blob: ead3f7f111b377cad8339eb5a898ced76b828f4a [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * Health-checks functions.
3 *
Willy Tarreau26c25062009-03-08 09:38:41 +01004 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02005 * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl>
Willy Tarreaubaaee002006-06-26 02:48:02 +02006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 */
13
Willy Tarreaub8816082008-01-18 12:18:15 +010014#include <assert.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020015#include <ctype.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020016#include <errno.h>
17#include <fcntl.h>
Willy Tarreau9b39dc52014-07-08 00:54:10 +020018#include <signal.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090019#include <stdarg.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020020#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020022#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020023#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <unistd.h>
Willy Tarreau9f6dc722019-03-01 11:15:10 +010025#include <sys/resource.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020026#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090028#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020029#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020030#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020031#include <arpa/inet.h>
32
Willy Tarreauc7e42382012-08-24 19:22:53 +020033#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/compat.h>
35#include <common/config.h>
36#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020037#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020038#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020039#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040
41#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020042#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010043#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020044
45#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020046#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010047#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020048#include <proto/fd.h>
49#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020050#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020052#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020063#include <proto/ssl_sock.h>
Olivier Houchard9130a962017-10-17 17:33:43 +020064
Willy Tarreaubd741542010-03-16 18:46:54 +010065static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090066static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020067static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020068static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020069static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020070static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020071static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Willy Tarreau8ceae722018-11-26 11:58:30 +010073DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert));
74DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule));
Christopher Faulet31dff9b2017-10-23 15:45:20 +020075
76
Simon Horman63a4a822012-03-19 07:24:41 +090077static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010078 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
79 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020080 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020081
Willy Tarreau23964182014-05-20 20:56:30 +020082 /* Below we have finished checks */
83 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010085
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
89 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
90 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
93 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
94 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020095
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010096 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
97 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020098
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020099 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100101 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
102 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
103 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900104
105 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
106 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200107 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200108};
109
Cyril Bontéac92a062014-12-27 22:28:38 +0100110const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
111 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
Christopher Faulet8d18db72020-04-26 09:50:31 +0200118 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_ADDR },
119 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_UINT },
Cyril Bontéac92a062014-12-27 22:28:38 +0100120 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
122};
123
Simon Horman63a4a822012-03-19 07:24:41 +0900124static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100125 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
126
127 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
128 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
129
130 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
131 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
132 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
133 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
134
135 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
136 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
137 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
138};
139
Willy Tarreaufdc79b92019-12-27 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));
Willy Tarreaud027f8c2021-02-17 15:15:15 +0100476 if (tick_is_lt(expire, s->check.task->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 Tarreaufdc79b92019-12-27 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 Tarreaufdc79b92019-12-27 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 Tarreaufdc79b92019-12-27 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 Tarreaufdc79b92019-12-27 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 Tarreaufdc79b92019-12-27 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 }
686 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == 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 }
701 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
702 /* 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 Houchard81badc42019-07-09 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 Houchard81badc42019-07-09 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
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200763 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100764 if (!(conn->flags & CO_FL_ERROR))
765 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200766 goto out;
767 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100768
Willy Tarreau20a18342013-12-05 00:31:46 +0100769 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200770 chk_report_conn_err(check, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100771 goto out_wakeup;
772 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100773
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200774 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100775 /* if the output is closed, we can't do anything */
776 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200777 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100778 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200779 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200780
Willy Tarreau06559ac2013-12-05 01:53:08 +0100781 /* here, we know that the connection is established. That's enough for
782 * a pure TCP check.
783 */
784 if (!check->type)
785 goto out_wakeup;
786
Willy Tarreauc09572f2017-10-04 11:58:22 +0200787 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100788 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200789 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200790
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200791 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200792 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200793 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200794 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200795 chk_report_conn_err(check, errno, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100796 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200797 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200798 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100799 conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200800 goto out;
801 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100802 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200803
Willy Tarreau100cc4b2019-09-05 18:24:46 +0200804 if (!b_data(&check->bo))
805 conn_xprt_stop_send(conn);
806
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100807 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
808 if (s->proxy->timeout.check) {
809 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
810 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200811 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200812 goto out;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100813
Willy Tarreau83749182007-04-15 20:56:27 +0200814 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200815 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200816 out:
817 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200818}
819
Willy Tarreaubaaee002006-06-26 02:48:02 +0200820/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200821 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200822 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900823 * set_server_check_status() to update check->status, check->duration
824 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200825
826 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
827 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
828 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
829 * response to an SSL HELLO (the principle is that this is enough to
830 * distinguish between an SSL server and a pure TCP relay). All other cases will
831 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
832 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100833 *
834 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200835 * via the out label.
836 *
837 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200838 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200839static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200840{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200841 struct connection *conn = cs->conn;
842 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900843 struct server *s = check->server;
844 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200845 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100846 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200847 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200848
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100849 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200850 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200851
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200852 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard06f68112019-03-28 17:32:42 +0100853 if (!(conn->flags & CO_FL_ERROR))
854 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200855 goto out;
856 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200857
Willy Tarreauc09572f2017-10-04 11:58:22 +0200858 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100859 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200860 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200861
Willy Tarreau83749182007-04-15 20:56:27 +0200862 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
863 * but the connection was closed on the remote end. Fortunately, recv still
864 * works correctly and we don't need to do the getsockopt() on linux.
865 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000866
867 /* Set buffer to point to the end of the data already read, and check
868 * that there is free space remaining. If the buffer is full, proceed
869 * with running the checks without attempting another socket read.
870 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000871
Willy Tarreau03938182010-03-17 21:52:07 +0100872 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000873
Olivier Houchard511efea2018-08-16 15:30:32 +0200874 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200875 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100876 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200877 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200878 /* Report network errors only if we got no other data. Otherwise
879 * we'll let the upper layers decide whether the response is OK
880 * or not. It is very common that an RST sent by the server is
881 * reported as an error just after the last data chunk.
882 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200883 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100884 goto out_wakeup;
885 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200886 }
887
Willy Tarreau489bbd32019-09-24 10:43:03 +0200888 /* the rest of the code below expects the connection to be ready! */
889 if (!(conn->flags & CO_FL_CONNECTED) && !done)
890 goto wait_more_data;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100891
Willy Tarreau03938182010-03-17 21:52:07 +0100892 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200893 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100894 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200895 if (b_data(&check->bi) < b_size(&check->bi))
896 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100897 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200898 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100899 done = 1; /* buffer full, don't wait for more data */
900 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200901
Nick Chalk57b1bf72010-03-16 15:50:46 +0000902 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900903 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200904 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200905 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100906 goto wait_more_data;
907
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100908 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200909 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
910 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
911 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
912 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
913 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
914 cut_crlf(b_head(&check->bi));
915 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200916
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100917 goto out_wakeup;
918 }
919
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200920 check->code = str2uic(b_head(&check->bi) + 9);
921 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200922
Willy Tarreaubd741542010-03-16 18:46:54 +0100923 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200924 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000925 /* 404 may be accepted as "stopping" only if the server was up */
926 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000928 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100929 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
930 /* Run content verification check... We know we have at least 13 chars */
931 if (!httpchk_expect(s, done))
932 goto wait_more_data;
933 }
934 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200935 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100936 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900937 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100938 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000939 else {
940 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900941 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000942 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200943 break;
944
945 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200946 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100947 goto wait_more_data;
948
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100949 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200950 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900951 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200952 else
Simon Horman4a741432013-02-23 15:35:38 +0900953 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200954 break;
955
956 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200957 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100958 goto wait_more_data;
959
Willy Tarreau55706782019-08-06 16:26:31 +0200960 /* do not reset when closing, servers don't like this */
961 if (conn_ctrl_ready(cs->conn))
962 fdtab[cs->conn->handle.fd].linger_risk = 0;
963
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200964 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200965 if ((b_data(&check->bi) < strlen("000\r")) ||
966 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
967 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
968 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
969 cut_crlf(b_head(&check->bi));
970 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200971 goto out_wakeup;
972 }
973
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200974 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200975
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200976 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200977 cut_crlf(desc);
978
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100979 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200980 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900981 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200982 else
Simon Horman4a741432013-02-23 15:35:38 +0900983 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200984 break;
985
Simon Hormana2b9dad2013-02-12 10:45:54 +0900986 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100987 int status = HCHK_STATUS_CHECKED;
988 const char *hs = NULL; /* health status */
989 const char *as = NULL; /* admin status */
990 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200991 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100992 const char *err = NULL; /* first error to report */
993 const char *wrn = NULL; /* first warning to report */
994 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900995
Willy Tarreau81f5d942013-12-09 20:51:51 +0100996 /* We're getting an agent check response. The agent could
997 * have been disabled in the mean time with a long check
998 * still pending. It is important that we ignore the whole
999 * response.
1000 */
1001 if (!(check->server->agent.state & CHK_ST_ENABLED))
1002 break;
1003
1004 /* The agent supports strings made of a single line ended by the
1005 * first CR ('\r') or LF ('\n'). This line is composed of words
1006 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
1007 * line may optionally contained a description of a state change
1008 * after a sharp ('#'), which is only considered if a health state
1009 * is announced.
1010 *
1011 * Words may be composed of :
1012 * - a numeric weight suffixed by the percent character ('%').
1013 * - a health status among "up", "down", "stopped", and "fail".
1014 * - an admin status among "ready", "drain", "maint".
1015 *
1016 * These words may appear in any order. If multiple words of the
1017 * same category appear, the last one wins.
1018 */
1019
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001020 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +01001021 while (*p && *p != '\n' && *p != '\r')
1022 p++;
1023
1024 if (!*p) {
1025 if (!done)
1026 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001027
Willy Tarreau9809b782013-12-11 21:40:11 +01001028 /* at least inform the admin that the agent is mis-behaving */
1029 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1030 break;
1031 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001032
Willy Tarreau9809b782013-12-11 21:40:11 +01001033 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001034 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001035
Willy Tarreau81f5d942013-12-09 20:51:51 +01001036 while (*cmd) {
1037 /* look for next word */
1038 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1039 cmd++;
1040 continue;
1041 }
Simon Horman671b6f02013-11-25 10:46:39 +09001042
Willy Tarreau81f5d942013-12-09 20:51:51 +01001043 if (*cmd == '#') {
1044 /* this is the beginning of a health status description,
1045 * skip the sharp and blanks.
1046 */
1047 cmd++;
1048 while (*cmd == '\t' || *cmd == ' ')
1049 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001050 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001051 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001052
1053 /* find the end of the word so that we have a null-terminated
1054 * word between <cmd> and <p>.
1055 */
1056 p = cmd + 1;
1057 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1058 p++;
1059 if (*p)
1060 *p++ = 0;
1061
1062 /* first, health statuses */
1063 if (strcasecmp(cmd, "up") == 0) {
1064 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001065 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001066 hs = cmd;
1067 }
1068 else if (strcasecmp(cmd, "down") == 0) {
1069 check->health = 0;
1070 status = HCHK_STATUS_L7STS;
1071 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001072 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001073 else if (strcasecmp(cmd, "stopped") == 0) {
1074 check->health = 0;
1075 status = HCHK_STATUS_L7STS;
1076 hs = cmd;
1077 }
1078 else if (strcasecmp(cmd, "fail") == 0) {
1079 check->health = 0;
1080 status = HCHK_STATUS_L7STS;
1081 hs = cmd;
1082 }
1083 /* admin statuses */
1084 else if (strcasecmp(cmd, "ready") == 0) {
1085 as = cmd;
1086 }
1087 else if (strcasecmp(cmd, "drain") == 0) {
1088 as = cmd;
1089 }
1090 else if (strcasecmp(cmd, "maint") == 0) {
1091 as = cmd;
1092 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001093 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001094 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1095 ps = cmd;
1096 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001097 /* try to parse a maxconn here */
1098 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1099 cs = cmd;
1100 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001101 else {
1102 /* keep a copy of the first error */
1103 if (!err)
1104 err = cmd;
1105 }
1106 /* skip to next word */
1107 cmd = p;
1108 }
1109 /* here, cmd points either to \0 or to the beginning of a
1110 * description. Skip possible leading spaces.
1111 */
1112 while (*cmd == ' ' || *cmd == '\n')
1113 cmd++;
1114
1115 /* First, update the admin status so that we avoid sending other
1116 * possibly useless warnings and can also update the health if
1117 * present after going back up.
1118 */
1119 if (as) {
1120 if (strcasecmp(as, "drain") == 0)
1121 srv_adm_set_drain(check->server);
1122 else if (strcasecmp(as, "maint") == 0)
1123 srv_adm_set_maint(check->server);
1124 else
1125 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001126 }
1127
Willy Tarreau81f5d942013-12-09 20:51:51 +01001128 /* now change weights */
1129 if (ps) {
1130 const char *msg;
1131
1132 msg = server_parse_weight_change_request(s, ps);
1133 if (!wrn || !*wrn)
1134 wrn = msg;
1135 }
1136
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001137 if (cs) {
1138 const char *msg;
1139
1140 cs += strlen("maxconn:");
1141
1142 msg = server_parse_maxconn_change_request(s, cs);
1143 if (!wrn || !*wrn)
1144 wrn = msg;
1145 }
1146
Willy Tarreau81f5d942013-12-09 20:51:51 +01001147 /* and finally health status */
1148 if (hs) {
1149 /* We'll report some of the warnings and errors we have
1150 * here. Down reports are critical, we leave them untouched.
1151 * Lack of report, or report of 'UP' leaves the room for
1152 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001153 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001154 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001155 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001156
1157 if (!*msg || status == HCHK_STATUS_L7OKD) {
1158 if (err && *err)
1159 msg = err;
1160 else if (wrn && *wrn)
1161 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001162 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001163
1164 t = get_trash_chunk();
1165 chunk_printf(t, "via agent : %s%s%s%s",
1166 hs, *msg ? " (" : "",
1167 msg, *msg ? ")" : "");
1168
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001169 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001170 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001171 else if (err && *err) {
1172 /* No status change but we'd like to report something odd.
1173 * Just report the current state and copy the message.
1174 */
1175 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001176 set_server_check_status(check, status/*check->status*/,
1177 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001178
Willy Tarreau81f5d942013-12-09 20:51:51 +01001179 }
1180 else if (wrn && *wrn) {
1181 /* No status change but we'd like to report something odd.
1182 * Just report the current state and copy the message.
1183 */
1184 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001185 set_server_check_status(check, status/*check->status*/,
1186 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001187 }
1188 else
1189 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001190 break;
1191 }
1192
Willy Tarreau1620ec32011-08-06 17:05:02 +02001193 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001194 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001195 goto wait_more_data;
1196
Willy Tarreau55706782019-08-06 16:26:31 +02001197 /* do not reset when closing, servers don't like this */
1198 if (conn_ctrl_ready(cs->conn))
1199 fdtab[cs->conn->handle.fd].linger_risk = 0;
1200
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001201 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001202 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001203 }
1204 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1206 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001207 else
1208 desc = "PostgreSQL unknown error";
1209
Simon Horman4a741432013-02-23 15:35:38 +09001210 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001211 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001212 break;
1213
1214 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001215 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001216 goto wait_more_data;
1217
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001218 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001219 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001220 }
1221 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001222 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001223 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001224 break;
1225
1226 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001227 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001228 goto wait_more_data;
1229
Willy Tarreau55706782019-08-06 16:26:31 +02001230 /* do not reset when closing, servers don't like this */
1231 if (conn_ctrl_ready(cs->conn))
1232 fdtab[cs->conn->handle.fd].linger_risk = 0;
1233
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001234 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001235 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001236 /* We set the MySQL Version in description for information purpose
1237 * FIXME : it can be cool to use MySQL Version for other purpose,
1238 * like mark as down old MySQL server.
1239 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001240 if (b_data(&check->bi) > 51) {
1241 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001242 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001243 }
1244 else {
1245 if (!done)
1246 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001247
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001248 /* it seems we have a OK packet but without a valid length,
1249 * it must be a protocol error
1250 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001251 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001252 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001253 }
1254 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001255 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001256 desc = ltrim(b_head(&check->bi) + 7, ' ');
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 }
1259 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001260 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1261 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1262 (((unsigned int) *(b_head(&check->bi) + 2)) << 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) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001265 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001266 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001267 /* We have only one MySQL packet and it is a Handshake Initialization packet
1268 * but we need to have a second packet to know if it is alright
1269 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001270 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001271 goto wait_more_data;
1272 }
1273 else {
1274 /* We have only one packet and it is an Error packet,
1275 * an error message is attached, so we can display it
1276 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001277 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001278 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001279 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001280 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001281 } else if (b_data(&check->bi) > first_packet_len + 4) {
1282 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1283 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1284 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001285
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001286 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001287 /* We have 2 packets and that's good */
1288 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001289 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001290 /* No error packet */
1291 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001292 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001293 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001294 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001295 }
1296 else {
1297 /* An error message is attached in the Error packet
1298 * so we can display it ! :)
1299 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001300 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001301 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001302 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001303 }
1304 }
1305 }
1306 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001307 if (!done)
1308 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001309
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001310 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001311 * it must be a protocol error
1312 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001313 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001314 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001315 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001316 }
1317 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001318 break;
1319
1320 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001321 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001322 goto wait_more_data;
1323
1324 /* Check if the server speaks LDAP (ASN.1/BER)
1325 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1326 * http://tools.ietf.org/html/rfc4511
1327 */
1328
1329 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1330 * LDAPMessage: 0x30: SEQUENCE
1331 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001332 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001333 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001334 }
1335 else {
1336 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001337 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001338
1339 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1340 * messageID: 0x02 0x01 0x01: INTEGER 1
1341 * protocolOp: 0x61: bindResponse
1342 */
1343 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001344 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001345 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001346 goto out_wakeup;
1347 }
1348
1349 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001350 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001351
1352 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1353 * ldapResult: 0x0a 0x01: ENUMERATION
1354 */
1355 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001356 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001357 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001358 goto out_wakeup;
1359 }
1360
1361 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1362 * resultCode
1363 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001364 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001365 if (check->code) {
1366 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 +02001367 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001368 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001369 }
1370 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001371 break;
1372
Christopher Fauletba7bc162016-11-07 21:07:38 +01001373 case PR_O2_SPOP_CHK: {
1374 unsigned int framesz;
1375 char err[HCHK_DESC_LEN];
1376
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001377 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001378 goto wait_more_data;
1379
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001380 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001381 framesz = ntohl(framesz);
1382
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001383 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001384 goto wait_more_data;
1385
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001386 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001387 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1388 else
1389 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1390 break;
1391 }
1392
Willy Tarreau1620ec32011-08-06 17:05:02 +02001393 default:
Willy Tarreaufbeee552019-09-05 18:43:22 +02001394 /* good connection is enough for pure TCP check */
1395 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Christopher Faulet68d35ae2020-03-27 18:55:49 +01001396 if (check->use_ssl == 1)
Willy Tarreaufbeee552019-09-05 18:43:22 +02001397 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
1398 else
1399 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
1400 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001401 break;
1402 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001403
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001404 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001405 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001406 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001407 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001408
Nick Chalk57b1bf72010-03-16 15:50:46 +00001409 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001410 *b_head(&check->bi) = '\0';
1411 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001412
Steven Davidovitz544d4812017-03-08 11:06:20 -08001413 /* Close the connection... We still attempt to nicely close if,
1414 * for instance, SSL needs to send a "close notify." Later, we perform
1415 * a hard close and reset the connection if some data are pending,
1416 * otherwise we end up with many TIME_WAITs and eat all the source port
1417 * range quickly. To avoid sending RSTs all the time, we first try to
1418 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001419 */
Olivier Houchard8aa445d2019-07-02 16:35:18 +02001420 /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the
1421 * connection, to make sure cs_shutw() will not lead to a shutdown()
1422 * that would provoke TIME_WAITs.
1423 */
1424 cs_shutr(cs, CS_SHR_DRAIN);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001425 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001426
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001427 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001428 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001429 conn->flags |= CO_FL_ERROR;
1430
Willy Tarreaufdccded2008-08-29 18:19:04 +02001431 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001432out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001433 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001434
1435 wait_more_data:
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001436 cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001437 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001438}
1439
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001440/*
1441 * This function is used only for server health-checks. It handles connection
1442 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001443 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1444 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001445 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001446static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001447{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001448 struct connection *conn = cs->conn;
1449 struct check *check = cs->data;
Olivier Houchard0923fa42019-01-11 18:43:04 +01001450 struct email_alertq *q = container_of(check, typeof(*q), check);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001451 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001452
Olivier Houchard0923fa42019-01-11 18:43:04 +01001453 if (check->server)
1454 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
1455 else
1456 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001457
Willy Tarreauc09572f2017-10-04 11:58:22 +02001458 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001459 if (check->type == PR_O2_TCPCHK_CHK) {
1460 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001461 cs = check->cs;
Willy Tarreau543abd42018-09-20 11:25:12 +02001462 conn = cs->conn;
Willy Tarreauab160a42019-09-05 17:38:40 +02001463 } else {
1464 if (!(check->wait_list.events & SUB_RETRY_SEND))
1465 __event_srv_chk_w(cs);
1466 if (!(check->wait_list.events & SUB_RETRY_RECV))
1467 __event_srv_chk_r(cs);
1468 }
Willy Tarreauc09572f2017-10-04 11:58:22 +02001469
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001470 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001471 /* We may get error reports bypassing the I/O handlers, typically
1472 * the case when sending a pure TCP check which fails, then the I/O
1473 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001474 * main processing task so let's simply wake it up. If we get here,
1475 * we expect errno to still be valid.
1476 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001477 chk_report_conn_err(check, errno, 0);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001478 task_wakeup(check->task, TASK_WOKEN_IO);
1479 }
Olivier Houchard53216e72018-10-10 15:46:36 +02001480 else if (!(conn->flags & CO_FL_HANDSHAKE) && !check->type) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001481 /* we may get here if only a connection probe was required : we
1482 * don't have any data to send nor anything expected in response,
1483 * so the completion of the connection establishment is enough.
1484 */
1485 task_wakeup(check->task, TASK_WOKEN_IO);
1486 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001487
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001488 if (check->result != CHK_RES_UNKNOWN) {
Christopher Faulet774c4862019-01-21 14:15:50 +01001489 /* Check complete or aborted. If connection not yet closed do it
1490 * now and wake the check task up to be sure the result is
1491 * handled ASAP. */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001492 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001493 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001494 ret = -1;
Olivier Houchard865ed9b2019-07-02 17:42:22 +02001495 /* We may have been scheduled to run, and the
1496 * I/O handler expects to have a cs, so remove
1497 * the tasklet
1498 */
1499 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Christopher Faulet774c4862019-01-21 14:15:50 +01001500 task_wakeup(check->task, TASK_WOKEN_IO);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001501 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001502
Olivier Houchard0923fa42019-01-11 18:43:04 +01001503 if (check->server)
1504 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
1505 else
1506 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001507
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001508 /* if a connection got replaced, we must absolutely prevent the connection
1509 * handler from touching its fd, and perform the FD polling updates ourselves
1510 */
1511 if (ret < 0)
1512 conn_cond_update_polling(conn);
1513
1514 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001515}
1516
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001517struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001518 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001519 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001520};
1521
Willy Tarreaubaaee002006-06-26 02:48:02 +02001522/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001523 * updates the server's weight during a warmup stage. Once the final weight is
1524 * reached, the task automatically stops. Note that any server status change
1525 * must have updated s->last_change accordingly.
1526 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001527static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001528{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001529 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001530
1531 /* by default, plan on stopping the task */
1532 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001533 if ((s->next_admin & SRV_ADMF_MAINT) ||
1534 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001535 return t;
1536
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001537 HA_SPIN_LOCK(SERVER_LOCK, &s->lock);
1538
Willy Tarreau892337c2014-05-13 23:41:20 +02001539 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001540 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001541
1542 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001543 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001544
Willy Tarreau4fc49a92019-05-05 06:54:22 +02001545 HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock);
1546
Willy Tarreau2e993902011-10-31 11:53:20 +01001547 /* get back there in 1 second or 1/20th of the slowstart interval,
1548 * whichever is greater, resulting in small 5% steps.
1549 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001550 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001551 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1552 return t;
1553}
1554
Willy Tarreau894c6422017-10-04 15:58:52 +02001555/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1556 * none was found.
1557 */
1558static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1559{
1560 struct tcpcheck_rule *r;
1561
1562 list_for_each_entry(r, list, list) {
1563 if (r->action != TCPCHK_ACT_COMMENT)
1564 return r;
1565 }
1566 return NULL;
1567}
1568
Willy Tarreau2e993902011-10-31 11:53:20 +01001569/*
Simon Horman98637e52014-06-20 12:30:16 +09001570 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001571 *
1572 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001573 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1574 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1575 * - SF_ERR_SRVTO if there are no more servers
1576 * - SF_ERR_SRVCL if the connection was refused by the server
1577 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1578 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1579 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001580 * - 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 +01001581 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001582 * Note that we try to prevent the network stack from sending the ACK during the
1583 * connect() when a pure TCP check is used (without PROXY protocol).
1584 */
Simon Horman98637e52014-06-20 12:30:16 +09001585static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001586{
1587 struct check *check = t->context;
1588 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001589 struct conn_stream *cs = check->cs;
1590 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001591 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001592 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001593 int ret;
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001594 int connflags = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001595
Willy Tarreau00149122017-10-04 18:05:01 +02001596 /* we cannot have a connection here */
1597 if (conn)
1598 return SF_ERR_INTERNAL;
1599
Simon Hormanb00d17a2014-06-13 16:18:16 +09001600 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001601 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001602 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001603 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1604 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001605
1606 /* prepare the check buffer.
1607 * This should not be used if check is the secondary agent check
1608 * of a server as s->proxy->check_req will relate to the
1609 * configuration of the primary check. Similarly, tcp-check uses
1610 * its own strings.
1611 */
1612 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001613 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001614
1615 /* we want to check if this host replies to HTTP or SSLv3 requests
1616 * so we'll send the request, and won't wake the checker up now.
1617 */
1618 if ((check->type) == PR_O2_SSL3_CHK) {
1619 /* SSL requires that we put Unix time in the request */
1620 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001621 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001622 }
1623 else if ((check->type) == PR_O2_HTTP_CHK) {
Cyril Bonté32602d22015-01-30 00:07:07 +01001624 /* prevent HTTP keep-alive when "http-check expect" is used */
1625 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001626 b_putist(&check->bo, ist("Connection: close\r\n"));
Christopher Fauletf304ad32020-04-09 08:44:06 +02001627
1628 /* If there is a body, add its content-length */
1629 if (s->proxy->check_body_len)
1630 chunk_appendf(&check->bo, "Content-Length: %s\r\n", ultoa(s->proxy->check_body_len));
1631
1632 /* Add configured headers */
1633 if (s->proxy->check_hdrs)
1634 b_putblk(&check->bo, s->proxy->check_hdrs, s->proxy->check_hdrs_len);
1635
1636 /* Add send-state header */
1637 if (s->proxy->options2 & PR_O2_CHK_SNDST)
1638 b_putblk(&check->bo, trash.area,
1639 httpchk_build_status_header(s, trash.area, trash.size));
1640
1641 /* end-of-header */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001642 b_putist(&check->bo, ist("\r\n"));
Christopher Fauletf304ad32020-04-09 08:44:06 +02001643
1644 /* Add the body */
1645 if (s->proxy->check_body)
1646 b_putblk(&check->bo, s->proxy->check_body, s->proxy->check_body_len);
1647
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001648 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001649 }
1650 }
1651
James Brown55f9ff12015-10-21 18:19:05 -07001652 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001653 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001654 }
1655
Willy Tarreauf411cce2017-10-04 16:21:19 +02001656 /* for tcp-checks, the initial connection setup is handled separately as
1657 * it may be sent to a specific port and not to the server's.
1658 */
1659 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1660 tcpcheck_main(check);
1661 return SF_ERR_UP;
1662 }
1663
Simon Hormanb00d17a2014-06-13 16:18:16 +09001664 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001665 cs = check->cs = cs_new(NULL);
1666 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001667 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001668 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001669 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01001670 check->wait_list.events = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001671
Simon Horman41f58762015-01-30 11:22:56 +09001672 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001673 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001674 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001675 }
1676 else {
1677 /* we'll connect to the addr on the server */
1678 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001679 }
1680
Alexander Liu2a54bb72019-05-22 19:44:48 +08001681 if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) {
1682 conn->send_proxy_ofs = 1;
1683 conn->flags |= CO_FL_SOCKS4;
1684 }
1685
Olivier Houchard6377a002017-12-01 22:04:05 +01001686 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001687 conn->target = &s->obj_type;
1688
Willy Tarreaube373152018-09-06 11:45:30 +02001689 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001690 int i = 0;
1691
1692 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001693 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001694 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001695
1696 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001697 }
1698
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001699 /* no client address */
1700 clear_addr(&conn->addr.from);
1701
Willy Tarreaube373152018-09-06 11:45:30 +02001702 conn_prepare(conn, proto, check->xprt);
Olivier Houchardf67be932019-01-29 15:47:43 +01001703 if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0)
1704 return SF_ERR_RESOURCE;
Willy Tarreaube373152018-09-06 11:45:30 +02001705 cs_attach(cs, check, &check_conn_cb);
1706
Willy Tarreauf3d34822014-12-08 12:11:28 +01001707 /* only plain tcp-check supports quick ACK */
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001708 if (check->type != 0)
1709 connflags |= CONNECT_HAS_DATA;
1710 if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) &&
1711 (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT))
1712 connflags |= CONNECT_DELACK_ALWAYS;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001713
Willy Tarreaue7dff022015-04-03 01:14:29 +02001714 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001715 if (proto && proto->connect)
Olivier Houchardfdcb0072019-05-06 18:32:29 +02001716 ret = proto->connect(conn, connflags);
Willy Tarreau16257f62017-11-02 15:45:00 +01001717
Willy Tarreau16257f62017-11-02 15:45:00 +01001718
Olivier Houchard9130a962017-10-17 17:33:43 +02001719#ifdef USE_OPENSSL
Olivier Houcharda48437b2019-01-29 16:37:52 +01001720 if (ret == SF_ERR_NONE) {
1721 if (s->check.sni)
1722 ssl_sock_set_servername(conn, s->check.sni);
1723 if (s->check.alpn_str)
1724 ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str,
1725 s->check.alpn_len);
1726 }
Olivier Houchard9130a962017-10-17 17:33:43 +02001727#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001728 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001729 conn->send_proxy_ofs = 1;
1730 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardb4c5e932019-12-30 15:13:42 +01001731 }
1732 if (conn->flags & (CO_FL_SEND_PROXY | CO_FL_SOCKS4) &&
1733 conn_ctrl_ready(conn)) {
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02001734 if (xprt_add_hs(conn) < 0)
1735 ret = SF_ERR_RESOURCE;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001736 }
1737
1738 return ret;
1739}
1740
Simon Horman98637e52014-06-20 12:30:16 +09001741static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001742static struct pool_head *pool_head_pid_list;
Willy Tarreau86abe442018-11-25 20:12:18 +01001743__decl_spinlock(pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001744
1745void block_sigchld(void)
1746{
1747 sigset_t set;
1748 sigemptyset(&set);
1749 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001750 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001751}
1752
1753void unblock_sigchld(void)
1754{
1755 sigset_t set;
1756 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001757 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001758 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001759}
1760
Simon Horman98637e52014-06-20 12:30:16 +09001761static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1762{
1763 struct pid_list *elem;
1764 struct check *check = t->context;
1765
Willy Tarreaubafbe012017-11-24 17:34:44 +01001766 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001767 if (!elem)
1768 return NULL;
1769 elem->pid = pid;
1770 elem->t = t;
1771 elem->exited = 0;
1772 check->curpid = elem;
1773 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001774
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001775 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001776 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001777 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001778
Simon Horman98637e52014-06-20 12:30:16 +09001779 return elem;
1780}
1781
Simon Horman98637e52014-06-20 12:30:16 +09001782static void pid_list_del(struct pid_list *elem)
1783{
1784 struct check *check;
1785
1786 if (!elem)
1787 return;
1788
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001789 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001790 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001791 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001792
Simon Horman98637e52014-06-20 12:30:16 +09001793 if (!elem->exited)
1794 kill(elem->pid, SIGTERM);
1795
1796 check = elem->t->context;
1797 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001798 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001799}
1800
1801/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1802static void pid_list_expire(pid_t pid, int status)
1803{
1804 struct pid_list *elem;
1805
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001806 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001807 list_for_each_entry(elem, &pid_list, list) {
1808 if (elem->pid == pid) {
1809 elem->t->expire = now_ms;
1810 elem->status = status;
1811 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001812 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001813 break;
Simon Horman98637e52014-06-20 12:30:16 +09001814 }
1815 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001816 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001817}
1818
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001819static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001820{
1821 pid_t pid;
1822 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001823
Simon Horman98637e52014-06-20 12:30:16 +09001824 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1825 pid_list_expire(pid, status);
1826}
1827
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001828static int init_pid_list(void)
1829{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001830 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001831 /* Nothing to do */
1832 return 0;
1833
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001834 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001835 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1836 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001837 return 1;
1838 }
1839
Willy Tarreaubafbe012017-11-24 17:34:44 +01001840 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1841 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001842 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1843 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001844 return 1;
1845 }
1846
1847 return 0;
1848}
1849
Cyril Bontéac92a062014-12-27 22:28:38 +01001850/* helper macro to set an environment variable and jump to a specific label on failure. */
1851#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001852
1853/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001854 * helper function to allocate enough memory to store an environment variable.
1855 * It will also check that the environment variable is updatable, and silently
1856 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001857 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001858static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001859{
1860 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001861 char *envname;
1862 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001863
Cyril Bontéac92a062014-12-27 22:28:38 +01001864 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001865 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001866 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001867 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001868
1869 envname = extcheck_envs[idx].name;
1870 vmaxlen = extcheck_envs[idx].vmaxlen;
1871
1872 /* Check if the environment variable is already set, and silently reject
1873 * the update if this one is not updatable. */
1874 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1875 return 0;
1876
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001877 /* Instead of sending NOT_USED, sending an empty value is preferable */
1878 if (strcmp(value, "NOT_USED") == 0) {
1879 value = "";
1880 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001881
1882 len = strlen(envname) + 1;
1883 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1884 len += strlen(value);
1885 else
1886 len += vmaxlen;
1887
1888 if (!check->envp[idx])
1889 check->envp[idx] = malloc(len + 1);
1890
1891 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001892 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001893 return 1;
1894 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001895 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001896 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001897 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001898 return 1;
1899 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001900 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001901 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001902 return 1;
1903 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001904 return 0;
1905}
Simon Horman98637e52014-06-20 12:30:16 +09001906
1907static int prepare_external_check(struct check *check)
1908{
1909 struct server *s = check->server;
1910 struct proxy *px = s->proxy;
1911 struct listener *listener = NULL, *l;
1912 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001913 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001914 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001915
1916 list_for_each_entry(l, &px->conf.listeners, by_fe)
1917 /* Use the first INET, INET6 or UNIX listener */
1918 if (l->addr.ss_family == AF_INET ||
1919 l->addr.ss_family == AF_INET6 ||
1920 l->addr.ss_family == AF_UNIX) {
1921 listener = l;
1922 break;
1923 }
1924
Simon Horman98637e52014-06-20 12:30:16 +09001925 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001926 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1927 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001928 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001929 goto err;
1930 }
Simon Horman98637e52014-06-20 12:30:16 +09001931
Cyril Bontéac92a062014-12-27 22:28:38 +01001932 check->argv = calloc(6, sizeof(char *));
1933 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001934 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001935 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001936 }
Simon Horman98637e52014-06-20 12:30:16 +09001937
1938 check->argv[0] = px->check_command;
1939
Cyril Bonté777be862014-12-02 21:21:35 +01001940 if (!listener) {
1941 check->argv[1] = strdup("NOT_USED");
1942 check->argv[2] = strdup("NOT_USED");
1943 }
1944 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001945 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001946 addr_to_str(&listener->addr, buf, sizeof(buf));
1947 check->argv[1] = strdup(buf);
1948 port_to_str(&listener->addr, buf, sizeof(buf));
1949 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001950 }
1951 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001952 const struct sockaddr_un *un;
1953
1954 un = (struct sockaddr_un *)&listener->addr;
1955 check->argv[1] = strdup(un->sun_path);
1956 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001957 }
1958 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001959 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001960 goto err;
1961 }
1962
Christopher Faulet8d18db72020-04-26 09:50:31 +02001963 if (!check->argv[1] || !check->argv[2]) {
1964 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1965 goto err;
1966 }
1967
1968 check->argv[3] = calloc(EXTCHK_SIZE_ADDR, sizeof(*check->argv[3]));
1969 check->argv[4] = calloc(EXTCHK_SIZE_UINT, sizeof(*check->argv[4]));
1970 if (!check->argv[3] || !check->argv[4]) {
1971 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1972 goto err;
1973 }
Willy Tarreau04276f32017-01-06 17:41:29 +01001974
Christopher Faulet8d18db72020-04-26 09:50:31 +02001975 addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR);
Willy Tarreau04276f32017-01-06 17:41:29 +01001976 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
Christopher Faulet8d18db72020-04-26 09:50:31 +02001977 snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port);
Simon Horman98637e52014-06-20 12:30:16 +09001978
Cyril Bontéac92a062014-12-27 22:28:38 +01001979 for (i = 0; i < 5; i++) {
1980 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001981 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001982 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001983 }
1984 }
Simon Horman98637e52014-06-20 12:30:16 +09001985
Cyril Bontéac92a062014-12-27 22:28:38 +01001986 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001987 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001988 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1989 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1990 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1991 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001992 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001993 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1994 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1995 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1996 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1997 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1998 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1999
2000 /* Ensure that we don't leave any hole in check->envp */
2001 for (i = 0; i < EXTCHK_SIZE; i++)
2002 if (!check->envp[i])
2003 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01002004
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002005 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09002006err:
2007 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01002008 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01002009 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09002010 free(check->envp);
2011 check->envp = NULL;
2012 }
2013
2014 if (check->argv) {
2015 for (i = 1; i < 5; i++)
2016 free(check->argv[i]);
2017 free(check->argv);
2018 check->argv = NULL;
2019 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002020 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09002021}
2022
Simon Hormanb00d17a2014-06-13 16:18:16 +09002023/*
Simon Horman98637e52014-06-20 12:30:16 +09002024 * establish a server health-check that makes use of a process.
2025 *
2026 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002027 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02002028 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01002029 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09002030 *
2031 * Blocks and then unblocks SIGCHLD
2032 */
2033static int connect_proc_chk(struct task *t)
2034{
Cyril Bontéac92a062014-12-27 22:28:38 +01002035 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09002036 struct check *check = t->context;
2037 struct server *s = check->server;
2038 struct proxy *px = s->proxy;
2039 int status;
2040 pid_t pid;
2041
Willy Tarreaue7dff022015-04-03 01:14:29 +02002042 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09002043
2044 block_sigchld();
2045
2046 pid = fork();
2047 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002048 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
2049 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09002050 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2051 goto out;
2052 }
2053 if (pid == 0) {
2054 /* Child */
2055 extern char **environ;
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002056 struct rlimit limit;
Willy Tarreaub7b24782016-06-21 15:32:29 +02002057 int fd;
2058
2059 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
2060 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
2061
Willy Tarreau2555ccf2019-02-21 22:22:06 +01002062 my_closefrom(fd);
Willy Tarreaub7b24782016-06-21 15:32:29 +02002063
Willy Tarreau9f6dc722019-03-01 11:15:10 +01002064 /* restore the initial FD limits */
2065 limit.rlim_cur = rlim_fd_cur_at_boot;
2066 limit.rlim_max = rlim_fd_max_at_boot;
2067 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
2068 getrlimit(RLIMIT_NOFILE, &limit);
2069 ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
2070 rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
2071 (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
2072 }
2073
Simon Horman98637e52014-06-20 12:30:16 +09002074 environ = check->envp;
Christopher Faulet8d18db72020-04-26 09:50:31 +02002075
2076 /* Update some environment variables and command args: curconn, server addr and server port */
Willy Tarreau2814afb2020-10-24 13:07:39 +02002077 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), fail);
Christopher Faulet8d18db72020-04-26 09:50:31 +02002078
2079 addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR);
Willy Tarreau2814afb2020-10-24 13:07:39 +02002080 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], fail);
Christopher Faulet8d18db72020-04-26 09:50:31 +02002081
2082 *check->argv[4] = 0;
2083 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
2084 snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port);
Willy Tarreau2814afb2020-10-24 13:07:39 +02002085 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], fail);
Christopher Faulet8d18db72020-04-26 09:50:31 +02002086
Willy Tarreauafc313e2019-07-01 07:51:29 +02002087 haproxy_unblock_signals();
Simon Horman98637e52014-06-20 12:30:16 +09002088 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01002089 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
2090 strerror(errno));
Willy Tarreau2814afb2020-10-24 13:07:39 +02002091 fail:
Simon Horman98637e52014-06-20 12:30:16 +09002092 exit(-1);
2093 }
2094
2095 /* Parent */
2096 if (check->result == CHK_RES_UNKNOWN) {
2097 if (pid_list_add(pid, t) != NULL) {
2098 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2099
2100 if (px->timeout.check && px->timeout.connect) {
2101 int t_con = tick_add(now_ms, px->timeout.connect);
2102 t->expire = tick_first(t->expire, t_con);
2103 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02002104 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09002105 goto out;
2106 }
2107 else {
2108 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2109 }
2110 kill(pid, SIGTERM); /* process creation error */
2111 }
2112 else
2113 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
2114
2115out:
2116 unblock_sigchld();
2117 return status;
2118}
2119
2120/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002121 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02002122 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002123 *
2124 * Please do NOT place any return statement in this function and only leave
2125 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002126 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002127static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002128{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002129 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002130 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002131 int rv;
2132 int ret;
2133 int expired = tick_is_expired(t->expire, now_ms);
2134
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002135 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002136 if (!(check->state & CHK_ST_INPROGRESS)) {
2137 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002138 if (!expired) /* woke up too early */
2139 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002140
2141 /* we don't send any health-checks when the proxy is
2142 * stopped, the server should not be checked or the check
2143 * is disabled.
2144 */
2145 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2146 s->proxy->state == PR_STSTOPPED)
2147 goto reschedule;
2148
2149 /* we'll initiate a new check */
2150 set_server_check_status(check, HCHK_STATUS_START, NULL);
2151
2152 check->state |= CHK_ST_INPROGRESS;
2153
Simon Hormandbf70192015-01-30 11:22:53 +09002154 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002155 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002156 /* the process was forked, we allow up to min(inter,
2157 * timeout.connect) for it to report its status, but
2158 * only when timeout.check is set as it may be to short
2159 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002160 */
2161 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2162
2163 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2164 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2165 t->expire = tick_first(t->expire, t_con);
2166 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002167 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002168 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002169 }
2170
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002171 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002172
2173 check->state &= ~CHK_ST_INPROGRESS;
2174 check_notify_failure(check);
2175
2176 /* we allow up to min(inter, timeout.connect) for a connection
2177 * to establish but only when timeout.check is set
2178 * as it may be to short for a full check otherwise
2179 */
2180 while (tick_is_expired(t->expire, now_ms)) {
2181 int t_con;
2182
2183 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2184 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2185
2186 if (s->proxy->timeout.check)
2187 t->expire = tick_first(t->expire, t_con);
2188 }
2189 }
2190 else {
2191 /* there was a test running.
2192 * First, let's check whether there was an uncaught error,
2193 * which can happen on connect timeout or error.
2194 */
2195 if (check->result == CHK_RES_UNKNOWN) {
2196 /* good connection is enough for pure TCP check */
2197 struct pid_list *elem = check->curpid;
2198 int status = HCHK_STATUS_UNKNOWN;
2199
2200 if (elem->exited) {
2201 status = elem->status; /* Save in case the process exits between use below */
2202 if (!WIFEXITED(status))
2203 check->code = -1;
2204 else
2205 check->code = WEXITSTATUS(status);
2206 if (!WIFEXITED(status) || WEXITSTATUS(status))
2207 status = HCHK_STATUS_PROCERR;
2208 else
2209 status = HCHK_STATUS_PROCOK;
2210 } else if (expired) {
2211 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002212 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002213 kill(elem->pid, SIGTERM);
2214 }
2215 set_server_check_status(check, status, NULL);
2216 }
2217
2218 if (check->result == CHK_RES_FAILED) {
2219 /* a failure or timeout detected */
2220 check_notify_failure(check);
2221 }
2222 else if (check->result == CHK_RES_CONDPASS) {
2223 /* check is OK but asks for stopping mode */
2224 check_notify_stopping(check);
2225 }
2226 else if (check->result == CHK_RES_PASSED) {
2227 /* a success was detected */
2228 check_notify_success(check);
2229 }
Willy Tarreaub1437112019-09-03 18:55:02 +02002230 task_set_affinity(t, 1);
Simon Horman98637e52014-06-20 12:30:16 +09002231 check->state &= ~CHK_ST_INPROGRESS;
2232
2233 pid_list_del(check->curpid);
2234
2235 rv = 0;
2236 if (global.spread_checks > 0) {
2237 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreauc35faee2020-03-08 17:53:53 +01002238 rv -= (int) (2 * rv * (ha_random32() / 4294967295.0));
Simon Horman98637e52014-06-20 12:30:16 +09002239 }
2240 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2241 }
2242
2243 reschedule:
2244 while (tick_is_expired(t->expire, now_ms))
2245 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002246
2247 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002248 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002249 return t;
2250}
2251
2252/*
2253 * manages a server health-check that uses a connection. Returns
2254 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002255 *
2256 * Please do NOT place any return statement in this function and only leave
2257 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002258 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002259static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002260{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002261 struct check *check = context;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002262 struct proxy *proxy = check->proxy;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002263 struct conn_stream *cs = check->cs;
2264 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002265 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002266 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002267 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002268
Olivier Houchard0923fa42019-01-11 18:43:04 +01002269 if (check->server)
2270 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002271 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002272 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002273 if (!expired) /* woke up too early */
2274 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002275
Simon Horman671b6f02013-11-25 10:46:39 +09002276 /* we don't send any health-checks when the proxy is
2277 * stopped, the server should not be checked or the check
2278 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002279 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002280 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002281 proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002282 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002283
2284 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002285 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002286
Willy Tarreau2c115e52013-12-11 19:41:16 +01002287 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002288 b_reset(&check->bi);
2289 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002290
Olivier Houchard68123bc2019-11-29 16:18:51 +01002291 task_set_affinity(t, tid_bit);
Simon Hormandbf70192015-01-30 11:22:53 +09002292 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002293 cs = check->cs;
2294 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002295
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002296 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002297 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002298 goto out_unlock;
2299
Willy Tarreaue7dff022015-04-03 01:14:29 +02002300 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002301 /* we allow up to min(inter, timeout.connect) for a connection
2302 * to establish but only when timeout.check is set
2303 * as it may be to short for a full check otherwise
2304 */
Simon Horman4a741432013-02-23 15:35:38 +09002305 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002306
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002307 if (proxy->timeout.check && proxy->timeout.connect) {
2308 int t_con = tick_add(now_ms, proxy->timeout.connect);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002309 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002310 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002311
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002312 if (check->type) {
2313 /* send the request if we have one. We avoid receiving
2314 * if not connected, unless we didn't subscribe for
2315 * sending since otherwise we won't be woken up.
2316 */
2317 __event_srv_chk_w(cs);
Willy Tarreauab160a42019-09-05 17:38:40 +02002318 if (!(conn->flags & CO_FL_WAIT_L4_CONN) ||
2319 !(check->wait_list.events & SUB_RETRY_SEND))
2320 __event_srv_chk_r(cs);
Willy Tarreaud19a3b92019-09-05 17:51:30 +02002321 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002322
Willy Tarreau5a78f362012-11-23 12:47:05 +01002323 goto reschedule;
2324
Willy Tarreaue7dff022015-04-03 01:14:29 +02002325 case SF_ERR_SRVTO: /* ETIMEDOUT */
2326 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002327 if (conn)
2328 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002329 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002330 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002331 /* should share same code than cases below */
2332 case SF_ERR_CHK_PORT:
2333 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002334 case SF_ERR_PRXCOND:
2335 case SF_ERR_RESOURCE:
2336 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002337 if (conn)
2338 conn->flags |= CO_FL_ERROR;
2339 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002340 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002341 }
2342
Willy Tarreau5a78f362012-11-23 12:47:05 +01002343 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard68123bc2019-11-29 16:18:51 +01002344 task_set_affinity(t, MAX_THREADS_MASK);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002345 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002346 if (check->wait_list.events)
2347 cs->conn->xprt->unsubscribe(cs->conn,
2348 cs->conn->xprt_ctx,
2349 check->wait_list.events,
2350 &check->wait_list);
2351 /* We may have been scheduled to run, and the
2352 * I/O handler expects to have a cs, so remove
2353 * the tasklet
2354 */
Willy Tarreau86eded62019-06-14 14:47:49 +02002355 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002356 cs_destroy(cs);
2357 cs = check->cs = NULL;
2358 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002359 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002360
Willy Tarreau2c115e52013-12-11 19:41:16 +01002361 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002362 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002363
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002364 /* we allow up to min(inter, timeout.connect) for a connection
2365 * to establish but only when timeout.check is set
2366 * as it may be to short for a full check otherwise
2367 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002368 while (tick_is_expired(t->expire, now_ms)) {
2369 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002370
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002371 t_con = tick_add(t->expire, proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002372 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002373
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002374 if (proxy->timeout.check)
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002375 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002376 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002377 }
2378 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002379 /* there was a test running.
2380 * First, let's check whether there was an uncaught error,
2381 * which can happen on connect timeout or error.
2382 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002383 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002384 /* good connection is enough for pure TCP check */
2385 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Christopher Faulet68d35ae2020-03-27 18:55:49 +01002386 if (check->use_ssl == 1)
Simon Horman4a741432013-02-23 15:35:38 +09002387 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002388 else
Simon Horman4a741432013-02-23 15:35:38 +09002389 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002390 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002391 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002392 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002393 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002394 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002395 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002396 }
2397
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002398 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002399 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002400 /* The check was aborted and the connection was not yet closed.
2401 * This can happen upon timeout, or when an external event such
2402 * as a failed response coupled with "observe layer7" caused the
2403 * server state to be suddenly changed.
2404 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002405 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002406 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002407 }
2408
Willy Tarreauac59f362017-10-08 11:10:19 +02002409 if (cs) {
Olivier Houchard49065542019-05-31 19:20:36 +02002410 if (check->wait_list.events)
2411 cs->conn->xprt->unsubscribe(cs->conn,
2412 cs->conn->xprt_ctx,
2413 check->wait_list.events,
2414 &check->wait_list);
2415 /* We may have been scheduled to run, and the
Willy Tarreau86eded62019-06-14 14:47:49 +02002416 * I/O handler expects to have a cs, so remove
2417 * the tasklet
2418 */
2419 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002420 cs_destroy(cs);
2421 cs = check->cs = NULL;
2422 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002423 }
2424
Olivier Houchard0923fa42019-01-11 18:43:04 +01002425 if (check->server) {
2426 if (check->result == CHK_RES_FAILED) {
2427 /* a failure or timeout detected */
2428 check_notify_failure(check);
2429 }
2430 else if (check->result == CHK_RES_CONDPASS) {
2431 /* check is OK but asks for stopping mode */
2432 check_notify_stopping(check);
2433 }
2434 else if (check->result == CHK_RES_PASSED) {
2435 /* a success was detected */
2436 check_notify_success(check);
2437 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002438 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002439 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002440 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002441
Olivier Houchard0923fa42019-01-11 18:43:04 +01002442 if (check->server) {
2443 rv = 0;
2444 if (global.spread_checks > 0) {
2445 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreauc35faee2020-03-08 17:53:53 +01002446 rv -= (int) (2 * rv * (ha_random32() / 4294967295.0));
Olivier Houchard0923fa42019-01-11 18:43:04 +01002447 }
2448 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002449 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002450 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002451
2452 reschedule:
2453 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002454 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002455 out_unlock:
Olivier Houchard0923fa42019-01-11 18:43:04 +01002456 if (check->server)
2457 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002458 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002459}
2460
Simon Horman98637e52014-06-20 12:30:16 +09002461/*
2462 * manages a server health-check. Returns
2463 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2464 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002465static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002466{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002467 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002468
2469 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002470 return process_chk_proc(t, context, state);
2471 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002472
Simon Horman98637e52014-06-20 12:30:16 +09002473}
2474
Simon Horman5c942422013-11-25 10:46:32 +09002475static int start_check_task(struct check *check, int mininter,
2476 int nbcheck, int srvpos)
2477{
2478 struct task *t;
Willy Tarreaub1437112019-09-03 18:55:02 +02002479 unsigned long thread_mask = MAX_THREADS_MASK;
2480
2481 if (check->type == PR_O2_EXT_CHK)
2482 thread_mask = 1;
2483
Simon Horman5c942422013-11-25 10:46:32 +09002484 /* task for the check */
Willy Tarreaub1437112019-09-03 18:55:02 +02002485 if ((t = task_new(thread_mask)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002486 ha_alert("Starting [%s:%s] check: out of memory.\n",
2487 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002488 return 0;
2489 }
2490
2491 check->task = t;
2492 t->process = process_chk;
2493 t->context = check;
2494
Willy Tarreau1746eec2014-04-25 10:46:47 +02002495 if (mininter < srv_getinter(check))
2496 mininter = srv_getinter(check);
2497
2498 if (global.max_spread_checks && mininter > global.max_spread_checks)
2499 mininter = global.max_spread_checks;
2500
Simon Horman5c942422013-11-25 10:46:32 +09002501 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002502 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002503 check->start = now;
2504 task_queue(t);
2505
2506 return 1;
2507}
2508
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002509/*
2510 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002511 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002512 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002513static int start_checks()
2514{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002515
2516 struct proxy *px;
2517 struct server *s;
2518 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002519 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002520
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002521 /* 1- count the checkers to run simultaneously.
2522 * We also determine the minimum interval among all of those which
2523 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2524 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002525 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002526 * too short an interval for all others.
2527 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002528 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002529 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002530 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002531 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002532 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002533 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002534 }
2535 /* We need a warmup task that will be called when the server
2536 * state switches from down to up.
2537 */
2538 s->warmup = t;
2539 t->process = server_warmup;
2540 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002541 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002542 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002543 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 +01002544 }
2545
Willy Tarreaud8514a22013-12-11 21:10:14 +01002546 if (s->check.state & CHK_ST_CONFIGURED) {
2547 nbcheck++;
2548 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2549 (!mininter || mininter > srv_getinter(&s->check)))
2550 mininter = srv_getinter(&s->check);
2551 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002552
Willy Tarreaud8514a22013-12-11 21:10:14 +01002553 if (s->agent.state & CHK_ST_CONFIGURED) {
2554 nbcheck++;
2555 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2556 (!mininter || mininter > srv_getinter(&s->agent)))
2557 mininter = srv_getinter(&s->agent);
2558 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002559 }
2560 }
2561
Simon Horman4a741432013-02-23 15:35:38 +09002562 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002563 return 0;
2564
2565 srand((unsigned)time(NULL));
2566
2567 /*
2568 * 2- start them as far as possible from each others. For this, we will
2569 * start them after their interval set to the min interval divided by
2570 * the number of servers, weighted by the server's position in the list.
2571 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002572 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002573 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2574 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002575 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002576 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002577 }
2578 }
2579
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002580 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002581 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002582 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002583 if (s->check.type == PR_O2_EXT_CHK) {
2584 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002585 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002586 }
Simon Hormand60d6912013-11-25 10:46:36 +09002587 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002588 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002589 srvpos++;
2590 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002591
Simon Hormand60d6912013-11-25 10:46:36 +09002592 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002593 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002594 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002595 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002596 }
2597 srvpos++;
2598 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002599 }
2600 }
2601 return 0;
2602}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002603
2604/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002605 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002606 * The buffer MUST be terminated by a null byte before calling this function.
2607 * Sets server status appropriately. The caller is responsible for ensuring
2608 * that the buffer contains at least 13 characters. If <done> is zero, we may
2609 * return 0 to indicate that data is required to decide of a match.
2610 */
2611static int httpchk_expect(struct server *s, int done)
2612{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002613 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002614 char status_code[] = "000";
2615 char *contentptr;
2616 int crlf;
2617 int ret;
2618
2619 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2620 case PR_O2_EXP_STS:
2621 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002622 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2623 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002624
2625 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2626 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2627 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002628 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002629
2630 /* we necessarily have the response, so there are no partial failures */
2631 if (s->proxy->options2 & PR_O2_EXP_INV)
2632 ret = !ret;
2633
Simon Horman4a741432013-02-23 15:35:38 +09002634 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002635 break;
2636
2637 case PR_O2_EXP_STR:
2638 case PR_O2_EXP_RSTR:
2639 /* very simple response parser: ignore CR and only count consecutive LFs,
2640 * stop with contentptr pointing to first char after the double CRLF or
2641 * to '\0' if crlf < 2.
2642 */
2643 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002644 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002645 if (crlf >= 2)
2646 break;
2647 if (*contentptr == '\r')
2648 continue;
2649 else if (*contentptr == '\n')
2650 crlf++;
2651 else
2652 crlf = 0;
2653 }
2654
2655 /* Check that response contains a body... */
2656 if (crlf < 2) {
2657 if (!done)
2658 return 0;
2659
Simon Horman4a741432013-02-23 15:35:38 +09002660 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002661 "HTTP content check could not find a response body");
2662 return 1;
2663 }
2664
2665 /* Check that response body is not empty... */
2666 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002667 if (!done)
2668 return 0;
2669
Simon Horman4a741432013-02-23 15:35:38 +09002670 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002671 "HTTP content check found empty response body");
2672 return 1;
2673 }
2674
2675 /* Check the response content against the supplied string
2676 * or regex... */
2677 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2678 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2679 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002680 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002681
2682 /* if we don't match, we may need to wait more */
2683 if (!ret && !done)
2684 return 0;
2685
2686 if (ret) {
2687 /* content matched */
2688 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002689 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002690 "HTTP check matched unwanted content");
2691 else
Simon Horman4a741432013-02-23 15:35:38 +09002692 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002693 "HTTP content check matched");
2694 }
2695 else {
2696 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002697 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002698 "HTTP check did not match unwanted content");
2699 else
Simon Horman4a741432013-02-23 15:35:38 +09002700 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002701 "HTTP content check did not match");
2702 }
2703 break;
2704 }
2705 return 1;
2706}
2707
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002708/*
2709 * return the id of a step in a send/expect session
2710 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002711static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002712{
2713 struct tcpcheck_rule *cur = NULL, *next = NULL;
2714 int i = 0;
2715
Willy Tarreau213c6782014-10-02 14:51:02 +02002716 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002717 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002718 return 0;
2719
Simon Hormane16c1b32015-01-30 11:22:57 +09002720 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002721
2722 /* no step => first step */
2723 if (cur == NULL)
2724 return 1;
2725
2726 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002727 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002728 if (next->list.p == &cur->list)
2729 break;
2730 ++i;
2731 }
2732
2733 return i;
2734}
2735
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002736/*
2737 * return the latest known comment before (including) the given stepid
2738 * returns NULL if no comment found
2739 */
2740static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2741{
2742 struct tcpcheck_rule *cur = NULL;
2743 char *ret = NULL;
2744 int i = 0;
2745
2746 /* not even started anything yet, return latest comment found before any action */
2747 if (!check->current_step) {
2748 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2749 if (cur->action == TCPCHK_ACT_COMMENT)
2750 ret = cur->comment;
2751 else
2752 goto return_comment;
2753 }
2754 }
2755
2756 i = 1;
2757 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2758 if (cur->comment)
2759 ret = cur->comment;
2760
2761 if (i >= stepid)
2762 goto return_comment;
2763
2764 ++i;
2765 }
2766
2767 return_comment:
2768 return ret;
2769}
2770
Willy Tarreaube74b882017-10-04 16:22:49 +02002771/* proceed with next steps for the TCP checks <check>. Note that this is called
2772 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002773 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2774 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002775 *
2776 * Please do NOT place any return statement in this function and only leave
Willy Tarreau495c3a42019-07-17 18:48:07 +02002777 * via the out_end_tcpcheck label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002778 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002779static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002780{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002781 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002782 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002783 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002784 struct conn_stream *cs = check->cs;
2785 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002786 struct server *s = check->server;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002787 struct proxy *proxy = check->proxy;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002788 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002789 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002790 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002791
Willy Tarreauef953952014-10-02 14:30:14 +02002792 /* here, we know that the check is complete or that it failed */
2793 if (check->result != CHK_RES_UNKNOWN)
2794 goto out_end_tcpcheck;
2795
2796 /* We have 4 possibilities here :
2797 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002798 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002799 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2800 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002801 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002802 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002803 * be waiting for a connection attempt to complete. conn!=NULL.
2804 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002805 *
2806 * #2 and #3 are quite similar, we want both the connection and the
2807 * handshake to complete before going any further. Thus we must always
2808 * wait for a connection to complete unless we're before and existing
2809 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002810 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002811
2812 /* find first rule and skip comments */
2813 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2814 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2815 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2816
Willy Tarreau00149122017-10-04 18:05:01 +02002817 if ((check->current_step || &next->list == head) &&
2818 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002819 /* we allow up to min(inter, timeout.connect) for a connection
2820 * to establish but only when timeout.check is set
2821 * as it may be to short for a full check otherwise
2822 */
2823 while (tick_is_expired(t->expire, now_ms)) {
2824 int t_con;
2825
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002826 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002827 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2828
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002829 if (proxy->timeout.check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002830 t->expire = tick_first(t->expire, t_con);
2831 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002832 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002833 }
2834
Willy Tarreauef953952014-10-02 14:30:14 +02002835 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002836 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002837 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002838 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002839 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002840
Willy Tarreau213c6782014-10-02 14:51:02 +02002841 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002842 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002843 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002844 b_reset(&check->bo);
2845 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002846 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002847 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01002848 if (proxy->timeout.check)
2849 t->expire = tick_add_ifset(now_ms, proxy->timeout.check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002850 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002851
Willy Tarreauabca5b62013-12-06 14:19:25 +01002852 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002853 /* We have to try to flush the output buffer before reading, at
2854 * the end, or if we're about to send a string that does not fit
2855 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002856 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002857 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002858 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002859 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002860 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002861 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002862 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002863
Olivier Houcharded0f2072018-08-16 15:41:52 +02002864 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002865 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002866
2867 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002868 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002869 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002870 goto out_end_tcpcheck;
2871 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002872 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002873 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002874 if (b_data(&check->bo)) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002875 cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002876 goto out;
2877 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002878 }
2879
Willy Tarreau263013d2015-05-13 11:59:14 +02002880 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002881 break;
2882
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002883 /* have 'next' point to the next rule or NULL if we're on the
2884 * last one, connect() needs this.
2885 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002886 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002887
2888 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002889 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002890 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002891
2892 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002893 if (&next->list == head)
2894 next = NULL;
2895
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002896 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2897 struct protocol *proto;
2898 struct xprt_ops *xprt;
2899
Willy Tarreau00149122017-10-04 18:05:01 +02002900 /* For a connect action we'll create a new connection.
2901 * We may also have to kill a previous one. But we don't
2902 * want to leave *without* a connection if we came here
2903 * from the connection layer, hence with a connection.
2904 * Thus we'll proceed in the following order :
2905 * 1: close but not release previous connection
2906 * 2: try to get a new connection
2907 * 3: release and replace the old one on success
2908 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002909 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002910 cs_close(check->cs);
2911 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002912 }
2913
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002914 /* mark the step as started */
2915 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002916
2917 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002918 cs = cs_new(NULL);
2919 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002920 step = tcpcheck_get_step_id(check);
2921 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2922 comment = tcpcheck_get_step_comment(check, step);
2923 if (comment)
2924 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002925 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2926 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002927 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002928 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002929 }
2930
Olivier Houchard49065542019-05-31 19:20:36 +02002931 if (check->cs) {
2932 if (check->wait_list.events)
2933 cs->conn->xprt->unsubscribe(cs->conn,
2934 cs->conn->xprt_ctx,
2935 check->wait_list.events,
2936 &check->wait_list);
Willy Tarreau86eded62019-06-14 14:47:49 +02002937 /* We may have been scheduled to run, and the
2938 * I/O handler expects to have a cs, so remove
2939 * the tasklet
2940 */
2941 tasklet_remove_from_tasklet_list(check->wait_list.tasklet);
Willy Tarreauac59f362017-10-08 11:10:19 +02002942 cs_destroy(check->cs);
Olivier Houchard49065542019-05-31 19:20:36 +02002943 }
Willy Tarreau00149122017-10-04 18:05:01 +02002944
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002945 check->cs = cs;
2946 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002947 /* Maybe there were an older connection we were waiting on */
Willy Tarreau4f6516d2018-12-19 13:59:17 +01002948 check->wait_list.events = 0;
Olivier Houchard0923fa42019-01-11 18:43:04 +01002949 conn->target = s ? &s->obj_type : &proxy->obj_type;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002950
2951 /* no client address */
2952 clear_addr(&conn->addr.from);
2953
Simon Horman41f58762015-01-30 11:22:56 +09002954 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002955 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002956 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002957 }
2958 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002959 /* we'll connect to the addr on the server */
2960 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002961 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002962 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002963
2964 /* port */
2965 if (check->current_step->port)
2966 set_host_port(&conn->addr.to, check->current_step->port);
2967 else if (check->port)
2968 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002969 else if (s->svc_port)
2970 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002971
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002972 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002973 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002974 }
2975 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002976 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002977 }
Willy Tarreaube373152018-09-06 11:45:30 +02002978
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002979 conn_prepare(conn, proto, xprt);
Willy Tarreau495c3a42019-07-17 18:48:07 +02002980
2981 if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) {
2982 ret = SF_ERR_RESOURCE;
2983 goto fail_check;
2984 }
2985
Willy Tarreaube373152018-09-06 11:45:30 +02002986 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002987
Willy Tarreaue7dff022015-04-03 01:14:29 +02002988 ret = SF_ERR_INTERNAL;
Ricardo Nabinger Sanchez4bccea92019-03-28 21:42:23 -03002989 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002990 ret = proto->connect(conn,
Willy Tarreau9ca8c1d2020-01-24 17:52:37 +01002991 CONNECT_HAS_DATA /* I/O polling is always needed */ | ((next && next->action == TCPCHK_ACT_EXPECT) ? 0 : CONNECT_DELACK_ALWAYS));
Olivier Houchardb4c5e932019-12-30 15:13:42 +01002992 if (conn_ctrl_ready(conn) &&
2993 check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002994 conn->send_proxy_ofs = 1;
2995 conn->flags |= CO_FL_SEND_PROXY;
Olivier Houchardfe50bfb2019-05-27 12:09:19 +02002996 if (xprt_add_hs(conn) < 0)
2997 ret = SF_ERR_RESOURCE;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002998 }
2999
3000 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02003001 * - SF_ERR_NONE if everything's OK
3002 * - SF_ERR_SRVTO if there are no more servers
3003 * - SF_ERR_SRVCL if the connection was refused by the server
3004 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
3005 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
3006 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01003007 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003008 * Note that we try to prevent the network stack from sending the ACK during the
3009 * connect() when a pure TCP check is used (without PROXY protocol).
3010 */
Willy Tarreau495c3a42019-07-17 18:48:07 +02003011 fail_check:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003012 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02003013 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003014 /* we allow up to min(inter, timeout.connect) for a connection
3015 * to establish but only when timeout.check is set
3016 * as it may be to short for a full check otherwise
3017 */
3018 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
3019
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003020 if (proxy->timeout.check && proxy->timeout.connect) {
3021 int t_con = tick_add(now_ms, proxy->timeout.connect);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003022 t->expire = tick_first(t->expire, t_con);
3023 }
3024 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02003025 case SF_ERR_SRVTO: /* ETIMEDOUT */
3026 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003027 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003028 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003029 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003030 comment = tcpcheck_get_step_comment(check, step);
3031 if (comment)
3032 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003033 set_server_check_status(check, HCHK_STATUS_L4CON,
3034 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003035 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02003036 case SF_ERR_PRXCOND:
3037 case SF_ERR_RESOURCE:
3038 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003039 step = tcpcheck_get_step_id(check);
3040 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003041 comment = tcpcheck_get_step_comment(check, step);
3042 if (comment)
3043 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003044 set_server_check_status(check, HCHK_STATUS_SOCKERR,
3045 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003046 goto out_end_tcpcheck;
3047 }
3048
3049 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003050 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003051
3052 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003053 while (&check->current_step->list != head &&
3054 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003055 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003056
Willy Tarreauf2c87352015-05-13 12:08:21 +02003057 if (&check->current_step->list == head)
3058 break;
3059
Willy Tarreau7199e962019-07-15 10:57:51 +02003060 /* don't do anything until the connection is established */
3061 if (!(conn->flags & CO_FL_CONNECTED))
3062 break;
3063
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003064 } /* end 'connect' */
3065 else if (check->current_step->action == TCPCHK_ACT_SEND) {
3066 /* mark the step as started */
3067 check->last_started_step = check->current_step;
3068
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003069 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003070 if (*b_head(&check->bi) != '\0') {
3071 *b_head(&check->bi) = '\0';
3072 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003073 }
3074
Willy Tarreaubbae3f02017-08-30 09:59:52 +02003075 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003076 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003077 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003078 goto out_end_tcpcheck;
3079 }
3080
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003081 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02003082 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003083 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09003084 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003085 set_server_check_status(check, HCHK_STATUS_L7RSP,
3086 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01003087 goto out_end_tcpcheck;
3088 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003089
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003090 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003091 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003092 continue;
3093
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003094 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
3095 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003096
Willy Tarreauabca5b62013-12-06 14:19:25 +01003097 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02003098 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003099
3100 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003101 while (&check->current_step->list != head &&
3102 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003103 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02003104
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003105 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003106 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003107 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003108 goto out_end_tcpcheck;
3109
Olivier Houchard48bf0572019-07-18 15:09:08 +02003110 /* If we already subscribed, then we tried to received
3111 * and failed, so there's no point trying again.
3112 */
3113 if (check->wait_list.events & SUB_RETRY_RECV)
3114 break;
Olivier Houchard511efea2018-08-16 15:30:32 +02003115 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003116 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003117 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003118 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003119 /* Report network errors only if we got no other data. Otherwise
3120 * we'll let the upper layers decide whether the response is OK
3121 * or not. It is very common that an RST sent by the server is
3122 * reported as an error just after the last data chunk.
3123 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003124 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003125 goto out_end_tcpcheck;
3126 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003127 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003128 else {
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003129 conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02003130 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003131 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003132 }
3133
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003134 /* mark the step as started */
3135 check->last_started_step = check->current_step;
3136
3137
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003138 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003139 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003140 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003141 if (b_data(&check->bi) < b_size(&check->bi)) {
3142 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003143 }
3144 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003145 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003146 done = 1; /* buffer full, don't wait for more data */
3147 }
3148
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003149 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003150
3151 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003152 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003153 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003154 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003155
3156 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003157 step = tcpcheck_get_step_id(check);
3158 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003159 comment = tcpcheck_get_step_comment(check, step);
3160 if (comment)
3161 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003162 set_server_check_status(check, HCHK_STATUS_L7RSP,
3163 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003164
3165 goto out_end_tcpcheck;
3166 }
3167
Gaetan Rivet9f2b1192020-02-13 10:30:01 +01003168 tcpcheck_expect:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003169 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01003170 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003171
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003172 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003173 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 +02003174 else if (check->current_step->expect_regex != NULL)
3175 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003176
3177 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01003178 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003179
3180 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003181 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003182 if (ret) {
3183 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003184 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003185 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003186 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003187 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003188 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003189 }
3190 else {
3191 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003192 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003193 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003194 comment = tcpcheck_get_step_comment(check, step);
3195 if (comment)
3196 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003197 set_server_check_status(check, HCHK_STATUS_L7RSP,
3198 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003199 goto out_end_tcpcheck;
3200 }
3201 /* matched and was supposed to => OK, next step */
3202 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003203 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003204 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003205
3206 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003207 while (&check->current_step->list != head &&
3208 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003209 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003210
Willy Tarreauf2c87352015-05-13 12:08:21 +02003211 if (&check->current_step->list == head)
3212 break;
3213
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003214 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003215 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003216 }
3217 }
3218 else {
3219 /* not matched */
3220 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003221 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003222 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003223 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003224
3225 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003226 while (&check->current_step->list != head &&
3227 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003228 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003229
Willy Tarreauf2c87352015-05-13 12:08:21 +02003230 if (&check->current_step->list == head)
3231 break;
3232
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003233 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003234 goto tcpcheck_expect;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003235 }
3236 /* not matched but was supposed to => ERROR */
3237 else {
3238 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003239 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003240 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003241 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003242 }
3243 else {
3244 /* we were looking for a regex */
3245 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003246 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003247 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003248 comment = tcpcheck_get_step_comment(check, step);
3249 if (comment)
3250 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003251 set_server_check_status(check, HCHK_STATUS_L7RSP,
3252 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003253 goto out_end_tcpcheck;
3254 }
3255 }
3256 } /* end expect */
3257 } /* end loop over double chained step list */
3258
Baptiste Assmann248f1172018-03-01 21:49:01 +01003259 /* don't do anything until the connection is established */
3260 if (!(conn->flags & CO_FL_CONNECTED)) {
3261 /* update expire time, should be done by process_chk */
3262 /* we allow up to min(inter, timeout.connect) for a connection
3263 * to establish but only when timeout.check is set
3264 * as it may be to short for a full check otherwise
3265 */
3266 while (tick_is_expired(t->expire, now_ms)) {
3267 int t_con;
3268
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003269 t_con = tick_add(t->expire, proxy->timeout.connect);
Baptiste Assmann248f1172018-03-01 21:49:01 +01003270 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3271
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003272 if (proxy->timeout.check)
Baptiste Assmann248f1172018-03-01 21:49:01 +01003273 t->expire = tick_first(t->expire, t_con);
3274 }
3275 goto out;
3276 }
3277
Willy Tarreau263013d2015-05-13 11:59:14 +02003278 /* We're waiting for some I/O to complete, we've reached the end of the
3279 * rules, or both. Do what we have to do, otherwise we're done.
3280 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003281 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003282 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3283 goto out_end_tcpcheck;
3284 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003285
Willy Tarreau53c5a042015-05-13 11:38:17 +02003286 if (&check->current_step->list != head &&
Olivier Houchard53216e72018-10-10 15:46:36 +02003287 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003288 __event_srv_chk_r(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003289 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003290
3291 out_end_tcpcheck:
3292 /* collect possible new errors */
Willy Tarreau3d84b1c2019-07-23 14:37:47 +02003293 if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR))
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003294 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003295
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003296 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003297 check->current_step = NULL;
3298
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003299 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003300 conn->flags |= CO_FL_ERROR;
3301
Christopher Fauletb6102852017-11-28 10:06:29 +01003302 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003303 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003304}
3305
Simon Hormanb1900d52015-01-30 11:22:54 +09003306const char *init_check(struct check *check, int type)
3307{
3308 check->type = type;
3309
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003310 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3311 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003312
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003313 check->bi.area = calloc(check->bi.size, sizeof(char));
3314 check->bo.area = calloc(check->bo.size, sizeof(char));
3315
3316 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003317 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003318
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003319 check->wait_list.tasklet = tasklet_new();
3320 if (!check->wait_list.tasklet)
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003321 return "out of memroy while allocating check tasklet";
Willy Tarreau4f6516d2018-12-19 13:59:17 +01003322 check->wait_list.events = 0;
Willy Tarreau3c39a7d2019-06-14 14:42:29 +02003323 check->wait_list.tasklet->process = event_srv_chk_io;
3324 check->wait_list.tasklet->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003325 return NULL;
3326}
3327
Simon Hormanbfb5d332015-01-30 11:22:55 +09003328void free_check(struct check *check)
3329{
Christopher Fauletc6e07702020-03-26 19:48:20 +01003330 task_destroy(check->task);
3331 if (check->wait_list.tasklet)
3332 tasklet_free(check->wait_list.tasklet);
3333
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003334 free(check->bi.area);
3335 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003336 if (check->cs) {
3337 free(check->cs->conn);
3338 check->cs->conn = NULL;
3339 cs_free(check->cs);
3340 check->cs = NULL;
3341 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003342}
3343
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003344void email_alert_free(struct email_alert *alert)
3345{
3346 struct tcpcheck_rule *rule, *back;
3347
3348 if (!alert)
3349 return;
3350
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003351 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3352 LIST_DEL(&rule->list);
3353 free(rule->comment);
3354 free(rule->string);
Dragan Dosen26743032019-04-30 15:54:36 +02003355 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003356 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003357 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003358 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003359}
3360
Olivier Houchard9f6af332018-05-25 14:04:04 +02003361static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003362{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003363 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003364 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003365 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003366
3367 q = container_of(check, typeof(*q), check);
3368
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003369 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003370 while (1) {
3371 if (!(check->state & CHK_ST_ENABLED)) {
3372 if (LIST_ISEMPTY(&q->email_alerts)) {
3373 /* All alerts processed, queue the task */
3374 t->expire = TICK_ETERNITY;
3375 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003376 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003377 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003378
3379 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003380 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003381 t->expire = now_ms;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003382 check->tcpcheck_rules = &alert->tcpcheck_rules;
Olivier Houchard0923fa42019-01-11 18:43:04 +01003383 check->status = HCHK_STATUS_INI;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003384 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003385 }
3386
Olivier Houchard9f6af332018-05-25 14:04:04 +02003387 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003388 if (check->state & CHK_ST_INPROGRESS)
3389 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003390
3391 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3392 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003393 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003394 check->server = NULL;
3395 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003396 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003397 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003398 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003399 return t;
3400}
3401
Christopher Faulet0108bb32017-10-20 21:34:32 +02003402/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3403 *
3404 * The function returns 1 in success case, otherwise, it returns 0 and err is
3405 * filled.
3406 */
3407int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003408{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003409 struct mailer *mailer;
3410 struct email_alertq *queues;
3411 const char *err_str;
3412 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003413
Christopher Faulet0108bb32017-10-20 21:34:32 +02003414 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3415 memprintf(err, "out of memory while allocating mailer alerts queues");
mildis5ab01cb2018-10-02 16:46:34 +02003416 goto fail_no_queue;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003417 }
3418
Christopher Faulet0108bb32017-10-20 21:34:32 +02003419 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3420 struct email_alertq *q = &queues[i];
3421 struct check *check = &q->check;
3422 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003423
3424 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003425 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003426 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003427 check->rise = DEF_AGENT_RISETIME;
Olivier Houchardc98aa1f2019-01-11 18:17:17 +01003428 check->proxy = p;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003429 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003430 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3431 memprintf(err, "%s", err_str);
3432 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003433 }
3434
3435 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003436 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003437 check->port = get_host_port(&mailer->addr);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003438
Emeric Brunc60def82017-09-27 14:59:38 +02003439 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003440 memprintf(err, "out of memory while allocating mailer alerts task");
3441 goto error;
3442 }
3443
3444 check->task = t;
3445 t->process = process_email_alert;
3446 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003447
Christopher Faulet0108bb32017-10-20 21:34:32 +02003448 /* check this in one ms */
3449 t->expire = TICK_ETERNITY;
3450 check->start = now;
3451 task_queue(t);
3452 }
3453
3454 mls->users++;
3455 free(p->email_alert.mailers.name);
3456 p->email_alert.mailers.m = mls;
3457 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003458 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003459
3460 error:
3461 for (i = 0; i < mls->count; i++) {
3462 struct email_alertq *q = &queues[i];
3463 struct check *check = &q->check;
3464
Christopher Faulet0108bb32017-10-20 21:34:32 +02003465 free_check(check);
3466 }
3467 free(queues);
mildis5ab01cb2018-10-02 16:46:34 +02003468 fail_no_queue:
Christopher Faulet0108bb32017-10-20 21:34:32 +02003469 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003470}
3471
3472
3473static int add_tcpcheck_expect_str(struct list *list, const char *str)
3474{
3475 struct tcpcheck_rule *tcpcheck;
3476
Willy Tarreaubafbe012017-11-24 17:34:44 +01003477 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003479 memset(tcpcheck, 0, sizeof(*tcpcheck));
3480 tcpcheck->action = TCPCHK_ACT_EXPECT;
3481 tcpcheck->string = strdup(str);
3482 tcpcheck->expect_regex = NULL;
3483 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003484 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003485 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003486 return 0;
3487 }
3488
3489 LIST_ADDQ(list, &tcpcheck->list);
3490 return 1;
3491}
3492
3493static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3494{
3495 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003496 const char *in;
3497 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003498 int i;
3499
Willy Tarreaubafbe012017-11-24 17:34:44 +01003500 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003501 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003502 memset(tcpcheck, 0, sizeof(*tcpcheck));
3503 tcpcheck->action = TCPCHK_ACT_SEND;
3504 tcpcheck->expect_regex = NULL;
3505 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003506 tcpcheck->string_len = 0;
3507 for (i = 0; strs[i]; i++)
3508 tcpcheck->string_len += strlen(strs[i]);
3509
3510 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3511 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003512 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003513 return 0;
3514 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003515
Willy Tarreau64345aa2016-08-10 19:29:09 +02003516 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003517 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003518 for (in = strs[i]; (*dst = *in++); dst++);
3519 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003520
3521 LIST_ADDQ(list, &tcpcheck->list);
3522 return 1;
3523}
3524
Christopher Faulet0108bb32017-10-20 21:34:32 +02003525static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3526 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003527{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003528 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003529 struct tcpcheck_rule *tcpcheck;
3530 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003531
Willy Tarreaubafbe012017-11-24 17:34:44 +01003532 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003533 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003534 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003535 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003536 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003537
Willy Tarreaubafbe012017-11-24 17:34:44 +01003538 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003539 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003540 memset(tcpcheck, 0, sizeof(*tcpcheck));
3541 tcpcheck->action = TCPCHK_ACT_CONNECT;
3542 tcpcheck->comment = NULL;
3543 tcpcheck->string = NULL;
3544 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003545 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3546
3547 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3548 goto error;
3549
3550 {
3551 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
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[4] = { "MAIL FROM:<", p->email_alert.from, ">\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, "250 "))
3566 goto error;
3567
3568 {
3569 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3570 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3571 goto error;
3572 }
3573
3574 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3575 goto error;
3576
3577 {
3578 const char * const strs[2] = { "DATA\r\n" };
3579 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3580 goto error;
3581 }
3582
3583 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3584 goto error;
3585
3586 {
3587 struct tm tm;
3588 char datestr[48];
3589 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003590 "From: ", p->email_alert.from, "\r\n",
3591 "To: ", p->email_alert.to, "\r\n",
3592 "Date: ", datestr, "\r\n",
3593 "Subject: [HAproxy Alert] ", msg, "\r\n",
3594 "\r\n",
3595 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003596 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003597 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003598 NULL
3599 };
3600
3601 get_localtime(date.tv_sec, &tm);
3602
3603 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3604 goto error;
3605 }
3606
3607 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3608 goto error;
3609 }
3610
3611 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3612 goto error;
3613
3614 {
3615 const char * const strs[2] = { "QUIT\r\n" };
3616 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3617 goto error;
3618 }
3619
3620 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3621 goto error;
3622
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003623 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003624 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003625 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003626 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003627 return 1;
3628
3629error:
3630 email_alert_free(alert);
3631 return 0;
3632}
3633
Christopher Faulet0108bb32017-10-20 21:34:32 +02003634static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003635{
3636 int i;
3637 struct mailer *mailer;
3638
3639 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3640 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003641 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003642 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003643 return;
3644 }
3645 }
3646
3647 return;
3648}
3649
3650/*
3651 * Send email alert if configured.
3652 */
Simon Horman64e34162015-02-06 11:11:57 +09003653void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003654{
3655 va_list argp;
3656 char buf[1024];
3657 int len;
3658 struct proxy *p = s->proxy;
3659
Christopher Faulet0108bb32017-10-20 21:34:32 +02003660 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003661 return;
3662
3663 va_start(argp, format);
3664 len = vsnprintf(buf, sizeof(buf), format, argp);
3665 va_end(argp);
3666
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003667 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003668 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003669 return;
3670 }
3671
Christopher Faulet0108bb32017-10-20 21:34:32 +02003672 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003673}
3674
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003675/*
3676 * Return value:
3677 * the port to be used for the health check
3678 * 0 in case no port could be found for the check
3679 */
3680int srv_check_healthcheck_port(struct check *chk)
3681{
3682 int i = 0;
3683 struct server *srv = NULL;
3684
3685 srv = chk->server;
3686
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003687 /* by default, we use the health check port ocnfigured */
3688 if (chk->port > 0)
3689 return chk->port;
3690
3691 /* try to get the port from check_core.addr if check.port not set */
3692 i = get_host_port(&chk->addr);
3693 if (i > 0)
3694 return i;
3695
3696 /* try to get the port from server address */
3697 /* prevent MAPPORTS from working at this point, since checks could
3698 * not be performed in such case (MAPPORTS impose a relative ports
3699 * based on live traffic)
3700 */
3701 if (srv->flags & SRV_F_MAPPORTS)
3702 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003703
3704 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003705 if (i > 0)
3706 return i;
3707
3708 return 0;
3709}
3710
Christopher Fauletc6e07702020-03-26 19:48:20 +01003711static int init_srv_check(struct server *srv)
3712{
3713 const char *err;
3714 struct tcpcheck_rule *r;
3715 int ret = 0;
3716
3717 if (!srv->do_check)
3718 goto out;
3719
3720
3721 /* If neither a port nor an addr was specified and no check transport
3722 * layer is forced, then the transport layer used by the checks is the
3723 * same as for the production traffic. Otherwise we use raw_sock by
3724 * default, unless one is specified.
3725 */
3726 if (!srv->check.port && !is_addr(&srv->check.addr)) {
3727 if (!srv->check.use_ssl && srv->use_ssl != -1) {
3728 srv->check.use_ssl = srv->use_ssl;
3729 srv->check.xprt = srv->xprt;
3730 }
3731 else if (srv->check.use_ssl == 1)
3732 srv->check.xprt = xprt_get(XPRT_SSL);
3733
3734 srv->check.send_proxy |= (srv->pp_opts);
3735 }
Christopher Faulet98bfaf72020-05-20 22:36:24 +02003736 else if (srv->check.use_ssl == 1)
3737 srv->check.xprt = xprt_get(XPRT_SSL);
Christopher Fauletc6e07702020-03-26 19:48:20 +01003738
3739 /* validate <srv> server health-check settings */
3740
3741 /* We need at least a service port, a check port or the first tcp-check
3742 * rule must be a 'connect' one when checking an IPv4/IPv6 server.
3743 */
3744 if ((srv_check_healthcheck_port(&srv->check) != 0) ||
3745 (!is_inet_addr(&srv->check.addr) && (is_addr(&srv->check.addr) || !is_inet_addr(&srv->addr))))
3746 goto init;
3747
3748 if (!LIST_ISEMPTY(&srv->proxy->tcpcheck_rules)) {
3749 ha_alert("config: %s '%s': server '%s' has neither service port nor check port.\n",
3750 proxy_type_str(srv->proxy), srv->proxy->id, srv->id);
3751 ret |= ERR_ALERT | ERR_ABORT;
3752 goto out;
3753 }
3754
3755 /* search the first action (connect / send / expect) in the list */
3756 r = get_first_tcpcheck_rule(&srv->proxy->tcpcheck_rules);
3757 if (!r || (r->action != TCPCHK_ACT_CONNECT) || !r->port) {
3758 ha_alert("config: %s '%s': server '%s' has neither service port nor check port "
3759 "nor tcp_check rule 'connect' with port information.\n",
3760 proxy_type_str(srv->proxy), srv->proxy->id, srv->id);
3761 ret |= ERR_ALERT | ERR_ABORT;
3762 goto out;
3763 }
3764
3765 /* scan the tcp-check ruleset to ensure a port has been configured */
3766 list_for_each_entry(r, &srv->proxy->tcpcheck_rules, list) {
3767 if ((r->action == TCPCHK_ACT_CONNECT) && (!r->port)) {
3768 ha_alert("config: %s '%s': server '%s' has neither service port nor check port, "
3769 "and a tcp_check rule 'connect' with no port information.\n",
3770 proxy_type_str(srv->proxy), srv->proxy->id, srv->id);
3771 ret |= ERR_ALERT | ERR_ABORT;
3772 goto out;
3773 }
3774 }
3775
3776 init:
3777 err = init_check(&srv->check, srv->proxy->options2 & PR_O2_CHK_ANY);
3778 if (err) {
3779 ha_alert("config: %s '%s': unable to init check for server '%s' (%s).\n",
3780 proxy_type_str(srv->proxy), srv->proxy->id, srv->id, err);
3781 ret |= ERR_ALERT | ERR_ABORT;
3782 goto out;
3783 }
3784 srv->check.state |= CHK_ST_CONFIGURED | CHK_ST_ENABLED;
3785 global.maxsock++;
3786
3787 out:
3788 return ret;
3789}
3790
3791static int init_srv_agent_check(struct server *srv)
3792{
3793 const char *err;
3794 int ret = 0;
3795
3796 if (!srv->do_agent)
3797 goto out;
3798
3799 err = init_check(&srv->agent, PR_O2_LB_AGENT_CHK);
3800 if (err) {
3801 ha_alert("config: %s '%s': unable to init agent-check for server '%s' (%s).\n",
3802 proxy_type_str(srv->proxy), srv->proxy->id, srv->id, err);
3803 ret |= ERR_ALERT | ERR_ABORT;
3804 goto out;
3805 }
3806
3807 if (!srv->agent.inter)
3808 srv->agent.inter = srv->check.inter;
3809
3810 srv->agent.state |= CHK_ST_CONFIGURED | CHK_ST_ENABLED | CHK_ST_AGENT;
3811 global.maxsock++;
3812
3813 out:
3814 return ret;
3815}
3816
3817void deinit_srv_check(struct server *srv)
3818{
3819 if (srv->do_check)
3820 free_check(&srv->check);
3821}
3822
3823
3824void deinit_srv_agent_check(struct server *srv)
3825{
3826 if (srv->do_agent)
3827 free_check(&srv->agent);
3828 free(srv->agent.send_string);
3829}
3830
3831static int init_servers_checks()
3832{
3833 struct proxy *px;
3834 struct server *s;
3835 int ret = 0;
3836
3837 for (px = proxies_list; px; px = px->next) {
3838 for (s = px->srv; s; s = s->next) {
3839 ret |= init_srv_check(s);
3840 if (ret & (ERR_ABORT|ERR_FATAL))
3841 goto end;
3842 ret |= init_srv_agent_check(s);
3843 if (ret & (ERR_ABORT|ERR_FATAL))
3844 goto end;
3845 }
3846 }
Christopher Fauleta377ce92020-05-12 09:11:09 +02003847 ret = start_checks();
Christopher Fauletc6e07702020-03-26 19:48:20 +01003848 end:
3849 return ret;
3850}
3851
Christopher Fauletc6e07702020-03-26 19:48:20 +01003852REGISTER_POST_CHECK(init_servers_checks);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003853
Willy Tarreaubd741542010-03-16 18:46:54 +01003854/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003855 * Local variables:
3856 * c-indent-level: 8
3857 * c-basic-offset: 8
3858 * End:
3859 */