blob: 6d19eef3f3528bcfc0131ae3a9c01c9ec2faa873 [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>
25#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040026#include <sys/types.h>
Simon Horman98637e52014-06-20 12:30:16 +090027#include <sys/wait.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020028#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020029#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020030#include <arpa/inet.h>
31
Willy Tarreauc7e42382012-08-24 19:22:53 +020032#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020033#include <common/compat.h>
34#include <common/config.h>
35#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020036#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020037#include <common/time.h>
Christopher Fauletcfda8472017-10-20 15:40:23 +020038#include <common/hathreads.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020039
40#include <types/global.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020041#include <types/dns.h>
William Lallemand9ed62032016-11-21 17:49:11 +010042#include <types/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043
44#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020045#include <proto/checks.h>
William Lallemand9ed62032016-11-21 17:49:11 +010046#include <proto/stats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020047#include <proto/fd.h>
48#include <proto/log.h>
Willy Tarreau53a47662017-08-28 10:53:00 +020049#include <proto/mux_pt.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020050#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020051#include <proto/port_range.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010052#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010053#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010054#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020055#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020056#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010057#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020058#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020059#include <proto/log.h>
60#include <proto/dns.h>
61#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020062
Olivier Houchard9130a962017-10-17 17:33:43 +020063#ifdef USE_OPENSSL
64#include <proto/ssl_sock.h>
65#endif /* USE_OPENSSL */
66
Willy Tarreaubd741542010-03-16 18:46:54 +010067static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090068static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020069static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020070static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020071static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020072static int wake_srv_chk(struct conn_stream *cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +020073static void __event_srv_chk_r(struct conn_stream *cs);
Willy Tarreaubd741542010-03-16 18:46:54 +010074
Willy Tarreaubafbe012017-11-24 17:34:44 +010075static struct pool_head *pool_head_email_alert = NULL;
76static struct pool_head *pool_head_tcpcheck_rule = NULL;
Christopher Faulet31dff9b2017-10-23 15:45:20 +020077
78
Simon Horman63a4a822012-03-19 07:24:41 +090079static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010080 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
81 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020082 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020083
Willy Tarreau23964182014-05-20 20:56:30 +020084 /* Below we have finished checks */
85 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010086 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020089
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010090 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
91 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
92 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020093
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010094 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
95 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
96 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020097
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010098 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
99 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200100
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200101 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200102
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100103 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
104 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
105 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900106
107 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
108 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200109 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200110};
111
Cyril Bontéac92a062014-12-27 22:28:38 +0100112const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
113 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
118 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
119 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
120 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
121 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
122 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
123 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
124};
125
Simon Horman63a4a822012-03-19 07:24:41 +0900126static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100127 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
128
129 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
130 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
131
132 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
133 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
134 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
135 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
136
137 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
138 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
139 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
140};
141
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200142/*
143 * Convert check_status code to description
144 */
145const char *get_check_status_description(short check_status) {
146
147 const char *desc;
148
149 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200150 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200151 else
152 desc = NULL;
153
154 if (desc && *desc)
155 return desc;
156 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200157 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200158}
159
160/*
161 * Convert check_status code to short info
162 */
163const char *get_check_status_info(short check_status) {
164
165 const char *info;
166
167 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200168 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200169 else
170 info = NULL;
171
172 if (info && *info)
173 return info;
174 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200175 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200176}
177
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100178const char *get_analyze_status(short analyze_status) {
179
180 const char *desc;
181
182 if (analyze_status < HANA_STATUS_SIZE)
183 desc = analyze_statuses[analyze_status].desc;
184 else
185 desc = NULL;
186
187 if (desc && *desc)
188 return desc;
189 else
190 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
191}
192
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200193/*
Simon Horman4a741432013-02-23 15:35:38 +0900194 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200195 * an adequate CHK_RES_* value. The new check->health is computed based
196 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200197 *
198 * Show information in logs about failed health check if server is UP
199 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200200 */
Simon Horman4a741432013-02-23 15:35:38 +0900201static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100202{
Simon Horman4a741432013-02-23 15:35:38 +0900203 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200204 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200205 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900206
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200207 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100208 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900209 check->desc[0] = '\0';
210 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200211 return;
212 }
213
Simon Horman4a741432013-02-23 15:35:38 +0900214 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200215 return;
216
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200217 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900218 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
219 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200220 } else
Simon Horman4a741432013-02-23 15:35:38 +0900221 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200222
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200224 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900225 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200226
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100227 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900228 check->duration = -1;
229 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200230 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900231 check->duration = tv_ms_elapsed(&check->start, &now);
232 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200233 }
234
Willy Tarreau23964182014-05-20 20:56:30 +0200235 /* no change is expected if no state change occurred */
236 if (check->result == CHK_RES_NEUTRAL)
237 return;
238
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200239 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200240
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200241 switch (check->result) {
242 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200243 /* Failure to connect to the agent as a secondary check should not
244 * cause the server to be marked down.
245 */
246 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900247 (check->status >= HCHK_STATUS_L57DATA)) &&
Christopher Fauletb119a792018-05-02 12:12:45 +0200248 (check->health > 0)) {
Christopher Faulet29f77e82017-06-08 14:04:45 +0200249 HA_ATOMIC_ADD(&s->counters.failed_checks, 1);
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200250 report = 1;
251 check->health--;
252 if (check->health < check->rise)
253 check->health = 0;
254 }
255 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200256
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200257 case CHK_RES_PASSED:
258 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
259 if ((check->health < check->rise + check->fall - 1) &&
260 (check->result == CHK_RES_PASSED || check->health > 0)) {
261 report = 1;
262 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200263
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200264 if (check->health >= check->rise)
265 check->health = check->rise + check->fall - 1; /* OK now */
266 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200267
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200268 /* clear consecutive_errors if observing is enabled */
269 if (s->onerror)
270 s->consecutive_errors = 0;
271 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100272
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200273 default:
274 break;
275 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200276
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200277 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
278 (status != prev_status || report)) {
279 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200280 "%s check for %sserver %s/%s %s%s",
281 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200282 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100283 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100284 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200285 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200286
Emeric Brun5a133512017-10-19 14:42:30 +0200287 srv_append_status(&trash, s, check, -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200288
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100289 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200290 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
291 (check->health >= check->rise) ? check->fall : check->rise,
292 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200293
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200294 ha_warning("%s.\n", trash.area);
295 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area);
296 send_email_alert(s, LOG_INFO, "%s", trash.area);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200297 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200298}
299
Willy Tarreau4eec5472014-05-20 22:32:27 +0200300/* Marks the check <check>'s server down if the current check is already failed
301 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200302 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200303static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200304{
Simon Horman4a741432013-02-23 15:35:38 +0900305 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900306
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200307 /* The agent secondary check should only cause a server to be marked
308 * as down if check->status is HCHK_STATUS_L7STS, which indicates
309 * that the agent returned "fail", "stopped" or "down".
310 * The implication here is that failure to connect to the agent
311 * as a secondary check should not cause the server to be marked
312 * down. */
313 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
314 return;
315
Willy Tarreau4eec5472014-05-20 22:32:27 +0200316 if (check->health > 0)
317 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100318
Willy Tarreau4eec5472014-05-20 22:32:27 +0200319 /* We only report a reason for the check if we did not do so previously */
Emeric Brun5a133512017-10-19 14:42:30 +0200320 srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200321}
322
Willy Tarreauaf549582014-05-16 17:37:50 +0200323/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200324 * it isn't in maintenance, it is not tracking a down server and other checks
325 * comply. The rule is simple : by default, a server is up, unless any of the
326 * following conditions is true :
327 * - health check failed (check->health < rise)
328 * - agent check failed (agent->health < rise)
329 * - the server tracks a down server (track && track->state == STOPPED)
330 * Note that if the server has a slowstart, it will switch to STARTING instead
331 * of RUNNING. Also, only the health checks support the nolb mode, so the
332 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200333 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200334static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200335{
Simon Horman4a741432013-02-23 15:35:38 +0900336 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100337
Emeric Brun52a91d32017-08-31 14:41:55 +0200338 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200339 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100340
Emeric Brun52a91d32017-08-31 14:41:55 +0200341 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200342 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100343
Willy Tarreau3e048382014-05-21 10:30:54 +0200344 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
345 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100346
Willy Tarreau3e048382014-05-21 10:30:54 +0200347 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
348 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200349
Emeric Brun52a91d32017-08-31 14:41:55 +0200350 if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING)
Willy Tarreau3e048382014-05-21 10:30:54 +0200351 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100352
Emeric Brun5a133512017-10-19 14:42:30 +0200353 srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100354}
355
Willy Tarreaudb58b792014-05-21 13:57:23 +0200356/* Marks the check <check> as valid and tries to set its server into stopping mode
357 * if it was running or starting, and provided it isn't in maintenance and other
358 * checks comply. The conditions for the server to be marked in stopping mode are
359 * the same as for it to be turned up. Also, only the health checks support the
360 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200361 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200362static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200363{
Simon Horman4a741432013-02-23 15:35:38 +0900364 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100365
Emeric Brun52a91d32017-08-31 14:41:55 +0200366 if (s->next_admin & SRV_ADMF_MAINT)
Willy Tarreauaf549582014-05-16 17:37:50 +0200367 return;
368
Willy Tarreaudb58b792014-05-21 13:57:23 +0200369 if (check->state & CHK_ST_AGENT)
370 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100371
Emeric Brun52a91d32017-08-31 14:41:55 +0200372 if (s->track && s->track->next_state == SRV_ST_STOPPED)
Willy Tarreaudb58b792014-05-21 13:57:23 +0200373 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100374
Willy Tarreaudb58b792014-05-21 13:57:23 +0200375 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
376 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100377
Willy Tarreaudb58b792014-05-21 13:57:23 +0200378 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
379 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100380
Willy Tarreaub26881a2017-12-23 11:16:49 +0100381 srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100382}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200383
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100384/* note: use health_adjust() only, which first checks that the observe mode is
385 * enabled.
386 */
387void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100388{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100389 int failed;
390 int expire;
391
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100392 if (s->observe >= HANA_OBS_SIZE)
393 return;
394
Willy Tarreaubb956662013-01-24 00:37:39 +0100395 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100396 return;
397
398 switch (analyze_statuses[status].lr[s->observe - 1]) {
399 case 1:
400 failed = 1;
401 break;
402
403 case 2:
404 failed = 0;
405 break;
406
407 default:
408 return;
409 }
410
411 if (!failed) {
412 /* good: clear consecutive_errors */
413 s->consecutive_errors = 0;
414 return;
415 }
416
Christopher Faulet29f77e82017-06-08 14:04:45 +0200417 HA_ATOMIC_ADD(&s->consecutive_errors, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100418
419 if (s->consecutive_errors < s->consecutive_errors_limit)
420 return;
421
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100422 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
423 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100424
425 switch (s->onerror) {
426 case HANA_ONERR_FASTINTER:
427 /* force fastinter - nothing to do here as all modes force it */
428 break;
429
430 case HANA_ONERR_SUDDTH:
431 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900432 if (s->check.health > s->check.rise)
433 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100434
435 /* no break - fall through */
436
437 case HANA_ONERR_FAILCHK:
438 /* simulate a failed health check */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200439 set_server_check_status(&s->check, HCHK_STATUS_HANA,
440 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200441 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100442 break;
443
444 case HANA_ONERR_MARKDWN:
445 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900446 s->check.health = s->check.rise;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200447 set_server_check_status(&s->check, HCHK_STATUS_HANA,
448 trash.area);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200449 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100450 break;
451
452 default:
453 /* write a warning? */
454 break;
455 }
456
457 s->consecutive_errors = 0;
Christopher Faulet29f77e82017-06-08 14:04:45 +0200458 HA_ATOMIC_ADD(&s->counters.failed_hana, 1);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100459
Simon Horman66183002013-02-23 10:16:43 +0900460 if (s->check.fastinter) {
461 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300462 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200463 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300464 /* requeue check task with new expire */
465 task_queue(s->check.task);
466 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100467 }
468}
469
Willy Tarreaua1dab552014-04-14 15:04:54 +0200470static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100471{
472 int sv_state;
473 int ratio;
474 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800475 char addr[46];
476 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100477 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
478 "UP %d/%d", "UP",
479 "NOLB %d/%d", "NOLB",
480 "no check" };
481
482 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
483 hlen += 24;
484
Willy Tarreauff5ae352013-12-11 20:36:34 +0100485 if (!(s->check.state & CHK_ST_ENABLED))
486 sv_state = 6;
Emeric Brun52a91d32017-08-31 14:41:55 +0200487 else if (s->cur_state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900488 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100489 sv_state = 3; /* UP */
490 else
491 sv_state = 2; /* going down */
492
Emeric Brun52a91d32017-08-31 14:41:55 +0200493 if (s->cur_state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100494 sv_state += 2;
495 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900496 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100497 sv_state = 1; /* going up */
498 else
499 sv_state = 0; /* DOWN */
500 }
501
Willy Tarreaua1dab552014-04-14 15:04:54 +0200502 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100503 srv_hlt_st[sv_state],
Emeric Brun52a91d32017-08-31 14:41:55 +0200504 (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
505 (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100506
Joseph Lynch514061c2015-01-15 17:52:59 -0800507 addr_to_str(&s->addr, addr, sizeof(addr));
Willy Tarreau04276f32017-01-06 17:41:29 +0100508 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
509 snprintf(port, sizeof(port), "%u", s->svc_port);
510 else
511 *port = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800512
513 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
514 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100515 global.node,
Emeric Brun52a91d32017-08-31 14:41:55 +0200516 (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
Willy Tarreauef781042010-01-27 11:53:01 +0100517 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
518 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
519 s->nbpend);
520
Emeric Brun52a91d32017-08-31 14:41:55 +0200521 if ((s->cur_state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100522 now.tv_sec < s->last_change + s->slowstart &&
523 now.tv_sec >= s->last_change) {
524 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200525 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100526 }
527
528 buffer[hlen++] = '\r';
529 buffer[hlen++] = '\n';
530
531 return hlen;
532}
533
Willy Tarreau20a18342013-12-05 00:31:46 +0100534/* Check the connection. If an error has already been reported or the socket is
535 * closed, keep errno intact as it is supposed to contain the valid error code.
536 * If no error is reported, check the socket's error queue using getsockopt().
537 * Warning, this must be done only once when returning from poll, and never
538 * after an I/O error was attempted, otherwise the error queue might contain
539 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
540 * socket. Returns non-zero if an error was reported, zero if everything is
541 * clean (including a properly closed socket).
542 */
543static int retrieve_errno_from_socket(struct connection *conn)
544{
545 int skerr;
546 socklen_t lskerr = sizeof(skerr);
547
548 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
549 return 1;
550
Willy Tarreau3c728722014-01-23 13:50:42 +0100551 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100552 return 0;
553
Willy Tarreau585744b2017-08-24 14:31:19 +0200554 if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
Willy Tarreau20a18342013-12-05 00:31:46 +0100555 errno = skerr;
556
557 if (errno == EAGAIN)
558 errno = 0;
559
560 if (!errno) {
561 /* we could not retrieve an error, that does not mean there is
562 * none. Just don't change anything and only report the prior
563 * error if any.
564 */
565 if (conn->flags & CO_FL_ERROR)
566 return 1;
567 else
568 return 0;
569 }
570
571 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
572 return 1;
573}
574
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100575/* Try to collect as much information as possible on the connection status,
576 * and adjust the server status accordingly. It may make use of <errno_bck>
577 * if non-null when the caller is absolutely certain of its validity (eg:
578 * checked just after a syscall). If the caller doesn't have a valid errno,
579 * it can pass zero, and retrieve_errno_from_socket() will be called to try
580 * to extract errno from the socket. If no error is reported, it will consider
581 * the <expired> flag. This is intended to be used when a connection error was
582 * reported in conn->flags or when a timeout was reported in <expired>. The
583 * function takes care of not updating a server status which was already set.
584 * All situations where at least one of <expired> or CO_FL_ERROR are set
585 * produce a status.
586 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200587static void chk_report_conn_err(struct check *check, int errno_bck, int expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100588{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200589 struct conn_stream *cs = check->cs;
590 struct connection *conn = cs_conn(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100591 const char *err_msg;
Willy Tarreau83061a82018-07-13 11:56:34 +0200592 struct buffer *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200593 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200594 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100595
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100596 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100597 return;
598
599 errno = errno_bck;
Willy Tarreau00149122017-10-04 18:05:01 +0200600 if (conn && (!errno || errno == EAGAIN))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100601 retrieve_errno_from_socket(conn);
602
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200603 if (conn && !(conn->flags & CO_FL_ERROR) &&
604 !(cs->flags & CS_FL_ERROR) && !expired)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100605 return;
606
607 /* we'll try to build a meaningful error message depending on the
608 * context of the error possibly present in conn->err_code, and the
609 * socket error possibly collected above. This is useful to know the
610 * exact step of the L6 layer (eg: SSL handshake).
611 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200612 chk = get_trash_chunk();
613
614 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900615 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200616 if (!step)
617 chunk_printf(chk, " at initial connection step of tcp-check");
618 else {
619 chunk_printf(chk, " at step %d of tcp-check", step);
620 /* we were looking for a string */
621 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
622 if (check->last_started_step->port)
623 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
624 else
625 chunk_appendf(chk, " (connect)");
626 }
627 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
628 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200629 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200630 else if (check->last_started_step->expect_regex)
631 chunk_appendf(chk, " (expect regex)");
632 }
633 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
634 chunk_appendf(chk, " (send)");
635 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200636
637 comment = tcpcheck_get_step_comment(check, step);
638 if (comment)
639 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200640 }
641 }
642
Willy Tarreau00149122017-10-04 18:05:01 +0200643 if (conn && conn->err_code) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100644 if (errno && errno != EAGAIN)
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200645 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno),
646 chk->area);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100647 else
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200648 chunk_printf(&trash, "%s%s", conn_err_code_str(conn),
649 chk->area);
650 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100651 }
652 else {
653 if (errno && errno != EAGAIN) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200654 chunk_printf(&trash, "%s%s", strerror(errno),
655 chk->area);
656 err_msg = trash.area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100657 }
658 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200659 err_msg = chk->area;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100660 }
661 }
662
Willy Tarreau00149122017-10-04 18:05:01 +0200663 if (check->state & CHK_ST_PORT_MISS) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +0200664 /* NOTE: this is reported after <fall> tries */
665 chunk_printf(chk, "No port available for the TCP connection");
666 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
667 }
668
Willy Tarreau00149122017-10-04 18:05:01 +0200669 if (!conn) {
670 /* connection allocation error before the connection was established */
671 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
672 }
673 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100674 /* L4 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200675 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100676 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
677 else if (expired)
678 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200679
680 /*
681 * might be due to a server IP change.
682 * Let's trigger a DNS resolution if none are currently running.
683 */
Christopher Faulet67957bd2017-09-27 11:00:59 +0200684 dns_trigger_resolution(check->server->dns_requester);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200685
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100686 }
687 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
688 /* L6 not established (yet) */
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200689 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100690 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
691 else if (expired)
692 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
693 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200694 else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100695 /* I/O error after connection was established and before we could diagnose */
696 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
697 }
698 else if (expired) {
699 /* connection established but expired check */
700 if (check->type == PR_O2_SSL3_CHK)
701 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
702 else /* HTTP, SMTP, ... */
703 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
704 }
705
706 return;
707}
708
Olivier Houchard5c110b92018-08-14 17:04:58 +0200709/* This function checks if any I/O is wanted, and if so, attempts to do so */
710static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200711{
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200712 struct check *check = ctx;
713 struct conn_stream *cs = check->cs;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200714
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200715 if (!(check->wait_list.wait_reason & SUB_CAN_SEND))
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200716 wake_srv_chk(cs);
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200717 if (!(check->wait_list.wait_reason & SUB_CAN_RECV)) {
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200718 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
719 __event_srv_chk_r(cs);
720 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
721 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200722 return NULL;
723}
724
725/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100726 *
727 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200728 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
729 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200731static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200732{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200733 struct connection *conn = cs->conn;
734 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900735 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900736 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200737
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100738 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100739 goto out_wakeup;
740
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200741 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200742 cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200743 goto out;
744 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100745
Willy Tarreau20a18342013-12-05 00:31:46 +0100746 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200747 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200748 __cs_stop_both(cs);
Willy Tarreau20a18342013-12-05 00:31:46 +0100749 goto out_wakeup;
750 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100751
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200752 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100753 /* if the output is closed, we can't do anything */
754 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200755 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100756 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200757 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200758
Willy Tarreau06559ac2013-12-05 01:53:08 +0100759 /* here, we know that the connection is established. That's enough for
760 * a pure TCP check.
761 */
762 if (!check->type)
763 goto out_wakeup;
764
Willy Tarreauc09572f2017-10-04 11:58:22 +0200765 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100766 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200767 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200768
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200769 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200770 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200771 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200772 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200773 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200774 __cs_stop_both(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100775 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200776 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200777 if (b_data(&check->bo)) {
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200778 conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200779 goto out;
780 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100781 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200782
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100783 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
784 if (s->proxy->timeout.check) {
785 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
786 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200787 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100788 goto out_nowake;
789
Willy Tarreau83749182007-04-15 20:56:27 +0200790 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200791 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200792 out_nowake:
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200793 __cs_stop_send(cs); /* nothing more to write */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200794 out:
795 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200796}
797
Willy Tarreaubaaee002006-06-26 02:48:02 +0200798/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200799 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200800 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900801 * set_server_check_status() to update check->status, check->duration
802 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200803
804 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
805 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
806 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
807 * response to an SSL HELLO (the principle is that this is enough to
808 * distinguish between an SSL server and a pure TCP relay). All other cases will
809 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
810 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100811 *
812 * Please do NOT place any return statement in this function and only leave
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200813 * via the out label.
814 *
815 * This must be called with the server lock held.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200816 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200817static void __event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200818{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200819 struct connection *conn = cs->conn;
820 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900821 struct server *s = check->server;
822 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200823 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100824 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200825 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200826
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100827 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200828 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200829
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200830 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard26e1a8f2018-09-12 15:15:12 +0200831 cs->conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200832 goto out;
833 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200834
Willy Tarreauc09572f2017-10-04 11:58:22 +0200835 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100836 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200837 goto out;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200838
Willy Tarreau83749182007-04-15 20:56:27 +0200839 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
840 * but the connection was closed on the remote end. Fortunately, recv still
841 * works correctly and we don't need to do the getsockopt() on linux.
842 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000843
844 /* Set buffer to point to the end of the data already read, and check
845 * that there is free space remaining. If the buffer is full, proceed
846 * with running the checks without attempting another socket read.
847 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000848
Willy Tarreau03938182010-03-17 21:52:07 +0100849 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000850
Olivier Houchard511efea2018-08-16 15:30:32 +0200851 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200852 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100853 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200854 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200855 /* Report network errors only if we got no other data. Otherwise
856 * we'll let the upper layers decide whether the response is OK
857 * or not. It is very common that an RST sent by the server is
858 * reported as an error just after the last data chunk.
859 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200860 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100861 goto out_wakeup;
862 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200863 }
864
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100865
Willy Tarreau03938182010-03-17 21:52:07 +0100866 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200867 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100868 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200869 if (b_data(&check->bi) < b_size(&check->bi))
870 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100871 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200872 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100873 done = 1; /* buffer full, don't wait for more data */
874 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200875
Nick Chalk57b1bf72010-03-16 15:50:46 +0000876 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900877 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200878 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200879 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100880 goto wait_more_data;
881
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100882 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200883 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
884 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
885 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
886 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
887 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
888 cut_crlf(b_head(&check->bi));
889 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200890
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100891 goto out_wakeup;
892 }
893
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200894 check->code = str2uic(b_head(&check->bi) + 9);
895 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200896
Willy Tarreaubd741542010-03-16 18:46:54 +0100897 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200898 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000899 /* 404 may be accepted as "stopping" only if the server was up */
900 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900901 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000902 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100903 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
904 /* Run content verification check... We know we have at least 13 chars */
905 if (!httpchk_expect(s, done))
906 goto wait_more_data;
907 }
908 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200909 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100910 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900911 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100912 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000913 else {
914 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900915 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000916 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200917 break;
918
919 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200920 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100921 goto wait_more_data;
922
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100923 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200924 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900925 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200926 else
Simon Horman4a741432013-02-23 15:35:38 +0900927 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200928 break;
929
930 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200931 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100932 goto wait_more_data;
933
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200934 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200935 if ((b_data(&check->bi) < strlen("000\r")) ||
936 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
937 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
938 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
939 cut_crlf(b_head(&check->bi));
940 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200941 goto out_wakeup;
942 }
943
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200944 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200945
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200946 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200947 cut_crlf(desc);
948
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100949 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200950 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900951 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
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_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200954 break;
955
Simon Hormana2b9dad2013-02-12 10:45:54 +0900956 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100957 int status = HCHK_STATUS_CHECKED;
958 const char *hs = NULL; /* health status */
959 const char *as = NULL; /* admin status */
960 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200961 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100962 const char *err = NULL; /* first error to report */
963 const char *wrn = NULL; /* first warning to report */
964 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900965
Willy Tarreau81f5d942013-12-09 20:51:51 +0100966 /* We're getting an agent check response. The agent could
967 * have been disabled in the mean time with a long check
968 * still pending. It is important that we ignore the whole
969 * response.
970 */
971 if (!(check->server->agent.state & CHK_ST_ENABLED))
972 break;
973
974 /* The agent supports strings made of a single line ended by the
975 * first CR ('\r') or LF ('\n'). This line is composed of words
976 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
977 * line may optionally contained a description of a state change
978 * after a sharp ('#'), which is only considered if a health state
979 * is announced.
980 *
981 * Words may be composed of :
982 * - a numeric weight suffixed by the percent character ('%').
983 * - a health status among "up", "down", "stopped", and "fail".
984 * - an admin status among "ready", "drain", "maint".
985 *
986 * These words may appear in any order. If multiple words of the
987 * same category appear, the last one wins.
988 */
989
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200990 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +0100991 while (*p && *p != '\n' && *p != '\r')
992 p++;
993
994 if (!*p) {
995 if (!done)
996 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900997
Willy Tarreau9809b782013-12-11 21:40:11 +0100998 /* at least inform the admin that the agent is mis-behaving */
999 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
1000 break;
1001 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001002
Willy Tarreau9809b782013-12-11 21:40:11 +01001003 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001004 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001005
Willy Tarreau81f5d942013-12-09 20:51:51 +01001006 while (*cmd) {
1007 /* look for next word */
1008 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1009 cmd++;
1010 continue;
1011 }
Simon Horman671b6f02013-11-25 10:46:39 +09001012
Willy Tarreau81f5d942013-12-09 20:51:51 +01001013 if (*cmd == '#') {
1014 /* this is the beginning of a health status description,
1015 * skip the sharp and blanks.
1016 */
1017 cmd++;
1018 while (*cmd == '\t' || *cmd == ' ')
1019 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001020 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001021 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001022
1023 /* find the end of the word so that we have a null-terminated
1024 * word between <cmd> and <p>.
1025 */
1026 p = cmd + 1;
1027 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1028 p++;
1029 if (*p)
1030 *p++ = 0;
1031
1032 /* first, health statuses */
1033 if (strcasecmp(cmd, "up") == 0) {
1034 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001035 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001036 hs = cmd;
1037 }
1038 else if (strcasecmp(cmd, "down") == 0) {
1039 check->health = 0;
1040 status = HCHK_STATUS_L7STS;
1041 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001042 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001043 else if (strcasecmp(cmd, "stopped") == 0) {
1044 check->health = 0;
1045 status = HCHK_STATUS_L7STS;
1046 hs = cmd;
1047 }
1048 else if (strcasecmp(cmd, "fail") == 0) {
1049 check->health = 0;
1050 status = HCHK_STATUS_L7STS;
1051 hs = cmd;
1052 }
1053 /* admin statuses */
1054 else if (strcasecmp(cmd, "ready") == 0) {
1055 as = cmd;
1056 }
1057 else if (strcasecmp(cmd, "drain") == 0) {
1058 as = cmd;
1059 }
1060 else if (strcasecmp(cmd, "maint") == 0) {
1061 as = cmd;
1062 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001063 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001064 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1065 ps = cmd;
1066 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001067 /* try to parse a maxconn here */
1068 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1069 cs = cmd;
1070 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001071 else {
1072 /* keep a copy of the first error */
1073 if (!err)
1074 err = cmd;
1075 }
1076 /* skip to next word */
1077 cmd = p;
1078 }
1079 /* here, cmd points either to \0 or to the beginning of a
1080 * description. Skip possible leading spaces.
1081 */
1082 while (*cmd == ' ' || *cmd == '\n')
1083 cmd++;
1084
1085 /* First, update the admin status so that we avoid sending other
1086 * possibly useless warnings and can also update the health if
1087 * present after going back up.
1088 */
1089 if (as) {
1090 if (strcasecmp(as, "drain") == 0)
1091 srv_adm_set_drain(check->server);
1092 else if (strcasecmp(as, "maint") == 0)
1093 srv_adm_set_maint(check->server);
1094 else
1095 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001096 }
1097
Willy Tarreau81f5d942013-12-09 20:51:51 +01001098 /* now change weights */
1099 if (ps) {
1100 const char *msg;
1101
1102 msg = server_parse_weight_change_request(s, ps);
1103 if (!wrn || !*wrn)
1104 wrn = msg;
1105 }
1106
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001107 if (cs) {
1108 const char *msg;
1109
1110 cs += strlen("maxconn:");
1111
1112 msg = server_parse_maxconn_change_request(s, cs);
1113 if (!wrn || !*wrn)
1114 wrn = msg;
1115 }
1116
Willy Tarreau81f5d942013-12-09 20:51:51 +01001117 /* and finally health status */
1118 if (hs) {
1119 /* We'll report some of the warnings and errors we have
1120 * here. Down reports are critical, we leave them untouched.
1121 * Lack of report, or report of 'UP' leaves the room for
1122 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001123 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001124 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001125 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001126
1127 if (!*msg || status == HCHK_STATUS_L7OKD) {
1128 if (err && *err)
1129 msg = err;
1130 else if (wrn && *wrn)
1131 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001132 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001133
1134 t = get_trash_chunk();
1135 chunk_printf(t, "via agent : %s%s%s%s",
1136 hs, *msg ? " (" : "",
1137 msg, *msg ? ")" : "");
1138
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001139 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001140 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001141 else if (err && *err) {
1142 /* No status change but we'd like to report something odd.
1143 * Just report the current state and copy the message.
1144 */
1145 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001146 set_server_check_status(check, status/*check->status*/,
1147 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001148
Willy Tarreau81f5d942013-12-09 20:51:51 +01001149 }
1150 else if (wrn && *wrn) {
1151 /* No status change but we'd like to report something odd.
1152 * Just report the current state and copy the message.
1153 */
1154 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001155 set_server_check_status(check, status/*check->status*/,
1156 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001157 }
1158 else
1159 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001160 break;
1161 }
1162
Willy Tarreau1620ec32011-08-06 17:05:02 +02001163 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001164 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001165 goto wait_more_data;
1166
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001167 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001168 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001169 }
1170 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001171 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1172 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001173 else
1174 desc = "PostgreSQL unknown error";
1175
Simon Horman4a741432013-02-23 15:35:38 +09001176 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001177 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001178 break;
1179
1180 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001181 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001182 goto wait_more_data;
1183
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001184 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001185 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001186 }
1187 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001188 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001189 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001190 break;
1191
1192 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001193 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001194 goto wait_more_data;
1195
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001196 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001197 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001198 /* We set the MySQL Version in description for information purpose
1199 * FIXME : it can be cool to use MySQL Version for other purpose,
1200 * like mark as down old MySQL server.
1201 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001202 if (b_data(&check->bi) > 51) {
1203 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001204 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001205 }
1206 else {
1207 if (!done)
1208 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001209
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001210 /* it seems we have a OK packet but without a valid length,
1211 * it must be a protocol error
1212 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001213 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001214 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001215 }
1216 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001217 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001218 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001219 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001220 }
1221 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001222 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1223 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1224 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001225
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001226 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001227 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001228 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001229 /* We have only one MySQL packet and it is a Handshake Initialization packet
1230 * but we need to have a second packet to know if it is alright
1231 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001232 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001233 goto wait_more_data;
1234 }
1235 else {
1236 /* We have only one packet and it is an Error packet,
1237 * an error message is attached, so we can display it
1238 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001239 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001240 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001241 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001242 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001243 } else if (b_data(&check->bi) > first_packet_len + 4) {
1244 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1245 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1246 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001247
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001248 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249 /* We have 2 packets and that's good */
1250 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001251 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001252 /* No error packet */
1253 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001254 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001255 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001256 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001257 }
1258 else {
1259 /* An error message is attached in the Error packet
1260 * so we can display it ! :)
1261 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001262 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001263 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001264 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001265 }
1266 }
1267 }
1268 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001269 if (!done)
1270 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001271
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001272 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001273 * it must be a protocol error
1274 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001275 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001276 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001277 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001278 }
1279 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001280 break;
1281
1282 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001283 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001284 goto wait_more_data;
1285
1286 /* Check if the server speaks LDAP (ASN.1/BER)
1287 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1288 * http://tools.ietf.org/html/rfc4511
1289 */
1290
1291 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1292 * LDAPMessage: 0x30: SEQUENCE
1293 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001294 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001295 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001296 }
1297 else {
1298 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001299 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001300
1301 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1302 * messageID: 0x02 0x01 0x01: INTEGER 1
1303 * protocolOp: 0x61: bindResponse
1304 */
1305 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001306 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001307 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001308 goto out_wakeup;
1309 }
1310
1311 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001312 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001313
1314 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1315 * ldapResult: 0x0a 0x01: ENUMERATION
1316 */
1317 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001318 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001319 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001320 goto out_wakeup;
1321 }
1322
1323 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1324 * resultCode
1325 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001326 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001327 if (check->code) {
1328 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 +02001329 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001330 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001331 }
1332 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001333 break;
1334
Christopher Fauletba7bc162016-11-07 21:07:38 +01001335 case PR_O2_SPOP_CHK: {
1336 unsigned int framesz;
1337 char err[HCHK_DESC_LEN];
1338
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001339 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001340 goto wait_more_data;
1341
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001342 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001343 framesz = ntohl(framesz);
1344
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001345 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001346 goto wait_more_data;
1347
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001348 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001349 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1350 else
1351 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1352 break;
1353 }
1354
Willy Tarreau1620ec32011-08-06 17:05:02 +02001355 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001356 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001357 break;
1358 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001359
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001360 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001361 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001362 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001363 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001364
Nick Chalk57b1bf72010-03-16 15:50:46 +00001365 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001366 *b_head(&check->bi) = '\0';
1367 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001368
Steven Davidovitz544d4812017-03-08 11:06:20 -08001369 /* Close the connection... We still attempt to nicely close if,
1370 * for instance, SSL needs to send a "close notify." Later, we perform
1371 * a hard close and reset the connection if some data are pending,
1372 * otherwise we end up with many TIME_WAITs and eat all the source port
1373 * range quickly. To avoid sending RSTs all the time, we first try to
1374 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001375 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001376 __cs_stop_both(cs);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001377 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001378
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001379 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001380 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001381 conn->flags |= CO_FL_ERROR;
1382
Willy Tarreaufdccded2008-08-29 18:19:04 +02001383 task_wakeup(t, TASK_WOKEN_IO);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001384out:
Willy Tarreau3267d362012-08-17 23:53:56 +02001385 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001386
1387 wait_more_data:
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001388 __cs_want_recv(cs);
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001389 cs->conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02001390 goto out;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001391}
1392
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001393/*
1394 * This function is used only for server health-checks. It handles connection
1395 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001396 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1397 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001398 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001399static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001400{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001401 struct connection *conn = cs->conn;
1402 struct check *check = cs->data;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001403 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001404
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001405 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001406
Willy Tarreauc09572f2017-10-04 11:58:22 +02001407 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001408 if (check->type == PR_O2_TCPCHK_CHK) {
1409 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001410 cs = check->cs;
1411 conn = cs_conn(cs);
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001412 } else if (!(check->wait_list.wait_reason & SUB_CAN_SEND))
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001413 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001414
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001415 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001416 /* We may get error reports bypassing the I/O handlers, typically
1417 * the case when sending a pure TCP check which fails, then the I/O
1418 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001419 * main processing task so let's simply wake it up. If we get here,
1420 * we expect errno to still be valid.
1421 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001422 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001423 __cs_stop_both(cs);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001424 task_wakeup(check->task, TASK_WOKEN_IO);
1425 }
Willy Tarreau2500fc22018-04-03 19:31:38 +02001426 else if (!(conn->flags & CO_FL_HANDSHAKE) && !(cs->flags & (CS_FL_DATA_RD_ENA|CS_FL_DATA_WR_ENA))) {
Willy Tarreau3be293f2014-02-05 18:31:24 +01001427 /* we may get here if only a connection probe was required : we
1428 * don't have any data to send nor anything expected in response,
1429 * so the completion of the connection establishment is enough.
1430 */
1431 task_wakeup(check->task, TASK_WOKEN_IO);
1432 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001433
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001434 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001435 /* We're here because nobody wants to handle the error, so we
1436 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001437 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001438 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001439 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001440 ret = -1;
Willy Tarreau2d351b62013-12-05 02:36:25 +01001441 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001442
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001443 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001444
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001445 /* if a connection got replaced, we must absolutely prevent the connection
1446 * handler from touching its fd, and perform the FD polling updates ourselves
1447 */
1448 if (ret < 0)
1449 conn_cond_update_polling(conn);
1450
1451 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001452}
1453
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001454struct data_cb check_conn_cb = {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001455 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001456 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001457};
1458
Willy Tarreaubaaee002006-06-26 02:48:02 +02001459/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001460 * updates the server's weight during a warmup stage. Once the final weight is
1461 * reached, the task automatically stops. Note that any server status change
1462 * must have updated s->last_change accordingly.
1463 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001464static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001465{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001466 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001467
1468 /* by default, plan on stopping the task */
1469 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001470 if ((s->next_admin & SRV_ADMF_MAINT) ||
1471 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001472 return t;
1473
Willy Tarreau892337c2014-05-13 23:41:20 +02001474 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001475 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001476
1477 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001478 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001479
1480 /* get back there in 1 second or 1/20th of the slowstart interval,
1481 * whichever is greater, resulting in small 5% steps.
1482 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001483 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001484 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1485 return t;
1486}
1487
Willy Tarreau894c6422017-10-04 15:58:52 +02001488/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1489 * none was found.
1490 */
1491static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1492{
1493 struct tcpcheck_rule *r;
1494
1495 list_for_each_entry(r, list, list) {
1496 if (r->action != TCPCHK_ACT_COMMENT)
1497 return r;
1498 }
1499 return NULL;
1500}
1501
Willy Tarreau2e993902011-10-31 11:53:20 +01001502/*
Simon Horman98637e52014-06-20 12:30:16 +09001503 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001504 *
1505 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001506 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1507 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1508 * - SF_ERR_SRVTO if there are no more servers
1509 * - SF_ERR_SRVCL if the connection was refused by the server
1510 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1511 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1512 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001513 * - 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 +01001514 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001515 * Note that we try to prevent the network stack from sending the ACK during the
1516 * connect() when a pure TCP check is used (without PROXY protocol).
1517 */
Simon Horman98637e52014-06-20 12:30:16 +09001518static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001519{
1520 struct check *check = t->context;
1521 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001522 struct conn_stream *cs = check->cs;
1523 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001524 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001525 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001526 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001527 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001528
Willy Tarreau00149122017-10-04 18:05:01 +02001529 /* we cannot have a connection here */
1530 if (conn)
1531 return SF_ERR_INTERNAL;
1532
Simon Hormanb00d17a2014-06-13 16:18:16 +09001533 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001534 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001535 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001536 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1537 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001538
1539 /* prepare the check buffer.
1540 * This should not be used if check is the secondary agent check
1541 * of a server as s->proxy->check_req will relate to the
1542 * configuration of the primary check. Similarly, tcp-check uses
1543 * its own strings.
1544 */
1545 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001546 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001547
1548 /* we want to check if this host replies to HTTP or SSLv3 requests
1549 * so we'll send the request, and won't wake the checker up now.
1550 */
1551 if ((check->type) == PR_O2_SSL3_CHK) {
1552 /* SSL requires that we put Unix time in the request */
1553 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001554 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001555 }
1556 else if ((check->type) == PR_O2_HTTP_CHK) {
1557 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001558 b_putblk(&check->bo, trash.area,
1559 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001560 /* prevent HTTP keep-alive when "http-check expect" is used */
1561 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001562 b_putist(&check->bo, ist("Connection: close\r\n"));
1563 b_putist(&check->bo, ist("\r\n"));
1564 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001565 }
1566 }
1567
James Brown55f9ff12015-10-21 18:19:05 -07001568 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001569 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001570 }
1571
Willy Tarreauf411cce2017-10-04 16:21:19 +02001572 /* for tcp-checks, the initial connection setup is handled separately as
1573 * it may be sent to a specific port and not to the server's.
1574 */
1575 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1576 tcpcheck_main(check);
1577 return SF_ERR_UP;
1578 }
1579
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001581 cs = check->cs = cs_new(NULL);
1582 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001583 return SF_ERR_RESOURCE;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001584 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02001585 /* Maybe there were an older connection we were waiting on */
1586 check->wait_list.wait_reason = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001587
Simon Horman41f58762015-01-30 11:22:56 +09001588 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001589 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001590 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001591 }
1592 else {
1593 /* we'll connect to the addr on the server */
1594 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001595 }
1596
Olivier Houchard6377a002017-12-01 22:04:05 +01001597 proto = protocol_by_family(conn->addr.to.ss_family);
Olivier Houchard6377a002017-12-01 22:04:05 +01001598 conn->target = &s->obj_type;
1599
Willy Tarreaube373152018-09-06 11:45:30 +02001600 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001601 int i = 0;
1602
1603 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001604 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001605 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001606
1607 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001608 }
1609
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001610 /* no client address */
1611 clear_addr(&conn->addr.from);
1612
Willy Tarreaube373152018-09-06 11:45:30 +02001613 conn_prepare(conn, proto, check->xprt);
Willy Tarreau175a2bb2018-09-12 12:02:05 +02001614 conn_install_mux(conn, &mux_pt_ops, cs, s->proxy);
Willy Tarreaube373152018-09-06 11:45:30 +02001615 cs_attach(cs, check, &check_conn_cb);
1616
Willy Tarreauf3d34822014-12-08 12:11:28 +01001617 /* only plain tcp-check supports quick ACK */
1618 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1619
Willy Tarreauf411cce2017-10-04 16:21:19 +02001620 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1621 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001622
Willy Tarreaue7dff022015-04-03 01:14:29 +02001623 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001624 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001625 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001626
1627 if (check->type)
1628 cs_want_send(cs);
1629
Olivier Houchard9130a962017-10-17 17:33:43 +02001630#ifdef USE_OPENSSL
1631 if (s->check.sni)
1632 ssl_sock_set_servername(conn, s->check.sni);
1633#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001634 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001635 conn->send_proxy_ofs = 1;
1636 conn->flags |= CO_FL_SEND_PROXY;
1637 }
1638
1639 return ret;
1640}
1641
Simon Horman98637e52014-06-20 12:30:16 +09001642static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001643static struct pool_head *pool_head_pid_list;
Christopher Faulet9dcf9b62017-11-13 10:34:01 +01001644__decl_hathreads(HA_SPINLOCK_T pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001645
1646void block_sigchld(void)
1647{
1648 sigset_t set;
1649 sigemptyset(&set);
1650 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001651 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001652}
1653
1654void unblock_sigchld(void)
1655{
1656 sigset_t set;
1657 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001658 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001659 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001660}
1661
Simon Horman98637e52014-06-20 12:30:16 +09001662static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1663{
1664 struct pid_list *elem;
1665 struct check *check = t->context;
1666
Willy Tarreaubafbe012017-11-24 17:34:44 +01001667 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001668 if (!elem)
1669 return NULL;
1670 elem->pid = pid;
1671 elem->t = t;
1672 elem->exited = 0;
1673 check->curpid = elem;
1674 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001675
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001676 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001677 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001678 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001679
Simon Horman98637e52014-06-20 12:30:16 +09001680 return elem;
1681}
1682
Simon Horman98637e52014-06-20 12:30:16 +09001683static void pid_list_del(struct pid_list *elem)
1684{
1685 struct check *check;
1686
1687 if (!elem)
1688 return;
1689
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001690 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001691 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001692 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001693
Simon Horman98637e52014-06-20 12:30:16 +09001694 if (!elem->exited)
1695 kill(elem->pid, SIGTERM);
1696
1697 check = elem->t->context;
1698 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001699 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001700}
1701
1702/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1703static void pid_list_expire(pid_t pid, int status)
1704{
1705 struct pid_list *elem;
1706
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001707 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001708 list_for_each_entry(elem, &pid_list, list) {
1709 if (elem->pid == pid) {
1710 elem->t->expire = now_ms;
1711 elem->status = status;
1712 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001713 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001714 break;
Simon Horman98637e52014-06-20 12:30:16 +09001715 }
1716 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001717 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001718}
1719
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001720static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001721{
1722 pid_t pid;
1723 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001724
Simon Horman98637e52014-06-20 12:30:16 +09001725 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1726 pid_list_expire(pid, status);
1727}
1728
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001729static int init_pid_list(void)
1730{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001731 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001732 /* Nothing to do */
1733 return 0;
1734
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001735 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001736 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1737 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001738 return 1;
1739 }
1740
Willy Tarreaubafbe012017-11-24 17:34:44 +01001741 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1742 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001743 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1744 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001745 return 1;
1746 }
1747
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001748 HA_SPIN_INIT(&pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001749
Simon Horman98637e52014-06-20 12:30:16 +09001750 return 0;
1751}
1752
Cyril Bontéac92a062014-12-27 22:28:38 +01001753/* helper macro to set an environment variable and jump to a specific label on failure. */
1754#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001755
1756/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001757 * helper function to allocate enough memory to store an environment variable.
1758 * It will also check that the environment variable is updatable, and silently
1759 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001760 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001761static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001762{
1763 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001764 char *envname;
1765 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001766
Cyril Bontéac92a062014-12-27 22:28:38 +01001767 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001768 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001769 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001770 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001771
1772 envname = extcheck_envs[idx].name;
1773 vmaxlen = extcheck_envs[idx].vmaxlen;
1774
1775 /* Check if the environment variable is already set, and silently reject
1776 * the update if this one is not updatable. */
1777 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1778 return 0;
1779
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001780 /* Instead of sending NOT_USED, sending an empty value is preferable */
1781 if (strcmp(value, "NOT_USED") == 0) {
1782 value = "";
1783 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001784
1785 len = strlen(envname) + 1;
1786 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1787 len += strlen(value);
1788 else
1789 len += vmaxlen;
1790
1791 if (!check->envp[idx])
1792 check->envp[idx] = malloc(len + 1);
1793
1794 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001795 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001796 return 1;
1797 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001798 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001799 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001800 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001801 return 1;
1802 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001803 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001804 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001805 return 1;
1806 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001807 return 0;
1808}
Simon Horman98637e52014-06-20 12:30:16 +09001809
1810static int prepare_external_check(struct check *check)
1811{
1812 struct server *s = check->server;
1813 struct proxy *px = s->proxy;
1814 struct listener *listener = NULL, *l;
1815 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001816 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001817 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001818
1819 list_for_each_entry(l, &px->conf.listeners, by_fe)
1820 /* Use the first INET, INET6 or UNIX listener */
1821 if (l->addr.ss_family == AF_INET ||
1822 l->addr.ss_family == AF_INET6 ||
1823 l->addr.ss_family == AF_UNIX) {
1824 listener = l;
1825 break;
1826 }
1827
Simon Horman98637e52014-06-20 12:30:16 +09001828 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001829 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1830 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001831 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001832 goto err;
1833 }
Simon Horman98637e52014-06-20 12:30:16 +09001834
Cyril Bontéac92a062014-12-27 22:28:38 +01001835 check->argv = calloc(6, sizeof(char *));
1836 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001837 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001838 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001839 }
Simon Horman98637e52014-06-20 12:30:16 +09001840
1841 check->argv[0] = px->check_command;
1842
Cyril Bonté777be862014-12-02 21:21:35 +01001843 if (!listener) {
1844 check->argv[1] = strdup("NOT_USED");
1845 check->argv[2] = strdup("NOT_USED");
1846 }
1847 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001848 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001849 addr_to_str(&listener->addr, buf, sizeof(buf));
1850 check->argv[1] = strdup(buf);
1851 port_to_str(&listener->addr, buf, sizeof(buf));
1852 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001853 }
1854 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001855 const struct sockaddr_un *un;
1856
1857 un = (struct sockaddr_un *)&listener->addr;
1858 check->argv[1] = strdup(un->sun_path);
1859 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001860 }
1861 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001862 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001863 goto err;
1864 }
1865
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001866 addr_to_str(&s->addr, buf, sizeof(buf));
1867 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001868
1869 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1870 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1871 else
1872 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001873 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001874
Cyril Bontéac92a062014-12-27 22:28:38 +01001875 for (i = 0; i < 5; i++) {
1876 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001877 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001878 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001879 }
1880 }
Simon Horman98637e52014-06-20 12:30:16 +09001881
Cyril Bontéac92a062014-12-27 22:28:38 +01001882 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001883 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001884 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1885 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1886 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1887 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001888 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001889 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1890 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1891 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1892 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1893 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1894 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1895
1896 /* Ensure that we don't leave any hole in check->envp */
1897 for (i = 0; i < EXTCHK_SIZE; i++)
1898 if (!check->envp[i])
1899 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001900
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001901 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001902err:
1903 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001904 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001905 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001906 free(check->envp);
1907 check->envp = NULL;
1908 }
1909
1910 if (check->argv) {
1911 for (i = 1; i < 5; i++)
1912 free(check->argv[i]);
1913 free(check->argv);
1914 check->argv = NULL;
1915 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001916 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001917}
1918
Simon Hormanb00d17a2014-06-13 16:18:16 +09001919/*
Simon Horman98637e52014-06-20 12:30:16 +09001920 * establish a server health-check that makes use of a process.
1921 *
1922 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001923 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001924 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001925 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001926 *
1927 * Blocks and then unblocks SIGCHLD
1928 */
1929static int connect_proc_chk(struct task *t)
1930{
Cyril Bontéac92a062014-12-27 22:28:38 +01001931 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001932 struct check *check = t->context;
1933 struct server *s = check->server;
1934 struct proxy *px = s->proxy;
1935 int status;
1936 pid_t pid;
1937
Willy Tarreaue7dff022015-04-03 01:14:29 +02001938 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001939
1940 block_sigchld();
1941
1942 pid = fork();
1943 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001944 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1945 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001946 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1947 goto out;
1948 }
1949 if (pid == 0) {
1950 /* Child */
1951 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001952 int fd;
1953
1954 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1955 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1956
1957 while (fd < global.rlimit_nofile)
1958 close(fd++);
1959
Simon Horman98637e52014-06-20 12:30:16 +09001960 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001961 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001962 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001963 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1964 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001965 exit(-1);
1966 }
1967
1968 /* Parent */
1969 if (check->result == CHK_RES_UNKNOWN) {
1970 if (pid_list_add(pid, t) != NULL) {
1971 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1972
1973 if (px->timeout.check && px->timeout.connect) {
1974 int t_con = tick_add(now_ms, px->timeout.connect);
1975 t->expire = tick_first(t->expire, t_con);
1976 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001977 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001978 goto out;
1979 }
1980 else {
1981 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1982 }
1983 kill(pid, SIGTERM); /* process creation error */
1984 }
1985 else
1986 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1987
1988out:
1989 unblock_sigchld();
1990 return status;
1991}
1992
1993/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02001994 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001995 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001996 *
1997 * Please do NOT place any return statement in this function and only leave
1998 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001999 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002000static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002001{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002002 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002003 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09002004 int rv;
2005 int ret;
2006 int expired = tick_is_expired(t->expire, now_ms);
2007
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002008 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002009 if (!(check->state & CHK_ST_INPROGRESS)) {
2010 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002011 if (!expired) /* woke up too early */
2012 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002013
2014 /* we don't send any health-checks when the proxy is
2015 * stopped, the server should not be checked or the check
2016 * is disabled.
2017 */
2018 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2019 s->proxy->state == PR_STSTOPPED)
2020 goto reschedule;
2021
2022 /* we'll initiate a new check */
2023 set_server_check_status(check, HCHK_STATUS_START, NULL);
2024
2025 check->state |= CHK_ST_INPROGRESS;
2026
Simon Hormandbf70192015-01-30 11:22:53 +09002027 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002028 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002029 /* the process was forked, we allow up to min(inter,
2030 * timeout.connect) for it to report its status, but
2031 * only when timeout.check is set as it may be to short
2032 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002033 */
2034 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2035
2036 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2037 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2038 t->expire = tick_first(t->expire, t_con);
2039 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002040 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002041 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002042 }
2043
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002044 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002045
2046 check->state &= ~CHK_ST_INPROGRESS;
2047 check_notify_failure(check);
2048
2049 /* we allow up to min(inter, timeout.connect) for a connection
2050 * to establish but only when timeout.check is set
2051 * as it may be to short for a full check otherwise
2052 */
2053 while (tick_is_expired(t->expire, now_ms)) {
2054 int t_con;
2055
2056 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2057 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2058
2059 if (s->proxy->timeout.check)
2060 t->expire = tick_first(t->expire, t_con);
2061 }
2062 }
2063 else {
2064 /* there was a test running.
2065 * First, let's check whether there was an uncaught error,
2066 * which can happen on connect timeout or error.
2067 */
2068 if (check->result == CHK_RES_UNKNOWN) {
2069 /* good connection is enough for pure TCP check */
2070 struct pid_list *elem = check->curpid;
2071 int status = HCHK_STATUS_UNKNOWN;
2072
2073 if (elem->exited) {
2074 status = elem->status; /* Save in case the process exits between use below */
2075 if (!WIFEXITED(status))
2076 check->code = -1;
2077 else
2078 check->code = WEXITSTATUS(status);
2079 if (!WIFEXITED(status) || WEXITSTATUS(status))
2080 status = HCHK_STATUS_PROCERR;
2081 else
2082 status = HCHK_STATUS_PROCOK;
2083 } else if (expired) {
2084 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002085 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002086 kill(elem->pid, SIGTERM);
2087 }
2088 set_server_check_status(check, status, NULL);
2089 }
2090
2091 if (check->result == CHK_RES_FAILED) {
2092 /* a failure or timeout detected */
2093 check_notify_failure(check);
2094 }
2095 else if (check->result == CHK_RES_CONDPASS) {
2096 /* check is OK but asks for stopping mode */
2097 check_notify_stopping(check);
2098 }
2099 else if (check->result == CHK_RES_PASSED) {
2100 /* a success was detected */
2101 check_notify_success(check);
2102 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002103 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002104 check->state &= ~CHK_ST_INPROGRESS;
2105
2106 pid_list_del(check->curpid);
2107
2108 rv = 0;
2109 if (global.spread_checks > 0) {
2110 rv = srv_getinter(check) * global.spread_checks / 100;
2111 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2112 }
2113 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2114 }
2115
2116 reschedule:
2117 while (tick_is_expired(t->expire, now_ms))
2118 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002119
2120 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002121 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002122 return t;
2123}
2124
2125/*
2126 * manages a server health-check that uses a connection. Returns
2127 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002128 *
2129 * Please do NOT place any return statement in this function and only leave
2130 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002131 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002132static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002133{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002134 struct check *check = context;
Simon Horman4a741432013-02-23 15:35:38 +09002135 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002136 struct conn_stream *cs = check->cs;
2137 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002138 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002139 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002140 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002141
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002142 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002143 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002144 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002145 if (!expired) /* woke up too early */
2146 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002147
Simon Horman671b6f02013-11-25 10:46:39 +09002148 /* we don't send any health-checks when the proxy is
2149 * stopped, the server should not be checked or the check
2150 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002151 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002152 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002153 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002154 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002155
2156 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002157 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002158
Willy Tarreau2c115e52013-12-11 19:41:16 +01002159 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002160 b_reset(&check->bi);
2161 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002162
Simon Hormandbf70192015-01-30 11:22:53 +09002163 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002164 cs = check->cs;
2165 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002166
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002167 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002168 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002169 goto out_unlock;
2170
Willy Tarreaue7dff022015-04-03 01:14:29 +02002171 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002172 /* we allow up to min(inter, timeout.connect) for a connection
2173 * to establish but only when timeout.check is set
2174 * as it may be to short for a full check otherwise
2175 */
Simon Horman4a741432013-02-23 15:35:38 +09002176 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002177
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002178 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2179 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2180 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002181 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002182
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002183 if (check->type) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002184 cs_want_recv(cs); /* prepare for reading a possible reply */
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002185 __event_srv_chk_r(cs);
2186 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002187
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002188 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002189 goto reschedule;
2190
Willy Tarreaue7dff022015-04-03 01:14:29 +02002191 case SF_ERR_SRVTO: /* ETIMEDOUT */
2192 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002193 if (conn)
2194 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002195 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002196 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002197 /* should share same code than cases below */
2198 case SF_ERR_CHK_PORT:
2199 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002200 case SF_ERR_PRXCOND:
2201 case SF_ERR_RESOURCE:
2202 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002203 if (conn)
2204 conn->flags |= CO_FL_ERROR;
2205 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002206 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002207 }
2208
Willy Tarreau5a78f362012-11-23 12:47:05 +01002209 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002210 if (cs) {
2211 cs_destroy(cs);
2212 cs = check->cs = NULL;
2213 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002214 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002215
Willy Tarreau2c115e52013-12-11 19:41:16 +01002216 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002217 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002218
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002219 /* we allow up to min(inter, timeout.connect) for a connection
2220 * to establish but only when timeout.check is set
2221 * as it may be to short for a full check otherwise
2222 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002223 while (tick_is_expired(t->expire, now_ms)) {
2224 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002225
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002226 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002227 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002228
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002229 if (s->proxy->timeout.check)
2230 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002231 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002232 }
2233 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002234 /* there was a test running.
2235 * First, let's check whether there was an uncaught error,
2236 * which can happen on connect timeout or error.
2237 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002238 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002239 /* good connection is enough for pure TCP check */
2240 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002241 if (check->use_ssl)
2242 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002243 else
Simon Horman4a741432013-02-23 15:35:38 +09002244 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002245 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002246 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002247 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002248 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002249 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002250 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002251 }
2252
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002253 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002254 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002255 /* The check was aborted and the connection was not yet closed.
2256 * This can happen upon timeout, or when an external event such
2257 * as a failed response coupled with "observe layer7" caused the
2258 * server state to be suddenly changed.
2259 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002260 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002261 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002262 }
2263
Willy Tarreauac59f362017-10-08 11:10:19 +02002264 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002265 cs_destroy(cs);
2266 cs = check->cs = NULL;
2267 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002268 }
2269
Willy Tarreauaf549582014-05-16 17:37:50 +02002270 if (check->result == CHK_RES_FAILED) {
2271 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002272 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002273 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002274 else if (check->result == CHK_RES_CONDPASS) {
2275 /* check is OK but asks for stopping mode */
2276 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002277 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002278 else if (check->result == CHK_RES_PASSED) {
2279 /* a success was detected */
2280 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002281 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002282 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002283 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002284
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002285 rv = 0;
2286 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002287 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002288 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002289 }
Simon Horman4a741432013-02-23 15:35:38 +09002290 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002291 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002292
2293 reschedule:
2294 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002295 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002296 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002297 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002298 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002299}
2300
Simon Horman98637e52014-06-20 12:30:16 +09002301/*
2302 * manages a server health-check. Returns
2303 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2304 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002305static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002306{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002307 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002308
2309 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002310 return process_chk_proc(t, context, state);
2311 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002312
Simon Horman98637e52014-06-20 12:30:16 +09002313}
2314
Simon Horman5c942422013-11-25 10:46:32 +09002315static int start_check_task(struct check *check, int mininter,
2316 int nbcheck, int srvpos)
2317{
2318 struct task *t;
2319 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002320 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002321 ha_alert("Starting [%s:%s] check: out of memory.\n",
2322 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002323 return 0;
2324 }
2325
2326 check->task = t;
2327 t->process = process_chk;
2328 t->context = check;
2329
Willy Tarreau1746eec2014-04-25 10:46:47 +02002330 if (mininter < srv_getinter(check))
2331 mininter = srv_getinter(check);
2332
2333 if (global.max_spread_checks && mininter > global.max_spread_checks)
2334 mininter = global.max_spread_checks;
2335
Simon Horman5c942422013-11-25 10:46:32 +09002336 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002337 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002338 check->start = now;
2339 task_queue(t);
2340
2341 return 1;
2342}
2343
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002344/*
2345 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002346 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002347 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002348static int start_checks()
2349{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002350
2351 struct proxy *px;
2352 struct server *s;
2353 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002354 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002355
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002356 /* 1- count the checkers to run simultaneously.
2357 * We also determine the minimum interval among all of those which
2358 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2359 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002360 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002361 * too short an interval for all others.
2362 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002363 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002364 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002365 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002366 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002367 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002368 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002369 }
2370 /* We need a warmup task that will be called when the server
2371 * state switches from down to up.
2372 */
2373 s->warmup = t;
2374 t->process = server_warmup;
2375 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002376 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002377 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002378 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 +01002379 }
2380
Willy Tarreaud8514a22013-12-11 21:10:14 +01002381 if (s->check.state & CHK_ST_CONFIGURED) {
2382 nbcheck++;
2383 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2384 (!mininter || mininter > srv_getinter(&s->check)))
2385 mininter = srv_getinter(&s->check);
2386 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002387
Willy Tarreaud8514a22013-12-11 21:10:14 +01002388 if (s->agent.state & CHK_ST_CONFIGURED) {
2389 nbcheck++;
2390 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2391 (!mininter || mininter > srv_getinter(&s->agent)))
2392 mininter = srv_getinter(&s->agent);
2393 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002394 }
2395 }
2396
Simon Horman4a741432013-02-23 15:35:38 +09002397 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002398 return 0;
2399
2400 srand((unsigned)time(NULL));
2401
2402 /*
2403 * 2- start them as far as possible from each others. For this, we will
2404 * start them after their interval set to the min interval divided by
2405 * the number of servers, weighted by the server's position in the list.
2406 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002407 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002408 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2409 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002410 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002411 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002412 }
2413 }
2414
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002415 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002416 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002417 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002418 if (s->check.type == PR_O2_EXT_CHK) {
2419 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002420 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002421 }
Simon Hormand60d6912013-11-25 10:46:36 +09002422 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002423 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002424 srvpos++;
2425 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002426
Simon Hormand60d6912013-11-25 10:46:36 +09002427 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002428 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002429 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002430 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002431 }
2432 srvpos++;
2433 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002434 }
2435 }
2436 return 0;
2437}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002438
2439/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002440 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002441 * The buffer MUST be terminated by a null byte before calling this function.
2442 * Sets server status appropriately. The caller is responsible for ensuring
2443 * that the buffer contains at least 13 characters. If <done> is zero, we may
2444 * return 0 to indicate that data is required to decide of a match.
2445 */
2446static int httpchk_expect(struct server *s, int done)
2447{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002448 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002449 char status_code[] = "000";
2450 char *contentptr;
2451 int crlf;
2452 int ret;
2453
2454 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2455 case PR_O2_EXP_STS:
2456 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002457 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2458 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002459
2460 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2461 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2462 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002463 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002464
2465 /* we necessarily have the response, so there are no partial failures */
2466 if (s->proxy->options2 & PR_O2_EXP_INV)
2467 ret = !ret;
2468
Simon Horman4a741432013-02-23 15:35:38 +09002469 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002470 break;
2471
2472 case PR_O2_EXP_STR:
2473 case PR_O2_EXP_RSTR:
2474 /* very simple response parser: ignore CR and only count consecutive LFs,
2475 * stop with contentptr pointing to first char after the double CRLF or
2476 * to '\0' if crlf < 2.
2477 */
2478 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002479 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002480 if (crlf >= 2)
2481 break;
2482 if (*contentptr == '\r')
2483 continue;
2484 else if (*contentptr == '\n')
2485 crlf++;
2486 else
2487 crlf = 0;
2488 }
2489
2490 /* Check that response contains a body... */
2491 if (crlf < 2) {
2492 if (!done)
2493 return 0;
2494
Simon Horman4a741432013-02-23 15:35:38 +09002495 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002496 "HTTP content check could not find a response body");
2497 return 1;
2498 }
2499
2500 /* Check that response body is not empty... */
2501 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002502 if (!done)
2503 return 0;
2504
Simon Horman4a741432013-02-23 15:35:38 +09002505 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002506 "HTTP content check found empty response body");
2507 return 1;
2508 }
2509
2510 /* Check the response content against the supplied string
2511 * or regex... */
2512 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2513 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2514 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002515 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002516
2517 /* if we don't match, we may need to wait more */
2518 if (!ret && !done)
2519 return 0;
2520
2521 if (ret) {
2522 /* content matched */
2523 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002524 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002525 "HTTP check matched unwanted content");
2526 else
Simon Horman4a741432013-02-23 15:35:38 +09002527 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002528 "HTTP content check matched");
2529 }
2530 else {
2531 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002532 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002533 "HTTP check did not match unwanted content");
2534 else
Simon Horman4a741432013-02-23 15:35:38 +09002535 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002536 "HTTP content check did not match");
2537 }
2538 break;
2539 }
2540 return 1;
2541}
2542
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002543/*
2544 * return the id of a step in a send/expect session
2545 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002546static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002547{
2548 struct tcpcheck_rule *cur = NULL, *next = NULL;
2549 int i = 0;
2550
Willy Tarreau213c6782014-10-02 14:51:02 +02002551 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002552 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002553 return 0;
2554
Simon Hormane16c1b32015-01-30 11:22:57 +09002555 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002556
2557 /* no step => first step */
2558 if (cur == NULL)
2559 return 1;
2560
2561 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002562 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002563 if (next->list.p == &cur->list)
2564 break;
2565 ++i;
2566 }
2567
2568 return i;
2569}
2570
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002571/*
2572 * return the latest known comment before (including) the given stepid
2573 * returns NULL if no comment found
2574 */
2575static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2576{
2577 struct tcpcheck_rule *cur = NULL;
2578 char *ret = NULL;
2579 int i = 0;
2580
2581 /* not even started anything yet, return latest comment found before any action */
2582 if (!check->current_step) {
2583 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2584 if (cur->action == TCPCHK_ACT_COMMENT)
2585 ret = cur->comment;
2586 else
2587 goto return_comment;
2588 }
2589 }
2590
2591 i = 1;
2592 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2593 if (cur->comment)
2594 ret = cur->comment;
2595
2596 if (i >= stepid)
2597 goto return_comment;
2598
2599 ++i;
2600 }
2601
2602 return_comment:
2603 return ret;
2604}
2605
Willy Tarreaube74b882017-10-04 16:22:49 +02002606/* proceed with next steps for the TCP checks <check>. Note that this is called
2607 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002608 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2609 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002610 *
2611 * Please do NOT place any return statement in this function and only leave
2612 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002613 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002614static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002615{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002616 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002617 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002618 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002619 struct conn_stream *cs = check->cs;
2620 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002621 struct server *s = check->server;
2622 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002623 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002624 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002625
Willy Tarreauef953952014-10-02 14:30:14 +02002626 /* here, we know that the check is complete or that it failed */
2627 if (check->result != CHK_RES_UNKNOWN)
2628 goto out_end_tcpcheck;
2629
2630 /* We have 4 possibilities here :
2631 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002632 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002633 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2634 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002635 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002636 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002637 * be waiting for a connection attempt to complete. conn!=NULL.
2638 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002639 *
2640 * #2 and #3 are quite similar, we want both the connection and the
2641 * handshake to complete before going any further. Thus we must always
2642 * wait for a connection to complete unless we're before and existing
2643 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002644 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002645
2646 /* find first rule and skip comments */
2647 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2648 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2649 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2650
Willy Tarreau00149122017-10-04 18:05:01 +02002651 if ((check->current_step || &next->list == head) &&
2652 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002653 /* we allow up to min(inter, timeout.connect) for a connection
2654 * to establish but only when timeout.check is set
2655 * as it may be to short for a full check otherwise
2656 */
2657 while (tick_is_expired(t->expire, now_ms)) {
2658 int t_con;
2659
2660 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2661 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2662
2663 if (s->proxy->timeout.check)
2664 t->expire = tick_first(t->expire, t_con);
2665 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002666 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002667 }
2668
Willy Tarreauef953952014-10-02 14:30:14 +02002669 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002670 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002671 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002672 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002673 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002674
Willy Tarreau213c6782014-10-02 14:51:02 +02002675 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002676 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002677 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002678 b_reset(&check->bo);
2679 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002680 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002681 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2682 if (s->proxy->timeout.check)
2683 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2684 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002685
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002686 /* It's only the rules which will enable send/recv */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002687 if (cs)
2688 cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002689
Willy Tarreauabca5b62013-12-06 14:19:25 +01002690 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002691 /* We have to try to flush the output buffer before reading, at
2692 * the end, or if we're about to send a string that does not fit
2693 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002694 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002695 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002696 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002697 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002698 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002699 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002700 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002701
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002702 __cs_want_send(cs);
Olivier Houcharded0f2072018-08-16 15:41:52 +02002703 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002704 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002705
2706 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002707 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002708 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002709 __cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002710 goto out_end_tcpcheck;
2711 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002712 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002713 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002714 if (b_data(&check->bo)) {
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002715 cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list);
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002716 goto out;
2717 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002718 }
2719
Willy Tarreau263013d2015-05-13 11:59:14 +02002720 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002721 break;
2722
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002723 /* have 'next' point to the next rule or NULL if we're on the
2724 * last one, connect() needs this.
2725 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002726 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002727
2728 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002729 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002730 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002731
2732 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002733 if (&next->list == head)
2734 next = NULL;
2735
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002736 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2737 struct protocol *proto;
2738 struct xprt_ops *xprt;
2739
Willy Tarreau00149122017-10-04 18:05:01 +02002740 /* For a connect action we'll create a new connection.
2741 * We may also have to kill a previous one. But we don't
2742 * want to leave *without* a connection if we came here
2743 * from the connection layer, hence with a connection.
2744 * Thus we'll proceed in the following order :
2745 * 1: close but not release previous connection
2746 * 2: try to get a new connection
2747 * 3: release and replace the old one on success
2748 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002749 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002750 cs_close(check->cs);
2751 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002752 }
2753
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002754 /* mark the step as started */
2755 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002756
2757 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002758 cs = cs_new(NULL);
2759 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002760 step = tcpcheck_get_step_id(check);
2761 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2762 comment = tcpcheck_get_step_comment(check, step);
2763 if (comment)
2764 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002765 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2766 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002767 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002768 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002769 }
2770
Willy Tarreauac59f362017-10-08 11:10:19 +02002771 if (check->cs)
2772 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002773
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002774 check->cs = cs;
2775 conn = cs->conn;
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002776 /* Maybe there were an older connection we were waiting on */
2777 check->wait_list.wait_reason = 0;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002778 conn->target = &s->obj_type;
2779
2780 /* no client address */
2781 clear_addr(&conn->addr.from);
2782
Simon Horman41f58762015-01-30 11:22:56 +09002783 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002784 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002785 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002786 }
2787 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002788 /* we'll connect to the addr on the server */
2789 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002790 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002791 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002792
2793 /* port */
2794 if (check->current_step->port)
2795 set_host_port(&conn->addr.to, check->current_step->port);
2796 else if (check->port)
2797 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002798 else if (s->svc_port)
2799 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002800
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002801 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002802 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002803 }
2804 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002805 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002806 }
Willy Tarreaube373152018-09-06 11:45:30 +02002807
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002808 conn_prepare(conn, proto, xprt);
Willy Tarreau175a2bb2018-09-12 12:02:05 +02002809 conn_install_mux(conn, &mux_pt_ops, cs, s->proxy);
Willy Tarreaube373152018-09-06 11:45:30 +02002810 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002811
Willy Tarreaue7dff022015-04-03 01:14:29 +02002812 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002813 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002814 ret = proto->connect(conn,
2815 1 /* I/O polling is always needed */,
2816 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002817 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2818 conn->send_proxy_ofs = 1;
2819 conn->flags |= CO_FL_SEND_PROXY;
2820 }
2821
2822 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002823 * - SF_ERR_NONE if everything's OK
2824 * - SF_ERR_SRVTO if there are no more servers
2825 * - SF_ERR_SRVCL if the connection was refused by the server
2826 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2827 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2828 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002829 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002830 * Note that we try to prevent the network stack from sending the ACK during the
2831 * connect() when a pure TCP check is used (without PROXY protocol).
2832 */
2833 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002834 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002835 /* we allow up to min(inter, timeout.connect) for a connection
2836 * to establish but only when timeout.check is set
2837 * as it may be to short for a full check otherwise
2838 */
2839 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2840
2841 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2842 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2843 t->expire = tick_first(t->expire, t_con);
2844 }
2845 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002846 case SF_ERR_SRVTO: /* ETIMEDOUT */
2847 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002848 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002849 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002850 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002851 comment = tcpcheck_get_step_comment(check, step);
2852 if (comment)
2853 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002854 set_server_check_status(check, HCHK_STATUS_L4CON,
2855 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002856 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002857 case SF_ERR_PRXCOND:
2858 case SF_ERR_RESOURCE:
2859 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002860 step = tcpcheck_get_step_id(check);
2861 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002862 comment = tcpcheck_get_step_comment(check, step);
2863 if (comment)
2864 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002865 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2866 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002867 goto out_end_tcpcheck;
2868 }
2869
2870 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002871 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002872
2873 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002874 while (&check->current_step->list != head &&
2875 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002876 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002877
Willy Tarreauf2c87352015-05-13 12:08:21 +02002878 if (&check->current_step->list == head)
2879 break;
2880
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002881 } /* end 'connect' */
2882 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2883 /* mark the step as started */
2884 check->last_started_step = check->current_step;
2885
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002886 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002887 if (*b_head(&check->bi) != '\0') {
2888 *b_head(&check->bi) = '\0';
2889 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002890 }
2891
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002892 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002893 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002894 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002895 goto out_end_tcpcheck;
2896 }
2897
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002898 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002899 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002900 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002901 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002902 set_server_check_status(check, HCHK_STATUS_L7RSP,
2903 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002904 goto out_end_tcpcheck;
2905 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002906
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002907 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002908 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002909 continue;
2910
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002911 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2912 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002913
Willy Tarreauabca5b62013-12-06 14:19:25 +01002914 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002915 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002916
2917 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002918 while (&check->current_step->list != head &&
2919 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002920 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002921
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002922 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002923 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002924 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002925 goto out_end_tcpcheck;
2926
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002927 __cs_want_recv(cs);
Olivier Houchard511efea2018-08-16 15:30:32 +02002928 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002929 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002930 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002931 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002932 /* Report network errors only if we got no other data. Otherwise
2933 * we'll let the upper layers decide whether the response is OK
2934 * or not. It is very common that an RST sent by the server is
2935 * reported as an error just after the last data chunk.
2936 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002937 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002938 goto out_end_tcpcheck;
2939 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002940 }
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002941 else {
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02002942 conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list);
Willy Tarreau263013d2015-05-13 11:59:14 +02002943 break;
Olivier Houchardaf4021e2018-08-09 13:06:55 +02002944 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002945 }
2946
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002947 /* mark the step as started */
2948 check->last_started_step = check->current_step;
2949
2950
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002951 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002952 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002953 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002954 if (b_data(&check->bi) < b_size(&check->bi)) {
2955 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002956 }
2957 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002958 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002959 done = 1; /* buffer full, don't wait for more data */
2960 }
2961
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002962 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002963
2964 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002965 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002966 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002967 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002968
2969 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002970 step = tcpcheck_get_step_id(check);
2971 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002972 comment = tcpcheck_get_step_comment(check, step);
2973 if (comment)
2974 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002975 set_server_check_status(check, HCHK_STATUS_L7RSP,
2976 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977
2978 goto out_end_tcpcheck;
2979 }
2980
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002981 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002982 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002983
Willy Tarreaua970c282013-12-06 12:47:19 +01002984 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002985 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002986 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 +02002987 else if (check->current_step->expect_regex != NULL)
2988 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002989
2990 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002991 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002992
2993 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002994 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002995 if (ret) {
2996 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002997 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002998 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002999 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003000 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003001 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003002 }
3003 else {
3004 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003005 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003006 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003007 comment = tcpcheck_get_step_comment(check, step);
3008 if (comment)
3009 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003010 set_server_check_status(check, HCHK_STATUS_L7RSP,
3011 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003012 goto out_end_tcpcheck;
3013 }
3014 /* matched and was supposed to => OK, next step */
3015 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003016 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003017 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003018
3019 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003020 while (&check->current_step->list != head &&
3021 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003022 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003023
Willy Tarreauf2c87352015-05-13 12:08:21 +02003024 if (&check->current_step->list == head)
3025 break;
3026
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003027 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003028 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003029 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003030 }
3031 }
3032 else {
3033 /* not matched */
3034 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003035 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003036 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003037 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003038
3039 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003040 while (&check->current_step->list != head &&
3041 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003042 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003043
Willy Tarreauf2c87352015-05-13 12:08:21 +02003044 if (&check->current_step->list == head)
3045 break;
3046
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003047 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003048 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003049 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003050 }
3051 /* not matched but was supposed to => ERROR */
3052 else {
3053 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003054 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003055 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003056 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003057 }
3058 else {
3059 /* we were looking for a regex */
3060 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003061 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003062 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003063 comment = tcpcheck_get_step_comment(check, step);
3064 if (comment)
3065 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003066 set_server_check_status(check, HCHK_STATUS_L7RSP,
3067 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003068 goto out_end_tcpcheck;
3069 }
3070 }
3071 } /* end expect */
3072 } /* end loop over double chained step list */
3073
Baptiste Assmann248f1172018-03-01 21:49:01 +01003074 /* don't do anything until the connection is established */
3075 if (!(conn->flags & CO_FL_CONNECTED)) {
3076 /* update expire time, should be done by process_chk */
3077 /* we allow up to min(inter, timeout.connect) for a connection
3078 * to establish but only when timeout.check is set
3079 * as it may be to short for a full check otherwise
3080 */
3081 while (tick_is_expired(t->expire, now_ms)) {
3082 int t_con;
3083
3084 t_con = tick_add(t->expire, s->proxy->timeout.connect);
3085 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3086
3087 if (s->proxy->timeout.check)
3088 t->expire = tick_first(t->expire, t_con);
3089 }
3090 goto out;
3091 }
3092
Willy Tarreau263013d2015-05-13 11:59:14 +02003093 /* We're waiting for some I/O to complete, we've reached the end of the
3094 * rules, or both. Do what we have to do, otherwise we're done.
3095 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003096 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003097 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3098 goto out_end_tcpcheck;
3099 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003100
Willy Tarreau53c5a042015-05-13 11:38:17 +02003101 /* warning, current_step may now point to the head */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003102 if (b_data(&check->bo))
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003103 __cs_want_send(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003104
Willy Tarreau53c5a042015-05-13 11:38:17 +02003105 if (&check->current_step->list != head &&
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003106 check->current_step->action == TCPCHK_ACT_EXPECT) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003107 __cs_want_recv(cs);
Olivier Houchardaf4021e2018-08-09 13:06:55 +02003108 __event_srv_chk_r(cs);
3109 }
Christopher Fauletb6102852017-11-28 10:06:29 +01003110 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003111
3112 out_end_tcpcheck:
3113 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003114 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003115 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003116
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003117 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003118 check->current_step = NULL;
3119
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003120 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003121 conn->flags |= CO_FL_ERROR;
3122
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003123 __cs_stop_both(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01003124
Christopher Fauletb6102852017-11-28 10:06:29 +01003125 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003126 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003127}
3128
Simon Hormanb1900d52015-01-30 11:22:54 +09003129const char *init_check(struct check *check, int type)
3130{
3131 check->type = type;
3132
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003133 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3134 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003135
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003136 check->bi.area = calloc(check->bi.size, sizeof(char));
3137 check->bo.area = calloc(check->bo.size, sizeof(char));
3138
3139 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003140 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003141
Olivier Houchard26e1a8f2018-09-12 15:15:12 +02003142 check->wait_list.task = tasklet_new();
3143 if (!check->wait_list.task)
3144 return "out of memroy while allocating check tasklet";
3145 LIST_INIT(&check->wait_list.list);
3146 check->wait_list.wait_reason = 0;
3147 check->wait_list.task->process = event_srv_chk_io;
3148 check->wait_list.task->context = check;
Simon Hormanb1900d52015-01-30 11:22:54 +09003149 return NULL;
3150}
3151
Simon Hormanbfb5d332015-01-30 11:22:55 +09003152void free_check(struct check *check)
3153{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003154 free(check->bi.area);
3155 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003156 if (check->cs) {
3157 free(check->cs->conn);
3158 check->cs->conn = NULL;
3159 cs_free(check->cs);
3160 check->cs = NULL;
3161 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003162}
3163
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003164void email_alert_free(struct email_alert *alert)
3165{
3166 struct tcpcheck_rule *rule, *back;
3167
3168 if (!alert)
3169 return;
3170
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003171 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3172 LIST_DEL(&rule->list);
3173 free(rule->comment);
3174 free(rule->string);
3175 if (rule->expect_regex)
3176 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003177 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003178 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003179 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003180}
3181
Olivier Houchard9f6af332018-05-25 14:04:04 +02003182static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003183{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003184 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003185 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003186 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003187
3188 q = container_of(check, typeof(*q), check);
3189
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003190 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003191 while (1) {
3192 if (!(check->state & CHK_ST_ENABLED)) {
3193 if (LIST_ISEMPTY(&q->email_alerts)) {
3194 /* All alerts processed, queue the task */
3195 t->expire = TICK_ETERNITY;
3196 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003197 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003198 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003199
3200 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003201 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003202 t->expire = now_ms;
3203 check->server = alert->srv;
3204 check->tcpcheck_rules = &alert->tcpcheck_rules;
PiBa-NL1714b9f2017-12-06 01:35:43 +01003205 check->status = HCHK_STATUS_UNKNOWN; // the UNKNOWN status is used to exit set_server_check_status(.) early
Christopher Faulet0108bb32017-10-20 21:34:32 +02003206 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003207 }
3208
Olivier Houchard9f6af332018-05-25 14:04:04 +02003209 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003210 if (check->state & CHK_ST_INPROGRESS)
3211 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003212
3213 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3214 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003215 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003216 check->server = NULL;
3217 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003218 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003219 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003220 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003221 return t;
3222}
3223
Christopher Faulet0108bb32017-10-20 21:34:32 +02003224/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3225 *
3226 * The function returns 1 in success case, otherwise, it returns 0 and err is
3227 * filled.
3228 */
3229int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003230{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003231 struct mailer *mailer;
3232 struct email_alertq *queues;
3233 const char *err_str;
3234 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003235
Christopher Faulet0108bb32017-10-20 21:34:32 +02003236 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3237 memprintf(err, "out of memory while allocating mailer alerts queues");
3238 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003239 }
3240
Christopher Faulet0108bb32017-10-20 21:34:32 +02003241 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3242 struct email_alertq *q = &queues[i];
3243 struct check *check = &q->check;
3244 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003245
3246 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003247 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003248 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003249 check->rise = DEF_AGENT_RISETIME;
3250 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003251 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3252 memprintf(err, "%s", err_str);
3253 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003254 }
3255
3256 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003257 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003258 check->port = get_host_port(&mailer->addr);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003259 //check->server = s;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003260
Emeric Brunc60def82017-09-27 14:59:38 +02003261 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003262 memprintf(err, "out of memory while allocating mailer alerts task");
3263 goto error;
3264 }
3265
3266 check->task = t;
3267 t->process = process_email_alert;
3268 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003269
Christopher Faulet0108bb32017-10-20 21:34:32 +02003270 /* check this in one ms */
3271 t->expire = TICK_ETERNITY;
3272 check->start = now;
3273 task_queue(t);
3274 }
3275
3276 mls->users++;
3277 free(p->email_alert.mailers.name);
3278 p->email_alert.mailers.m = mls;
3279 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003280 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003281
3282 error:
3283 for (i = 0; i < mls->count; i++) {
3284 struct email_alertq *q = &queues[i];
3285 struct check *check = &q->check;
3286
3287 if (check->task) {
3288 task_delete(check->task);
3289 task_free(check->task);
3290 check->task = NULL;
3291 }
3292 free_check(check);
3293 }
3294 free(queues);
3295 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003296}
3297
3298
3299static int add_tcpcheck_expect_str(struct list *list, const char *str)
3300{
3301 struct tcpcheck_rule *tcpcheck;
3302
Willy Tarreaubafbe012017-11-24 17:34:44 +01003303 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003304 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003305 memset(tcpcheck, 0, sizeof(*tcpcheck));
3306 tcpcheck->action = TCPCHK_ACT_EXPECT;
3307 tcpcheck->string = strdup(str);
3308 tcpcheck->expect_regex = NULL;
3309 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003310 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003311 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003312 return 0;
3313 }
3314
3315 LIST_ADDQ(list, &tcpcheck->list);
3316 return 1;
3317}
3318
3319static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3320{
3321 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003322 const char *in;
3323 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003324 int i;
3325
Willy Tarreaubafbe012017-11-24 17:34:44 +01003326 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003327 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003328 memset(tcpcheck, 0, sizeof(*tcpcheck));
3329 tcpcheck->action = TCPCHK_ACT_SEND;
3330 tcpcheck->expect_regex = NULL;
3331 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003332 tcpcheck->string_len = 0;
3333 for (i = 0; strs[i]; i++)
3334 tcpcheck->string_len += strlen(strs[i]);
3335
3336 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3337 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003338 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003339 return 0;
3340 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003341
Willy Tarreau64345aa2016-08-10 19:29:09 +02003342 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003343 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003344 for (in = strs[i]; (*dst = *in++); dst++);
3345 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003346
3347 LIST_ADDQ(list, &tcpcheck->list);
3348 return 1;
3349}
3350
Christopher Faulet0108bb32017-10-20 21:34:32 +02003351static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3352 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003353{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003354 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003355 struct tcpcheck_rule *tcpcheck;
3356 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003357
Willy Tarreaubafbe012017-11-24 17:34:44 +01003358 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003359 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003360 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003361 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003362 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003363
Willy Tarreaubafbe012017-11-24 17:34:44 +01003364 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003365 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003366 memset(tcpcheck, 0, sizeof(*tcpcheck));
3367 tcpcheck->action = TCPCHK_ACT_CONNECT;
3368 tcpcheck->comment = NULL;
3369 tcpcheck->string = NULL;
3370 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003371 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3372
3373 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3374 goto error;
3375
3376 {
3377 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3378 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3379 goto error;
3380 }
3381
3382 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3383 goto error;
3384
3385 {
3386 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3387 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3388 goto error;
3389 }
3390
3391 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3392 goto error;
3393
3394 {
3395 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3396 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3397 goto error;
3398 }
3399
3400 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3401 goto error;
3402
3403 {
3404 const char * const strs[2] = { "DATA\r\n" };
3405 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3406 goto error;
3407 }
3408
3409 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3410 goto error;
3411
3412 {
3413 struct tm tm;
3414 char datestr[48];
3415 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003416 "From: ", p->email_alert.from, "\r\n",
3417 "To: ", p->email_alert.to, "\r\n",
3418 "Date: ", datestr, "\r\n",
3419 "Subject: [HAproxy Alert] ", msg, "\r\n",
3420 "\r\n",
3421 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003422 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003423 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003424 NULL
3425 };
3426
3427 get_localtime(date.tv_sec, &tm);
3428
3429 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3430 goto error;
3431 }
3432
3433 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3434 goto error;
3435 }
3436
3437 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3438 goto error;
3439
3440 {
3441 const char * const strs[2] = { "QUIT\r\n" };
3442 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3443 goto error;
3444 }
3445
3446 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3447 goto error;
3448
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003449 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003450 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003451 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003452 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003453 return 1;
3454
3455error:
3456 email_alert_free(alert);
3457 return 0;
3458}
3459
Christopher Faulet0108bb32017-10-20 21:34:32 +02003460static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003461{
3462 int i;
3463 struct mailer *mailer;
3464
3465 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3466 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003467 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003468 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003469 return;
3470 }
3471 }
3472
3473 return;
3474}
3475
3476/*
3477 * Send email alert if configured.
3478 */
Simon Horman64e34162015-02-06 11:11:57 +09003479void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003480{
3481 va_list argp;
3482 char buf[1024];
3483 int len;
3484 struct proxy *p = s->proxy;
3485
Christopher Faulet0108bb32017-10-20 21:34:32 +02003486 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003487 return;
3488
3489 va_start(argp, format);
3490 len = vsnprintf(buf, sizeof(buf), format, argp);
3491 va_end(argp);
3492
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003493 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003494 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003495 return;
3496 }
3497
Christopher Faulet0108bb32017-10-20 21:34:32 +02003498 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003499}
3500
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003501/*
3502 * Return value:
3503 * the port to be used for the health check
3504 * 0 in case no port could be found for the check
3505 */
3506int srv_check_healthcheck_port(struct check *chk)
3507{
3508 int i = 0;
3509 struct server *srv = NULL;
3510
3511 srv = chk->server;
3512
3513 /* If neither a port nor an addr was specified and no check transport
3514 * layer is forced, then the transport layer used by the checks is the
3515 * same as for the production traffic. Otherwise we use raw_sock by
3516 * default, unless one is specified.
3517 */
3518 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003519 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003520 chk->send_proxy |= (srv->pp_opts);
3521 }
3522
3523 /* by default, we use the health check port ocnfigured */
3524 if (chk->port > 0)
3525 return chk->port;
3526
3527 /* try to get the port from check_core.addr if check.port not set */
3528 i = get_host_port(&chk->addr);
3529 if (i > 0)
3530 return i;
3531
3532 /* try to get the port from server address */
3533 /* prevent MAPPORTS from working at this point, since checks could
3534 * not be performed in such case (MAPPORTS impose a relative ports
3535 * based on live traffic)
3536 */
3537 if (srv->flags & SRV_F_MAPPORTS)
3538 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003539
3540 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003541 if (i > 0)
3542 return i;
3543
3544 return 0;
3545}
3546
Willy Tarreau865c5142016-12-21 20:04:48 +01003547__attribute__((constructor))
3548static void __check_init(void)
3549{
3550 hap_register_post_check(start_checks);
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003551
Willy Tarreaubafbe012017-11-24 17:34:44 +01003552 pool_head_email_alert = create_pool("email_alert", sizeof(struct email_alert), MEM_F_SHARED);
3553 pool_head_tcpcheck_rule = create_pool("tcpcheck_rule", sizeof(struct tcpcheck_rule), MEM_F_SHARED);
Willy Tarreau865c5142016-12-21 20:04:48 +01003554}
3555
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003556
Willy Tarreaubd741542010-03-16 18:46:54 +01003557/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003558 * Local variables:
3559 * c-indent-level: 8
3560 * c-basic-offset: 8
3561 * End:
3562 */