blob: 71debb6d499657f3864255b55a76756bed98ba03 [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>
18#include <stdio.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020019#include <stdlib.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020020#include <string.h>
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +020021#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020022#include <unistd.h>
23#include <sys/socket.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040024#include <sys/types.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020025#include <netinet/in.h>
Willy Tarreau1274bc42009-07-15 07:16:31 +020026#include <netinet/tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020027#include <arpa/inet.h>
28
Willy Tarreauc7e42382012-08-24 19:22:53 +020029#include <common/chunk.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020030#include <common/compat.h>
31#include <common/config.h>
32#include <common/mini-clist.h>
Willy Tarreau83749182007-04-15 20:56:27 +020033#include <common/standard.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020035
36#include <types/global.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020037
Baptiste Assmann69e273f2013-12-11 00:52:19 +010038#ifdef USE_OPENSSL
39#include <types/ssl_sock.h>
40#include <proto/ssl_sock.h>
41#endif /* USE_OPENSSL */
42
Willy Tarreaubaaee002006-06-26 02:48:02 +020043#include <proto/backend.h>
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020044#include <proto/checks.h>
Simon Hormana2b9dad2013-02-12 10:45:54 +090045#include <proto/dumpstats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020046#include <proto/fd.h>
47#include <proto/log.h>
48#include <proto/queue.h>
Willy Tarreauc6f4ce82009-06-10 11:09:37 +020049#include <proto/port_range.h>
Willy Tarreau3d300592007-03-18 18:34:41 +010050#include <proto/proto_http.h>
Willy Tarreaue8c66af2008-01-13 18:40:14 +010051#include <proto/proto_tcp.h>
Baptiste Assmann69e273f2013-12-11 00:52:19 +010052#include <proto/protocol.h>
Willy Tarreau2b5652f2006-12-31 17:46:05 +010053#include <proto/proxy.h>
Willy Tarreaufb56aab2012-09-28 14:40:02 +020054#include <proto/raw_sock.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020055#include <proto/server.h>
Simon Hormane0d1bfb2011-06-21 14:34:58 +090056#include <proto/session.h>
Willy Tarreau9e000c62011-03-10 14:03:36 +010057#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020058#include <proto/task.h>
59
Willy Tarreaubd741542010-03-16 18:46:54 +010060static int httpchk_expect(struct server *s, int done);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +020061static int tcpcheck_get_step_id(struct server *);
62static void tcpcheck_main(struct connection *);
Willy Tarreaubd741542010-03-16 18:46:54 +010063
Simon Horman63a4a822012-03-19 07:24:41 +090064static const struct check_status check_statuses[HCHK_STATUS_SIZE] = {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010065 [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" },
66 [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" },
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020067 [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020068
Willy Tarreau23964182014-05-20 20:56:30 +020069 /* Below we have finished checks */
70 [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" },
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010071 [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" },
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010072
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010073 [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020074
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010075 [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" },
76 [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" },
77 [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020078
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010079 [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" },
80 [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" },
81 [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020082
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010083 [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" },
84 [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020085
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +020086 [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020087
Willy Tarreau6aaa1b82013-12-11 17:09:34 +010088 [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" },
89 [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" },
90 [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" },
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +020091};
92
Simon Horman63a4a822012-03-19 07:24:41 +090093static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +010094 [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }},
95
96 [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }},
97 [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }},
98
99 [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }},
100 [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }},
101 [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }},
102 [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }},
103
104 [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }},
105 [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }},
106 [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }},
107};
108
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200109/*
110 * Convert check_status code to description
111 */
112const char *get_check_status_description(short check_status) {
113
114 const char *desc;
115
116 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200117 desc = check_statuses[check_status].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200118 else
119 desc = NULL;
120
121 if (desc && *desc)
122 return desc;
123 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200124 return check_statuses[HCHK_STATUS_UNKNOWN].desc;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200125}
126
127/*
128 * Convert check_status code to short info
129 */
130const char *get_check_status_info(short check_status) {
131
132 const char *info;
133
134 if (check_status < HCHK_STATUS_SIZE)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200135 info = check_statuses[check_status].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200136 else
137 info = NULL;
138
139 if (info && *info)
140 return info;
141 else
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200142 return check_statuses[HCHK_STATUS_UNKNOWN].info;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200143}
144
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100145const char *get_analyze_status(short analyze_status) {
146
147 const char *desc;
148
149 if (analyze_status < HANA_STATUS_SIZE)
150 desc = analyze_statuses[analyze_status].desc;
151 else
152 desc = NULL;
153
154 if (desc && *desc)
155 return desc;
156 else
157 return analyze_statuses[HANA_STATUS_UNKNOWN].desc;
158}
159
Willy Tarreaua150cf12014-05-20 21:57:23 +0200160/* Builds a string containing some information about the health check's result.
161 * The output string is allocated from the trash chunks. If the check is NULL,
162 * NULL is returned. This is designed to be used when emitting logs about health
163 * checks.
Willy Tarreauddd329c2014-05-16 16:46:12 +0200164 */
Willy Tarreaua150cf12014-05-20 21:57:23 +0200165static const char *check_reason_string(struct check *check)
Willy Tarreauddd329c2014-05-16 16:46:12 +0200166{
Willy Tarreaua150cf12014-05-20 21:57:23 +0200167 struct chunk *msg;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200168
Willy Tarreaua150cf12014-05-20 21:57:23 +0200169 if (!check)
170 return NULL;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200171
Willy Tarreaua150cf12014-05-20 21:57:23 +0200172 msg = get_trash_chunk();
173 chunk_printf(msg, "reason: %s", get_check_status_description(check->status));
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200174
Willy Tarreaua150cf12014-05-20 21:57:23 +0200175 if (check->status >= HCHK_STATUS_L57DATA)
176 chunk_appendf(msg, ", code: %d", check->code);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200177
Willy Tarreaua150cf12014-05-20 21:57:23 +0200178 if (*check->desc) {
179 struct chunk src;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200180
Willy Tarreaua150cf12014-05-20 21:57:23 +0200181 chunk_appendf(msg, ", info: \"");
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200182
Willy Tarreaua150cf12014-05-20 21:57:23 +0200183 chunk_initlen(&src, check->desc, 0, strlen(check->desc));
184 chunk_asciiencode(msg, &src, '"');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200185
Willy Tarreaua150cf12014-05-20 21:57:23 +0200186 chunk_appendf(msg, "\"");
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200187 }
188
Willy Tarreaua150cf12014-05-20 21:57:23 +0200189 if (check->duration >= 0)
190 chunk_appendf(msg, ", check duration: %ldms", check->duration);
191
192 return msg->str;
Krzysztof Piotr Oledzki99ab5f82009-09-27 17:28:21 +0200193}
194
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200195/*
Simon Horman4a741432013-02-23 15:35:38 +0900196 * Set check->status, update check->duration and fill check->result with
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200197 * an adequate CHK_RES_* value. The new check->health is computed based
198 * on the result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200199 *
200 * Show information in logs about failed health check if server is UP
201 * or succeeded health checks if server is DOWN.
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200202 */
Simon Horman4a741432013-02-23 15:35:38 +0900203static void set_server_check_status(struct check *check, short status, const char *desc)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100204{
Simon Horman4a741432013-02-23 15:35:38 +0900205 struct server *s = check->server;
Willy Tarreaubef1b322014-05-13 21:01:39 +0200206 short prev_status = check->status;
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200207 int report = 0;
Simon Horman4a741432013-02-23 15:35:38 +0900208
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200209 if (status == HCHK_STATUS_START) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100210 check->result = CHK_RES_UNKNOWN; /* no result yet */
Simon Horman4a741432013-02-23 15:35:38 +0900211 check->desc[0] = '\0';
212 check->start = now;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200213 return;
214 }
215
Simon Horman4a741432013-02-23 15:35:38 +0900216 if (!check->status)
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200217 return;
218
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200219 if (desc && *desc) {
Simon Horman4a741432013-02-23 15:35:38 +0900220 strncpy(check->desc, desc, HCHK_DESC_LEN-1);
221 check->desc[HCHK_DESC_LEN-1] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200222 } else
Simon Horman4a741432013-02-23 15:35:38 +0900223 check->desc[0] = '\0';
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200224
Simon Horman4a741432013-02-23 15:35:38 +0900225 check->status = status;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200226 if (check_statuses[status].result)
Simon Horman4a741432013-02-23 15:35:38 +0900227 check->result = check_statuses[status].result;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200228
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100229 if (status == HCHK_STATUS_HANA)
Simon Horman4a741432013-02-23 15:35:38 +0900230 check->duration = -1;
231 else if (!tv_iszero(&check->start)) {
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200232 /* set_server_check_status() may be called more than once */
Simon Horman4a741432013-02-23 15:35:38 +0900233 check->duration = tv_ms_elapsed(&check->start, &now);
234 tv_zero(&check->start);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200235 }
236
Willy Tarreau23964182014-05-20 20:56:30 +0200237 /* no change is expected if no state change occurred */
238 if (check->result == CHK_RES_NEUTRAL)
239 return;
240
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200241 report = 0;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200242
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200243 switch (check->result) {
244 case CHK_RES_FAILED:
Willy Tarreau12634e12014-05-23 11:32:36 +0200245 /* Failure to connect to the agent as a secondary check should not
246 * cause the server to be marked down.
247 */
248 if ((!(check->state & CHK_ST_AGENT) ||
Simon Hormanbfb8f882015-02-26 11:26:17 +0900249 (check->status >= HCHK_STATUS_L57DATA)) &&
Willy Tarreau12634e12014-05-23 11:32:36 +0200250 (check->health >= check->rise)) {
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200251 s->counters.failed_checks++;
252 report = 1;
253 check->health--;
254 if (check->health < check->rise)
255 check->health = 0;
256 }
257 break;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200258
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200259 case CHK_RES_PASSED:
260 case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */
261 if ((check->health < check->rise + check->fall - 1) &&
262 (check->result == CHK_RES_PASSED || check->health > 0)) {
263 report = 1;
264 check->health++;
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200265
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200266 if (check->health >= check->rise)
267 check->health = check->rise + check->fall - 1; /* OK now */
268 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200269
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200270 /* clear consecutive_errors if observing is enabled */
271 if (s->onerror)
272 s->consecutive_errors = 0;
273 break;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100274
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200275 default:
276 break;
277 }
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200278
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200279 if (s->proxy->options2 & PR_O2_LOGHCHKS &&
280 (status != prev_status || report)) {
281 chunk_printf(&trash,
Willy Tarreau12634e12014-05-23 11:32:36 +0200282 "%s check for %sserver %s/%s %s%s",
283 (check->state & CHK_ST_AGENT) ? "Agent" : "Health",
Willy Tarreauc93cd162014-05-13 15:54:22 +0200284 s->flags & SRV_F_BACKUP ? "backup " : "",
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100285 s->proxy->id, s->id,
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100286 (check->result == CHK_RES_CONDPASS) ? "conditionally ":"",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200287 (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200288
Willy Tarreaua150cf12014-05-20 21:57:23 +0200289 srv_append_status(&trash, s, check_reason_string(check), -1, 0);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200290
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100291 chunk_appendf(&trash, ", status: %d/%d %s",
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200292 (check->health >= check->rise) ? check->health - check->rise + 1 : check->health,
293 (check->health >= check->rise) ? check->fall : check->rise,
294 (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN");
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200295
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100296 Warning("%s.\n", trash.str);
297 send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.str);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200298 }
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200299}
300
Willy Tarreau4eec5472014-05-20 22:32:27 +0200301/* Marks the check <check>'s server down if the current check is already failed
302 * and the server is not down yet nor in maintenance.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200303 */
Willy Tarreau4eec5472014-05-20 22:32:27 +0200304static void check_notify_failure(struct check *check)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200305{
Simon Horman4a741432013-02-23 15:35:38 +0900306 struct server *s = check->server;
Simon Hormane0d1bfb2011-06-21 14:34:58 +0900307
Willy Tarreau7b1d47c2014-05-20 14:55:13 +0200308 /* The agent secondary check should only cause a server to be marked
309 * as down if check->status is HCHK_STATUS_L7STS, which indicates
310 * that the agent returned "fail", "stopped" or "down".
311 * The implication here is that failure to connect to the agent
312 * as a secondary check should not cause the server to be marked
313 * down. */
314 if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS)
315 return;
316
Willy Tarreau4eec5472014-05-20 22:32:27 +0200317 if (check->health > 0)
318 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100319
Willy Tarreau4eec5472014-05-20 22:32:27 +0200320 /* We only report a reason for the check if we did not do so previously */
321 srv_set_stopped(s, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check_reason_string(check) : NULL);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200322}
323
Willy Tarreauaf549582014-05-16 17:37:50 +0200324/* Marks the check <check> as valid and tries to set its server up, provided
Willy Tarreau3e048382014-05-21 10:30:54 +0200325 * it isn't in maintenance, it is not tracking a down server and other checks
326 * comply. The rule is simple : by default, a server is up, unless any of the
327 * following conditions is true :
328 * - health check failed (check->health < rise)
329 * - agent check failed (agent->health < rise)
330 * - the server tracks a down server (track && track->state == STOPPED)
331 * Note that if the server has a slowstart, it will switch to STARTING instead
332 * of RUNNING. Also, only the health checks support the nolb mode, so the
333 * agent's success may not take the server out of this mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200334 */
Willy Tarreau3e048382014-05-21 10:30:54 +0200335static void check_notify_success(struct check *check)
Willy Tarreauaf549582014-05-16 17:37:50 +0200336{
Simon Horman4a741432013-02-23 15:35:38 +0900337 struct server *s = check->server;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100338
Willy Tarreauaf549582014-05-16 17:37:50 +0200339 if (s->admin & SRV_ADMF_MAINT)
340 return;
Cyril Bontécd19e512010-01-31 22:34:03 +0100341
Willy Tarreau3e048382014-05-21 10:30:54 +0200342 if (s->track && s->track->state == SRV_ST_STOPPED)
Willy Tarreauaf549582014-05-16 17:37:50 +0200343 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100344
Willy Tarreau3e048382014-05-21 10:30:54 +0200345 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
346 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100347
Willy Tarreau3e048382014-05-21 10:30:54 +0200348 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
349 return;
Willy Tarreauaf549582014-05-16 17:37:50 +0200350
Willy Tarreau3e048382014-05-21 10:30:54 +0200351 if ((check->state & CHK_ST_AGENT) && s->state == SRV_ST_STOPPING)
352 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100353
Willy Tarreau3e048382014-05-21 10:30:54 +0200354 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 +0100355}
356
Willy Tarreaudb58b792014-05-21 13:57:23 +0200357/* Marks the check <check> as valid and tries to set its server into stopping mode
358 * if it was running or starting, and provided it isn't in maintenance and other
359 * checks comply. The conditions for the server to be marked in stopping mode are
360 * the same as for it to be turned up. Also, only the health checks support the
361 * nolb mode.
Willy Tarreauaf549582014-05-16 17:37:50 +0200362 */
Willy Tarreaudb58b792014-05-21 13:57:23 +0200363static void check_notify_stopping(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;
369
Willy Tarreaudb58b792014-05-21 13:57:23 +0200370 if (check->state & CHK_ST_AGENT)
371 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100372
Willy Tarreaudb58b792014-05-21 13:57:23 +0200373 if (s->track && s->track->state == SRV_ST_STOPPED)
374 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100375
Willy Tarreaudb58b792014-05-21 13:57:23 +0200376 if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise))
377 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100378
Willy Tarreaudb58b792014-05-21 13:57:23 +0200379 if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise))
380 return;
Krzysztof Piotr Oledzkic8b16fc2008-02-18 01:26:35 +0100381
Willy Tarreaudb58b792014-05-21 13:57:23 +0200382 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 +0100383}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200384
Willy Tarreau9fe7aae2013-12-31 23:47:37 +0100385/* note: use health_adjust() only, which first checks that the observe mode is
386 * enabled.
387 */
388void __health_adjust(struct server *s, short status)
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100389{
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100390 int failed;
391 int expire;
392
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100393 if (s->observe >= HANA_OBS_SIZE)
394 return;
395
Willy Tarreaubb956662013-01-24 00:37:39 +0100396 if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc)
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100397 return;
398
399 switch (analyze_statuses[status].lr[s->observe - 1]) {
400 case 1:
401 failed = 1;
402 break;
403
404 case 2:
405 failed = 0;
406 break;
407
408 default:
409 return;
410 }
411
412 if (!failed) {
413 /* good: clear consecutive_errors */
414 s->consecutive_errors = 0;
415 return;
416 }
417
418 s->consecutive_errors++;
419
420 if (s->consecutive_errors < s->consecutive_errors_limit)
421 return;
422
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100423 chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s",
424 s->consecutive_errors, get_analyze_status(status));
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100425
426 switch (s->onerror) {
427 case HANA_ONERR_FASTINTER:
428 /* force fastinter - nothing to do here as all modes force it */
429 break;
430
431 case HANA_ONERR_SUDDTH:
432 /* simulate a pre-fatal failed health check */
Simon Horman58c32972013-11-25 10:46:38 +0900433 if (s->check.health > s->check.rise)
434 s->check.health = s->check.rise + 1;
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100435
436 /* no break - fall through */
437
438 case HANA_ONERR_FAILCHK:
439 /* simulate a failed health check */
Simon Horman4a741432013-02-23 15:35:38 +0900440 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200441 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100442 break;
443
444 case HANA_ONERR_MARKDWN:
445 /* mark server down */
Simon Horman58c32972013-11-25 10:46:38 +0900446 s->check.health = s->check.rise;
Simon Horman4a741432013-02-23 15:35:38 +0900447 set_server_check_status(&s->check, HCHK_STATUS_HANA, trash.str);
Willy Tarreau4eec5472014-05-20 22:32:27 +0200448 check_notify_failure(&s->check);
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100449 break;
450
451 default:
452 /* write a warning? */
453 break;
454 }
455
456 s->consecutive_errors = 0;
457 s->counters.failed_hana++;
458
Simon Horman66183002013-02-23 10:16:43 +0900459 if (s->check.fastinter) {
460 expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter));
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300461 if (s->check.task->expire > expire) {
Willy Tarreau5b3a2022012-09-28 15:01:02 +0200462 s->check.task->expire = expire;
Sergiy Prykhodko1d57e502013-09-21 12:05:00 +0300463 /* requeue check task with new expire */
464 task_queue(s->check.task);
465 }
Krzysztof Piotr Oledzki97f07b82009-12-15 22:31:24 +0100466 }
467}
468
Willy Tarreaua1dab552014-04-14 15:04:54 +0200469static int httpchk_build_status_header(struct server *s, char *buffer, int size)
Willy Tarreauef781042010-01-27 11:53:01 +0100470{
471 int sv_state;
472 int ratio;
473 int hlen = 0;
474 const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d",
475 "UP %d/%d", "UP",
476 "NOLB %d/%d", "NOLB",
477 "no check" };
478
479 memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24);
480 hlen += 24;
481
Willy Tarreauff5ae352013-12-11 20:36:34 +0100482 if (!(s->check.state & CHK_ST_ENABLED))
483 sv_state = 6;
Willy Tarreau892337c2014-05-13 23:41:20 +0200484 else if (s->state != SRV_ST_STOPPED) {
Simon Horman58c32972013-11-25 10:46:38 +0900485 if (s->check.health == s->check.rise + s->check.fall - 1)
Willy Tarreauef781042010-01-27 11:53:01 +0100486 sv_state = 3; /* UP */
487 else
488 sv_state = 2; /* going down */
489
Willy Tarreau892337c2014-05-13 23:41:20 +0200490 if (s->state == SRV_ST_STOPPING)
Willy Tarreauef781042010-01-27 11:53:01 +0100491 sv_state += 2;
492 } else {
Simon Horman125d0992013-02-24 17:23:38 +0900493 if (s->check.health)
Willy Tarreauef781042010-01-27 11:53:01 +0100494 sv_state = 1; /* going up */
495 else
496 sv_state = 0; /* DOWN */
497 }
498
Willy Tarreaua1dab552014-04-14 15:04:54 +0200499 hlen += snprintf(buffer + hlen, size - hlen,
Willy Tarreauef781042010-01-27 11:53:01 +0100500 srv_hlt_st[sv_state],
Willy Tarreau892337c2014-05-13 23:41:20 +0200501 (s->state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health),
502 (s->state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise));
Willy Tarreauef781042010-01-27 11:53:01 +0100503
Willy Tarreaua1dab552014-04-14 15:04:54 +0200504 hlen += snprintf(buffer + hlen, size - hlen, "; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d",
Willy Tarreauef781042010-01-27 11:53:01 +0100505 s->proxy->id, s->id,
506 global.node,
507 (s->eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
508 (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv,
509 s->cur_sess, s->proxy->beconn - s->proxy->nbpend,
510 s->nbpend);
511
Willy Tarreau892337c2014-05-13 23:41:20 +0200512 if ((s->state == SRV_ST_STARTING) &&
Willy Tarreauef781042010-01-27 11:53:01 +0100513 now.tv_sec < s->last_change + s->slowstart &&
514 now.tv_sec >= s->last_change) {
515 ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart);
Willy Tarreaua1dab552014-04-14 15:04:54 +0200516 hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio);
Willy Tarreauef781042010-01-27 11:53:01 +0100517 }
518
519 buffer[hlen++] = '\r';
520 buffer[hlen++] = '\n';
521
522 return hlen;
523}
524
Willy Tarreau20a18342013-12-05 00:31:46 +0100525/* Check the connection. If an error has already been reported or the socket is
526 * closed, keep errno intact as it is supposed to contain the valid error code.
527 * If no error is reported, check the socket's error queue using getsockopt().
528 * Warning, this must be done only once when returning from poll, and never
529 * after an I/O error was attempted, otherwise the error queue might contain
530 * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the
531 * socket. Returns non-zero if an error was reported, zero if everything is
532 * clean (including a properly closed socket).
533 */
534static int retrieve_errno_from_socket(struct connection *conn)
535{
536 int skerr;
537 socklen_t lskerr = sizeof(skerr);
538
539 if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl))
540 return 1;
541
Willy Tarreau3c728722014-01-23 13:50:42 +0100542 if (!conn_ctrl_ready(conn))
Willy Tarreau20a18342013-12-05 00:31:46 +0100543 return 0;
544
545 if (getsockopt(conn->t.sock.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0)
546 errno = skerr;
547
548 if (errno == EAGAIN)
549 errno = 0;
550
551 if (!errno) {
552 /* we could not retrieve an error, that does not mean there is
553 * none. Just don't change anything and only report the prior
554 * error if any.
555 */
556 if (conn->flags & CO_FL_ERROR)
557 return 1;
558 else
559 return 0;
560 }
561
562 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH;
563 return 1;
564}
565
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100566/* Try to collect as much information as possible on the connection status,
567 * and adjust the server status accordingly. It may make use of <errno_bck>
568 * if non-null when the caller is absolutely certain of its validity (eg:
569 * checked just after a syscall). If the caller doesn't have a valid errno,
570 * it can pass zero, and retrieve_errno_from_socket() will be called to try
571 * to extract errno from the socket. If no error is reported, it will consider
572 * the <expired> flag. This is intended to be used when a connection error was
573 * reported in conn->flags or when a timeout was reported in <expired>. The
574 * function takes care of not updating a server status which was already set.
575 * All situations where at least one of <expired> or CO_FL_ERROR are set
576 * produce a status.
577 */
578static void chk_report_conn_err(struct connection *conn, int errno_bck, int expired)
579{
580 struct check *check = conn->owner;
581 const char *err_msg;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200582 struct chunk *chk;
Willy Tarreau90055f22014-10-02 14:51:02 +0200583 int step;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100584
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100585 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100586 return;
587
588 errno = errno_bck;
589 if (!errno || errno == EAGAIN)
590 retrieve_errno_from_socket(conn);
591
592 if (!(conn->flags & CO_FL_ERROR) && !expired)
593 return;
594
595 /* we'll try to build a meaningful error message depending on the
596 * context of the error possibly present in conn->err_code, and the
597 * socket error possibly collected above. This is useful to know the
598 * exact step of the L6 layer (eg: SSL handshake).
599 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200600 chk = get_trash_chunk();
601
602 if (check->type == PR_O2_TCPCHK_CHK) {
Willy Tarreau90055f22014-10-02 14:51:02 +0200603 step = tcpcheck_get_step_id(check->server);
604 if (!step)
605 chunk_printf(chk, " at initial connection step of tcp-check");
606 else {
607 chunk_printf(chk, " at step %d of tcp-check", step);
608 /* we were looking for a string */
609 if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) {
610 if (check->last_started_step->port)
611 chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port);
612 else
613 chunk_appendf(chk, " (connect)");
614 }
615 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) {
616 if (check->last_started_step->string)
617 chunk_appendf(chk, " (string '%s')", check->last_started_step->string);
618 else if (check->last_started_step->expect_regex)
619 chunk_appendf(chk, " (expect regex)");
620 }
621 else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) {
622 chunk_appendf(chk, " (send)");
623 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200624 }
625 }
626
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100627 if (conn->err_code) {
628 if (errno && errno != EAGAIN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200629 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100630 else
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200631 chunk_printf(&trash, "%s%s", conn_err_code_str(conn), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100632 err_msg = trash.str;
633 }
634 else {
635 if (errno && errno != EAGAIN) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200636 chunk_printf(&trash, "%s%s", strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100637 err_msg = trash.str;
638 }
639 else {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200640 err_msg = chk->str;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100641 }
642 }
643
644 if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
645 /* L4 not established (yet) */
646 if (conn->flags & CO_FL_ERROR)
647 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
648 else if (expired)
649 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
650 }
651 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
652 /* L6 not established (yet) */
653 if (conn->flags & CO_FL_ERROR)
654 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
655 else if (expired)
656 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
657 }
658 else if (conn->flags & CO_FL_ERROR) {
659 /* I/O error after connection was established and before we could diagnose */
660 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
661 }
662 else if (expired) {
663 /* connection established but expired check */
664 if (check->type == PR_O2_SSL3_CHK)
665 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
666 else /* HTTP, SMTP, ... */
667 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
668 }
669
670 return;
671}
672
Willy Tarreaubaaee002006-06-26 02:48:02 +0200673/*
674 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200675 * the connection acknowledgement. If the proxy requires L7 health-checks,
676 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900677 * to set check->status, check->duration and check->result.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200678 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200679static void event_srv_chk_w(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200680{
Simon Horman4a741432013-02-23 15:35:38 +0900681 struct check *check = conn->owner;
682 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900683 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200684
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100685 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100686 goto out_wakeup;
687
Willy Tarreau310987a2014-01-22 19:46:33 +0100688 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100689 return;
690
Willy Tarreau20a18342013-12-05 00:31:46 +0100691 if (retrieve_errno_from_socket(conn)) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100692 chk_report_conn_err(conn, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100693 __conn_data_stop_both(conn);
694 goto out_wakeup;
695 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100696
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100697 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100698 /* if the output is closed, we can't do anything */
699 conn->flags |= CO_FL_ERROR;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100700 chk_report_conn_err(conn, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100701 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200702 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200703
Willy Tarreau06559ac2013-12-05 01:53:08 +0100704 /* here, we know that the connection is established. That's enough for
705 * a pure TCP check.
706 */
707 if (!check->type)
708 goto out_wakeup;
709
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200710 if (check->type == PR_O2_TCPCHK_CHK) {
711 tcpcheck_main(conn);
712 return;
713 }
714
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100715 if (check->bo->o) {
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100716 conn->xprt->snd_buf(conn, check->bo, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100717 if (conn->flags & CO_FL_ERROR) {
718 chk_report_conn_err(conn, errno, 0);
719 __conn_data_stop_both(conn);
720 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200721 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100722 if (check->bo->o)
723 return;
724 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200725
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100726 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
727 if (s->proxy->timeout.check) {
728 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
729 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100731 goto out_nowake;
732
Willy Tarreau83749182007-04-15 20:56:27 +0200733 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200734 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200735 out_nowake:
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200736 __conn_data_stop_send(conn); /* nothing more to write */
Willy Tarreaubaaee002006-06-26 02:48:02 +0200737}
738
Willy Tarreaubaaee002006-06-26 02:48:02 +0200739/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200740 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200741 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900742 * set_server_check_status() to update check->status, check->duration
743 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200744
745 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
746 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
747 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
748 * response to an SSL HELLO (the principle is that this is enough to
749 * distinguish between an SSL server and a pure TCP relay). All other cases will
750 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
751 * etc.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200752 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200753static void event_srv_chk_r(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200754{
Simon Horman4a741432013-02-23 15:35:38 +0900755 struct check *check = conn->owner;
756 struct server *s = check->server;
757 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200758 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100759 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200760 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200761
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100762 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200763 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200764
Willy Tarreau310987a2014-01-22 19:46:33 +0100765 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200766 return;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200767
768 if (check->type == PR_O2_TCPCHK_CHK) {
769 tcpcheck_main(conn);
770 return;
771 }
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200772
Willy Tarreau83749182007-04-15 20:56:27 +0200773 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
774 * but the connection was closed on the remote end. Fortunately, recv still
775 * works correctly and we don't need to do the getsockopt() on linux.
776 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000777
778 /* Set buffer to point to the end of the data already read, and check
779 * that there is free space remaining. If the buffer is full, proceed
780 * with running the checks without attempting another socket read.
781 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000782
Willy Tarreau03938182010-03-17 21:52:07 +0100783 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000784
Simon Horman4a741432013-02-23 15:35:38 +0900785 conn->xprt->rcv_buf(conn, check->bi, check->bi->size);
Willy Tarreauf1503172012-09-28 19:39:36 +0200786 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
Willy Tarreau03938182010-03-17 21:52:07 +0100787 done = 1;
Simon Horman4a741432013-02-23 15:35:38 +0900788 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200789 /* Report network errors only if we got no other data. Otherwise
790 * we'll let the upper layers decide whether the response is OK
791 * or not. It is very common that an RST sent by the server is
792 * reported as an error just after the last data chunk.
793 */
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100794 chk_report_conn_err(conn, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100795 goto out_wakeup;
796 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200797 }
798
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100799
Willy Tarreau03938182010-03-17 21:52:07 +0100800 /* Intermediate or complete response received.
Simon Horman4a741432013-02-23 15:35:38 +0900801 * Terminate string in check->bi->data buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100802 */
Simon Horman4a741432013-02-23 15:35:38 +0900803 if (check->bi->i < check->bi->size)
804 check->bi->data[check->bi->i] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100805 else {
Simon Horman4a741432013-02-23 15:35:38 +0900806 check->bi->data[check->bi->i - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100807 done = 1; /* buffer full, don't wait for more data */
808 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200809
Nick Chalk57b1bf72010-03-16 15:50:46 +0000810 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900811 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200812 case PR_O2_HTTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900813 if (!done && check->bi->i < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100814 goto wait_more_data;
815
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100816 /* Check if the server speaks HTTP 1.X */
Simon Horman4a741432013-02-23 15:35:38 +0900817 if ((check->bi->i < strlen("HTTP/1.0 000\r")) ||
818 (memcmp(check->bi->data, "HTTP/1.", 7) != 0 ||
819 (*(check->bi->data + 12) != ' ' && *(check->bi->data + 12) != '\r')) ||
820 !isdigit((unsigned char) *(check->bi->data + 9)) || !isdigit((unsigned char) *(check->bi->data + 10)) ||
821 !isdigit((unsigned char) *(check->bi->data + 11))) {
822 cut_crlf(check->bi->data);
823 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200824
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100825 goto out_wakeup;
826 }
827
Simon Horman4a741432013-02-23 15:35:38 +0900828 check->code = str2uic(check->bi->data + 9);
829 desc = ltrim(check->bi->data + 12, ' ');
Nick Chalk57b1bf72010-03-16 15:50:46 +0000830
Willy Tarreaubd741542010-03-16 18:46:54 +0100831 if ((s->proxy->options & PR_O_DISABLE404) &&
Willy Tarreau892337c2014-05-13 23:41:20 +0200832 (s->state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000833 /* 404 may be accepted as "stopping" only if the server was up */
834 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900835 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000836 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100837 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
838 /* Run content verification check... We know we have at least 13 chars */
839 if (!httpchk_expect(s, done))
840 goto wait_more_data;
841 }
842 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Simon Horman4a741432013-02-23 15:35:38 +0900843 else if (*(check->bi->data + 9) == '2' || *(check->bi->data + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100844 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900845 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100846 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000847 else {
848 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900849 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000850 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200851 break;
852
853 case PR_O2_SSL3_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900854 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100855 goto wait_more_data;
856
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100857 /* Check for SSLv3 alert or handshake */
Simon Horman4a741432013-02-23 15:35:38 +0900858 if ((check->bi->i >= 5) && (*check->bi->data == 0x15 || *check->bi->data == 0x16))
859 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200860 else
Simon Horman4a741432013-02-23 15:35:38 +0900861 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200862 break;
863
864 case PR_O2_SMTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900865 if (!done && check->bi->i < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100866 goto wait_more_data;
867
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200868 /* Check if the server speaks SMTP */
Simon Horman4a741432013-02-23 15:35:38 +0900869 if ((check->bi->i < strlen("000\r")) ||
870 (*(check->bi->data + 3) != ' ' && *(check->bi->data + 3) != '\r') ||
871 !isdigit((unsigned char) *check->bi->data) || !isdigit((unsigned char) *(check->bi->data + 1)) ||
872 !isdigit((unsigned char) *(check->bi->data + 2))) {
873 cut_crlf(check->bi->data);
874 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200875
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200876 goto out_wakeup;
877 }
878
Simon Horman4a741432013-02-23 15:35:38 +0900879 check->code = str2uic(check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200880
Simon Horman4a741432013-02-23 15:35:38 +0900881 desc = ltrim(check->bi->data + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200882 cut_crlf(desc);
883
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100884 /* Check for SMTP code 2xx (should be 250) */
Simon Horman4a741432013-02-23 15:35:38 +0900885 if (*check->bi->data == '2')
886 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200887 else
Simon Horman4a741432013-02-23 15:35:38 +0900888 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200889 break;
890
Simon Hormana2b9dad2013-02-12 10:45:54 +0900891 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100892 int status = HCHK_STATUS_CHECKED;
893 const char *hs = NULL; /* health status */
894 const char *as = NULL; /* admin status */
895 const char *ps = NULL; /* performance status */
896 const char *err = NULL; /* first error to report */
897 const char *wrn = NULL; /* first warning to report */
898 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900899
Willy Tarreau81f5d942013-12-09 20:51:51 +0100900 /* We're getting an agent check response. The agent could
901 * have been disabled in the mean time with a long check
902 * still pending. It is important that we ignore the whole
903 * response.
904 */
905 if (!(check->server->agent.state & CHK_ST_ENABLED))
906 break;
907
908 /* The agent supports strings made of a single line ended by the
909 * first CR ('\r') or LF ('\n'). This line is composed of words
910 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
911 * line may optionally contained a description of a state change
912 * after a sharp ('#'), which is only considered if a health state
913 * is announced.
914 *
915 * Words may be composed of :
916 * - a numeric weight suffixed by the percent character ('%').
917 * - a health status among "up", "down", "stopped", and "fail".
918 * - an admin status among "ready", "drain", "maint".
919 *
920 * These words may appear in any order. If multiple words of the
921 * same category appear, the last one wins.
922 */
923
Willy Tarreau9809b782013-12-11 21:40:11 +0100924 p = check->bi->data;
925 while (*p && *p != '\n' && *p != '\r')
926 p++;
927
928 if (!*p) {
929 if (!done)
930 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900931
Willy Tarreau9809b782013-12-11 21:40:11 +0100932 /* at least inform the admin that the agent is mis-behaving */
933 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
934 break;
935 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100936
Willy Tarreau9809b782013-12-11 21:40:11 +0100937 *p = 0;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100938 cmd = check->bi->data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900939
Willy Tarreau81f5d942013-12-09 20:51:51 +0100940 while (*cmd) {
941 /* look for next word */
942 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
943 cmd++;
944 continue;
945 }
Simon Horman671b6f02013-11-25 10:46:39 +0900946
Willy Tarreau81f5d942013-12-09 20:51:51 +0100947 if (*cmd == '#') {
948 /* this is the beginning of a health status description,
949 * skip the sharp and blanks.
950 */
951 cmd++;
952 while (*cmd == '\t' || *cmd == ' ')
953 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +0900954 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900955 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100956
957 /* find the end of the word so that we have a null-terminated
958 * word between <cmd> and <p>.
959 */
960 p = cmd + 1;
961 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
962 p++;
963 if (*p)
964 *p++ = 0;
965
966 /* first, health statuses */
967 if (strcasecmp(cmd, "up") == 0) {
968 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900969 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100970 hs = cmd;
971 }
972 else if (strcasecmp(cmd, "down") == 0) {
973 check->health = 0;
974 status = HCHK_STATUS_L7STS;
975 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900976 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100977 else if (strcasecmp(cmd, "stopped") == 0) {
978 check->health = 0;
979 status = HCHK_STATUS_L7STS;
980 hs = cmd;
981 }
982 else if (strcasecmp(cmd, "fail") == 0) {
983 check->health = 0;
984 status = HCHK_STATUS_L7STS;
985 hs = cmd;
986 }
987 /* admin statuses */
988 else if (strcasecmp(cmd, "ready") == 0) {
989 as = cmd;
990 }
991 else if (strcasecmp(cmd, "drain") == 0) {
992 as = cmd;
993 }
994 else if (strcasecmp(cmd, "maint") == 0) {
995 as = cmd;
996 }
997 /* else try to parse a weight here and keep the last one */
998 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
999 ps = cmd;
1000 }
1001 else {
1002 /* keep a copy of the first error */
1003 if (!err)
1004 err = cmd;
1005 }
1006 /* skip to next word */
1007 cmd = p;
1008 }
1009 /* here, cmd points either to \0 or to the beginning of a
1010 * description. Skip possible leading spaces.
1011 */
1012 while (*cmd == ' ' || *cmd == '\n')
1013 cmd++;
1014
1015 /* First, update the admin status so that we avoid sending other
1016 * possibly useless warnings and can also update the health if
1017 * present after going back up.
1018 */
1019 if (as) {
1020 if (strcasecmp(as, "drain") == 0)
1021 srv_adm_set_drain(check->server);
1022 else if (strcasecmp(as, "maint") == 0)
1023 srv_adm_set_maint(check->server);
1024 else
1025 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001026 }
1027
Willy Tarreau81f5d942013-12-09 20:51:51 +01001028 /* now change weights */
1029 if (ps) {
1030 const char *msg;
1031
1032 msg = server_parse_weight_change_request(s, ps);
1033 if (!wrn || !*wrn)
1034 wrn = msg;
1035 }
1036
1037 /* and finally health status */
1038 if (hs) {
1039 /* We'll report some of the warnings and errors we have
1040 * here. Down reports are critical, we leave them untouched.
1041 * Lack of report, or report of 'UP' leaves the room for
1042 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001043 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001044 const char *msg = cmd;
1045 struct chunk *t;
1046
1047 if (!*msg || status == HCHK_STATUS_L7OKD) {
1048 if (err && *err)
1049 msg = err;
1050 else if (wrn && *wrn)
1051 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001052 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001053
1054 t = get_trash_chunk();
1055 chunk_printf(t, "via agent : %s%s%s%s",
1056 hs, *msg ? " (" : "",
1057 msg, *msg ? ")" : "");
1058
1059 set_server_check_status(check, status, t->str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001060 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001061 else if (err && *err) {
1062 /* No status change but we'd like to report something odd.
1063 * Just report the current state and copy the message.
1064 */
1065 chunk_printf(&trash, "agent reports an error : %s", err);
1066 set_server_check_status(check, status/*check->status*/, trash.str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001067
Willy Tarreau81f5d942013-12-09 20:51:51 +01001068 }
1069 else if (wrn && *wrn) {
1070 /* No status change but we'd like to report something odd.
1071 * Just report the current state and copy the message.
1072 */
1073 chunk_printf(&trash, "agent warns : %s", wrn);
1074 set_server_check_status(check, status/*check->status*/, trash.str);
1075 }
1076 else
1077 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001078 break;
1079 }
1080
Willy Tarreau1620ec32011-08-06 17:05:02 +02001081 case PR_O2_PGSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001082 if (!done && check->bi->i < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001083 goto wait_more_data;
1084
Simon Horman4a741432013-02-23 15:35:38 +09001085 if (check->bi->data[0] == 'R') {
1086 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001087 }
1088 else {
Simon Horman4a741432013-02-23 15:35:38 +09001089 if ((check->bi->data[0] == 'E') && (check->bi->data[5]!=0) && (check->bi->data[6]!=0))
1090 desc = &check->bi->data[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001091 else
1092 desc = "PostgreSQL unknown error";
1093
Simon Horman4a741432013-02-23 15:35:38 +09001094 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001095 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001096 break;
1097
1098 case PR_O2_REDIS_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001099 if (!done && check->bi->i < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001100 goto wait_more_data;
1101
Simon Horman4a741432013-02-23 15:35:38 +09001102 if (strcmp(check->bi->data, "+PONG\r\n") == 0) {
1103 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001104 }
1105 else {
Simon Horman4a741432013-02-23 15:35:38 +09001106 set_server_check_status(check, HCHK_STATUS_L7STS, check->bi->data);
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001107 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001108 break;
1109
1110 case PR_O2_MYSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001111 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001112 goto wait_more_data;
1113
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001114 if (s->proxy->check_len == 0) { // old mode
Simon Horman4a741432013-02-23 15:35:38 +09001115 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001116 /* We set the MySQL Version in description for information purpose
1117 * FIXME : it can be cool to use MySQL Version for other purpose,
1118 * like mark as down old MySQL server.
1119 */
Simon Horman4a741432013-02-23 15:35:38 +09001120 if (check->bi->i > 51) {
1121 desc = ltrim(check->bi->data + 5, ' ');
1122 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001123 }
1124 else {
1125 if (!done)
1126 goto wait_more_data;
1127 /* it seems we have a OK packet but without a valid length,
1128 * it must be a protocol error
1129 */
Simon Horman4a741432013-02-23 15:35:38 +09001130 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001131 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001132 }
1133 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001134 /* An error message is attached in the Error packet */
Simon Horman4a741432013-02-23 15:35:38 +09001135 desc = ltrim(check->bi->data + 7, ' ');
1136 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001137 }
1138 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001139 unsigned int first_packet_len = ((unsigned int) *check->bi->data) +
1140 (((unsigned int) *(check->bi->data + 1)) << 8) +
1141 (((unsigned int) *(check->bi->data + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001142
Simon Horman4a741432013-02-23 15:35:38 +09001143 if (check->bi->i == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001144 /* MySQL Error packet always begin with field_count = 0xff */
Simon Horman4a741432013-02-23 15:35:38 +09001145 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001146 /* We have only one MySQL packet and it is a Handshake Initialization packet
1147 * but we need to have a second packet to know if it is alright
1148 */
Simon Horman4a741432013-02-23 15:35:38 +09001149 if (!done && check->bi->i < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001150 goto wait_more_data;
1151 }
1152 else {
1153 /* We have only one packet and it is an Error packet,
1154 * an error message is attached, so we can display it
1155 */
Simon Horman4a741432013-02-23 15:35:38 +09001156 desc = &check->bi->data[7];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001157 //Warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001158 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001159 }
Simon Horman4a741432013-02-23 15:35:38 +09001160 } else if (check->bi->i > first_packet_len + 4) {
1161 unsigned int second_packet_len = ((unsigned int) *(check->bi->data + first_packet_len + 4)) +
1162 (((unsigned int) *(check->bi->data + first_packet_len + 5)) << 8) +
1163 (((unsigned int) *(check->bi->data + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001164
Simon Horman4a741432013-02-23 15:35:38 +09001165 if (check->bi->i == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001166 /* We have 2 packets and that's good */
1167 /* Check if the second packet is a MySQL Error packet or not */
Simon Horman4a741432013-02-23 15:35:38 +09001168 if (*(check->bi->data + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001169 /* No error packet */
1170 /* We set the MySQL Version in description for information purpose */
Simon Horman4a741432013-02-23 15:35:38 +09001171 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001172 //Warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001173 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001174 }
1175 else {
1176 /* An error message is attached in the Error packet
1177 * so we can display it ! :)
1178 */
Simon Horman4a741432013-02-23 15:35:38 +09001179 desc = &check->bi->data[first_packet_len+11];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001180 //Warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001181 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001182 }
1183 }
1184 }
1185 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001186 if (!done)
1187 goto wait_more_data;
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001188 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001189 * it must be a protocol error
1190 */
Simon Horman4a741432013-02-23 15:35:38 +09001191 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001192 //Warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001193 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001194 }
1195 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001196 break;
1197
1198 case PR_O2_LDAP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001199 if (!done && check->bi->i < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001200 goto wait_more_data;
1201
1202 /* Check if the server speaks LDAP (ASN.1/BER)
1203 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1204 * http://tools.ietf.org/html/rfc4511
1205 */
1206
1207 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1208 * LDAPMessage: 0x30: SEQUENCE
1209 */
Simon Horman4a741432013-02-23 15:35:38 +09001210 if ((check->bi->i < 14) || (*(check->bi->data) != '\x30')) {
1211 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001212 }
1213 else {
1214 /* size of LDAPMessage */
Simon Horman4a741432013-02-23 15:35:38 +09001215 msglen = (*(check->bi->data + 1) & 0x80) ? (*(check->bi->data + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001216
1217 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1218 * messageID: 0x02 0x01 0x01: INTEGER 1
1219 * protocolOp: 0x61: bindResponse
1220 */
1221 if ((msglen > 2) ||
Simon Horman4a741432013-02-23 15:35:38 +09001222 (memcmp(check->bi->data + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
1223 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001224
1225 goto out_wakeup;
1226 }
1227
1228 /* size of bindResponse */
Simon Horman4a741432013-02-23 15:35:38 +09001229 msglen += (*(check->bi->data + msglen + 6) & 0x80) ? (*(check->bi->data + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001230
1231 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1232 * ldapResult: 0x0a 0x01: ENUMERATION
1233 */
1234 if ((msglen > 4) ||
Simon Horman4a741432013-02-23 15:35:38 +09001235 (memcmp(check->bi->data + 7 + msglen, "\x0a\x01", 2) != 0)) {
1236 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001237
1238 goto out_wakeup;
1239 }
1240
1241 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1242 * resultCode
1243 */
Simon Horman4a741432013-02-23 15:35:38 +09001244 check->code = *(check->bi->data + msglen + 9);
1245 if (check->code) {
1246 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 +02001247 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001248 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001249 }
1250 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001251 break;
1252
1253 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001254 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001255 break;
1256 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001257
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001258 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001259 /* collect possible new errors */
1260 if (conn->flags & CO_FL_ERROR)
1261 chk_report_conn_err(conn, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001262
Nick Chalk57b1bf72010-03-16 15:50:46 +00001263 /* Reset the check buffer... */
Simon Horman4a741432013-02-23 15:35:38 +09001264 *check->bi->data = '\0';
1265 check->bi->i = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +00001266
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001267 /* Close the connection... We absolutely want to perform a hard close
1268 * and reset the connection if some data are pending, otherwise we end
1269 * up with many TIME_WAITs and eat all the source port range quickly.
1270 * To avoid sending RSTs all the time, we first try to drain pending
1271 * data.
1272 */
Willy Tarreauf1503172012-09-28 19:39:36 +02001273 if (conn->xprt && conn->xprt->shutw)
1274 conn->xprt->shutw(conn, 0);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001275
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001276 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001277 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001278 conn->flags |= CO_FL_ERROR;
1279
Willy Tarreaua522f802012-11-23 08:56:35 +01001280 __conn_data_stop_both(conn);
Willy Tarreaufdccded2008-08-29 18:19:04 +02001281 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau3267d362012-08-17 23:53:56 +02001282 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001283
1284 wait_more_data:
Willy Tarreauf817e9f2014-01-10 16:58:45 +01001285 __conn_data_want_recv(conn);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001286}
1287
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001288/*
1289 * This function is used only for server health-checks. It handles connection
1290 * status updates including errors. If necessary, it wakes the check task up.
1291 * It always returns 0.
1292 */
1293static int wake_srv_chk(struct connection *conn)
Willy Tarreau20bea422012-07-06 12:00:49 +02001294{
Simon Horman4a741432013-02-23 15:35:38 +09001295 struct check *check = conn->owner;
Willy Tarreau20bea422012-07-06 12:00:49 +02001296
Willy Tarreau6c560da2012-11-24 11:14:45 +01001297 if (unlikely(conn->flags & CO_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001298 /* We may get error reports bypassing the I/O handlers, typically
1299 * the case when sending a pure TCP check which fails, then the I/O
1300 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001301 * main processing task so let's simply wake it up. If we get here,
1302 * we expect errno to still be valid.
1303 */
1304 chk_report_conn_err(conn, errno, 0);
1305
Willy Tarreau2d351b62013-12-05 02:36:25 +01001306 __conn_data_stop_both(conn);
1307 task_wakeup(check->task, TASK_WOKEN_IO);
1308 }
Willy Tarreau3be293f2014-02-05 18:31:24 +01001309 else if (!(conn->flags & (CO_FL_DATA_RD_ENA|CO_FL_DATA_WR_ENA|CO_FL_HANDSHAKE))) {
1310 /* we may get here if only a connection probe was required : we
1311 * don't have any data to send nor anything expected in response,
1312 * so the completion of the connection establishment is enough.
1313 */
1314 task_wakeup(check->task, TASK_WOKEN_IO);
1315 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001316
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001317 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001318 /* We're here because nobody wants to handle the error, so we
1319 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001320 */
Willy Tarreau46be2e52014-01-20 12:10:52 +01001321 conn_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001322 conn_force_close(conn);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001323 }
Willy Tarreau3267d362012-08-17 23:53:56 +02001324 return 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001325}
1326
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001327struct data_cb check_conn_cb = {
1328 .recv = event_srv_chk_r,
1329 .send = event_srv_chk_w,
1330 .wake = wake_srv_chk,
1331};
1332
Willy Tarreaubaaee002006-06-26 02:48:02 +02001333/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001334 * updates the server's weight during a warmup stage. Once the final weight is
1335 * reached, the task automatically stops. Note that any server status change
1336 * must have updated s->last_change accordingly.
1337 */
1338static struct task *server_warmup(struct task *t)
1339{
1340 struct server *s = t->context;
1341
1342 /* by default, plan on stopping the task */
1343 t->expire = TICK_ETERNITY;
Willy Tarreau20125212014-05-13 19:44:56 +02001344 if ((s->admin & SRV_ADMF_MAINT) ||
Willy Tarreau892337c2014-05-13 23:41:20 +02001345 (s->state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001346 return t;
1347
Willy Tarreau892337c2014-05-13 23:41:20 +02001348 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001349 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001350
1351 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001352 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001353
1354 /* get back there in 1 second or 1/20th of the slowstart interval,
1355 * whichever is greater, resulting in small 5% steps.
1356 */
Willy Tarreau892337c2014-05-13 23:41:20 +02001357 if (s->state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001358 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1359 return t;
1360}
1361
1362/*
Simon Hormanb00d17a2014-06-13 16:18:16 +09001363 * establish a server health-check.
1364 *
1365 * It can return one of :
1366 * - SN_ERR_NONE if everything's OK and tcpcheck_main() was not called
1367 * - SN_ERR_UP if if everything's OK and tcpcheck_main() was called
1368 * - SN_ERR_SRVTO if there are no more servers
1369 * - SN_ERR_SRVCL if the connection was refused by the server
1370 * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1371 * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1372 * - SN_ERR_INTERNAL for any other purely internal errors
1373 * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted.
1374 * Note that we try to prevent the network stack from sending the ACK during the
1375 * connect() when a pure TCP check is used (without PROXY protocol).
1376 */
1377static int connect_chk(struct task *t)
1378{
1379 struct check *check = t->context;
1380 struct server *s = check->server;
1381 struct connection *conn = check->conn;
1382 struct protocol *proto;
1383 int ret;
Willy Tarreau02c62122014-12-08 12:11:28 +01001384 int quickack;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001385
1386 /* tcpcheck send/expect initialisation */
1387 if (check->type == PR_O2_TCPCHK_CHK)
1388 check->current_step = NULL;
1389
1390 /* prepare the check buffer.
1391 * This should not be used if check is the secondary agent check
1392 * of a server as s->proxy->check_req will relate to the
1393 * configuration of the primary check. Similarly, tcp-check uses
1394 * its own strings.
1395 */
1396 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
1397 bo_putblk(check->bo, s->proxy->check_req, s->proxy->check_len);
1398
1399 /* we want to check if this host replies to HTTP or SSLv3 requests
1400 * so we'll send the request, and won't wake the checker up now.
1401 */
1402 if ((check->type) == PR_O2_SSL3_CHK) {
1403 /* SSL requires that we put Unix time in the request */
1404 int gmt_time = htonl(date.tv_sec);
1405 memcpy(check->bo->data + 11, &gmt_time, 4);
1406 }
1407 else if ((check->type) == PR_O2_HTTP_CHK) {
1408 if (s->proxy->options2 & PR_O2_CHK_SNDST)
1409 bo_putblk(check->bo, trash.str, httpchk_build_status_header(s, trash.str, trash.size));
Cyril Bontéa448e162015-01-30 00:07:07 +01001410 /* prevent HTTP keep-alive when "http-check expect" is used */
1411 if (s->proxy->options2 & PR_O2_EXP_TYPE)
1412 bo_putstr(check->bo, "Connection: close\r\n");
Simon Hormanb00d17a2014-06-13 16:18:16 +09001413 bo_putstr(check->bo, "\r\n");
1414 *check->bo->p = '\0'; /* to make gdb output easier to read */
1415 }
1416 }
1417
1418 /* prepare a new connection */
1419 conn_init(conn);
Cyril Bonté1f96a872014-11-15 22:41:27 +01001420 conn_prepare(conn, s->check_common.proto, check->xprt);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001421 conn_attach(conn, check, &check_conn_cb);
1422 conn->target = &s->obj_type;
1423
1424 /* no client address */
1425 clear_addr(&conn->addr.from);
1426
1427 if (is_addr(&s->check_common.addr)) {
1428
1429 /* we'll connect to the check addr specified on the server */
1430 conn->addr.to = s->check_common.addr;
1431 proto = s->check_common.proto;
1432 }
1433 else {
1434 /* we'll connect to the addr on the server */
1435 conn->addr.to = s->addr;
1436 proto = s->proto;
1437 }
1438
1439 if (check->port) {
1440 set_host_port(&conn->addr.to, check->port);
1441 }
1442
Willy Tarreau02c62122014-12-08 12:11:28 +01001443 /* only plain tcp-check supports quick ACK */
1444 quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK;
1445
Willy Tarreau619a6ae2014-12-08 11:52:28 +01001446 if (check->type == PR_O2_TCPCHK_CHK && !LIST_ISEMPTY(&s->proxy->tcpcheck_rules)) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001447 struct tcpcheck_rule *r = (struct tcpcheck_rule *) s->proxy->tcpcheck_rules.n;
1448 /* if first step is a 'connect', then tcpcheck_main must run it */
1449 if (r->action == TCPCHK_ACT_CONNECT) {
1450 tcpcheck_main(conn);
1451 return SN_ERR_UP;
1452 }
Willy Tarreau02c62122014-12-08 12:11:28 +01001453 if (r->action == TCPCHK_ACT_EXPECT)
1454 quickack = 0;
Simon Hormanb00d17a2014-06-13 16:18:16 +09001455 }
1456
1457 ret = SN_ERR_INTERNAL;
1458 if (proto->connect)
Willy Tarreau02c62122014-12-08 12:11:28 +01001459 ret = proto->connect(conn, check->type, quickack ? 2 : 0);
Simon Hormanb00d17a2014-06-13 16:18:16 +09001460 conn->flags |= CO_FL_WAKE_DATA;
1461 if (s->check.send_proxy) {
1462 conn->send_proxy_ofs = 1;
1463 conn->flags |= CO_FL_SEND_PROXY;
1464 }
1465
1466 return ret;
1467}
1468
1469/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02001470 * manages a server health-check. Returns
1471 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
1472 */
Simon Horman63a4a822012-03-19 07:24:41 +09001473static struct task *process_chk(struct task *t)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001474{
Simon Horman4a741432013-02-23 15:35:38 +09001475 struct check *check = t->context;
1476 struct server *s = check->server;
1477 struct connection *conn = check->conn;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001478 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001479 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001480 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001481
Willy Tarreau2c115e52013-12-11 19:41:16 +01001482 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01001483 /* no check currently running */
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001484 if (!expired) /* woke up too early */
Willy Tarreau26c25062009-03-08 09:38:41 +01001485 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001486
Simon Horman671b6f02013-11-25 10:46:39 +09001487 /* we don't send any health-checks when the proxy is
1488 * stopped, the server should not be checked or the check
1489 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001490 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01001491 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01001492 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01001493 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001494
1495 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09001496 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001497
Willy Tarreau2c115e52013-12-11 19:41:16 +01001498 check->state |= CHK_ST_INPROGRESS;
Simon Horman4a741432013-02-23 15:35:38 +09001499 check->bi->p = check->bi->data;
1500 check->bi->i = 0;
1501 check->bo->p = check->bo->data;
1502 check->bo->o = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001503
Simon Hormanb00d17a2014-06-13 16:18:16 +09001504 ret = connect_chk(t);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001505 switch (ret) {
Simon Hormanb00d17a2014-06-13 16:18:16 +09001506 case SN_ERR_UP:
1507 return t;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001508 case SN_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001509 /* we allow up to min(inter, timeout.connect) for a connection
1510 * to establish but only when timeout.check is set
1511 * as it may be to short for a full check otherwise
1512 */
Simon Horman4a741432013-02-23 15:35:38 +09001513 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02001514
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001515 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
1516 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
1517 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001518 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01001519
1520 if (check->type)
1521 conn_data_want_recv(conn); /* prepare for reading a possible reply */
1522
Willy Tarreau5a78f362012-11-23 12:47:05 +01001523 goto reschedule;
1524
1525 case SN_ERR_SRVTO: /* ETIMEDOUT */
1526 case SN_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau4bd07de2014-01-24 16:10:57 +01001527 conn->flags |= CO_FL_ERROR;
1528 chk_report_conn_err(conn, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01001529 break;
1530 case SN_ERR_PRXCOND:
1531 case SN_ERR_RESOURCE:
1532 case SN_ERR_INTERNAL:
Willy Tarreau4bd07de2014-01-24 16:10:57 +01001533 conn->flags |= CO_FL_ERROR;
1534 chk_report_conn_err(conn, 0, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01001535 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001536 }
1537
Willy Tarreau5a78f362012-11-23 12:47:05 +01001538 /* here, we have seen a synchronous error, no fd was allocated */
Willy Tarreau6b0a8502012-11-23 08:51:32 +01001539
Willy Tarreau2c115e52013-12-11 19:41:16 +01001540 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02001541 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001542
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001543 /* we allow up to min(inter, timeout.connect) for a connection
1544 * to establish but only when timeout.check is set
1545 * as it may be to short for a full check otherwise
1546 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001547 while (tick_is_expired(t->expire, now_ms)) {
1548 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001549
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001550 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09001551 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001552
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001553 if (s->proxy->timeout.check)
1554 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001555 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001556 }
1557 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02001558 /* there was a test running.
1559 * First, let's check whether there was an uncaught error,
1560 * which can happen on connect timeout or error.
1561 */
Simon Horman9ac7cab2014-06-20 12:29:47 +09001562 if (check->result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001563 /* good connection is enough for pure TCP check */
1564 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09001565 if (check->use_ssl)
1566 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02001567 else
Simon Horman4a741432013-02-23 15:35:38 +09001568 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001569 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001570 else if ((conn->flags & CO_FL_ERROR) || expired) {
1571 chk_report_conn_err(conn, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02001572 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001573 else
1574 goto out_wait; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02001575 }
1576
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001577 /* check complete or aborted */
Willy Tarreau5ba04f62013-02-12 15:23:12 +01001578 if (conn->xprt) {
1579 /* The check was aborted and the connection was not yet closed.
1580 * This can happen upon timeout, or when an external event such
1581 * as a failed response coupled with "observe layer7" caused the
1582 * server state to be suddenly changed.
1583 */
Willy Tarreau46be2e52014-01-20 12:10:52 +01001584 conn_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001585 conn_force_close(conn);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01001586 }
1587
Willy Tarreauaf549582014-05-16 17:37:50 +02001588 if (check->result == CHK_RES_FAILED) {
1589 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02001590 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02001591 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02001592 else if (check->result == CHK_RES_CONDPASS) {
1593 /* check is OK but asks for stopping mode */
1594 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02001595 }
Willy Tarreau3e048382014-05-21 10:30:54 +02001596 else if (check->result == CHK_RES_PASSED) {
1597 /* a success was detected */
1598 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001599 }
Willy Tarreau2c115e52013-12-11 19:41:16 +01001600 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001601
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001602 rv = 0;
1603 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001604 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001605 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001606 }
Simon Horman4a741432013-02-23 15:35:38 +09001607 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001608 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01001609
1610 reschedule:
1611 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09001612 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001613 out_wait:
Willy Tarreau26c25062009-03-08 09:38:41 +01001614 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001615}
1616
Simon Horman5c942422013-11-25 10:46:32 +09001617static int start_check_task(struct check *check, int mininter,
1618 int nbcheck, int srvpos)
1619{
1620 struct task *t;
1621 /* task for the check */
1622 if ((t = task_new()) == NULL) {
1623 Alert("Starting [%s:%s] check: out of memory.\n",
1624 check->server->proxy->id, check->server->id);
1625 return 0;
1626 }
1627
1628 check->task = t;
1629 t->process = process_chk;
1630 t->context = check;
1631
Willy Tarreau1746eec2014-04-25 10:46:47 +02001632 if (mininter < srv_getinter(check))
1633 mininter = srv_getinter(check);
1634
1635 if (global.max_spread_checks && mininter > global.max_spread_checks)
1636 mininter = global.max_spread_checks;
1637
Simon Horman5c942422013-11-25 10:46:32 +09001638 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02001639 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09001640 check->start = now;
1641 task_queue(t);
1642
1643 return 1;
1644}
1645
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001646/*
1647 * Start health-check.
1648 * Returns 0 if OK, -1 if error, and prints the error in this case.
1649 */
1650int start_checks() {
1651
1652 struct proxy *px;
1653 struct server *s;
1654 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09001655 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001656
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02001657 /* 1- count the checkers to run simultaneously.
1658 * We also determine the minimum interval among all of those which
1659 * have an interval larger than SRV_CHK_INTER_THRES. This interval
1660 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04001661 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02001662 * too short an interval for all others.
1663 */
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001664 for (px = proxy; px; px = px->next) {
1665 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01001666 if (s->slowstart) {
1667 if ((t = task_new()) == NULL) {
1668 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1669 return -1;
1670 }
1671 /* We need a warmup task that will be called when the server
1672 * state switches from down to up.
1673 */
1674 s->warmup = t;
1675 t->process = server_warmup;
1676 t->context = s;
1677 t->expire = TICK_ETERNITY;
1678 }
1679
Willy Tarreaud8514a22013-12-11 21:10:14 +01001680 if (s->check.state & CHK_ST_CONFIGURED) {
1681 nbcheck++;
1682 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
1683 (!mininter || mininter > srv_getinter(&s->check)))
1684 mininter = srv_getinter(&s->check);
1685 }
Willy Tarreau15f39102013-12-11 20:41:18 +01001686
Willy Tarreaud8514a22013-12-11 21:10:14 +01001687 if (s->agent.state & CHK_ST_CONFIGURED) {
1688 nbcheck++;
1689 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
1690 (!mininter || mininter > srv_getinter(&s->agent)))
1691 mininter = srv_getinter(&s->agent);
1692 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001693 }
1694 }
1695
Simon Horman4a741432013-02-23 15:35:38 +09001696 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001697 return 0;
1698
1699 srand((unsigned)time(NULL));
1700
1701 /*
1702 * 2- start them as far as possible from each others. For this, we will
1703 * start them after their interval set to the min interval divided by
1704 * the number of servers, weighted by the server's position in the list.
1705 */
1706 for (px = proxy; px; px = px->next) {
1707 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09001708 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01001709 if (s->check.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09001710 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
1711 return -1;
1712 srvpos++;
1713 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001714
Simon Hormand60d6912013-11-25 10:46:36 +09001715 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01001716 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09001717 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
1718 return -1;
1719 }
1720 srvpos++;
1721 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001722 }
1723 }
1724 return 0;
1725}
Willy Tarreaubaaee002006-06-26 02:48:02 +02001726
1727/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02001728 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01001729 * The buffer MUST be terminated by a null byte before calling this function.
1730 * Sets server status appropriately. The caller is responsible for ensuring
1731 * that the buffer contains at least 13 characters. If <done> is zero, we may
1732 * return 0 to indicate that data is required to decide of a match.
1733 */
1734static int httpchk_expect(struct server *s, int done)
1735{
1736 static char status_msg[] = "HTTP status check returned code <000>";
1737 char status_code[] = "000";
1738 char *contentptr;
1739 int crlf;
1740 int ret;
1741
1742 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
1743 case PR_O2_EXP_STS:
1744 case PR_O2_EXP_RSTS:
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001745 memcpy(status_code, s->check.bi->data + 9, 3);
1746 memcpy(status_msg + strlen(status_msg) - 4, s->check.bi->data + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01001747
1748 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
1749 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
1750 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02001751 ret = regex_exec(s->proxy->expect_regex, status_code);
Willy Tarreaubd741542010-03-16 18:46:54 +01001752
1753 /* we necessarily have the response, so there are no partial failures */
1754 if (s->proxy->options2 & PR_O2_EXP_INV)
1755 ret = !ret;
1756
Simon Horman4a741432013-02-23 15:35:38 +09001757 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01001758 break;
1759
1760 case PR_O2_EXP_STR:
1761 case PR_O2_EXP_RSTR:
1762 /* very simple response parser: ignore CR and only count consecutive LFs,
1763 * stop with contentptr pointing to first char after the double CRLF or
1764 * to '\0' if crlf < 2.
1765 */
1766 crlf = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001767 for (contentptr = s->check.bi->data; *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01001768 if (crlf >= 2)
1769 break;
1770 if (*contentptr == '\r')
1771 continue;
1772 else if (*contentptr == '\n')
1773 crlf++;
1774 else
1775 crlf = 0;
1776 }
1777
1778 /* Check that response contains a body... */
1779 if (crlf < 2) {
1780 if (!done)
1781 return 0;
1782
Simon Horman4a741432013-02-23 15:35:38 +09001783 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001784 "HTTP content check could not find a response body");
1785 return 1;
1786 }
1787
1788 /* Check that response body is not empty... */
1789 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02001790 if (!done)
1791 return 0;
1792
Simon Horman4a741432013-02-23 15:35:38 +09001793 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001794 "HTTP content check found empty response body");
1795 return 1;
1796 }
1797
1798 /* Check the response content against the supplied string
1799 * or regex... */
1800 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
1801 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
1802 else
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02001803 ret = regex_exec(s->proxy->expect_regex, contentptr);
Willy Tarreaubd741542010-03-16 18:46:54 +01001804
1805 /* if we don't match, we may need to wait more */
1806 if (!ret && !done)
1807 return 0;
1808
1809 if (ret) {
1810 /* content matched */
1811 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09001812 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001813 "HTTP check matched unwanted content");
1814 else
Simon Horman4a741432013-02-23 15:35:38 +09001815 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01001816 "HTTP content check matched");
1817 }
1818 else {
1819 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09001820 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01001821 "HTTP check did not match unwanted content");
1822 else
Simon Horman4a741432013-02-23 15:35:38 +09001823 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001824 "HTTP content check did not match");
1825 }
1826 break;
1827 }
1828 return 1;
1829}
1830
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001831/*
1832 * return the id of a step in a send/expect session
1833 */
1834static int tcpcheck_get_step_id(struct server *s)
1835{
1836 struct tcpcheck_rule *cur = NULL, *next = NULL;
1837 int i = 0;
1838
Willy Tarreau90055f22014-10-02 14:51:02 +02001839 /* not even started anything yet => step 0 = initial connect */
1840 if (!s->check.current_step)
1841 return 0;
1842
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001843 cur = s->check.last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001844
1845 /* no step => first step */
1846 if (cur == NULL)
1847 return 1;
1848
1849 /* increment i until current step */
1850 list_for_each_entry(next, &s->proxy->tcpcheck_rules, list) {
1851 if (next->list.p == &cur->list)
1852 break;
1853 ++i;
1854 }
1855
1856 return i;
1857}
1858
1859static void tcpcheck_main(struct connection *conn)
1860{
1861 char *contentptr;
Willy Tarreau02c62122014-12-08 12:11:28 +01001862 struct tcpcheck_rule *cur, *next;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001863 int done = 0, ret = 0;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001864 struct check *check = conn->owner;
1865 struct server *s = check->server;
1866 struct task *t = check->task;
Willy Tarreaue61737a2014-10-02 14:30:14 +02001867 struct list *head = &s->proxy->tcpcheck_rules;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001868
Willy Tarreaue61737a2014-10-02 14:30:14 +02001869 /* here, we know that the check is complete or that it failed */
1870 if (check->result != CHK_RES_UNKNOWN)
1871 goto out_end_tcpcheck;
1872
1873 /* We have 4 possibilities here :
1874 * 1. we've not yet attempted step 1, and step 1 is a connect, so no
1875 * connection attempt was made yet ;
1876 * 2. we've not yet attempted step 1, and step 1 is a not connect or
1877 * does not exist (no rule), so a connection attempt was made
1878 * before coming here.
1879 * 3. we're coming back after having started with step 1, so we may
1880 * be waiting for a connection attempt to complete.
1881 * 4. the connection + handshake are complete
1882 *
1883 * #2 and #3 are quite similar, we want both the connection and the
1884 * handshake to complete before going any further. Thus we must always
1885 * wait for a connection to complete unless we're before and existing
1886 * step 1.
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001887 */
Willy Tarreaue61737a2014-10-02 14:30:14 +02001888 if ((!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE)) &&
1889 (check->current_step || LIST_ISEMPTY(head))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001890 /* we allow up to min(inter, timeout.connect) for a connection
1891 * to establish but only when timeout.check is set
1892 * as it may be to short for a full check otherwise
1893 */
1894 while (tick_is_expired(t->expire, now_ms)) {
1895 int t_con;
1896
1897 t_con = tick_add(t->expire, s->proxy->timeout.connect);
1898 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
1899
1900 if (s->proxy->timeout.check)
1901 t->expire = tick_first(t->expire, t_con);
1902 }
1903 return;
1904 }
1905
Willy Tarreaue61737a2014-10-02 14:30:14 +02001906 /* special case: option tcp-check with no rule, a connect is enough */
1907 if (LIST_ISEMPTY(head)) {
1908 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001909 goto out_end_tcpcheck;
Willy Tarreaue61737a2014-10-02 14:30:14 +02001910 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001911
Willy Tarreau90055f22014-10-02 14:51:02 +02001912 /* no step means first step initialisation */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001913 if (check->current_step == NULL) {
Willy Tarreau90055f22014-10-02 14:51:02 +02001914 check->last_started_step = NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001915 check->bo->p = check->bo->data;
1916 check->bo->o = 0;
1917 check->bi->p = check->bi->data;
1918 check->bi->i = 0;
1919 cur = check->current_step = LIST_ELEM(head->n, struct tcpcheck_rule *, list);
1920 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1921 if (s->proxy->timeout.check)
1922 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
1923 }
1924 /* keep on processing step */
1925 else {
1926 cur = check->current_step;
1927 }
1928
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001929 /* It's only the rules which will enable send/recv */
1930 __conn_data_stop_both(conn);
1931
Willy Tarreauabca5b62013-12-06 14:19:25 +01001932 while (1) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001933 /* we have to try to flush the output buffer before reading, at the end,
1934 * or if we're about to send a string that does not fit in the remaining space.
1935 */
1936 if (check->bo->o &&
1937 (&cur->list == head ||
1938 check->current_step->action != TCPCHK_ACT_SEND ||
1939 check->current_step->string_len >= buffer_total_space(check->bo))) {
1940
Willy Tarreau1049b1f2014-02-02 01:51:17 +01001941 if (conn->xprt->snd_buf(conn, check->bo, 0) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001942 if (conn->flags & CO_FL_ERROR) {
1943 chk_report_conn_err(conn, errno, 0);
1944 __conn_data_stop_both(conn);
1945 goto out_end_tcpcheck;
1946 }
1947 goto out_need_io;
Willy Tarreauabca5b62013-12-06 14:19:25 +01001948 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01001949 }
1950
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001951 /* did we reach the end ? If so, let's check that everything was sent */
1952 if (&cur->list == head) {
1953 if (check->bo->o)
1954 goto out_need_io;
Willy Tarreauabca5b62013-12-06 14:19:25 +01001955 break;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001956 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01001957
Willy Tarreau02c62122014-12-08 12:11:28 +01001958 /* have 'next' point to the next rule or NULL if we're on the last one */
1959 next = (struct tcpcheck_rule *)cur->list.n;
1960 if (&next->list == head)
1961 next = NULL;
1962
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001963 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
1964 struct protocol *proto;
1965 struct xprt_ops *xprt;
1966
1967 /* mark the step as started */
1968 check->last_started_step = check->current_step;
1969 /* first, shut existing connection */
1970 conn_force_close(conn);
1971
1972 /* prepare new connection */
1973 /* initialization */
1974 conn_init(conn);
1975 conn_attach(conn, check, &check_conn_cb);
1976 conn->target = &s->obj_type;
1977
1978 /* no client address */
1979 clear_addr(&conn->addr.from);
1980
Willy Tarreau640556c2014-05-09 23:38:15 +02001981 if (is_addr(&s->check_common.addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001982 /* we'll connect to the check addr specified on the server */
1983 conn->addr.to = s->check_common.addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001984 proto = s->check_common.proto;
1985 }
1986 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001987 /* we'll connect to the addr on the server */
1988 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001989 proto = s->proto;
1990 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001991
1992 /* port */
1993 if (check->current_step->port)
1994 set_host_port(&conn->addr.to, check->current_step->port);
1995 else if (check->port)
1996 set_host_port(&conn->addr.to, check->port);
1997
1998#ifdef USE_OPENSSL
1999 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
2000 xprt = &ssl_sock;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002001 }
2002 else {
2003 xprt = &raw_sock;
2004 }
2005#else /* USE_OPENSSL */
2006 xprt = &raw_sock;
2007#endif /* USE_OPENSSL */
2008 conn_prepare(conn, proto, xprt);
2009
2010 ret = SN_ERR_INTERNAL;
2011 if (proto->connect)
Willy Tarreau02c62122014-12-08 12:11:28 +01002012 ret = proto->connect(conn,
2013 1 /* I/O polling is always needed */,
2014 (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2);
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002015 conn->flags |= CO_FL_WAKE_DATA;
2016 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
2017 conn->send_proxy_ofs = 1;
2018 conn->flags |= CO_FL_SEND_PROXY;
2019 }
2020
2021 /* It can return one of :
2022 * - SN_ERR_NONE if everything's OK
2023 * - SN_ERR_SRVTO if there are no more servers
2024 * - SN_ERR_SRVCL if the connection was refused by the server
2025 * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
2026 * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
2027 * - SN_ERR_INTERNAL for any other purely internal errors
2028 * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted.
2029 * Note that we try to prevent the network stack from sending the ACK during the
2030 * connect() when a pure TCP check is used (without PROXY protocol).
2031 */
2032 switch (ret) {
2033 case SN_ERR_NONE:
2034 /* we allow up to min(inter, timeout.connect) for a connection
2035 * to establish but only when timeout.check is set
2036 * as it may be to short for a full check otherwise
2037 */
2038 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
2039
2040 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
2041 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
2042 t->expire = tick_first(t->expire, t_con);
2043 }
2044 break;
2045 case SN_ERR_SRVTO: /* ETIMEDOUT */
2046 case SN_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
2047 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
2048 tcpcheck_get_step_id(s), strerror(errno));
2049 set_server_check_status(check, HCHK_STATUS_L4CON, trash.str);
2050 goto out_end_tcpcheck;
2051 case SN_ERR_PRXCOND:
2052 case SN_ERR_RESOURCE:
2053 case SN_ERR_INTERNAL:
2054 chunk_printf(&trash, "TCPCHK error establishing connection at step %d",
2055 tcpcheck_get_step_id(s));
2056 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2057 goto out_end_tcpcheck;
2058 }
2059
2060 /* allow next rule */
2061 cur = (struct tcpcheck_rule *)cur->list.n;
2062 check->current_step = cur;
2063
2064 /* don't do anything until the connection is established */
2065 if (!(conn->flags & CO_FL_CONNECTED)) {
2066 /* update expire time, should be done by process_chk */
2067 /* we allow up to min(inter, timeout.connect) for a connection
2068 * to establish but only when timeout.check is set
2069 * as it may be to short for a full check otherwise
2070 */
2071 while (tick_is_expired(t->expire, now_ms)) {
2072 int t_con;
2073
2074 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2075 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2076
2077 if (s->proxy->timeout.check)
2078 t->expire = tick_first(t->expire, t_con);
2079 }
2080 return;
2081 }
2082
2083 } /* end 'connect' */
2084 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2085 /* mark the step as started */
2086 check->last_started_step = check->current_step;
2087
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002088 /* reset the read buffer */
2089 if (*check->bi->data != '\0') {
2090 *check->bi->data = '\0';
2091 check->bi->i = 0;
2092 }
2093
2094 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
2095 conn->flags |= CO_FL_ERROR;
2096 chk_report_conn_err(conn, 0, 0);
2097 goto out_end_tcpcheck;
2098 }
2099
Willy Tarreauabca5b62013-12-06 14:19:25 +01002100 if (check->current_step->string_len >= check->bo->size) {
2101 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%d) at step %d",
2102 check->current_step->string_len, check->bo->size,
2103 tcpcheck_get_step_id(s));
2104 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2105 goto out_end_tcpcheck;
2106 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002107
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002108 /* do not try to send if there is no space */
2109 if (check->current_step->string_len >= buffer_total_space(check->bo))
2110 continue;
2111
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002112 bo_putblk(check->bo, check->current_step->string, check->current_step->string_len);
2113 *check->bo->p = '\0'; /* to make gdb output easier to read */
2114
Willy Tarreauabca5b62013-12-06 14:19:25 +01002115 /* go to next rule and try to send */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002116 cur = (struct tcpcheck_rule *)cur->list.n;
2117 check->current_step = cur;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002118 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002119 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002120 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002121 goto out_end_tcpcheck;
2122
Willy Tarreau310987a2014-01-22 19:46:33 +01002123 if (conn->xprt->rcv_buf(conn, check->bi, check->bi->size) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002124 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
2125 done = 1;
2126 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
2127 /* Report network errors only if we got no other data. Otherwise
2128 * we'll let the upper layers decide whether the response is OK
2129 * or not. It is very common that an RST sent by the server is
2130 * reported as an error just after the last data chunk.
2131 */
2132 chk_report_conn_err(conn, errno, 0);
2133 goto out_end_tcpcheck;
2134 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002135 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002136 else
2137 goto out_need_io;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002138 }
2139
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002140 /* mark the step as started */
2141 check->last_started_step = check->current_step;
2142
2143
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002144 /* Intermediate or complete response received.
2145 * Terminate string in check->bi->data buffer.
2146 */
2147 if (check->bi->i < check->bi->size) {
2148 check->bi->data[check->bi->i] = '\0';
2149 }
2150 else {
2151 check->bi->data[check->bi->i - 1] = '\0';
2152 done = 1; /* buffer full, don't wait for more data */
2153 }
2154
2155 contentptr = check->bi->data;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002156
2157 /* Check that response body is not empty... */
Willy Tarreauec6b0122014-05-13 17:57:29 +02002158 if (!check->bi->i) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002159 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002160 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002161
2162 /* empty response */
2163 chunk_printf(&trash, "TCPCHK got an empty response at step %d",
2164 tcpcheck_get_step_id(s));
2165 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2166
2167 goto out_end_tcpcheck;
2168 }
2169
2170 if (!done && (cur->string != NULL) && (check->bi->i < cur->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002171 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002172
Willy Tarreaua970c282013-12-06 12:47:19 +01002173 tcpcheck_expect:
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002174 if (cur->string != NULL)
Willy Tarreauec6b0122014-05-13 17:57:29 +02002175 ret = my_memmem(contentptr, check->bi->i, cur->string, cur->string_len) != NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002176 else if (cur->expect_regex != NULL)
Thierry FOURNIER09af0d62014-06-18 11:35:54 +02002177 ret = regex_exec(cur->expect_regex, contentptr);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002178
2179 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002180 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002181
2182 /* matched */
2183 if (ret) {
2184 /* matched but we did not want to => ERROR */
2185 if (cur->inverse) {
2186 /* we were looking for a string */
2187 if (cur->string != NULL) {
2188 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
2189 cur->string, tcpcheck_get_step_id(s));
2190 }
2191 else {
2192 /* we were looking for a regex */
2193 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d",
2194 tcpcheck_get_step_id(s));
2195 }
2196 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2197 goto out_end_tcpcheck;
2198 }
2199 /* matched and was supposed to => OK, next step */
2200 else {
2201 cur = (struct tcpcheck_rule*)cur->list.n;
2202 check->current_step = cur;
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002203 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002204 goto tcpcheck_expect;
2205 __conn_data_stop_recv(conn);
2206 }
2207 }
2208 else {
2209 /* not matched */
2210 /* not matched and was not supposed to => OK, next step */
2211 if (cur->inverse) {
2212 cur = (struct tcpcheck_rule*)cur->list.n;
2213 check->current_step = cur;
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002214 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002215 goto tcpcheck_expect;
2216 __conn_data_stop_recv(conn);
2217 }
2218 /* not matched but was supposed to => ERROR */
2219 else {
2220 /* we were looking for a string */
2221 if (cur->string != NULL) {
2222 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
2223 cur->string, tcpcheck_get_step_id(s));
2224 }
2225 else {
2226 /* we were looking for a regex */
2227 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
2228 tcpcheck_get_step_id(s));
2229 }
2230 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2231 goto out_end_tcpcheck;
2232 }
2233 }
2234 } /* end expect */
2235 } /* end loop over double chained step list */
2236
2237 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
2238 goto out_end_tcpcheck;
2239
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002240 out_need_io:
2241 if (check->bo->o)
2242 __conn_data_want_send(conn);
2243
2244 if (check->current_step->action == TCPCHK_ACT_EXPECT)
2245 __conn_data_want_recv(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002246 return;
2247
2248 out_end_tcpcheck:
2249 /* collect possible new errors */
2250 if (conn->flags & CO_FL_ERROR)
2251 chk_report_conn_err(conn, 0, 0);
2252
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002253 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002254 check->current_step = NULL;
2255
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002256 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002257 conn->flags |= CO_FL_ERROR;
2258
2259 __conn_data_stop_both(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002260
2261 return;
2262}
2263
2264
Willy Tarreaubd741542010-03-16 18:46:54 +01002265/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02002266 * Local variables:
2267 * c-indent-level: 8
2268 * c-basic-offset: 8
2269 * End:
2270 */