blob: 51864e38914eaf9f0498220084ac5f71cccecce3 [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>
Willy Tarreaubaaee002006-06-26 02:48:02 +020038
39#include <types/global.h>
Simon Horman0ba0e4a2015-01-30 11:23:00 +090040#include <types/mailers.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020041#include <types/dns.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020042
Baptiste Assmann69e273f2013-12-11 00:52:19 +010043#ifdef USE_OPENSSL
44#include <types/ssl_sock.h>
45#include <proto/ssl_sock.h>
46#endif /* USE_OPENSSL */
47
Willy Tarreaubaaee002006-06-26 02:48:02 +020048#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020049#include <proto/checks.h>
Simon Hormana2b9dad2013-02-12 10:45:54 +090050#include <proto/dumpstats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/fd.h>
52#include <proto/log.h>
53#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020054#include <proto/port_range.h>
Willy Tarreau3d300592007-03-18 18:34:41 +010055#include <proto/proto_http.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010056#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010057#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010058#include <proto/proxy.h>
Willy Tarreaufb56aab2012-09-28 14:40:02 +020059#include <proto/raw_sock.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020060#include <proto/server.h>
Willy Tarreau48d6bf22016-06-21 16:27:34 +020061#include <proto/signal.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010062#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020063#include <proto/task.h>
Baptiste Assmanna68ca962015-04-14 01:15:08 +020064#include <proto/log.h>
65#include <proto/dns.h>
66#include <proto/proto_udp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020067
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);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +020071static void tcpcheck_main(struct connection *);
Willy Tarreaubd741542010-03-16 18:46:54 +010072
Simon Horman63a4a822012-03-19 07:24:41 +090073static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010074 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
75 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020076 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020077
Willy Tarreau23964182014-05-20 20:56:30 +020078 /* Below we have finished checks */
79 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010080 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010081
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010082 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020083
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010084 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
85 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
86 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
89 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
90 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010092 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
93 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020094
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020095 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020096
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010097 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
98 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
99 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Simon Horman98637e52014-06-20 12:30:16 +0900100
101 [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" },
102 [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" },
Cyril Bonté77010d82014-08-07 01:55:37 +0200103 [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200104};
105
Cyril Bontéac92a062014-12-27 22:28:38 +0100106const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = {
107 [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT },
108 [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT },
109 [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT },
110 [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT },
111 [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT },
112 [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT },
113 [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT },
114 [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT },
115 [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT },
116 [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT },
117 [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG },
118};
119
Simon Horman63a4a822012-03-19 07:24:41 +0900120static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100121 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
122
123 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
124 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
125
126 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
127 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
128 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
129 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
130
131 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
132 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
133 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
134};
135
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200136/*
137 * Convert check_status code to description
138 */
139const char *get_check_status_description(short check_status) {
140
141 const char *desc;
142
143 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200144 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200145 else
146 desc = NULL;
147
148 if (desc && *desc)
149 return desc;
150 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200151 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200152}
153
154/*
155 * Convert check_status code to short info
156 */
157const char *get_check_status_info(short check_status) {
158
159 const char *info;
160
161 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200162 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200163 else
164 info = NULL;
165
166 if (info && *info)
167 return info;
168 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200169 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200170}
171
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100172const char *get_analyze_status(short analyze_status) {
173
174 const char *desc;
175
176 if (analyze_status < HANA_STATUS_SIZE)
177 desc = analyze_statuses[analyze_status].desc;
178 else
179 desc = NULL;
180
181 if (desc && *desc)
182 return desc;
183 else
184 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
185}
186
Willy Tarreaua150cf12014-05-20 21:57:23 +0200187/* Builds a string containing some information about the health check's result.
188 * The output string is allocated from the trash chunks. If the check is NULL,
189 * NULL is returned. This is designed to be used when emitting logs about health
190 * checks.
Willy Tarreauddd329c2014-05-16 16:46:12 +0200191 */
Willy Tarreaua150cf12014-05-20 21:57:23 +0200192static const char *check_reason_string(struct check *check)
Willy Tarreauddd329c2014-05-16 16:46:12 +0200193{
Willy Tarreaua150cf12014-05-20 21:57:23 +0200194 struct chunk *msg;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200195
Willy Tarreaua150cf12014-05-20 21:57:23 +0200196 if (!check)
197 return NULL;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200198
Willy Tarreaua150cf12014-05-20 21:57:23 +0200199 msg = get_trash_chunk();
200 chunk_printf(msg, "reason: %s", get_check_status_description(check->status));
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200201
Willy Tarreaua150cf12014-05-20 21:57:23 +0200202 if (check->status >= HCHK_STATUS_L57DATA)
203 chunk_appendf(msg, ", code: %d", check->code);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200204
Willy Tarreaua150cf12014-05-20 21:57:23 +0200205 if (*check->desc) {
206 struct chunk src;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200207
Willy Tarreaua150cf12014-05-20 21:57:23 +0200208 chunk_appendf(msg, ", info: \"");
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200209
Willy Tarreaua150cf12014-05-20 21:57:23 +0200210 chunk_initlen(&src, check->desc, 0, strlen(check->desc));
211 chunk_asciiencode(msg, &src, '"');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200212
Willy Tarreaua150cf12014-05-20 21:57:23 +0200213 chunk_appendf(msg, "\"");
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200214 }
215
Willy Tarreaua150cf12014-05-20 21:57:23 +0200216 if (check->duration >= 0)
217 chunk_appendf(msg, ", check duration: %ldms", check->duration);
218
219 return msg->str;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200220}
221
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200222/*
Simon Horman4a741432013-02-23 15:35:38 +0900223 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200224 * an adequate CHK_RES_* value. The new check->health is computed based
225 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200226 *
227 * Show information in logs about failed health check if server is UP
228 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200229 */
Simon Horman4a741432013-02-23 15:35:38 +0900230static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100231{
Simon Horman4a741432013-02-23 15:35:38 +0900232 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200233 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200234 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900235
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200236 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100237 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900238 check->desc[0] = '\0';
239 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200240 return;
241 }
242
Simon Horman4a741432013-02-23 15:35:38 +0900243 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200244 return;
245
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200246 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900247 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
248 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200249 } else
Simon Horman4a741432013-02-23 15:35:38 +0900250 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200251
Simon Horman4a741432013-02-23 15:35:38 +0900252 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200253 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900254 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200255
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100256 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900257 check->duration = -1;
258 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200259 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900260 check->duration = tv_ms_elapsed(&check->start, &now);
261 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200262 }
263
Willy Tarreau23964182014-05-20 20:56:30 +0200264 /* no change is expected if no state change occurred */
265 if (check->result == CHK_RES_NEUTRAL)
266 return;
267
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200268 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200269
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200270 switch (check->result) {
271 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200272 /* Failure to connect to the agent as a secondary check should not
273 * cause the server to be marked down.
274 */
275 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormaneaabd522015-02-26 11:26:17 +0900276 (check->status >= HCHK_STATUS_L57DATA)) &&
Willy Tarreau12634e12014-05-23 11:32:36 +0200277 (check->health >= check->rise)) {
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200278 s->counters.failed_checks++;
279 report = 1;
280 check->health--;
281 if (check->health < check->rise)
282 check->health = 0;
283 }
284 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200285
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200286 case CHK_RES_PASSED:
287 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
288 if ((check->health < check->rise + check->fall - 1) &&
289 (check->result == CHK_RES_PASSED || check->health > 0)) {
290 report = 1;
291 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200292
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200293 if (check->health >= check->rise)
294 check->health = check->rise + check->fall - 1; /* OK now */
295 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200296
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200297 /* clear consecutive_errors if observing is enabled */
298 if (s->onerror)
299 s->consecutive_errors = 0;
300 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100301
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200302 default:
303 break;
304 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200305
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200306 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
307 (status != prev_status || report)) {
308 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200309 "%s check for %sserver %s/%s %s%s",
310 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200311 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100312 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100313 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200314 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200315
Willy Tarreaua150cf12014-05-20 21:57:23 +0200316 srv_append_status(&trash, s, check_reason_string(check), -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200317
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100318 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200319 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
320 (check->health >= check->rise) ? check->fall : check->rise,
321 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200322
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100323 Warning("%s.\n", trash.str);
324 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
Simon Horman7ea9be02015-04-30 13:10:33 +0900325 send_email_alert(s, LOG_INFO, "%s", trash.str);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200326 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200327}
328
Willy Tarreau4eec5472014-05-20 22:32:27 +0200329/* Marks the check <check>'s server down if the current check is already failed
330 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200331 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200332static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200333{
Simon Horman4a741432013-02-23 15:35:38 +0900334 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900335
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200336 /* The agent secondary check should only cause a server to be marked
337 * as down if check->status is HCHK_STATUS_L7STS, which indicates
338 * that the agent returned "fail", "stopped" or "down".
339 * The implication here is that failure to connect to the agent
340 * as a secondary check should not cause the server to be marked
341 * down. */
342 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
343 return;
344
Willy Tarreau4eec5472014-05-20 22:32:27 +0200345 if (check->health > 0)
346 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100347
Willy Tarreau4eec5472014-05-20 22:32:27 +0200348 /* We only report a reason for the check if we did not do so previously */
349 srv_set_stopped(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200350}
351
Willy Tarreauaf549582014-05-16 17:37:50 +0200352/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200353 * it isn't in maintenance, it is not tracking a down server and other checks
354 * comply. The rule is simple : by default, a server is up, unless any of the
355 * following conditions is true :
356 * - health check failed (check->health < rise)
357 * - agent check failed (agent->health < rise)
358 * - the server tracks a down server (track && track->state == STOPPED)
359 * Note that if the server has a slowstart, it will switch to STARTING instead
360 * of RUNNING. Also, only the health checks support the nolb mode, so the
361 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200362 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200363static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200364{
Simon Horman4a741432013-02-23 15:35:38 +0900365 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100366
Willy Tarreauaf549582014-05-16 17:37:50 +0200367 if (s->admin & SRV_ADMF_MAINT)
368 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100369
Willy Tarreau3e048382014-05-21 10:30:54 +0200370 if (s->track && s->track->state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200371 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100372
Willy Tarreau3e048382014-05-21 10:30:54 +0200373 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
374 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100375
Willy Tarreau3e048382014-05-21 10:30:54 +0200376 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
377 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200378
Willy Tarreau3e048382014-05-21 10:30:54 +0200379 if ((check->state & CHK_ST_AGENT) && s->state == SRV_ST_STOPPING)
380 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100381
Willy Tarreau3e048382014-05-21 10:30:54 +0200382 srv_set_running(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100383}
384
Willy Tarreaudb58b792014-05-21 13:57:23 +0200385/* Marks the check <check> as valid and tries to set its server into stopping mode
386 * if it was running or starting, and provided it isn't in maintenance and other
387 * checks comply. The conditions for the server to be marked in stopping mode are
388 * the same as for it to be turned up. Also, only the health checks support the
389 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200390 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200391static void check_notify_stopping(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200392{
Simon Horman4a741432013-02-23 15:35:38 +0900393 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100394
Willy Tarreauaf549582014-05-16 17:37:50 +0200395 if (s->admin & SRV_ADMF_MAINT)
396 return;
397
Willy Tarreaudb58b792014-05-21 13:57:23 +0200398 if (check->state & CHK_ST_AGENT)
399 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100400
Willy Tarreaudb58b792014-05-21 13:57:23 +0200401 if (s->track && s->track->state == SRV_ST_STOPPED)
402 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100403
Willy Tarreaudb58b792014-05-21 13:57:23 +0200404 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
405 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100406
Willy Tarreaudb58b792014-05-21 13:57:23 +0200407 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
408 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100409
Willy Tarreaudb58b792014-05-21 13:57:23 +0200410 srv_set_stopping(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100411}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200412
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100413/* note: use health_adjust() only, which first checks that the observe mode is
414 * enabled.
415 */
416void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100417{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100418 int failed;
419 int expire;
420
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100421 if (s->observe >= HANA_OBS_SIZE)
422 return;
423
Willy Tarreaubb956662013-01-24 00:37:39 +0100424 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100425 return;
426
427 switch (analyze_statuses[status].lr[s->observe - 1]) {
428 case 1:
429 failed = 1;
430 break;
431
432 case 2:
433 failed = 0;
434 break;
435
436 default:
437 return;
438 }
439
440 if (!failed) {
441 /* good: clear consecutive_errors */
442 s->consecutive_errors = 0;
443 return;
444 }
445
446 s->consecutive_errors++;
447
448 if (s->consecutive_errors < s->consecutive_errors_limit)
449 return;
450
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100451 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
452 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100453
454 switch (s->onerror) {
455 case HANA_ONERR_FASTINTER:
456 /* force fastinter - nothing to do here as all modes force it */
457 break;
458
459 case HANA_ONERR_SUDDTH:
460 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900461 if (s->check.health > s->check.rise)
462 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100463
464 /* no break - fall through */
465
466 case HANA_ONERR_FAILCHK:
467 /* simulate a failed health check */
Simon Horman4a741432013-02-23 15:35:38 +0900468 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200469 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100470 break;
471
472 case HANA_ONERR_MARKDWN:
473 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900474 s->check.health = s->check.rise;
Simon Horman4a741432013-02-23 15:35:38 +0900475 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200476 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100477 break;
478
479 default:
480 /* write a warning? */
481 break;
482 }
483
484 s->consecutive_errors = 0;
485 s->counters.failed_hana++;
486
Simon Horman66183002013-02-23 10:16:43 +0900487 if (s->check.fastinter) {
488 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300489 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200490 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300491 /* requeue check task with new expire */
492 task_queue(s->check.task);
493 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100494 }
495}
496
Willy Tarreaua1dab552014-04-14 15:04:54 +0200497static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100498{
499 int sv_state;
500 int ratio;
501 int hlen = 0;
Joseph Lynch514061c2015-01-15 17:52:59 -0800502 char addr[46];
503 char port[6];
Willy Tarreauef781042010-01-27 11:53:01 +0100504 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
505 "UP %d/%d", "UP",
506 "NOLB %d/%d", "NOLB",
507 "no check" };
508
509 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
510 hlen += 24;
511
Willy Tarreauff5ae352013-12-11 20:36:34 +0100512 if (!(s->check.state & CHK_ST_ENABLED))
513 sv_state = 6;
Willy Tarreau892337c2014-05-13 23:41:20 +0200514 else if (s->state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900515 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100516 sv_state = 3; /* UP */
517 else
518 sv_state = 2; /* going down */
519
Willy Tarreau892337c2014-05-13 23:41:20 +0200520 if (s->state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100521 sv_state += 2;
522 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900523 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100524 sv_state = 1; /* going up */
525 else
526 sv_state = 0; /* DOWN */
527 }
528
Willy Tarreaua1dab552014-04-14 15:04:54 +0200529 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100530 srv_hlt_st[sv_state],
Willy Tarreau892337c2014-05-13 23:41:20 +0200531 (s->state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
532 (s->state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100533
Joseph Lynch514061c2015-01-15 17:52:59 -0800534 addr_to_str(&s->addr, addr, sizeof(addr));
535 port_to_str(&s->addr, port, sizeof(port));
536
537 hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
538 addr, port, s->proxy->id, s->id,
Willy Tarreauef781042010-01-27 11:53:01 +0100539 global.node,
540 (s->eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
541 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
542 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
543 s->nbpend);
544
Willy Tarreau892337c2014-05-13 23:41:20 +0200545 if ((s->state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100546 now.tv_sec < s->last_change + s->slowstart &&
547 now.tv_sec >= s->last_change) {
548 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200549 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100550 }
551
552 buffer[hlen++] = '\r';
553 buffer[hlen++] = '\n';
554
555 return hlen;
556}
557
Willy Tarreau20a18342013-12-05 00:31:46 +0100558/* Check the connection. If an error has already been reported or the socket is
559 * closed, keep errno intact as it is supposed to contain the valid error code.
560 * If no error is reported, check the socket's error queue using getsockopt().
561 * Warning, this must be done only once when returning from poll, and never
562 * after an I/O error was attempted, otherwise the error queue might contain
563 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
564 * socket. Returns non-zero if an error was reported, zero if everything is
565 * clean (including a properly closed socket).
566 */
567static int retrieve_errno_from_socket(struct connection *conn)
568{
569 int skerr;
570 socklen_t lskerr = sizeof(skerr);
571
572 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
573 return 1;
574
Willy Tarreau3c728722014-01-23 13:50:42 +0100575 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100576 return 0;
577
578 if (getsockopt(conn->t.sock.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
579 errno = skerr;
580
581 if (errno == EAGAIN)
582 errno = 0;
583
584 if (!errno) {
585 /* we could not retrieve an error, that does not mean there is
586 * none. Just don't change anything and only report the prior
587 * error if any.
588 */
589 if (conn->flags & CO_FL_ERROR)
590 return 1;
591 else
592 return 0;
593 }
594
595 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
596 return 1;
597}
598
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100599/* Try to collect as much information as possible on the connection status,
600 * and adjust the server status accordingly. It may make use of <errno_bck>
601 * if non-null when the caller is absolutely certain of its validity (eg:
602 * checked just after a syscall). If the caller doesn't have a valid errno,
603 * it can pass zero, and retrieve_errno_from_socket() will be called to try
604 * to extract errno from the socket. If no error is reported, it will consider
605 * the <expired> flag. This is intended to be used when a connection error was
606 * reported in conn->flags or when a timeout was reported in <expired>. The
607 * function takes care of not updating a server status which was already set.
608 * All situations where at least one of <expired> or CO_FL_ERROR are set
609 * produce a status.
610 */
611static void chk_report_conn_err(struct connection *conn, int errno_bck, int expired)
612{
613 struct check *check = conn->owner;
614 const char *err_msg;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200615 struct chunk *chk;
Willy Tarreau213c6782014-10-02 14:51:02 +0200616 int step;
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200617 char *comment;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100618
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100619 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100620 return;
621
622 errno = errno_bck;
623 if (!errno || errno == EAGAIN)
624 retrieve_errno_from_socket(conn);
625
626 if (!(conn->flags & CO_FL_ERROR) && !expired)
627 return;
628
629 /* we'll try to build a meaningful error message depending on the
630 * context of the error possibly present in conn->err_code, and the
631 * socket error possibly collected above. This is useful to know the
632 * exact step of the L6 layer (eg: SSL handshake).
633 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200634 chk = get_trash_chunk();
635
636 if (check->type == PR_O2_TCPCHK_CHK) {
Simon Hormane16c1b32015-01-30 11:22:57 +0900637 step = tcpcheck_get_step_id(check);
Willy Tarreau213c6782014-10-02 14:51:02 +0200638 if (!step)
639 chunk_printf(chk, " at initial connection step of tcp-check");
640 else {
641 chunk_printf(chk, " at step %d of tcp-check", step);
642 /* we were looking for a string */
643 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
644 if (check->last_started_step->port)
645 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
646 else
647 chunk_appendf(chk, " (connect)");
648 }
649 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
650 if (check->last_started_step->string)
Baptiste Assmann96a5c9b2015-05-01 08:09:29 +0200651 chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string);
Willy Tarreau213c6782014-10-02 14:51:02 +0200652 else if (check->last_started_step->expect_regex)
653 chunk_appendf(chk, " (expect regex)");
654 }
655 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
656 chunk_appendf(chk, " (send)");
657 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +0200658
659 comment = tcpcheck_get_step_comment(check, step);
660 if (comment)
661 chunk_appendf(chk, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200662 }
663 }
664
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100665 if (conn->err_code) {
666 if (errno && errno != EAGAIN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200667 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100668 else
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200669 chunk_printf(&trash, "%s%s", conn_err_code_str(conn), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100670 err_msg = trash.str;
671 }
672 else {
673 if (errno && errno != EAGAIN) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200674 chunk_printf(&trash, "%s%s", strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100675 err_msg = trash.str;
676 }
677 else {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200678 err_msg = chk->str;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100679 }
680 }
681
682 if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
683 /* L4 not established (yet) */
684 if (conn->flags & CO_FL_ERROR)
685 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
686 else if (expired)
687 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200688
689 /*
690 * might be due to a server IP change.
691 * Let's trigger a DNS resolution if none are currently running.
692 */
693 if ((check->server->resolution) && (check->server->resolution->step == RSLV_STEP_NONE))
694 trigger_resolution(check->server);
695
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100696 }
697 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
698 /* L6 not established (yet) */
699 if (conn->flags & CO_FL_ERROR)
700 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
701 else if (expired)
702 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
703 }
704 else if (conn->flags & CO_FL_ERROR) {
705 /* I/O error after connection was established and before we could diagnose */
706 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
707 }
708 else if (expired) {
709 /* connection established but expired check */
710 if (check->type == PR_O2_SSL3_CHK)
711 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
712 else /* HTTP, SMTP, ... */
713 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
714 }
715
716 return;
717}
718
Willy Tarreaubaaee002006-06-26 02:48:02 +0200719/*
720 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200721 * the connection acknowledgement. If the proxy requires L7 health-checks,
722 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900723 * to set check->status, check->duration and check->result.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200724 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200725static void event_srv_chk_w(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200726{
Simon Horman4a741432013-02-23 15:35:38 +0900727 struct check *check = conn->owner;
728 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900729 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200730
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100731 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100732 goto out_wakeup;
733
Willy Tarreau310987a2014-01-22 19:46:33 +0100734 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100735 return;
736
Willy Tarreau20a18342013-12-05 00:31:46 +0100737 if (retrieve_errno_from_socket(conn)) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100738 chk_report_conn_err(conn, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100739 __conn_data_stop_both(conn);
740 goto out_wakeup;
741 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100742
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100743 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100744 /* if the output is closed, we can't do anything */
745 conn->flags |= CO_FL_ERROR;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100746 chk_report_conn_err(conn, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100747 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200748 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200749
Willy Tarreau06559ac2013-12-05 01:53:08 +0100750 /* here, we know that the connection is established. That's enough for
751 * a pure TCP check.
752 */
753 if (!check->type)
754 goto out_wakeup;
755
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200756 if (check->type == PR_O2_TCPCHK_CHK) {
757 tcpcheck_main(conn);
758 return;
759 }
760
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100761 if (check->bo->o) {
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100762 conn->xprt->snd_buf(conn, check->bo, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100763 if (conn->flags & CO_FL_ERROR) {
764 chk_report_conn_err(conn, errno, 0);
765 __conn_data_stop_both(conn);
766 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200767 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100768 if (check->bo->o)
769 return;
770 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200771
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100772 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
773 if (s->proxy->timeout.check) {
774 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
775 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200776 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100777 goto out_nowake;
778
Willy Tarreau83749182007-04-15 20:56:27 +0200779 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200780 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200781 out_nowake:
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200782 __conn_data_stop_send(conn); /* nothing more to write */
Willy Tarreaubaaee002006-06-26 02:48:02 +0200783}
784
Willy Tarreaubaaee002006-06-26 02:48:02 +0200785/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200786 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200787 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900788 * set_server_check_status() to update check->status, check->duration
789 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200790
791 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
792 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
793 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
794 * response to an SSL HELLO (the principle is that this is enough to
795 * distinguish between an SSL server and a pure TCP relay). All other cases will
796 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
797 * etc.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200798 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200799static void event_srv_chk_r(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200800{
Simon Horman4a741432013-02-23 15:35:38 +0900801 struct check *check = conn->owner;
802 struct server *s = check->server;
803 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200804 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100805 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200806 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200807
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100808 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200809 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200810
Willy Tarreau310987a2014-01-22 19:46:33 +0100811 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200812 return;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200813
814 if (check->type == PR_O2_TCPCHK_CHK) {
815 tcpcheck_main(conn);
816 return;
817 }
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200818
Willy Tarreau83749182007-04-15 20:56:27 +0200819 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
820 * but the connection was closed on the remote end. Fortunately, recv still
821 * works correctly and we don't need to do the getsockopt() on linux.
822 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000823
824 /* Set buffer to point to the end of the data already read, and check
825 * that there is free space remaining. If the buffer is full, proceed
826 * with running the checks without attempting another socket read.
827 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000828
Willy Tarreau03938182010-03-17 21:52:07 +0100829 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000830
Simon Horman4a741432013-02-23 15:35:38 +0900831 conn->xprt->rcv_buf(conn, check->bi, check->bi->size);
Willy Tarreauf1503172012-09-28 19:39:36 +0200832 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
Willy Tarreau03938182010-03-17 21:52:07 +0100833 done = 1;
Simon Horman4a741432013-02-23 15:35:38 +0900834 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200835 /* Report network errors only if we got no other data. Otherwise
836 * we'll let the upper layers decide whether the response is OK
837 * or not. It is very common that an RST sent by the server is
838 * reported as an error just after the last data chunk.
839 */
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100840 chk_report_conn_err(conn, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100841 goto out_wakeup;
842 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200843 }
844
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100845
Willy Tarreau03938182010-03-17 21:52:07 +0100846 /* Intermediate or complete response received.
Simon Horman4a741432013-02-23 15:35:38 +0900847 * Terminate string in check->bi->data buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100848 */
Simon Horman4a741432013-02-23 15:35:38 +0900849 if (check->bi->i < check->bi->size)
850 check->bi->data[check->bi->i] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100851 else {
Simon Horman4a741432013-02-23 15:35:38 +0900852 check->bi->data[check->bi->i - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100853 done = 1; /* buffer full, don't wait for more data */
854 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200855
Nick Chalk57b1bf72010-03-16 15:50:46 +0000856 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900857 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200858 case PR_O2_HTTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900859 if (!done && check->bi->i < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100860 goto wait_more_data;
861
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100862 /* Check if the server speaks HTTP 1.X */
Simon Horman4a741432013-02-23 15:35:38 +0900863 if ((check->bi->i < strlen("HTTP/1.0 000\r")) ||
864 (memcmp(check->bi->data, "HTTP/1.", 7) != 0 ||
865 (*(check->bi->data + 12) != ' ' && *(check->bi->data + 12) != '\r')) ||
866 !isdigit((unsigned char) *(check->bi->data + 9)) || !isdigit((unsigned char) *(check->bi->data + 10)) ||
867 !isdigit((unsigned char) *(check->bi->data + 11))) {
868 cut_crlf(check->bi->data);
869 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200870
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100871 goto out_wakeup;
872 }
873
Simon Horman4a741432013-02-23 15:35:38 +0900874 check->code = str2uic(check->bi->data + 9);
875 desc = ltrim(check->bi->data + 12, ' ');
Nick Chalk57b1bf72010-03-16 15:50:46 +0000876
Willy Tarreaubd741542010-03-16 18:46:54 +0100877 if ((s->proxy->options & PR_O_DISABLE404) &&
Willy Tarreau892337c2014-05-13 23:41:20 +0200878 (s->state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000879 /* 404 may be accepted as "stopping" only if the server was up */
880 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900881 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000882 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100883 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
884 /* Run content verification check... We know we have at least 13 chars */
885 if (!httpchk_expect(s, done))
886 goto wait_more_data;
887 }
888 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Simon Horman4a741432013-02-23 15:35:38 +0900889 else if (*(check->bi->data + 9) == '2' || *(check->bi->data + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100890 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900891 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100892 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000893 else {
894 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900895 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000896 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200897 break;
898
899 case PR_O2_SSL3_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900900 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100901 goto wait_more_data;
902
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100903 /* Check for SSLv3 alert or handshake */
Simon Horman4a741432013-02-23 15:35:38 +0900904 if ((check->bi->i >= 5) && (*check->bi->data == 0x15 || *check->bi->data == 0x16))
905 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200906 else
Simon Horman4a741432013-02-23 15:35:38 +0900907 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200908 break;
909
910 case PR_O2_SMTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900911 if (!done && check->bi->i < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100912 goto wait_more_data;
913
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200914 /* Check if the server speaks SMTP */
Simon Horman4a741432013-02-23 15:35:38 +0900915 if ((check->bi->i < strlen("000\r")) ||
916 (*(check->bi->data + 3) != ' ' && *(check->bi->data + 3) != '\r') ||
917 !isdigit((unsigned char) *check->bi->data) || !isdigit((unsigned char) *(check->bi->data + 1)) ||
918 !isdigit((unsigned char) *(check->bi->data + 2))) {
919 cut_crlf(check->bi->data);
920 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200921
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200922 goto out_wakeup;
923 }
924
Simon Horman4a741432013-02-23 15:35:38 +0900925 check->code = str2uic(check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200926
Simon Horman4a741432013-02-23 15:35:38 +0900927 desc = ltrim(check->bi->data + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200928 cut_crlf(desc);
929
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100930 /* Check for SMTP code 2xx (should be 250) */
Simon Horman4a741432013-02-23 15:35:38 +0900931 if (*check->bi->data == '2')
932 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200933 else
Simon Horman4a741432013-02-23 15:35:38 +0900934 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200935 break;
936
Simon Hormana2b9dad2013-02-12 10:45:54 +0900937 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100938 int status = HCHK_STATUS_CHECKED;
939 const char *hs = NULL; /* health status */
940 const char *as = NULL; /* admin status */
941 const char *ps = NULL; /* performance status */
Nenad Merdanovic174dd372016-04-24 23:10:06 +0200942 const char *cs = NULL; /* maxconn */
Willy Tarreau81f5d942013-12-09 20:51:51 +0100943 const char *err = NULL; /* first error to report */
944 const char *wrn = NULL; /* first warning to report */
945 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900946
Willy Tarreau81f5d942013-12-09 20:51:51 +0100947 /* We're getting an agent check response. The agent could
948 * have been disabled in the mean time with a long check
949 * still pending. It is important that we ignore the whole
950 * response.
951 */
952 if (!(check->server->agent.state & CHK_ST_ENABLED))
953 break;
954
955 /* The agent supports strings made of a single line ended by the
956 * first CR ('\r') or LF ('\n'). This line is composed of words
957 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
958 * line may optionally contained a description of a state change
959 * after a sharp ('#'), which is only considered if a health state
960 * is announced.
961 *
962 * Words may be composed of :
963 * - a numeric weight suffixed by the percent character ('%').
964 * - a health status among "up", "down", "stopped", and "fail".
965 * - an admin status among "ready", "drain", "maint".
966 *
967 * These words may appear in any order. If multiple words of the
968 * same category appear, the last one wins.
969 */
970
Willy Tarreau9809b782013-12-11 21:40:11 +0100971 p = check->bi->data;
972 while (*p && *p != '\n' && *p != '\r')
973 p++;
974
975 if (!*p) {
976 if (!done)
977 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900978
Willy Tarreau9809b782013-12-11 21:40:11 +0100979 /* at least inform the admin that the agent is mis-behaving */
980 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
981 break;
982 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100983
Willy Tarreau9809b782013-12-11 21:40:11 +0100984 *p = 0;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100985 cmd = check->bi->data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900986
Willy Tarreau81f5d942013-12-09 20:51:51 +0100987 while (*cmd) {
988 /* look for next word */
989 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
990 cmd++;
991 continue;
992 }
Simon Horman671b6f02013-11-25 10:46:39 +0900993
Willy Tarreau81f5d942013-12-09 20:51:51 +0100994 if (*cmd == '#') {
995 /* this is the beginning of a health status description,
996 * skip the sharp and blanks.
997 */
998 cmd++;
999 while (*cmd == '\t' || *cmd == ' ')
1000 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +09001001 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001002 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001003
1004 /* find the end of the word so that we have a null-terminated
1005 * word between <cmd> and <p>.
1006 */
1007 p = cmd + 1;
1008 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
1009 p++;
1010 if (*p)
1011 *p++ = 0;
1012
1013 /* first, health statuses */
1014 if (strcasecmp(cmd, "up") == 0) {
1015 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001016 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +01001017 hs = cmd;
1018 }
1019 else if (strcasecmp(cmd, "down") == 0) {
1020 check->health = 0;
1021 status = HCHK_STATUS_L7STS;
1022 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001023 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001024 else if (strcasecmp(cmd, "stopped") == 0) {
1025 check->health = 0;
1026 status = HCHK_STATUS_L7STS;
1027 hs = cmd;
1028 }
1029 else if (strcasecmp(cmd, "fail") == 0) {
1030 check->health = 0;
1031 status = HCHK_STATUS_L7STS;
1032 hs = cmd;
1033 }
1034 /* admin statuses */
1035 else if (strcasecmp(cmd, "ready") == 0) {
1036 as = cmd;
1037 }
1038 else if (strcasecmp(cmd, "drain") == 0) {
1039 as = cmd;
1040 }
1041 else if (strcasecmp(cmd, "maint") == 0) {
1042 as = cmd;
1043 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001044 /* try to parse a weight here and keep the last one */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001045 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
1046 ps = cmd;
1047 }
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001048 /* try to parse a maxconn here */
1049 else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) {
1050 cs = cmd;
1051 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001052 else {
1053 /* keep a copy of the first error */
1054 if (!err)
1055 err = cmd;
1056 }
1057 /* skip to next word */
1058 cmd = p;
1059 }
1060 /* here, cmd points either to \0 or to the beginning of a
1061 * description. Skip possible leading spaces.
1062 */
1063 while (*cmd == ' ' || *cmd == '\n')
1064 cmd++;
1065
1066 /* First, update the admin status so that we avoid sending other
1067 * possibly useless warnings and can also update the health if
1068 * present after going back up.
1069 */
1070 if (as) {
1071 if (strcasecmp(as, "drain") == 0)
1072 srv_adm_set_drain(check->server);
1073 else if (strcasecmp(as, "maint") == 0)
1074 srv_adm_set_maint(check->server);
1075 else
1076 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001077 }
1078
Willy Tarreau81f5d942013-12-09 20:51:51 +01001079 /* now change weights */
1080 if (ps) {
1081 const char *msg;
1082
1083 msg = server_parse_weight_change_request(s, ps);
1084 if (!wrn || !*wrn)
1085 wrn = msg;
1086 }
1087
Nenad Merdanovic174dd372016-04-24 23:10:06 +02001088 if (cs) {
1089 const char *msg;
1090
1091 cs += strlen("maxconn:");
1092
1093 msg = server_parse_maxconn_change_request(s, cs);
1094 if (!wrn || !*wrn)
1095 wrn = msg;
1096 }
1097
Willy Tarreau81f5d942013-12-09 20:51:51 +01001098 /* and finally health status */
1099 if (hs) {
1100 /* We'll report some of the warnings and errors we have
1101 * here. Down reports are critical, we leave them untouched.
1102 * Lack of report, or report of 'UP' leaves the room for
1103 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001104 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001105 const char *msg = cmd;
1106 struct chunk *t;
1107
1108 if (!*msg || status == HCHK_STATUS_L7OKD) {
1109 if (err && *err)
1110 msg = err;
1111 else if (wrn && *wrn)
1112 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001113 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001114
1115 t = get_trash_chunk();
1116 chunk_printf(t, "via agent : %s%s%s%s",
1117 hs, *msg ? " (" : "",
1118 msg, *msg ? ")" : "");
1119
1120 set_server_check_status(check, status, t->str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001121 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001122 else if (err && *err) {
1123 /* No status change but we'd like to report something odd.
1124 * Just report the current state and copy the message.
1125 */
1126 chunk_printf(&trash, "agent reports an error : %s", err);
1127 set_server_check_status(check, status/*check->status*/, trash.str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001128
Willy Tarreau81f5d942013-12-09 20:51:51 +01001129 }
1130 else if (wrn && *wrn) {
1131 /* No status change but we'd like to report something odd.
1132 * Just report the current state and copy the message.
1133 */
1134 chunk_printf(&trash, "agent warns : %s", wrn);
1135 set_server_check_status(check, status/*check->status*/, trash.str);
1136 }
1137 else
1138 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001139 break;
1140 }
1141
Willy Tarreau1620ec32011-08-06 17:05:02 +02001142 case PR_O2_PGSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001143 if (!done && check->bi->i < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001144 goto wait_more_data;
1145
Simon Horman4a741432013-02-23 15:35:38 +09001146 if (check->bi->data[0] == 'R') {
1147 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001148 }
1149 else {
Simon Horman4a741432013-02-23 15:35:38 +09001150 if ((check->bi->data[0] == 'E') && (check->bi->data[5]!=0) && (check->bi->data[6]!=0))
1151 desc = &check->bi->data[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001152 else
1153 desc = "PostgreSQL unknown error";
1154
Simon Horman4a741432013-02-23 15:35:38 +09001155 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001156 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001157 break;
1158
1159 case PR_O2_REDIS_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001160 if (!done && check->bi->i < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001161 goto wait_more_data;
1162
Simon Horman4a741432013-02-23 15:35:38 +09001163 if (strcmp(check->bi->data, "+PONG\r\n") == 0) {
1164 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001165 }
1166 else {
Simon Horman4a741432013-02-23 15:35:38 +09001167 set_server_check_status(check, HCHK_STATUS_L7STS, check->bi->data);
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001168 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001169 break;
1170
1171 case PR_O2_MYSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001172 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001173 goto wait_more_data;
1174
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001175 if (s->proxy->check_len == 0) { // old mode
Simon Horman4a741432013-02-23 15:35:38 +09001176 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001177 /* We set the MySQL Version in description for information purpose
1178 * FIXME : it can be cool to use MySQL Version for other purpose,
1179 * like mark as down old MySQL server.
1180 */
Simon Horman4a741432013-02-23 15:35:38 +09001181 if (check->bi->i > 51) {
1182 desc = ltrim(check->bi->data + 5, ' ');
1183 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001184 }
1185 else {
1186 if (!done)
1187 goto wait_more_data;
1188 /* it seems we have a OK packet but without a valid length,
1189 * it must be a protocol error
1190 */
Simon Horman4a741432013-02-23 15:35:38 +09001191 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001192 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001193 }
1194 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001195 /* An error message is attached in the Error packet */
Simon Horman4a741432013-02-23 15:35:38 +09001196 desc = ltrim(check->bi->data + 7, ' ');
1197 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001198 }
1199 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001200 unsigned int first_packet_len = ((unsigned int) *check->bi->data) +
1201 (((unsigned int) *(check->bi->data + 1)) << 8) +
1202 (((unsigned int) *(check->bi->data + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001203
Simon Horman4a741432013-02-23 15:35:38 +09001204 if (check->bi->i == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001205 /* MySQL Error packet always begin with field_count = 0xff */
Simon Horman4a741432013-02-23 15:35:38 +09001206 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001207 /* We have only one MySQL packet and it is a Handshake Initialization packet
1208 * but we need to have a second packet to know if it is alright
1209 */
Simon Horman4a741432013-02-23 15:35:38 +09001210 if (!done && check->bi->i < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001211 goto wait_more_data;
1212 }
1213 else {
1214 /* We have only one packet and it is an Error packet,
1215 * an error message is attached, so we can display it
1216 */
Simon Horman4a741432013-02-23 15:35:38 +09001217 desc = &check->bi->data[7];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001218 //Warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001219 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001220 }
Simon Horman4a741432013-02-23 15:35:38 +09001221 } else if (check->bi->i > first_packet_len + 4) {
1222 unsigned int second_packet_len = ((unsigned int) *(check->bi->data + first_packet_len + 4)) +
1223 (((unsigned int) *(check->bi->data + first_packet_len + 5)) << 8) +
1224 (((unsigned int) *(check->bi->data + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001225
Simon Horman4a741432013-02-23 15:35:38 +09001226 if (check->bi->i == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001227 /* We have 2 packets and that's good */
1228 /* Check if the second packet is a MySQL Error packet or not */
Simon Horman4a741432013-02-23 15:35:38 +09001229 if (*(check->bi->data + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001230 /* No error packet */
1231 /* We set the MySQL Version in description for information purpose */
Simon Horman4a741432013-02-23 15:35:38 +09001232 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001233 //Warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001234 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001235 }
1236 else {
1237 /* An error message is attached in the Error packet
1238 * so we can display it ! :)
1239 */
Simon Horman4a741432013-02-23 15:35:38 +09001240 desc = &check->bi->data[first_packet_len+11];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001241 //Warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001242 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001243 }
1244 }
1245 }
1246 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001247 if (!done)
1248 goto wait_more_data;
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001249 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001250 * it must be a protocol error
1251 */
Simon Horman4a741432013-02-23 15:35:38 +09001252 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001253 //Warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001254 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001255 }
1256 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001257 break;
1258
1259 case PR_O2_LDAP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001260 if (!done && check->bi->i < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001261 goto wait_more_data;
1262
1263 /* Check if the server speaks LDAP (ASN.1/BER)
1264 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1265 * http://tools.ietf.org/html/rfc4511
1266 */
1267
1268 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1269 * LDAPMessage: 0x30: SEQUENCE
1270 */
Simon Horman4a741432013-02-23 15:35:38 +09001271 if ((check->bi->i < 14) || (*(check->bi->data) != '\x30')) {
1272 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001273 }
1274 else {
1275 /* size of LDAPMessage */
Simon Horman4a741432013-02-23 15:35:38 +09001276 msglen = (*(check->bi->data + 1) & 0x80) ? (*(check->bi->data + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001277
1278 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1279 * messageID: 0x02 0x01 0x01: INTEGER 1
1280 * protocolOp: 0x61: bindResponse
1281 */
1282 if ((msglen > 2) ||
Simon Horman4a741432013-02-23 15:35:38 +09001283 (memcmp(check->bi->data + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
1284 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001285
1286 goto out_wakeup;
1287 }
1288
1289 /* size of bindResponse */
Simon Horman4a741432013-02-23 15:35:38 +09001290 msglen += (*(check->bi->data + msglen + 6) & 0x80) ? (*(check->bi->data + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001291
1292 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1293 * ldapResult: 0x0a 0x01: ENUMERATION
1294 */
1295 if ((msglen > 4) ||
Simon Horman4a741432013-02-23 15:35:38 +09001296 (memcmp(check->bi->data + 7 + msglen, "\x0a\x01", 2) != 0)) {
1297 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001298
1299 goto out_wakeup;
1300 }
1301
1302 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1303 * resultCode
1304 */
Simon Horman4a741432013-02-23 15:35:38 +09001305 check->code = *(check->bi->data + msglen + 9);
1306 if (check->code) {
1307 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 +02001308 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001309 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001310 }
1311 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001312 break;
1313
1314 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001315 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001316 break;
1317 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001318
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001319 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001320 /* collect possible new errors */
1321 if (conn->flags & CO_FL_ERROR)
1322 chk_report_conn_err(conn, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001323
Nick Chalk57b1bf72010-03-16 15:50:46 +00001324 /* Reset the check buffer... */
Simon Horman4a741432013-02-23 15:35:38 +09001325 *check->bi->data = '\0';
1326 check->bi->i = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +00001327
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001328 /* Close the connection... We absolutely want to perform a hard close
1329 * and reset the connection if some data are pending, otherwise we end
1330 * up with many TIME_WAITs and eat all the source port range quickly.
1331 * To avoid sending RSTs all the time, we first try to drain pending
1332 * data.
1333 */
Willy Tarreaub4017d02015-03-12 23:11:26 +01001334 __conn_data_stop_both(conn);
1335 conn_data_shutw_hard(conn);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001336
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001337 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001338 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001339 conn->flags |= CO_FL_ERROR;
1340
Willy Tarreaufdccded2008-08-29 18:19:04 +02001341 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau3267d362012-08-17 23:53:56 +02001342 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001343
1344 wait_more_data:
Willy Tarreauf817e9f2014-01-10 16:58:45 +01001345 __conn_data_want_recv(conn);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001346}
1347
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001348/*
1349 * This function is used only for server health-checks. It handles connection
1350 * status updates including errors. If necessary, it wakes the check task up.
1351 * It always returns 0.
1352 */
1353static int wake_srv_chk(struct connection *conn)
Willy Tarreau20bea422012-07-06 12:00:49 +02001354{
Simon Horman4a741432013-02-23 15:35:38 +09001355 struct check *check = conn->owner;
Willy Tarreau20bea422012-07-06 12:00:49 +02001356
Willy Tarreau6c560da2012-11-24 11:14:45 +01001357 if (unlikely(conn->flags & CO_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001358 /* We may get error reports bypassing the I/O handlers, typically
1359 * the case when sending a pure TCP check which fails, then the I/O
1360 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001361 * main processing task so let's simply wake it up. If we get here,
1362 * we expect errno to still be valid.
1363 */
1364 chk_report_conn_err(conn, errno, 0);
1365
Willy Tarreau2d351b62013-12-05 02:36:25 +01001366 __conn_data_stop_both(conn);
1367 task_wakeup(check->task, TASK_WOKEN_IO);
1368 }
Willy Tarreau3be293f2014-02-05 18:31:24 +01001369 else if (!(conn->flags & (CO_FL_DATA_RD_ENA|CO_FL_DATA_WR_ENA|CO_FL_HANDSHAKE))) {
1370 /* we may get here if only a connection probe was required : we
1371 * don't have any data to send nor anything expected in response,
1372 * so the completion of the connection establishment is enough.
1373 */
1374 task_wakeup(check->task, TASK_WOKEN_IO);
1375 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001376
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001377 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001378 /* We're here because nobody wants to handle the error, so we
1379 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001380 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01001381 conn_sock_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001382 conn_force_close(conn);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001383 }
Willy Tarreau3267d362012-08-17 23:53:56 +02001384 return 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001385}
1386
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001387struct data_cb check_conn_cb = {
1388 .recv = event_srv_chk_r,
1389 .send = event_srv_chk_w,
1390 .wake = wake_srv_chk,
1391};
1392
Willy Tarreaubaaee002006-06-26 02:48:02 +02001393/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001394 * updates the server's weight during a warmup stage. Once the final weight is
1395 * reached, the task automatically stops. Note that any server status change
1396 * must have updated s->last_change accordingly.
1397 */
1398static struct task *server_warmup(struct task *t)
1399{
1400 struct server *s = t->context;
1401
1402 /* by default, plan on stopping the task */
1403 t->expire = TICK_ETERNITY;
Willy Tarreau20125212014-05-13 19:44:56 +02001404 if ((s->admin & SRV_ADMF_MAINT) ||
Willy Tarreau892337c2014-05-13 23:41:20 +02001405 (s->state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001406 return t;
1407
Willy Tarreau892337c2014-05-13 23:41:20 +02001408 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001409 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001410
1411 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001412 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001413
1414 /* get back there in 1 second or 1/20th of the slowstart interval,
1415 * whichever is greater, resulting in small 5% steps.
1416 */
Willy Tarreau892337c2014-05-13 23:41:20 +02001417 if (s->state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001418 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1419 return t;
1420}
1421
1422/*
Simon Horman98637e52014-06-20 12:30:16 +09001423 * establish a server health-check that makes use of a connection.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001424 *
1425 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001426 * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called
1427 * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called
1428 * - SF_ERR_SRVTO if there are no more servers
1429 * - SF_ERR_SRVCL if the connection was refused by the server
1430 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1431 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1432 * - SF_ERR_INTERNAL for any other purely internal errors
1433 * Additionnally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Hormanb00d17a2014-06-13 16:18:16 +09001434 * Note that we try to prevent the network stack from sending the ACK during the
1435 * connect() when a pure TCP check is used (without PROXY protocol).
1436 */
Simon Horman98637e52014-06-20 12:30:16 +09001437static int connect_conn_chk(struct task *t)
Simon Hormanb00d17a2014-06-13 16:18:16 +09001438{
1439 struct check *check = t->context;
1440 struct server *s = check->server;
1441 struct connection *conn = check->conn;
1442 struct protocol *proto;
1443 int ret;
Willy Tarreauf3d34822014-12-08 12:11:28 +01001444 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001445
1446 /* tcpcheck send/expect initialisation */
1447 if (check->type == PR_O2_TCPCHK_CHK)
1448 check->current_step = NULL;
1449
1450 /* prepare the check buffer.
1451 * This should not be used if check is the secondary agent check
1452 * of a server as s->proxy->check_req will relate to the
1453 * configuration of the primary check. Similarly, tcp-check uses
1454 * its own strings.
1455 */
1456 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
1457 bo_putblk(check->bo, s->proxy->check_req, s->proxy->check_len);
1458
1459 /* we want to check if this host replies to HTTP or SSLv3 requests
1460 * so we'll send the request, and won't wake the checker up now.
1461 */
1462 if ((check->type) == PR_O2_SSL3_CHK) {
1463 /* SSL requires that we put Unix time in the request */
1464 int gmt_time = htonl(date.tv_sec);
1465 memcpy(check->bo->data + 11, &gmt_time, 4);
1466 }
1467 else if ((check->type) == PR_O2_HTTP_CHK) {
1468 if (s->proxy->options2 & PR_O2_CHK_SNDST)
1469 bo_putblk(check->bo, trash.str, httpchk_build_status_header(s, trash.str, trash.size));
Cyril Bonté32602d22015-01-30 00:07:07 +01001470 /* prevent HTTP keep-alive when "http-check expect" is used */
1471 if (s->proxy->options2 & PR_O2_EXP_TYPE)
1472 bo_putstr(check->bo, "Connection: close\r\n");
Simon Hormanb00d17a2014-06-13 16:18:16 +09001473 bo_putstr(check->bo, "\r\n");
1474 *check->bo->p = '\0'; /* to make gdb output easier to read */
1475 }
1476 }
1477
James Brown55f9ff12015-10-21 18:19:05 -07001478 if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) {
1479 bo_putblk(check->bo, check->send_string, check->send_string_len);
1480 }
1481
Simon Hormanb00d17a2014-06-13 16:18:16 +09001482 /* prepare a new connection */
1483 conn_init(conn);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001484
Simon Horman41f58762015-01-30 11:22:56 +09001485 if (is_addr(&check->addr)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001486 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09001487 conn->addr.to = check->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001488 }
1489 else {
1490 /* we'll connect to the addr on the server */
1491 conn->addr.to = s->addr;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001492 }
1493
1494 if (check->port) {
1495 set_host_port(&conn->addr.to, check->port);
1496 }
1497
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01001498 proto = protocol_by_family(conn->addr.to.ss_family);
1499
1500 conn_prepare(conn, proto, check->xprt);
1501 conn_attach(conn, check, &check_conn_cb);
1502 conn->target = &s->obj_type;
1503
1504 /* no client address */
1505 clear_addr(&conn->addr.from);
1506
Willy Tarreauf3d34822014-12-08 12:11:28 +01001507 /* only plain tcp-check supports quick ACK */
1508 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1509
Simon Hormane16c1b32015-01-30 11:22:57 +09001510 if (check->type == PR_O2_TCPCHK_CHK && !LIST_ISEMPTY(check->tcpcheck_rules)) {
Willy Tarreau5581c272015-05-13 12:24:53 +02001511 struct tcpcheck_rule *r;
1512
1513 r = LIST_NEXT(check->tcpcheck_rules, struct tcpcheck_rule *, list);
1514
Simon Hormanb00d17a2014-06-13 16:18:16 +09001515 /* if first step is a 'connect', then tcpcheck_main must run it */
1516 if (r->action == TCPCHK_ACT_CONNECT) {
1517 tcpcheck_main(conn);
Willy Tarreaue7dff022015-04-03 01:14:29 +02001518 return SF_ERR_UP;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001519 }
Willy Tarreauf3d34822014-12-08 12:11:28 +01001520 if (r->action == TCPCHK_ACT_EXPECT)
1521 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001522 }
1523
Willy Tarreaue7dff022015-04-03 01:14:29 +02001524 ret = SF_ERR_INTERNAL;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001525 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01001526 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001527 conn->flags |= CO_FL_WAKE_DATA;
1528 if (s->check.send_proxy) {
1529 conn->send_proxy_ofs = 1;
1530 conn->flags |= CO_FL_SEND_PROXY;
1531 }
1532
1533 return ret;
1534}
1535
Simon Horman98637e52014-06-20 12:30:16 +09001536static struct list pid_list = LIST_HEAD_INIT(pid_list);
1537static struct pool_head *pool2_pid_list;
1538
1539void block_sigchld(void)
1540{
1541 sigset_t set;
1542 sigemptyset(&set);
1543 sigaddset(&set, SIGCHLD);
Willy Tarreauebc92442016-06-21 17:29:46 +02001544 assert(sigprocmask(SIG_BLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001545}
1546
1547void unblock_sigchld(void)
1548{
1549 sigset_t set;
1550 sigemptyset(&set);
Willy Tarreauebc92442016-06-21 17:29:46 +02001551 sigaddset(&set, SIGCHLD);
1552 assert(sigprocmask(SIG_UNBLOCK, &set, NULL) == 0);
Simon Horman98637e52014-06-20 12:30:16 +09001553}
1554
Simon Horman98637e52014-06-20 12:30:16 +09001555static struct pid_list *pid_list_add(pid_t pid, struct task *t)
1556{
1557 struct pid_list *elem;
1558 struct check *check = t->context;
1559
1560 elem = pool_alloc2(pool2_pid_list);
1561 if (!elem)
1562 return NULL;
1563 elem->pid = pid;
1564 elem->t = t;
1565 elem->exited = 0;
1566 check->curpid = elem;
1567 LIST_INIT(&elem->list);
1568 LIST_ADD(&pid_list, &elem->list);
1569 return elem;
1570}
1571
Simon Horman98637e52014-06-20 12:30:16 +09001572static void pid_list_del(struct pid_list *elem)
1573{
1574 struct check *check;
1575
1576 if (!elem)
1577 return;
1578
Simon Horman98637e52014-06-20 12:30:16 +09001579 LIST_DEL(&elem->list);
Simon Horman98637e52014-06-20 12:30:16 +09001580 if (!elem->exited)
1581 kill(elem->pid, SIGTERM);
1582
1583 check = elem->t->context;
1584 check->curpid = NULL;
1585 pool_free2(pool2_pid_list, elem);
1586}
1587
1588/* Called from inside SIGCHLD handler, SIGCHLD is blocked */
1589static void pid_list_expire(pid_t pid, int status)
1590{
1591 struct pid_list *elem;
1592
1593 list_for_each_entry(elem, &pid_list, list) {
1594 if (elem->pid == pid) {
1595 elem->t->expire = now_ms;
1596 elem->status = status;
1597 elem->exited = 1;
Cyril Bonté9dbcfab2014-08-07 01:55:39 +02001598 task_wakeup(elem->t, TASK_WOKEN_IO);
Simon Horman98637e52014-06-20 12:30:16 +09001599 return;
1600 }
1601 }
1602}
1603
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001604static void sigchld_handler(struct sig_handler *sh)
Simon Horman98637e52014-06-20 12:30:16 +09001605{
1606 pid_t pid;
1607 int status;
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001608
Simon Horman98637e52014-06-20 12:30:16 +09001609 while ((pid = waitpid(0, &status, WNOHANG)) > 0)
1610 pid_list_expire(pid, status);
1611}
1612
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001613static int init_pid_list(void)
1614{
Simon Horman98637e52014-06-20 12:30:16 +09001615 if (pool2_pid_list != NULL)
1616 /* Nothing to do */
1617 return 0;
1618
Willy Tarreau48d6bf22016-06-21 16:27:34 +02001619 if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) {
Simon Horman98637e52014-06-20 12:30:16 +09001620 Alert("Failed to set signal handler for external health checks: %s. Aborting.\n",
1621 strerror(errno));
1622 return 1;
1623 }
1624
1625 pool2_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED);
1626 if (pool2_pid_list == NULL) {
1627 Alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n",
1628 strerror(errno));
1629 return 1;
1630 }
1631
1632 return 0;
1633}
1634
Cyril Bontéac92a062014-12-27 22:28:38 +01001635/* helper macro to set an environment variable and jump to a specific label on failure. */
1636#define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001637
1638/*
Cyril Bontéac92a062014-12-27 22:28:38 +01001639 * helper function to allocate enough memory to store an environment variable.
1640 * It will also check that the environment variable is updatable, and silently
1641 * fail if not.
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001642 */
Cyril Bontéac92a062014-12-27 22:28:38 +01001643static int extchk_setenv(struct check *check, int idx, const char *value)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001644{
1645 int len, ret;
Cyril Bontéac92a062014-12-27 22:28:38 +01001646 char *envname;
1647 int vmaxlen;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001648
Cyril Bontéac92a062014-12-27 22:28:38 +01001649 if (idx < 0 || idx >= EXTCHK_SIZE) {
1650 Alert("Illegal environment variable index %d. Aborting.\n", idx);
1651 return 1;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001652 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001653
1654 envname = extcheck_envs[idx].name;
1655 vmaxlen = extcheck_envs[idx].vmaxlen;
1656
1657 /* Check if the environment variable is already set, and silently reject
1658 * the update if this one is not updatable. */
1659 if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx]))
1660 return 0;
1661
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001662 /* Instead of sending NOT_USED, sending an empty value is preferable */
1663 if (strcmp(value, "NOT_USED") == 0) {
1664 value = "";
1665 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001666
1667 len = strlen(envname) + 1;
1668 if (vmaxlen == EXTCHK_SIZE_EVAL_INIT)
1669 len += strlen(value);
1670 else
1671 len += vmaxlen;
1672
1673 if (!check->envp[idx])
1674 check->envp[idx] = malloc(len + 1);
1675
1676 if (!check->envp[idx]) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001677 Alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname);
1678 return 1;
1679 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001680 ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001681 if (ret < 0) {
1682 Alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno));
1683 return 1;
1684 }
Cyril Bontéac92a062014-12-27 22:28:38 +01001685 else if (ret > len) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001686 Alert("Environment variable '%s' was truncated. Aborting.\n", envname);
1687 return 1;
1688 }
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001689 return 0;
1690}
Simon Horman98637e52014-06-20 12:30:16 +09001691
1692static int prepare_external_check(struct check *check)
1693{
1694 struct server *s = check->server;
1695 struct proxy *px = s->proxy;
1696 struct listener *listener = NULL, *l;
1697 int i;
Simon Horman98637e52014-06-20 12:30:16 +09001698 const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH;
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001699 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001700
1701 list_for_each_entry(l, &px->conf.listeners, by_fe)
1702 /* Use the first INET, INET6 or UNIX listener */
1703 if (l->addr.ss_family == AF_INET ||
1704 l->addr.ss_family == AF_INET6 ||
1705 l->addr.ss_family == AF_UNIX) {
1706 listener = l;
1707 break;
1708 }
1709
Simon Horman98637e52014-06-20 12:30:16 +09001710 check->curpid = NULL;
Cyril Bontéac92a062014-12-27 22:28:38 +01001711 check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *));
1712 if (!check->envp) {
1713 Alert("Failed to allocate memory for environment variables. Aborting\n");
1714 goto err;
1715 }
Simon Horman98637e52014-06-20 12:30:16 +09001716
Cyril Bontéac92a062014-12-27 22:28:38 +01001717 check->argv = calloc(6, sizeof(char *));
1718 if (!check->argv) {
1719 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001720 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001721 }
Simon Horman98637e52014-06-20 12:30:16 +09001722
1723 check->argv[0] = px->check_command;
1724
Cyril Bonté777be862014-12-02 21:21:35 +01001725 if (!listener) {
1726 check->argv[1] = strdup("NOT_USED");
1727 check->argv[2] = strdup("NOT_USED");
1728 }
1729 else if (listener->addr.ss_family == AF_INET ||
Simon Horman98637e52014-06-20 12:30:16 +09001730 listener->addr.ss_family == AF_INET6) {
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001731 addr_to_str(&listener->addr, buf, sizeof(buf));
1732 check->argv[1] = strdup(buf);
1733 port_to_str(&listener->addr, buf, sizeof(buf));
1734 check->argv[2] = strdup(buf);
Cyril Bonté777be862014-12-02 21:21:35 +01001735 }
1736 else if (listener->addr.ss_family == AF_UNIX) {
Simon Horman98637e52014-06-20 12:30:16 +09001737 const struct sockaddr_un *un;
1738
1739 un = (struct sockaddr_un *)&listener->addr;
1740 check->argv[1] = strdup(un->sun_path);
1741 check->argv[2] = strdup("NOT_USED");
Cyril Bonté777be862014-12-02 21:21:35 +01001742 }
1743 else {
Cyril Bontéac92a062014-12-27 22:28:38 +01001744 Alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001745 goto err;
1746 }
1747
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001748 addr_to_str(&s->addr, buf, sizeof(buf));
1749 check->argv[3] = strdup(buf);
1750 port_to_str(&s->addr, buf, sizeof(buf));
1751 check->argv[4] = strdup(buf);
Simon Horman98637e52014-06-20 12:30:16 +09001752
Cyril Bontéac92a062014-12-27 22:28:38 +01001753 for (i = 0; i < 5; i++) {
1754 if (!check->argv[i]) {
1755 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
Simon Horman98637e52014-06-20 12:30:16 +09001756 goto err;
Cyril Bontéac92a062014-12-27 22:28:38 +01001757 }
1758 }
Simon Horman98637e52014-06-20 12:30:16 +09001759
Cyril Bontéac92a062014-12-27 22:28:38 +01001760 EXTCHK_SETENV(check, EXTCHK_PATH, path, err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001761 /* Add proxy environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001762 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err);
1763 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err);
1764 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err);
1765 EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001766 /* Add server environment variables */
Cyril Bontéac92a062014-12-27 22:28:38 +01001767 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err);
1768 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err);
1769 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err);
1770 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err);
1771 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err);
1772 EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err);
1773
1774 /* Ensure that we don't leave any hole in check->envp */
1775 for (i = 0; i < EXTCHK_SIZE; i++)
1776 if (!check->envp[i])
1777 EXTCHK_SETENV(check, i, "", err);
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001778
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001779 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09001780err:
1781 if (check->envp) {
Cyril Bontéac92a062014-12-27 22:28:38 +01001782 for (i = 0; i < EXTCHK_SIZE; i++)
Cyril Bonté9ede66b2014-12-02 21:21:36 +01001783 free(check->envp[i]);
Simon Horman98637e52014-06-20 12:30:16 +09001784 free(check->envp);
1785 check->envp = NULL;
1786 }
1787
1788 if (check->argv) {
1789 for (i = 1; i < 5; i++)
1790 free(check->argv[i]);
1791 free(check->argv);
1792 check->argv = NULL;
1793 }
Cyril Bonté99c5bf52014-08-07 01:55:38 +02001794 return 0;
Simon Horman98637e52014-06-20 12:30:16 +09001795}
1796
Simon Hormanb00d17a2014-06-13 16:18:16 +09001797/*
Simon Horman98637e52014-06-20 12:30:16 +09001798 * establish a server health-check that makes use of a process.
1799 *
1800 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02001801 * - SF_ERR_NONE if everything's OK
1802 * - SF_ERR_SRVTO if there are no more servers
1803 * - SF_ERR_SRVCL if the connection was refused by the server
1804 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1805 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1806 * - SF_ERR_INTERNAL for any other purely internal errors
1807 * Additionnally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Simon Horman98637e52014-06-20 12:30:16 +09001808 *
1809 * Blocks and then unblocks SIGCHLD
1810 */
1811static int connect_proc_chk(struct task *t)
1812{
Cyril Bontéac92a062014-12-27 22:28:38 +01001813 char buf[256];
Simon Horman98637e52014-06-20 12:30:16 +09001814 struct check *check = t->context;
1815 struct server *s = check->server;
1816 struct proxy *px = s->proxy;
1817 int status;
1818 pid_t pid;
1819
Willy Tarreaue7dff022015-04-03 01:14:29 +02001820 status = SF_ERR_RESOURCE;
Simon Horman98637e52014-06-20 12:30:16 +09001821
1822 block_sigchld();
1823
1824 pid = fork();
1825 if (pid < 0) {
1826 Alert("Failed to fork process for external health check: %s. Aborting.\n",
1827 strerror(errno));
1828 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1829 goto out;
1830 }
1831 if (pid == 0) {
1832 /* Child */
1833 extern char **environ;
Willy Tarreaub7b24782016-06-21 15:32:29 +02001834 int fd;
1835
1836 /* close all FDs. Keep stdin/stdout/stderr in verbose mode */
1837 fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3;
1838
1839 while (fd < global.rlimit_nofile)
1840 close(fd++);
1841
Simon Horman98637e52014-06-20 12:30:16 +09001842 environ = check->envp;
Cyril Bontéac92a062014-12-27 22:28:38 +01001843 extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)));
Simon Horman98637e52014-06-20 12:30:16 +09001844 execvp(px->check_command, check->argv);
1845 Alert("Failed to exec process for external health check: %s. Aborting.\n",
1846 strerror(errno));
1847 exit(-1);
1848 }
1849
1850 /* Parent */
1851 if (check->result == CHK_RES_UNKNOWN) {
1852 if (pid_list_add(pid, t) != NULL) {
1853 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1854
1855 if (px->timeout.check && px->timeout.connect) {
1856 int t_con = tick_add(now_ms, px->timeout.connect);
1857 t->expire = tick_first(t->expire, t_con);
1858 }
Willy Tarreaue7dff022015-04-03 01:14:29 +02001859 status = SF_ERR_NONE;
Simon Horman98637e52014-06-20 12:30:16 +09001860 goto out;
1861 }
1862 else {
1863 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1864 }
1865 kill(pid, SIGTERM); /* process creation error */
1866 }
1867 else
1868 set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno));
1869
1870out:
1871 unblock_sigchld();
1872 return status;
1873}
1874
1875/*
Simon Horman98637e52014-06-20 12:30:16 +09001876 * manages a server health-check that uses a process. Returns
Willy Tarreaubaaee002006-06-26 02:48:02 +02001877 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
1878 */
Simon Horman98637e52014-06-20 12:30:16 +09001879static struct task *process_chk_proc(struct task *t)
1880{
1881 struct check *check = t->context;
1882 struct server *s = check->server;
1883 struct connection *conn = check->conn;
1884 int rv;
1885 int ret;
1886 int expired = tick_is_expired(t->expire, now_ms);
1887
1888 if (!(check->state & CHK_ST_INPROGRESS)) {
1889 /* no check currently running */
1890 if (!expired) /* woke up too early */
1891 return t;
1892
1893 /* we don't send any health-checks when the proxy is
1894 * stopped, the server should not be checked or the check
1895 * is disabled.
1896 */
1897 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
1898 s->proxy->state == PR_STSTOPPED)
1899 goto reschedule;
1900
1901 /* we'll initiate a new check */
1902 set_server_check_status(check, HCHK_STATUS_START, NULL);
1903
1904 check->state |= CHK_ST_INPROGRESS;
1905
Simon Hormandbf70192015-01-30 11:22:53 +09001906 ret = connect_proc_chk(t);
Simon Horman98637e52014-06-20 12:30:16 +09001907 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02001908 case SF_ERR_UP:
Simon Horman98637e52014-06-20 12:30:16 +09001909 return t;
Willy Tarreaue7dff022015-04-03 01:14:29 +02001910 case SF_ERR_NONE:
Simon Horman98637e52014-06-20 12:30:16 +09001911 /* we allow up to min(inter, timeout.connect) for a connection
1912 * to establish but only when timeout.check is set
1913 * as it may be to short for a full check otherwise
1914 */
1915 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1916
1917 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
1918 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
1919 t->expire = tick_first(t->expire, t_con);
1920 }
1921
1922 goto reschedule;
1923
Willy Tarreaue7dff022015-04-03 01:14:29 +02001924 case SF_ERR_SRVTO: /* ETIMEDOUT */
1925 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Simon Horman98637e52014-06-20 12:30:16 +09001926 conn->flags |= CO_FL_ERROR;
1927 chk_report_conn_err(conn, errno, 0);
1928 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02001929 case SF_ERR_PRXCOND:
1930 case SF_ERR_RESOURCE:
1931 case SF_ERR_INTERNAL:
Simon Horman98637e52014-06-20 12:30:16 +09001932 conn->flags |= CO_FL_ERROR;
1933 chk_report_conn_err(conn, 0, 0);
1934 break;
1935 }
1936
1937 /* here, we have seen a synchronous error, no fd was allocated */
1938
1939 check->state &= ~CHK_ST_INPROGRESS;
1940 check_notify_failure(check);
1941
1942 /* we allow up to min(inter, timeout.connect) for a connection
1943 * to establish but only when timeout.check is set
1944 * as it may be to short for a full check otherwise
1945 */
1946 while (tick_is_expired(t->expire, now_ms)) {
1947 int t_con;
1948
1949 t_con = tick_add(t->expire, s->proxy->timeout.connect);
1950 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
1951
1952 if (s->proxy->timeout.check)
1953 t->expire = tick_first(t->expire, t_con);
1954 }
1955 }
1956 else {
1957 /* there was a test running.
1958 * First, let's check whether there was an uncaught error,
1959 * which can happen on connect timeout or error.
1960 */
1961 if (check->result == CHK_RES_UNKNOWN) {
1962 /* good connection is enough for pure TCP check */
1963 struct pid_list *elem = check->curpid;
1964 int status = HCHK_STATUS_UNKNOWN;
1965
1966 if (elem->exited) {
1967 status = elem->status; /* Save in case the process exits between use below */
1968 if (!WIFEXITED(status))
1969 check->code = -1;
1970 else
1971 check->code = WEXITSTATUS(status);
1972 if (!WIFEXITED(status) || WEXITSTATUS(status))
1973 status = HCHK_STATUS_PROCERR;
1974 else
1975 status = HCHK_STATUS_PROCOK;
1976 } else if (expired) {
1977 status = HCHK_STATUS_PROCTOUT;
Willy Tarreaudc3d1902014-07-08 00:56:27 +02001978 Warning("kill %d\n", (int)elem->pid);
Simon Horman98637e52014-06-20 12:30:16 +09001979 kill(elem->pid, SIGTERM);
1980 }
1981 set_server_check_status(check, status, NULL);
1982 }
1983
1984 if (check->result == CHK_RES_FAILED) {
1985 /* a failure or timeout detected */
1986 check_notify_failure(check);
1987 }
1988 else if (check->result == CHK_RES_CONDPASS) {
1989 /* check is OK but asks for stopping mode */
1990 check_notify_stopping(check);
1991 }
1992 else if (check->result == CHK_RES_PASSED) {
1993 /* a success was detected */
1994 check_notify_success(check);
1995 }
1996 check->state &= ~CHK_ST_INPROGRESS;
1997
1998 pid_list_del(check->curpid);
1999
2000 rv = 0;
2001 if (global.spread_checks > 0) {
2002 rv = srv_getinter(check) * global.spread_checks / 100;
2003 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
2004 }
2005 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
2006 }
2007
2008 reschedule:
2009 while (tick_is_expired(t->expire, now_ms))
2010 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2011 return t;
2012}
2013
2014/*
2015 * manages a server health-check that uses a connection. Returns
2016 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2017 */
2018static struct task *process_chk_conn(struct task *t)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002019{
Simon Horman4a741432013-02-23 15:35:38 +09002020 struct check *check = t->context;
2021 struct server *s = check->server;
2022 struct connection *conn = check->conn;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002023 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002024 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002025 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002026
Willy Tarreau2c115e52013-12-11 19:41:16 +01002027 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01002028 /* no check currently running */
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002029 if (!expired) /* woke up too early */
Willy Tarreau26c25062009-03-08 09:38:41 +01002030 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002031
Simon Horman671b6f02013-11-25 10:46:39 +09002032 /* we don't send any health-checks when the proxy is
2033 * stopped, the server should not be checked or the check
2034 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002035 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01002036 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01002037 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01002038 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002039
2040 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09002041 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002042
Willy Tarreau2c115e52013-12-11 19:41:16 +01002043 check->state |= CHK_ST_INPROGRESS;
Simon Horman4a741432013-02-23 15:35:38 +09002044 check->bi->p = check->bi->data;
2045 check->bi->i = 0;
2046 check->bo->p = check->bo->data;
2047 check->bo->o = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002048
Simon Hormandbf70192015-01-30 11:22:53 +09002049 ret = connect_conn_chk(t);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002050 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002051 case SF_ERR_UP:
Simon Hormanb00d17a2014-06-13 16:18:16 +09002052 return t;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002053 case SF_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002054 /* we allow up to min(inter, timeout.connect) for a connection
2055 * to establish but only when timeout.check is set
2056 * as it may be to short for a full check otherwise
2057 */
Simon Horman4a741432013-02-23 15:35:38 +09002058 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02002059
Willy Tarreaufb56aab2012-09-28 14:40:02 +02002060 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2061 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2062 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002063 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01002064
2065 if (check->type)
2066 conn_data_want_recv(conn); /* prepare for reading a possible reply */
2067
Willy Tarreau5a78f362012-11-23 12:47:05 +01002068 goto reschedule;
2069
Willy Tarreaue7dff022015-04-03 01:14:29 +02002070 case SF_ERR_SRVTO: /* ETIMEDOUT */
2071 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau4bd07de2014-01-24 16:10:57 +01002072 conn->flags |= CO_FL_ERROR;
2073 chk_report_conn_err(conn, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002074 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002075 case SF_ERR_PRXCOND:
2076 case SF_ERR_RESOURCE:
2077 case SF_ERR_INTERNAL:
Willy Tarreau4bd07de2014-01-24 16:10:57 +01002078 conn->flags |= CO_FL_ERROR;
2079 chk_report_conn_err(conn, 0, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01002080 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002081 }
2082
Willy Tarreau5a78f362012-11-23 12:47:05 +01002083 /* here, we have seen a synchronous error, no fd was allocated */
Willy Tarreau6b0a8502012-11-23 08:51:32 +01002084
Willy Tarreau2c115e52013-12-11 19:41:16 +01002085 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02002086 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002087
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002088 /* we allow up to min(inter, timeout.connect) for a connection
2089 * to establish but only when timeout.check is set
2090 * as it may be to short for a full check otherwise
2091 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002092 while (tick_is_expired(t->expire, now_ms)) {
2093 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002094
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002095 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09002096 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002097
Willy Tarreau0c303ee2008-07-07 00:09:58 +02002098 if (s->proxy->timeout.check)
2099 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01002100 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002101 }
2102 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02002103 /* there was a test running.
2104 * First, let's check whether there was an uncaught error,
2105 * which can happen on connect timeout or error.
2106 */
Simon Hormanccaabcd2014-06-20 12:29:47 +09002107 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002108 /* good connection is enough for pure TCP check */
2109 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09002110 if (check->use_ssl)
2111 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02002112 else
Simon Horman4a741432013-02-23 15:35:38 +09002113 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01002114 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +01002115 else if ((conn->flags & CO_FL_ERROR) || expired) {
2116 chk_report_conn_err(conn, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02002117 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002118 else
2119 goto out_wait; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02002120 }
2121
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002122 /* check complete or aborted */
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002123 if (conn->xprt) {
2124 /* The check was aborted and the connection was not yet closed.
2125 * This can happen upon timeout, or when an external event such
2126 * as a failed response coupled with "observe layer7" caused the
2127 * server state to be suddenly changed.
2128 */
Willy Tarreaud85c4852015-03-13 00:40:28 +01002129 conn_sock_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02002130 conn_force_close(conn);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01002131 }
2132
Willy Tarreauaf549582014-05-16 17:37:50 +02002133 if (check->result == CHK_RES_FAILED) {
2134 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02002135 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002136 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02002137 else if (check->result == CHK_RES_CONDPASS) {
2138 /* check is OK but asks for stopping mode */
2139 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02002140 }
Willy Tarreau3e048382014-05-21 10:30:54 +02002141 else if (check->result == CHK_RES_PASSED) {
2142 /* a success was detected */
2143 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002144 }
Willy Tarreau2c115e52013-12-11 19:41:16 +01002145 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002146
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002147 rv = 0;
2148 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09002149 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002150 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002151 }
Simon Horman4a741432013-02-23 15:35:38 +09002152 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02002153 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01002154
2155 reschedule:
2156 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09002157 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01002158 out_wait:
Willy Tarreau26c25062009-03-08 09:38:41 +01002159 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002160}
2161
Simon Horman98637e52014-06-20 12:30:16 +09002162/*
2163 * manages a server health-check. Returns
2164 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
2165 */
2166static struct task *process_chk(struct task *t)
2167{
2168 struct check *check = t->context;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002169 struct server *s = check->server;
2170 struct dns_resolution *resolution = s->resolution;
2171
2172 /* trigger name resolution */
2173 if ((s->check.state & CHK_ST_ENABLED) && (resolution)) {
2174 /* check if a no resolution is running for this server */
2175 if (resolution->step == RSLV_STEP_NONE) {
2176 /*
2177 * if there has not been any name resolution for a longer period than
2178 * hold.valid, let's trigger a new one.
2179 */
Baptiste Assmannf0d93702015-09-03 10:59:39 +02002180 if (!resolution->last_resolution || tick_is_expired(tick_add(resolution->last_resolution, resolution->resolvers->hold.valid), now_ms)) {
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002181 trigger_resolution(s);
2182 }
2183 }
2184 }
Simon Horman98637e52014-06-20 12:30:16 +09002185
2186 if (check->type == PR_O2_EXT_CHK)
2187 return process_chk_proc(t);
2188 return process_chk_conn(t);
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002189
2190}
2191
2192/*
2193 * Initiates a new name resolution:
2194 * - generates a query id
2195 * - configure the resolution structure
2196 * - startup the resolvers task if required
2197 *
2198 * returns:
2199 * - 0 in case of error or if resolution already running
2200 * - 1 if everything started properly
2201 */
2202int trigger_resolution(struct server *s)
2203{
2204 struct dns_resolution *resolution;
2205 struct dns_resolvers *resolvers;
2206 int query_id;
2207 int i;
2208
2209 resolution = s->resolution;
2210 resolvers = resolution->resolvers;
2211
2212 /*
2213 * check if a resolution has already been started for this server
2214 * return directly to avoid resolution pill up
2215 */
2216 if (resolution->step != RSLV_STEP_NONE)
2217 return 0;
2218
2219 /* generates a query id */
2220 i = 0;
2221 do {
2222 query_id = dns_rnd16();
2223 /* we do try only 100 times to find a free query id */
2224 if (i++ > 100) {
2225 chunk_printf(&trash, "could not generate a query id for %s/%s, in resolvers %s",
2226 s->proxy->id, s->id, resolvers->id);
2227
2228 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
2229 return 0;
2230 }
2231 } while (eb32_lookup(&resolvers->query_ids, query_id));
2232
2233 LIST_ADDQ(&resolvers->curr_resolution, &resolution->list);
2234
2235 /* now update resolution parameters */
2236 resolution->query_id = query_id;
2237 resolution->qid.key = query_id;
2238 resolution->step = RSLV_STEP_RUNNING;
Thierry Fournierada34842016-02-17 21:25:09 +01002239 resolution->opts = &s->dns_opts;
2240 if (resolution->opts->family_prio == AF_INET) {
Andrew Hayworthe6a4a322015-10-19 22:29:51 +00002241 resolution->query_type = DNS_RTYPE_A;
2242 } else {
2243 resolution->query_type = DNS_RTYPE_AAAA;
2244 }
Baptiste Assmannf778bb42015-09-09 00:54:38 +02002245 resolution->try = resolvers->resolve_retries;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002246 resolution->try_cname = 0;
2247 resolution->nb_responses = 0;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002248 eb32_insert(&resolvers->query_ids, &resolution->qid);
2249
2250 dns_send_query(resolution);
Baptiste Assmannf778bb42015-09-09 00:54:38 +02002251 resolution->try -= 1;
Baptiste Assmanna68ca962015-04-14 01:15:08 +02002252
2253 /* update wakeup date if this resolution is the only one in the FIFO list */
2254 if (dns_check_resolution_queue(resolvers) == 1) {
2255 /* update task timeout */
2256 dns_update_resolvers_timeout(resolvers);
2257 task_queue(resolvers->t);
2258 }
2259
2260 return 1;
Simon Horman98637e52014-06-20 12:30:16 +09002261}
2262
Simon Horman5c942422013-11-25 10:46:32 +09002263static int start_check_task(struct check *check, int mininter,
2264 int nbcheck, int srvpos)
2265{
2266 struct task *t;
2267 /* task for the check */
2268 if ((t = task_new()) == NULL) {
2269 Alert("Starting [%s:%s] check: out of memory.\n",
2270 check->server->proxy->id, check->server->id);
2271 return 0;
2272 }
2273
2274 check->task = t;
2275 t->process = process_chk;
2276 t->context = check;
2277
Willy Tarreau1746eec2014-04-25 10:46:47 +02002278 if (mininter < srv_getinter(check))
2279 mininter = srv_getinter(check);
2280
2281 if (global.max_spread_checks && mininter > global.max_spread_checks)
2282 mininter = global.max_spread_checks;
2283
Simon Horman5c942422013-11-25 10:46:32 +09002284 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02002285 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09002286 check->start = now;
2287 task_queue(t);
2288
2289 return 1;
2290}
2291
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002292/*
2293 * Start health-check.
2294 * Returns 0 if OK, -1 if error, and prints the error in this case.
2295 */
2296int start_checks() {
2297
2298 struct proxy *px;
2299 struct server *s;
2300 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09002301 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002302
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002303 /* 1- count the checkers to run simultaneously.
2304 * We also determine the minimum interval among all of those which
2305 * have an interval larger than SRV_CHK_INTER_THRES. This interval
2306 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04002307 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02002308 * too short an interval for all others.
2309 */
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002310 for (px = proxy; px; px = px->next) {
2311 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01002312 if (s->slowstart) {
2313 if ((t = task_new()) == NULL) {
2314 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
2315 return -1;
2316 }
2317 /* We need a warmup task that will be called when the server
2318 * state switches from down to up.
2319 */
2320 s->warmup = t;
2321 t->process = server_warmup;
2322 t->context = s;
2323 t->expire = TICK_ETERNITY;
Baptiste Assmann6076d1c2015-09-17 22:53:59 +02002324 /* server can be in this state only because of */
2325 if (s->state == SRV_ST_STARTING)
2326 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 +01002327 }
2328
Willy Tarreaud8514a22013-12-11 21:10:14 +01002329 if (s->check.state & CHK_ST_CONFIGURED) {
2330 nbcheck++;
2331 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
2332 (!mininter || mininter > srv_getinter(&s->check)))
2333 mininter = srv_getinter(&s->check);
2334 }
Willy Tarreau15f39102013-12-11 20:41:18 +01002335
Willy Tarreaud8514a22013-12-11 21:10:14 +01002336 if (s->agent.state & CHK_ST_CONFIGURED) {
2337 nbcheck++;
2338 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
2339 (!mininter || mininter > srv_getinter(&s->agent)))
2340 mininter = srv_getinter(&s->agent);
2341 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002342 }
2343 }
2344
Simon Horman4a741432013-02-23 15:35:38 +09002345 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002346 return 0;
2347
2348 srand((unsigned)time(NULL));
2349
2350 /*
2351 * 2- start them as far as possible from each others. For this, we will
2352 * start them after their interval set to the min interval divided by
2353 * the number of servers, weighted by the server's position in the list.
2354 */
2355 for (px = proxy; px; px = px->next) {
Simon Horman98637e52014-06-20 12:30:16 +09002356 if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) {
2357 if (init_pid_list()) {
2358 Alert("Starting [%s] check: out of memory.\n", px->id);
2359 return -1;
2360 }
2361 }
2362
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002363 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09002364 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002365 if (s->check.state & CHK_ST_CONFIGURED) {
Cyril Bonté99c5bf52014-08-07 01:55:38 +02002366 if (s->check.type == PR_O2_EXT_CHK) {
2367 if (!prepare_external_check(&s->check))
2368 return -1;
2369 }
Simon Hormand60d6912013-11-25 10:46:36 +09002370 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
2371 return -1;
2372 srvpos++;
2373 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002374
Simon Hormand60d6912013-11-25 10:46:36 +09002375 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01002376 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09002377 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
2378 return -1;
2379 }
2380 srvpos++;
2381 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02002382 }
2383 }
2384 return 0;
2385}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002386
2387/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02002388 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01002389 * The buffer MUST be terminated by a null byte before calling this function.
2390 * Sets server status appropriately. The caller is responsible for ensuring
2391 * that the buffer contains at least 13 characters. If <done> is zero, we may
2392 * return 0 to indicate that data is required to decide of a match.
2393 */
2394static int httpchk_expect(struct server *s, int done)
2395{
2396 static char status_msg[] = "HTTP status check returned code <000>";
2397 char status_code[] = "000";
2398 char *contentptr;
2399 int crlf;
2400 int ret;
2401
2402 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
2403 case PR_O2_EXP_STS:
2404 case PR_O2_EXP_RSTS:
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002405 memcpy(status_code, s->check.bi->data + 9, 3);
2406 memcpy(status_msg + strlen(status_msg) - 4, s->check.bi->data + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01002407
2408 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
2409 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
2410 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002411 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01002412
2413 /* we necessarily have the response, so there are no partial failures */
2414 if (s->proxy->options2 & PR_O2_EXP_INV)
2415 ret = !ret;
2416
Simon Horman4a741432013-02-23 15:35:38 +09002417 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01002418 break;
2419
2420 case PR_O2_EXP_STR:
2421 case PR_O2_EXP_RSTR:
2422 /* very simple response parser: ignore CR and only count consecutive LFs,
2423 * stop with contentptr pointing to first char after the double CRLF or
2424 * to '\0' if crlf < 2.
2425 */
2426 crlf = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02002427 for (contentptr = s->check.bi->data; *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01002428 if (crlf >= 2)
2429 break;
2430 if (*contentptr == '\r')
2431 continue;
2432 else if (*contentptr == '\n')
2433 crlf++;
2434 else
2435 crlf = 0;
2436 }
2437
2438 /* Check that response contains a body... */
2439 if (crlf < 2) {
2440 if (!done)
2441 return 0;
2442
Simon Horman4a741432013-02-23 15:35:38 +09002443 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002444 "HTTP content check could not find a response body");
2445 return 1;
2446 }
2447
2448 /* Check that response body is not empty... */
2449 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02002450 if (!done)
2451 return 0;
2452
Simon Horman4a741432013-02-23 15:35:38 +09002453 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002454 "HTTP content check found empty response body");
2455 return 1;
2456 }
2457
2458 /* Check the response content against the supplied string
2459 * or regex... */
2460 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
2461 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
2462 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002463 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01002464
2465 /* if we don't match, we may need to wait more */
2466 if (!ret && !done)
2467 return 0;
2468
2469 if (ret) {
2470 /* content matched */
2471 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002472 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002473 "HTTP check matched unwanted content");
2474 else
Simon Horman4a741432013-02-23 15:35:38 +09002475 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002476 "HTTP content check matched");
2477 }
2478 else {
2479 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09002480 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01002481 "HTTP check did not match unwanted content");
2482 else
Simon Horman4a741432013-02-23 15:35:38 +09002483 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01002484 "HTTP content check did not match");
2485 }
2486 break;
2487 }
2488 return 1;
2489}
2490
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002491/*
2492 * return the id of a step in a send/expect session
2493 */
Simon Hormane16c1b32015-01-30 11:22:57 +09002494static int tcpcheck_get_step_id(struct check *check)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002495{
2496 struct tcpcheck_rule *cur = NULL, *next = NULL;
2497 int i = 0;
2498
Willy Tarreau213c6782014-10-02 14:51:02 +02002499 /* not even started anything yet => step 0 = initial connect */
Baptiste Assmannf95bc8e2015-04-25 16:03:06 +02002500 if (!check->current_step)
Willy Tarreau213c6782014-10-02 14:51:02 +02002501 return 0;
2502
Simon Hormane16c1b32015-01-30 11:22:57 +09002503 cur = check->last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002504
2505 /* no step => first step */
2506 if (cur == NULL)
2507 return 1;
2508
2509 /* increment i until current step */
Simon Hormane16c1b32015-01-30 11:22:57 +09002510 list_for_each_entry(next, check->tcpcheck_rules, list) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002511 if (next->list.p == &cur->list)
2512 break;
2513 ++i;
2514 }
2515
2516 return i;
2517}
2518
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002519/*
2520 * return the latest known comment before (including) the given stepid
2521 * returns NULL if no comment found
2522 */
2523static char * tcpcheck_get_step_comment(struct check *check, int stepid)
2524{
2525 struct tcpcheck_rule *cur = NULL;
2526 char *ret = NULL;
2527 int i = 0;
2528
2529 /* not even started anything yet, return latest comment found before any action */
2530 if (!check->current_step) {
2531 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2532 if (cur->action == TCPCHK_ACT_COMMENT)
2533 ret = cur->comment;
2534 else
2535 goto return_comment;
2536 }
2537 }
2538
2539 i = 1;
2540 list_for_each_entry(cur, check->tcpcheck_rules, list) {
2541 if (cur->comment)
2542 ret = cur->comment;
2543
2544 if (i >= stepid)
2545 goto return_comment;
2546
2547 ++i;
2548 }
2549
2550 return_comment:
2551 return ret;
2552}
2553
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002554static void tcpcheck_main(struct connection *conn)
2555{
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002556 char *contentptr, *comment;
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002557 struct tcpcheck_rule *next;
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002558 int done = 0, ret = 0, step = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002559 struct check *check = conn->owner;
2560 struct server *s = check->server;
2561 struct task *t = check->task;
Simon Hormane16c1b32015-01-30 11:22:57 +09002562 struct list *head = check->tcpcheck_rules;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002563
Willy Tarreauef953952014-10-02 14:30:14 +02002564 /* here, we know that the check is complete or that it failed */
2565 if (check->result != CHK_RES_UNKNOWN)
2566 goto out_end_tcpcheck;
2567
2568 /* We have 4 possibilities here :
2569 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
2570 * connection attempt was made yet ;
2571 * 2. we've not yet attempted step 1, and step 1 is a not connect or
2572 * does not exist (no rule), so a connection attempt was made
2573 * before coming here.
2574 * 3. we're coming back after having started with step 1, so we may
2575 * be waiting for a connection attempt to complete.
2576 * 4. the connection + handshake are complete
2577 *
2578 * #2 and #3 are quite similar, we want both the connection and the
2579 * handshake to complete before going any further. Thus we must always
2580 * wait for a connection to complete unless we're before and existing
2581 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002582 */
Willy Tarreau449f9522015-05-13 15:39:48 +02002583
2584 /* find first rule and skip comments */
2585 next = LIST_NEXT(head, struct tcpcheck_rule *, list);
2586 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
2587 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
2588
Willy Tarreauef953952014-10-02 14:30:14 +02002589 if ((!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE)) &&
Willy Tarreau449f9522015-05-13 15:39:48 +02002590 (check->current_step || &next->list == head)) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002591 /* we allow up to min(inter, timeout.connect) for a connection
2592 * to establish but only when timeout.check is set
2593 * as it may be to short for a full check otherwise
2594 */
2595 while (tick_is_expired(t->expire, now_ms)) {
2596 int t_con;
2597
2598 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2599 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2600
2601 if (s->proxy->timeout.check)
2602 t->expire = tick_first(t->expire, t_con);
2603 }
2604 return;
2605 }
2606
Willy Tarreauef953952014-10-02 14:30:14 +02002607 /* special case: option tcp-check with no rule, a connect is enough */
Willy Tarreau449f9522015-05-13 15:39:48 +02002608 if (&next->list == head) {
Willy Tarreauef953952014-10-02 14:30:14 +02002609 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002610 goto out_end_tcpcheck;
Willy Tarreauef953952014-10-02 14:30:14 +02002611 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002612
Willy Tarreau213c6782014-10-02 14:51:02 +02002613 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002614 if (check->current_step == NULL) {
Willy Tarreau213c6782014-10-02 14:51:02 +02002615 check->last_started_step = NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002616 check->bo->p = check->bo->data;
2617 check->bo->o = 0;
2618 check->bi->p = check->bi->data;
2619 check->bi->i = 0;
Willy Tarreau449f9522015-05-13 15:39:48 +02002620 check->current_step = next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002621 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2622 if (s->proxy->timeout.check)
2623 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
2624 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002625
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002626 /* It's only the rules which will enable send/recv */
2627 __conn_data_stop_both(conn);
2628
Willy Tarreauabca5b62013-12-06 14:19:25 +01002629 while (1) {
Willy Tarreau263013d2015-05-13 11:59:14 +02002630 /* We have to try to flush the output buffer before reading, at
2631 * the end, or if we're about to send a string that does not fit
2632 * in the remaining space. That explains why we break out of the
2633 * loop after this control.
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002634 */
2635 if (check->bo->o &&
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002636 (&check->current_step->list == head ||
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002637 check->current_step->action != TCPCHK_ACT_SEND ||
2638 check->current_step->string_len >= buffer_total_space(check->bo))) {
2639
Willy Tarreau1049b1f2014-02-02 01:51:17 +01002640 if (conn->xprt->snd_buf(conn, check->bo, 0) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002641 if (conn->flags & CO_FL_ERROR) {
2642 chk_report_conn_err(conn, errno, 0);
2643 __conn_data_stop_both(conn);
2644 goto out_end_tcpcheck;
2645 }
Willy Tarreau263013d2015-05-13 11:59:14 +02002646 break;
Willy Tarreauabca5b62013-12-06 14:19:25 +01002647 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01002648 }
2649
Willy Tarreau263013d2015-05-13 11:59:14 +02002650 if (&check->current_step->list == head)
Willy Tarreauabca5b62013-12-06 14:19:25 +01002651 break;
2652
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002653 /* have 'next' point to the next rule or NULL if we're on the
2654 * last one, connect() needs this.
2655 */
Willy Tarreau5581c272015-05-13 12:24:53 +02002656 next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002657
2658 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002659 while (&next->list != head && next->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002660 next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002661
2662 /* NULL if we're on the last rule */
Willy Tarreauf3d34822014-12-08 12:11:28 +01002663 if (&next->list == head)
2664 next = NULL;
2665
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002666 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
2667 struct protocol *proto;
2668 struct xprt_ops *xprt;
2669
2670 /* mark the step as started */
2671 check->last_started_step = check->current_step;
2672 /* first, shut existing connection */
2673 conn_force_close(conn);
2674
2675 /* prepare new connection */
2676 /* initialization */
2677 conn_init(conn);
2678 conn_attach(conn, check, &check_conn_cb);
2679 conn->target = &s->obj_type;
2680
2681 /* no client address */
2682 clear_addr(&conn->addr.from);
2683
Simon Horman41f58762015-01-30 11:22:56 +09002684 if (is_addr(&check->addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002685 /* we'll connect to the check addr specified on the server */
Simon Horman41f58762015-01-30 11:22:56 +09002686 conn->addr.to = check->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002687 }
2688 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002689 /* we'll connect to the addr on the server */
2690 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02002691 }
Thierry FOURNIERbb2ae642015-01-14 11:31:49 +01002692 proto = protocol_by_family(conn->addr.to.ss_family);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002693
2694 /* port */
2695 if (check->current_step->port)
2696 set_host_port(&conn->addr.to, check->current_step->port);
2697 else if (check->port)
2698 set_host_port(&conn->addr.to, check->port);
2699
2700#ifdef USE_OPENSSL
2701 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
2702 xprt = &ssl_sock;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002703 }
2704 else {
2705 xprt = &raw_sock;
2706 }
2707#else /* USE_OPENSSL */
2708 xprt = &raw_sock;
2709#endif /* USE_OPENSSL */
2710 conn_prepare(conn, proto, xprt);
2711
Willy Tarreaue7dff022015-04-03 01:14:29 +02002712 ret = SF_ERR_INTERNAL;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002713 if (proto->connect)
Willy Tarreauf3d34822014-12-08 12:11:28 +01002714 ret = proto->connect(conn,
2715 1 /* I/O polling is always needed */,
2716 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002717 conn->flags |= CO_FL_WAKE_DATA;
2718 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2719 conn->send_proxy_ofs = 1;
2720 conn->flags |= CO_FL_SEND_PROXY;
2721 }
2722
2723 /* It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +02002724 * - SF_ERR_NONE if everything's OK
2725 * - SF_ERR_SRVTO if there are no more servers
2726 * - SF_ERR_SRVCL if the connection was refused by the server
2727 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2728 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2729 * - SF_ERR_INTERNAL for any other purely internal errors
2730 * Additionnally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002731 * Note that we try to prevent the network stack from sending the ACK during the
2732 * connect() when a pure TCP check is used (without PROXY protocol).
2733 */
2734 switch (ret) {
Willy Tarreaue7dff022015-04-03 01:14:29 +02002735 case SF_ERR_NONE:
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002736 /* we allow up to min(inter, timeout.connect) for a connection
2737 * to establish but only when timeout.check is set
2738 * as it may be to short for a full check otherwise
2739 */
2740 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2741
2742 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2743 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2744 t->expire = tick_first(t->expire, t_con);
2745 }
2746 break;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002747 case SF_ERR_SRVTO: /* ETIMEDOUT */
2748 case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002749 step = tcpcheck_get_step_id(check);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002750 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002751 step, strerror(errno));
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002752 comment = tcpcheck_get_step_comment(check, step);
2753 if (comment)
2754 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002755 set_server_check_status(check, HCHK_STATUS_L4CON, trash.str);
2756 goto out_end_tcpcheck;
Willy Tarreaue7dff022015-04-03 01:14:29 +02002757 case SF_ERR_PRXCOND:
2758 case SF_ERR_RESOURCE:
2759 case SF_ERR_INTERNAL:
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002760 step = tcpcheck_get_step_id(check);
2761 chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002762 comment = tcpcheck_get_step_comment(check, step);
2763 if (comment)
2764 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002765 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2766 goto out_end_tcpcheck;
2767 }
2768
2769 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002770 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002771
2772 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002773 while (&check->current_step->list != head &&
2774 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002775 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002776
Willy Tarreauf2c87352015-05-13 12:08:21 +02002777 if (&check->current_step->list == head)
2778 break;
2779
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002780 /* don't do anything until the connection is established */
2781 if (!(conn->flags & CO_FL_CONNECTED)) {
2782 /* update expire time, should be done by process_chk */
2783 /* we allow up to min(inter, timeout.connect) for a connection
2784 * to establish but only when timeout.check is set
2785 * as it may be to short for a full check otherwise
2786 */
2787 while (tick_is_expired(t->expire, now_ms)) {
2788 int t_con;
2789
2790 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2791 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2792
2793 if (s->proxy->timeout.check)
2794 t->expire = tick_first(t->expire, t_con);
2795 }
2796 return;
2797 }
2798
2799 } /* end 'connect' */
2800 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2801 /* mark the step as started */
2802 check->last_started_step = check->current_step;
2803
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002804 /* reset the read buffer */
2805 if (*check->bi->data != '\0') {
2806 *check->bi->data = '\0';
2807 check->bi->i = 0;
2808 }
2809
2810 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
2811 conn->flags |= CO_FL_ERROR;
2812 chk_report_conn_err(conn, 0, 0);
2813 goto out_end_tcpcheck;
2814 }
2815
Willy Tarreauabca5b62013-12-06 14:19:25 +01002816 if (check->current_step->string_len >= check->bo->size) {
2817 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%d) at step %d",
2818 check->current_step->string_len, check->bo->size,
Simon Hormane16c1b32015-01-30 11:22:57 +09002819 tcpcheck_get_step_id(check));
Willy Tarreauabca5b62013-12-06 14:19:25 +01002820 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2821 goto out_end_tcpcheck;
2822 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002823
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002824 /* do not try to send if there is no space */
2825 if (check->current_step->string_len >= buffer_total_space(check->bo))
2826 continue;
2827
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002828 bo_putblk(check->bo, check->current_step->string, check->current_step->string_len);
2829 *check->bo->p = '\0'; /* to make gdb output easier to read */
2830
Willy Tarreauabca5b62013-12-06 14:19:25 +01002831 /* go to next rule and try to send */
Willy Tarreau5581c272015-05-13 12:24:53 +02002832 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002833
2834 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002835 while (&check->current_step->list != head &&
2836 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002837 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Willy Tarreauf2c87352015-05-13 12:08:21 +02002838
2839 if (&check->current_step->list == head)
2840 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002841 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002842 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002843 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002844 goto out_end_tcpcheck;
2845
Willy Tarreau310987a2014-01-22 19:46:33 +01002846 if (conn->xprt->rcv_buf(conn, check->bi, check->bi->size) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002847 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
2848 done = 1;
2849 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
2850 /* Report network errors only if we got no other data. Otherwise
2851 * we'll let the upper layers decide whether the response is OK
2852 * or not. It is very common that an RST sent by the server is
2853 * reported as an error just after the last data chunk.
2854 */
2855 chk_report_conn_err(conn, errno, 0);
2856 goto out_end_tcpcheck;
2857 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002858 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002859 else
Willy Tarreau263013d2015-05-13 11:59:14 +02002860 break;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002861 }
2862
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002863 /* mark the step as started */
2864 check->last_started_step = check->current_step;
2865
2866
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002867 /* Intermediate or complete response received.
2868 * Terminate string in check->bi->data buffer.
2869 */
2870 if (check->bi->i < check->bi->size) {
2871 check->bi->data[check->bi->i] = '\0';
2872 }
2873 else {
2874 check->bi->data[check->bi->i - 1] = '\0';
2875 done = 1; /* buffer full, don't wait for more data */
2876 }
2877
2878 contentptr = check->bi->data;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002879
2880 /* Check that response body is not empty... */
Willy Tarreauec6b0122014-05-13 17:57:29 +02002881 if (!check->bi->i) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002882 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002883 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002884
2885 /* empty response */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002886 step = tcpcheck_get_step_id(check);
2887 chunk_printf(&trash, "TCPCHK got an empty response at step %d", step);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002888 comment = tcpcheck_get_step_comment(check, step);
2889 if (comment)
2890 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002891 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2892
2893 goto out_end_tcpcheck;
2894 }
2895
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002896 if (!done && (check->current_step->string != NULL) && (check->bi->i < check->current_step->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002897 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002898
Willy Tarreaua970c282013-12-06 12:47:19 +01002899 tcpcheck_expect:
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002900 if (check->current_step->string != NULL)
2901 ret = my_memmem(contentptr, check->bi->i, check->current_step->string, check->current_step->string_len) != NULL;
2902 else if (check->current_step->expect_regex != NULL)
2903 ret = regex_exec(check->current_step->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002904
2905 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002906 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002907
2908 /* matched */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002909 step = tcpcheck_get_step_id(check);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002910 if (ret) {
2911 /* matched but we did not want to => ERROR */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002912 if (check->current_step->inverse) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002913 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002914 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002915 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002916 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002917 }
2918 else {
2919 /* we were looking for a regex */
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002920 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002921 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002922 comment = tcpcheck_get_step_comment(check, step);
2923 if (comment)
2924 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002925 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2926 goto out_end_tcpcheck;
2927 }
2928 /* matched and was supposed to => OK, next step */
2929 else {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002930 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002931 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002932
2933 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002934 while (&check->current_step->list != head &&
2935 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002936 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002937
Willy Tarreauf2c87352015-05-13 12:08:21 +02002938 if (&check->current_step->list == head)
2939 break;
2940
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002941 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002942 goto tcpcheck_expect;
2943 __conn_data_stop_recv(conn);
2944 }
2945 }
2946 else {
2947 /* not matched */
2948 /* not matched and was not supposed to => OK, next step */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002949 if (check->current_step->inverse) {
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002950 /* allow next rule */
Willy Tarreau5581c272015-05-13 12:24:53 +02002951 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002952
2953 /* bypass all comment rules */
Willy Tarreauf2c87352015-05-13 12:08:21 +02002954 while (&check->current_step->list != head &&
2955 check->current_step->action == TCPCHK_ACT_COMMENT)
Willy Tarreau5581c272015-05-13 12:24:53 +02002956 check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list);
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002957
Willy Tarreauf2c87352015-05-13 12:08:21 +02002958 if (&check->current_step->list == head)
2959 break;
2960
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002961 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002962 goto tcpcheck_expect;
2963 __conn_data_stop_recv(conn);
2964 }
2965 /* not matched but was supposed to => ERROR */
2966 else {
2967 /* we were looking for a string */
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002968 if (check->current_step->string != NULL) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002969 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
Willy Tarreauce8c42a2015-05-13 11:23:01 +02002970 check->current_step->string, step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002971 }
2972 else {
2973 /* we were looking for a regex */
2974 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
Baptiste Assmanncfbd1b82015-05-02 09:00:23 +02002975 step);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002976 }
Baptiste Assmann22b09d22015-05-01 08:03:04 +02002977 comment = tcpcheck_get_step_comment(check, step);
2978 if (comment)
2979 chunk_appendf(&trash, " comment: '%s'", comment);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002980 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2981 goto out_end_tcpcheck;
2982 }
2983 }
2984 } /* end expect */
2985 } /* end loop over double chained step list */
2986
Willy Tarreau263013d2015-05-13 11:59:14 +02002987 /* We're waiting for some I/O to complete, we've reached the end of the
2988 * rules, or both. Do what we have to do, otherwise we're done.
2989 */
2990 if (&check->current_step->list == head && !check->bo->o) {
2991 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
2992 goto out_end_tcpcheck;
2993 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002994
Willy Tarreau53c5a042015-05-13 11:38:17 +02002995 /* warning, current_step may now point to the head */
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002996 if (check->bo->o)
2997 __conn_data_want_send(conn);
2998
Willy Tarreau53c5a042015-05-13 11:38:17 +02002999 if (&check->current_step->list != head &&
3000 check->current_step->action == TCPCHK_ACT_EXPECT)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01003001 __conn_data_want_recv(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003002 return;
3003
3004 out_end_tcpcheck:
3005 /* collect possible new errors */
3006 if (conn->flags & CO_FL_ERROR)
3007 chk_report_conn_err(conn, 0, 0);
3008
Baptiste Assmann69e273f2013-12-11 00:52:19 +01003009 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003010 check->current_step = NULL;
3011
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01003012 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003013 conn->flags |= CO_FL_ERROR;
3014
3015 __conn_data_stop_both(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003016 return;
3017}
3018
Simon Hormanb1900d52015-01-30 11:22:54 +09003019const char *init_check(struct check *check, int type)
3020{
3021 check->type = type;
3022
3023 /* Allocate buffer for requests... */
3024 if ((check->bi = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3025 return "out of memory while allocating check buffer";
3026 }
3027 check->bi->size = global.tune.chksize;
3028
3029 /* Allocate buffer for responses... */
3030 if ((check->bo = calloc(sizeof(struct buffer) + global.tune.chksize, sizeof(char))) == NULL) {
3031 return "out of memory while allocating check buffer";
3032 }
3033 check->bo->size = global.tune.chksize;
3034
3035 /* Allocate buffer for partial results... */
3036 if ((check->conn = calloc(1, sizeof(struct connection))) == NULL) {
3037 return "out of memory while allocating check connection";
3038 }
3039
3040 check->conn->t.sock.fd = -1; /* no agent in progress yet */
3041
3042 return NULL;
3043}
3044
Simon Hormanbfb5d332015-01-30 11:22:55 +09003045void free_check(struct check *check)
3046{
3047 free(check->bi);
3048 free(check->bo);
3049 free(check->conn);
3050}
3051
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003052void email_alert_free(struct email_alert *alert)
3053{
3054 struct tcpcheck_rule *rule, *back;
3055
3056 if (!alert)
3057 return;
3058
3059 list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list)
3060 free(rule);
3061 free(alert);
3062}
3063
3064static struct task *process_email_alert(struct task *t)
3065{
3066 struct check *check = t->context;
3067 struct email_alertq *q;
3068
3069 q = container_of(check, typeof(*q), check);
3070
3071 if (!(check->state & CHK_ST_ENABLED)) {
3072 if (LIST_ISEMPTY(&q->email_alerts)) {
3073 /* All alerts processed, delete check */
3074 task_delete(t);
3075 task_free(t);
3076 check->task = NULL;
3077 return NULL;
3078 } else {
3079 struct email_alert *alert;
3080
3081 alert = LIST_NEXT(&q->email_alerts, typeof(alert), list);
3082 check->tcpcheck_rules = &alert->tcpcheck_rules;
3083 LIST_DEL(&alert->list);
3084
3085 check->state |= CHK_ST_ENABLED;
3086 }
3087
3088 }
3089
3090 process_chk(t);
3091
3092 if (!(check->state & CHK_ST_INPROGRESS) && check->tcpcheck_rules) {
3093 struct email_alert *alert;
3094
3095 alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules);
3096 email_alert_free(alert);
3097
3098 check->tcpcheck_rules = NULL;
3099 check->state &= ~CHK_ST_ENABLED;
3100 }
3101 return t;
3102}
3103
3104static int init_email_alert_checks(struct server *s)
3105{
3106 int i;
3107 struct mailer *mailer;
3108 const char *err_str;
3109 struct proxy *p = s->proxy;
3110
3111 if (p->email_alert.queues)
3112 /* Already initialised, nothing to do */
3113 return 1;
3114
3115 p->email_alert.queues = calloc(p->email_alert.mailers.m->count, sizeof *p->email_alert.queues);
3116 if (!p->email_alert.queues) {
3117 err_str = "out of memory while allocating checks array";
3118 goto error_alert;
3119 }
3120
3121 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3122 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
3123 struct email_alertq *q = &p->email_alert.queues[i];
3124 struct check *check = &q->check;
3125
3126
3127 LIST_INIT(&q->email_alerts);
3128
Pieter Baauw235fcfc2016-02-13 15:33:40 +01003129 check->inter = p->email_alert.mailers.m->timeout.mail;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003130 check->rise = DEF_AGENT_RISETIME;
3131 check->fall = DEF_AGENT_FALLTIME;
3132 err_str = init_check(check, PR_O2_TCPCHK_CHK);
3133 if (err_str) {
3134 goto error_free;
3135 }
3136
3137 check->xprt = mailer->xprt;
3138 if (!get_host_port(&mailer->addr))
3139 /* Default to submission port */
3140 check->port = 587;
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003141 check->addr = mailer->addr;
3142 check->server = s;
3143 }
3144
3145 return 1;
3146
3147error_free:
3148 while (i-- > 1)
3149 task_free(p->email_alert.queues[i].check.task);
3150 free(p->email_alert.queues);
3151 p->email_alert.queues = NULL;
3152error_alert:
3153 Alert("Email alert [%s] could not be initialised: %s\n", p->id, err_str);
3154 return 0;
3155}
3156
3157
3158static int add_tcpcheck_expect_str(struct list *list, const char *str)
3159{
3160 struct tcpcheck_rule *tcpcheck;
3161
3162 tcpcheck = calloc(1, sizeof *tcpcheck);
3163 if (!tcpcheck)
3164 return 0;
3165
3166 tcpcheck->action = TCPCHK_ACT_EXPECT;
3167 tcpcheck->string = strdup(str);
3168 if (!tcpcheck->string) {
3169 free(tcpcheck);
3170 return 0;
3171 }
3172
3173 LIST_ADDQ(list, &tcpcheck->list);
3174 return 1;
3175}
3176
3177static int add_tcpcheck_send_strs(struct list *list, const char * const *strs)
3178{
3179 struct tcpcheck_rule *tcpcheck;
3180 int i;
3181
3182 tcpcheck = calloc(1, sizeof *tcpcheck);
3183 if (!tcpcheck)
3184 return 0;
3185
3186 tcpcheck->action = TCPCHK_ACT_SEND;
3187
3188 tcpcheck->string_len = 0;
3189 for (i = 0; strs[i]; i++)
3190 tcpcheck->string_len += strlen(strs[i]);
3191
3192 tcpcheck->string = malloc(tcpcheck->string_len + 1);
3193 if (!tcpcheck->string) {
3194 free(tcpcheck);
3195 return 0;
3196 }
3197 tcpcheck->string[0] = '\0';
3198
3199 for (i = 0; strs[i]; i++)
3200 strcat(tcpcheck->string, strs[i]);
3201
3202 LIST_ADDQ(list, &tcpcheck->list);
3203 return 1;
3204}
3205
3206static int enqueue_one_email_alert(struct email_alertq *q, const char *msg)
3207{
3208 struct email_alert *alert = NULL;
3209 struct tcpcheck_rule *tcpcheck;
3210 struct check *check = &q->check;
3211 struct proxy *p = check->server->proxy;
3212
3213 alert = calloc(1, sizeof *alert);
3214 if (!alert) {
3215 goto error;
3216 }
3217 LIST_INIT(&alert->tcpcheck_rules);
3218
3219 tcpcheck = calloc(1, sizeof *tcpcheck);
3220 if (!tcpcheck)
3221 goto error;
3222 tcpcheck->action = TCPCHK_ACT_CONNECT;
3223 LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list);
3224
3225 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 "))
3226 goto error;
3227
3228 {
3229 const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" };
3230 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3231 goto error;
3232 }
3233
3234 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3235 goto error;
3236
3237 {
3238 const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" };
3239 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3240 goto error;
3241 }
3242
3243 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3244 goto error;
3245
3246 {
3247 const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" };
3248 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3249 goto error;
3250 }
3251
3252 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3253 goto error;
3254
3255 {
3256 const char * const strs[2] = { "DATA\r\n" };
3257 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3258 goto error;
3259 }
3260
3261 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 "))
3262 goto error;
3263
3264 {
3265 struct tm tm;
3266 char datestr[48];
3267 const char * const strs[18] = {
Pieter Baauw5e0964e2016-02-13 16:27:35 +01003268 "From: ", p->email_alert.from, "\r\n",
3269 "To: ", p->email_alert.to, "\r\n",
3270 "Date: ", datestr, "\r\n",
3271 "Subject: [HAproxy Alert] ", msg, "\r\n",
3272 "\r\n",
3273 msg, "\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003274 "\r\n",
Pieter Baauwed35c372015-07-22 19:51:54 +02003275 ".\r\n",
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003276 NULL
3277 };
3278
3279 get_localtime(date.tv_sec, &tm);
3280
3281 if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) {
3282 goto error;
3283 }
3284
3285 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3286 goto error;
3287 }
3288
3289 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 "))
3290 goto error;
3291
3292 {
3293 const char * const strs[2] = { "QUIT\r\n" };
3294 if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs))
3295 goto error;
3296 }
3297
3298 if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 "))
3299 goto error;
3300
3301 if (!check->task) {
3302 struct task *t;
3303
3304 if ((t = task_new()) == NULL)
3305 goto error;
3306
3307 check->task = t;
3308 t->process = process_email_alert;
3309 t->context = check;
3310
3311 /* check this in one ms */
3312 t->expire = tick_add(now_ms, MS_TO_TICKS(1));
3313 check->start = now;
3314 task_queue(t);
3315 }
3316
3317 LIST_ADDQ(&q->email_alerts, &alert->list);
3318
3319 return 1;
3320
3321error:
3322 email_alert_free(alert);
3323 return 0;
3324}
3325
3326static void enqueue_email_alert(struct proxy *p, const char *msg)
3327{
3328 int i;
3329 struct mailer *mailer;
3330
3331 for (i = 0, mailer = p->email_alert.mailers.m->mailer_list;
3332 i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) {
3333 if (!enqueue_one_email_alert(&p->email_alert.queues[i], msg)) {
3334 Alert("Email alert [%s] could not be enqueued: out of memory\n", p->id);
3335 return;
3336 }
3337 }
3338
3339 return;
3340}
3341
3342/*
3343 * Send email alert if configured.
3344 */
Simon Horman64e34162015-02-06 11:11:57 +09003345void send_email_alert(struct server *s, int level, const char *format, ...)
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003346{
3347 va_list argp;
3348 char buf[1024];
3349 int len;
3350 struct proxy *p = s->proxy;
3351
Simon Horman64e34162015-02-06 11:11:57 +09003352 if (!p->email_alert.mailers.m || level > p->email_alert.level ||
3353 format == NULL || !init_email_alert_checks(s))
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003354 return;
3355
3356 va_start(argp, format);
3357 len = vsnprintf(buf, sizeof(buf), format, argp);
3358 va_end(argp);
3359
3360 if (len < 0) {
Cyril Bontéb65e0332015-12-04 03:07:08 +01003361 Alert("Email alert [%s] could not format message\n", p->id);
Simon Horman0ba0e4a2015-01-30 11:23:00 +09003362 return;
3363 }
3364
3365 enqueue_email_alert(p, buf);
3366}
3367
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02003368
Willy Tarreaubd741542010-03-16 18:46:54 +01003369/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02003370 * Local variables:
3371 * c-indent-level: 8
3372 * c-basic-offset: 8
3373 * End:
3374 */