blob: 1c71ea618fea2c3a8addfed7f41d93e8d4cb88ab [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
Willy Tarreau9f6dc722019-03-01 11:15:10 +010025#include <sys/resource.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020026#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090028#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020029#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020030#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020031#include <arpa/inet.h>
32
Willy Tarreauc7e42382012-08-24 19:22:53 +020033#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/compat.h>
35#include <common/config.h>
36#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020037#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020038#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020039#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040
41#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020042#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010043#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020044
45#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020046#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010047#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020048#include <proto/fd.h>
49#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020050#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020052#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020063#include <proto/ssl_sock.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020064
Willy Tarreaubd741542010-03-16 18:46:54 +010065static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090066static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020067static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020068static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020069static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020070static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020071static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Willy Tarreau8ceae722018-11-26 11:58:30 +010073DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
74DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020075
76
Simon Horman63a4a822012-03-19 07:24:41 +090077static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010078 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
79 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020080 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020081
Willy Tarreau23964182014-05-20 20:56:30 +020082 /* Below we have finished checks */
83 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010085
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
89 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
90 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
93 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
94 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020095
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010096 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
97 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020099 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100101 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
102 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
103 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900104
105 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
106 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200107 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200108};
109
Cyril Bontéac92a062014-12-27 22:28:38 +0100110const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
111 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
122};
123
Simon Horman63a4a822012-03-19 07:24:41 +0900124static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100125 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
126
127 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
128 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
129
130 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
131 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
132 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
133 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
134
135 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
136 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
138};
139
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100140/* checks if <err> is a real error for errno or one that can be ignored, and
141 * return 0 for these ones or <err> for real ones.
142 */
143static inline int unclean_errno(int err)
144{
145 if (err == EAGAIN || err == EINPROGRESS ||
146 err == EISCONN || err == EALREADY)
147 return 0;
148 return err;
149}
150
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200151/*
152 * Convert check_status code to description
153 */
154const char *get_check_status_description(short check_status) {
155
156 const char *desc;
157
158 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200159 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200160 else
161 desc = NULL;
162
163 if (desc && *desc)
164 return desc;
165 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200166 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200167}
168
169/*
170 * Convert check_status code to short info
171 */
172const char *get_check_status_info(short check_status) {
173
174 const char *info;
175
176 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200177 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200178 else
179 info = NULL;
180
181 if (info && *info)
182 return info;
183 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200184 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200185}
186
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100187const char *get_analyze_status(short analyze_status) {
188
189 const char *desc;
190
191 if (analyze_status < HANA_STATUS_SIZE)
192 desc = analyze_statuses[analyze_status].desc;
193 else
194 desc = NULL;
195
196 if (desc && *desc)
197 return desc;
198 else
199 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
200}
201
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200202/*
Simon Horman4a741432013-02-23 15:35:38 +0900203 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200204 * an adequate CHK_RES_* value. The new check->health is computed based
205 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200206 *
207 * Show information in logs about failed health check if server is UP
208 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200209 */
Simon Horman4a741432013-02-23 15:35:38 +0900210static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100211{
Simon Horman4a741432013-02-23 15:35:38 +0900212 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200213 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200214 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900215
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200216 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100217 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900218 check->desc[0] = '\0';
219 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200220 return;
221 }
222
Simon Horman4a741432013-02-23 15:35:38 +0900223 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200224 return;
225
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200226 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900227 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
228 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200229 } else
Simon Horman4a741432013-02-23 15:35:38 +0900230 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200231
Simon Horman4a741432013-02-23 15:35:38 +0900232 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200233 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900234 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200235
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100236 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900237 check->duration = -1;
238 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200239 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900240 check->duration = tv_ms_elapsed(&check->start, &now);
241 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200242 }
243
Willy Tarreau23964182014-05-20 20:56:30 +0200244 /* no change is expected if no state change occurred */
245 if (check->result == CHK_RES_NEUTRAL)
246 return;
247
Olivier Houchard0923fa42019-01-11 18:43:04 +0100248 /* If the check was really just sending a mail, it won't have an
249 * associated server, so we're done now.
250 */
251 if (!s)
252 return;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200253 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200254
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200255 switch (check->result) {
256 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200257 /* Failure to connect to the agent as a secondary check should not
258 * cause the server to be marked down.
259 */
260 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900261 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200262 (check->health > 0)) {
Olivier Houchard7059c552019-03-08 18:49:32 +0100263 _HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200264 report = 1;
265 check->health--;
266 if (check->health < check->rise)
267 check->health = 0;
268 }
269 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200270
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200271 case CHK_RES_PASSED:
272 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
273 if ((check->health < check->rise + check->fall - 1) &&
274 (check->result == CHK_RES_PASSED || check->health > 0)) {
275 report = 1;
276 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200277
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200278 if (check->health >= check->rise)
279 check->health = check->rise + check->fall - 1; /* OK now */
280 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200281
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200282 /* clear consecutive_errors if observing is enabled */
283 if (s->onerror)
284 s->consecutive_errors = 0;
285 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100286
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200287 default:
288 break;
289 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200290
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200291 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
292 (status != prev_status || report)) {
293 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200294 "%s check for %sserver %s/%s %s%s",
295 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200296 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100297 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100298 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200299 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200300
Emeric Brun5a133512017-10-19 14:42:30 +0200301 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200302
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100303 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200304 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
305 (check->health >= check->rise) ? check->fall : check->rise,
306 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200307
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200308 ha_warning("%s.\n", trash.area);
309 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
310 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200311 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200312}
313
Willy Tarreau4eec5472014-05-20 22:32:27 +0200314/* Marks the check <check>'s server down if the current check is already failed
315 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200316 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200317static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200318{
Simon Horman4a741432013-02-23 15:35:38 +0900319 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900320
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200321 /* The agent secondary check should only cause a server to be marked
322 * as down if check->status is HCHK_STATUS_L7STS, which indicates
323 * that the agent returned "fail", "stopped" or "down".
324 * The implication here is that failure to connect to the agent
325 * as a secondary check should not cause the server to be marked
326 * down. */
327 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
328 return;
329
Willy Tarreau4eec5472014-05-20 22:32:27 +0200330 if (check->health > 0)
331 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100332
Willy Tarreau4eec5472014-05-20 22:32:27 +0200333 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200334 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200335}
336
Willy Tarreauaf549582014-05-16 17:37:50 +0200337/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200338 * it isn't in maintenance, it is not tracking a down server and other checks
339 * comply. The rule is simple : by default, a server is up, unless any of the
340 * following conditions is true :
341 * - health check failed (check->health < rise)
342 * - agent check failed (agent->health < rise)
343 * - the server tracks a down server (track && track->state == STOPPED)
344 * Note that if the server has a slowstart, it will switch to STARTING instead
345 * of RUNNING. Also, only the health checks support the nolb mode, so the
346 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200347 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200348static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200349{
Simon Horman4a741432013-02-23 15:35:38 +0900350 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100351
Emeric Brun52a91d32017-08-31 14:41:55 +0200352 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200353 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100354
Emeric Brun52a91d32017-08-31 14:41:55 +0200355 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200356 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100357
Willy Tarreau3e048382014-05-21 10:30:54 +0200358 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
359 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100360
Willy Tarreau3e048382014-05-21 10:30:54 +0200361 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
362 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200363
Emeric Brun52a91d32017-08-31 14:41:55 +0200364 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200365 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100366
Emeric Brun5a133512017-10-19 14:42:30 +0200367 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100368}
369
Willy Tarreaudb58b792014-05-21 13:57:23 +0200370/* Marks the check <check> as valid and tries to set its server into stopping mode
371 * if it was running or starting, and provided it isn't in maintenance and other
372 * checks comply. The conditions for the server to be marked in stopping mode are
373 * the same as for it to be turned up. Also, only the health checks support the
374 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200375 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200376static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200377{
Simon Horman4a741432013-02-23 15:35:38 +0900378 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100379
Emeric Brun52a91d32017-08-31 14:41:55 +0200380 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200381 return;
382
Willy Tarreaudb58b792014-05-21 13:57:23 +0200383 if (check->state & CHK_ST_AGENT)
384 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100385
Emeric Brun52a91d32017-08-31 14:41:55 +0200386 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200387 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100388
Willy Tarreaudb58b792014-05-21 13:57:23 +0200389 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
390 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100391
Willy Tarreaudb58b792014-05-21 13:57:23 +0200392 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
393 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100394
Willy Tarreaub26881a2017-12-23 11:16:49 +0100395 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100396}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200397
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100398/* note: use health_adjust() only, which first checks that the observe mode is
399 * enabled.
400 */
401void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100402{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100403 int failed;
404 int expire;
405
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100406 if (s->observe >= HANA_OBS_SIZE)
407 return;
408
Willy Tarreaubb956662013-01-24 00:37:39 +0100409 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100410 return;
411
412 switch (analyze_statuses[status].lr[s->observe - 1]) {
413 case 1:
414 failed = 1;
415 break;
416
417 case 2:
418 failed = 0;
419 break;
420
421 default:
422 return;
423 }
424
425 if (!failed) {
426 /* good: clear consecutive_errors */
427 s->consecutive_errors = 0;
428 return;
429 }
430
Olivier Houchard7059c552019-03-08 18:49:32 +0100431 _HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100432
433 if (s->consecutive_errors < s->consecutive_errors_limit)
434 return;
435
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100436 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
437 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100438
439 switch (s->onerror) {
440 case HANA_ONERR_FASTINTER:
441 /* force fastinter - nothing to do here as all modes force it */
442 break;
443
444 case HANA_ONERR_SUDDTH:
445 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900446 if (s->check.health > s->check.rise)
447 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100448
449 /* no break - fall through */
450
451 case HANA_ONERR_FAILCHK:
452 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200453 set_server_check_status(&s->check, HCHK_STATUS_HANA,
454 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200455 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100456 break;
457
458 case HANA_ONERR_MARKDWN:
459 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900460 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200461 set_server_check_status(&s->check, HCHK_STATUS_HANA,
462 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200463 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100464 break;
465
466 default:
467 /* write a warning? */
468 break;
469 }
470
471 s->consecutive_errors = 0;
Olivier Houchard7059c552019-03-08 18:49:32 +0100472 _HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100473
Simon Horman66183002013-02-23 10:16:43 +0900474 if (s->check.fastinter) {
475 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300476 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200477 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300478 /* requeue check task with new expire */
479 task_queue(s->check.task);
480 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100481 }
482}
483
Willy Tarreaua1dab552014-04-14 15:04:54 +0200484static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100485{
486 int sv_state;
487 int ratio;
488 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800489 char addr[46];
490 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100491 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
492 "UP %d/%d", "UP",
493 "NOLB %d/%d", "NOLB",
494 "no check" };
495
496 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
497 hlen += 24;
498
Willy Tarreauff5ae352013-12-11 20:36:34 +0100499 if (!(s->check.state & CHK_ST_ENABLED))
500 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200501 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900502 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100503 sv_state = 3; /* UP */
504 else
505 sv_state = 2; /* going down */
506
Emeric Brun52a91d32017-08-31 14:41:55 +0200507 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100508 sv_state += 2;
509 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900510 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100511 sv_state = 1; /* going up */
512 else
513 sv_state = 0; /* DOWN */
514 }
515
Willy Tarreaua1dab552014-04-14 15:04:54 +0200516 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100517 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200518 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
519 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100520
Joseph Lynch514061c2015-01-15 17:52:59 -0800521 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100522 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
523 snprintf(port, sizeof(port), "%u", s->svc_port);
524 else
525 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800526
527 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
528 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100529 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200530 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100531 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
532 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
533 s->nbpend);
534
Emeric Brun52a91d32017-08-31 14:41:55 +0200535 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100536 now.tv_sec < s->last_change + s->slowstart &&
537 now.tv_sec >= s->last_change) {
538 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200539 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100540 }
541
542 buffer[hlen++] = '\r';
543 buffer[hlen++] = '\n';
544
545 return hlen;
546}
547
Willy Tarreau20a18342013-12-05 00:31:46 +0100548/* Check the connection. If an error has already been reported or the socket is
549 * closed, keep errno intact as it is supposed to contain the valid error code.
550 * If no error is reported, check the socket's error queue using getsockopt().
551 * Warning, this must be done only once when returning from poll, and never
552 * after an I/O error was attempted, otherwise the error queue might contain
553 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
554 * socket. Returns non-zero if an error was reported, zero if everything is
555 * clean (including a properly closed socket).
556 */
557static int retrieve_errno_from_socket(struct connection *conn)
558{
559 int skerr;
560 socklen_t lskerr = sizeof(skerr);
561
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100562 if (conn->flags & CO_FL_ERROR && (unclean_errno(errno) || !conn->ctrl))
Willy Tarreau20a18342013-12-05 00:31:46 +0100563 return 1;
564
Willy Tarreau3c728722014-01-23 13:50:42 +0100565 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100566 return 0;
567
Willy Tarreau585744b2017-08-24 14:31:19 +0200568 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100569 errno = skerr;
570
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100571 errno = unclean_errno(errno);
Willy Tarreau20a18342013-12-05 00:31:46 +0100572
573 if (!errno) {
574 /* we could not retrieve an error, that does not mean there is
575 * none. Just don't change anything and only report the prior
576 * error if any.
577 */
578 if (conn->flags & CO_FL_ERROR)
579 return 1;
580 else
581 return 0;
582 }
583
584 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
585 return 1;
586}
587
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100588/* Try to collect as much information as possible on the connection status,
589 * and adjust the server status accordingly. It may make use of <errno_bck>
590 * if non-null when the caller is absolutely certain of its validity (eg:
591 * checked just after a syscall). If the caller doesn't have a valid errno,
592 * it can pass zero, and retrieve_errno_from_socket() will be called to try
593 * to extract errno from the socket. If no error is reported, it will consider
594 * the <expired> flag. This is intended to be used when a connection error was
595 * reported in conn->flags or when a timeout was reported in <expired>. The
596 * function takes care of not updating a server status which was already set.
597 * All situations where at least one of <expired> or CO_FL_ERROR are set
598 * produce a status.
599 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200600static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200602 struct conn_stream *cs = check->cs;
603 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100604 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200605 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200606 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200607 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100608
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100609 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100610 return;
611
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100612 errno = unclean_errno(errno_bck);
613 if (conn && errno)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100614 retrieve_errno_from_socket(conn);
615
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200616 if (conn && !(conn->flags & CO_FL_ERROR) &&
617 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100618 return;
619
620 /* we'll try to build a meaningful error message depending on the
621 * context of the error possibly present in conn->err_code, and the
622 * socket error possibly collected above. This is useful to know the
623 * exact step of the L6 layer (eg: SSL handshake).
624 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200625 chk = get_trash_chunk();
626
627 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900628 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200629 if (!step)
630 chunk_printf(chk, " at initial connection step of tcp-check");
631 else {
632 chunk_printf(chk, " at step %d of tcp-check", step);
633 /* we were looking for a string */
634 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
635 if (check->last_started_step->port)
636 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
637 else
638 chunk_appendf(chk, " (connect)");
639 }
640 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
641 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200642 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200643 else if (check->last_started_step->expect_regex)
644 chunk_appendf(chk, " (expect regex)");
645 }
646 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
647 chunk_appendf(chk, " (send)");
648 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200649
650 comment = tcpcheck_get_step_comment(check, step);
651 if (comment)
652 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200653 }
654 }
655
Willy Tarreau00149122017-10-04 18:05:01 +0200656 if (conn && conn->err_code) {
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100657 if (unclean_errno(errno))
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200658 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
659 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100660 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200661 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
662 chk->area);
663 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100664 }
665 else {
Willy Tarreauc8dc20a2019-12-27 12:03:27 +0100666 if (unclean_errno(errno)) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200667 chunk_printf(&trash, "%s%s", strerror(errno),
668 chk->area);
669 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100670 }
671 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200672 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100673 }
674 }
675
Willy Tarreau00149122017-10-04 18:05:01 +0200676 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200677 /* NOTE: this is reported after <fall> tries */
678 chunk_printf(chk, "No port available for the TCP connection");
679 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
680 }
681
Willy Tarreau00149122017-10-04 18:05:01 +0200682 if (!conn) {
683 /* connection allocation error before the connection was established */
684 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
685 }
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100686 else if (conn->flags & CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100687 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200688 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100689 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
690 else if (expired)
691 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200692
693 /*
694 * might be due to a server IP change.
695 * Let's trigger a DNS resolution if none are currently running.
696 */
Olivier Houchard0923fa42019-01-11 18:43:04 +0100697 if (check->server)
698 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200699
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100700 }
Willy Tarreauc192b0a2020-01-23 09:11:58 +0100701 else if (conn->flags & CO_FL_WAIT_L6_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100702 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200703 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100704 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
705 else if (expired)
706 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
707 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200708 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100709 /* I/O error after connection was established and before we could diagnose */
710 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
711 }
712 else if (expired) {
713 /* connection established but expired check */
714 if (check->type == PR_O2_SSL3_CHK)
715 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
716 else /* HTTP, SMTP, ... */
717 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
718 }
719
720 return;
721}
722
Olivier Houchard5c110b92018-08-14 17:04:58 +0200723/* This function checks if any I/O is wanted, and if so, attempts to do so */
724static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200725{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200726 struct check *check = ctx;
727 struct conn_stream *cs = check->cs;
Olivier Houchard0923fa42019-01-11 18:43:04 +0100728 struct email_alertq *q = container_of(check, typeof(*q), check);
Olivier Houchardbc89ad82019-07-09 17:28:51 +0200729 int ret = 0;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200730
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100731 if (!(check->wait_list.events & SUB_RETRY_SEND))
Olivier Houchardbc89ad82019-07-09 17:28:51 +0200732 ret = wake_srv_chk(cs);
733 if (ret == 0 && !(check->wait_list.events & SUB_RETRY_RECV)) {
Olivier Houchard0923fa42019-01-11 18:43:04 +0100734 if (check->server)
735 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
736 else
737 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200738 __event_srv_chk_r(cs);
Olivier Houchard0923fa42019-01-11 18:43:04 +0100739 if (check->server)
740 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
741 else
742 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200743 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200744 return NULL;
745}
746
747/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100748 *
749 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200750 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
751 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200752 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200753static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200754{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200755 struct connection *conn = cs->conn;
756 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900757 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900758 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200759
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100760 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100761 goto out_wakeup;
762
Willy Tarreau20a18342013-12-05 00:31:46 +0100763 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200764 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100765 goto out_wakeup;
766 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100767
Willy Tarreau06559ac2013-12-05 01:53:08 +0100768 /* here, we know that the connection is established. That's enough for
769 * a pure TCP check.
770 */
771 if (!check->type)
772 goto out_wakeup;
773
Willy Tarreauc09572f2017-10-04 11:58:22 +0200774 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100775 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200776 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200777
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200778 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200779 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200780 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200781 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200782 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100783 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200784 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200785 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100786 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200787 goto out;
788 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100789 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200790
Willy 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
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001928 addr_to_str(&s->addr, buf, sizeof(buf));
1929 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001930
1931 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1932 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1933 else
1934 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001935 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001936
Cyril Bontéac92a062014-12-27 22:28:38 +01001937 for (i = 0; i < 5; i++) {
1938 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001939 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001940 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001941 }
1942 }
Simon Horman98637e52014-06-20 12:30:16 +09001943
Cyril Bontéac92a062014-12-27 22:28:38 +01001944 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001945 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001946 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1947 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1948 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1949 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001950 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001951 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1952 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1953 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1954 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1955 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1956 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1957
1958 /* Ensure that we don't leave any hole in check->envp */
1959 for (i = 0; i < EXTCHK_SIZE; i++)
1960 if (!check->envp[i])
1961 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001962
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001963 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001964err:
1965 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001966 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001967 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001968 free(check->envp);
1969 check->envp = NULL;
1970 }
1971
1972 if (check->argv) {
1973 for (i = 1; i < 5; i++)
1974 free(check->argv[i]);
1975 free(check->argv);
1976 check->argv = NULL;
1977 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001978 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001979}
1980
Simon Hormanb00d17a2014-06-13 16:18:16 +09001981/*
Simon Horman98637e52014-06-20 12:30:16 +09001982 * establish a server health-check that makes use of a process.
1983 *
1984 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001985 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001986 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001987 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001988 *
1989 * Blocks and then unblocks SIGCHLD
1990 */
1991static int connect_proc_chk(struct task *t)
1992{
Cyril Bontéac92a062014-12-27 22:28:38 +01001993 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001994 struct check *check = t->context;
1995 struct server *s = check->server;
1996 struct proxy *px = s->proxy;
1997 int status;
1998 pid_t pid;
1999
Willy Tarreaue7dff022015-04-03 01:14:29 +02002000 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002001
2002 block_sigchld();
2003
2004 pid = fork();
2005 if (pid < 0) {
Willy Tarreaud96f1122019-12-03 07:07:36 +01002006 ha_alert("Failed to fork process for external health check%s: %s. Aborting.\n",
2007 (global.tune.options & GTUNE_INSECURE_FORK) ?
2008 "" : " (likely caused by missing 'insecure-fork-wanted')",
Christopher Faulet767a84b2017-11-24 16:50:31 +01002009 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002010 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2011 goto out;
2012 }
2013 if (pid == 0) {
2014 /* Child */
2015 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002016 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002017 int fd;
2018
2019 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2020 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2021
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002022 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002023
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002024 /* restore the initial FD limits */
2025 limit.rlim_cur = rlim_fd_cur_at_boot;
2026 limit.rlim_max = rlim_fd_max_at_boot;
2027 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2028 getrlimit(RLIMIT_NOFILE, &limit);
2029 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2030 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2031 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2032 }
2033
Simon Horman98637e52014-06-20 12:30:16 +09002034 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01002035 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Willy Tarreau2df8cad2019-07-01 07:51:29 +02002036 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002037 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002038 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2039 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002040 exit(-1);
2041 }
2042
2043 /* Parent */
2044 if (check->result == CHK_RES_UNKNOWN) {
2045 if (pid_list_add(pid, t) != NULL) {
2046 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2047
2048 if (px->timeout.check && px->timeout.connect) {
2049 int t_con = tick_add(now_ms, px->timeout.connect);
2050 t->expire = tick_first(t->expire, t_con);
2051 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002052 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002053 goto out;
2054 }
2055 else {
2056 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2057 }
2058 kill(pid, SIGTERM); /* process creation error */
2059 }
2060 else
2061 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2062
2063out:
2064 unblock_sigchld();
2065 return status;
2066}
2067
2068/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002069 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002070 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002071 *
2072 * Please do NOT place any return statement in this function and only leave
2073 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002074 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002075static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002076{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002077 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002078 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002079 int rv;
2080 int ret;
2081 int expired = tick_is_expired(t->expire, now_ms);
2082
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002083 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002084 if (!(check->state & CHK_ST_INPROGRESS)) {
2085 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002086 if (!expired) /* woke up too early */
2087 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002088
2089 /* we don't send any health-checks when the proxy is
2090 * stopped, the server should not be checked or the check
2091 * is disabled.
2092 */
2093 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2094 s->proxy->state == PR_STSTOPPED)
2095 goto reschedule;
2096
2097 /* we'll initiate a new check */
2098 set_server_check_status(check, HCHK_STATUS_START, NULL);
2099
2100 check->state |= CHK_ST_INPROGRESS;
2101
Simon Hormandbf70192015-01-30 11:22:53 +09002102 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002103 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002104 /* the process was forked, we allow up to min(inter,
2105 * timeout.connect) for it to report its status, but
2106 * only when timeout.check is set as it may be to short
2107 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002108 */
2109 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2110
2111 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2112 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2113 t->expire = tick_first(t->expire, t_con);
2114 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002115 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002116 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002117 }
2118
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002119 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002120
2121 check->state &= ~CHK_ST_INPROGRESS;
2122 check_notify_failure(check);
2123
2124 /* we allow up to min(inter, timeout.connect) for a connection
2125 * to establish but only when timeout.check is set
2126 * as it may be to short for a full check otherwise
2127 */
2128 while (tick_is_expired(t->expire, now_ms)) {
2129 int t_con;
2130
2131 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2132 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2133
2134 if (s->proxy->timeout.check)
2135 t->expire = tick_first(t->expire, t_con);
2136 }
2137 }
2138 else {
2139 /* there was a test running.
2140 * First, let's check whether there was an uncaught error,
2141 * which can happen on connect timeout or error.
2142 */
2143 if (check->result == CHK_RES_UNKNOWN) {
2144 /* good connection is enough for pure TCP check */
2145 struct pid_list *elem = check->curpid;
2146 int status = HCHK_STATUS_UNKNOWN;
2147
2148 if (elem->exited) {
2149 status = elem->status; /* Save in case the process exits between use below */
2150 if (!WIFEXITED(status))
2151 check->code = -1;
2152 else
2153 check->code = WEXITSTATUS(status);
2154 if (!WIFEXITED(status) || WEXITSTATUS(status))
2155 status = HCHK_STATUS_PROCERR;
2156 else
2157 status = HCHK_STATUS_PROCOK;
2158 } else if (expired) {
2159 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002160 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002161 kill(elem->pid, SIGTERM);
2162 }
2163 set_server_check_status(check, status, NULL);
2164 }
2165
2166 if (check->result == CHK_RES_FAILED) {
2167 /* a failure or timeout detected */
2168 check_notify_failure(check);
2169 }
2170 else if (check->result == CHK_RES_CONDPASS) {
2171 /* check is OK but asks for stopping mode */
2172 check_notify_stopping(check);
2173 }
2174 else if (check->result == CHK_RES_PASSED) {
2175 /* a success was detected */
2176 check_notify_success(check);
2177 }
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002178 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002179 check->state &= ~CHK_ST_INPROGRESS;
2180
2181 pid_list_del(check->curpid);
2182
2183 rv = 0;
2184 if (global.spread_checks > 0) {
2185 rv = srv_getinter(check) * global.spread_checks / 100;
2186 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2187 }
2188 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2189 }
2190
2191 reschedule:
2192 while (tick_is_expired(t->expire, now_ms))
2193 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002194
2195 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002196 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002197 return t;
2198}
2199
2200/*
2201 * manages a server health-check that uses a connection. Returns
2202 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002203 *
2204 * Please do NOT place any return statement in this function and only leave
2205 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002206 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002207static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002208{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002209 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002210 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002211 struct conn_stream *cs = check->cs;
2212 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002213 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002214 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002215 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002216
Olivier Houchard0923fa42019-01-11 18:43:04 +01002217 if (check->server)
2218 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002219 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002220 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002221 if (!expired) /* woke up too early */
2222 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002223
Simon Horman671b6f02013-11-25 10:46:39 +09002224 /* we don't send any health-checks when the proxy is
2225 * stopped, the server should not be checked or the check
2226 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002227 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002228 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002229 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002230 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002231
2232 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002233 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002234
Willy Tarreau2c115e52013-12-11 19:41:16 +01002235 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002236 b_reset(&check->bi);
2237 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002238
Olivier Houchardaebeff72019-11-29 16:18:51 +01002239 task_set_affinity(t, tid_bit);
Simon Hormandbf70192015-01-30 11:22:53 +09002240 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002241 cs = check->cs;
2242 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002243
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002244 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002245 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002246 goto out_unlock;
2247
Willy Tarreaue7dff022015-04-03 01:14:29 +02002248 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002249 /* we allow up to min(inter, timeout.connect) for a connection
2250 * to establish but only when timeout.check is set
2251 * as it may be to short for a full check otherwise
2252 */
Simon Horman4a741432013-02-23 15:35:38 +09002253 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002254
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002255 if (proxy->timeout.check && proxy->timeout.connect) {
2256 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002257 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002258 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002259
Willy Tarreaucc705a62019-09-05 17:51:30 +02002260 if (check->type) {
2261 /* send the request if we have one. We avoid receiving
2262 * if not connected, unless we didn't subscribe for
2263 * sending since otherwise we won't be woken up.
2264 */
2265 __event_srv_chk_w(cs);
Willy Tarreau911db9b2020-01-23 16:27:54 +01002266 if (!(conn->flags & CO_FL_WAIT_XPRT) ||
Willy Tarreauc5940392019-09-05 17:38:40 +02002267 !(check->wait_list.events & SUB_RETRY_SEND))
2268 __event_srv_chk_r(cs);
Willy Tarreaucc705a62019-09-05 17:51:30 +02002269 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002270
Willy Tarreau5a78f362012-11-23 12:47:05 +01002271 goto reschedule;
2272
Willy Tarreaue7dff022015-04-03 01:14:29 +02002273 case SF_ERR_SRVTO: /* ETIMEDOUT */
2274 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002275 if (conn)
2276 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002277 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002278 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002279 /* should share same code than cases below */
2280 case SF_ERR_CHK_PORT:
2281 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002282 case SF_ERR_PRXCOND:
2283 case SF_ERR_RESOURCE:
2284 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002285 if (conn)
2286 conn->flags |= CO_FL_ERROR;
2287 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002288 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002289 }
2290
Willy Tarreau5a78f362012-11-23 12:47:05 +01002291 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchardaebeff72019-11-29 16:18:51 +01002292 task_set_affinity(t, MAX_THREADS_MASK);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002293 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002294 if (check->wait_list.events)
2295 cs->conn->xprt->unsubscribe(cs->conn,
2296 cs->conn->xprt_ctx,
2297 check->wait_list.events,
2298 &check->wait_list);
2299 /* We may have been scheduled to run, and the
2300 * I/O handler expects to have a cs, so remove
2301 * the tasklet
2302 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002303 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002304 cs_destroy(cs);
2305 cs = check->cs = NULL;
2306 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002307 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002308
Willy Tarreau2c115e52013-12-11 19:41:16 +01002309 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002310 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002311
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002312 /* we allow up to min(inter, timeout.connect) for a connection
2313 * to establish but only when timeout.check is set
2314 * as it may be to short for a full check otherwise
2315 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002316 while (tick_is_expired(t->expire, now_ms)) {
2317 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002318
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002319 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002320 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002321
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002322 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002323 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002324 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002325 }
2326 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002327 /* there was a test running.
2328 * First, let's check whether there was an uncaught error,
2329 * which can happen on connect timeout or error.
2330 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002331 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002332 /* good connection is enough for pure TCP check */
Willy Tarreau911db9b2020-01-23 16:27:54 +01002333 if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002334 if (check->use_ssl)
2335 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002336 else
Simon Horman4a741432013-02-23 15:35:38 +09002337 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002338 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002339 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002340 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002341 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002342 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002343 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002344 }
2345
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002346 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002347 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002348 /* The check was aborted and the connection was not yet closed.
2349 * This can happen upon timeout, or when an external event such
2350 * as a failed response coupled with "observe layer7" caused the
2351 * server state to be suddenly changed.
2352 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002353 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002354 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002355 }
2356
Willy Tarreauac59f362017-10-08 11:10:19 +02002357 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002358 if (check->wait_list.events)
2359 cs->conn->xprt->unsubscribe(cs->conn,
2360 cs->conn->xprt_ctx,
2361 check->wait_list.events,
2362 &check->wait_list);
2363 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002364 * I/O handler expects to have a cs, so remove
2365 * the tasklet
2366 */
2367 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002368 cs_destroy(cs);
2369 cs = check->cs = NULL;
2370 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002371 }
2372
Olivier Houchard0923fa42019-01-11 18:43:04 +01002373 if (check->server) {
2374 if (check->result == CHK_RES_FAILED) {
2375 /* a failure or timeout detected */
2376 check_notify_failure(check);
2377 }
2378 else if (check->result == CHK_RES_CONDPASS) {
2379 /* check is OK but asks for stopping mode */
2380 check_notify_stopping(check);
2381 }
2382 else if (check->result == CHK_RES_PASSED) {
2383 /* a success was detected */
2384 check_notify_success(check);
2385 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002386 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002387 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002388 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002389
Olivier Houchard0923fa42019-01-11 18:43:04 +01002390 if (check->server) {
2391 rv = 0;
2392 if (global.spread_checks > 0) {
2393 rv = srv_getinter(check) * global.spread_checks / 100;
2394 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2395 }
2396 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002397 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002398 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002399
2400 reschedule:
2401 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002402 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002403 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002404 if (check->server)
2405 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002406 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002407}
2408
Simon Horman98637e52014-06-20 12:30:16 +09002409/*
2410 * manages a server health-check. Returns
2411 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2412 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002413static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002414{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002415 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002416
2417 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002418 return process_chk_proc(t, context, state);
2419 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002420
Simon Horman98637e52014-06-20 12:30:16 +09002421}
2422
Simon Horman5c942422013-11-25 10:46:32 +09002423static int start_check_task(struct check *check, int mininter,
2424 int nbcheck, int srvpos)
2425{
2426 struct task *t;
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002427 unsigned long thread_mask = MAX_THREADS_MASK;
2428
2429 if (check->type == PR_O2_EXT_CHK)
2430 thread_mask = 1;
2431
Simon Horman5c942422013-11-25 10:46:32 +09002432 /* task for the check */
Willy Tarreau6dd4ac82019-09-03 18:55:02 +02002433 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002434 ha_alert("Starting [%s:%s] check: out of memory.\n",
2435 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002436 return 0;
2437 }
2438
2439 check->task = t;
2440 t->process = process_chk;
2441 t->context = check;
2442
Willy Tarreau1746eec2014-04-25 10:46:47 +02002443 if (mininter < srv_getinter(check))
2444 mininter = srv_getinter(check);
2445
2446 if (global.max_spread_checks && mininter > global.max_spread_checks)
2447 mininter = global.max_spread_checks;
2448
Simon Horman5c942422013-11-25 10:46:32 +09002449 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002450 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002451 check->start = now;
2452 task_queue(t);
2453
2454 return 1;
2455}
2456
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002457/*
2458 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002459 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002460 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002461static int start_checks()
2462{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002463
2464 struct proxy *px;
2465 struct server *s;
2466 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002467 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002468
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002469 /* 1- count the checkers to run simultaneously.
2470 * We also determine the minimum interval among all of those which
2471 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2472 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002473 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002474 * too short an interval for all others.
2475 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002476 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002477 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002478 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002479 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002480 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002481 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002482 }
2483 /* We need a warmup task that will be called when the server
2484 * state switches from down to up.
2485 */
2486 s->warmup = t;
2487 t->process = server_warmup;
2488 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002489 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002490 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002491 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 +01002492 }
2493
Willy Tarreaud8514a22013-12-11 21:10:14 +01002494 if (s->check.state & CHK_ST_CONFIGURED) {
2495 nbcheck++;
2496 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2497 (!mininter || mininter > srv_getinter(&s->check)))
2498 mininter = srv_getinter(&s->check);
2499 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002500
Willy Tarreaud8514a22013-12-11 21:10:14 +01002501 if (s->agent.state & CHK_ST_CONFIGURED) {
2502 nbcheck++;
2503 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2504 (!mininter || mininter > srv_getinter(&s->agent)))
2505 mininter = srv_getinter(&s->agent);
2506 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002507 }
2508 }
2509
Simon Horman4a741432013-02-23 15:35:38 +09002510 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002511 return 0;
2512
2513 srand((unsigned)time(NULL));
2514
2515 /*
2516 * 2- start them as far as possible from each others. For this, we will
2517 * start them after their interval set to the min interval divided by
2518 * the number of servers, weighted by the server's position in the list.
2519 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002520 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002521 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2522 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002523 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002524 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002525 }
2526 }
2527
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002528 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002529 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002530 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002531 if (s->check.type == PR_O2_EXT_CHK) {
2532 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002533 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002534 }
Simon Hormand60d6912013-11-25 10:46:36 +09002535 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002536 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002537 srvpos++;
2538 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002539
Simon Hormand60d6912013-11-25 10:46:36 +09002540 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002541 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002542 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002543 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002544 }
2545 srvpos++;
2546 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002547 }
2548 }
2549 return 0;
2550}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002551
2552/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002553 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002554 * The buffer MUST be terminated by a null byte before calling this function.
2555 * Sets server status appropriately. The caller is responsible for ensuring
2556 * that the buffer contains at least 13 characters. If <done> is zero, we may
2557 * return 0 to indicate that data is required to decide of a match.
2558 */
2559static int httpchk_expect(struct server *s, int done)
2560{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002561 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002562 char status_code[] = "000";
2563 char *contentptr;
2564 int crlf;
2565 int ret;
2566
2567 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2568 case PR_O2_EXP_STS:
2569 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002570 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2571 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002572
2573 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2574 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2575 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002576 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002577
2578 /* we necessarily have the response, so there are no partial failures */
2579 if (s->proxy->options2 & PR_O2_EXP_INV)
2580 ret = !ret;
2581
Simon Horman4a741432013-02-23 15:35:38 +09002582 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002583 break;
2584
2585 case PR_O2_EXP_STR:
2586 case PR_O2_EXP_RSTR:
2587 /* very simple response parser: ignore CR and only count consecutive LFs,
2588 * stop with contentptr pointing to first char after the double CRLF or
2589 * to '\0' if crlf < 2.
2590 */
2591 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002592 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002593 if (crlf >= 2)
2594 break;
2595 if (*contentptr == '\r')
2596 continue;
2597 else if (*contentptr == '\n')
2598 crlf++;
2599 else
2600 crlf = 0;
2601 }
2602
2603 /* Check that response contains a body... */
2604 if (crlf < 2) {
2605 if (!done)
2606 return 0;
2607
Simon Horman4a741432013-02-23 15:35:38 +09002608 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002609 "HTTP content check could not find a response body");
2610 return 1;
2611 }
2612
2613 /* Check that response body is not empty... */
2614 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002615 if (!done)
2616 return 0;
2617
Simon Horman4a741432013-02-23 15:35:38 +09002618 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002619 "HTTP content check found empty response body");
2620 return 1;
2621 }
2622
2623 /* Check the response content against the supplied string
2624 * or regex... */
2625 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2626 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2627 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002628 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002629
2630 /* if we don't match, we may need to wait more */
2631 if (!ret && !done)
2632 return 0;
2633
2634 if (ret) {
2635 /* content matched */
2636 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002637 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002638 "HTTP check matched unwanted content");
2639 else
Simon Horman4a741432013-02-23 15:35:38 +09002640 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002641 "HTTP content check matched");
2642 }
2643 else {
2644 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002645 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002646 "HTTP check did not match unwanted content");
2647 else
Simon Horman4a741432013-02-23 15:35:38 +09002648 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002649 "HTTP content check did not match");
2650 }
2651 break;
2652 }
2653 return 1;
2654}
2655
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002656/*
2657 * return the id of a step in a send/expect session
2658 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002659static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002660{
2661 struct tcpcheck_rule *cur = NULL, *next = NULL;
2662 int i = 0;
2663
Willy Tarreau213c6782014-10-02 14:51:02 +02002664 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002665 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002666 return 0;
2667
Simon Hormane16c1b32015-01-30 11:22:57 +09002668 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002669
2670 /* no step => first step */
2671 if (cur == NULL)
2672 return 1;
2673
2674 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002675 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002676 if (next->list.p == &cur->list)
2677 break;
2678 ++i;
2679 }
2680
2681 return i;
2682}
2683
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002684/*
2685 * return the latest known comment before (including) the given stepid
2686 * returns NULL if no comment found
2687 */
2688static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2689{
2690 struct tcpcheck_rule *cur = NULL;
2691 char *ret = NULL;
2692 int i = 0;
2693
2694 /* not even started anything yet, return latest comment found before any action */
2695 if (!check->current_step) {
2696 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2697 if (cur->action == TCPCHK_ACT_COMMENT)
2698 ret = cur->comment;
2699 else
2700 goto return_comment;
2701 }
2702 }
2703
2704 i = 1;
2705 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2706 if (cur->comment)
2707 ret = cur->comment;
2708
2709 if (i >= stepid)
2710 goto return_comment;
2711
2712 ++i;
2713 }
2714
2715 return_comment:
2716 return ret;
2717}
2718
Willy Tarreaube74b882017-10-04 16:22:49 +02002719/* proceed with next steps for the TCP checks <check>. Note that this is called
2720 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002721 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2722 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002723 *
2724 * Please do NOT place any return statement in this function and only leave
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002725 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002726 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002727static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002728{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002729 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002730 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002731 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002732 struct conn_stream *cs = check->cs;
2733 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002734 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002735 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002736 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002737 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002738 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002739
Willy Tarreauef953952014-10-02 14:30:14 +02002740 /* here, we know that the check is complete or that it failed */
2741 if (check->result != CHK_RES_UNKNOWN)
2742 goto out_end_tcpcheck;
2743
2744 /* We have 4 possibilities here :
2745 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002746 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002747 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2748 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002749 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002750 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002751 * be waiting for a connection attempt to complete. conn!=NULL.
2752 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002753 *
2754 * #2 and #3 are quite similar, we want both the connection and the
2755 * handshake to complete before going any further. Thus we must always
2756 * wait for a connection to complete unless we're before and existing
2757 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002758 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002759
2760 /* find first rule and skip comments */
2761 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2762 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2763 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2764
Willy Tarreau00149122017-10-04 18:05:01 +02002765 if ((check->current_step || &next->list == head) &&
Willy Tarreau911db9b2020-01-23 16:27:54 +01002766 (conn->flags & CO_FL_WAIT_XPRT)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002767 /* we allow up to min(inter, timeout.connect) for a connection
2768 * to establish but only when timeout.check is set
2769 * as it may be to short for a full check otherwise
2770 */
2771 while (tick_is_expired(t->expire, now_ms)) {
2772 int t_con;
2773
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002774 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002775 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2776
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002777 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002778 t->expire = tick_first(t->expire, t_con);
2779 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002780 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002781 }
2782
Willy Tarreauef953952014-10-02 14:30:14 +02002783 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002784 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002785 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002786 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002787 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002788
Willy Tarreau213c6782014-10-02 14:51:02 +02002789 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002790 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002791 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002792 b_reset(&check->bo);
2793 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002794 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002795 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002796 if (proxy->timeout.check)
2797 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002798 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002799
Willy Tarreauabca5b62013-12-06 14:19:25 +01002800 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002801 /* We have to try to flush the output buffer before reading, at
2802 * the end, or if we're about to send a string that does not fit
2803 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002804 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002805 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002806 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002807 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002808 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002809 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002810 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002811
Olivier Houcharded0f2072018-08-16 15:41:52 +02002812 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002813 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002814
2815 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002816 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002817 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002818 goto out_end_tcpcheck;
2819 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002820 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002821 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002822 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002823 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002824 goto out;
2825 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002826 }
2827
Willy Tarreau263013d2015-05-13 11:59:14 +02002828 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002829 break;
2830
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002831 /* have 'next' point to the next rule or NULL if we're on the
2832 * last one, connect() needs this.
2833 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002834 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002835
2836 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002837 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002838 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002839
2840 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002841 if (&next->list == head)
2842 next = NULL;
2843
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002844 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2845 struct protocol *proto;
2846 struct xprt_ops *xprt;
2847
Willy Tarreau00149122017-10-04 18:05:01 +02002848 /* For a connect action we'll create a new connection.
2849 * We may also have to kill a previous one. But we don't
2850 * want to leave *without* a connection if we came here
2851 * from the connection layer, hence with a connection.
2852 * Thus we'll proceed in the following order :
2853 * 1: close but not release previous connection
2854 * 2: try to get a new connection
2855 * 3: release and replace the old one on success
2856 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002857 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002858 cs_close(check->cs);
2859 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002860 }
2861
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002862 /* mark the step as started */
2863 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002864
2865 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002866 cs = cs_new(NULL);
2867 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002868 step = tcpcheck_get_step_id(check);
2869 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2870 comment = tcpcheck_get_step_comment(check, step);
2871 if (comment)
2872 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002873 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2874 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002875 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002876 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002877 }
2878
Olivier Houchard49065542019-05-31 19:20:36 +02002879 if (check->cs) {
2880 if (check->wait_list.events)
2881 cs->conn->xprt->unsubscribe(cs->conn,
2882 cs->conn->xprt_ctx,
2883 check->wait_list.events,
2884 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002885 /* We may have been scheduled to run, and the
2886 * I/O handler expects to have a cs, so remove
2887 * the tasklet
2888 */
2889 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002890 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002891 }
Willy Tarreau00149122017-10-04 18:05:01 +02002892
Olivier Houchardff1e9f32019-09-20 17:18:35 +02002893 tasklet_set_tid(check->wait_list.tasklet, tid);
2894
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002895 check->cs = cs;
2896 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002897 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002898 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002899 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002900
2901 /* no client address */
Willy Tarreauca79f592019-07-17 19:04:47 +02002902
2903 if (!sockaddr_alloc(&conn->dst)) {
2904 ret = SF_ERR_RESOURCE;
2905 goto fail_check;
2906 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002907
Simon Horman41f58762015-01-30 11:22:56 +09002908 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002909 /* we'll connect to the check addr specified on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002910 *conn->dst = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002911 }
2912 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002913 /* we'll connect to the addr on the server */
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002914 *conn->dst = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002915 }
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002916 proto = protocol_by_family(conn->dst->ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002917
2918 /* port */
2919 if (check->current_step->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002920 set_host_port(conn->dst, check->current_step->port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002921 else if (check->port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002922 set_host_port(conn->dst, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002923 else if (s->svc_port)
Willy Tarreaub3c81cb2019-07-17 16:54:52 +02002924 set_host_port(conn->dst, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002925
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002926 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002927 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002928 }
2929 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002930 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002931 }
Willy Tarreaube373152018-09-06 11:45:30 +02002932
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002933 conn_prepare(conn, proto, xprt);
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002934
2935 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2936 ret = SF_ERR_RESOURCE;
2937 goto fail_check;
2938 }
2939
Willy Tarreaube373152018-09-06 11:45:30 +02002940 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002941
Willy Tarreaue7dff022015-04-03 01:14:29 +02002942 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002943 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002944 ret = proto->connect(conn,
Willy Tarreau74ab7d22020-01-24 17:52:37 +01002945 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 +01002946 if (conn_ctrl_ready(conn) &&
2947 check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002948 conn->send_proxy_ofs = 1;
2949 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002950 if (xprt_add_hs(conn) < 0)
2951 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002952 }
2953
2954 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002955 * - SF_ERR_NONE if everything's OK
2956 * - SF_ERR_SRVTO if there are no more servers
2957 * - SF_ERR_SRVCL if the connection was refused by the server
2958 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2959 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2960 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002961 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002962 * Note that we try to prevent the network stack from sending the ACK during the
2963 * connect() when a pure TCP check is used (without PROXY protocol).
2964 */
Willy Tarreau2ab5c382019-07-17 18:48:07 +02002965 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002966 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002967 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002968 /* we allow up to min(inter, timeout.connect) for a connection
2969 * to establish but only when timeout.check is set
2970 * as it may be to short for a full check otherwise
2971 */
2972 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2973
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002974 if (proxy->timeout.check && proxy->timeout.connect) {
2975 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002976 t->expire = tick_first(t->expire, t_con);
2977 }
2978 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002979 case SF_ERR_SRVTO: /* ETIMEDOUT */
2980 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002981 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002982 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002983 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002984 comment = tcpcheck_get_step_comment(check, step);
2985 if (comment)
2986 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002987 set_server_check_status(check, HCHK_STATUS_L4CON,
2988 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002989 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002990 case SF_ERR_PRXCOND:
2991 case SF_ERR_RESOURCE:
2992 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002993 step = tcpcheck_get_step_id(check);
2994 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002995 comment = tcpcheck_get_step_comment(check, step);
2996 if (comment)
2997 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002998 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2999 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003000 goto out_end_tcpcheck;
3001 }
3002
3003 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003004 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003005
3006 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003007 while (&check->current_step->list != head &&
3008 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003009 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003010
Willy Tarreauf2c87352015-05-13 12:08:21 +02003011 if (&check->current_step->list == head)
3012 break;
3013
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003014 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003015 if (conn->flags & CO_FL_WAIT_XPRT)
Willy Tarreau7df8ca62019-07-15 10:57:51 +02003016 break;
3017
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003018 } /* end 'connect' */
3019 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3020 /* mark the step as started */
3021 check->last_started_step = check->current_step;
3022
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003023 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003024 if (*b_head(&check->bi) != '\0') {
3025 *b_head(&check->bi) = '\0';
3026 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003027 }
3028
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003029 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003030 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003031 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003032 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003033 set_server_check_status(check, HCHK_STATUS_L7RSP,
3034 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003035 goto out_end_tcpcheck;
3036 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003037
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003038 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003039 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003040 continue;
3041
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003042 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3043 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003044
Willy Tarreauabca5b62013-12-06 14:19:25 +01003045 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003046 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003047
3048 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003049 while (&check->current_step->list != head &&
3050 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003051 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003052
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003053 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003054 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003055 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003056 goto out_end_tcpcheck;
3057
Olivier Houchard0ba6c852019-07-18 15:09:08 +02003058 /* If we already subscribed, then we tried to received
3059 * and failed, so there's no point trying again.
3060 */
3061 if (check->wait_list.events & SUB_RETRY_RECV)
3062 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003063 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003064 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003065 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003066 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003067 /* Report network errors only if we got no other data. Otherwise
3068 * we'll let the upper layers decide whether the response is OK
3069 * or not. It is very common that an RST sent by the server is
3070 * reported as an error just after the last data chunk.
3071 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003072 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003073 goto out_end_tcpcheck;
3074 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003075 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003076 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003077 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003078 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003079 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003080 }
3081
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003082 /* mark the step as started */
3083 check->last_started_step = check->current_step;
3084
3085
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003086 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003087 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003088 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003089 if (b_data(&check->bi) < b_size(&check->bi)) {
3090 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003091 }
3092 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003093 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003094 done = 1; /* buffer full, don't wait for more data */
3095 }
3096
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003097 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003098
3099 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003100 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003101 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003102 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003103
3104 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003105 step = tcpcheck_get_step_id(check);
3106 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003107 comment = tcpcheck_get_step_comment(check, step);
3108 if (comment)
3109 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003110 set_server_check_status(check, HCHK_STATUS_L7RSP,
3111 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003112
3113 goto out_end_tcpcheck;
3114 }
3115
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003116 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003117 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003118
Willy Tarreaua970c282013-12-06 12:47:19 +01003119 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003120 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003121 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 +02003122 else if (check->current_step->expect_regex != NULL)
3123 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003124
3125 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003126 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003127
3128 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003129 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003130 if (ret) {
3131 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003132 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003133 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003134 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003135 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003136 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003137 }
3138 else {
3139 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003140 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003141 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003142 comment = tcpcheck_get_step_comment(check, step);
3143 if (comment)
3144 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003145 set_server_check_status(check, HCHK_STATUS_L7RSP,
3146 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003147 goto out_end_tcpcheck;
3148 }
3149 /* matched and was supposed to => OK, next step */
3150 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003151 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003152 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003153
3154 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003155 while (&check->current_step->list != head &&
3156 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003157 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003158
Willy Tarreauf2c87352015-05-13 12:08:21 +02003159 if (&check->current_step->list == head)
3160 break;
3161
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003162 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003163 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003164 }
3165 }
3166 else {
3167 /* not matched */
3168 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003169 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003170 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003171 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003172
3173 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003174 while (&check->current_step->list != head &&
3175 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003176 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003177
Willy Tarreauf2c87352015-05-13 12:08:21 +02003178 if (&check->current_step->list == head)
3179 break;
3180
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003181 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003182 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003183 }
3184 /* not matched but was supposed to => ERROR */
3185 else {
3186 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003187 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003188 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003189 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003190 }
3191 else {
3192 /* we were looking for a regex */
3193 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003194 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003195 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003196 comment = tcpcheck_get_step_comment(check, step);
3197 if (comment)
3198 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003199 set_server_check_status(check, HCHK_STATUS_L7RSP,
3200 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003201 goto out_end_tcpcheck;
3202 }
3203 }
3204 } /* end expect */
3205 } /* end loop over double chained step list */
3206
Baptiste Assmann248f1172018-03-01 21:49:01 +01003207 /* don't do anything until the connection is established */
Willy Tarreau911db9b2020-01-23 16:27:54 +01003208 if (conn->flags & CO_FL_WAIT_XPRT) {
Baptiste Assmann248f1172018-03-01 21:49:01 +01003209 /* update expire time, should be done by process_chk */
3210 /* we allow up to min(inter, timeout.connect) for a connection
3211 * to establish but only when timeout.check is set
3212 * as it may be to short for a full check otherwise
3213 */
3214 while (tick_is_expired(t->expire, now_ms)) {
3215 int t_con;
3216
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003217 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003218 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3219
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003220 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003221 t->expire = tick_first(t->expire, t_con);
3222 }
3223 goto out;
3224 }
3225
Willy Tarreau263013d2015-05-13 11:59:14 +02003226 /* We're waiting for some I/O to complete, we've reached the end of the
3227 * rules, or both. Do what we have to do, otherwise we're done.
3228 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003229 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003230 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3231 goto out_end_tcpcheck;
3232 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003233
Willy Tarreau53c5a042015-05-13 11:38:17 +02003234 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003235 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003236 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003237 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003238
3239 out_end_tcpcheck:
3240 /* collect possible new errors */
Willy Tarreauef91c932019-07-23 14:37:47 +02003241 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003242 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003243
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003244 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003245 check->current_step = NULL;
3246
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003247 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003248 conn->flags |= CO_FL_ERROR;
3249
Christopher Fauletb6102852017-11-28 10:06:29 +01003250 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003251 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003252}
3253
Simon Hormanb1900d52015-01-30 11:22:54 +09003254const char *init_check(struct check *check, int type)
3255{
3256 check->type = type;
3257
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003258 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3259 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003260
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003261 check->bi.area = calloc(check->bi.size, sizeof(char));
3262 check->bo.area = calloc(check->bo.size, sizeof(char));
3263
3264 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003265 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003266
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003267 check->wait_list.tasklet = tasklet_new();
3268 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003269 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003270 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003271 check->wait_list.tasklet->process = event_srv_chk_io;
3272 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003273 return NULL;
3274}
3275
Simon Hormanbfb5d332015-01-30 11:22:55 +09003276void free_check(struct check *check)
3277{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003278 free(check->bi.area);
3279 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003280 if (check->cs) {
3281 free(check->cs->conn);
3282 check->cs->conn = NULL;
3283 cs_free(check->cs);
3284 check->cs = NULL;
3285 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003286}
3287
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003288void email_alert_free(struct email_alert *alert)
3289{
3290 struct tcpcheck_rule *rule, *back;
3291
3292 if (!alert)
3293 return;
3294
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003295 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3296 LIST_DEL(&rule->list);
3297 free(rule->comment);
3298 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003299 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003300 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003301 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003302 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003303}
3304
Olivier Houchard9f6af332018-05-25 14:04:04 +02003305static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003307 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003308 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003309 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003310
3311 q = container_of(check, typeof(*q), check);
3312
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003313 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003314 while (1) {
3315 if (!(check->state & CHK_ST_ENABLED)) {
3316 if (LIST_ISEMPTY(&q->email_alerts)) {
3317 /* All alerts processed, queue the task */
3318 t->expire = TICK_ETERNITY;
3319 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003320 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003321 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003322
3323 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003324 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003325 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003326 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003327 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003328 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003329 }
3330
Olivier Houchard9f6af332018-05-25 14:04:04 +02003331 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003332 if (check->state & CHK_ST_INPROGRESS)
3333 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003334
3335 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3336 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003337 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003338 check->server = NULL;
3339 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003340 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003341 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003342 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003343 return t;
3344}
3345
Christopher Faulet0108bb32017-10-20 21:34:32 +02003346/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3347 *
3348 * The function returns 1 in success case, otherwise, it returns 0 and err is
3349 * filled.
3350 */
3351int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003352{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003353 struct mailer *mailer;
3354 struct email_alertq *queues;
3355 const char *err_str;
3356 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003357
Christopher Faulet0108bb32017-10-20 21:34:32 +02003358 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3359 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003360 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003361 }
3362
Christopher Faulet0108bb32017-10-20 21:34:32 +02003363 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3364 struct email_alertq *q = &queues[i];
3365 struct check *check = &q->check;
3366 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003367
3368 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003369 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003370 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003371 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003372 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003373 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003374 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3375 memprintf(err, "%s", err_str);
3376 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003377 }
3378
3379 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003380 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003381 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003382
Emeric Brunc60def82017-09-27 14:59:38 +02003383 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003384 memprintf(err, "out of memory while allocating mailer alerts task");
3385 goto error;
3386 }
3387
3388 check->task = t;
3389 t->process = process_email_alert;
3390 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003391
Christopher Faulet0108bb32017-10-20 21:34:32 +02003392 /* check this in one ms */
3393 t->expire = TICK_ETERNITY;
3394 check->start = now;
3395 task_queue(t);
3396 }
3397
3398 mls->users++;
3399 free(p->email_alert.mailers.name);
3400 p->email_alert.mailers.m = mls;
3401 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003402 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003403
3404 error:
3405 for (i = 0; i < mls->count; i++) {
3406 struct email_alertq *q = &queues[i];
3407 struct check *check = &q->check;
3408
Willy Tarreauf6562792019-05-07 19:05:35 +02003409 task_destroy(check->task);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003410 free_check(check);
3411 }
3412 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003413 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003414 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003415}
3416
3417
3418static int add_tcpcheck_expect_str(struct list *list, const char *str)
3419{
3420 struct tcpcheck_rule *tcpcheck;
3421
Willy Tarreaubafbe012017-11-24 17:34:44 +01003422 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003423 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003424 memset(tcpcheck, 0, sizeof(*tcpcheck));
3425 tcpcheck->action = TCPCHK_ACT_EXPECT;
3426 tcpcheck->string = strdup(str);
3427 tcpcheck->expect_regex = NULL;
3428 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003429 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003430 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003431 return 0;
3432 }
3433
3434 LIST_ADDQ(list, &tcpcheck->list);
3435 return 1;
3436}
3437
3438static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3439{
3440 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003441 const char *in;
3442 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003443 int i;
3444
Willy Tarreaubafbe012017-11-24 17:34:44 +01003445 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003446 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003447 memset(tcpcheck, 0, sizeof(*tcpcheck));
3448 tcpcheck->action = TCPCHK_ACT_SEND;
3449 tcpcheck->expect_regex = NULL;
3450 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003451 tcpcheck->string_len = 0;
3452 for (i = 0; strs[i]; i++)
3453 tcpcheck->string_len += strlen(strs[i]);
3454
3455 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3456 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003457 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003458 return 0;
3459 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003460
Willy Tarreau64345aa2016-08-10 19:29:09 +02003461 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003462 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003463 for (in = strs[i]; (*dst = *in++); dst++);
3464 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003465
3466 LIST_ADDQ(list, &tcpcheck->list);
3467 return 1;
3468}
3469
Christopher Faulet0108bb32017-10-20 21:34:32 +02003470static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3471 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003472{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003473 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003474 struct tcpcheck_rule *tcpcheck;
3475 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003476
Willy Tarreaubafbe012017-11-24 17:34:44 +01003477 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003479 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003480 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003481 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003482
Willy Tarreaubafbe012017-11-24 17:34:44 +01003483 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003484 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003485 memset(tcpcheck, 0, sizeof(*tcpcheck));
3486 tcpcheck->action = TCPCHK_ACT_CONNECT;
3487 tcpcheck->comment = NULL;
3488 tcpcheck->string = NULL;
3489 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003490 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3491
3492 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3493 goto error;
3494
3495 {
3496 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3497 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3498 goto error;
3499 }
3500
3501 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3502 goto error;
3503
3504 {
3505 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3506 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3507 goto error;
3508 }
3509
3510 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3511 goto error;
3512
3513 {
3514 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\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[2] = { "DATA\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, "354 "))
3529 goto error;
3530
3531 {
3532 struct tm tm;
3533 char datestr[48];
3534 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003535 "From: ", p->email_alert.from, "\r\n",
3536 "To: ", p->email_alert.to, "\r\n",
3537 "Date: ", datestr, "\r\n",
3538 "Subject: [HAproxy Alert] ", msg, "\r\n",
3539 "\r\n",
3540 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003541 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003542 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003543 NULL
3544 };
3545
3546 get_localtime(date.tv_sec, &tm);
3547
3548 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3549 goto error;
3550 }
3551
3552 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3553 goto error;
3554 }
3555
3556 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3557 goto error;
3558
3559 {
3560 const char * const strs[2] = { "QUIT\r\n" };
3561 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3562 goto error;
3563 }
3564
3565 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3566 goto error;
3567
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003568 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003569 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003570 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003571 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003572 return 1;
3573
3574error:
3575 email_alert_free(alert);
3576 return 0;
3577}
3578
Christopher Faulet0108bb32017-10-20 21:34:32 +02003579static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003580{
3581 int i;
3582 struct mailer *mailer;
3583
3584 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3585 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003586 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003587 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003588 return;
3589 }
3590 }
3591
3592 return;
3593}
3594
3595/*
3596 * Send email alert if configured.
3597 */
Simon Horman64e34162015-02-06 11:11:57 +09003598void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003599{
3600 va_list argp;
3601 char buf[1024];
3602 int len;
3603 struct proxy *p = s->proxy;
3604
Christopher Faulet0108bb32017-10-20 21:34:32 +02003605 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003606 return;
3607
3608 va_start(argp, format);
3609 len = vsnprintf(buf, sizeof(buf), format, argp);
3610 va_end(argp);
3611
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003612 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003613 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003614 return;
3615 }
3616
Christopher Faulet0108bb32017-10-20 21:34:32 +02003617 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003618}
3619
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003620/*
3621 * Return value:
3622 * the port to be used for the health check
3623 * 0 in case no port could be found for the check
3624 */
3625int srv_check_healthcheck_port(struct check *chk)
3626{
3627 int i = 0;
3628 struct server *srv = NULL;
3629
3630 srv = chk->server;
3631
3632 /* If neither a port nor an addr was specified and no check transport
3633 * layer is forced, then the transport layer used by the checks is the
3634 * same as for the production traffic. Otherwise we use raw_sock by
3635 * default, unless one is specified.
3636 */
3637 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003638 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003639 chk->send_proxy |= (srv->pp_opts);
3640 }
3641
3642 /* by default, we use the health check port ocnfigured */
3643 if (chk->port > 0)
3644 return chk->port;
3645
3646 /* try to get the port from check_core.addr if check.port not set */
3647 i = get_host_port(&chk->addr);
3648 if (i > 0)
3649 return i;
3650
3651 /* try to get the port from server address */
3652 /* prevent MAPPORTS from working at this point, since checks could
3653 * not be performed in such case (MAPPORTS impose a relative ports
3654 * based on live traffic)
3655 */
3656 if (srv->flags & SRV_F_MAPPORTS)
3657 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003658
3659 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003660 if (i > 0)
3661 return i;
3662
3663 return 0;
3664}
3665
Willy Tarreau172f5ce2018-11-26 11:21:50 +01003666REGISTER_POST_CHECK(start_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003667
Willy Tarreaubd741542010-03-16 18:46:54 +01003668/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003669 * Local variables:
3670 * c-indent-level: 8
3671 * c-basic-offset: 8
3672 * End:
3673 */