blob: 3a74996cf55ae47b85db9e2c0f7375515c48e995 [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>
Olivier Houchard9130a962017-10-17 17:33:43 +020063#include <proto/ssl_sock.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020064
Willy Tarreaubd741542010-03-16 18:46:54 +010065static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090066static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020067static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020068static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020069static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020070static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020071static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Willy Tarreau8ceae722018-11-26 11:58:30 +010073DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
74DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020075
76
Simon Horman63a4a822012-03-19 07:24:41 +090077static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010078 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
79 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020080 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020081
Willy Tarreau23964182014-05-20 20:56:30 +020082 /* Below we have finished checks */
83 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010085
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
89 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
90 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
93 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
94 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020095
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010096 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
97 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020099 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100101 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
102 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
103 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900104
105 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
106 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200107 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200108};
109
Cyril Bontéac92a062014-12-27 22:28:38 +0100110const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
111 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
122};
123
Simon Horman63a4a822012-03-19 07:24:41 +0900124static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100125 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
126
127 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
128 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
129
130 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
131 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
132 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
133 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
134
135 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
136 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
138};
139
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100140/* checks if <err> is a real error for errno or one that can be ignored, and
141 * return 0 for these ones or <err> for real ones.
142 */
143static inline int unclean_errno(int err)
144{
145 if (err == EAGAIN || err == EINPROGRESS ||
146 err == EISCONN || err == EALREADY)
147 return 0;
148 return err;
149}
150
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200151/*
152 * Convert check_status code to description
153 */
154const char *get_check_status_description(short check_status) {
155
156 const char *desc;
157
158 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200159 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200160 else
161 desc = NULL;
162
163 if (desc && *desc)
164 return desc;
165 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200166 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200167}
168
169/*
170 * Convert check_status code to short info
171 */
172const char *get_check_status_info(short check_status) {
173
174 const char *info;
175
176 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200177 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200178 else
179 info = NULL;
180
181 if (info && *info)
182 return info;
183 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200184 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200185}
186
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100187const char *get_analyze_status(short analyze_status) {
188
189 const char *desc;
190
191 if (analyze_status < HANA_STATUS_SIZE)
192 desc = analyze_statuses[analyze_status].desc;
193 else
194 desc = NULL;
195
196 if (desc && *desc)
197 return desc;
198 else
199 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
200}
201
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200202/*
Simon Horman4a741432013-02-23 15:35:38 +0900203 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200204 * an adequate CHK_RES_* value. The new check->health is computed based
205 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200206 *
207 * Show information in logs about failed health check if server is UP
208 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200209 */
Simon Horman4a741432013-02-23 15:35:38 +0900210static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100211{
Simon Horman4a741432013-02-23 15:35:38 +0900212 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200213 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200214 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900215
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200216 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100217 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900218 check->desc[0] = '\0';
219 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200220 return;
221 }
222
Simon Horman4a741432013-02-23 15:35:38 +0900223 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200224 return;
225
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200226 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900227 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
228 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200229 } else
Simon Horman4a741432013-02-23 15:35:38 +0900230 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200231
Simon Horman4a741432013-02-23 15:35:38 +0900232 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200233 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900234 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200235
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100236 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900237 check->duration = -1;
238 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200239 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900240 check->duration = tv_ms_elapsed(&check->start, &now);
241 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200242 }
243
Willy Tarreau23964182014-05-20 20:56:30 +0200244 /* no change is expected if no state change occurred */
245 if (check->result == CHK_RES_NEUTRAL)
246 return;
247
Olivier Houchard0923fa42019-01-11 18:43:04 +0100248 /* If the check was really just sending a mail, it won't have an
249 * associated server, so we're done now.
250 */
251 if (!s)
252 return;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200253 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200254
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200255 switch (check->result) {
256 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200257 /* Failure to connect to the agent as a secondary check should not
258 * cause the server to be marked down.
259 */
260 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900261 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200262 (check->health > 0)) {
Olivier Houchard7059c552019-03-08 18:49:32 +0100263 _HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200264 report = 1;
265 check->health--;
266 if (check->health < check->rise)
267 check->health = 0;
268 }
269 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200270
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200271 case CHK_RES_PASSED:
272 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
273 if ((check->health < check->rise + check->fall - 1) &&
274 (check->result == CHK_RES_PASSED || check->health > 0)) {
275 report = 1;
276 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200277
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200278 if (check->health >= check->rise)
279 check->health = check->rise + check->fall - 1; /* OK now */
280 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200281
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200282 /* clear consecutive_errors if observing is enabled */
283 if (s->onerror)
284 s->consecutive_errors = 0;
285 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100286
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200287 default:
288 break;
289 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200290
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200291 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
292 (status != prev_status || report)) {
293 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200294 "%s check for %sserver %s/%s %s%s",
295 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200296 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100297 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100298 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200299 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200300
Emeric Brun5a133512017-10-19 14:42:30 +0200301 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200302
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100303 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200304 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
305 (check->health >= check->rise) ? check->fall : check->rise,
306 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200307
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200308 ha_warning("%s.\n", trash.area);
309 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
310 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200311 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200312}
313
Willy Tarreau4eec5472014-05-20 22:32:27 +0200314/* Marks the check <check>'s server down if the current check is already failed
315 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200316 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200317static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200318{
Simon Horman4a741432013-02-23 15:35:38 +0900319 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900320
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200321 /* The agent secondary check should only cause a server to be marked
322 * as down if check->status is HCHK_STATUS_L7STS, which indicates
323 * that the agent returned "fail", "stopped" or "down".
324 * The implication here is that failure to connect to the agent
325 * as a secondary check should not cause the server to be marked
326 * down. */
327 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
328 return;
329
Willy Tarreau4eec5472014-05-20 22:32:27 +0200330 if (check->health > 0)
331 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100332
Willy Tarreau4eec5472014-05-20 22:32:27 +0200333 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200334 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200335}
336
Willy Tarreauaf549582014-05-16 17:37:50 +0200337/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200338 * it isn't in maintenance, it is not tracking a down server and other checks
339 * comply. The rule is simple : by default, a server is up, unless any of the
340 * following conditions is true :
341 * - health check failed (check->health < rise)
342 * - agent check failed (agent->health < rise)
343 * - the server tracks a down server (track && track->state == STOPPED)
344 * Note that if the server has a slowstart, it will switch to STARTING instead
345 * of RUNNING. Also, only the health checks support the nolb mode, so the
346 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200347 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200348static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200349{
Simon Horman4a741432013-02-23 15:35:38 +0900350 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100351
Emeric Brun52a91d32017-08-31 14:41:55 +0200352 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200353 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100354
Emeric Brun52a91d32017-08-31 14:41:55 +0200355 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200356 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100357
Willy Tarreau3e048382014-05-21 10:30:54 +0200358 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
359 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100360
Willy Tarreau3e048382014-05-21 10:30:54 +0200361 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
362 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200363
Emeric Brun52a91d32017-08-31 14:41:55 +0200364 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200365 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100366
Emeric Brun5a133512017-10-19 14:42:30 +0200367 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100368}
369
Willy Tarreaudb58b792014-05-21 13:57:23 +0200370/* Marks the check <check> as valid and tries to set its server into stopping mode
371 * if it was running or starting, and provided it isn't in maintenance and other
372 * checks comply. The conditions for the server to be marked in stopping mode are
373 * the same as for it to be turned up. Also, only the health checks support the
374 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200375 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200376static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200377{
Simon Horman4a741432013-02-23 15:35:38 +0900378 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100379
Emeric Brun52a91d32017-08-31 14:41:55 +0200380 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200381 return;
382
Willy Tarreaudb58b792014-05-21 13:57:23 +0200383 if (check->state & CHK_ST_AGENT)
384 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100385
Emeric Brun52a91d32017-08-31 14:41:55 +0200386 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200387 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100388
Willy Tarreaudb58b792014-05-21 13:57:23 +0200389 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
390 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100391
Willy Tarreaudb58b792014-05-21 13:57:23 +0200392 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
393 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100394
Willy Tarreaub26881a2017-12-23 11:16:49 +0100395 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100396}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200397
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100398/* note: use health_adjust() only, which first checks that the observe mode is
399 * enabled.
400 */
401void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100402{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100403 int failed;
404 int expire;
405
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100406 if (s->observe >= HANA_OBS_SIZE)
407 return;
408
Willy Tarreaubb956662013-01-24 00:37:39 +0100409 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100410 return;
411
412 switch (analyze_statuses[status].lr[s->observe - 1]) {
413 case 1:
414 failed = 1;
415 break;
416
417 case 2:
418 failed = 0;
419 break;
420
421 default:
422 return;
423 }
424
425 if (!failed) {
426 /* good: clear consecutive_errors */
427 s->consecutive_errors = 0;
428 return;
429 }
430
Olivier Houchard7059c552019-03-08 18:49:32 +0100431 _HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100432
433 if (s->consecutive_errors < s->consecutive_errors_limit)
434 return;
435
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100436 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
437 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100438
439 switch (s->onerror) {
440 case HANA_ONERR_FASTINTER:
441 /* force fastinter - nothing to do here as all modes force it */
442 break;
443
444 case HANA_ONERR_SUDDTH:
445 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900446 if (s->check.health > s->check.rise)
447 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100448
449 /* no break - fall through */
450
451 case HANA_ONERR_FAILCHK:
452 /* simulate a failed health check */
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 case HANA_ONERR_MARKDWN:
459 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900460 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200461 set_server_check_status(&s->check, HCHK_STATUS_HANA,
462 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200463 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100464 break;
465
466 default:
467 /* write a warning? */
468 break;
469 }
470
471 s->consecutive_errors = 0;
Olivier Houchard7059c552019-03-08 18:49:32 +0100472 _HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100473
Simon Horman66183002013-02-23 10:16:43 +0900474 if (s->check.fastinter) {
475 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300476 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200477 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300478 /* requeue check task with new expire */
479 task_queue(s->check.task);
480 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100481 }
482}
483
Willy Tarreaua1dab552014-04-14 15:04:54 +0200484static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100485{
486 int sv_state;
487 int ratio;
488 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800489 char addr[46];
490 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100491 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
492 "UP %d/%d", "UP",
493 "NOLB %d/%d", "NOLB",
494 "no check" };
495
496 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
497 hlen += 24;
498
Willy Tarreauff5ae352013-12-11 20:36:34 +0100499 if (!(s->check.state & CHK_ST_ENABLED))
500 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200501 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900502 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100503 sv_state = 3; /* UP */
504 else
505 sv_state = 2; /* going down */
506
Emeric Brun52a91d32017-08-31 14:41:55 +0200507 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100508 sv_state += 2;
509 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900510 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100511 sv_state = 1; /* going up */
512 else
513 sv_state = 0; /* DOWN */
514 }
515
Willy Tarreaua1dab552014-04-14 15:04:54 +0200516 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100517 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200518 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
519 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100520
Joseph Lynch514061c2015-01-15 17:52:59 -0800521 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100522 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
523 snprintf(port, sizeof(port), "%u", s->svc_port);
524 else
525 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800526
527 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
528 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100529 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200530 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100531 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
532 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
533 s->nbpend);
534
Emeric Brun52a91d32017-08-31 14:41:55 +0200535 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100536 now.tv_sec < s->last_change + s->slowstart &&
537 now.tv_sec >= s->last_change) {
538 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200539 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100540 }
541
542 buffer[hlen++] = '\r';
543 buffer[hlen++] = '\n';
544
545 return hlen;
546}
547
Willy Tarreau20a18342013-12-05 00:31:46 +0100548/* Check the connection. If an error has already been reported or the socket is
549 * closed, keep errno intact as it is supposed to contain the valid error code.
550 * If no error is reported, check the socket's error queue using getsockopt().
551 * Warning, this must be done only once when returning from poll, and never
552 * after an I/O error was attempted, otherwise the error queue might contain
553 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
554 * socket. Returns non-zero if an error was reported, zero if everything is
555 * clean (including a properly closed socket).
556 */
557static int retrieve_errno_from_socket(struct connection *conn)
558{
559 int skerr;
560 socklen_t lskerr = sizeof(skerr);
561
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100562 if (conn->flags & CO_FL_ERROR && (unclean_errno(errno) || !conn->ctrl))
Willy Tarreau20a18342013-12-05 00:31:46 +0100563 return 1;
564
Willy Tarreau3c728722014-01-23 13:50:42 +0100565 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100566 return 0;
567
Willy Tarreau585744b2017-08-24 14:31:19 +0200568 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100569 errno = skerr;
570
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100571 errno = unclean_errno(errno);
Willy Tarreau20a18342013-12-05 00:31:46 +0100572
573 if (!errno) {
574 /* we could not retrieve an error, that does not mean there is
575 * none. Just don't change anything and only report the prior
576 * error if any.
577 */
578 if (conn->flags & CO_FL_ERROR)
579 return 1;
580 else
581 return 0;
582 }
583
584 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
585 return 1;
586}
587
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100588/* Try to collect as much information as possible on the connection status,
589 * and adjust the server status accordingly. It may make use of <errno_bck>
590 * if non-null when the caller is absolutely certain of its validity (eg:
591 * checked just after a syscall). If the caller doesn't have a valid errno,
592 * it can pass zero, and retrieve_errno_from_socket() will be called to try
593 * to extract errno from the socket. If no error is reported, it will consider
594 * the <expired> flag. This is intended to be used when a connection error was
595 * reported in conn->flags or when a timeout was reported in <expired>. The
596 * function takes care of not updating a server status which was already set.
597 * All situations where at least one of <expired> or CO_FL_ERROR are set
598 * produce a status.
599 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200600static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200602 struct conn_stream *cs = check->cs;
603 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100604 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200605 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200606 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200607 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100608
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100609 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100610 return;
611
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100612 errno = unclean_errno(errno_bck);
613 if (conn && errno)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100614 retrieve_errno_from_socket(conn);
615
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200616 if (conn && !(conn->flags & CO_FL_ERROR) &&
617 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100618 return;
619
620 /* we'll try to build a meaningful error message depending on the
621 * context of the error possibly present in conn->err_code, and the
622 * socket error possibly collected above. This is useful to know the
623 * exact step of the L6 layer (eg: SSL handshake).
624 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200625 chk = get_trash_chunk();
626
627 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900628 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200629 if (!step)
630 chunk_printf(chk, " at initial connection step of tcp-check");
631 else {
632 chunk_printf(chk, " at step %d of tcp-check", step);
633 /* we were looking for a string */
634 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
635 if (check->last_started_step->port)
636 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
637 else
638 chunk_appendf(chk, " (connect)");
639 }
640 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
641 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200642 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200643 else if (check->last_started_step->expect_regex)
644 chunk_appendf(chk, " (expect regex)");
645 }
646 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
647 chunk_appendf(chk, " (send)");
648 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200649
650 comment = tcpcheck_get_step_comment(check, step);
651 if (comment)
652 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200653 }
654 }
655
Willy Tarreau00149122017-10-04 18:05:01 +0200656 if (conn && conn->err_code) {
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100657 if (unclean_errno(errno))
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200658 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
659 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100660 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200661 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
662 chk->area);
663 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100664 }
665 else {
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100666 if (unclean_errno(errno)) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200667 chunk_printf(&trash, "%s%s", strerror(errno),
668 chk->area);
669 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100670 }
671 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200672 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100673 }
674 }
675
Willy Tarreau00149122017-10-04 18:05:01 +0200676 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200677 /* NOTE: this is reported after <fall> tries */
678 chunk_printf(chk, "No port available for the TCP connection");
679 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
680 }
681
Willy Tarreau00149122017-10-04 18:05:01 +0200682 if (!conn) {
683 /* connection allocation error before the connection was established */
684 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
685 }
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100686 else if (conn->flags & CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100687 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200688 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100689 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
690 else if (expired)
691 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200692
693 /*
694 * might be due to a server IP change.
695 * Let's trigger a DNS resolution if none are currently running.
696 */
Olivier Houchard0923fa42019-01-11 18:43:04 +0100697 if (check->server)
698 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200699
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100700 }
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100701 else if (conn->flags & CO_FL_WAIT_L6_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100702 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200703 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100704 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
705 else if (expired)
706 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
707 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200708 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100709 /* I/O error after connection was established and before we could diagnose */
710 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
711 }
712 else if (expired) {
713 /* connection established but expired check */
714 if (check->type == PR_O2_SSL3_CHK)
715 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
716 else /* HTTP, SMTP, ... */
717 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
718 }
719
720 return;
721}
722
Olivier Houchard5c110b92018-08-14 17:04:58 +0200723/* This function checks if any I/O is wanted, and if so, attempts to do so */
724static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200725{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200726 struct check *check = ctx;
727 struct conn_stream *cs = check->cs;
Olivier Houchard0923fa42019-01-11 18:43:04 +0100728 struct email_alertq *q = container_of(check, typeof(*q), check);
Olivier Houchardbc89ad82019-07-09 17:28:51 +0200729 int ret = 0;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200730
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100731 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchardbc89ad82019-07-09 17:28:51 +0200732 ret = wake_srv_chk(cs);
733 if (ret == 0 && !(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100734 if (check->server)
735 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
736 else
737 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200738 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100739 if (check->server)
740 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
741 else
742 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200743 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200744 return NULL;
745}
746
747/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100748 *
749 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200750 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
751 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200752 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200753static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200754{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200755 struct connection *conn = cs->conn;
756 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900757 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900758 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200759
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100760 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100761 goto out_wakeup;
762
Willy Tarreau20a18342013-12-05 00:31:46 +0100763 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200764 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100765 goto out_wakeup;
766 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100767
Willy Tarreau06559ac2013-12-05 01:53:08 +0100768 /* here, we know that the connection is established. That's enough for
769 * a pure TCP check.
770 */
771 if (!check->type)
772 goto out_wakeup;
773
Willy Tarreauc09572f2017-10-04 11:58:22 +0200774 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100775 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200776 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200777
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200778 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200779 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200780 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200781 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200782 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100783 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200784 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200785 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100786 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200787 goto out;
788 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100789 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200790
Willy Tarreau59093802019-09-05 18:24:46 +0200791 if (!b_data(&check->bo))
792 conn_xprt_stop_send(conn);
793
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100794 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
795 if (s->proxy->timeout.check) {
796 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
797 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200798 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200799 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100800
Willy Tarreau83749182007-04-15 20:56:27 +0200801 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200802 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200803 out:
804 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200805}
806
Willy Tarreaubaaee002006-06-26 02:48:02 +0200807/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200808 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200809 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900810 * set_server_check_status() to update check->status, check->duration
811 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200812
813 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
814 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
815 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
816 * response to an SSL HELLO (the principle is that this is enough to
817 * distinguish between an SSL server and a pure TCP relay). All other cases will
818 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
819 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100820 *
821 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200822 * via the out label.
823 *
824 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200825 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200826static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200827{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200828 struct connection *conn = cs->conn;
829 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900830 struct server *s = check->server;
831 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200832 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100833 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200834 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200835
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100836 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200837 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200838
Willy Tarreauc09572f2017-10-04 11:58:22 +0200839 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100840 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200841 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200842
Willy Tarreau83749182007-04-15 20:56:27 +0200843 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
844 * but the connection was closed on the remote end. Fortunately, recv still
845 * works correctly and we don't need to do the getsockopt() on linux.
846 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000847
848 /* Set buffer to point to the end of the data already read, and check
849 * that there is free space remaining. If the buffer is full, proceed
850 * with running the checks without attempting another socket read.
851 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000852
Willy Tarreau03938182010-03-17 21:52:07 +0100853 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000854
Olivier Houchard511efea2018-08-16 15:30:32 +0200855 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200856 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100857 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200858 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200859 /* Report network errors only if we got no other data. Otherwise
860 * we'll let the upper layers decide whether the response is OK
861 * or not. It is very common that an RST sent by the server is
862 * reported as an error just after the last data chunk.
863 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200864 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100865 goto out_wakeup;
866 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200867 }
868
Willy Tarreau0f0393f2019-09-24 10:43:03 +0200869 /* the rest of the code below expects the connection to be ready! */
Willy Tarreau911db9b2020-01-23 16:27:54 +0100870 if (conn->flags & CO_FL_WAIT_XPRT && !done)
Willy Tarreau0f0393f2019-09-24 10:43:03 +0200871 goto wait_more_data;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100872
Willy Tarreau03938182010-03-17 21:52:07 +0100873 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200874 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100875 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200876 if (b_data(&check->bi) < b_size(&check->bi))
877 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100878 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100880 done = 1; /* buffer full, don't wait for more data */
881 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200882
Nick Chalk57b1bf72010-03-16 15:50:46 +0000883 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900884 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200885 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200886 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100887 goto wait_more_data;
888
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100889 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200890 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
891 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
892 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
893 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
894 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
895 cut_crlf(b_head(&check->bi));
896 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200897
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100898 goto out_wakeup;
899 }
900
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200901 check->code = str2uic(b_head(&check->bi) + 9);
902 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200903
Willy Tarreaubd741542010-03-16 18:46:54 +0100904 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200905 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000906 /* 404 may be accepted as "stopping" only if the server was up */
907 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900908 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000909 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100910 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
911 /* Run content verification check... We know we have at least 13 chars */
912 if (!httpchk_expect(s, done))
913 goto wait_more_data;
914 }
915 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200916 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100917 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900918 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100919 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000920 else {
921 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900922 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000923 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200924 break;
925
926 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200927 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100928 goto wait_more_data;
929
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100930 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200931 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900932 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200933 else
Simon Horman4a741432013-02-23 15:35:38 +0900934 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200935 break;
936
937 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200938 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100939 goto wait_more_data;
940
Willy Tarreau5488a622019-08-06 16:26:31 +0200941 /* do not reset when closing, servers don't like this */
942 if (conn_ctrl_ready(cs->conn))
943 fdtab[cs->conn->handle.fd].linger_risk = 0;
944
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200945 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200946 if ((b_data(&check->bi) < strlen("000\r")) ||
947 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
948 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
949 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
950 cut_crlf(b_head(&check->bi));
951 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200952 goto out_wakeup;
953 }
954
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200955 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200956
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200957 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200958 cut_crlf(desc);
959
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100960 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200961 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900962 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200963 else
Simon Horman4a741432013-02-23 15:35:38 +0900964 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200965 break;
966
Simon Hormana2b9dad2013-02-12 10:45:54 +0900967 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100968 int status = HCHK_STATUS_CHECKED;
969 const char *hs = NULL; /* health status */
970 const char *as = NULL; /* admin status */
971 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200972 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100973 const char *err = NULL; /* first error to report */
974 const char *wrn = NULL; /* first warning to report */
975 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900976
Willy Tarreau81f5d942013-12-09 20:51:51 +0100977 /* We're getting an agent check response. The agent could
978 * have been disabled in the mean time with a long check
979 * still pending. It is important that we ignore the whole
980 * response.
981 */
982 if (!(check->server->agent.state & CHK_ST_ENABLED))
983 break;
984
985 /* The agent supports strings made of a single line ended by the
986 * first CR ('\r') or LF ('\n'). This line is composed of words
987 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
988 * line may optionally contained a description of a state change
989 * after a sharp ('#'), which is only considered if a health state
990 * is announced.
991 *
992 * Words may be composed of :
993 * - a numeric weight suffixed by the percent character ('%').
994 * - a health status among "up", "down", "stopped", and "fail".
995 * - an admin status among "ready", "drain", "maint".
996 *
997 * These words may appear in any order. If multiple words of the
998 * same category appear, the last one wins.
999 */
1000
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001001 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001002 while (*p && *p != '\n' && *p != '\r')
1003 p++;
1004
1005 if (!*p) {
1006 if (!done)
1007 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001008
Willy Tarreau9809b782013-12-11 21:40:11 +01001009 /* at least inform the admin that the agent is mis-behaving */
1010 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1011 break;
1012 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001013
Willy Tarreau9809b782013-12-11 21:40:11 +01001014 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001015 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001016
Willy Tarreau81f5d942013-12-09 20:51:51 +01001017 while (*cmd) {
1018 /* look for next word */
1019 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1020 cmd++;
1021 continue;
1022 }
Simon Horman671b6f02013-11-25 10:46:39 +09001023
Willy Tarreau81f5d942013-12-09 20:51:51 +01001024 if (*cmd == '#') {
1025 /* this is the beginning of a health status description,
1026 * skip the sharp and blanks.
1027 */
1028 cmd++;
1029 while (*cmd == '\t' || *cmd == ' ')
1030 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001031 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001032 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001033
1034 /* find the end of the word so that we have a null-terminated
1035 * word between <cmd> and <p>.
1036 */
1037 p = cmd + 1;
1038 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1039 p++;
1040 if (*p)
1041 *p++ = 0;
1042
1043 /* first, health statuses */
1044 if (strcasecmp(cmd, "up") == 0) {
1045 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001046 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001047 hs = cmd;
1048 }
1049 else if (strcasecmp(cmd, "down") == 0) {
1050 check->health = 0;
1051 status = HCHK_STATUS_L7STS;
1052 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001053 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001054 else if (strcasecmp(cmd, "stopped") == 0) {
1055 check->health = 0;
1056 status = HCHK_STATUS_L7STS;
1057 hs = cmd;
1058 }
1059 else if (strcasecmp(cmd, "fail") == 0) {
1060 check->health = 0;
1061 status = HCHK_STATUS_L7STS;
1062 hs = cmd;
1063 }
1064 /* admin statuses */
1065 else if (strcasecmp(cmd, "ready") == 0) {
1066 as = cmd;
1067 }
1068 else if (strcasecmp(cmd, "drain") == 0) {
1069 as = cmd;
1070 }
1071 else if (strcasecmp(cmd, "maint") == 0) {
1072 as = cmd;
1073 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001074 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001075 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1076 ps = cmd;
1077 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001078 /* try to parse a maxconn here */
1079 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1080 cs = cmd;
1081 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001082 else {
1083 /* keep a copy of the first error */
1084 if (!err)
1085 err = cmd;
1086 }
1087 /* skip to next word */
1088 cmd = p;
1089 }
1090 /* here, cmd points either to \0 or to the beginning of a
1091 * description. Skip possible leading spaces.
1092 */
1093 while (*cmd == ' ' || *cmd == '\n')
1094 cmd++;
1095
1096 /* First, update the admin status so that we avoid sending other
1097 * possibly useless warnings and can also update the health if
1098 * present after going back up.
1099 */
1100 if (as) {
1101 if (strcasecmp(as, "drain") == 0)
1102 srv_adm_set_drain(check->server);
1103 else if (strcasecmp(as, "maint") == 0)
1104 srv_adm_set_maint(check->server);
1105 else
1106 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001107 }
1108
Willy Tarreau81f5d942013-12-09 20:51:51 +01001109 /* now change weights */
1110 if (ps) {
1111 const char *msg;
1112
1113 msg = server_parse_weight_change_request(s, ps);
1114 if (!wrn || !*wrn)
1115 wrn = msg;
1116 }
1117
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001118 if (cs) {
1119 const char *msg;
1120
1121 cs += strlen("maxconn:");
1122
1123 msg = server_parse_maxconn_change_request(s, cs);
1124 if (!wrn || !*wrn)
1125 wrn = msg;
1126 }
1127
Willy Tarreau81f5d942013-12-09 20:51:51 +01001128 /* and finally health status */
1129 if (hs) {
1130 /* We'll report some of the warnings and errors we have
1131 * here. Down reports are critical, we leave them untouched.
1132 * Lack of report, or report of 'UP' leaves the room for
1133 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001134 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001135 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001136 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001137
1138 if (!*msg || status == HCHK_STATUS_L7OKD) {
1139 if (err && *err)
1140 msg = err;
1141 else if (wrn && *wrn)
1142 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001143 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001144
1145 t = get_trash_chunk();
1146 chunk_printf(t, "via agent : %s%s%s%s",
1147 hs, *msg ? " (" : "",
1148 msg, *msg ? ")" : "");
1149
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001150 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001151 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001152 else if (err && *err) {
1153 /* No status change but we'd like to report something odd.
1154 * Just report the current state and copy the message.
1155 */
1156 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001157 set_server_check_status(check, status/*check->status*/,
1158 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001159
Willy Tarreau81f5d942013-12-09 20:51:51 +01001160 }
1161 else if (wrn && *wrn) {
1162 /* No status change but we'd like to report something odd.
1163 * Just report the current state and copy the message.
1164 */
1165 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001166 set_server_check_status(check, status/*check->status*/,
1167 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001168 }
1169 else
1170 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001171 break;
1172 }
1173
Willy Tarreau1620ec32011-08-06 17:05:02 +02001174 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001175 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001176 goto wait_more_data;
1177
Willy Tarreau5488a622019-08-06 16:26:31 +02001178 /* do not reset when closing, servers don't like this */
1179 if (conn_ctrl_ready(cs->conn))
1180 fdtab[cs->conn->handle.fd].linger_risk = 0;
1181
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001182 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001183 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001184 }
1185 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001186 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1187 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001188 else
1189 desc = "PostgreSQL unknown error";
1190
Simon Horman4a741432013-02-23 15:35:38 +09001191 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001192 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001193 break;
1194
1195 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001196 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001197 goto wait_more_data;
1198
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001199 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001200 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001201 }
1202 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001203 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001204 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001205 break;
1206
1207 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001208 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001209 goto wait_more_data;
1210
Willy Tarreau5488a622019-08-06 16:26:31 +02001211 /* do not reset when closing, servers don't like this */
1212 if (conn_ctrl_ready(cs->conn))
1213 fdtab[cs->conn->handle.fd].linger_risk = 0;
1214
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001215 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001216 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001217 /* We set the MySQL Version in description for information purpose
1218 * FIXME : it can be cool to use MySQL Version for other purpose,
1219 * like mark as down old MySQL server.
1220 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001221 if (b_data(&check->bi) > 51) {
1222 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001223 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001224 }
1225 else {
1226 if (!done)
1227 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001228
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 /* it seems we have a OK packet but without a valid length,
1230 * it must be a protocol error
1231 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001232 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001233 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001234 }
1235 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001236 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001237 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001238 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239 }
1240 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001241 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1242 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1243 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001244
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001245 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001246 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001247 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001248 /* We have only one MySQL packet and it is a Handshake Initialization packet
1249 * but we need to have a second packet to know if it is alright
1250 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001251 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001252 goto wait_more_data;
1253 }
1254 else {
1255 /* We have only one packet and it is an Error packet,
1256 * an error message is attached, so we can display it
1257 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001258 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001259 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001260 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001261 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001262 } else if (b_data(&check->bi) > first_packet_len + 4) {
1263 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1264 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1265 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001266
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001267 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001268 /* We have 2 packets and that's good */
1269 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001270 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001271 /* No error packet */
1272 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001273 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001274 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001275 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001276 }
1277 else {
1278 /* An error message is attached in the Error packet
1279 * so we can display it ! :)
1280 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001281 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001282 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001283 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001284 }
1285 }
1286 }
1287 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001288 if (!done)
1289 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001290
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001291 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001292 * it must be a protocol error
1293 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001294 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001295 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001296 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001297 }
1298 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001299 break;
1300
1301 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001302 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001303 goto wait_more_data;
1304
1305 /* Check if the server speaks LDAP (ASN.1/BER)
1306 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1307 * http://tools.ietf.org/html/rfc4511
1308 */
1309
1310 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1311 * LDAPMessage: 0x30: SEQUENCE
1312 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001313 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001314 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001315 }
1316 else {
1317 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001318 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001319
1320 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1321 * messageID: 0x02 0x01 0x01: INTEGER 1
1322 * protocolOp: 0x61: bindResponse
1323 */
1324 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001325 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001326 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001327 goto out_wakeup;
1328 }
1329
1330 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001331 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001332
1333 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1334 * ldapResult: 0x0a 0x01: ENUMERATION
1335 */
1336 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001337 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001338 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001339 goto out_wakeup;
1340 }
1341
1342 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1343 * resultCode
1344 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001345 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001346 if (check->code) {
1347 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 +02001348 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001349 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001350 }
1351 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001352 break;
1353
Christopher Fauletba7bc162016-11-07 21:07:38 +01001354 case PR_O2_SPOP_CHK: {
1355 unsigned int framesz;
1356 char err[HCHK_DESC_LEN];
1357
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001358 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001359 goto wait_more_data;
1360
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001361 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001362 framesz = ntohl(framesz);
1363
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001364 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001365 goto wait_more_data;
1366
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001367 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001368 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1369 else
1370 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1371 break;
1372 }
1373
Willy Tarreau1620ec32011-08-06 17:05:02 +02001374 default:
Willy Tarreau4c1a2b32019-09-05 18:43:22 +02001375 /* good connection is enough for pure TCP check */
Willy Tarreau911db9b2020-01-23 16:27:54 +01001376 if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Willy Tarreau4c1a2b32019-09-05 18:43:22 +02001377 if (check->use_ssl)
1378 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
1379 else
1380 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
1381 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001382 break;
1383 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001384
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001385 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001386 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001387 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001388 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001389
Nick Chalk57b1bf72010-03-16 15:50:46 +00001390 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001391 *b_head(&check->bi) = '\0';
1392 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001393
Steven Davidovitz544d4812017-03-08 11:06:20 -08001394 /* Close the connection... We still attempt to nicely close if,
1395 * for instance, SSL needs to send a "close notify." Later, we perform
1396 * a hard close and reset the connection if some data are pending,
1397 * otherwise we end up with many TIME_WAITs and eat all the source port
1398 * range quickly. To avoid sending RSTs all the time, we first try to
1399 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001400 */
Olivier Houchard6c7e96a2019-07-02 16:35:18 +02001401 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1402 * connection, to make sure cs_shutw() will not lead to a shutdown()
1403 * that would provoke TIME_WAITs.
1404 */
1405 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001406 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001407
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001408 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001409 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001410 conn->flags |= CO_FL_ERROR;
1411
Willy Tarreaufdccded2008-08-29 18:19:04 +02001412 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001413out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001414 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001415
1416 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001417 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001418 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001419}
1420
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001421/*
1422 * This function is used only for server health-checks. It handles connection
1423 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001424 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1425 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001426 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001427static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001428{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001429 struct connection *conn = cs->conn;
1430 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001431 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001432 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001433
Olivier Houchard0923fa42019-01-11 18:43:04 +01001434 if (check->server)
1435 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1436 else
1437 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001438
Willy Tarreauc09572f2017-10-04 11:58:22 +02001439 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001440 if (check->type == PR_O2_TCPCHK_CHK) {
1441 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001442 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001443 conn = cs->conn;
Willy Tarreauc5940392019-09-05 17:38:40 +02001444 } else {
1445 if (!(check->wait_list.events & SUB_RETRY_SEND))
1446 __event_srv_chk_w(cs);
1447 if (!(check->wait_list.events & SUB_RETRY_RECV))
1448 __event_srv_chk_r(cs);
1449 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001450
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001451 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001452 /* We may get error reports bypassing the I/O handlers, typically
1453 * the case when sending a pure TCP check which fails, then the I/O
1454 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001455 * main processing task so let's simply wake it up. If we get here,
1456 * we expect errno to still be valid.
1457 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001458 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001459 task_wakeup(check->task, TASK_WOKEN_IO);
1460 }
Willy Tarreau911db9b2020-01-23 16:27:54 +01001461 else if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001462 /* we may get here if only a connection probe was required : we
1463 * don't have any data to send nor anything expected in response,
1464 * so the completion of the connection establishment is enough.
1465 */
1466 task_wakeup(check->task, TASK_WOKEN_IO);
1467 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001468
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001469 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001470 /* Check complete or aborted. If connection not yet closed do it
1471 * now and wake the check task up to be sure the result is
1472 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001473 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001474 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001475 ret = -1;
Olivier Houchardf4949572019-07-02 17:42:22 +02001476 /* We may have been scheduled to run, and the
1477 * I/O handler expects to have a cs, so remove
1478 * the tasklet
1479 */
1480 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001481 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001482 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001483
Olivier Houchard0923fa42019-01-11 18:43:04 +01001484 if (check->server)
1485 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1486 else
1487 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001488
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001489 /* if a connection got replaced, we must absolutely prevent the connection
1490 * handler from touching its fd, and perform the FD polling updates ourselves
1491 */
1492 if (ret < 0)
1493 conn_cond_update_polling(conn);
1494
1495 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001496}
1497
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001498struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001499 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001500 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001501};
1502
Willy Tarreaubaaee002006-06-26 02:48:02 +02001503/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001504 * updates the server's weight during a warmup stage. Once the final weight is
1505 * reached, the task automatically stops. Note that any server status change
1506 * must have updated s->last_change accordingly.
1507 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001508static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001509{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001510 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001511
1512 /* by default, plan on stopping the task */
1513 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001514 if ((s->next_admin & SRV_ADMF_MAINT) ||
1515 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001516 return t;
1517
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001518 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1519
Willy Tarreau892337c2014-05-13 23:41:20 +02001520 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001521 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001522
1523 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001524 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001525
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001526 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1527
Willy Tarreau2e993902011-10-31 11:53:20 +01001528 /* get back there in 1 second or 1/20th of the slowstart interval,
1529 * whichever is greater, resulting in small 5% steps.
1530 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001531 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001532 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1533 return t;
1534}
1535
Willy Tarreau894c6422017-10-04 15:58:52 +02001536/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1537 * none was found.
1538 */
1539static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1540{
1541 struct tcpcheck_rule *r;
1542
1543 list_for_each_entry(r, list, list) {
1544 if (r->action != TCPCHK_ACT_COMMENT)
1545 return r;
1546 }
1547 return NULL;
1548}
1549
Willy Tarreau2e993902011-10-31 11:53:20 +01001550/*
Simon Horman98637e52014-06-20 12:30:16 +09001551 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001552 *
1553 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001554 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1555 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1556 * - SF_ERR_SRVTO if there are no more servers
1557 * - SF_ERR_SRVCL if the connection was refused by the server
1558 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1559 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1560 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001561 * - 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 +01001562 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001563 * Note that we try to prevent the network stack from sending the ACK during the
1564 * connect() when a pure TCP check is used (without PROXY protocol).
1565 */
Simon Horman98637e52014-06-20 12:30:16 +09001566static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001567{
1568 struct check *check = t->context;
1569 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001570 struct conn_stream *cs = check->cs;
1571 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001572 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001573 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001574 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001575 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001576
Willy Tarreau00149122017-10-04 18:05:01 +02001577 /* we cannot have a connection here */
1578 if (conn)
1579 return SF_ERR_INTERNAL;
1580
Simon Hormanb00d17a2014-06-13 16:18:16 +09001581 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001582 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001583 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001584 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1585 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001586
1587 /* prepare the check buffer.
1588 * This should not be used if check is the secondary agent check
1589 * of a server as s->proxy->check_req will relate to the
1590 * configuration of the primary check. Similarly, tcp-check uses
1591 * its own strings.
1592 */
1593 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001594 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001595
1596 /* we want to check if this host replies to HTTP or SSLv3 requests
1597 * so we'll send the request, and won't wake the checker up now.
1598 */
1599 if ((check->type) == PR_O2_SSL3_CHK) {
1600 /* SSL requires that we put Unix time in the request */
1601 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001602 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001603 }
1604 else if ((check->type) == PR_O2_HTTP_CHK) {
1605 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001606 b_putblk(&check->bo, trash.area,
1607 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001608 /* prevent HTTP keep-alive when "http-check expect" is used */
1609 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001610 b_putist(&check->bo, ist("Connection: close\r\n"));
1611 b_putist(&check->bo, ist("\r\n"));
1612 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001613 }
1614 }
1615
James Brown55f9ff12015-10-21 18:19:05 -07001616 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001617 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001618 }
1619
Willy Tarreauf411cce2017-10-04 16:21:19 +02001620 /* for tcp-checks, the initial connection setup is handled separately as
1621 * it may be sent to a specific port and not to the server's.
1622 */
1623 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1624 tcpcheck_main(check);
1625 return SF_ERR_UP;
1626 }
1627
Simon Hormanb00d17a2014-06-13 16:18:16 +09001628 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001629 cs = check->cs = cs_new(NULL);
1630 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001631 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001632 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001633 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001634 check->wait_list.events = 0;
Olivier Houchardff1e9f32019-09-20 17:18:35 +02001635 tasklet_set_tid(check->wait_list.tasklet, tid);
1636
Simon Hormanb00d17a2014-06-13 16:18:16 +09001637
Willy Tarreauca79f592019-07-17 19:04:47 +02001638 if (!sockaddr_alloc(&conn->dst))
1639 return SF_ERR_RESOURCE;
1640
Simon Horman41f58762015-01-30 11:22:56 +09001641 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001642 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001643 *conn->dst = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001644 }
1645 else {
1646 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001647 *conn->dst = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001648 }
1649
Alexander Liu2a54bb72019-05-22 19:44:48 +08001650 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1651 conn->send_proxy_ofs = 1;
1652 conn->flags |= CO_FL_SOCKS4;
1653 }
1654
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001655 proto = protocol_by_family(conn->dst->ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001656 conn->target = &s->obj_type;
1657
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001658 if ((conn->dst->ss_family == AF_INET) || (conn->dst->ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001659 int i = 0;
1660
1661 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001662 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001663 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001664
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001665 set_host_port(conn->dst, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001666 }
1667
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001668 /* no client address */
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001669
Willy Tarreaube373152018-09-06 11:45:30 +02001670 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001671 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1672 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001673 cs_attach(cs, check, &check_conn_cb);
1674
Willy Tarreauf3d34822014-12-08 12:11:28 +01001675 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001676 if (check->type != 0)
1677 connflags |= CONNECT_HAS_DATA;
1678 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1679 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1680 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001681
Willy Tarreaue7dff022015-04-03 01:14:29 +02001682 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001683 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001684 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001685
Willy Tarreau16257f62017-11-02 15:45:00 +01001686
Olivier Houchard9130a962017-10-17 17:33:43 +02001687#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001688 if (ret == SF_ERR_NONE) {
1689 if (s->check.sni)
1690 ssl_sock_set_servername(conn, s->check.sni);
1691 if (s->check.alpn_str)
1692 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1693 s->check.alpn_len);
1694 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001695#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001696 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001697 conn->send_proxy_ofs = 1;
1698 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchard37d78972019-12-30 15:13:42 +01001699 }
1700 if (conn->flags & (CO_FL_SEND_PROXY | CO_FL_SOCKS4) &&
1701 conn_ctrl_ready(conn)) {
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001702 if (xprt_add_hs(conn) < 0)
1703 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001704 }
1705
1706 return ret;
1707}
1708
Simon Horman98637e52014-06-20 12:30:16 +09001709static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001710static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001711__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001712
1713void block_sigchld(void)
1714{
1715 sigset_t set;
1716 sigemptyset(&set);
1717 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001718 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001719}
1720
1721void unblock_sigchld(void)
1722{
1723 sigset_t set;
1724 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001725 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001726 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001727}
1728
Simon Horman98637e52014-06-20 12:30:16 +09001729static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1730{
1731 struct pid_list *elem;
1732 struct check *check = t->context;
1733
Willy Tarreaubafbe012017-11-24 17:34:44 +01001734 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001735 if (!elem)
1736 return NULL;
1737 elem->pid = pid;
1738 elem->t = t;
1739 elem->exited = 0;
1740 check->curpid = elem;
1741 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001742
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001743 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001744 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001745 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001746
Simon Horman98637e52014-06-20 12:30:16 +09001747 return elem;
1748}
1749
Simon Horman98637e52014-06-20 12:30:16 +09001750static void pid_list_del(struct pid_list *elem)
1751{
1752 struct check *check;
1753
1754 if (!elem)
1755 return;
1756
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001757 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001758 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001759 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001760
Simon Horman98637e52014-06-20 12:30:16 +09001761 if (!elem->exited)
1762 kill(elem->pid, SIGTERM);
1763
1764 check = elem->t->context;
1765 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001766 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001767}
1768
1769/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1770static void pid_list_expire(pid_t pid, int status)
1771{
1772 struct pid_list *elem;
1773
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001774 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001775 list_for_each_entry(elem, &pid_list, list) {
1776 if (elem->pid == pid) {
1777 elem->t->expire = now_ms;
1778 elem->status = status;
1779 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001780 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001781 break;
Simon Horman98637e52014-06-20 12:30:16 +09001782 }
1783 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001784 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001785}
1786
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001787static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001788{
1789 pid_t pid;
1790 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001791
Simon Horman98637e52014-06-20 12:30:16 +09001792 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1793 pid_list_expire(pid, status);
1794}
1795
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001796static int init_pid_list(void)
1797{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001798 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001799 /* Nothing to do */
1800 return 0;
1801
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001802 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001803 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1804 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001805 return 1;
1806 }
1807
Willy Tarreaubafbe012017-11-24 17:34:44 +01001808 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1809 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001810 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1811 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001812 return 1;
1813 }
1814
1815 return 0;
1816}
1817
Cyril Bontéac92a062014-12-27 22:28:38 +01001818/* helper macro to set an environment variable and jump to a specific label on failure. */
1819#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001820
1821/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001822 * helper function to allocate enough memory to store an environment variable.
1823 * It will also check that the environment variable is updatable, and silently
1824 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001825 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001826static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001827{
1828 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001829 char *envname;
1830 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001831
Cyril Bontéac92a062014-12-27 22:28:38 +01001832 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001833 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001834 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001835 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001836
1837 envname = extcheck_envs[idx].name;
1838 vmaxlen = extcheck_envs[idx].vmaxlen;
1839
1840 /* Check if the environment variable is already set, and silently reject
1841 * the update if this one is not updatable. */
1842 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1843 return 0;
1844
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001845 /* Instead of sending NOT_USED, sending an empty value is preferable */
1846 if (strcmp(value, "NOT_USED") == 0) {
1847 value = "";
1848 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001849
1850 len = strlen(envname) + 1;
1851 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1852 len += strlen(value);
1853 else
1854 len += vmaxlen;
1855
1856 if (!check->envp[idx])
1857 check->envp[idx] = malloc(len + 1);
1858
1859 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001860 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001861 return 1;
1862 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001863 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001864 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001865 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001866 return 1;
1867 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001868 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001869 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001870 return 1;
1871 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001872 return 0;
1873}
Simon Horman98637e52014-06-20 12:30:16 +09001874
1875static int prepare_external_check(struct check *check)
1876{
1877 struct server *s = check->server;
1878 struct proxy *px = s->proxy;
1879 struct listener *listener = NULL, *l;
1880 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001881 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001882 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001883
1884 list_for_each_entry(l, &px->conf.listeners, by_fe)
1885 /* Use the first INET, INET6 or UNIX listener */
1886 if (l->addr.ss_family == AF_INET ||
1887 l->addr.ss_family == AF_INET6 ||
1888 l->addr.ss_family == AF_UNIX) {
1889 listener = l;
1890 break;
1891 }
1892
Simon Horman98637e52014-06-20 12:30:16 +09001893 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001894 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1895 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001896 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001897 goto err;
1898 }
Simon Horman98637e52014-06-20 12:30:16 +09001899
Cyril Bontéac92a062014-12-27 22:28:38 +01001900 check->argv = calloc(6, sizeof(char *));
1901 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001902 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001903 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001904 }
Simon Horman98637e52014-06-20 12:30:16 +09001905
1906 check->argv[0] = px->check_command;
1907
Cyril Bonté777be862014-12-02 21:21:35 +01001908 if (!listener) {
1909 check->argv[1] = strdup("NOT_USED");
1910 check->argv[2] = strdup("NOT_USED");
1911 }
1912 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001913 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001914 addr_to_str(&listener->addr, buf, sizeof(buf));
1915 check->argv[1] = strdup(buf);
1916 port_to_str(&listener->addr, buf, sizeof(buf));
1917 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001918 }
1919 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001920 const struct sockaddr_un *un;
1921
1922 un = (struct sockaddr_un *)&listener->addr;
1923 check->argv[1] = strdup(un->sun_path);
1924 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001925 }
1926 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001927 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001928 goto err;
1929 }
1930
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001931 addr_to_str(&s->addr, buf, sizeof(buf));
1932 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001933
1934 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1935 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1936 else
1937 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001938 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001939
Cyril Bontéac92a062014-12-27 22:28:38 +01001940 for (i = 0; i < 5; i++) {
1941 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001942 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001943 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001944 }
1945 }
Simon Horman98637e52014-06-20 12:30:16 +09001946
Cyril Bontéac92a062014-12-27 22:28:38 +01001947 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001948 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001949 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1950 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1951 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1952 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001953 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001954 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1955 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1956 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1957 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1958 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1959 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1960
1961 /* Ensure that we don't leave any hole in check->envp */
1962 for (i = 0; i < EXTCHK_SIZE; i++)
1963 if (!check->envp[i])
1964 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001965
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001966 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001967err:
1968 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001969 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001970 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001971 free(check->envp);
1972 check->envp = NULL;
1973 }
1974
1975 if (check->argv) {
1976 for (i = 1; i < 5; i++)
1977 free(check->argv[i]);
1978 free(check->argv);
1979 check->argv = NULL;
1980 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001981 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001982}
1983
Simon Hormanb00d17a2014-06-13 16:18:16 +09001984/*
Simon Horman98637e52014-06-20 12:30:16 +09001985 * establish a server health-check that makes use of a process.
1986 *
1987 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001988 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001989 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001990 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001991 *
1992 * Blocks and then unblocks SIGCHLD
1993 */
1994static int connect_proc_chk(struct task *t)
1995{
Cyril Bontéac92a062014-12-27 22:28:38 +01001996 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001997 struct check *check = t->context;
1998 struct server *s = check->server;
1999 struct proxy *px = s->proxy;
2000 int status;
2001 pid_t pid;
2002
Willy Tarreaue7dff022015-04-03 01:14:29 +02002003 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002004
2005 block_sigchld();
2006
2007 pid = fork();
2008 if (pid < 0) {
Willy Tarreaud96f1122019-12-03 07:07:36 +01002009 ha_alert("Failed to fork process for external health check%s: %s. Aborting.\n",
2010 (global.tune.options & GTUNE_INSECURE_FORK) ?
2011 "" : " (likely caused by missing 'insecure-fork-wanted')",
Christopher Faulet767a84b2017-11-24 16:50:31 +01002012 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002013 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2014 goto out;
2015 }
2016 if (pid == 0) {
2017 /* Child */
2018 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002019 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002020 int fd;
2021
2022 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2023 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2024
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002025 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002026
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002027 /* restore the initial FD limits */
2028 limit.rlim_cur = rlim_fd_cur_at_boot;
2029 limit.rlim_max = rlim_fd_max_at_boot;
2030 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2031 getrlimit(RLIMIT_NOFILE, &limit);
2032 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2033 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2034 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2035 }
2036
Simon Horman98637e52014-06-20 12:30:16 +09002037 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01002038 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Willy Tarreau2df8cad2019-07-01 07:51:29 +02002039 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002040 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002041 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2042 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002043 exit(-1);
2044 }
2045
2046 /* Parent */
2047 if (check->result == CHK_RES_UNKNOWN) {
2048 if (pid_list_add(pid, t) != NULL) {
2049 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2050
2051 if (px->timeout.check && px->timeout.connect) {
2052 int t_con = tick_add(now_ms, px->timeout.connect);
2053 t->expire = tick_first(t->expire, t_con);
2054 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002055 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002056 goto out;
2057 }
2058 else {
2059 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2060 }
2061 kill(pid, SIGTERM); /* process creation error */
2062 }
2063 else
2064 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2065
2066out:
2067 unblock_sigchld();
2068 return status;
2069}
2070
2071/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002072 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002073 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002074 *
2075 * Please do NOT place any return statement in this function and only leave
2076 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002077 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002078static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002079{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002080 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002081 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002082 int rv;
2083 int ret;
2084 int expired = tick_is_expired(t->expire, now_ms);
2085
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002086 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002087 if (!(check->state & CHK_ST_INPROGRESS)) {
2088 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002089 if (!expired) /* woke up too early */
2090 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002091
2092 /* we don't send any health-checks when the proxy is
2093 * stopped, the server should not be checked or the check
2094 * is disabled.
2095 */
2096 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2097 s->proxy->state == PR_STSTOPPED)
2098 goto reschedule;
2099
2100 /* we'll initiate a new check */
2101 set_server_check_status(check, HCHK_STATUS_START, NULL);
2102
2103 check->state |= CHK_ST_INPROGRESS;
2104
Simon Hormandbf70192015-01-30 11:22:53 +09002105 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002106 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002107 /* the process was forked, we allow up to min(inter,
2108 * timeout.connect) for it to report its status, but
2109 * only when timeout.check is set as it may be to short
2110 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002111 */
2112 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2113
2114 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2115 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2116 t->expire = tick_first(t->expire, t_con);
2117 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002118 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002119 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002120 }
2121
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002122 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002123
2124 check->state &= ~CHK_ST_INPROGRESS;
2125 check_notify_failure(check);
2126
2127 /* we allow up to min(inter, timeout.connect) for a connection
2128 * to establish but only when timeout.check is set
2129 * as it may be to short for a full check otherwise
2130 */
2131 while (tick_is_expired(t->expire, now_ms)) {
2132 int t_con;
2133
2134 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2135 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2136
2137 if (s->proxy->timeout.check)
2138 t->expire = tick_first(t->expire, t_con);
2139 }
2140 }
2141 else {
2142 /* there was a test running.
2143 * First, let's check whether there was an uncaught error,
2144 * which can happen on connect timeout or error.
2145 */
2146 if (check->result == CHK_RES_UNKNOWN) {
2147 /* good connection is enough for pure TCP check */
2148 struct pid_list *elem = check->curpid;
2149 int status = HCHK_STATUS_UNKNOWN;
2150
2151 if (elem->exited) {
2152 status = elem->status; /* Save in case the process exits between use below */
2153 if (!WIFEXITED(status))
2154 check->code = -1;
2155 else
2156 check->code = WEXITSTATUS(status);
2157 if (!WIFEXITED(status) || WEXITSTATUS(status))
2158 status = HCHK_STATUS_PROCERR;
2159 else
2160 status = HCHK_STATUS_PROCOK;
2161 } else if (expired) {
2162 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002163 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002164 kill(elem->pid, SIGTERM);
2165 }
2166 set_server_check_status(check, status, NULL);
2167 }
2168
2169 if (check->result == CHK_RES_FAILED) {
2170 /* a failure or timeout detected */
2171 check_notify_failure(check);
2172 }
2173 else if (check->result == CHK_RES_CONDPASS) {
2174 /* check is OK but asks for stopping mode */
2175 check_notify_stopping(check);
2176 }
2177 else if (check->result == CHK_RES_PASSED) {
2178 /* a success was detected */
2179 check_notify_success(check);
2180 }
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002181 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002182 check->state &= ~CHK_ST_INPROGRESS;
2183
2184 pid_list_del(check->curpid);
2185
2186 rv = 0;
2187 if (global.spread_checks > 0) {
2188 rv = srv_getinter(check) * global.spread_checks / 100;
2189 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2190 }
2191 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2192 }
2193
2194 reschedule:
2195 while (tick_is_expired(t->expire, now_ms))
2196 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002197
2198 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002199 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002200 return t;
2201}
2202
2203/*
2204 * manages a server health-check that uses a connection. Returns
2205 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002206 *
2207 * Please do NOT place any return statement in this function and only leave
2208 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002209 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002210static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002211{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002212 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002213 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002214 struct conn_stream *cs = check->cs;
2215 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002216 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002217 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002218 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002219
Olivier Houchard0923fa42019-01-11 18:43:04 +01002220 if (check->server)
2221 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002222 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002223 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002224 if (!expired) /* woke up too early */
2225 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002226
Simon Horman671b6f02013-11-25 10:46:39 +09002227 /* we don't send any health-checks when the proxy is
2228 * stopped, the server should not be checked or the check
2229 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002230 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002231 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002232 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002233 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002234
2235 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002236 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002237
Willy Tarreau2c115e52013-12-11 19:41:16 +01002238 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002239 b_reset(&check->bi);
2240 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002241
Olivier Houchardaebeff72019-11-29 16:18:51 +01002242 task_set_affinity(t, tid_bit);
Simon Hormandbf70192015-01-30 11:22:53 +09002243 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002244 cs = check->cs;
2245 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002246
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002247 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002248 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002249 goto out_unlock;
2250
Willy Tarreaue7dff022015-04-03 01:14:29 +02002251 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002252 /* 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 */
Simon Horman4a741432013-02-23 15:35:38 +09002256 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002257
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002258 if (proxy->timeout.check && proxy->timeout.connect) {
2259 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002260 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002261 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002262
Willy Tarreaucc705a62019-09-05 17:51:30 +02002263 if (check->type) {
2264 /* send the request if we have one. We avoid receiving
2265 * if not connected, unless we didn't subscribe for
2266 * sending since otherwise we won't be woken up.
2267 */
2268 __event_srv_chk_w(cs);
Willy Tarreau911db9b2020-01-23 16:27:54 +01002269 if (!(conn->flags & CO_FL_WAIT_XPRT) ||
Willy Tarreauc5940392019-09-05 17:38:40 +02002270 !(check->wait_list.events & SUB_RETRY_SEND))
2271 __event_srv_chk_r(cs);
Willy Tarreaucc705a62019-09-05 17:51:30 +02002272 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002273
Willy Tarreau5a78f362012-11-23 12:47:05 +01002274 goto reschedule;
2275
Willy Tarreaue7dff022015-04-03 01:14:29 +02002276 case SF_ERR_SRVTO: /* ETIMEDOUT */
2277 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002278 if (conn)
2279 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002280 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002281 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002282 /* should share same code than cases below */
2283 case SF_ERR_CHK_PORT:
2284 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002285 case SF_ERR_PRXCOND:
2286 case SF_ERR_RESOURCE:
2287 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002288 if (conn)
2289 conn->flags |= CO_FL_ERROR;
2290 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002291 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002292 }
2293
Willy Tarreau5a78f362012-11-23 12:47:05 +01002294 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchardaebeff72019-11-29 16:18:51 +01002295 task_set_affinity(t, MAX_THREADS_MASK);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002296 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002297 if (check->wait_list.events)
2298 cs->conn->xprt->unsubscribe(cs->conn,
2299 cs->conn->xprt_ctx,
2300 check->wait_list.events,
2301 &check->wait_list);
2302 /* We may have been scheduled to run, and the
2303 * I/O handler expects to have a cs, so remove
2304 * the tasklet
2305 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002306 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002307 cs_destroy(cs);
2308 cs = check->cs = NULL;
2309 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002310 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002311
Willy Tarreau2c115e52013-12-11 19:41:16 +01002312 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002313 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002314
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002315 /* we allow up to min(inter, timeout.connect) for a connection
2316 * to establish but only when timeout.check is set
2317 * as it may be to short for a full check otherwise
2318 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002319 while (tick_is_expired(t->expire, now_ms)) {
2320 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002321
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002322 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002323 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002324
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002325 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002326 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002327 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002328 }
2329 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002330 /* there was a test running.
2331 * First, let's check whether there was an uncaught error,
2332 * which can happen on connect timeout or error.
2333 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002334 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002335 /* good connection is enough for pure TCP check */
Willy Tarreau911db9b2020-01-23 16:27:54 +01002336 if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002337 if (check->use_ssl)
2338 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002339 else
Simon Horman4a741432013-02-23 15:35:38 +09002340 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002341 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002342 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002343 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002344 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002345 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002346 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002347 }
2348
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002349 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002350 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002351 /* The check was aborted and the connection was not yet closed.
2352 * This can happen upon timeout, or when an external event such
2353 * as a failed response coupled with "observe layer7" caused the
2354 * server state to be suddenly changed.
2355 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002356 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002357 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002358 }
2359
Willy Tarreauac59f362017-10-08 11:10:19 +02002360 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002361 if (check->wait_list.events)
2362 cs->conn->xprt->unsubscribe(cs->conn,
2363 cs->conn->xprt_ctx,
2364 check->wait_list.events,
2365 &check->wait_list);
2366 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002367 * I/O handler expects to have a cs, so remove
2368 * the tasklet
2369 */
2370 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002371 cs_destroy(cs);
2372 cs = check->cs = NULL;
2373 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002374 }
2375
Olivier Houchard0923fa42019-01-11 18:43:04 +01002376 if (check->server) {
2377 if (check->result == CHK_RES_FAILED) {
2378 /* a failure or timeout detected */
2379 check_notify_failure(check);
2380 }
2381 else if (check->result == CHK_RES_CONDPASS) {
2382 /* check is OK but asks for stopping mode */
2383 check_notify_stopping(check);
2384 }
2385 else if (check->result == CHK_RES_PASSED) {
2386 /* a success was detected */
2387 check_notify_success(check);
2388 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002389 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002390 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002391 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002392
Olivier Houchard0923fa42019-01-11 18:43:04 +01002393 if (check->server) {
2394 rv = 0;
2395 if (global.spread_checks > 0) {
2396 rv = srv_getinter(check) * global.spread_checks / 100;
2397 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2398 }
2399 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002400 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002401 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002402
2403 reschedule:
2404 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002405 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002406 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002407 if (check->server)
2408 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002409 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002410}
2411
Simon Horman98637e52014-06-20 12:30:16 +09002412/*
2413 * manages a server health-check. Returns
2414 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2415 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002416static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002417{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002418 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002419
2420 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002421 return process_chk_proc(t, context, state);
2422 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002423
Simon Horman98637e52014-06-20 12:30:16 +09002424}
2425
Simon Horman5c942422013-11-25 10:46:32 +09002426static int start_check_task(struct check *check, int mininter,
2427 int nbcheck, int srvpos)
2428{
2429 struct task *t;
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002430 unsigned long thread_mask = MAX_THREADS_MASK;
2431
2432 if (check->type == PR_O2_EXT_CHK)
2433 thread_mask = 1;
2434
Simon Horman5c942422013-11-25 10:46:32 +09002435 /* task for the check */
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002436 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002437 ha_alert("Starting [%s:%s] check: out of memory.\n",
2438 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002439 return 0;
2440 }
2441
2442 check->task = t;
2443 t->process = process_chk;
2444 t->context = check;
2445
Willy Tarreau1746eec2014-04-25 10:46:47 +02002446 if (mininter < srv_getinter(check))
2447 mininter = srv_getinter(check);
2448
2449 if (global.max_spread_checks && mininter > global.max_spread_checks)
2450 mininter = global.max_spread_checks;
2451
Simon Horman5c942422013-11-25 10:46:32 +09002452 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002453 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002454 check->start = now;
2455 task_queue(t);
2456
2457 return 1;
2458}
2459
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002460/*
2461 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002462 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002463 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002464static int start_checks()
2465{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002466
2467 struct proxy *px;
2468 struct server *s;
2469 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002470 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002471
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002472 /* 1- count the checkers to run simultaneously.
2473 * We also determine the minimum interval among all of those which
2474 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2475 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002476 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002477 * too short an interval for all others.
2478 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002479 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002480 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002481 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002482 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002483 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002484 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002485 }
2486 /* We need a warmup task that will be called when the server
2487 * state switches from down to up.
2488 */
2489 s->warmup = t;
2490 t->process = server_warmup;
2491 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002492 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002493 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002494 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 +01002495 }
2496
Willy Tarreaud8514a22013-12-11 21:10:14 +01002497 if (s->check.state & CHK_ST_CONFIGURED) {
2498 nbcheck++;
2499 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2500 (!mininter || mininter > srv_getinter(&s->check)))
2501 mininter = srv_getinter(&s->check);
2502 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002503
Willy Tarreaud8514a22013-12-11 21:10:14 +01002504 if (s->agent.state & CHK_ST_CONFIGURED) {
2505 nbcheck++;
2506 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2507 (!mininter || mininter > srv_getinter(&s->agent)))
2508 mininter = srv_getinter(&s->agent);
2509 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002510 }
2511 }
2512
Simon Horman4a741432013-02-23 15:35:38 +09002513 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002514 return 0;
2515
2516 srand((unsigned)time(NULL));
2517
2518 /*
2519 * 2- start them as far as possible from each others. For this, we will
2520 * start them after their interval set to the min interval divided by
2521 * the number of servers, weighted by the server's position in the list.
2522 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002523 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002524 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2525 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002526 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002527 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002528 }
2529 }
2530
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002531 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002532 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002533 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002534 if (s->check.type == PR_O2_EXT_CHK) {
2535 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002536 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002537 }
Simon Hormand60d6912013-11-25 10:46:36 +09002538 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002539 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002540 srvpos++;
2541 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002542
Simon Hormand60d6912013-11-25 10:46:36 +09002543 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002544 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002545 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002546 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002547 }
2548 srvpos++;
2549 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002550 }
2551 }
2552 return 0;
2553}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002554
2555/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002556 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002557 * The buffer MUST be terminated by a null byte before calling this function.
2558 * Sets server status appropriately. The caller is responsible for ensuring
2559 * that the buffer contains at least 13 characters. If <done> is zero, we may
2560 * return 0 to indicate that data is required to decide of a match.
2561 */
2562static int httpchk_expect(struct server *s, int done)
2563{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002564 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002565 char status_code[] = "000";
2566 char *contentptr;
2567 int crlf;
2568 int ret;
2569
2570 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2571 case PR_O2_EXP_STS:
2572 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002573 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2574 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002575
2576 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2577 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2578 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002579 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002580
2581 /* we necessarily have the response, so there are no partial failures */
2582 if (s->proxy->options2 & PR_O2_EXP_INV)
2583 ret = !ret;
2584
Simon Horman4a741432013-02-23 15:35:38 +09002585 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002586 break;
2587
2588 case PR_O2_EXP_STR:
2589 case PR_O2_EXP_RSTR:
2590 /* very simple response parser: ignore CR and only count consecutive LFs,
2591 * stop with contentptr pointing to first char after the double CRLF or
2592 * to '\0' if crlf < 2.
2593 */
2594 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002595 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002596 if (crlf >= 2)
2597 break;
2598 if (*contentptr == '\r')
2599 continue;
2600 else if (*contentptr == '\n')
2601 crlf++;
2602 else
2603 crlf = 0;
2604 }
2605
2606 /* Check that response contains a body... */
2607 if (crlf < 2) {
2608 if (!done)
2609 return 0;
2610
Simon Horman4a741432013-02-23 15:35:38 +09002611 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002612 "HTTP content check could not find a response body");
2613 return 1;
2614 }
2615
2616 /* Check that response body is not empty... */
2617 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002618 if (!done)
2619 return 0;
2620
Simon Horman4a741432013-02-23 15:35:38 +09002621 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002622 "HTTP content check found empty response body");
2623 return 1;
2624 }
2625
2626 /* Check the response content against the supplied string
2627 * or regex... */
2628 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2629 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2630 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002631 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002632
2633 /* if we don't match, we may need to wait more */
2634 if (!ret && !done)
2635 return 0;
2636
2637 if (ret) {
2638 /* content matched */
2639 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002640 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002641 "HTTP check matched unwanted content");
2642 else
Simon Horman4a741432013-02-23 15:35:38 +09002643 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002644 "HTTP content check matched");
2645 }
2646 else {
2647 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002648 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002649 "HTTP check did not match unwanted content");
2650 else
Simon Horman4a741432013-02-23 15:35:38 +09002651 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002652 "HTTP content check did not match");
2653 }
2654 break;
2655 }
2656 return 1;
2657}
2658
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002659/*
2660 * return the id of a step in a send/expect session
2661 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002662static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663{
2664 struct tcpcheck_rule *cur = NULL, *next = NULL;
2665 int i = 0;
2666
Willy Tarreau213c6782014-10-02 14:51:02 +02002667 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002668 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002669 return 0;
2670
Simon Hormane16c1b32015-01-30 11:22:57 +09002671 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002672
2673 /* no step => first step */
2674 if (cur == NULL)
2675 return 1;
2676
2677 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002678 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002679 if (next->list.p == &cur->list)
2680 break;
2681 ++i;
2682 }
2683
2684 return i;
2685}
2686
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002687/*
2688 * return the latest known comment before (including) the given stepid
2689 * returns NULL if no comment found
2690 */
2691static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2692{
2693 struct tcpcheck_rule *cur = NULL;
2694 char *ret = NULL;
2695 int i = 0;
2696
2697 /* not even started anything yet, return latest comment found before any action */
2698 if (!check->current_step) {
2699 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2700 if (cur->action == TCPCHK_ACT_COMMENT)
2701 ret = cur->comment;
2702 else
2703 goto return_comment;
2704 }
2705 }
2706
2707 i = 1;
2708 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2709 if (cur->comment)
2710 ret = cur->comment;
2711
2712 if (i >= stepid)
2713 goto return_comment;
2714
2715 ++i;
2716 }
2717
2718 return_comment:
2719 return ret;
2720}
2721
Willy Tarreaube74b882017-10-04 16:22:49 +02002722/* proceed with next steps for the TCP checks <check>. Note that this is called
2723 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002724 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2725 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002726 *
2727 * Please do NOT place any return statement in this function and only leave
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002728 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002729 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002730static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002731{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002732 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002733 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002734 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002735 struct conn_stream *cs = check->cs;
2736 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002737 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002738 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002739 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002740 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002741 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002742
Willy Tarreauef953952014-10-02 14:30:14 +02002743 /* here, we know that the check is complete or that it failed */
2744 if (check->result != CHK_RES_UNKNOWN)
2745 goto out_end_tcpcheck;
2746
2747 /* We have 4 possibilities here :
2748 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002749 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002750 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2751 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002752 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002753 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002754 * be waiting for a connection attempt to complete. conn!=NULL.
2755 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002756 *
2757 * #2 and #3 are quite similar, we want both the connection and the
2758 * handshake to complete before going any further. Thus we must always
2759 * wait for a connection to complete unless we're before and existing
2760 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002761 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002762
2763 /* find first rule and skip comments */
2764 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2765 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2766 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2767
Willy Tarreau00149122017-10-04 18:05:01 +02002768 if ((check->current_step || &next->list == head) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +01002769 (conn->flags & CO_FL_WAIT_XPRT)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002770 /* we allow up to min(inter, timeout.connect) for a connection
2771 * to establish but only when timeout.check is set
2772 * as it may be to short for a full check otherwise
2773 */
2774 while (tick_is_expired(t->expire, now_ms)) {
2775 int t_con;
2776
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002777 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002778 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2779
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002780 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002781 t->expire = tick_first(t->expire, t_con);
2782 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002783 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002784 }
2785
Willy Tarreauef953952014-10-02 14:30:14 +02002786 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002787 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002788 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002789 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002790 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002791
Willy Tarreau213c6782014-10-02 14:51:02 +02002792 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002793 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002794 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002795 b_reset(&check->bo);
2796 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002797 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002798 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002799 if (proxy->timeout.check)
2800 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002801 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002802
Willy Tarreauabca5b62013-12-06 14:19:25 +01002803 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002804 /* We have to try to flush the output buffer before reading, at
2805 * the end, or if we're about to send a string that does not fit
2806 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002807 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002808 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002809 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002810 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002811 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002812 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002813 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002814
Olivier Houcharded0f2072018-08-16 15:41:52 +02002815 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002816 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002817
2818 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002819 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002820 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002821 goto out_end_tcpcheck;
2822 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002823 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002824 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002825 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002826 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002827 goto out;
2828 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002829 }
2830
Willy Tarreau263013d2015-05-13 11:59:14 +02002831 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002832 break;
2833
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002834 /* have 'next' point to the next rule or NULL if we're on the
2835 * last one, connect() needs this.
2836 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002837 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002838
2839 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002840 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002841 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002842
2843 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002844 if (&next->list == head)
2845 next = NULL;
2846
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002847 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2848 struct protocol *proto;
2849 struct xprt_ops *xprt;
2850
Willy Tarreau00149122017-10-04 18:05:01 +02002851 /* For a connect action we'll create a new connection.
2852 * We may also have to kill a previous one. But we don't
2853 * want to leave *without* a connection if we came here
2854 * from the connection layer, hence with a connection.
2855 * Thus we'll proceed in the following order :
2856 * 1: close but not release previous connection
2857 * 2: try to get a new connection
2858 * 3: release and replace the old one on success
2859 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002860 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002861 cs_close(check->cs);
2862 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002863 }
2864
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002865 /* mark the step as started */
2866 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002867
2868 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002869 cs = cs_new(NULL);
2870 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002871 step = tcpcheck_get_step_id(check);
2872 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2873 comment = tcpcheck_get_step_comment(check, step);
2874 if (comment)
2875 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002876 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2877 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002878 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002879 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002880 }
2881
Olivier Houchard49065542019-05-31 19:20:36 +02002882 if (check->cs) {
2883 if (check->wait_list.events)
2884 cs->conn->xprt->unsubscribe(cs->conn,
2885 cs->conn->xprt_ctx,
2886 check->wait_list.events,
2887 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002888 /* We may have been scheduled to run, and the
2889 * I/O handler expects to have a cs, so remove
2890 * the tasklet
2891 */
2892 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002893 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002894 }
Willy Tarreau00149122017-10-04 18:05:01 +02002895
Olivier Houchardff1e9f32019-09-20 17:18:35 +02002896 tasklet_set_tid(check->wait_list.tasklet, tid);
2897
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002898 check->cs = cs;
2899 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002900 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002901 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002902 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002903
2904 /* no client address */
Willy Tarreauca79f592019-07-17 19:04:47 +02002905
2906 if (!sockaddr_alloc(&conn->dst)) {
2907 ret = SF_ERR_RESOURCE;
2908 goto fail_check;
2909 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002910
Simon Horman41f58762015-01-30 11:22:56 +09002911 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002912 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002913 *conn->dst = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002914 }
2915 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002916 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002917 *conn->dst = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002918 }
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002919 proto = protocol_by_family(conn->dst->ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002920
2921 /* port */
2922 if (check->current_step->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002923 set_host_port(conn->dst, check->current_step->port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002924 else if (check->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002925 set_host_port(conn->dst, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002926 else if (s->svc_port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002927 set_host_port(conn->dst, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002928
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002929 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002930 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002931 }
2932 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002933 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002934 }
Willy Tarreaube373152018-09-06 11:45:30 +02002935
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002936 conn_prepare(conn, proto, xprt);
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002937
2938 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2939 ret = SF_ERR_RESOURCE;
2940 goto fail_check;
2941 }
2942
Willy Tarreaube373152018-09-06 11:45:30 +02002943 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002944
Willy Tarreaue7dff022015-04-03 01:14:29 +02002945 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002946 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002947 ret = proto->connect(conn,
Willy Tarreau74ab7d22020-01-24 17:52:37 +01002948 CONNECT_HAS_DATA /* I/O polling is always needed */ | ((next && next->action == TCPCHK_ACT_EXPECT) ? 0 : CONNECT_DELACK_ALWAYS));
Olivier Houchard37d78972019-12-30 15:13:42 +01002949 if (conn_ctrl_ready(conn) &&
2950 check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002951 conn->send_proxy_ofs = 1;
2952 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002953 if (xprt_add_hs(conn) < 0)
2954 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002955 }
2956
2957 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002958 * - SF_ERR_NONE if everything's OK
2959 * - SF_ERR_SRVTO if there are no more servers
2960 * - SF_ERR_SRVCL if the connection was refused by the server
2961 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2962 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2963 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002964 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002965 * Note that we try to prevent the network stack from sending the ACK during the
2966 * connect() when a pure TCP check is used (without PROXY protocol).
2967 */
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002968 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002969 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002970 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002971 /* we allow up to min(inter, timeout.connect) for a connection
2972 * to establish but only when timeout.check is set
2973 * as it may be to short for a full check otherwise
2974 */
2975 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2976
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002977 if (proxy->timeout.check && proxy->timeout.connect) {
2978 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002979 t->expire = tick_first(t->expire, t_con);
2980 }
2981 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002982 case SF_ERR_SRVTO: /* ETIMEDOUT */
2983 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002984 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002985 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002986 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002987 comment = tcpcheck_get_step_comment(check, step);
2988 if (comment)
2989 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002990 set_server_check_status(check, HCHK_STATUS_L4CON,
2991 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002992 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002993 case SF_ERR_PRXCOND:
2994 case SF_ERR_RESOURCE:
2995 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002996 step = tcpcheck_get_step_id(check);
2997 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002998 comment = tcpcheck_get_step_comment(check, step);
2999 if (comment)
3000 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003001 set_server_check_status(check, HCHK_STATUS_SOCKERR,
3002 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003003 goto out_end_tcpcheck;
3004 }
3005
3006 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003007 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003008
3009 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003010 while (&check->current_step->list != head &&
3011 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003012 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003013
Willy Tarreauf2c87352015-05-13 12:08:21 +02003014 if (&check->current_step->list == head)
3015 break;
3016
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003017 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003018 if (conn->flags & CO_FL_WAIT_XPRT)
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003019 break;
3020
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003021 } /* end 'connect' */
3022 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3023 /* mark the step as started */
3024 check->last_started_step = check->current_step;
3025
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003026 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003027 if (*b_head(&check->bi) != '\0') {
3028 *b_head(&check->bi) = '\0';
3029 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003030 }
3031
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003032 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003033 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003034 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003035 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003036 set_server_check_status(check, HCHK_STATUS_L7RSP,
3037 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003038 goto out_end_tcpcheck;
3039 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003040
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003041 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003042 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003043 continue;
3044
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003045 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3046 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003047
Willy Tarreauabca5b62013-12-06 14:19:25 +01003048 /* go to next rule and try to send */
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);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003055
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003056 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003057 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003058 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003059 goto out_end_tcpcheck;
3060
Olivier Houchard0ba6c852019-07-18 15:09:08 +02003061 /* If we already subscribed, then we tried to received
3062 * and failed, so there's no point trying again.
3063 */
3064 if (check->wait_list.events & SUB_RETRY_RECV)
3065 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003066 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003067 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003068 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003069 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003070 /* Report network errors only if we got no other data. Otherwise
3071 * we'll let the upper layers decide whether the response is OK
3072 * or not. It is very common that an RST sent by the server is
3073 * reported as an error just after the last data chunk.
3074 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003075 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003076 goto out_end_tcpcheck;
3077 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003078 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003079 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003080 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003081 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003082 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003083 }
3084
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003085 /* mark the step as started */
3086 check->last_started_step = check->current_step;
3087
3088
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003089 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003090 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003091 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003092 if (b_data(&check->bi) < b_size(&check->bi)) {
3093 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003094 }
3095 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003096 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003097 done = 1; /* buffer full, don't wait for more data */
3098 }
3099
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003100 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003101
3102 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003103 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003104 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003105 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003106
3107 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003108 step = tcpcheck_get_step_id(check);
3109 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003110 comment = tcpcheck_get_step_comment(check, step);
3111 if (comment)
3112 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003113 set_server_check_status(check, HCHK_STATUS_L7RSP,
3114 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003115
3116 goto out_end_tcpcheck;
3117 }
3118
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003119 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003120 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003121
Willy Tarreaua970c282013-12-06 12:47:19 +01003122 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003123 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003124 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 +02003125 else if (check->current_step->expect_regex != NULL)
3126 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003127
3128 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003129 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003130
3131 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003132 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003133 if (ret) {
3134 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003135 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003136 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003137 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003138 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003139 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003140 }
3141 else {
3142 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003143 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003144 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003145 comment = tcpcheck_get_step_comment(check, step);
3146 if (comment)
3147 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003148 set_server_check_status(check, HCHK_STATUS_L7RSP,
3149 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003150 goto out_end_tcpcheck;
3151 }
3152 /* matched and was supposed to => OK, next step */
3153 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003154 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003155 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003156
3157 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003158 while (&check->current_step->list != head &&
3159 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003160 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003161
Willy Tarreauf2c87352015-05-13 12:08:21 +02003162 if (&check->current_step->list == head)
3163 break;
3164
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003165 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003166 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003167 }
3168 }
3169 else {
3170 /* not matched */
3171 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003172 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003173 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003174 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003175
3176 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003177 while (&check->current_step->list != head &&
3178 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003179 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003180
Willy Tarreauf2c87352015-05-13 12:08:21 +02003181 if (&check->current_step->list == head)
3182 break;
3183
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003184 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003185 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003186 }
3187 /* not matched but was supposed to => ERROR */
3188 else {
3189 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003190 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003191 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003192 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003193 }
3194 else {
3195 /* we were looking for a regex */
3196 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003197 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003198 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003199 comment = tcpcheck_get_step_comment(check, step);
3200 if (comment)
3201 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003202 set_server_check_status(check, HCHK_STATUS_L7RSP,
3203 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003204 goto out_end_tcpcheck;
3205 }
3206 }
3207 } /* end expect */
3208 } /* end loop over double chained step list */
3209
Baptiste Assmann248f1172018-03-01 21:49:01 +01003210 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003211 if (conn->flags & CO_FL_WAIT_XPRT) {
Baptiste Assmann248f1172018-03-01 21:49:01 +01003212 /* update expire time, should be done by process_chk */
3213 /* we allow up to min(inter, timeout.connect) for a connection
3214 * to establish but only when timeout.check is set
3215 * as it may be to short for a full check otherwise
3216 */
3217 while (tick_is_expired(t->expire, now_ms)) {
3218 int t_con;
3219
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003220 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003221 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3222
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003223 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003224 t->expire = tick_first(t->expire, t_con);
3225 }
3226 goto out;
3227 }
3228
Willy Tarreau263013d2015-05-13 11:59:14 +02003229 /* We're waiting for some I/O to complete, we've reached the end of the
3230 * rules, or both. Do what we have to do, otherwise we're done.
3231 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003232 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003233 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3234 goto out_end_tcpcheck;
3235 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003236
Willy Tarreau53c5a042015-05-13 11:38:17 +02003237 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003238 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003239 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003240 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003241
3242 out_end_tcpcheck:
3243 /* collect possible new errors */
Willy Tarreauef91c932019-07-23 14:37:47 +02003244 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003245 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003246
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003247 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003248 check->current_step = NULL;
3249
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003250 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003251 conn->flags |= CO_FL_ERROR;
3252
Christopher Fauletb6102852017-11-28 10:06:29 +01003253 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003254 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003255}
3256
Simon Hormanb1900d52015-01-30 11:22:54 +09003257const char *init_check(struct check *check, int type)
3258{
3259 check->type = type;
3260
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003261 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3262 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003263
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003264 check->bi.area = calloc(check->bi.size, sizeof(char));
3265 check->bo.area = calloc(check->bo.size, sizeof(char));
3266
3267 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003268 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003269
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003270 check->wait_list.tasklet = tasklet_new();
3271 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003272 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003273 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003274 check->wait_list.tasklet->process = event_srv_chk_io;
3275 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003276 return NULL;
3277}
3278
Simon Hormanbfb5d332015-01-30 11:22:55 +09003279void free_check(struct check *check)
3280{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003281 free(check->bi.area);
3282 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003283 if (check->cs) {
3284 free(check->cs->conn);
3285 check->cs->conn = NULL;
3286 cs_free(check->cs);
3287 check->cs = NULL;
3288 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003289}
3290
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003291void email_alert_free(struct email_alert *alert)
3292{
3293 struct tcpcheck_rule *rule, *back;
3294
3295 if (!alert)
3296 return;
3297
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003298 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3299 LIST_DEL(&rule->list);
3300 free(rule->comment);
3301 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003302 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003303 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003304 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003305 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306}
3307
Olivier Houchard9f6af332018-05-25 14:04:04 +02003308static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003309{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003310 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003311 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003312 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003313
3314 q = container_of(check, typeof(*q), check);
3315
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003316 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003317 while (1) {
3318 if (!(check->state & CHK_ST_ENABLED)) {
3319 if (LIST_ISEMPTY(&q->email_alerts)) {
3320 /* All alerts processed, queue the task */
3321 t->expire = TICK_ETERNITY;
3322 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003323 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003324 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003325
3326 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003328 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003329 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003330 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003331 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003332 }
3333
Olivier Houchard9f6af332018-05-25 14:04:04 +02003334 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003335 if (check->state & CHK_ST_INPROGRESS)
3336 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003337
3338 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3339 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003340 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003341 check->server = NULL;
3342 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003343 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003344 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003345 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003346 return t;
3347}
3348
Christopher Faulet0108bb32017-10-20 21:34:32 +02003349/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3350 *
3351 * The function returns 1 in success case, otherwise, it returns 0 and err is
3352 * filled.
3353 */
3354int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003355{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003356 struct mailer *mailer;
3357 struct email_alertq *queues;
3358 const char *err_str;
3359 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003360
Christopher Faulet0108bb32017-10-20 21:34:32 +02003361 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3362 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003363 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003364 }
3365
Christopher Faulet0108bb32017-10-20 21:34:32 +02003366 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3367 struct email_alertq *q = &queues[i];
3368 struct check *check = &q->check;
3369 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003370
3371 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003372 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003373 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003374 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003375 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003376 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003377 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3378 memprintf(err, "%s", err_str);
3379 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380 }
3381
3382 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003383 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003384 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003385
Emeric Brunc60def82017-09-27 14:59:38 +02003386 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003387 memprintf(err, "out of memory while allocating mailer alerts task");
3388 goto error;
3389 }
3390
3391 check->task = t;
3392 t->process = process_email_alert;
3393 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003394
Christopher Faulet0108bb32017-10-20 21:34:32 +02003395 /* check this in one ms */
3396 t->expire = TICK_ETERNITY;
3397 check->start = now;
3398 task_queue(t);
3399 }
3400
3401 mls->users++;
3402 free(p->email_alert.mailers.name);
3403 p->email_alert.mailers.m = mls;
3404 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003405 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003406
3407 error:
3408 for (i = 0; i < mls->count; i++) {
3409 struct email_alertq *q = &queues[i];
3410 struct check *check = &q->check;
3411
Willy Tarreauf6562792019-05-07 19:05:35 +02003412 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003413 free_check(check);
3414 }
3415 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003416 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003417 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003418}
3419
3420
3421static int add_tcpcheck_expect_str(struct list *list, const char *str)
3422{
3423 struct tcpcheck_rule *tcpcheck;
3424
Willy Tarreaubafbe012017-11-24 17:34:44 +01003425 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003426 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003427 memset(tcpcheck, 0, sizeof(*tcpcheck));
3428 tcpcheck->action = TCPCHK_ACT_EXPECT;
3429 tcpcheck->string = strdup(str);
3430 tcpcheck->expect_regex = NULL;
3431 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003432 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003433 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003434 return 0;
3435 }
3436
3437 LIST_ADDQ(list, &tcpcheck->list);
3438 return 1;
3439}
3440
3441static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3442{
3443 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003444 const char *in;
3445 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003446 int i;
3447
Willy Tarreaubafbe012017-11-24 17:34:44 +01003448 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003449 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003450 memset(tcpcheck, 0, sizeof(*tcpcheck));
3451 tcpcheck->action = TCPCHK_ACT_SEND;
3452 tcpcheck->expect_regex = NULL;
3453 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003454 tcpcheck->string_len = 0;
3455 for (i = 0; strs[i]; i++)
3456 tcpcheck->string_len += strlen(strs[i]);
3457
3458 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3459 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003460 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003461 return 0;
3462 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003463
Willy Tarreau64345aa2016-08-10 19:29:09 +02003464 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003465 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003466 for (in = strs[i]; (*dst = *in++); dst++);
3467 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003468
3469 LIST_ADDQ(list, &tcpcheck->list);
3470 return 1;
3471}
3472
Christopher Faulet0108bb32017-10-20 21:34:32 +02003473static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3474 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003475{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003476 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003477 struct tcpcheck_rule *tcpcheck;
3478 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003479
Willy Tarreaubafbe012017-11-24 17:34:44 +01003480 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003481 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003482 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003483 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003484 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003485
Willy Tarreaubafbe012017-11-24 17:34:44 +01003486 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003487 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003488 memset(tcpcheck, 0, sizeof(*tcpcheck));
3489 tcpcheck->action = TCPCHK_ACT_CONNECT;
3490 tcpcheck->comment = NULL;
3491 tcpcheck->string = NULL;
3492 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003493 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3494
3495 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3496 goto error;
3497
3498 {
3499 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3500 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3501 goto error;
3502 }
3503
3504 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3505 goto error;
3506
3507 {
3508 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3509 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3510 goto error;
3511 }
3512
3513 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3514 goto error;
3515
3516 {
3517 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3518 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3519 goto error;
3520 }
3521
3522 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3523 goto error;
3524
3525 {
3526 const char * const strs[2] = { "DATA\r\n" };
3527 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3528 goto error;
3529 }
3530
3531 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3532 goto error;
3533
3534 {
3535 struct tm tm;
3536 char datestr[48];
3537 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003538 "From: ", p->email_alert.from, "\r\n",
3539 "To: ", p->email_alert.to, "\r\n",
3540 "Date: ", datestr, "\r\n",
3541 "Subject: [HAproxy Alert] ", msg, "\r\n",
3542 "\r\n",
3543 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003544 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003545 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003546 NULL
3547 };
3548
3549 get_localtime(date.tv_sec, &tm);
3550
3551 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3552 goto error;
3553 }
3554
3555 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3556 goto error;
3557 }
3558
3559 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3560 goto error;
3561
3562 {
3563 const char * const strs[2] = { "QUIT\r\n" };
3564 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3565 goto error;
3566 }
3567
3568 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3569 goto error;
3570
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003571 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003572 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003573 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003574 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003575 return 1;
3576
3577error:
3578 email_alert_free(alert);
3579 return 0;
3580}
3581
Christopher Faulet0108bb32017-10-20 21:34:32 +02003582static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003583{
3584 int i;
3585 struct mailer *mailer;
3586
3587 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3588 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003589 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003590 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003591 return;
3592 }
3593 }
3594
3595 return;
3596}
3597
3598/*
3599 * Send email alert if configured.
3600 */
Simon Horman64e34162015-02-06 11:11:57 +09003601void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003602{
3603 va_list argp;
3604 char buf[1024];
3605 int len;
3606 struct proxy *p = s->proxy;
3607
Christopher Faulet0108bb32017-10-20 21:34:32 +02003608 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003609 return;
3610
3611 va_start(argp, format);
3612 len = vsnprintf(buf, sizeof(buf), format, argp);
3613 va_end(argp);
3614
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003615 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003616 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003617 return;
3618 }
3619
Christopher Faulet0108bb32017-10-20 21:34:32 +02003620 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003621}
3622
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003623/*
3624 * Return value:
3625 * the port to be used for the health check
3626 * 0 in case no port could be found for the check
3627 */
3628int srv_check_healthcheck_port(struct check *chk)
3629{
3630 int i = 0;
3631 struct server *srv = NULL;
3632
3633 srv = chk->server;
3634
3635 /* If neither a port nor an addr was specified and no check transport
3636 * layer is forced, then the transport layer used by the checks is the
3637 * same as for the production traffic. Otherwise we use raw_sock by
3638 * default, unless one is specified.
3639 */
3640 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003641 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003642 chk->send_proxy |= (srv->pp_opts);
3643 }
3644
3645 /* by default, we use the health check port ocnfigured */
3646 if (chk->port > 0)
3647 return chk->port;
3648
3649 /* try to get the port from check_core.addr if check.port not set */
3650 i = get_host_port(&chk->addr);
3651 if (i > 0)
3652 return i;
3653
3654 /* try to get the port from server address */
3655 /* prevent MAPPORTS from working at this point, since checks could
3656 * not be performed in such case (MAPPORTS impose a relative ports
3657 * based on live traffic)
3658 */
3659 if (srv->flags & SRV_F_MAPPORTS)
3660 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003661
3662 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003663 if (i > 0)
3664 return i;
3665
3666 return 0;
3667}
3668
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003669REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003670
Willy Tarreaubd741542010-03-16 18:46:54 +01003671/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003672 * Local variables:
3673 * c-indent-level: 8
3674 * c-basic-offset: 8
3675 * End:
3676 */