blob: fc1384580f71a0b8c05415c419f045ae95b85b55 [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 Tarreau3d300592007-03-18 18:34:41 +010052#include <proto/proto_http.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010053#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010054#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010055#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020057#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010058#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020059#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020060#include <proto/log.h>
61#include <proto/dns.h>
62#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020063
Olivier Houchard9130a962017-10-17 17:33:43 +020064#ifdef USE_OPENSSL
65#include <proto/ssl_sock.h>
66#endif /* USE_OPENSSL */
67
Willy Tarreaubd741542010-03-16 18:46:54 +010068static int httpchk_expect(struct server *s, int done);
Simon Hormane16c1b32015-01-30 11:22:57 +090069static int tcpcheck_get_step_id(struct check *);
Baptiste Assmann22b09d22015-05-01 08:03:04 +020070static char * tcpcheck_get_step_comment(struct check *, int);
Willy Tarreau6bdcab02017-10-04 18:41:00 +020071static int tcpcheck_main(struct check *);
Olivier Houchard910b2bc2018-07-17 18:49:38 +020072static void __event_srv_chk_w(struct conn_stream *cs);
Olivier Houchard4501c3e2018-08-28 19:36:18 +020073static int wake_srv_chk(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{
712 struct conn_stream *cs = ctx;
Olivier Houchard4501c3e2018-08-28 19:36:18 +0200713
714 if (!(cs->wait_list.wait_reason & SUB_CAN_SEND))
715 wake_srv_chk(cs);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200716 return NULL;
717}
718
719/* same as above but protected by the server lock.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100720 *
721 * Please do NOT place any return statement in this function and only leave
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200722 * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT
723 * TO USE event_srv_chk_w() instead.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200724 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200725static void __event_srv_chk_w(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200726{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200727 struct connection *conn = cs->conn;
728 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900729 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900730 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200731
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100732 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100733 goto out_wakeup;
734
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200735 if (conn->flags & CO_FL_HANDSHAKE) {
Olivier Houchard5c110b92018-08-14 17:04:58 +0200736 cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &cs->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200737 goto out;
738 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100739
Willy Tarreau20a18342013-12-05 00:31:46 +0100740 if (retrieve_errno_from_socket(conn)) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200741 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200742 __cs_stop_both(cs);
Willy Tarreau20a18342013-12-05 00:31:46 +0100743 goto out_wakeup;
744 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100745
Willy Tarreaubbae3f02017-08-30 09:59:52 +0200746 if (conn->flags & CO_FL_SOCK_WR_SH) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100747 /* if the output is closed, we can't do anything */
748 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200749 chk_report_conn_err(check, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100750 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200751 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200752
Willy Tarreau06559ac2013-12-05 01:53:08 +0100753 /* here, we know that the connection is established. That's enough for
754 * a pure TCP check.
755 */
756 if (!check->type)
757 goto out_wakeup;
758
Willy Tarreauc09572f2017-10-04 11:58:22 +0200759 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100760 if (check->type == PR_O2_TCPCHK_CHK)
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200761 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200762
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200763 if (b_data(&check->bo)) {
Olivier Houcharded0f2072018-08-16 15:41:52 +0200764 cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200765 b_realign_if_empty(&check->bo);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200766 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200767 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200768 __cs_stop_both(cs);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100769 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200770 }
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200771 if (b_data(&check->bo)) {
Olivier Houchard5c110b92018-08-14 17:04:58 +0200772 conn->mux->subscribe(cs, SUB_CAN_SEND, &cs->wait_list);
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200773 goto out;
774 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100775 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200776
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100777 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
778 if (s->proxy->timeout.check) {
779 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
780 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200781 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100782 goto out_nowake;
783
Willy Tarreau83749182007-04-15 20:56:27 +0200784 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200785 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200786 out_nowake:
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200787 __cs_stop_send(cs); /* nothing more to write */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200788 out:
789 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200790}
791
Willy Tarreaubaaee002006-06-26 02:48:02 +0200792/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200793 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200794 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900795 * set_server_check_status() to update check->status, check->duration
796 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200797
798 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
799 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
800 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
801 * response to an SSL HELLO (the principle is that this is enough to
802 * distinguish between an SSL server and a pure TCP relay). All other cases will
803 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
804 * etc.
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100805 *
806 * Please do NOT place any return statement in this function and only leave
807 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200808 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200809static void event_srv_chk_r(struct conn_stream *cs)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200810{
Olivier Houchard9aaf7782017-09-13 18:30:23 +0200811 struct connection *conn = cs->conn;
812 struct check *check = cs->data;
Simon Horman4a741432013-02-23 15:35:38 +0900813 struct server *s = check->server;
814 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200815 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100816 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200817 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200818
Christopher Faulet2a944ee2017-11-07 10:42:54 +0100819 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brun9f0b4582017-10-23 14:39:51 +0200820
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100821 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200822 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200823
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100824 if (conn->flags & CO_FL_HANDSHAKE)
825 goto out_unlock;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200826
Willy Tarreauc09572f2017-10-04 11:58:22 +0200827 /* wake() will take care of calling tcpcheck_main() */
Willy Tarreau62ac84f2017-11-05 10:11:13 +0100828 if (check->type == PR_O2_TCPCHK_CHK)
829 goto out_unlock;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200830
Willy Tarreau83749182007-04-15 20:56:27 +0200831 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
832 * but the connection was closed on the remote end. Fortunately, recv still
833 * works correctly and we don't need to do the getsockopt() on linux.
834 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000835
836 /* Set buffer to point to the end of the data already read, and check
837 * that there is free space remaining. If the buffer is full, proceed
838 * with running the checks without attempting another socket read.
839 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000840
Willy Tarreau03938182010-03-17 21:52:07 +0100841 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000842
Olivier Houchard511efea2018-08-16 15:30:32 +0200843 cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0);
Willy Tarreau4ff3b892017-10-16 15:17:17 +0200844 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreau03938182010-03-17 21:52:07 +0100845 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200846 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200847 /* Report network errors only if we got no other data. Otherwise
848 * we'll let the upper layers decide whether the response is OK
849 * or not. It is very common that an RST sent by the server is
850 * reported as an error just after the last data chunk.
851 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +0200852 chk_report_conn_err(check, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100853 goto out_wakeup;
854 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200855 }
856
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100857
Willy Tarreau03938182010-03-17 21:52:07 +0100858 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200859 * Terminate string in b_head(&check->bi) buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100860 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200861 if (b_data(&check->bi) < b_size(&check->bi))
862 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100863 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200864 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100865 done = 1; /* buffer full, don't wait for more data */
866 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200867
Nick Chalk57b1bf72010-03-16 15:50:46 +0000868 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900869 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200870 case PR_O2_HTTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200871 if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100872 goto wait_more_data;
873
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100874 /* Check if the server speaks HTTP 1.X */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200875 if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) ||
876 (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 ||
877 (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) ||
878 !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) ||
879 !isdigit((unsigned char) *(b_head(&check->bi) + 11))) {
880 cut_crlf(b_head(&check->bi));
881 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200882
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100883 goto out_wakeup;
884 }
885
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200886 check->code = str2uic(b_head(&check->bi) + 9);
887 desc = ltrim(b_head(&check->bi) + 12, ' ');
Christopher Fauletc2a89a62017-10-23 15:54:24 +0200888
Willy Tarreaubd741542010-03-16 18:46:54 +0100889 if ((s->proxy->options & PR_O_DISABLE404) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200890 (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000891 /* 404 may be accepted as "stopping" only if the server was up */
892 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900893 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000894 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100895 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
896 /* Run content verification check... We know we have at least 13 chars */
897 if (!httpchk_expect(s, done))
898 goto wait_more_data;
899 }
900 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200901 else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100902 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900903 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100904 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000905 else {
906 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900907 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000908 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200909 break;
910
911 case PR_O2_SSL3_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200912 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100913 goto wait_more_data;
914
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100915 /* Check for SSLv3 alert or handshake */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200916 if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16))
Simon Horman4a741432013-02-23 15:35:38 +0900917 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200918 else
Simon Horman4a741432013-02-23 15:35:38 +0900919 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200920 break;
921
922 case PR_O2_SMTP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200923 if (!done && b_data(&check->bi) < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100924 goto wait_more_data;
925
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200926 /* Check if the server speaks SMTP */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200927 if ((b_data(&check->bi) < strlen("000\r")) ||
928 (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') ||
929 !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) ||
930 !isdigit((unsigned char) *(b_head(&check->bi) + 2))) {
931 cut_crlf(b_head(&check->bi));
932 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200933 goto out_wakeup;
934 }
935
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200936 check->code = str2uic(b_head(&check->bi));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200937
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200938 desc = ltrim(b_head(&check->bi) + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200939 cut_crlf(desc);
940
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100941 /* Check for SMTP code 2xx (should be 250) */
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200942 if (*b_head(&check->bi) == '2')
Simon Horman4a741432013-02-23 15:35:38 +0900943 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200944 else
Simon Horman4a741432013-02-23 15:35:38 +0900945 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200946 break;
947
Simon Hormana2b9dad2013-02-12 10:45:54 +0900948 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100949 int status = HCHK_STATUS_CHECKED;
950 const char *hs = NULL; /* health status */
951 const char *as = NULL; /* admin status */
952 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200953 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100954 const char *err = NULL; /* first error to report */
955 const char *wrn = NULL; /* first warning to report */
956 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900957
Willy Tarreau81f5d942013-12-09 20:51:51 +0100958 /* We're getting an agent check response. The agent could
959 * have been disabled in the mean time with a long check
960 * still pending. It is important that we ignore the whole
961 * response.
962 */
963 if (!(check->server->agent.state & CHK_ST_ENABLED))
964 break;
965
966 /* The agent supports strings made of a single line ended by the
967 * first CR ('\r') or LF ('\n'). This line is composed of words
968 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
969 * line may optionally contained a description of a state change
970 * after a sharp ('#'), which is only considered if a health state
971 * is announced.
972 *
973 * Words may be composed of :
974 * - a numeric weight suffixed by the percent character ('%').
975 * - a health status among "up", "down", "stopped", and "fail".
976 * - an admin status among "ready", "drain", "maint".
977 *
978 * These words may appear in any order. If multiple words of the
979 * same category appear, the last one wins.
980 */
981
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200982 p = b_head(&check->bi);
Willy Tarreau9809b782013-12-11 21:40:11 +0100983 while (*p && *p != '\n' && *p != '\r')
984 p++;
985
986 if (!*p) {
987 if (!done)
988 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900989
Willy Tarreau9809b782013-12-11 21:40:11 +0100990 /* at least inform the admin that the agent is mis-behaving */
991 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
992 break;
993 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100994
Willy Tarreau9809b782013-12-11 21:40:11 +0100995 *p = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +0200996 cmd = b_head(&check->bi);
Simon Hormana2b9dad2013-02-12 10:45:54 +0900997
Willy Tarreau81f5d942013-12-09 20:51:51 +0100998 while (*cmd) {
999 /* look for next word */
1000 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
1001 cmd++;
1002 continue;
1003 }
Simon Horman671b6f02013-11-25 10:46:39 +09001004
Willy Tarreau81f5d942013-12-09 20:51:51 +01001005 if (*cmd == '#') {
1006 /* this is the beginning of a health status description,
1007 * skip the sharp and blanks.
1008 */
1009 cmd++;
1010 while (*cmd == '\t' || *cmd == ' ')
1011 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001012 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001013 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001014
1015 /* find the end of the word so that we have a null-terminated
1016 * word between <cmd> and <p>.
1017 */
1018 p = cmd + 1;
1019 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1020 p++;
1021 if (*p)
1022 *p++ = 0;
1023
1024 /* first, health statuses */
1025 if (strcasecmp(cmd, "up") == 0) {
1026 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001027 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001028 hs = cmd;
1029 }
1030 else if (strcasecmp(cmd, "down") == 0) {
1031 check->health = 0;
1032 status = HCHK_STATUS_L7STS;
1033 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001034 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001035 else if (strcasecmp(cmd, "stopped") == 0) {
1036 check->health = 0;
1037 status = HCHK_STATUS_L7STS;
1038 hs = cmd;
1039 }
1040 else if (strcasecmp(cmd, "fail") == 0) {
1041 check->health = 0;
1042 status = HCHK_STATUS_L7STS;
1043 hs = cmd;
1044 }
1045 /* admin statuses */
1046 else if (strcasecmp(cmd, "ready") == 0) {
1047 as = cmd;
1048 }
1049 else if (strcasecmp(cmd, "drain") == 0) {
1050 as = cmd;
1051 }
1052 else if (strcasecmp(cmd, "maint") == 0) {
1053 as = cmd;
1054 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001055 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001056 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1057 ps = cmd;
1058 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001059 /* try to parse a maxconn here */
1060 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1061 cs = cmd;
1062 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001063 else {
1064 /* keep a copy of the first error */
1065 if (!err)
1066 err = cmd;
1067 }
1068 /* skip to next word */
1069 cmd = p;
1070 }
1071 /* here, cmd points either to \0 or to the beginning of a
1072 * description. Skip possible leading spaces.
1073 */
1074 while (*cmd == ' ' || *cmd == '\n')
1075 cmd++;
1076
1077 /* First, update the admin status so that we avoid sending other
1078 * possibly useless warnings and can also update the health if
1079 * present after going back up.
1080 */
1081 if (as) {
1082 if (strcasecmp(as, "drain") == 0)
1083 srv_adm_set_drain(check->server);
1084 else if (strcasecmp(as, "maint") == 0)
1085 srv_adm_set_maint(check->server);
1086 else
1087 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001088 }
1089
Willy Tarreau81f5d942013-12-09 20:51:51 +01001090 /* now change weights */
1091 if (ps) {
1092 const char *msg;
1093
1094 msg = server_parse_weight_change_request(s, ps);
1095 if (!wrn || !*wrn)
1096 wrn = msg;
1097 }
1098
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001099 if (cs) {
1100 const char *msg;
1101
1102 cs += strlen("maxconn:");
1103
1104 msg = server_parse_maxconn_change_request(s, cs);
1105 if (!wrn || !*wrn)
1106 wrn = msg;
1107 }
1108
Willy Tarreau81f5d942013-12-09 20:51:51 +01001109 /* and finally health status */
1110 if (hs) {
1111 /* We'll report some of the warnings and errors we have
1112 * here. Down reports are critical, we leave them untouched.
1113 * Lack of report, or report of 'UP' leaves the room for
1114 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001115 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001116 const char *msg = cmd;
Willy Tarreau83061a82018-07-13 11:56:34 +02001117 struct buffer *t;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001118
1119 if (!*msg || status == HCHK_STATUS_L7OKD) {
1120 if (err && *err)
1121 msg = err;
1122 else if (wrn && *wrn)
1123 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001124 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001125
1126 t = get_trash_chunk();
1127 chunk_printf(t, "via agent : %s%s%s%s",
1128 hs, *msg ? " (" : "",
1129 msg, *msg ? ")" : "");
1130
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001131 set_server_check_status(check, status, t->area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001132 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001133 else if (err && *err) {
1134 /* No status change but we'd like to report something odd.
1135 * Just report the current state and copy the message.
1136 */
1137 chunk_printf(&trash, "agent reports an error : %s", err);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001138 set_server_check_status(check, status/*check->status*/,
1139 trash.area);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001140
Willy Tarreau81f5d942013-12-09 20:51:51 +01001141 }
1142 else if (wrn && *wrn) {
1143 /* No status change but we'd like to report something odd.
1144 * Just report the current state and copy the message.
1145 */
1146 chunk_printf(&trash, "agent warns : %s", wrn);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001147 set_server_check_status(check, status/*check->status*/,
1148 trash.area);
Willy Tarreau81f5d942013-12-09 20:51:51 +01001149 }
1150 else
1151 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001152 break;
1153 }
1154
Willy Tarreau1620ec32011-08-06 17:05:02 +02001155 case PR_O2_PGSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001156 if (!done && b_data(&check->bi) < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001157 goto wait_more_data;
1158
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001159 if (b_head(&check->bi)[0] == 'R') {
Simon Horman4a741432013-02-23 15:35:38 +09001160 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001161 }
1162 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001163 if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0))
1164 desc = &b_head(&check->bi)[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001165 else
1166 desc = "PostgreSQL unknown error";
1167
Simon Horman4a741432013-02-23 15:35:38 +09001168 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001169 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001170 break;
1171
1172 case PR_O2_REDIS_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001173 if (!done && b_data(&check->bi) < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001174 goto wait_more_data;
1175
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001176 if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001177 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001178 }
1179 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001180 set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi));
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001181 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001182 break;
1183
1184 case PR_O2_MYSQL_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001185 if (!done && b_data(&check->bi) < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001186 goto wait_more_data;
1187
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001188 if (s->proxy->check_len == 0) { // old mode
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001189 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001190 /* We set the MySQL Version in description for information purpose
1191 * FIXME : it can be cool to use MySQL Version for other purpose,
1192 * like mark as down old MySQL server.
1193 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001194 if (b_data(&check->bi) > 51) {
1195 desc = ltrim(b_head(&check->bi) + 5, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001196 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001197 }
1198 else {
1199 if (!done)
1200 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001201
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001202 /* it seems we have a OK packet but without a valid length,
1203 * it must be a protocol error
1204 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001205 set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi));
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001206 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001207 }
1208 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001209 /* An error message is attached in the Error packet */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001210 desc = ltrim(b_head(&check->bi) + 7, ' ');
Simon Horman4a741432013-02-23 15:35:38 +09001211 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001212 }
1213 } else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001214 unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) +
1215 (((unsigned int) *(b_head(&check->bi) + 1)) << 8) +
1216 (((unsigned int) *(b_head(&check->bi) + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001217
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001218 if (b_data(&check->bi) == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001219 /* MySQL Error packet always begin with field_count = 0xff */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001220 if (*(b_head(&check->bi) + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001221 /* We have only one MySQL packet and it is a Handshake Initialization packet
1222 * but we need to have a second packet to know if it is alright
1223 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001224 if (!done && b_data(&check->bi) < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001225 goto wait_more_data;
1226 }
1227 else {
1228 /* We have only one packet and it is an Error packet,
1229 * an error message is attached, so we can display it
1230 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001231 desc = &b_head(&check->bi)[7];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001232 //ha_warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001233 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001234 }
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001235 } else if (b_data(&check->bi) > first_packet_len + 4) {
1236 unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) +
1237 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) +
1238 (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001239
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001240 if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001241 /* We have 2 packets and that's good */
1242 /* Check if the second packet is a MySQL Error packet or not */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001243 if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001244 /* No error packet */
1245 /* We set the MySQL Version in description for information purpose */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001246 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001247 //ha_warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001248 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249 }
1250 else {
1251 /* An error message is attached in the Error packet
1252 * so we can display it ! :)
1253 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001254 desc = &b_head(&check->bi)[first_packet_len+11];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001255 //ha_warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001256 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001257 }
1258 }
1259 }
1260 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001261 if (!done)
1262 goto wait_more_data;
Emeric Brun9f0b4582017-10-23 14:39:51 +02001263
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001264 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001265 * it must be a protocol error
1266 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001267 desc = &b_head(&check->bi)[5];
Christopher Faulet767a84b2017-11-24 16:50:31 +01001268 //ha_warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001269 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001270 }
1271 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001272 break;
1273
1274 case PR_O2_LDAP_CHK:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001275 if (!done && b_data(&check->bi) < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001276 goto wait_more_data;
1277
1278 /* Check if the server speaks LDAP (ASN.1/BER)
1279 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1280 * http://tools.ietf.org/html/rfc4511
1281 */
1282
1283 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1284 * LDAPMessage: 0x30: SEQUENCE
1285 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001286 if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) {
Simon Horman4a741432013-02-23 15:35:38 +09001287 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001288 }
1289 else {
1290 /* size of LDAPMessage */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001291 msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001292
1293 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1294 * messageID: 0x02 0x01 0x01: INTEGER 1
1295 * protocolOp: 0x61: bindResponse
1296 */
1297 if ((msglen > 2) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001298 (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001299 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001300 goto out_wakeup;
1301 }
1302
1303 /* size of bindResponse */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001304 msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001305
1306 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1307 * ldapResult: 0x0a 0x01: ENUMERATION
1308 */
1309 if ((msglen > 4) ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001310 (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) {
Simon Horman4a741432013-02-23 15:35:38 +09001311 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001312 goto out_wakeup;
1313 }
1314
1315 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1316 * resultCode
1317 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001318 check->code = *(b_head(&check->bi) + msglen + 9);
Simon Horman4a741432013-02-23 15:35:38 +09001319 if (check->code) {
1320 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 +02001321 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001322 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001323 }
1324 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001325 break;
1326
Christopher Fauletba7bc162016-11-07 21:07:38 +01001327 case PR_O2_SPOP_CHK: {
1328 unsigned int framesz;
1329 char err[HCHK_DESC_LEN];
1330
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001331 if (!done && b_data(&check->bi) < 4)
Christopher Fauletba7bc162016-11-07 21:07:38 +01001332 goto wait_more_data;
1333
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001334 memcpy(&framesz, b_head(&check->bi), 4);
Christopher Fauletba7bc162016-11-07 21:07:38 +01001335 framesz = ntohl(framesz);
1336
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001337 if (!done && b_data(&check->bi) < (4+framesz))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001338 goto wait_more_data;
1339
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001340 if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1))
Christopher Fauletba7bc162016-11-07 21:07:38 +01001341 set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok");
1342 else
1343 set_server_check_status(check, HCHK_STATUS_L7STS, err);
1344 break;
1345 }
1346
Willy Tarreau1620ec32011-08-06 17:05:02 +02001347 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001348 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001349 break;
1350 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001351
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001352 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001353 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001354 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001355 chk_report_conn_err(check, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001356
Nick Chalk57b1bf72010-03-16 15:50:46 +00001357 /* Reset the check buffer... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001358 *b_head(&check->bi) = '\0';
1359 b_reset(&check->bi);
Nick Chalk57b1bf72010-03-16 15:50:46 +00001360
Steven Davidovitz544d4812017-03-08 11:06:20 -08001361 /* Close the connection... We still attempt to nicely close if,
1362 * for instance, SSL needs to send a "close notify." Later, we perform
1363 * a hard close and reset the connection if some data are pending,
1364 * otherwise we end up with many TIME_WAITs and eat all the source port
1365 * range quickly. To avoid sending RSTs all the time, we first try to
1366 * drain pending data.
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001367 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001368 __cs_stop_both(cs);
Willy Tarreauecdb3fe2017-10-05 15:25:48 +02001369 cs_shutw(cs, CS_SHW_NORMAL);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001370
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001371 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001372 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001373 conn->flags |= CO_FL_ERROR;
1374
Willy Tarreaufdccded2008-08-29 18:19:04 +02001375 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001376 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001377 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau3267d362012-08-17 23:53:56 +02001378 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001379
1380 wait_more_data:
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001381 __cs_want_recv(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001382 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001383}
1384
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001385/*
1386 * This function is used only for server health-checks. It handles connection
1387 * status updates including errors. If necessary, it wakes the check task up.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001388 * It returns 0 on normal cases, <0 if at least one close() has happened on the
1389 * connection (eg: reconnect).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001390 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001391static int wake_srv_chk(struct conn_stream *cs)
Willy Tarreau20bea422012-07-06 12:00:49 +02001392{
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001393 struct connection *conn = cs->conn;
1394 struct check *check = cs->data;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001395 int ret = 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001396
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001397 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001398
Willy Tarreauc09572f2017-10-04 11:58:22 +02001399 /* we may have to make progress on the TCP checks */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001400 if (check->type == PR_O2_TCPCHK_CHK) {
1401 ret = tcpcheck_main(check);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001402 cs = check->cs;
1403 conn = cs_conn(cs);
Olivier Houchard910b2bc2018-07-17 18:49:38 +02001404 } else if (LIST_ISEMPTY(&cs->wait_list.list))
1405 __event_srv_chk_w(cs);
Willy Tarreauc09572f2017-10-04 11:58:22 +02001406
Willy Tarreau4ff3b892017-10-16 15:17:17 +02001407 if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001408 /* We may get error reports bypassing the I/O handlers, typically
1409 * the case when sending a pure TCP check which fails, then the I/O
1410 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001411 * main processing task so let's simply wake it up. If we get here,
1412 * we expect errno to still be valid.
1413 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02001414 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001415 __cs_stop_both(cs);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001416 task_wakeup(check->task, TASK_WOKEN_IO);
1417 }
Willy Tarreau2500fc22018-04-03 19:31:38 +02001418 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 +01001419 /* we may get here if only a connection probe was required : we
1420 * don't have any data to send nor anything expected in response,
1421 * so the completion of the connection establishment is enough.
1422 */
1423 task_wakeup(check->task, TASK_WOKEN_IO);
1424 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001425
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001426 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001427 /* We're here because nobody wants to handle the error, so we
1428 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001429 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001430 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02001431 cs_close(cs);
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001432 ret = -1;
Willy Tarreau2d351b62013-12-05 02:36:25 +01001433 }
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001434
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001435 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Emeric Brunf6ba17d2017-11-02 14:35:27 +01001436
Willy Tarreau6bdcab02017-10-04 18:41:00 +02001437 /* if a connection got replaced, we must absolutely prevent the connection
1438 * handler from touching its fd, and perform the FD polling updates ourselves
1439 */
1440 if (ret < 0)
1441 conn_cond_update_polling(conn);
1442
1443 return ret;
Willy Tarreau20bea422012-07-06 12:00:49 +02001444}
1445
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001446struct data_cb check_conn_cb = {
1447 .recv = event_srv_chk_r,
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001448 .wake = wake_srv_chk,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +01001449 .name = "CHCK",
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001450};
1451
Willy Tarreaubaaee002006-06-26 02:48:02 +02001452/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001453 * updates the server's weight during a warmup stage. Once the final weight is
1454 * reached, the task automatically stops. Note that any server status change
1455 * must have updated s->last_change accordingly.
1456 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001457static struct task *server_warmup(struct task *t, void *context, unsigned short state)
Willy Tarreau2e993902011-10-31 11:53:20 +01001458{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001459 struct server *s = context;
Willy Tarreau2e993902011-10-31 11:53:20 +01001460
1461 /* by default, plan on stopping the task */
1462 t->expire = TICK_ETERNITY;
Emeric Brun52a91d32017-08-31 14:41:55 +02001463 if ((s->next_admin & SRV_ADMF_MAINT) ||
1464 (s->next_state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001465 return t;
1466
Willy Tarreau892337c2014-05-13 23:41:20 +02001467 /* recalculate the weights and update the state */
Willy Tarreau3ff577e2018-08-02 11:48:52 +02001468 server_recalc_eweight(s, 1);
Willy Tarreau2e993902011-10-31 11:53:20 +01001469
1470 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001471 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001472
1473 /* get back there in 1 second or 1/20th of the slowstart interval,
1474 * whichever is greater, resulting in small 5% steps.
1475 */
Emeric Brun52a91d32017-08-31 14:41:55 +02001476 if (s->next_state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001477 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1478 return t;
1479}
1480
Willy Tarreau894c6422017-10-04 15:58:52 +02001481/* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if
1482 * none was found.
1483 */
1484static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list)
1485{
1486 struct tcpcheck_rule *r;
1487
1488 list_for_each_entry(r, list, list) {
1489 if (r->action != TCPCHK_ACT_COMMENT)
1490 return r;
1491 }
1492 return NULL;
1493}
1494
Willy Tarreau2e993902011-10-31 11:53:20 +01001495/*
Simon Horman98637e52014-06-20 12:30:16 +09001496 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001497 *
1498 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001499 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1500 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1501 * - SF_ERR_SRVTO if there are no more servers
1502 * - SF_ERR_SRVCL if the connection was refused by the server
1503 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1504 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1505 * - SF_ERR_INTERNAL for any other purely internal errors
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001506 * - 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 +01001507 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001508 * Note that we try to prevent the network stack from sending the ACK during the
1509 * connect() when a pure TCP check is used (without PROXY protocol).
1510 */
Simon Horman98637e52014-06-20 12:30:16 +09001511static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001512{
1513 struct check *check = t->context;
1514 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001515 struct conn_stream *cs = check->cs;
1516 struct connection *conn = cs_conn(cs);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001517 struct protocol *proto;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001518 struct tcpcheck_rule *tcp_rule = NULL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001519 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001520 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001521
Willy Tarreau00149122017-10-04 18:05:01 +02001522 /* we cannot have a connection here */
1523 if (conn)
1524 return SF_ERR_INTERNAL;
1525
Simon Hormanb00d17a2014-06-13 16:18:16 +09001526 /* tcpcheck send/expect initialisation */
Willy Tarreauf411cce2017-10-04 16:21:19 +02001527 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001528 check->current_step = NULL;
Willy Tarreauf411cce2017-10-04 16:21:19 +02001529 tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules);
1530 }
Simon Hormanb00d17a2014-06-13 16:18:16 +09001531
1532 /* prepare the check buffer.
1533 * This should not be used if check is the secondary agent check
1534 * of a server as s->proxy->check_req will relate to the
1535 * configuration of the primary check. Similarly, tcp-check uses
1536 * its own strings.
1537 */
1538 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001539 b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001540
1541 /* we want to check if this host replies to HTTP or SSLv3 requests
1542 * so we'll send the request, and won't wake the checker up now.
1543 */
1544 if ((check->type) == PR_O2_SSL3_CHK) {
1545 /* SSL requires that we put Unix time in the request */
1546 int gmt_time = htonl(date.tv_sec);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001547 memcpy(b_head(&check->bo) + 11, &gmt_time, 4);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001548 }
1549 else if ((check->type) == PR_O2_HTTP_CHK) {
1550 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001551 b_putblk(&check->bo, trash.area,
1552 httpchk_build_status_header(s, trash.area, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001553 /* prevent HTTP keep-alive when "http-check expect" is used */
1554 if (s->proxy->options2 & PR_O2_EXP_TYPE)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001555 b_putist(&check->bo, ist("Connection: close\r\n"));
1556 b_putist(&check->bo, ist("\r\n"));
1557 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Simon Hormanb00d17a2014-06-13 16:18:16 +09001558 }
1559 }
1560
James Brown55f9ff12015-10-21 18:19:05 -07001561 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02001562 b_putblk(&check->bo, check->send_string, check->send_string_len);
James Brown55f9ff12015-10-21 18:19:05 -07001563 }
1564
Willy Tarreauf411cce2017-10-04 16:21:19 +02001565 /* for tcp-checks, the initial connection setup is handled separately as
1566 * it may be sent to a specific port and not to the server's.
1567 */
1568 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) {
1569 tcpcheck_main(check);
1570 return SF_ERR_UP;
1571 }
1572
Simon Hormanb00d17a2014-06-13 16:18:16 +09001573 /* prepare a new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001574 cs = check->cs = cs_new(NULL);
1575 if (!check->cs)
Willy Tarreau00149122017-10-04 18:05:01 +02001576 return SF_ERR_RESOURCE;
Olivier Houchard5c110b92018-08-14 17:04:58 +02001577 cs->wait_list.task->process = event_srv_chk_io;
1578 cs->wait_list.task->context = cs;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001579 conn = cs->conn;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001580
Simon Horman41f58762015-01-30 11:22:56 +09001581 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001582 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001583 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001584 }
1585 else {
1586 /* we'll connect to the addr on the server */
1587 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001588 }
1589
Olivier Houchard6377a002017-12-01 22:04:05 +01001590 proto = protocol_by_family(conn->addr.to.ss_family);
1591
1592 conn_prepare(conn, proto, check->xprt);
1593 conn_install_mux(conn, &mux_pt_ops, cs);
1594 cs_attach(cs, check, &check_conn_cb);
1595 conn->target = &s->obj_type;
1596
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001597 if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) {
1598 int i = 0;
1599
1600 i = srv_check_healthcheck_port(check);
Olivier Houchard6377a002017-12-01 22:04:05 +01001601 if (i == 0)
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001602 return SF_ERR_CHK_PORT;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02001603
1604 set_host_port(&conn->addr.to, i);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001605 }
1606
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001607 /* no client address */
1608 clear_addr(&conn->addr.from);
1609
Willy Tarreauf3d34822014-12-08 12:11:28 +01001610 /* only plain tcp-check supports quick ACK */
1611 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1612
Willy Tarreauf411cce2017-10-04 16:21:19 +02001613 if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT)
1614 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001615
Willy Tarreaue7dff022015-04-03 01:14:29 +02001616 ret = SF_ERR_INTERNAL;
Olivier Houchardb68fda42017-08-04 18:39:01 +02001617 if (proto && proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001618 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Willy Tarreau16257f62017-11-02 15:45:00 +01001619
1620 if (check->type)
1621 cs_want_send(cs);
1622
Olivier Houchard9130a962017-10-17 17:33:43 +02001623#ifdef USE_OPENSSL
1624 if (s->check.sni)
1625 ssl_sock_set_servername(conn, s->check.sni);
1626#endif
Willy Tarreauf4949772017-05-06 08:45:28 +02001627 if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001628 conn->send_proxy_ofs = 1;
1629 conn->flags |= CO_FL_SEND_PROXY;
1630 }
1631
1632 return ret;
1633}
1634
Simon Horman98637e52014-06-20 12:30:16 +09001635static struct list pid_list = LIST_HEAD_INIT(pid_list);
Willy Tarreaubafbe012017-11-24 17:34:44 +01001636static struct pool_head *pool_head_pid_list;
Christopher Faulet9dcf9b62017-11-13 10:34:01 +01001637__decl_hathreads(HA_SPINLOCK_T pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001638
1639void block_sigchld(void)
1640{
1641 sigset_t set;
1642 sigemptyset(&set);
1643 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001644 assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001645}
1646
1647void unblock_sigchld(void)
1648{
1649 sigset_t set;
1650 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001651 sigaddset(&set, SIGCHLD);
William Lallemand6e1796e2018-06-07 11:23:40 +02001652 assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001653}
1654
Simon Horman98637e52014-06-20 12:30:16 +09001655static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1656{
1657 struct pid_list *elem;
1658 struct check *check = t->context;
1659
Willy Tarreaubafbe012017-11-24 17:34:44 +01001660 elem = pool_alloc(pool_head_pid_list);
Simon Horman98637e52014-06-20 12:30:16 +09001661 if (!elem)
1662 return NULL;
1663 elem->pid = pid;
1664 elem->t = t;
1665 elem->exited = 0;
1666 check->curpid = elem;
1667 LIST_INIT(&elem->list);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001668
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001669 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001670 LIST_ADD(&pid_list, &elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001671 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001672
Simon Horman98637e52014-06-20 12:30:16 +09001673 return elem;
1674}
1675
Simon Horman98637e52014-06-20 12:30:16 +09001676static void pid_list_del(struct pid_list *elem)
1677{
1678 struct check *check;
1679
1680 if (!elem)
1681 return;
1682
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001683 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001684 LIST_DEL(&elem->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001685 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001686
Simon Horman98637e52014-06-20 12:30:16 +09001687 if (!elem->exited)
1688 kill(elem->pid, SIGTERM);
1689
1690 check = elem->t->context;
1691 check->curpid = NULL;
Willy Tarreaubafbe012017-11-24 17:34:44 +01001692 pool_free(pool_head_pid_list, elem);
Simon Horman98637e52014-06-20 12:30:16 +09001693}
1694
1695/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1696static void pid_list_expire(pid_t pid, int status)
1697{
1698 struct pid_list *elem;
1699
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001700 HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001701 list_for_each_entry(elem, &pid_list, list) {
1702 if (elem->pid == pid) {
1703 elem->t->expire = now_ms;
1704 elem->status = status;
1705 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001706 task_wakeup(elem->t, TASK_WOKEN_IO);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001707 break;
Simon Horman98637e52014-06-20 12:30:16 +09001708 }
1709 }
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001710 HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock);
Simon Horman98637e52014-06-20 12:30:16 +09001711}
1712
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001713static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001714{
1715 pid_t pid;
1716 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001717
Simon Horman98637e52014-06-20 12:30:16 +09001718 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1719 pid_list_expire(pid, status);
1720}
1721
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001722static int init_pid_list(void)
1723{
Willy Tarreaubafbe012017-11-24 17:34:44 +01001724 if (pool_head_pid_list != NULL)
Simon Horman98637e52014-06-20 12:30:16 +09001725 /* Nothing to do */
1726 return 0;
1727
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001728 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001729 ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1730 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001731 return 1;
1732 }
1733
Willy Tarreaubafbe012017-11-24 17:34:44 +01001734 pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1735 if (pool_head_pid_list == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001736 ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1737 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001738 return 1;
1739 }
1740
Christopher Faulet2a944ee2017-11-07 10:42:54 +01001741 HA_SPIN_INIT(&pid_list_lock);
Christopher Fauletcfda8472017-10-20 15:40:23 +02001742
Simon Horman98637e52014-06-20 12:30:16 +09001743 return 0;
1744}
1745
Cyril Bontéac92a062014-12-27 22:28:38 +01001746/* helper macro to set an environment variable and jump to a specific label on failure. */
1747#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001748
1749/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001750 * helper function to allocate enough memory to store an environment variable.
1751 * It will also check that the environment variable is updatable, and silently
1752 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001753 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001754static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001755{
1756 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001757 char *envname;
1758 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001759
Cyril Bontéac92a062014-12-27 22:28:38 +01001760 if (idx < 0 || idx >= EXTCHK_SIZE) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001761 ha_alert("Illegal environment variable index %d. Aborting.\n", idx);
Cyril Bontéac92a062014-12-27 22:28:38 +01001762 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001763 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001764
1765 envname = extcheck_envs[idx].name;
1766 vmaxlen = extcheck_envs[idx].vmaxlen;
1767
1768 /* Check if the environment variable is already set, and silently reject
1769 * the update if this one is not updatable. */
1770 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1771 return 0;
1772
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001773 /* Instead of sending NOT_USED, sending an empty value is preferable */
1774 if (strcmp(value, "NOT_USED") == 0) {
1775 value = "";
1776 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001777
1778 len = strlen(envname) + 1;
1779 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1780 len += strlen(value);
1781 else
1782 len += vmaxlen;
1783
1784 if (!check->envp[idx])
1785 check->envp[idx] = malloc(len + 1);
1786
1787 if (!check->envp[idx]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001788 ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001789 return 1;
1790 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001791 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001792 if (ret < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001793 ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001794 return 1;
1795 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001796 else if (ret > len) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001797 ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001798 return 1;
1799 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001800 return 0;
1801}
Simon Horman98637e52014-06-20 12:30:16 +09001802
1803static int prepare_external_check(struct check *check)
1804{
1805 struct server *s = check->server;
1806 struct proxy *px = s->proxy;
1807 struct listener *listener = NULL, *l;
1808 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001809 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001810 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001811
1812 list_for_each_entry(l, &px->conf.listeners, by_fe)
1813 /* Use the first INET, INET6 or UNIX listener */
1814 if (l->addr.ss_family == AF_INET ||
1815 l->addr.ss_family == AF_INET6 ||
1816 l->addr.ss_family == AF_UNIX) {
1817 listener = l;
1818 break;
1819 }
1820
Simon Horman98637e52014-06-20 12:30:16 +09001821 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001822 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1823 if (!check->envp) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001824 ha_alert("Failed to allocate memory for environment variables. Aborting\n");
Cyril Bontéac92a062014-12-27 22:28:38 +01001825 goto err;
1826 }
Simon Horman98637e52014-06-20 12:30:16 +09001827
Cyril Bontéac92a062014-12-27 22:28:38 +01001828 check->argv = calloc(6, sizeof(char *));
1829 if (!check->argv) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001830 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001831 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001832 }
Simon Horman98637e52014-06-20 12:30:16 +09001833
1834 check->argv[0] = px->check_command;
1835
Cyril Bonté777be862014-12-02 21:21:35 +01001836 if (!listener) {
1837 check->argv[1] = strdup("NOT_USED");
1838 check->argv[2] = strdup("NOT_USED");
1839 }
1840 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001841 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001842 addr_to_str(&listener->addr, buf, sizeof(buf));
1843 check->argv[1] = strdup(buf);
1844 port_to_str(&listener->addr, buf, sizeof(buf));
1845 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001846 }
1847 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001848 const struct sockaddr_un *un;
1849
1850 un = (struct sockaddr_un *)&listener->addr;
1851 check->argv[1] = strdup(un->sun_path);
1852 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001853 }
1854 else {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001855 ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001856 goto err;
1857 }
1858
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001859 addr_to_str(&s->addr, buf, sizeof(buf));
1860 check->argv[3] = strdup(buf);
Willy Tarreau04276f32017-01-06 17:41:29 +01001861
1862 if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6)
1863 snprintf(buf, sizeof(buf), "%u", s->svc_port);
1864 else
1865 *buf = 0;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001866 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001867
Cyril Bontéac92a062014-12-27 22:28:38 +01001868 for (i = 0; i < 5; i++) {
1869 if (!check->argv[i]) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001870 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001871 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001872 }
1873 }
Simon Horman98637e52014-06-20 12:30:16 +09001874
Cyril Bontéac92a062014-12-27 22:28:38 +01001875 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001876 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001877 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1878 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1879 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1880 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001881 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001882 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1883 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1884 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1885 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1886 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1887 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1888
1889 /* Ensure that we don't leave any hole in check->envp */
1890 for (i = 0; i < EXTCHK_SIZE; i++)
1891 if (!check->envp[i])
1892 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001893
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001894 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001895err:
1896 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001897 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001898 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001899 free(check->envp);
1900 check->envp = NULL;
1901 }
1902
1903 if (check->argv) {
1904 for (i = 1; i < 5; i++)
1905 free(check->argv[i]);
1906 free(check->argv);
1907 check->argv = NULL;
1908 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001909 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001910}
1911
Simon Hormanb00d17a2014-06-13 16:18:16 +09001912/*
Simon Horman98637e52014-06-20 12:30:16 +09001913 * establish a server health-check that makes use of a process.
1914 *
1915 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001916 * - SF_ERR_NONE if everything's OK
Willy Tarreaue7dff022015-04-03 01:14:29 +02001917 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
Tim Düsterhus4896c442016-11-29 02:15:19 +01001918 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001919 *
1920 * Blocks and then unblocks SIGCHLD
1921 */
1922static int connect_proc_chk(struct task *t)
1923{
Cyril Bontéac92a062014-12-27 22:28:38 +01001924 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001925 struct check *check = t->context;
1926 struct server *s = check->server;
1927 struct proxy *px = s->proxy;
1928 int status;
1929 pid_t pid;
1930
Willy Tarreaue7dff022015-04-03 01:14:29 +02001931 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001932
1933 block_sigchld();
1934
1935 pid = fork();
1936 if (pid < 0) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001937 ha_alert("Failed to fork process for external health check: %s. Aborting.\n",
1938 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001939 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1940 goto out;
1941 }
1942 if (pid == 0) {
1943 /* Child */
1944 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001945 int fd;
1946
1947 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1948 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1949
1950 while (fd < global.rlimit_nofile)
1951 close(fd++);
1952
Simon Horman98637e52014-06-20 12:30:16 +09001953 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001954 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001955 execvp(px->check_command, check->argv);
Christopher Faulet767a84b2017-11-24 16:50:31 +01001956 ha_alert("Failed to exec process for external health check: %s. Aborting.\n",
1957 strerror(errno));
Simon Horman98637e52014-06-20 12:30:16 +09001958 exit(-1);
1959 }
1960
1961 /* Parent */
1962 if (check->result == CHK_RES_UNKNOWN) {
1963 if (pid_list_add(pid, t) != NULL) {
1964 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1965
1966 if (px->timeout.check && px->timeout.connect) {
1967 int t_con = tick_add(now_ms, px->timeout.connect);
1968 t->expire = tick_first(t->expire, t_con);
1969 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001970 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001971 goto out;
1972 }
1973 else {
1974 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1975 }
1976 kill(pid, SIGTERM); /* process creation error */
1977 }
1978 else
1979 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1980
1981out:
1982 unblock_sigchld();
1983 return status;
1984}
1985
1986/*
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02001987 * manages a server health-check that uses an external process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001988 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01001989 *
1990 * Please do NOT place any return statement in this function and only leave
1991 * via the out_unlock label.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001992 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02001993static struct task *process_chk_proc(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09001994{
Olivier Houchard9f6af332018-05-25 14:04:04 +02001995 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09001996 struct server *s = check->server;
Simon Horman98637e52014-06-20 12:30:16 +09001997 int rv;
1998 int ret;
1999 int expired = tick_is_expired(t->expire, now_ms);
2000
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002001 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002002 if (!(check->state & CHK_ST_INPROGRESS)) {
2003 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002004 if (!expired) /* woke up too early */
2005 goto out_unlock;
Simon Horman98637e52014-06-20 12:30:16 +09002006
2007 /* we don't send any health-checks when the proxy is
2008 * stopped, the server should not be checked or the check
2009 * is disabled.
2010 */
2011 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
2012 s->proxy->state == PR_STSTOPPED)
2013 goto reschedule;
2014
2015 /* we'll initiate a new check */
2016 set_server_check_status(check, HCHK_STATUS_START, NULL);
2017
2018 check->state |= CHK_ST_INPROGRESS;
2019
Simon Hormandbf70192015-01-30 11:22:53 +09002020 ret = connect_proc_chk(t);
Willy Tarreaud7c3fbd2017-10-04 15:19:26 +02002021 if (ret == SF_ERR_NONE) {
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002022 /* the process was forked, we allow up to min(inter,
2023 * timeout.connect) for it to report its status, but
2024 * only when timeout.check is set as it may be to short
2025 * for a full check otherwise.
Simon Horman98637e52014-06-20 12:30:16 +09002026 */
2027 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2028
2029 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2030 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2031 t->expire = tick_first(t->expire, t_con);
2032 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002033 task_set_affinity(t, tid_bit);
Simon Horman98637e52014-06-20 12:30:16 +09002034 goto reschedule;
Simon Horman98637e52014-06-20 12:30:16 +09002035 }
2036
Willy Tarreau1e62e2a2017-10-04 15:07:02 +02002037 /* here, we failed to start the check */
Simon Horman98637e52014-06-20 12:30:16 +09002038
2039 check->state &= ~CHK_ST_INPROGRESS;
2040 check_notify_failure(check);
2041
2042 /* we allow up to min(inter, timeout.connect) for a connection
2043 * to establish but only when timeout.check is set
2044 * as it may be to short for a full check otherwise
2045 */
2046 while (tick_is_expired(t->expire, now_ms)) {
2047 int t_con;
2048
2049 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2050 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2051
2052 if (s->proxy->timeout.check)
2053 t->expire = tick_first(t->expire, t_con);
2054 }
2055 }
2056 else {
2057 /* there was a test running.
2058 * First, let's check whether there was an uncaught error,
2059 * which can happen on connect timeout or error.
2060 */
2061 if (check->result == CHK_RES_UNKNOWN) {
2062 /* good connection is enough for pure TCP check */
2063 struct pid_list *elem = check->curpid;
2064 int status = HCHK_STATUS_UNKNOWN;
2065
2066 if (elem->exited) {
2067 status = elem->status; /* Save in case the process exits between use below */
2068 if (!WIFEXITED(status))
2069 check->code = -1;
2070 else
2071 check->code = WEXITSTATUS(status);
2072 if (!WIFEXITED(status) || WEXITSTATUS(status))
2073 status = HCHK_STATUS_PROCERR;
2074 else
2075 status = HCHK_STATUS_PROCOK;
2076 } else if (expired) {
2077 status = HCHK_STATUS_PROCTOUT;
Christopher Faulet767a84b2017-11-24 16:50:31 +01002078 ha_warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09002079 kill(elem->pid, SIGTERM);
2080 }
2081 set_server_check_status(check, status, NULL);
2082 }
2083
2084 if (check->result == CHK_RES_FAILED) {
2085 /* a failure or timeout detected */
2086 check_notify_failure(check);
2087 }
2088 else if (check->result == CHK_RES_CONDPASS) {
2089 /* check is OK but asks for stopping mode */
2090 check_notify_stopping(check);
2091 }
2092 else if (check->result == CHK_RES_PASSED) {
2093 /* a success was detected */
2094 check_notify_success(check);
2095 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002096 task_set_affinity(t, MAX_THREADS_MASK);
Simon Horman98637e52014-06-20 12:30:16 +09002097 check->state &= ~CHK_ST_INPROGRESS;
2098
2099 pid_list_del(check->curpid);
2100
2101 rv = 0;
2102 if (global.spread_checks > 0) {
2103 rv = srv_getinter(check) * global.spread_checks / 100;
2104 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2105 }
2106 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2107 }
2108
2109 reschedule:
2110 while (tick_is_expired(t->expire, now_ms))
2111 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002112
2113 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002114 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Simon Horman98637e52014-06-20 12:30:16 +09002115 return t;
2116}
2117
2118/*
2119 * manages a server health-check that uses a connection. Returns
2120 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002121 *
2122 * Please do NOT place any return statement in this function and only leave
2123 * via the out_unlock label.
Simon Horman98637e52014-06-20 12:30:16 +09002124 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002125static struct task *process_chk_conn(struct task *t, void *context, unsigned short state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002126{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002127 struct check *check = context;
Simon Horman4a741432013-02-23 15:35:38 +09002128 struct server *s = check->server;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002129 struct conn_stream *cs = check->cs;
2130 struct connection *conn = cs_conn(cs);
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002131 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002132 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002133 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002134
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002135 HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002136 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002137 /* no check currently running */
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002138 if (!expired) /* woke up too early */
2139 goto out_unlock;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002140
Simon Horman671b6f02013-11-25 10:46:39 +09002141 /* we don't send any health-checks when the proxy is
2142 * stopped, the server should not be checked or the check
2143 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002144 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002145 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002146 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002147 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002148
2149 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002150 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002151
Willy Tarreau2c115e52013-12-11 19:41:16 +01002152 check->state |= CHK_ST_INPROGRESS;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002153 b_reset(&check->bi);
2154 b_reset(&check->bo);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002155
Simon Hormandbf70192015-01-30 11:22:53 +09002156 ret = connect_conn_chk(t);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002157 cs = check->cs;
2158 conn = cs_conn(cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002159
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002160 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002161 case SF_ERR_UP:
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002162 goto out_unlock;
2163
Willy Tarreaue7dff022015-04-03 01:14:29 +02002164 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002165 /* we allow up to min(inter, timeout.connect) for a connection
2166 * to establish but only when timeout.check is set
2167 * as it may be to short for a full check otherwise
2168 */
Simon Horman4a741432013-02-23 15:35:38 +09002169 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002170
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002171 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2172 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2173 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002174 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002175
2176 if (check->type)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002177 cs_want_recv(cs); /* prepare for reading a possible reply */
Willy Tarreau06559ac2013-12-05 01:53:08 +01002178
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002179 task_set_affinity(t, tid_bit);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002180 goto reschedule;
2181
Willy Tarreaue7dff022015-04-03 01:14:29 +02002182 case SF_ERR_SRVTO: /* ETIMEDOUT */
2183 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau00149122017-10-04 18:05:01 +02002184 if (conn)
2185 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002186 chk_report_conn_err(check, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002187 break;
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02002188 /* should share same code than cases below */
2189 case SF_ERR_CHK_PORT:
2190 check->state |= CHK_ST_PORT_MISS;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002191 case SF_ERR_PRXCOND:
2192 case SF_ERR_RESOURCE:
2193 case SF_ERR_INTERNAL:
Willy Tarreau00149122017-10-04 18:05:01 +02002194 if (conn)
2195 conn->flags |= CO_FL_ERROR;
2196 chk_report_conn_err(check, conn ? 0 : ENOMEM, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002197 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002198 }
2199
Willy Tarreau5a78f362012-11-23 12:47:05 +01002200 /* here, we have seen a synchronous error, no fd was allocated */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002201 if (cs) {
2202 cs_destroy(cs);
2203 cs = check->cs = NULL;
2204 conn = NULL;
Olivier Houchard390485a2017-10-24 19:03:30 +02002205 }
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002206
Willy Tarreau2c115e52013-12-11 19:41:16 +01002207 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002208 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002209
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002210 /* we allow up to min(inter, timeout.connect) for a connection
2211 * to establish but only when timeout.check is set
2212 * as it may be to short for a full check otherwise
2213 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002214 while (tick_is_expired(t->expire, now_ms)) {
2215 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002216
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002217 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002218 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002219
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002220 if (s->proxy->timeout.check)
2221 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002222 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002223 }
2224 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002225 /* there was a test running.
2226 * First, let's check whether there was an uncaught error,
2227 * which can happen on connect timeout or error.
2228 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002229 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002230 /* good connection is enough for pure TCP check */
2231 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002232 if (check->use_ssl)
2233 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002234 else
Simon Horman4a741432013-02-23 15:35:38 +09002235 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002236 }
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002237 else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002238 chk_report_conn_err(check, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002239 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002240 else
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002241 goto out_unlock; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002242 }
2243
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002244 /* check complete or aborted */
Willy Tarreau00149122017-10-04 18:05:01 +02002245 if (conn && conn->xprt) {
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002246 /* The check was aborted and the connection was not yet closed.
2247 * This can happen upon timeout, or when an external event such
2248 * as a failed response coupled with "observe layer7" caused the
2249 * server state to be suddenly changed.
2250 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002251 conn_sock_drain(conn);
Willy Tarreaua553ae92017-10-05 18:52:17 +02002252 cs_close(cs);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002253 }
2254
Willy Tarreauac59f362017-10-08 11:10:19 +02002255 if (cs) {
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002256 cs_destroy(cs);
2257 cs = check->cs = NULL;
2258 conn = NULL;
Willy Tarreau00149122017-10-04 18:05:01 +02002259 }
2260
Willy Tarreauaf549582014-05-16 17:37:50 +02002261 if (check->result == CHK_RES_FAILED) {
2262 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002263 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002264 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002265 else if (check->result == CHK_RES_CONDPASS) {
2266 /* check is OK but asks for stopping mode */
2267 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002268 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002269 else if (check->result == CHK_RES_PASSED) {
2270 /* a success was detected */
2271 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002272 }
Christopher Faulet88ce5d12017-10-20 15:41:18 +02002273 task_set_affinity(t, MAX_THREADS_MASK);
Willy Tarreau2c115e52013-12-11 19:41:16 +01002274 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002275
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002276 rv = 0;
2277 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002278 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002279 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002280 }
Simon Horman4a741432013-02-23 15:35:38 +09002281 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002282 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002283
2284 reschedule:
2285 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002286 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002287 out_unlock:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01002288 HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock);
Willy Tarreau26c25062009-03-08 09:38:41 +01002289 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002290}
2291
Simon Horman98637e52014-06-20 12:30:16 +09002292/*
2293 * manages a server health-check. Returns
2294 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2295 */
Olivier Houchard9f6af332018-05-25 14:04:04 +02002296static struct task *process_chk(struct task *t, void *context, unsigned short state)
Simon Horman98637e52014-06-20 12:30:16 +09002297{
Olivier Houchard9f6af332018-05-25 14:04:04 +02002298 struct check *check = context;
Simon Horman98637e52014-06-20 12:30:16 +09002299
2300 if (check->type == PR_O2_EXT_CHK)
Olivier Houchard9f6af332018-05-25 14:04:04 +02002301 return process_chk_proc(t, context, state);
2302 return process_chk_conn(t, context, state);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002303
Simon Horman98637e52014-06-20 12:30:16 +09002304}
2305
Simon Horman5c942422013-11-25 10:46:32 +09002306static int start_check_task(struct check *check, int mininter,
2307 int nbcheck, int srvpos)
2308{
2309 struct task *t;
2310 /* task for the check */
Emeric Brunc60def82017-09-27 14:59:38 +02002311 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002312 ha_alert("Starting [%s:%s] check: out of memory.\n",
2313 check->server->proxy->id, check->server->id);
Simon Horman5c942422013-11-25 10:46:32 +09002314 return 0;
2315 }
2316
2317 check->task = t;
2318 t->process = process_chk;
2319 t->context = check;
2320
Willy Tarreau1746eec2014-04-25 10:46:47 +02002321 if (mininter < srv_getinter(check))
2322 mininter = srv_getinter(check);
2323
2324 if (global.max_spread_checks && mininter > global.max_spread_checks)
2325 mininter = global.max_spread_checks;
2326
Simon Horman5c942422013-11-25 10:46:32 +09002327 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002328 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002329 check->start = now;
2330 task_queue(t);
2331
2332 return 1;
2333}
2334
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002335/*
2336 * Start health-check.
Willy Tarreau865c5142016-12-21 20:04:48 +01002337 * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case.
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002338 */
Willy Tarreau865c5142016-12-21 20:04:48 +01002339static int start_checks()
2340{
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002341
2342 struct proxy *px;
2343 struct server *s;
2344 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002345 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002346
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002347 /* 1- count the checkers to run simultaneously.
2348 * We also determine the minimum interval among all of those which
2349 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2350 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002351 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002352 * too short an interval for all others.
2353 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002354 for (px = proxies_list; px; px = px->next) {
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002355 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002356 if (s->slowstart) {
Emeric Brunc60def82017-09-27 14:59:38 +02002357 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002358 ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002359 return ERR_ALERT | ERR_FATAL;
Willy Tarreaue7b73482013-11-21 11:50:50 +01002360 }
2361 /* We need a warmup task that will be called when the server
2362 * state switches from down to up.
2363 */
2364 s->warmup = t;
2365 t->process = server_warmup;
2366 t->context = s;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002367 /* server can be in this state only because of */
Emeric Brun52a91d32017-08-31 14:41:55 +02002368 if (s->next_state == SRV_ST_STARTING)
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002369 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 +01002370 }
2371
Willy Tarreaud8514a22013-12-11 21:10:14 +01002372 if (s->check.state & CHK_ST_CONFIGURED) {
2373 nbcheck++;
2374 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2375 (!mininter || mininter > srv_getinter(&s->check)))
2376 mininter = srv_getinter(&s->check);
2377 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002378
Willy Tarreaud8514a22013-12-11 21:10:14 +01002379 if (s->agent.state & CHK_ST_CONFIGURED) {
2380 nbcheck++;
2381 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2382 (!mininter || mininter > srv_getinter(&s->agent)))
2383 mininter = srv_getinter(&s->agent);
2384 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002385 }
2386 }
2387
Simon Horman4a741432013-02-23 15:35:38 +09002388 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002389 return 0;
2390
2391 srand((unsigned)time(NULL));
2392
2393 /*
2394 * 2- start them as far as possible from each others. For this, we will
2395 * start them after their interval set to the min interval divided by
2396 * the number of servers, weighted by the server's position in the list.
2397 */
Olivier Houchardfbc74e82017-11-24 16:54:05 +01002398 for (px = proxies_list; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002399 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2400 if (init_pid_list()) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01002401 ha_alert("Starting [%s] check: out of memory.\n", px->id);
Willy Tarreau865c5142016-12-21 20:04:48 +01002402 return ERR_ALERT | ERR_FATAL;
Simon Horman98637e52014-06-20 12:30:16 +09002403 }
2404 }
2405
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002406 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002407 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002408 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002409 if (s->check.type == PR_O2_EXT_CHK) {
2410 if (!prepare_external_check(&s->check))
Willy Tarreau865c5142016-12-21 20:04:48 +01002411 return ERR_ALERT | ERR_FATAL;
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002412 }
Simon Hormand60d6912013-11-25 10:46:36 +09002413 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
Willy Tarreau865c5142016-12-21 20:04:48 +01002414 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002415 srvpos++;
2416 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002417
Simon Hormand60d6912013-11-25 10:46:36 +09002418 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002419 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002420 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
Willy Tarreau865c5142016-12-21 20:04:48 +01002421 return ERR_ALERT | ERR_FATAL;
Simon Hormand60d6912013-11-25 10:46:36 +09002422 }
2423 srvpos++;
2424 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002425 }
2426 }
2427 return 0;
2428}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002429
2430/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002431 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002432 * The buffer MUST be terminated by a null byte before calling this function.
2433 * Sets server status appropriately. The caller is responsible for ensuring
2434 * that the buffer contains at least 13 characters. If <done> is zero, we may
2435 * return 0 to indicate that data is required to decide of a match.
2436 */
2437static int httpchk_expect(struct server *s, int done)
2438{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002439 static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>";
Willy Tarreaubd741542010-03-16 18:46:54 +01002440 char status_code[] = "000";
2441 char *contentptr;
2442 int crlf;
2443 int ret;
2444
2445 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2446 case PR_O2_EXP_STS:
2447 case PR_O2_EXP_RSTS:
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002448 memcpy(status_code, b_head(&s->check.bi) + 9, 3);
2449 memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002450
2451 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2452 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2453 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002454 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002455
2456 /* we necessarily have the response, so there are no partial failures */
2457 if (s->proxy->options2 & PR_O2_EXP_INV)
2458 ret = !ret;
2459
Simon Horman4a741432013-02-23 15:35:38 +09002460 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002461 break;
2462
2463 case PR_O2_EXP_STR:
2464 case PR_O2_EXP_RSTR:
2465 /* very simple response parser: ignore CR and only count consecutive LFs,
2466 * stop with contentptr pointing to first char after the double CRLF or
2467 * to '\0' if crlf < 2.
2468 */
2469 crlf = 0;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002470 for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002471 if (crlf >= 2)
2472 break;
2473 if (*contentptr == '\r')
2474 continue;
2475 else if (*contentptr == '\n')
2476 crlf++;
2477 else
2478 crlf = 0;
2479 }
2480
2481 /* Check that response contains a body... */
2482 if (crlf < 2) {
2483 if (!done)
2484 return 0;
2485
Simon Horman4a741432013-02-23 15:35:38 +09002486 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002487 "HTTP content check could not find a response body");
2488 return 1;
2489 }
2490
2491 /* Check that response body is not empty... */
2492 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002493 if (!done)
2494 return 0;
2495
Simon Horman4a741432013-02-23 15:35:38 +09002496 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002497 "HTTP content check found empty response body");
2498 return 1;
2499 }
2500
2501 /* Check the response content against the supplied string
2502 * or regex... */
2503 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2504 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2505 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002506 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002507
2508 /* if we don't match, we may need to wait more */
2509 if (!ret && !done)
2510 return 0;
2511
2512 if (ret) {
2513 /* content matched */
2514 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002515 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002516 "HTTP check matched unwanted content");
2517 else
Simon Horman4a741432013-02-23 15:35:38 +09002518 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002519 "HTTP content check matched");
2520 }
2521 else {
2522 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002523 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002524 "HTTP check did not match unwanted content");
2525 else
Simon Horman4a741432013-02-23 15:35:38 +09002526 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002527 "HTTP content check did not match");
2528 }
2529 break;
2530 }
2531 return 1;
2532}
2533
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002534/*
2535 * return the id of a step in a send/expect session
2536 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002537static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002538{
2539 struct tcpcheck_rule *cur = NULL, *next = NULL;
2540 int i = 0;
2541
Willy Tarreau213c6782014-10-02 14:51:02 +02002542 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002543 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002544 return 0;
2545
Simon Hormane16c1b32015-01-30 11:22:57 +09002546 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002547
2548 /* no step => first step */
2549 if (cur == NULL)
2550 return 1;
2551
2552 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002553 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002554 if (next->list.p == &cur->list)
2555 break;
2556 ++i;
2557 }
2558
2559 return i;
2560}
2561
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002562/*
2563 * return the latest known comment before (including) the given stepid
2564 * returns NULL if no comment found
2565 */
2566static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2567{
2568 struct tcpcheck_rule *cur = NULL;
2569 char *ret = NULL;
2570 int i = 0;
2571
2572 /* not even started anything yet, return latest comment found before any action */
2573 if (!check->current_step) {
2574 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2575 if (cur->action == TCPCHK_ACT_COMMENT)
2576 ret = cur->comment;
2577 else
2578 goto return_comment;
2579 }
2580 }
2581
2582 i = 1;
2583 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2584 if (cur->comment)
2585 ret = cur->comment;
2586
2587 if (i >= stepid)
2588 goto return_comment;
2589
2590 ++i;
2591 }
2592
2593 return_comment:
2594 return ret;
2595}
2596
Willy Tarreaube74b882017-10-04 16:22:49 +02002597/* proceed with next steps for the TCP checks <check>. Note that this is called
2598 * both from the connection's wake() callback and from the check scheduling task.
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002599 * It returns 0 on normal cases, or <0 if a close() has happened on an existing
2600 * connection, presenting the risk of an fd replacement.
Willy Tarreau62ac84f2017-11-05 10:11:13 +01002601 *
2602 * Please do NOT place any return statement in this function and only leave
2603 * via the out_unlock label after setting retcode.
Willy Tarreaube74b882017-10-04 16:22:49 +02002604 */
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002605static int tcpcheck_main(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002606{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002607 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002608 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002609 int done = 0, ret = 0, step = 0;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002610 struct conn_stream *cs = check->cs;
2611 struct connection *conn = cs_conn(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002612 struct server *s = check->server;
2613 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002614 struct list *head = check->tcpcheck_rules;
Willy Tarreau6bdcab02017-10-04 18:41:00 +02002615 int retcode = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002616
Willy Tarreauef953952014-10-02 14:30:14 +02002617 /* here, we know that the check is complete or that it failed */
2618 if (check->result != CHK_RES_UNKNOWN)
2619 goto out_end_tcpcheck;
2620
2621 /* We have 4 possibilities here :
2622 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
Willy Tarreau00149122017-10-04 18:05:01 +02002623 * connection attempt was made yet ; conn==NULL;current_step==NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002624 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2625 * does not exist (no rule), so a connection attempt was made
Willy Tarreau00149122017-10-04 18:05:01 +02002626 * before coming here, conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002627 * 3. we're coming back after having started with step 1, so we may
Willy Tarreau00149122017-10-04 18:05:01 +02002628 * be waiting for a connection attempt to complete. conn!=NULL.
2629 * 4. the connection + handshake are complete. conn!=NULL.
Willy Tarreauef953952014-10-02 14:30:14 +02002630 *
2631 * #2 and #3 are quite similar, we want both the connection and the
2632 * handshake to complete before going any further. Thus we must always
2633 * wait for a connection to complete unless we're before and existing
2634 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002635 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002636
2637 /* find first rule and skip comments */
2638 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2639 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2640 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2641
Willy Tarreau00149122017-10-04 18:05:01 +02002642 if ((check->current_step || &next->list == head) &&
2643 (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002644 /* we allow up to min(inter, timeout.connect) for a connection
2645 * to establish but only when timeout.check is set
2646 * as it may be to short for a full check otherwise
2647 */
2648 while (tick_is_expired(t->expire, now_ms)) {
2649 int t_con;
2650
2651 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2652 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2653
2654 if (s->proxy->timeout.check)
2655 t->expire = tick_first(t->expire, t_con);
2656 }
Christopher Fauletb6102852017-11-28 10:06:29 +01002657 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002658 }
2659
Willy Tarreauef953952014-10-02 14:30:14 +02002660 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002661 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002662 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002663 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002664 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002665
Willy Tarreau213c6782014-10-02 14:51:02 +02002666 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002667 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002668 check->last_started_step = NULL;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002669 b_reset(&check->bo);
2670 b_reset(&check->bi);
Willy Tarreau449f9522015-05-13 15:39:48 +02002671 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002672 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2673 if (s->proxy->timeout.check)
2674 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2675 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002676
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002677 /* It's only the rules which will enable send/recv */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002678 if (cs)
2679 cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002680
Willy Tarreauabca5b62013-12-06 14:19:25 +01002681 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002682 /* We have to try to flush the output buffer before reading, at
2683 * the end, or if we're about to send a string that does not fit
2684 * in the remaining space. That explains why we break out of the
Willy Tarreau00149122017-10-04 18:05:01 +02002685 * loop after this control. If we have data, conn is valid.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002686 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002687 if (b_data(&check->bo) &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002688 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002689 check->current_step->action != TCPCHK_ACT_SEND ||
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002690 check->current_step->string_len >= b_room(&check->bo))) {
Willy Tarreaudeccd112018-06-14 18:38:55 +02002691 int ret;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002692
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002693 __cs_want_send(cs);
Olivier Houcharded0f2072018-08-16 15:41:52 +02002694 ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0);
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002695 b_realign_if_empty(&check->bo);
Willy Tarreaudeccd112018-06-14 18:38:55 +02002696
2697 if (ret <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002698 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) {
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002699 chk_report_conn_err(check, errno, 0);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002700 __cs_stop_both(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002701 goto out_end_tcpcheck;
2702 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002703 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002704 }
Olivier Houchard4501c3e2018-08-28 19:36:18 +02002705 if (b_data(&check->bo)) {
2706 cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &cs->wait_list);
2707 goto out;
2708 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002709 }
2710
Willy Tarreau263013d2015-05-13 11:59:14 +02002711 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002712 break;
2713
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002714 /* have 'next' point to the next rule or NULL if we're on the
2715 * last one, connect() needs this.
2716 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002717 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002718
2719 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002720 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002721 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002722
2723 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002724 if (&next->list == head)
2725 next = NULL;
2726
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002727 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2728 struct protocol *proto;
2729 struct xprt_ops *xprt;
2730
Willy Tarreau00149122017-10-04 18:05:01 +02002731 /* For a connect action we'll create a new connection.
2732 * We may also have to kill a previous one. But we don't
2733 * want to leave *without* a connection if we came here
2734 * from the connection layer, hence with a connection.
2735 * Thus we'll proceed in the following order :
2736 * 1: close but not release previous connection
2737 * 2: try to get a new connection
2738 * 3: release and replace the old one on success
2739 */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002740 if (check->cs) {
Willy Tarreaua553ae92017-10-05 18:52:17 +02002741 cs_close(check->cs);
2742 retcode = -1; /* do not reuse the fd in the caller! */
Willy Tarreau00149122017-10-04 18:05:01 +02002743 }
2744
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002745 /* mark the step as started */
2746 check->last_started_step = check->current_step;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002747
2748 /* prepare new connection */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002749 cs = cs_new(NULL);
2750 if (!cs) {
Willy Tarreau00149122017-10-04 18:05:01 +02002751 step = tcpcheck_get_step_id(check);
2752 chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step);
2753 comment = tcpcheck_get_step_comment(check, step);
2754 if (comment)
2755 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002756 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2757 trash.area);
Willy Tarreau00149122017-10-04 18:05:01 +02002758 check->current_step = NULL;
Christopher Fauletb6102852017-11-28 10:06:29 +01002759 goto out;
Willy Tarreau00149122017-10-04 18:05:01 +02002760 }
Olivier Houchard5c110b92018-08-14 17:04:58 +02002761 cs->wait_list.task->process = event_srv_chk_io;
2762 cs->wait_list.task->context = cs;
Willy Tarreau00149122017-10-04 18:05:01 +02002763
Willy Tarreauac59f362017-10-08 11:10:19 +02002764 if (check->cs)
2765 cs_destroy(check->cs);
Willy Tarreau00149122017-10-04 18:05:01 +02002766
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002767 check->cs = cs;
2768 conn = cs->conn;
2769 cs_attach(cs, check, &check_conn_cb);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002770 conn->target = &s->obj_type;
2771
2772 /* no client address */
2773 clear_addr(&conn->addr.from);
2774
Simon Horman41f58762015-01-30 11:22:56 +09002775 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002776 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002777 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002778 }
2779 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002780 /* we'll connect to the addr on the server */
2781 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002782 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002783 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002784
2785 /* port */
2786 if (check->current_step->port)
2787 set_host_port(&conn->addr.to, check->current_step->port);
2788 else if (check->port)
2789 set_host_port(&conn->addr.to, check->port);
Baptiste Assmann2f3a56b2018-03-19 12:22:41 +01002790 else if (s->svc_port)
2791 set_host_port(&conn->addr.to, s->svc_port);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002792
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002793 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002794 xprt = xprt_get(XPRT_SSL);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002795 }
2796 else {
Willy Tarreaua261e9b2016-12-22 20:44:00 +01002797 xprt = xprt_get(XPRT_RAW);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002798 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002799 conn_prepare(conn, proto, xprt);
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002800 conn_install_mux(conn, &mux_pt_ops, cs);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002801
Willy Tarreaue7dff022015-04-03 01:14:29 +02002802 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002803 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002804 ret = proto->connect(conn,
2805 1 /* I/O polling is always needed */,
2806 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002807 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2808 conn->send_proxy_ofs = 1;
2809 conn->flags |= CO_FL_SEND_PROXY;
2810 }
2811
2812 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002813 * - SF_ERR_NONE if everything's OK
2814 * - SF_ERR_SRVTO if there are no more servers
2815 * - SF_ERR_SRVCL if the connection was refused by the server
2816 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2817 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2818 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +01002819 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002820 * Note that we try to prevent the network stack from sending the ACK during the
2821 * connect() when a pure TCP check is used (without PROXY protocol).
2822 */
2823 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002824 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002825 /* we allow up to min(inter, timeout.connect) for a connection
2826 * to establish but only when timeout.check is set
2827 * as it may be to short for a full check otherwise
2828 */
2829 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2830
2831 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2832 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2833 t->expire = tick_first(t->expire, t_con);
2834 }
2835 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002836 case SF_ERR_SRVTO: /* ETIMEDOUT */
2837 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002838 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002839 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002840 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002841 comment = tcpcheck_get_step_comment(check, step);
2842 if (comment)
2843 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002844 set_server_check_status(check, HCHK_STATUS_L4CON,
2845 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002846 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002847 case SF_ERR_PRXCOND:
2848 case SF_ERR_RESOURCE:
2849 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002850 step = tcpcheck_get_step_id(check);
2851 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002852 comment = tcpcheck_get_step_comment(check, step);
2853 if (comment)
2854 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002855 set_server_check_status(check, HCHK_STATUS_SOCKERR,
2856 trash.area);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002857 goto out_end_tcpcheck;
2858 }
2859
2860 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002861 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002862
2863 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002864 while (&check->current_step->list != head &&
2865 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002866 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002867
Willy Tarreauf2c87352015-05-13 12:08:21 +02002868 if (&check->current_step->list == head)
2869 break;
2870
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002871 } /* end 'connect' */
2872 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2873 /* mark the step as started */
2874 check->last_started_step = check->current_step;
2875
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002876 /* reset the read buffer */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002877 if (*b_head(&check->bi) != '\0') {
2878 *b_head(&check->bi) = '\0';
2879 b_reset(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002880 }
2881
Willy Tarreaubbae3f02017-08-30 09:59:52 +02002882 if (conn->flags & CO_FL_SOCK_WR_SH) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002883 conn->flags |= CO_FL_ERROR;
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002884 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002885 goto out_end_tcpcheck;
2886 }
2887
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002888 if (check->current_step->string_len >= b_size(&check->bo)) {
Willy Tarreau506a29a2018-07-18 10:07:58 +02002889 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d",
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002890 check->current_step->string_len, (unsigned int)b_size(&check->bo),
Simon Hormane16c1b32015-01-30 11:22:57 +09002891 tcpcheck_get_step_id(check));
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002892 set_server_check_status(check, HCHK_STATUS_L7RSP,
2893 trash.area);
Willy Tarreauabca5b62013-12-06 14:19:25 +01002894 goto out_end_tcpcheck;
2895 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002896
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002897 /* do not try to send if there is no space */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002898 if (check->current_step->string_len >= b_room(&check->bo))
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002899 continue;
2900
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002901 b_putblk(&check->bo, check->current_step->string, check->current_step->string_len);
2902 *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002903
Willy Tarreauabca5b62013-12-06 14:19:25 +01002904 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002905 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002906
2907 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002908 while (&check->current_step->list != head &&
2909 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002910 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002911
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002912 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002913 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002914 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002915 goto out_end_tcpcheck;
2916
Olivier Houchard9aaf7782017-09-13 18:30:23 +02002917 __cs_want_recv(cs);
Olivier Houchard511efea2018-08-16 15:30:32 +02002918 if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) {
Willy Tarreau4ff3b892017-10-16 15:17:17 +02002919 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002920 done = 1;
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002921 if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002922 /* Report network errors only if we got no other data. Otherwise
2923 * we'll let the upper layers decide whether the response is OK
2924 * or not. It is very common that an RST sent by the server is
2925 * reported as an error just after the last data chunk.
2926 */
Willy Tarreaub5259bf2017-10-04 14:47:29 +02002927 chk_report_conn_err(check, errno, 0);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002928 goto out_end_tcpcheck;
2929 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002930 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002931 else
Willy Tarreau263013d2015-05-13 11:59:14 +02002932 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002933 }
2934
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002935 /* mark the step as started */
2936 check->last_started_step = check->current_step;
2937
2938
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002939 /* Intermediate or complete response received.
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002940 * Terminate string in b_head(&check->bi) buffer.
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002941 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002942 if (b_data(&check->bi) < b_size(&check->bi)) {
2943 b_head(&check->bi)[b_data(&check->bi)] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002944 }
2945 else {
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002946 b_head(&check->bi)[b_data(&check->bi) - 1] = '\0';
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002947 done = 1; /* buffer full, don't wait for more data */
2948 }
2949
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002950 contentptr = b_head(&check->bi);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002951
2952 /* Check that response body is not empty... */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002953 if (!b_data(&check->bi)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002954 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002955 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002956
2957 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002958 step = tcpcheck_get_step_id(check);
2959 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002960 comment = tcpcheck_get_step_comment(check, step);
2961 if (comment)
2962 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002963 set_server_check_status(check, HCHK_STATUS_L7RSP,
2964 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002965
2966 goto out_end_tcpcheck;
2967 }
2968
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002969 if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002970 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002971
Willy Tarreaua970c282013-12-06 12:47:19 +01002972 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002973 if (check->current_step->string != NULL)
Willy Tarreauc9fa0482018-07-10 17:43:27 +02002974 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 +02002975 else if (check->current_step->expect_regex != NULL)
2976 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002977
2978 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002979 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980
2981 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002982 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002983 if (ret) {
2984 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002985 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002986 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002987 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002988 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002989 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002990 }
2991 else {
2992 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002993 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002994 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002995 comment = tcpcheck_get_step_comment(check, step);
2996 if (comment)
2997 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002998 set_server_check_status(check, HCHK_STATUS_L7RSP,
2999 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003000 goto out_end_tcpcheck;
3001 }
3002 /* matched and was supposed to => OK, next step */
3003 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003004 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003005 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003006
3007 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003008 while (&check->current_step->list != head &&
3009 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003010 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003011
Willy Tarreauf2c87352015-05-13 12:08:21 +02003012 if (&check->current_step->list == head)
3013 break;
3014
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003015 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003016 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003017 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003018 }
3019 }
3020 else {
3021 /* not matched */
3022 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003023 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003024 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02003025 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003026
3027 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02003028 while (&check->current_step->list != head &&
3029 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02003030 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003031
Willy Tarreauf2c87352015-05-13 12:08:21 +02003032 if (&check->current_step->list == head)
3033 break;
3034
Willy Tarreau98aec9f2013-12-06 16:16:41 +01003035 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003036 goto tcpcheck_expect;
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003037 __cs_stop_recv(cs);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003038 }
3039 /* not matched but was supposed to => ERROR */
3040 else {
3041 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003042 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003043 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02003044 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003045 }
3046 else {
3047 /* we were looking for a regex */
3048 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02003049 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003050 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02003051 comment = tcpcheck_get_step_comment(check, step);
3052 if (comment)
3053 chunk_appendf(&trash, " comment: '%s'", comment);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02003054 set_server_check_status(check, HCHK_STATUS_L7RSP,
3055 trash.area);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003056 goto out_end_tcpcheck;
3057 }
3058 }
3059 } /* end expect */
3060 } /* end loop over double chained step list */
3061
Baptiste Assmann248f1172018-03-01 21:49:01 +01003062 /* don't do anything until the connection is established */
3063 if (!(conn->flags & CO_FL_CONNECTED)) {
3064 /* update expire time, should be done by process_chk */
3065 /* we allow up to min(inter, timeout.connect) for a connection
3066 * to establish but only when timeout.check is set
3067 * as it may be to short for a full check otherwise
3068 */
3069 while (tick_is_expired(t->expire, now_ms)) {
3070 int t_con;
3071
3072 t_con = tick_add(t->expire, s->proxy->timeout.connect);
3073 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
3074
3075 if (s->proxy->timeout.check)
3076 t->expire = tick_first(t->expire, t_con);
3077 }
3078 goto out;
3079 }
3080
Willy Tarreau263013d2015-05-13 11:59:14 +02003081 /* We're waiting for some I/O to complete, we've reached the end of the
3082 * rules, or both. Do what we have to do, otherwise we're done.
3083 */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003084 if (&check->current_step->list == head && !b_data(&check->bo)) {
Willy Tarreau263013d2015-05-13 11:59:14 +02003085 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
3086 goto out_end_tcpcheck;
3087 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003088
Willy Tarreau53c5a042015-05-13 11:38:17 +02003089 /* warning, current_step may now point to the head */
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003090 if (b_data(&check->bo))
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003091 __cs_want_send(cs);
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003092
Willy Tarreau53c5a042015-05-13 11:38:17 +02003093 if (&check->current_step->list != head &&
3094 check->current_step->action == TCPCHK_ACT_EXPECT)
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003095 __cs_want_recv(cs);
Christopher Fauletb6102852017-11-28 10:06:29 +01003096 goto out;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003097
3098 out_end_tcpcheck:
3099 /* collect possible new errors */
Willy Tarreau4ff3b892017-10-16 15:17:17 +02003100 if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)
Willy Tarreaub5259bf2017-10-04 14:47:29 +02003101 chk_report_conn_err(check, 0, 0);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003102
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003103 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003104 check->current_step = NULL;
3105
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003106 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003107 conn->flags |= CO_FL_ERROR;
3108
Olivier Houchard9aaf7782017-09-13 18:30:23 +02003109 __cs_stop_both(cs);
Willy Tarreau62ac84f2017-11-05 10:11:13 +01003110
Christopher Fauletb6102852017-11-28 10:06:29 +01003111 out:
Willy Tarreau6bdcab02017-10-04 18:41:00 +02003112 return retcode;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003113}
3114
Simon Hormanb1900d52015-01-30 11:22:54 +09003115const char *init_check(struct check *check, int type)
3116{
3117 check->type = type;
3118
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003119 b_reset(&check->bi); check->bi.size = global.tune.chksize;
3120 b_reset(&check->bo); check->bo.size = global.tune.chksize;
Simon Hormanb1900d52015-01-30 11:22:54 +09003121
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003122 check->bi.area = calloc(check->bi.size, sizeof(char));
3123 check->bo.area = calloc(check->bo.size, sizeof(char));
3124
3125 if (!check->bi.area || !check->bo.area)
Simon Hormanb1900d52015-01-30 11:22:54 +09003126 return "out of memory while allocating check buffer";
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003127
Simon Hormanb1900d52015-01-30 11:22:54 +09003128 return NULL;
3129}
3130
Simon Hormanbfb5d332015-01-30 11:22:55 +09003131void free_check(struct check *check)
3132{
Willy Tarreauc9fa0482018-07-10 17:43:27 +02003133 free(check->bi.area);
3134 free(check->bo.area);
Christopher Faulet23d86d12018-01-25 11:36:35 +01003135 if (check->cs) {
3136 free(check->cs->conn);
3137 check->cs->conn = NULL;
3138 cs_free(check->cs);
3139 check->cs = NULL;
3140 }
Simon Hormanbfb5d332015-01-30 11:22:55 +09003141}
3142
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003143void email_alert_free(struct email_alert *alert)
3144{
3145 struct tcpcheck_rule *rule, *back;
3146
3147 if (!alert)
3148 return;
3149
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003150 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) {
3151 LIST_DEL(&rule->list);
3152 free(rule->comment);
3153 free(rule->string);
3154 if (rule->expect_regex)
3155 regex_free(rule->expect_regex);
Willy Tarreaubafbe012017-11-24 17:34:44 +01003156 pool_free(pool_head_tcpcheck_rule, rule);
Christopher Fauletde1a75b2017-10-23 15:38:19 +02003157 }
Willy Tarreaubafbe012017-11-24 17:34:44 +01003158 pool_free(pool_head_email_alert, alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003159}
3160
Olivier Houchard9f6af332018-05-25 14:04:04 +02003161static struct task *process_email_alert(struct task *t, void *context, unsigned short state)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003162{
Olivier Houchard9f6af332018-05-25 14:04:04 +02003163 struct check *check = context;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003164 struct email_alertq *q;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003165 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003166
3167 q = container_of(check, typeof(*q), check);
3168
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003169 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003170 while (1) {
3171 if (!(check->state & CHK_ST_ENABLED)) {
3172 if (LIST_ISEMPTY(&q->email_alerts)) {
3173 /* All alerts processed, queue the task */
3174 t->expire = TICK_ETERNITY;
3175 task_queue(t);
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003176 goto end;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003177 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003178
3179 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003180 LIST_DEL(&alert->list);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003181 t->expire = now_ms;
3182 check->server = alert->srv;
3183 check->tcpcheck_rules = &alert->tcpcheck_rules;
PiBa-NL1714b9f2017-12-06 01:35:43 +01003184 check->status = HCHK_STATUS_UNKNOWN; // the UNKNOWN status is used to exit set_server_check_status(.) early
Christopher Faulet0108bb32017-10-20 21:34:32 +02003185 check->state |= CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003186 }
3187
Olivier Houchard9f6af332018-05-25 14:04:04 +02003188 process_chk(t, context, state);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003189 if (check->state & CHK_ST_INPROGRESS)
3190 break;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003191
3192 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3193 email_alert_free(alert);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003194 check->tcpcheck_rules = NULL;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003195 check->server = NULL;
3196 check->state &= ~CHK_ST_ENABLED;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003197 }
Christopher Fauletc2a89a62017-10-23 15:54:24 +02003198 end:
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003199 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003200 return t;
3201}
3202
Christopher Faulet0108bb32017-10-20 21:34:32 +02003203/* Initializes mailer alerts for the proxy <p> using <mls> parameters.
3204 *
3205 * The function returns 1 in success case, otherwise, it returns 0 and err is
3206 * filled.
3207 */
3208int init_email_alert(struct mailers *mls, struct proxy *p, char **err)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003209{
Christopher Faulet0108bb32017-10-20 21:34:32 +02003210 struct mailer *mailer;
3211 struct email_alertq *queues;
3212 const char *err_str;
3213 int i = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003214
Christopher Faulet0108bb32017-10-20 21:34:32 +02003215 if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) {
3216 memprintf(err, "out of memory while allocating mailer alerts queues");
3217 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003218 }
3219
Christopher Faulet0108bb32017-10-20 21:34:32 +02003220 for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) {
3221 struct email_alertq *q = &queues[i];
3222 struct check *check = &q->check;
3223 struct task *t;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003224
3225 LIST_INIT(&q->email_alerts);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003226 HA_SPIN_INIT(&q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003227 check->inter = mls->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003228 check->rise = DEF_AGENT_RISETIME;
3229 check->fall = DEF_AGENT_FALLTIME;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003230 if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) {
3231 memprintf(err, "%s", err_str);
3232 goto error;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003233 }
3234
3235 check->xprt = mailer->xprt;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003236 check->addr = mailer->addr;
Christopher Fauletb797ae12018-03-27 15:35:35 +02003237 check->port = get_host_port(&mailer->addr);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003238 //check->server = s;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003239
Emeric Brunc60def82017-09-27 14:59:38 +02003240 if ((t = task_new(MAX_THREADS_MASK)) == NULL) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003241 memprintf(err, "out of memory while allocating mailer alerts task");
3242 goto error;
3243 }
3244
3245 check->task = t;
3246 t->process = process_email_alert;
3247 t->context = check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003248
Christopher Faulet0108bb32017-10-20 21:34:32 +02003249 /* check this in one ms */
3250 t->expire = TICK_ETERNITY;
3251 check->start = now;
3252 task_queue(t);
3253 }
3254
3255 mls->users++;
3256 free(p->email_alert.mailers.name);
3257 p->email_alert.mailers.m = mls;
3258 p->email_alert.queues = queues;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003259 return 0;
Christopher Faulet0108bb32017-10-20 21:34:32 +02003260
3261 error:
3262 for (i = 0; i < mls->count; i++) {
3263 struct email_alertq *q = &queues[i];
3264 struct check *check = &q->check;
3265
3266 if (check->task) {
3267 task_delete(check->task);
3268 task_free(check->task);
3269 check->task = NULL;
3270 }
3271 free_check(check);
3272 }
3273 free(queues);
3274 return 1;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003275}
3276
3277
3278static int add_tcpcheck_expect_str(struct list *list, const char *str)
3279{
3280 struct tcpcheck_rule *tcpcheck;
3281
Willy Tarreaubafbe012017-11-24 17:34:44 +01003282 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003283 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003284 memset(tcpcheck, 0, sizeof(*tcpcheck));
3285 tcpcheck->action = TCPCHK_ACT_EXPECT;
3286 tcpcheck->string = strdup(str);
3287 tcpcheck->expect_regex = NULL;
3288 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003289 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003290 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003291 return 0;
3292 }
3293
3294 LIST_ADDQ(list, &tcpcheck->list);
3295 return 1;
3296}
3297
3298static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3299{
3300 struct tcpcheck_rule *tcpcheck;
Willy Tarreau64345aa2016-08-10 19:29:09 +02003301 const char *in;
3302 char *dst;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003303 int i;
3304
Willy Tarreaubafbe012017-11-24 17:34:44 +01003305 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003306 return 0;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003307 memset(tcpcheck, 0, sizeof(*tcpcheck));
3308 tcpcheck->action = TCPCHK_ACT_SEND;
3309 tcpcheck->expect_regex = NULL;
3310 tcpcheck->comment = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003311 tcpcheck->string_len = 0;
3312 for (i = 0; strs[i]; i++)
3313 tcpcheck->string_len += strlen(strs[i]);
3314
3315 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3316 if (!tcpcheck->string) {
Willy Tarreaubafbe012017-11-24 17:34:44 +01003317 pool_free(pool_head_tcpcheck_rule, tcpcheck);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003318 return 0;
3319 }
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003320
Willy Tarreau64345aa2016-08-10 19:29:09 +02003321 dst = tcpcheck->string;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003322 for (i = 0; strs[i]; i++)
Willy Tarreau64345aa2016-08-10 19:29:09 +02003323 for (in = strs[i]; (*dst = *in++); dst++);
3324 *dst = 0;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003325
3326 LIST_ADDQ(list, &tcpcheck->list);
3327 return 1;
3328}
3329
Christopher Faulet0108bb32017-10-20 21:34:32 +02003330static int enqueue_one_email_alert(struct proxy *p, struct server *s,
3331 struct email_alertq *q, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003332{
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003333 struct email_alert *alert;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003334 struct tcpcheck_rule *tcpcheck;
3335 struct check *check = &q->check;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003336
Willy Tarreaubafbe012017-11-24 17:34:44 +01003337 if ((alert = pool_alloc(pool_head_email_alert)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003338 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003339 LIST_INIT(&alert->list);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003340 LIST_INIT(&alert->tcpcheck_rules);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003341 alert->srv = s;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003342
Willy Tarreaubafbe012017-11-24 17:34:44 +01003343 if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003344 goto error;
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003345 memset(tcpcheck, 0, sizeof(*tcpcheck));
3346 tcpcheck->action = TCPCHK_ACT_CONNECT;
3347 tcpcheck->comment = NULL;
3348 tcpcheck->string = NULL;
3349 tcpcheck->expect_regex = NULL;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003350 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3351
3352 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3353 goto error;
3354
3355 {
3356 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3357 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3358 goto error;
3359 }
3360
3361 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3362 goto error;
3363
3364 {
3365 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3366 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3367 goto error;
3368 }
3369
3370 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3371 goto error;
3372
3373 {
3374 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3375 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3376 goto error;
3377 }
3378
3379 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3380 goto error;
3381
3382 {
3383 const char * const strs[2] = { "DATA\r\n" };
3384 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3385 goto error;
3386 }
3387
3388 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3389 goto error;
3390
3391 {
3392 struct tm tm;
3393 char datestr[48];
3394 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003395 "From: ", p->email_alert.from, "\r\n",
3396 "To: ", p->email_alert.to, "\r\n",
3397 "Date: ", datestr, "\r\n",
3398 "Subject: [HAproxy Alert] ", msg, "\r\n",
3399 "\r\n",
3400 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003401 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003402 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003403 NULL
3404 };
3405
3406 get_localtime(date.tv_sec, &tm);
3407
3408 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3409 goto error;
3410 }
3411
3412 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3413 goto error;
3414 }
3415
3416 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3417 goto error;
3418
3419 {
3420 const char * const strs[2] = { "QUIT\r\n" };
3421 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3422 goto error;
3423 }
3424
3425 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3426 goto error;
3427
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003428 HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock);
Christopher Faulet0108bb32017-10-20 21:34:32 +02003429 task_wakeup(check->task, TASK_WOKEN_MSG);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003430 LIST_ADDQ(&q->email_alerts, &alert->list);
Christopher Faulet2a944ee2017-11-07 10:42:54 +01003431 HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003432 return 1;
3433
3434error:
3435 email_alert_free(alert);
3436 return 0;
3437}
3438
Christopher Faulet0108bb32017-10-20 21:34:32 +02003439static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003440{
3441 int i;
3442 struct mailer *mailer;
3443
3444 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3445 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
Christopher Faulet0108bb32017-10-20 21:34:32 +02003446 if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003447 ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003448 return;
3449 }
3450 }
3451
3452 return;
3453}
3454
3455/*
3456 * Send email alert if configured.
3457 */
Simon Horman64e34162015-02-06 11:11:57 +09003458void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003459{
3460 va_list argp;
3461 char buf[1024];
3462 int len;
3463 struct proxy *p = s->proxy;
3464
Christopher Faulet0108bb32017-10-20 21:34:32 +02003465 if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003466 return;
3467
3468 va_start(argp, format);
3469 len = vsnprintf(buf, sizeof(buf), format, argp);
3470 va_end(argp);
3471
Thierry FOURNIER62c8a212017-02-09 12:19:27 +01003472 if (len < 0 || len >= sizeof(buf)) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01003473 ha_alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003474 return;
3475 }
3476
Christopher Faulet0108bb32017-10-20 21:34:32 +02003477 enqueue_email_alert(p, s, buf);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003478}
3479
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003480/*
3481 * Return value:
3482 * the port to be used for the health check
3483 * 0 in case no port could be found for the check
3484 */
3485int srv_check_healthcheck_port(struct check *chk)
3486{
3487 int i = 0;
3488 struct server *srv = NULL;
3489
3490 srv = chk->server;
3491
3492 /* If neither a port nor an addr was specified and no check transport
3493 * layer is forced, then the transport layer used by the checks is the
3494 * same as for the production traffic. Otherwise we use raw_sock by
3495 * default, unless one is specified.
3496 */
3497 if (!chk->port && !is_addr(&chk->addr)) {
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003498 chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL));
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003499 chk->send_proxy |= (srv->pp_opts);
3500 }
3501
3502 /* by default, we use the health check port ocnfigured */
3503 if (chk->port > 0)
3504 return chk->port;
3505
3506 /* try to get the port from check_core.addr if check.port not set */
3507 i = get_host_port(&chk->addr);
3508 if (i > 0)
3509 return i;
3510
3511 /* try to get the port from server address */
3512 /* prevent MAPPORTS from working at this point, since checks could
3513 * not be performed in such case (MAPPORTS impose a relative ports
3514 * based on live traffic)
3515 */
3516 if (srv->flags & SRV_F_MAPPORTS)
3517 return 0;
Willy Tarreau04276f32017-01-06 17:41:29 +01003518
3519 i = srv->svc_port; /* by default */
Baptiste Assmann95db2bc2016-06-13 14:15:41 +02003520 if (i > 0)
3521 return i;
3522
3523 return 0;
3524}
3525
Willy Tarreau865c5142016-12-21 20:04:48 +01003526__attribute__((constructor))
3527static void __check_init(void)
3528{
3529 hap_register_post_check(start_checks);
Christopher Faulet31dff9b2017-10-23 15:45:20 +02003530
Willy Tarreaubafbe012017-11-24 17:34:44 +01003531 pool_head_email_alert = create_pool("email_alert", sizeof(struct email_alert), MEM_F_SHARED);
3532 pool_head_tcpcheck_rule = create_pool("tcpcheck_rule", sizeof(struct tcpcheck_rule), MEM_F_SHARED);
Willy Tarreau865c5142016-12-21 20:04:48 +01003533}
3534
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003535
Willy Tarreaubd741542010-03-16 18:46:54 +01003536/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003537 * Local variables:
3538 * c-indent-level: 8
3539 * c-basic-offset: 8
3540 * End:
3541 */