)
3041 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003042 set_server_check_status(check, HCHK_STATUS_L7RSP,
3043 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003044 goto out_end_tcpcheck;
3045 }
3046 /* matched and was supposed to => OK, next step */
3047 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003048 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003049 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003050
3051 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003052 while (&check->current_step->list != head &&
3053 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003054 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003055
Willy Tarreauf2c87352015-05-13 12:08:21 +02003056 if (&check->current_step->list == head)
3057 break;
3058
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003059 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003060 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003061 }
3062 }
3063 else {
3064 /* not matched */
3065 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003066 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003067 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003068 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003069
3070 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003071 while (&check->current_step->list != head &&
3072 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003073 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003074
Willy Tarreauf2c87352015-05-13 12:08:21 +02003075 if (&check->current_step->list == head)
3076 break;
3077
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003078 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003079 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003080 }
3081 /* not matched but was supposed to => ERROR */
3082 else {
3083 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003084 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003085 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003086 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003087 }
3088 else {
3089 /* we were looking for a regex */
3090 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003091 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003092 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003093 comment = tcpcheck_get_step_comment(check, step);
3094 if (comment)
3095 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003096 set_server_check_status(check, HCHK_STATUS_L7RSP,
3097 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003098 goto out_end_tcpcheck;
3099 }
3100 }
3101 } /* end expect */
3102 } /* end loop over double chained step list */
3103
Baptiste Assmann248f1172018-03-01 21:49:01 +01003104 /* don't do anything until the connection is established */
3105 if (!(conn->flags & CO_FL_CONNECTED)) {
3106 /* update expire time, should be done by process_chk */
3107 /* we allow up to min(inter, timeout.connect) for a connection
3108 * to establish but only when timeout.check is set
3109 * as it may be to short for a full check otherwise
3110 */
3111 while (tick_is_expired(t->expire, now_ms)) {
3112 int t_con;
3113
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003114 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003115 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3116
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003117 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003118 t->expire = tick_first(t->expire, t_con);
3119 }
3120 goto out;
3121 }
3122
Willy Tarreau263013d2015-05-13 11:59:14 +02003123 /* We're waiting for some I/O to complete, we've reached the end of the
3124 * rules, or both. Do what we have to do, otherwise we're done.
3125 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003126 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003127 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3128 goto out_end_tcpcheck;
3129 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003130
Willy Tarreau53c5a042015-05-13 11:38:17 +02003131 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003132 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003133 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003134 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003135
3136 out_end_tcpcheck:
3137 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003138 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003139 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003140
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003141 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003142 check->current_step = NULL;
3143
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003144 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003145 conn->flags |= CO_FL_ERROR;
3146
Christopher Fauletb6102852017-11-28 10:06:29 +01003147 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003148 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003149}
3150
Simon Hormanb1900d52015-01-30 11:22:54 +09003151const char *init_check(struct check *check, int type)
3152{
3153 check->type = type;
3154
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003155 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3156 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003157
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003158 check->bi.area = calloc(check->bi.size, sizeof(char));
3159 check->bo.area = calloc(check->bo.size, sizeof(char));
3160
3161 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003162 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003163
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003164 check->wait_list.task = tasklet_new();
3165 if (!check->wait_list.task)
3166 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003167 check->wait_list.events = 0;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003168 check->wait_list.task->process = event_srv_chk_io;
3169 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003170 return NULL;
3171}
3172
Simon Hormanbfb5d332015-01-30 11:22:55 +09003173void free_check(struct check *check)
3174{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003175 free(check->bi.area);
3176 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003177 if (check->cs) {
3178 free(check->cs->conn);
3179 check->cs->conn = NULL;
3180 cs_free(check->cs);
3181 check->cs = NULL;
3182 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003183}
3184
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003185void email_alert_free(struct email_alert *alert)
3186{
3187 struct tcpcheck_rule *rule, *back;
3188
3189 if (!alert)
3190 return;
3191
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003192 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3193 LIST_DEL(&rule->list);
3194 free(rule->comment);
3195 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003196 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003197 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003198 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003199 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003200}
3201
Olivier Houchard9f6af332018-05-25 14:04:04 +02003202static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003203{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003204 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003205 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003206 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003207
3208 q = container_of(check, typeof(*q), check);
3209
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003210 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003211 while (1) {
3212 if (!(check->state & CHK_ST_ENABLED)) {
3213 if (LIST_ISEMPTY(&q->email_alerts)) {
3214 /* All alerts processed, queue the task */
3215 t->expire = TICK_ETERNITY;
3216 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003217 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003218 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003219
3220 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003221 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003222 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003223 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003224 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003225 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003226 }
3227
Olivier Houchard9f6af332018-05-25 14:04:04 +02003228 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003229 if (check->state & CHK_ST_INPROGRESS)
3230 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003231
3232 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3233 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003234 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003235 check->server = NULL;
3236 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003237 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003238 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003239 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003240 return t;
3241}
3242
Christopher Faulet0108bb32017-10-20 21:34:32 +02003243/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3244 *
3245 * The function returns 1 in success case, otherwise, it returns 0 and err is
3246 * filled.
3247 */
3248int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003249{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003250 struct mailer *mailer;
3251 struct email_alertq *queues;
3252 const char *err_str;
3253 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003254
Christopher Faulet0108bb32017-10-20 21:34:32 +02003255 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3256 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003257 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003258 }
3259
Christopher Faulet0108bb32017-10-20 21:34:32 +02003260 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3261 struct email_alertq *q = &queues[i];
3262 struct check *check = &q->check;
3263 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003264
3265 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003266 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003267 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003268 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003269 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003270 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003271 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3272 memprintf(err, "%s", err_str);
3273 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003274 }
3275
3276 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003277 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003278 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003279
Emeric Brunc60def82017-09-27 14:59:38 +02003280 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003281 memprintf(err, "out of memory while allocating mailer alerts task");
3282 goto error;
3283 }
3284
3285 check->task = t;
3286 t->process = process_email_alert;
3287 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003288
Christopher Faulet0108bb32017-10-20 21:34:32 +02003289 /* check this in one ms */
3290 t->expire = TICK_ETERNITY;
3291 check->start = now;
3292 task_queue(t);
3293 }
3294
3295 mls->users++;
3296 free(p->email_alert.mailers.name);
3297 p->email_alert.mailers.m = mls;
3298 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003299 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003300
3301 error:
3302 for (i = 0; i < mls->count; i++) {
3303 struct email_alertq *q = &queues[i];
3304 struct check *check = &q->check;
3305
3306 if (check->task) {
Olivier Houchard3f795f72019-04-17 22:51:06 +02003307 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003308 check->task = NULL;
3309 }
3310 free_check(check);
3311 }
3312 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003313 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003314 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003315}
3316
3317
3318static int add_tcpcheck_expect_str(struct list *list, const char *str)
3319{
3320 struct tcpcheck_rule *tcpcheck;
3321
Willy Tarreaubafbe012017-11-24 17:34:44 +01003322 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003323 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003324 memset(tcpcheck, 0, sizeof(*tcpcheck));
3325 tcpcheck->action = TCPCHK_ACT_EXPECT;
3326 tcpcheck->string = strdup(str);
3327 tcpcheck->expect_regex = NULL;
3328 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003329 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003330 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003331 return 0;
3332 }
3333
3334 LIST_ADDQ(list, &tcpcheck->list);
3335 return 1;
3336}
3337
3338static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3339{
3340 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003341 const char *in;
3342 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003343 int i;
3344
Willy Tarreaubafbe012017-11-24 17:34:44 +01003345 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003346 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003347 memset(tcpcheck, 0, sizeof(*tcpcheck));
3348 tcpcheck->action = TCPCHK_ACT_SEND;
3349 tcpcheck->expect_regex = NULL;
3350 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003351 tcpcheck->string_len = 0;
3352 for (i = 0; strs[i]; i++)
3353 tcpcheck->string_len += strlen(strs[i]);
3354
3355 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3356 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003357 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003358 return 0;
3359 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003360
Willy Tarreau64345aa2016-08-10 19:29:09 +02003361 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003362 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003363 for (in = strs[i]; (*dst = *in++); dst++);
3364 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003365
3366 LIST_ADDQ(list, &tcpcheck->list);
3367 return 1;
3368}
3369
Christopher Faulet0108bb32017-10-20 21:34:32 +02003370static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3371 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003372{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003373 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003374 struct tcpcheck_rule *tcpcheck;
3375 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003376
Willy Tarreaubafbe012017-11-24 17:34:44 +01003377 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003378 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003379 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003381 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003382
Willy Tarreaubafbe012017-11-24 17:34:44 +01003383 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003384 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003385 memset(tcpcheck, 0, sizeof(*tcpcheck));
3386 tcpcheck->action = TCPCHK_ACT_CONNECT;
3387 tcpcheck->comment = NULL;
3388 tcpcheck->string = NULL;
3389 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003390 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3391
3392 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3393 goto error;
3394
3395 {
3396 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3397 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3398 goto error;
3399 }
3400
3401 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3402 goto error;
3403
3404 {
3405 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3406 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3407 goto error;
3408 }
3409
3410 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3411 goto error;
3412
3413 {
3414 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3415 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3416 goto error;
3417 }
3418
3419 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3420 goto error;
3421
3422 {
3423 const char * const strs[2] = { "DATA\r\n" };
3424 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3425 goto error;
3426 }
3427
3428 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3429 goto error;
3430
3431 {
3432 struct tm tm;
3433 char datestr[48];
3434 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003435 "From: ", p->email_alert.from, "\r\n",
3436 "To: ", p->email_alert.to, "\r\n",
3437 "Date: ", datestr, "\r\n",
3438 "Subject: [HAproxy Alert] ", msg, "\r\n",
3439 "\r\n",
3440 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003441 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003442 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003443 NULL
3444 };
3445
3446 get_localtime(date.tv_sec, &tm);
3447
3448 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3449 goto error;
3450 }
3451
3452 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3453 goto error;
3454 }
3455
3456 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3457 goto error;
3458
3459 {
3460 const char * const strs[2] = { "QUIT\r\n" };
3461 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3462 goto error;
3463 }
3464
3465 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3466 goto error;
3467
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003468 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003469 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003470 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003471 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003472 return 1;
3473
3474error:
3475 email_alert_free(alert);
3476 return 0;
3477}
3478
Christopher Faulet0108bb32017-10-20 21:34:32 +02003479static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003480{
3481 int i;
3482 struct mailer *mailer;
3483
3484 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3485 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003486 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003487 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003488 return;
3489 }
3490 }
3491
3492 return;
3493}
3494
3495/*
3496 * Send email alert if configured.
3497 */
Simon Horman64e34162015-02-06 11:11:57 +09003498void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003499{
3500 va_list argp;
3501 char buf[1024];
3502 int len;
3503 struct proxy *p = s->proxy;
3504
Christopher Faulet0108bb32017-10-20 21:34:32 +02003505 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003506 return;
3507
3508 va_start(argp, format);
3509 len = vsnprintf(buf, sizeof(buf), format, argp);
3510 va_end(argp);
3511
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003512 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003513 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003514 return;
3515 }
3516
Christopher Faulet0108bb32017-10-20 21:34:32 +02003517 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003518}
3519
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003520/*
3521 * Return value:
3522 * the port to be used for the health check
3523 * 0 in case no port could be found for the check
3524 */
3525int srv_check_healthcheck_port(struct check *chk)
3526{
3527 int i = 0;
3528 struct server *srv = NULL;
3529
3530 srv = chk->server;
3531
3532 /* If neither a port nor an addr was specified and no check transport
3533 * layer is forced, then the transport layer used by the checks is the
3534 * same as for the production traffic. Otherwise we use raw_sock by
3535 * default, unless one is specified.
3536 */
3537 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003538 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003539 chk->send_proxy |= (srv->pp_opts);
3540 }
3541
3542 /* by default, we use the health check port ocnfigured */
3543 if (chk->port > 0)
3544 return chk->port;
3545
3546 /* try to get the port from check_core.addr if check.port not set */
3547 i = get_host_port(&chk->addr);
3548 if (i > 0)
3549 return i;
3550
3551 /* try to get the port from server address */
3552 /* prevent MAPPORTS from working at this point, since checks could
3553 * not be performed in such case (MAPPORTS impose a relative ports
3554 * based on live traffic)
3555 */
3556 if (srv->flags & SRV_F_MAPPORTS)
3557 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003558
3559 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003560 if (i > 0)
3561 return i;
3562
3563 return 0;
3564}
3565
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003566REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003567
Willy Tarreaubd741542010-03-16 18:46:54 +01003568/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003569 * Local variables:
3570 * c-indent-level: 8
3571 * c-basic-offset: 8
3572 * End:
3573 */