blob: e7a889f161b3ca2277fffebe9277be1b873ecf6d [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 },
Christopher Fauletaaae9a02020-04-26 09:50:31 +0200118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_ADDR },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_UINT },
Cyril Bontéac92a062014-12-27 22:28:38 +0100120 [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 Tarreau25e2ab52013-12-04 11:17:05 +0100791 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
792 if (s->proxy->timeout.check) {
793 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
794 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200795 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200796 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100797
Willy Tarreau83749182007-04-15 20:56:27 +0200798 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200799 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200800 out:
801 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200802}
803
Willy Tarreaubaaee002006-06-26 02:48:02 +0200804/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200805 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200806 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900807 * set_server_check_status() to update check->status, check->duration
808 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200809
810 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
811 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
812 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
813 * response to an SSL HELLO (the principle is that this is enough to
814 * distinguish between an SSL server and a pure TCP relay). All other cases will
815 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
816 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100817 *
818 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200819 * via the out label.
820 *
821 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200822 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200823static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200824{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200825 struct connection *conn = cs->conn;
826 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900827 struct server *s = check->server;
828 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200829 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100830 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200831 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200832
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100833 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200834 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200835
Willy Tarreauc09572f2017-10-04 11:58:22 +0200836 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100837 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200838 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200839
Willy Tarreau83749182007-04-15 20:56:27 +0200840 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
841 * but the connection was closed on the remote end. Fortunately, recv still
842 * works correctly and we don't need to do the getsockopt() on linux.
843 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000844
845 /* Set buffer to point to the end of the data already read, and check
846 * that there is free space remaining. If the buffer is full, proceed
847 * with running the checks without attempting another socket read.
848 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000849
Willy Tarreau03938182010-03-17 21:52:07 +0100850 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000851
Olivier Houchard511efea2018-08-16 15:30:32 +0200852 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200853 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100854 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200855 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200856 /* Report network errors only if we got no other data. Otherwise
857 * we'll let the upper layers decide whether the response is OK
858 * or not. It is very common that an RST sent by the server is
859 * reported as an error just after the last data chunk.
860 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200861 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100862 goto out_wakeup;
863 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200864 }
865
Willy Tarreau0f0393f2019-09-24 10:43:03 +0200866 /* the rest of the code below expects the connection to be ready! */
Willy Tarreau911db9b2020-01-23 16:27:54 +0100867 if (conn->flags & CO_FL_WAIT_XPRT && !done)
Willy Tarreau0f0393f2019-09-24 10:43:03 +0200868 goto wait_more_data;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100869
Willy Tarreau03938182010-03-17 21:52:07 +0100870 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200871 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100872 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200873 if (b_data(&check->bi) < b_size(&check->bi))
874 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100875 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200876 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100877 done = 1; /* buffer full, don't wait for more data */
878 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200879
Nick Chalk57b1bf72010-03-16 15:50:46 +0000880 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900881 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200882 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200883 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100884 goto wait_more_data;
885
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100886 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200887 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
888 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
889 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
890 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
891 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
892 cut_crlf(b_head(&check->bi));
893 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200894
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100895 goto out_wakeup;
896 }
897
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200898 check->code = str2uic(b_head(&check->bi) + 9);
899 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200900
Willy Tarreaubd741542010-03-16 18:46:54 +0100901 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200902 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000903 /* 404 may be accepted as "stopping" only if the server was up */
904 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900905 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000906 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100907 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
908 /* Run content verification check... We know we have at least 13 chars */
909 if (!httpchk_expect(s, done))
910 goto wait_more_data;
911 }
912 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200913 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100914 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900915 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100916 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000917 else {
918 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900919 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000920 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200921 break;
922
923 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200924 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100925 goto wait_more_data;
926
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100927 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200928 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900929 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200930 else
Simon Horman4a741432013-02-23 15:35:38 +0900931 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200932 break;
933
934 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200935 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100936 goto wait_more_data;
937
Willy Tarreau5488a622019-08-06 16:26:31 +0200938 /* do not reset when closing, servers don't like this */
939 if (conn_ctrl_ready(cs->conn))
940 fdtab[cs->conn->handle.fd].linger_risk = 0;
941
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200942 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200943 if ((b_data(&check->bi) < strlen("000\r")) ||
944 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
945 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
946 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
947 cut_crlf(b_head(&check->bi));
948 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200949 goto out_wakeup;
950 }
951
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200952 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200953
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200954 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200955 cut_crlf(desc);
956
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100957 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200958 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900959 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200960 else
Simon Horman4a741432013-02-23 15:35:38 +0900961 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200962 break;
963
Simon Hormana2b9dad2013-02-12 10:45:54 +0900964 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100965 int status = HCHK_STATUS_CHECKED;
966 const char *hs = NULL; /* health status */
967 const char *as = NULL; /* admin status */
968 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200969 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100970 const char *err = NULL; /* first error to report */
971 const char *wrn = NULL; /* first warning to report */
972 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900973
Willy Tarreau81f5d942013-12-09 20:51:51 +0100974 /* We're getting an agent check response. The agent could
975 * have been disabled in the mean time with a long check
976 * still pending. It is important that we ignore the whole
977 * response.
978 */
979 if (!(check->server->agent.state & CHK_ST_ENABLED))
980 break;
981
982 /* The agent supports strings made of a single line ended by the
983 * first CR ('\r') or LF ('\n'). This line is composed of words
984 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
985 * line may optionally contained a description of a state change
986 * after a sharp ('#'), which is only considered if a health state
987 * is announced.
988 *
989 * Words may be composed of :
990 * - a numeric weight suffixed by the percent character ('%').
991 * - a health status among "up", "down", "stopped", and "fail".
992 * - an admin status among "ready", "drain", "maint".
993 *
994 * These words may appear in any order. If multiple words of the
995 * same category appear, the last one wins.
996 */
997
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200998 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +0100999 while (*p && *p != '\n' && *p != '\r')
1000 p++;
1001
1002 if (!*p) {
1003 if (!done)
1004 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001005
Willy Tarreau9809b782013-12-11 21:40:11 +01001006 /* at least inform the admin that the agent is mis-behaving */
1007 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1008 break;
1009 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001010
Willy Tarreau9809b782013-12-11 21:40:11 +01001011 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001012 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001013
Willy Tarreau81f5d942013-12-09 20:51:51 +01001014 while (*cmd) {
1015 /* look for next word */
1016 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1017 cmd++;
1018 continue;
1019 }
Simon Horman671b6f02013-11-25 10:46:39 +09001020
Willy Tarreau81f5d942013-12-09 20:51:51 +01001021 if (*cmd == '#') {
1022 /* this is the beginning of a health status description,
1023 * skip the sharp and blanks.
1024 */
1025 cmd++;
1026 while (*cmd == '\t' || *cmd == ' ')
1027 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001028 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001029 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001030
1031 /* find the end of the word so that we have a null-terminated
1032 * word between <cmd> and <p>.
1033 */
1034 p = cmd + 1;
1035 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1036 p++;
1037 if (*p)
1038 *p++ = 0;
1039
1040 /* first, health statuses */
1041 if (strcasecmp(cmd, "up") == 0) {
1042 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001043 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001044 hs = cmd;
1045 }
1046 else if (strcasecmp(cmd, "down") == 0) {
1047 check->health = 0;
1048 status = HCHK_STATUS_L7STS;
1049 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001050 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001051 else if (strcasecmp(cmd, "stopped") == 0) {
1052 check->health = 0;
1053 status = HCHK_STATUS_L7STS;
1054 hs = cmd;
1055 }
1056 else if (strcasecmp(cmd, "fail") == 0) {
1057 check->health = 0;
1058 status = HCHK_STATUS_L7STS;
1059 hs = cmd;
1060 }
1061 /* admin statuses */
1062 else if (strcasecmp(cmd, "ready") == 0) {
1063 as = cmd;
1064 }
1065 else if (strcasecmp(cmd, "drain") == 0) {
1066 as = cmd;
1067 }
1068 else if (strcasecmp(cmd, "maint") == 0) {
1069 as = cmd;
1070 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001071 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001072 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1073 ps = cmd;
1074 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001075 /* try to parse a maxconn here */
1076 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1077 cs = cmd;
1078 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001079 else {
1080 /* keep a copy of the first error */
1081 if (!err)
1082 err = cmd;
1083 }
1084 /* skip to next word */
1085 cmd = p;
1086 }
1087 /* here, cmd points either to \0 or to the beginning of a
1088 * description. Skip possible leading spaces.
1089 */
1090 while (*cmd == ' ' || *cmd == '\n')
1091 cmd++;
1092
1093 /* First, update the admin status so that we avoid sending other
1094 * possibly useless warnings and can also update the health if
1095 * present after going back up.
1096 */
1097 if (as) {
1098 if (strcasecmp(as, "drain") == 0)
1099 srv_adm_set_drain(check->server);
1100 else if (strcasecmp(as, "maint") == 0)
1101 srv_adm_set_maint(check->server);
1102 else
1103 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001104 }
1105
Willy Tarreau81f5d942013-12-09 20:51:51 +01001106 /* now change weights */
1107 if (ps) {
1108 const char *msg;
1109
1110 msg = server_parse_weight_change_request(s, ps);
1111 if (!wrn || !*wrn)
1112 wrn = msg;
1113 }
1114
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001115 if (cs) {
1116 const char *msg;
1117
1118 cs += strlen("maxconn:");
1119
1120 msg = server_parse_maxconn_change_request(s, cs);
1121 if (!wrn || !*wrn)
1122 wrn = msg;
1123 }
1124
Willy Tarreau81f5d942013-12-09 20:51:51 +01001125 /* and finally health status */
1126 if (hs) {
1127 /* We'll report some of the warnings and errors we have
1128 * here. Down reports are critical, we leave them untouched.
1129 * Lack of report, or report of 'UP' leaves the room for
1130 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001131 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001132 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001133 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001134
1135 if (!*msg || status == HCHK_STATUS_L7OKD) {
1136 if (err && *err)
1137 msg = err;
1138 else if (wrn && *wrn)
1139 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001140 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001141
1142 t = get_trash_chunk();
1143 chunk_printf(t, "via agent : %s%s%s%s",
1144 hs, *msg ? " (" : "",
1145 msg, *msg ? ")" : "");
1146
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001147 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001148 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001149 else if (err && *err) {
1150 /* No status change but we'd like to report something odd.
1151 * Just report the current state and copy the message.
1152 */
1153 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001154 set_server_check_status(check, status/*check->status*/,
1155 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001156
Willy Tarreau81f5d942013-12-09 20:51:51 +01001157 }
1158 else if (wrn && *wrn) {
1159 /* No status change but we'd like to report something odd.
1160 * Just report the current state and copy the message.
1161 */
1162 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001163 set_server_check_status(check, status/*check->status*/,
1164 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001165 }
1166 else
1167 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001168 break;
1169 }
1170
Willy Tarreau1620ec32011-08-06 17:05:02 +02001171 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001172 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001173 goto wait_more_data;
1174
Willy Tarreau5488a622019-08-06 16:26:31 +02001175 /* do not reset when closing, servers don't like this */
1176 if (conn_ctrl_ready(cs->conn))
1177 fdtab[cs->conn->handle.fd].linger_risk = 0;
1178
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001179 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001180 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001181 }
1182 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001183 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1184 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001185 else
1186 desc = "PostgreSQL unknown error";
1187
Simon Horman4a741432013-02-23 15:35:38 +09001188 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001189 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001190 break;
1191
1192 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001193 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001194 goto wait_more_data;
1195
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001196 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001197 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001198 }
1199 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001200 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001201 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001202 break;
1203
1204 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001206 goto wait_more_data;
1207
Willy Tarreau5488a622019-08-06 16:26:31 +02001208 /* do not reset when closing, servers don't like this */
1209 if (conn_ctrl_ready(cs->conn))
1210 fdtab[cs->conn->handle.fd].linger_risk = 0;
1211
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001212 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001213 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001214 /* We set the MySQL Version in description for information purpose
1215 * FIXME : it can be cool to use MySQL Version for other purpose,
1216 * like mark as down old MySQL server.
1217 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001218 if (b_data(&check->bi) > 51) {
1219 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001220 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001221 }
1222 else {
1223 if (!done)
1224 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001225
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001226 /* it seems we have a OK packet but without a valid length,
1227 * it must be a protocol error
1228 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001229 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001230 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001231 }
1232 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001233 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001234 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001235 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001236 }
1237 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001238 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1239 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1240 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001241
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001242 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001243 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001244 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001245 /* We have only one MySQL packet and it is a Handshake Initialization packet
1246 * but we need to have a second packet to know if it is alright
1247 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001248 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249 goto wait_more_data;
1250 }
1251 else {
1252 /* We have only one packet and it is an Error packet,
1253 * an error message is attached, so we can display it
1254 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001255 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001256 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001257 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001258 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001259 } else if (b_data(&check->bi) > first_packet_len + 4) {
1260 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1261 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1262 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001263
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001264 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001265 /* We have 2 packets and that's good */
1266 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001267 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001268 /* No error packet */
1269 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001270 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001271 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001272 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001273 }
1274 else {
1275 /* An error message is attached in the Error packet
1276 * so we can display it ! :)
1277 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001278 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001279 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001280 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001281 }
1282 }
1283 }
1284 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001285 if (!done)
1286 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001287
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001288 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001289 * it must be a protocol error
1290 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001291 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001292 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001293 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001294 }
1295 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001296 break;
1297
1298 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001299 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001300 goto wait_more_data;
1301
1302 /* Check if the server speaks LDAP (ASN.1/BER)
1303 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1304 * http://tools.ietf.org/html/rfc4511
1305 */
1306
1307 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1308 * LDAPMessage: 0x30: SEQUENCE
1309 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001310 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001311 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001312 }
1313 else {
1314 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001315 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001316
1317 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1318 * messageID: 0x02 0x01 0x01: INTEGER 1
1319 * protocolOp: 0x61: bindResponse
1320 */
1321 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001322 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001323 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001324 goto out_wakeup;
1325 }
1326
1327 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001328 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001329
1330 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1331 * ldapResult: 0x0a 0x01: ENUMERATION
1332 */
1333 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001334 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001335 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001336 goto out_wakeup;
1337 }
1338
1339 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1340 * resultCode
1341 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001342 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001343 if (check->code) {
1344 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 +02001345 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001346 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001347 }
1348 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001349 break;
1350
Christopher Fauletba7bc162016-11-07 21:07:38 +01001351 case PR_O2_SPOP_CHK: {
1352 unsigned int framesz;
1353 char err[HCHK_DESC_LEN];
1354
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001355 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001356 goto wait_more_data;
1357
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001358 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001359 framesz = ntohl(framesz);
1360
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001361 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001362 goto wait_more_data;
1363
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001364 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001365 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1366 else
1367 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1368 break;
1369 }
1370
Willy Tarreau1620ec32011-08-06 17:05:02 +02001371 default:
Willy Tarreau4c1a2b32019-09-05 18:43:22 +02001372 /* good connection is enough for pure TCP check */
Willy Tarreau911db9b2020-01-23 16:27:54 +01001373 if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Willy Tarreau4c1a2b32019-09-05 18:43:22 +02001374 if (check->use_ssl)
1375 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
1376 else
1377 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
1378 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001379 break;
1380 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001381
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001382 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001383 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001384 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001385 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001386
Nick Chalk57b1bf72010-03-16 15:50:46 +00001387 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001388 *b_head(&check->bi) = '\0';
1389 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001390
Steven Davidovitz544d4812017-03-08 11:06:20 -08001391 /* Close the connection... We still attempt to nicely close if,
1392 * for instance, SSL needs to send a "close notify." Later, we perform
1393 * a hard close and reset the connection if some data are pending,
1394 * otherwise we end up with many TIME_WAITs and eat all the source port
1395 * range quickly. To avoid sending RSTs all the time, we first try to
1396 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001397 */
Olivier Houchard6c7e96a2019-07-02 16:35:18 +02001398 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1399 * connection, to make sure cs_shutw() will not lead to a shutdown()
1400 * that would provoke TIME_WAITs.
1401 */
1402 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001403 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001404
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001405 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001406 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001407 conn->flags |= CO_FL_ERROR;
1408
Willy Tarreaufdccded2008-08-29 18:19:04 +02001409 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001410out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001411 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001412
1413 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001414 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001415 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001416}
1417
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001418/*
1419 * This function is used only for server health-checks. It handles connection
1420 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001421 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1422 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001423 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001424static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001425{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001426 struct connection *conn = cs->conn;
1427 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001428 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001429 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001430
Olivier Houchard0923fa42019-01-11 18:43:04 +01001431 if (check->server)
1432 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1433 else
1434 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001435
Willy Tarreauc09572f2017-10-04 11:58:22 +02001436 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001437 if (check->type == PR_O2_TCPCHK_CHK) {
1438 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001439 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001440 conn = cs->conn;
Willy Tarreauc5940392019-09-05 17:38:40 +02001441 } else {
1442 if (!(check->wait_list.events & SUB_RETRY_SEND))
1443 __event_srv_chk_w(cs);
1444 if (!(check->wait_list.events & SUB_RETRY_RECV))
1445 __event_srv_chk_r(cs);
1446 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001447
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001448 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001449 /* We may get error reports bypassing the I/O handlers, typically
1450 * the case when sending a pure TCP check which fails, then the I/O
1451 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001452 * main processing task so let's simply wake it up. If we get here,
1453 * we expect errno to still be valid.
1454 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001455 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001456 task_wakeup(check->task, TASK_WOKEN_IO);
1457 }
Willy Tarreau911db9b2020-01-23 16:27:54 +01001458 else if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001459 /* we may get here if only a connection probe was required : we
1460 * don't have any data to send nor anything expected in response,
1461 * so the completion of the connection establishment is enough.
1462 */
1463 task_wakeup(check->task, TASK_WOKEN_IO);
1464 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001465
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001466 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001467 /* Check complete or aborted. If connection not yet closed do it
1468 * now and wake the check task up to be sure the result is
1469 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001470 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001471 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001472 ret = -1;
Olivier Houchardf4949572019-07-02 17:42:22 +02001473 /* We may have been scheduled to run, and the
1474 * I/O handler expects to have a cs, so remove
1475 * the tasklet
1476 */
1477 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001478 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001479 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001480
Olivier Houchard0923fa42019-01-11 18:43:04 +01001481 if (check->server)
1482 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1483 else
1484 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001485
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001486 /* if a connection got replaced, we must absolutely prevent the connection
1487 * handler from touching its fd, and perform the FD polling updates ourselves
1488 */
1489 if (ret < 0)
1490 conn_cond_update_polling(conn);
1491
1492 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001493}
1494
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001495struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001496 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001497 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001498};
1499
Willy Tarreaubaaee002006-06-26 02:48:02 +02001500/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001501 * updates the server's weight during a warmup stage. Once the final weight is
1502 * reached, the task automatically stops. Note that any server status change
1503 * must have updated s->last_change accordingly.
1504 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001505static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001506{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001507 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001508
1509 /* by default, plan on stopping the task */
1510 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001511 if ((s->next_admin & SRV_ADMF_MAINT) ||
1512 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001513 return t;
1514
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001515 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1516
Willy Tarreau892337c2014-05-13 23:41:20 +02001517 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001518 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001519
1520 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001521 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001522
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001523 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1524
Willy Tarreau2e993902011-10-31 11:53:20 +01001525 /* get back there in 1 second or 1/20th of the slowstart interval,
1526 * whichever is greater, resulting in small 5% steps.
1527 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001528 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001529 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1530 return t;
1531}
1532
Willy Tarreau894c6422017-10-04 15:58:52 +02001533/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1534 * none was found.
1535 */
1536static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1537{
1538 struct tcpcheck_rule *r;
1539
1540 list_for_each_entry(r, list, list) {
1541 if (r->action != TCPCHK_ACT_COMMENT)
1542 return r;
1543 }
1544 return NULL;
1545}
1546
Willy Tarreau2e993902011-10-31 11:53:20 +01001547/*
Simon Horman98637e52014-06-20 12:30:16 +09001548 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001549 *
1550 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001551 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1552 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1553 * - SF_ERR_SRVTO if there are no more servers
1554 * - SF_ERR_SRVCL if the connection was refused by the server
1555 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1556 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1557 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001558 * - 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 +01001559 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001560 * Note that we try to prevent the network stack from sending the ACK during the
1561 * connect() when a pure TCP check is used (without PROXY protocol).
1562 */
Simon Horman98637e52014-06-20 12:30:16 +09001563static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001564{
1565 struct check *check = t->context;
1566 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001567 struct conn_stream *cs = check->cs;
1568 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001569 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001570 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001571 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001572 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001573
Willy Tarreau00149122017-10-04 18:05:01 +02001574 /* we cannot have a connection here */
1575 if (conn)
1576 return SF_ERR_INTERNAL;
1577
Simon Hormanb00d17a2014-06-13 16:18:16 +09001578 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001579 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001581 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1582 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001583
1584 /* prepare the check buffer.
1585 * This should not be used if check is the secondary agent check
1586 * of a server as s->proxy->check_req will relate to the
1587 * configuration of the primary check. Similarly, tcp-check uses
1588 * its own strings.
1589 */
1590 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001591 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001592
1593 /* we want to check if this host replies to HTTP or SSLv3 requests
1594 * so we'll send the request, and won't wake the checker up now.
1595 */
1596 if ((check->type) == PR_O2_SSL3_CHK) {
1597 /* SSL requires that we put Unix time in the request */
1598 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001599 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001600 }
1601 else if ((check->type) == PR_O2_HTTP_CHK) {
1602 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001603 b_putblk(&check->bo, trash.area,
1604 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001605 /* prevent HTTP keep-alive when "http-check expect" is used */
1606 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001607 b_putist(&check->bo, ist("Connection: close\r\n"));
1608 b_putist(&check->bo, ist("\r\n"));
1609 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001610 }
1611 }
1612
James Brown55f9ff12015-10-21 18:19:05 -07001613 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001614 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001615 }
1616
Willy Tarreauf411cce2017-10-04 16:21:19 +02001617 /* for tcp-checks, the initial connection setup is handled separately as
1618 * it may be sent to a specific port and not to the server's.
1619 */
1620 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1621 tcpcheck_main(check);
1622 return SF_ERR_UP;
1623 }
1624
Simon Hormanb00d17a2014-06-13 16:18:16 +09001625 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001626 cs = check->cs = cs_new(NULL);
1627 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001628 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001629 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001630 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001631 check->wait_list.events = 0;
Olivier Houchardff1e9f32019-09-20 17:18:35 +02001632 tasklet_set_tid(check->wait_list.tasklet, tid);
1633
Simon Hormanb00d17a2014-06-13 16:18:16 +09001634
Willy Tarreauca79f592019-07-17 19:04:47 +02001635 if (!sockaddr_alloc(&conn->dst))
1636 return SF_ERR_RESOURCE;
1637
Simon Horman41f58762015-01-30 11:22:56 +09001638 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001639 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001640 *conn->dst = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001641 }
1642 else {
1643 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001644 *conn->dst = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001645 }
1646
Alexander Liu2a54bb72019-05-22 19:44:48 +08001647 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1648 conn->send_proxy_ofs = 1;
1649 conn->flags |= CO_FL_SOCKS4;
1650 }
1651
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001652 proto = protocol_by_family(conn->dst->ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001653 conn->target = &s->obj_type;
1654
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001655 if ((conn->dst->ss_family == AF_INET) || (conn->dst->ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001656 int i = 0;
1657
1658 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001659 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001660 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001661
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02001662 set_host_port(conn->dst, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001663 }
1664
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001665 /* no client address */
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001666
Willy Tarreaube373152018-09-06 11:45:30 +02001667 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001668 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1669 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001670 cs_attach(cs, check, &check_conn_cb);
1671
Willy Tarreauf3d34822014-12-08 12:11:28 +01001672 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001673 if (check->type != 0)
1674 connflags |= CONNECT_HAS_DATA;
1675 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1676 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1677 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001678
Willy Tarreaue7dff022015-04-03 01:14:29 +02001679 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001680 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001681 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001682
Willy Tarreau16257f62017-11-02 15:45:00 +01001683
Olivier Houchard9130a962017-10-17 17:33:43 +02001684#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001685 if (ret == SF_ERR_NONE) {
1686 if (s->check.sni)
1687 ssl_sock_set_servername(conn, s->check.sni);
1688 if (s->check.alpn_str)
1689 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1690 s->check.alpn_len);
1691 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001692#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001693 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001694 conn->send_proxy_ofs = 1;
1695 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchard37d78972019-12-30 15:13:42 +01001696 }
1697 if (conn->flags & (CO_FL_SEND_PROXY | CO_FL_SOCKS4) &&
1698 conn_ctrl_ready(conn)) {
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001699 if (xprt_add_hs(conn) < 0)
1700 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001701 }
1702
1703 return ret;
1704}
1705
Simon Horman98637e52014-06-20 12:30:16 +09001706static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001707static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001708__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001709
1710void block_sigchld(void)
1711{
1712 sigset_t set;
1713 sigemptyset(&set);
1714 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001715 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001716}
1717
1718void unblock_sigchld(void)
1719{
1720 sigset_t set;
1721 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001722 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001723 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001724}
1725
Simon Horman98637e52014-06-20 12:30:16 +09001726static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1727{
1728 struct pid_list *elem;
1729 struct check *check = t->context;
1730
Willy Tarreaubafbe012017-11-24 17:34:44 +01001731 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001732 if (!elem)
1733 return NULL;
1734 elem->pid = pid;
1735 elem->t = t;
1736 elem->exited = 0;
1737 check->curpid = elem;
1738 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001739
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001740 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001741 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001742 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001743
Simon Horman98637e52014-06-20 12:30:16 +09001744 return elem;
1745}
1746
Simon Horman98637e52014-06-20 12:30:16 +09001747static void pid_list_del(struct pid_list *elem)
1748{
1749 struct check *check;
1750
1751 if (!elem)
1752 return;
1753
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001754 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001755 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001756 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001757
Simon Horman98637e52014-06-20 12:30:16 +09001758 if (!elem->exited)
1759 kill(elem->pid, SIGTERM);
1760
1761 check = elem->t->context;
1762 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001763 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001764}
1765
1766/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1767static void pid_list_expire(pid_t pid, int status)
1768{
1769 struct pid_list *elem;
1770
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001771 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001772 list_for_each_entry(elem, &pid_list, list) {
1773 if (elem->pid == pid) {
1774 elem->t->expire = now_ms;
1775 elem->status = status;
1776 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001777 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001778 break;
Simon Horman98637e52014-06-20 12:30:16 +09001779 }
1780 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001781 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001782}
1783
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001784static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001785{
1786 pid_t pid;
1787 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001788
Simon Horman98637e52014-06-20 12:30:16 +09001789 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1790 pid_list_expire(pid, status);
1791}
1792
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001793static int init_pid_list(void)
1794{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001795 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001796 /* Nothing to do */
1797 return 0;
1798
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001799 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001800 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1801 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001802 return 1;
1803 }
1804
Willy Tarreaubafbe012017-11-24 17:34:44 +01001805 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1806 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001807 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1808 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001809 return 1;
1810 }
1811
1812 return 0;
1813}
1814
Cyril Bontéac92a062014-12-27 22:28:38 +01001815/* helper macro to set an environment variable and jump to a specific label on failure. */
1816#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001817
1818/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001819 * helper function to allocate enough memory to store an environment variable.
1820 * It will also check that the environment variable is updatable, and silently
1821 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001822 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001823static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001824{
1825 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001826 char *envname;
1827 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001828
Cyril Bontéac92a062014-12-27 22:28:38 +01001829 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001830 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001831 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001832 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001833
1834 envname = extcheck_envs[idx].name;
1835 vmaxlen = extcheck_envs[idx].vmaxlen;
1836
1837 /* Check if the environment variable is already set, and silently reject
1838 * the update if this one is not updatable. */
1839 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1840 return 0;
1841
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001842 /* Instead of sending NOT_USED, sending an empty value is preferable */
1843 if (strcmp(value, "NOT_USED") == 0) {
1844 value = "";
1845 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001846
1847 len = strlen(envname) + 1;
1848 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1849 len += strlen(value);
1850 else
1851 len += vmaxlen;
1852
1853 if (!check->envp[idx])
1854 check->envp[idx] = malloc(len + 1);
1855
1856 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001857 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001858 return 1;
1859 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001860 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001861 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001862 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001863 return 1;
1864 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001865 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001866 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001867 return 1;
1868 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001869 return 0;
1870}
Simon Horman98637e52014-06-20 12:30:16 +09001871
1872static int prepare_external_check(struct check *check)
1873{
1874 struct server *s = check->server;
1875 struct proxy *px = s->proxy;
1876 struct listener *listener = NULL, *l;
1877 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001878 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001879 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001880
1881 list_for_each_entry(l, &px->conf.listeners, by_fe)
1882 /* Use the first INET, INET6 or UNIX listener */
1883 if (l->addr.ss_family == AF_INET ||
1884 l->addr.ss_family == AF_INET6 ||
1885 l->addr.ss_family == AF_UNIX) {
1886 listener = l;
1887 break;
1888 }
1889
Simon Horman98637e52014-06-20 12:30:16 +09001890 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001891 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1892 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001893 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001894 goto err;
1895 }
Simon Horman98637e52014-06-20 12:30:16 +09001896
Cyril Bontéac92a062014-12-27 22:28:38 +01001897 check->argv = calloc(6, sizeof(char *));
1898 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001899 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001900 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001901 }
Simon Horman98637e52014-06-20 12:30:16 +09001902
1903 check->argv[0] = px->check_command;
1904
Cyril Bonté777be862014-12-02 21:21:35 +01001905 if (!listener) {
1906 check->argv[1] = strdup("NOT_USED");
1907 check->argv[2] = strdup("NOT_USED");
1908 }
1909 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001910 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001911 addr_to_str(&listener->addr, buf, sizeof(buf));
1912 check->argv[1] = strdup(buf);
1913 port_to_str(&listener->addr, buf, sizeof(buf));
1914 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001915 }
1916 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001917 const struct sockaddr_un *un;
1918
1919 un = (struct sockaddr_un *)&listener->addr;
1920 check->argv[1] = strdup(un->sun_path);
1921 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001922 }
1923 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001924 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001925 goto err;
1926 }
1927
Christopher Fauletaaae9a02020-04-26 09:50:31 +02001928 if (!check->argv[1] || !check->argv[2]) {
1929 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1930 goto err;
1931 }
1932
1933 check->argv[3] = calloc(EXTCHK_SIZE_ADDR, sizeof(*check->argv[3]));
1934 check->argv[4] = calloc(EXTCHK_SIZE_UINT, sizeof(*check->argv[4]));
1935 if (!check->argv[3] || !check->argv[4]) {
1936 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1937 goto err;
1938 }
Willy Tarreau04276f32017-01-06 17:41:29 +01001939
Christopher Fauletaaae9a02020-04-26 09:50:31 +02001940 addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR);
Willy Tarreau04276f32017-01-06 17:41:29 +01001941 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
Christopher Fauletaaae9a02020-04-26 09:50:31 +02001942 snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port);
Simon Horman98637e52014-06-20 12:30:16 +09001943
Cyril Bontéac92a062014-12-27 22:28:38 +01001944 for (i = 0; i < 5; i++) {
1945 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001946 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001947 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001948 }
1949 }
Simon Horman98637e52014-06-20 12:30:16 +09001950
Cyril Bontéac92a062014-12-27 22:28:38 +01001951 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001952 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001953 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1954 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1955 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1956 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001957 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001958 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1959 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1960 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1961 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1962 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1963 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1964
1965 /* Ensure that we don't leave any hole in check->envp */
1966 for (i = 0; i < EXTCHK_SIZE; i++)
1967 if (!check->envp[i])
1968 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001969
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001970 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001971err:
1972 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001973 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001974 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001975 free(check->envp);
1976 check->envp = NULL;
1977 }
1978
1979 if (check->argv) {
1980 for (i = 1; i < 5; i++)
1981 free(check->argv[i]);
1982 free(check->argv);
1983 check->argv = NULL;
1984 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001985 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001986}
1987
Simon Hormanb00d17a2014-06-13 16:18:16 +09001988/*
Simon Horman98637e52014-06-20 12:30:16 +09001989 * establish a server health-check that makes use of a process.
1990 *
1991 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001992 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001993 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001994 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001995 *
1996 * Blocks and then unblocks SIGCHLD
1997 */
1998static int connect_proc_chk(struct task *t)
1999{
Cyril Bontéac92a062014-12-27 22:28:38 +01002000 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09002001 struct check *check = t->context;
2002 struct server *s = check->server;
2003 struct proxy *px = s->proxy;
2004 int status;
2005 pid_t pid;
2006
Willy Tarreaue7dff022015-04-03 01:14:29 +02002007 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002008
2009 block_sigchld();
2010
2011 pid = fork();
2012 if (pid < 0) {
Willy Tarreaud96f1122019-12-03 07:07:36 +01002013 ha_alert("Failed to fork process for external health check%s: %s. Aborting.\n",
2014 (global.tune.options & GTUNE_INSECURE_FORK) ?
2015 "" : " (likely caused by missing 'insecure-fork-wanted')",
Christopher Faulet767a84b2017-11-24 16:50:31 +01002016 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002017 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2018 goto out;
2019 }
2020 if (pid == 0) {
2021 /* Child */
2022 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002023 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002024 int fd;
2025
2026 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2027 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2028
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002029 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002030
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002031 /* restore the initial FD limits */
2032 limit.rlim_cur = rlim_fd_cur_at_boot;
2033 limit.rlim_max = rlim_fd_max_at_boot;
2034 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2035 getrlimit(RLIMIT_NOFILE, &limit);
2036 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2037 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2038 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2039 }
2040
Simon Horman98637e52014-06-20 12:30:16 +09002041 environ = check->envp;
Christopher Fauletaaae9a02020-04-26 09:50:31 +02002042
2043 /* Update some environment variables and command args: curconn, server addr and server port */
Cyril Bontéac92a062014-12-27 22:28:38 +01002044 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Christopher Fauletaaae9a02020-04-26 09:50:31 +02002045
2046 addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR);
2047 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3]);
2048
2049 *check->argv[4] = 0;
2050 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
2051 snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port);
2052 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4]);
2053
Willy Tarreau2df8cad2019-07-01 07:51:29 +02002054 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002055 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002056 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2057 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002058 exit(-1);
2059 }
2060
2061 /* Parent */
2062 if (check->result == CHK_RES_UNKNOWN) {
2063 if (pid_list_add(pid, t) != NULL) {
2064 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2065
2066 if (px->timeout.check && px->timeout.connect) {
2067 int t_con = tick_add(now_ms, px->timeout.connect);
2068 t->expire = tick_first(t->expire, t_con);
2069 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002070 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002071 goto out;
2072 }
2073 else {
2074 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2075 }
2076 kill(pid, SIGTERM); /* process creation error */
2077 }
2078 else
2079 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2080
2081out:
2082 unblock_sigchld();
2083 return status;
2084}
2085
2086/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002087 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002088 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002089 *
2090 * Please do NOT place any return statement in this function and only leave
2091 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002092 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002093static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002094{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002095 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002096 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002097 int rv;
2098 int ret;
2099 int expired = tick_is_expired(t->expire, now_ms);
2100
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002101 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002102 if (!(check->state & CHK_ST_INPROGRESS)) {
2103 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002104 if (!expired) /* woke up too early */
2105 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002106
2107 /* we don't send any health-checks when the proxy is
2108 * stopped, the server should not be checked or the check
2109 * is disabled.
2110 */
2111 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2112 s->proxy->state == PR_STSTOPPED)
2113 goto reschedule;
2114
2115 /* we'll initiate a new check */
2116 set_server_check_status(check, HCHK_STATUS_START, NULL);
2117
2118 check->state |= CHK_ST_INPROGRESS;
2119
Simon Hormandbf70192015-01-30 11:22:53 +09002120 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002121 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002122 /* the process was forked, we allow up to min(inter,
2123 * timeout.connect) for it to report its status, but
2124 * only when timeout.check is set as it may be to short
2125 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002126 */
2127 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2128
2129 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2130 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2131 t->expire = tick_first(t->expire, t_con);
2132 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002133 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002134 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002135 }
2136
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002137 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002138
2139 check->state &= ~CHK_ST_INPROGRESS;
2140 check_notify_failure(check);
2141
2142 /* we allow up to min(inter, timeout.connect) for a connection
2143 * to establish but only when timeout.check is set
2144 * as it may be to short for a full check otherwise
2145 */
2146 while (tick_is_expired(t->expire, now_ms)) {
2147 int t_con;
2148
2149 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2150 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2151
2152 if (s->proxy->timeout.check)
2153 t->expire = tick_first(t->expire, t_con);
2154 }
2155 }
2156 else {
2157 /* there was a test running.
2158 * First, let's check whether there was an uncaught error,
2159 * which can happen on connect timeout or error.
2160 */
2161 if (check->result == CHK_RES_UNKNOWN) {
2162 /* good connection is enough for pure TCP check */
2163 struct pid_list *elem = check->curpid;
2164 int status = HCHK_STATUS_UNKNOWN;
2165
2166 if (elem->exited) {
2167 status = elem->status; /* Save in case the process exits between use below */
2168 if (!WIFEXITED(status))
2169 check->code = -1;
2170 else
2171 check->code = WEXITSTATUS(status);
2172 if (!WIFEXITED(status) || WEXITSTATUS(status))
2173 status = HCHK_STATUS_PROCERR;
2174 else
2175 status = HCHK_STATUS_PROCOK;
2176 } else if (expired) {
2177 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002178 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002179 kill(elem->pid, SIGTERM);
2180 }
2181 set_server_check_status(check, status, NULL);
2182 }
2183
2184 if (check->result == CHK_RES_FAILED) {
2185 /* a failure or timeout detected */
2186 check_notify_failure(check);
2187 }
2188 else if (check->result == CHK_RES_CONDPASS) {
2189 /* check is OK but asks for stopping mode */
2190 check_notify_stopping(check);
2191 }
2192 else if (check->result == CHK_RES_PASSED) {
2193 /* a success was detected */
2194 check_notify_success(check);
2195 }
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002196 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002197 check->state &= ~CHK_ST_INPROGRESS;
2198
2199 pid_list_del(check->curpid);
2200
2201 rv = 0;
2202 if (global.spread_checks > 0) {
2203 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau5a6d3e72020-03-08 17:53:53 +01002204 rv -= (int) (2 * rv * (ha_random32() / 4294967295.0));
Simon Horman98637e52014-06-20 12:30:16 +09002205 }
2206 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2207 }
2208
2209 reschedule:
2210 while (tick_is_expired(t->expire, now_ms))
2211 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002212
2213 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002214 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002215 return t;
2216}
2217
2218/*
2219 * manages a server health-check that uses a connection. Returns
2220 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002221 *
2222 * Please do NOT place any return statement in this function and only leave
2223 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002224 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002225static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002226{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002227 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002228 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002229 struct conn_stream *cs = check->cs;
2230 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002231 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002232 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002233 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002234
Olivier Houchard0923fa42019-01-11 18:43:04 +01002235 if (check->server)
2236 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002237 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002238 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002239 if (!expired) /* woke up too early */
2240 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002241
Simon Horman671b6f02013-11-25 10:46:39 +09002242 /* we don't send any health-checks when the proxy is
2243 * stopped, the server should not be checked or the check
2244 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002245 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002246 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002247 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002248 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002249
2250 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002251 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002252
Willy Tarreau2c115e52013-12-11 19:41:16 +01002253 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002254 b_reset(&check->bi);
2255 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002256
Olivier Houchardaebeff72019-11-29 16:18:51 +01002257 task_set_affinity(t, tid_bit);
Simon Hormandbf70192015-01-30 11:22:53 +09002258 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002259 cs = check->cs;
2260 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002261
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002262 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002263 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002264 goto out_unlock;
2265
Willy Tarreaue7dff022015-04-03 01:14:29 +02002266 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002267 /* we allow up to min(inter, timeout.connect) for a connection
2268 * to establish but only when timeout.check is set
2269 * as it may be to short for a full check otherwise
2270 */
Simon Horman4a741432013-02-23 15:35:38 +09002271 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002272
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002273 if (proxy->timeout.check && proxy->timeout.connect) {
2274 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002275 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002276 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002277
Willy Tarreaucc705a62019-09-05 17:51:30 +02002278 if (check->type) {
2279 /* send the request if we have one. We avoid receiving
2280 * if not connected, unless we didn't subscribe for
2281 * sending since otherwise we won't be woken up.
2282 */
2283 __event_srv_chk_w(cs);
Willy Tarreau911db9b2020-01-23 16:27:54 +01002284 if (!(conn->flags & CO_FL_WAIT_XPRT) ||
Willy Tarreauc5940392019-09-05 17:38:40 +02002285 !(check->wait_list.events & SUB_RETRY_SEND))
2286 __event_srv_chk_r(cs);
Willy Tarreaucc705a62019-09-05 17:51:30 +02002287 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002288
Willy Tarreau5a78f362012-11-23 12:47:05 +01002289 goto reschedule;
2290
Willy Tarreaue7dff022015-04-03 01:14:29 +02002291 case SF_ERR_SRVTO: /* ETIMEDOUT */
2292 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002293 if (conn)
2294 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002295 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002296 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002297 /* should share same code than cases below */
2298 case SF_ERR_CHK_PORT:
2299 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002300 case SF_ERR_PRXCOND:
2301 case SF_ERR_RESOURCE:
2302 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002303 if (conn)
2304 conn->flags |= CO_FL_ERROR;
2305 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002306 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002307 }
2308
Willy Tarreau5a78f362012-11-23 12:47:05 +01002309 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchardaebeff72019-11-29 16:18:51 +01002310 task_set_affinity(t, MAX_THREADS_MASK);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002311 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002312 if (check->wait_list.events)
2313 cs->conn->xprt->unsubscribe(cs->conn,
2314 cs->conn->xprt_ctx,
2315 check->wait_list.events,
2316 &check->wait_list);
2317 /* We may have been scheduled to run, and the
2318 * I/O handler expects to have a cs, so remove
2319 * the tasklet
2320 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002321 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002322 cs_destroy(cs);
2323 cs = check->cs = NULL;
2324 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002325 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002326
Willy Tarreau2c115e52013-12-11 19:41:16 +01002327 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002328 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002329
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002330 /* we allow up to min(inter, timeout.connect) for a connection
2331 * to establish but only when timeout.check is set
2332 * as it may be to short for a full check otherwise
2333 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002334 while (tick_is_expired(t->expire, now_ms)) {
2335 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002336
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002337 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002338 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002339
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002340 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002341 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002342 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002343 }
2344 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002345 /* there was a test running.
2346 * First, let's check whether there was an uncaught error,
2347 * which can happen on connect timeout or error.
2348 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002349 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002350 /* good connection is enough for pure TCP check */
Willy Tarreau911db9b2020-01-23 16:27:54 +01002351 if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002352 if (check->use_ssl)
2353 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002354 else
Simon Horman4a741432013-02-23 15:35:38 +09002355 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002356 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002357 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002358 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002359 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002360 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002361 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002362 }
2363
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002364 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002365 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002366 /* The check was aborted and the connection was not yet closed.
2367 * This can happen upon timeout, or when an external event such
2368 * as a failed response coupled with "observe layer7" caused the
2369 * server state to be suddenly changed.
2370 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002371 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002372 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002373 }
2374
Willy Tarreauac59f362017-10-08 11:10:19 +02002375 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002376 if (check->wait_list.events)
2377 cs->conn->xprt->unsubscribe(cs->conn,
2378 cs->conn->xprt_ctx,
2379 check->wait_list.events,
2380 &check->wait_list);
2381 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002382 * I/O handler expects to have a cs, so remove
2383 * the tasklet
2384 */
2385 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002386 cs_destroy(cs);
2387 cs = check->cs = NULL;
2388 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002389 }
2390
Olivier Houchard0923fa42019-01-11 18:43:04 +01002391 if (check->server) {
2392 if (check->result == CHK_RES_FAILED) {
2393 /* a failure or timeout detected */
2394 check_notify_failure(check);
2395 }
2396 else if (check->result == CHK_RES_CONDPASS) {
2397 /* check is OK but asks for stopping mode */
2398 check_notify_stopping(check);
2399 }
2400 else if (check->result == CHK_RES_PASSED) {
2401 /* a success was detected */
2402 check_notify_success(check);
2403 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002404 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002405 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002406 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002407
Olivier Houchard0923fa42019-01-11 18:43:04 +01002408 if (check->server) {
2409 rv = 0;
2410 if (global.spread_checks > 0) {
2411 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau5a6d3e72020-03-08 17:53:53 +01002412 rv -= (int) (2 * rv * (ha_random32() / 4294967295.0));
Olivier Houchard0923fa42019-01-11 18:43:04 +01002413 }
2414 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002415 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002416 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002417
2418 reschedule:
2419 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002420 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002421 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002422 if (check->server)
2423 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002424 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002425}
2426
Simon Horman98637e52014-06-20 12:30:16 +09002427/*
2428 * manages a server health-check. Returns
2429 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2430 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002431static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002432{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002433 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002434
2435 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002436 return process_chk_proc(t, context, state);
2437 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002438
Simon Horman98637e52014-06-20 12:30:16 +09002439}
2440
Simon Horman5c942422013-11-25 10:46:32 +09002441static int start_check_task(struct check *check, int mininter,
2442 int nbcheck, int srvpos)
2443{
2444 struct task *t;
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002445 unsigned long thread_mask = MAX_THREADS_MASK;
2446
2447 if (check->type == PR_O2_EXT_CHK)
2448 thread_mask = 1;
2449
Simon Horman5c942422013-11-25 10:46:32 +09002450 /* task for the check */
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002451 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002452 ha_alert("Starting [%s:%s] check: out of memory.\n",
2453 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002454 return 0;
2455 }
2456
2457 check->task = t;
2458 t->process = process_chk;
2459 t->context = check;
2460
Willy Tarreau1746eec2014-04-25 10:46:47 +02002461 if (mininter < srv_getinter(check))
2462 mininter = srv_getinter(check);
2463
2464 if (global.max_spread_checks && mininter > global.max_spread_checks)
2465 mininter = global.max_spread_checks;
2466
Simon Horman5c942422013-11-25 10:46:32 +09002467 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002468 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002469 check->start = now;
2470 task_queue(t);
2471
2472 return 1;
2473}
2474
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002475/*
2476 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002477 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002478 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002479static int start_checks()
2480{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002481
2482 struct proxy *px;
2483 struct server *s;
2484 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002485 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002486
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002487 /* 1- count the checkers to run simultaneously.
2488 * We also determine the minimum interval among all of those which
2489 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2490 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002491 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002492 * too short an interval for all others.
2493 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002494 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002495 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002496 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002497 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002498 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002499 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002500 }
2501 /* We need a warmup task that will be called when the server
2502 * state switches from down to up.
2503 */
2504 s->warmup = t;
2505 t->process = server_warmup;
2506 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002507 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002508 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002509 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 +01002510 }
2511
Willy Tarreaud8514a22013-12-11 21:10:14 +01002512 if (s->check.state & CHK_ST_CONFIGURED) {
2513 nbcheck++;
2514 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2515 (!mininter || mininter > srv_getinter(&s->check)))
2516 mininter = srv_getinter(&s->check);
2517 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002518
Willy Tarreaud8514a22013-12-11 21:10:14 +01002519 if (s->agent.state & CHK_ST_CONFIGURED) {
2520 nbcheck++;
2521 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2522 (!mininter || mininter > srv_getinter(&s->agent)))
2523 mininter = srv_getinter(&s->agent);
2524 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002525 }
2526 }
2527
Simon Horman4a741432013-02-23 15:35:38 +09002528 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002529 return 0;
2530
2531 srand((unsigned)time(NULL));
2532
2533 /*
2534 * 2- start them as far as possible from each others. For this, we will
2535 * start them after their interval set to the min interval divided by
2536 * the number of servers, weighted by the server's position in the list.
2537 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002538 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002539 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2540 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002541 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002542 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002543 }
2544 }
2545
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002546 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002547 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002548 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002549 if (s->check.type == PR_O2_EXT_CHK) {
2550 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002551 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002552 }
Simon Hormand60d6912013-11-25 10:46:36 +09002553 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002554 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002555 srvpos++;
2556 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002557
Simon Hormand60d6912013-11-25 10:46:36 +09002558 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002559 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002560 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002561 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002562 }
2563 srvpos++;
2564 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002565 }
2566 }
2567 return 0;
2568}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002569
2570/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002571 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002572 * The buffer MUST be terminated by a null byte before calling this function.
2573 * Sets server status appropriately. The caller is responsible for ensuring
2574 * that the buffer contains at least 13 characters. If <done> is zero, we may
2575 * return 0 to indicate that data is required to decide of a match.
2576 */
2577static int httpchk_expect(struct server *s, int done)
2578{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002579 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002580 char status_code[] = "000";
2581 char *contentptr;
2582 int crlf;
2583 int ret;
2584
2585 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2586 case PR_O2_EXP_STS:
2587 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002588 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2589 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002590
2591 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2592 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2593 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002594 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002595
2596 /* we necessarily have the response, so there are no partial failures */
2597 if (s->proxy->options2 & PR_O2_EXP_INV)
2598 ret = !ret;
2599
Simon Horman4a741432013-02-23 15:35:38 +09002600 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002601 break;
2602
2603 case PR_O2_EXP_STR:
2604 case PR_O2_EXP_RSTR:
2605 /* very simple response parser: ignore CR and only count consecutive LFs,
2606 * stop with contentptr pointing to first char after the double CRLF or
2607 * to '\0' if crlf < 2.
2608 */
2609 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002610 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002611 if (crlf >= 2)
2612 break;
2613 if (*contentptr == '\r')
2614 continue;
2615 else if (*contentptr == '\n')
2616 crlf++;
2617 else
2618 crlf = 0;
2619 }
2620
2621 /* Check that response contains a body... */
2622 if (crlf < 2) {
2623 if (!done)
2624 return 0;
2625
Simon Horman4a741432013-02-23 15:35:38 +09002626 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002627 "HTTP content check could not find a response body");
2628 return 1;
2629 }
2630
2631 /* Check that response body is not empty... */
2632 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002633 if (!done)
2634 return 0;
2635
Simon Horman4a741432013-02-23 15:35:38 +09002636 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002637 "HTTP content check found empty response body");
2638 return 1;
2639 }
2640
2641 /* Check the response content against the supplied string
2642 * or regex... */
2643 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2644 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2645 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002646 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002647
2648 /* if we don't match, we may need to wait more */
2649 if (!ret && !done)
2650 return 0;
2651
2652 if (ret) {
2653 /* content matched */
2654 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002655 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002656 "HTTP check matched unwanted content");
2657 else
Simon Horman4a741432013-02-23 15:35:38 +09002658 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002659 "HTTP content check matched");
2660 }
2661 else {
2662 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002663 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002664 "HTTP check did not match unwanted content");
2665 else
Simon Horman4a741432013-02-23 15:35:38 +09002666 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002667 "HTTP content check did not match");
2668 }
2669 break;
2670 }
2671 return 1;
2672}
2673
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002674/*
2675 * return the id of a step in a send/expect session
2676 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002677static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002678{
2679 struct tcpcheck_rule *cur = NULL, *next = NULL;
2680 int i = 0;
2681
Willy Tarreau213c6782014-10-02 14:51:02 +02002682 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002683 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002684 return 0;
2685
Simon Hormane16c1b32015-01-30 11:22:57 +09002686 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002687
2688 /* no step => first step */
2689 if (cur == NULL)
2690 return 1;
2691
2692 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002693 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002694 if (next->list.p == &cur->list)
2695 break;
2696 ++i;
2697 }
2698
2699 return i;
2700}
2701
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002702/*
2703 * return the latest known comment before (including) the given stepid
2704 * returns NULL if no comment found
2705 */
2706static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2707{
2708 struct tcpcheck_rule *cur = NULL;
2709 char *ret = NULL;
2710 int i = 0;
2711
2712 /* not even started anything yet, return latest comment found before any action */
2713 if (!check->current_step) {
2714 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2715 if (cur->action == TCPCHK_ACT_COMMENT)
2716 ret = cur->comment;
2717 else
2718 goto return_comment;
2719 }
2720 }
2721
2722 i = 1;
2723 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2724 if (cur->comment)
2725 ret = cur->comment;
2726
2727 if (i >= stepid)
2728 goto return_comment;
2729
2730 ++i;
2731 }
2732
2733 return_comment:
2734 return ret;
2735}
2736
Willy Tarreaube74b882017-10-04 16:22:49 +02002737/* proceed with next steps for the TCP checks <check>. Note that this is called
2738 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002739 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2740 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002741 *
2742 * Please do NOT place any return statement in this function and only leave
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002743 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002744 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002745static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002746{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002747 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002748 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002749 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002750 struct conn_stream *cs = check->cs;
2751 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002752 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002753 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002754 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002755 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002756 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002757
Willy Tarreauef953952014-10-02 14:30:14 +02002758 /* here, we know that the check is complete or that it failed */
2759 if (check->result != CHK_RES_UNKNOWN)
2760 goto out_end_tcpcheck;
2761
2762 /* We have 4 possibilities here :
2763 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002764 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002765 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2766 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002767 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002768 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002769 * be waiting for a connection attempt to complete. conn!=NULL.
2770 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002771 *
2772 * #2 and #3 are quite similar, we want both the connection and the
2773 * handshake to complete before going any further. Thus we must always
2774 * wait for a connection to complete unless we're before and existing
2775 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002776 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002777
2778 /* find first rule and skip comments */
2779 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2780 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2781 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2782
Willy Tarreau00149122017-10-04 18:05:01 +02002783 if ((check->current_step || &next->list == head) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +01002784 (conn->flags & CO_FL_WAIT_XPRT)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002785 /* we allow up to min(inter, timeout.connect) for a connection
2786 * to establish but only when timeout.check is set
2787 * as it may be to short for a full check otherwise
2788 */
2789 while (tick_is_expired(t->expire, now_ms)) {
2790 int t_con;
2791
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002792 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002793 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2794
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002795 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002796 t->expire = tick_first(t->expire, t_con);
2797 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002798 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002799 }
2800
Willy Tarreauef953952014-10-02 14:30:14 +02002801 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002802 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002803 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002804 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002805 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002806
Willy Tarreau213c6782014-10-02 14:51:02 +02002807 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002808 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002809 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002810 b_reset(&check->bo);
2811 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002812 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002813 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002814 if (proxy->timeout.check)
2815 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002816 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002817
Willy Tarreauabca5b62013-12-06 14:19:25 +01002818 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002819 /* We have to try to flush the output buffer before reading, at
2820 * the end, or if we're about to send a string that does not fit
2821 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002822 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002823 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002824 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002825 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002826 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002827 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002828 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002829
Olivier Houcharded0f2072018-08-16 15:41:52 +02002830 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002831 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002832
2833 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002834 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002835 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002836 goto out_end_tcpcheck;
2837 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002838 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002839 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002840 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002841 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002842 goto out;
2843 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002844 }
2845
Willy Tarreau263013d2015-05-13 11:59:14 +02002846 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002847 break;
2848
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002849 /* have 'next' point to the next rule or NULL if we're on the
2850 * last one, connect() needs this.
2851 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002852 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002853
2854 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002855 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002856 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002857
2858 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002859 if (&next->list == head)
2860 next = NULL;
2861
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002862 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2863 struct protocol *proto;
2864 struct xprt_ops *xprt;
2865
Willy Tarreau00149122017-10-04 18:05:01 +02002866 /* For a connect action we'll create a new connection.
2867 * We may also have to kill a previous one. But we don't
2868 * want to leave *without* a connection if we came here
2869 * from the connection layer, hence with a connection.
2870 * Thus we'll proceed in the following order :
2871 * 1: close but not release previous connection
2872 * 2: try to get a new connection
2873 * 3: release and replace the old one on success
2874 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002875 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002876 cs_close(check->cs);
2877 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002878 }
2879
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002880 /* mark the step as started */
2881 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002882
2883 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002884 cs = cs_new(NULL);
2885 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002886 step = tcpcheck_get_step_id(check);
2887 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2888 comment = tcpcheck_get_step_comment(check, step);
2889 if (comment)
2890 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002891 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2892 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002893 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002894 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002895 }
2896
Olivier Houchard49065542019-05-31 19:20:36 +02002897 if (check->cs) {
2898 if (check->wait_list.events)
2899 cs->conn->xprt->unsubscribe(cs->conn,
2900 cs->conn->xprt_ctx,
2901 check->wait_list.events,
2902 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002903 /* We may have been scheduled to run, and the
2904 * I/O handler expects to have a cs, so remove
2905 * the tasklet
2906 */
2907 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002908 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002909 }
Willy Tarreau00149122017-10-04 18:05:01 +02002910
Olivier Houchardff1e9f32019-09-20 17:18:35 +02002911 tasklet_set_tid(check->wait_list.tasklet, tid);
2912
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002913 check->cs = cs;
2914 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002915 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002916 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002917 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002918
2919 /* no client address */
Willy Tarreauca79f592019-07-17 19:04:47 +02002920
2921 if (!sockaddr_alloc(&conn->dst)) {
2922 ret = SF_ERR_RESOURCE;
2923 goto fail_check;
2924 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002925
Simon Horman41f58762015-01-30 11:22:56 +09002926 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002927 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002928 *conn->dst = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002929 }
2930 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002931 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002932 *conn->dst = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002933 }
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002934 proto = protocol_by_family(conn->dst->ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002935
2936 /* port */
2937 if (check->current_step->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002938 set_host_port(conn->dst, check->current_step->port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002939 else if (check->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002940 set_host_port(conn->dst, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002941 else if (s->svc_port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002942 set_host_port(conn->dst, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002943
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002944 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002945 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002946 }
2947 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002948 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002949 }
Willy Tarreaube373152018-09-06 11:45:30 +02002950
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002951 conn_prepare(conn, proto, xprt);
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002952
2953 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2954 ret = SF_ERR_RESOURCE;
2955 goto fail_check;
2956 }
2957
Willy Tarreaube373152018-09-06 11:45:30 +02002958 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002959
Willy Tarreaue7dff022015-04-03 01:14:29 +02002960 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002961 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002962 ret = proto->connect(conn,
Willy Tarreau74ab7d22020-01-24 17:52:37 +01002963 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 +01002964 if (conn_ctrl_ready(conn) &&
2965 check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002966 conn->send_proxy_ofs = 1;
2967 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002968 if (xprt_add_hs(conn) < 0)
2969 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002970 }
2971
2972 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002973 * - SF_ERR_NONE if everything's OK
2974 * - SF_ERR_SRVTO if there are no more servers
2975 * - SF_ERR_SRVCL if the connection was refused by the server
2976 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2977 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2978 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002979 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002980 * Note that we try to prevent the network stack from sending the ACK during the
2981 * connect() when a pure TCP check is used (without PROXY protocol).
2982 */
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002983 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002984 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002985 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002986 /* we allow up to min(inter, timeout.connect) for a connection
2987 * to establish but only when timeout.check is set
2988 * as it may be to short for a full check otherwise
2989 */
2990 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2991
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002992 if (proxy->timeout.check && proxy->timeout.connect) {
2993 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002994 t->expire = tick_first(t->expire, t_con);
2995 }
2996 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002997 case SF_ERR_SRVTO: /* ETIMEDOUT */
2998 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002999 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003000 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003001 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003002 comment = tcpcheck_get_step_comment(check, step);
3003 if (comment)
3004 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003005 set_server_check_status(check, HCHK_STATUS_L4CON,
3006 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003007 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02003008 case SF_ERR_PRXCOND:
3009 case SF_ERR_RESOURCE:
3010 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003011 step = tcpcheck_get_step_id(check);
3012 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003013 comment = tcpcheck_get_step_comment(check, step);
3014 if (comment)
3015 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003016 set_server_check_status(check, HCHK_STATUS_SOCKERR,
3017 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003018 goto out_end_tcpcheck;
3019 }
3020
3021 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003022 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003023
3024 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003025 while (&check->current_step->list != head &&
3026 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003027 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003028
Willy Tarreauf2c87352015-05-13 12:08:21 +02003029 if (&check->current_step->list == head)
3030 break;
3031
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003032 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003033 if (conn->flags & CO_FL_WAIT_XPRT)
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003034 break;
3035
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003036 } /* end 'connect' */
3037 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3038 /* mark the step as started */
3039 check->last_started_step = check->current_step;
3040
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003041 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003042 if (*b_head(&check->bi) != '\0') {
3043 *b_head(&check->bi) = '\0';
3044 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003045 }
3046
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003047 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003048 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003049 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003050 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003051 set_server_check_status(check, HCHK_STATUS_L7RSP,
3052 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003053 goto out_end_tcpcheck;
3054 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003055
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003056 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003057 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003058 continue;
3059
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003060 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3061 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003062
Willy Tarreauabca5b62013-12-06 14:19:25 +01003063 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003064 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003065
3066 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003067 while (&check->current_step->list != head &&
3068 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003069 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003070
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003071 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003072 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003073 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003074 goto out_end_tcpcheck;
3075
Olivier Houchard0ba6c852019-07-18 15:09:08 +02003076 /* If we already subscribed, then we tried to received
3077 * and failed, so there's no point trying again.
3078 */
3079 if (check->wait_list.events & SUB_RETRY_RECV)
3080 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003081 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003082 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003083 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003084 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003085 /* Report network errors only if we got no other data. Otherwise
3086 * we'll let the upper layers decide whether the response is OK
3087 * or not. It is very common that an RST sent by the server is
3088 * reported as an error just after the last data chunk.
3089 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003090 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003091 goto out_end_tcpcheck;
3092 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003093 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003094 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003095 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003096 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003097 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003098 }
3099
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003100 /* mark the step as started */
3101 check->last_started_step = check->current_step;
3102
3103
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003104 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003105 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003106 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003107 if (b_data(&check->bi) < b_size(&check->bi)) {
3108 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003109 }
3110 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003111 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003112 done = 1; /* buffer full, don't wait for more data */
3113 }
3114
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003115 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003116
3117 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003118 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003119 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003120 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003121
3122 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003123 step = tcpcheck_get_step_id(check);
3124 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003125 comment = tcpcheck_get_step_comment(check, step);
3126 if (comment)
3127 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003128 set_server_check_status(check, HCHK_STATUS_L7RSP,
3129 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003130
3131 goto out_end_tcpcheck;
3132 }
3133
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003134 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003135 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003136
Willy Tarreaua970c282013-12-06 12:47:19 +01003137 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003138 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003139 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 +02003140 else if (check->current_step->expect_regex != NULL)
3141 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003142
3143 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003144 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003145
3146 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003147 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003148 if (ret) {
3149 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003150 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003151 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003152 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003153 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003154 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003155 }
3156 else {
3157 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003158 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003159 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003160 comment = tcpcheck_get_step_comment(check, step);
3161 if (comment)
3162 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003163 set_server_check_status(check, HCHK_STATUS_L7RSP,
3164 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003165 goto out_end_tcpcheck;
3166 }
3167 /* matched and was supposed to => OK, next step */
3168 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003169 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003170 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003171
3172 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003173 while (&check->current_step->list != head &&
3174 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003175 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003176
Willy Tarreauf2c87352015-05-13 12:08:21 +02003177 if (&check->current_step->list == head)
3178 break;
3179
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003180 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003181 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003182 }
3183 }
3184 else {
3185 /* not matched */
3186 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003187 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003188 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003189 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003190
3191 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003192 while (&check->current_step->list != head &&
3193 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003194 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003195
Willy Tarreauf2c87352015-05-13 12:08:21 +02003196 if (&check->current_step->list == head)
3197 break;
3198
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003199 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003200 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003201 }
3202 /* not matched but was supposed to => ERROR */
3203 else {
3204 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003205 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003206 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003207 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003208 }
3209 else {
3210 /* we were looking for a regex */
3211 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003212 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003213 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003214 comment = tcpcheck_get_step_comment(check, step);
3215 if (comment)
3216 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003217 set_server_check_status(check, HCHK_STATUS_L7RSP,
3218 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003219 goto out_end_tcpcheck;
3220 }
3221 }
3222 } /* end expect */
3223 } /* end loop over double chained step list */
3224
Baptiste Assmann248f1172018-03-01 21:49:01 +01003225 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003226 if (conn->flags & CO_FL_WAIT_XPRT) {
Baptiste Assmann248f1172018-03-01 21:49:01 +01003227 /* update expire time, should be done by process_chk */
3228 /* we allow up to min(inter, timeout.connect) for a connection
3229 * to establish but only when timeout.check is set
3230 * as it may be to short for a full check otherwise
3231 */
3232 while (tick_is_expired(t->expire, now_ms)) {
3233 int t_con;
3234
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003235 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003236 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3237
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003238 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003239 t->expire = tick_first(t->expire, t_con);
3240 }
3241 goto out;
3242 }
3243
Willy Tarreau263013d2015-05-13 11:59:14 +02003244 /* We're waiting for some I/O to complete, we've reached the end of the
3245 * rules, or both. Do what we have to do, otherwise we're done.
3246 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003247 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003248 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3249 goto out_end_tcpcheck;
3250 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003251
Willy Tarreau53c5a042015-05-13 11:38:17 +02003252 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003253 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003254 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003255 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003256
3257 out_end_tcpcheck:
3258 /* collect possible new errors */
Willy Tarreauef91c932019-07-23 14:37:47 +02003259 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003260 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003261
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003262 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003263 check->current_step = NULL;
3264
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003265 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003266 conn->flags |= CO_FL_ERROR;
3267
Christopher Fauletb6102852017-11-28 10:06:29 +01003268 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003269 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003270}
3271
Simon Hormanb1900d52015-01-30 11:22:54 +09003272const char *init_check(struct check *check, int type)
3273{
3274 check->type = type;
3275
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003276 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3277 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003278
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003279 check->bi.area = calloc(check->bi.size, sizeof(char));
3280 check->bo.area = calloc(check->bo.size, sizeof(char));
3281
3282 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003283 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003284
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003285 check->wait_list.tasklet = tasklet_new();
3286 if (!check->wait_list.tasklet)
Ilya Shipitsind4259502020-04-08 01:07:56 +05003287 return "out of memory while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003288 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003289 check->wait_list.tasklet->process = event_srv_chk_io;
3290 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003291 return NULL;
3292}
3293
Simon Hormanbfb5d332015-01-30 11:22:55 +09003294void free_check(struct check *check)
3295{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003296 free(check->bi.area);
3297 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003298 if (check->cs) {
3299 free(check->cs->conn);
3300 check->cs->conn = NULL;
3301 cs_free(check->cs);
3302 check->cs = NULL;
3303 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003304}
3305
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306void email_alert_free(struct email_alert *alert)
3307{
3308 struct tcpcheck_rule *rule, *back;
3309
3310 if (!alert)
3311 return;
3312
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003313 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3314 LIST_DEL(&rule->list);
3315 free(rule->comment);
3316 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003317 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003318 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003319 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003320 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003321}
3322
Olivier Houchard9f6af332018-05-25 14:04:04 +02003323static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003324{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003325 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003326 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003327 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003328
3329 q = container_of(check, typeof(*q), check);
3330
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003331 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003332 while (1) {
3333 if (!(check->state & CHK_ST_ENABLED)) {
3334 if (LIST_ISEMPTY(&q->email_alerts)) {
3335 /* All alerts processed, queue the task */
3336 t->expire = TICK_ETERNITY;
3337 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003338 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003339 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003340
3341 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003342 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003343 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003344 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003345 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003346 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003347 }
3348
Olivier Houchard9f6af332018-05-25 14:04:04 +02003349 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003350 if (check->state & CHK_ST_INPROGRESS)
3351 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003352
3353 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3354 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003355 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003356 check->server = NULL;
3357 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003358 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003359 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003360 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003361 return t;
3362}
3363
Christopher Faulet0108bb32017-10-20 21:34:32 +02003364/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3365 *
3366 * The function returns 1 in success case, otherwise, it returns 0 and err is
3367 * filled.
3368 */
3369int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003370{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003371 struct mailer *mailer;
3372 struct email_alertq *queues;
3373 const char *err_str;
3374 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003375
Christopher Faulet0108bb32017-10-20 21:34:32 +02003376 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3377 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003378 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003379 }
3380
Christopher Faulet0108bb32017-10-20 21:34:32 +02003381 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3382 struct email_alertq *q = &queues[i];
3383 struct check *check = &q->check;
3384 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003385
3386 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003387 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003388 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003389 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003390 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003391 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003392 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3393 memprintf(err, "%s", err_str);
3394 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003395 }
3396
3397 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003398 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003399 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003400
Emeric Brunc60def82017-09-27 14:59:38 +02003401 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003402 memprintf(err, "out of memory while allocating mailer alerts task");
3403 goto error;
3404 }
3405
3406 check->task = t;
3407 t->process = process_email_alert;
3408 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003409
Christopher Faulet0108bb32017-10-20 21:34:32 +02003410 /* check this in one ms */
3411 t->expire = TICK_ETERNITY;
3412 check->start = now;
3413 task_queue(t);
3414 }
3415
3416 mls->users++;
3417 free(p->email_alert.mailers.name);
3418 p->email_alert.mailers.m = mls;
3419 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003420 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003421
3422 error:
3423 for (i = 0; i < mls->count; i++) {
3424 struct email_alertq *q = &queues[i];
3425 struct check *check = &q->check;
3426
Willy Tarreauf6562792019-05-07 19:05:35 +02003427 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003428 free_check(check);
3429 }
3430 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003431 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003432 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003433}
3434
3435
3436static int add_tcpcheck_expect_str(struct list *list, const char *str)
3437{
3438 struct tcpcheck_rule *tcpcheck;
3439
Willy Tarreaubafbe012017-11-24 17:34:44 +01003440 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003441 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003442 memset(tcpcheck, 0, sizeof(*tcpcheck));
3443 tcpcheck->action = TCPCHK_ACT_EXPECT;
3444 tcpcheck->string = strdup(str);
3445 tcpcheck->expect_regex = NULL;
3446 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003447 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003448 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003449 return 0;
3450 }
3451
3452 LIST_ADDQ(list, &tcpcheck->list);
3453 return 1;
3454}
3455
3456static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3457{
3458 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003459 const char *in;
3460 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003461 int i;
3462
Willy Tarreaubafbe012017-11-24 17:34:44 +01003463 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003464 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003465 memset(tcpcheck, 0, sizeof(*tcpcheck));
3466 tcpcheck->action = TCPCHK_ACT_SEND;
3467 tcpcheck->expect_regex = NULL;
3468 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003469 tcpcheck->string_len = 0;
3470 for (i = 0; strs[i]; i++)
3471 tcpcheck->string_len += strlen(strs[i]);
3472
3473 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3474 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003475 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003476 return 0;
3477 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478
Willy Tarreau64345aa2016-08-10 19:29:09 +02003479 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003480 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003481 for (in = strs[i]; (*dst = *in++); dst++);
3482 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003483
3484 LIST_ADDQ(list, &tcpcheck->list);
3485 return 1;
3486}
3487
Christopher Faulet0108bb32017-10-20 21:34:32 +02003488static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3489 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003490{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003491 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003492 struct tcpcheck_rule *tcpcheck;
3493 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003494
Willy Tarreaubafbe012017-11-24 17:34:44 +01003495 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003496 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003497 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003498 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003499 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003500
Willy Tarreaubafbe012017-11-24 17:34:44 +01003501 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003502 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003503 memset(tcpcheck, 0, sizeof(*tcpcheck));
3504 tcpcheck->action = TCPCHK_ACT_CONNECT;
3505 tcpcheck->comment = NULL;
3506 tcpcheck->string = NULL;
3507 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003508 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3509
3510 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3511 goto error;
3512
3513 {
3514 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3515 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3516 goto error;
3517 }
3518
3519 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3520 goto error;
3521
3522 {
3523 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3524 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3525 goto error;
3526 }
3527
3528 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3529 goto error;
3530
3531 {
3532 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3533 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3534 goto error;
3535 }
3536
3537 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3538 goto error;
3539
3540 {
3541 const char * const strs[2] = { "DATA\r\n" };
3542 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3543 goto error;
3544 }
3545
3546 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3547 goto error;
3548
3549 {
3550 struct tm tm;
3551 char datestr[48];
3552 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003553 "From: ", p->email_alert.from, "\r\n",
3554 "To: ", p->email_alert.to, "\r\n",
3555 "Date: ", datestr, "\r\n",
3556 "Subject: [HAproxy Alert] ", msg, "\r\n",
3557 "\r\n",
3558 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003559 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003560 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003561 NULL
3562 };
3563
3564 get_localtime(date.tv_sec, &tm);
3565
3566 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3567 goto error;
3568 }
3569
3570 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3571 goto error;
3572 }
3573
3574 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3575 goto error;
3576
3577 {
3578 const char * const strs[2] = { "QUIT\r\n" };
3579 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3580 goto error;
3581 }
3582
3583 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3584 goto error;
3585
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003586 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003587 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003588 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003589 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003590 return 1;
3591
3592error:
3593 email_alert_free(alert);
3594 return 0;
3595}
3596
Christopher Faulet0108bb32017-10-20 21:34:32 +02003597static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003598{
3599 int i;
3600 struct mailer *mailer;
3601
3602 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3603 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003604 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003605 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003606 return;
3607 }
3608 }
3609
3610 return;
3611}
3612
3613/*
3614 * Send email alert if configured.
3615 */
Simon Horman64e34162015-02-06 11:11:57 +09003616void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003617{
3618 va_list argp;
3619 char buf[1024];
3620 int len;
3621 struct proxy *p = s->proxy;
3622
Christopher Faulet0108bb32017-10-20 21:34:32 +02003623 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003624 return;
3625
3626 va_start(argp, format);
3627 len = vsnprintf(buf, sizeof(buf), format, argp);
3628 va_end(argp);
3629
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003630 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003631 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003632 return;
3633 }
3634
Christopher Faulet0108bb32017-10-20 21:34:32 +02003635 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003636}
3637
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003638/*
3639 * Return value:
3640 * the port to be used for the health check
3641 * 0 in case no port could be found for the check
3642 */
3643int srv_check_healthcheck_port(struct check *chk)
3644{
3645 int i = 0;
3646 struct server *srv = NULL;
3647
3648 srv = chk->server;
3649
3650 /* If neither a port nor an addr was specified and no check transport
3651 * layer is forced, then the transport layer used by the checks is the
3652 * same as for the production traffic. Otherwise we use raw_sock by
3653 * default, unless one is specified.
3654 */
3655 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003656 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003657 chk->send_proxy |= (srv->pp_opts);
3658 }
3659
3660 /* by default, we use the health check port ocnfigured */
3661 if (chk->port > 0)
3662 return chk->port;
3663
3664 /* try to get the port from check_core.addr if check.port not set */
3665 i = get_host_port(&chk->addr);
3666 if (i > 0)
3667 return i;
3668
3669 /* try to get the port from server address */
3670 /* prevent MAPPORTS from working at this point, since checks could
3671 * not be performed in such case (MAPPORTS impose a relative ports
3672 * based on live traffic)
3673 */
3674 if (srv->flags & SRV_F_MAPPORTS)
3675 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003676
3677 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003678 if (i > 0)
3679 return i;
3680
3681 return 0;
3682}
3683
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003684REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003685
Willy Tarreaubd741542010-03-16 18:46:54 +01003686/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003687 * Local variables:
3688 * c-indent-level: 8
3689 * c-basic-offset: 8
3690 * End:
3691 */