blob: e54e46a7004c949a60ceb8372e141b6022d81aff [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) ||
249 (check->status >= HCHK_STATUS_L7TOUT)) &&
250 (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 Tarreau25e2ab52013-12-04 11:17:05 +0100583
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100584 if (check->result != CHK_RES_UNKNOWN)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100585 return;
586
587 errno = errno_bck;
588 if (!errno || errno == EAGAIN)
589 retrieve_errno_from_socket(conn);
590
591 if (!(conn->flags & CO_FL_ERROR) && !expired)
592 return;
593
594 /* we'll try to build a meaningful error message depending on the
595 * context of the error possibly present in conn->err_code, and the
596 * socket error possibly collected above. This is useful to know the
597 * exact step of the L6 layer (eg: SSL handshake).
598 */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200599 chk = get_trash_chunk();
600
601 if (check->type == PR_O2_TCPCHK_CHK) {
602 chunk_printf(chk, " at step %d of tcp-check", tcpcheck_get_step_id(check->server));
603 /* we were looking for a string */
Baptiste Assmann69e273f2013-12-11 00:52:19 +0100604 if (check->current_step && check->current_step->action == TCPCHK_ACT_CONNECT) {
605 chunk_appendf(chk, " (connect)");
606 }
607 else if (check->current_step && check->current_step->action == TCPCHK_ACT_EXPECT) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200608 if (check->current_step->string)
609 chunk_appendf(chk, " (string '%s')", check->current_step->string);
610 else if (check->current_step->expect_regex)
611 chunk_appendf(chk, " (expect regex)");
612 }
613 else if (check->current_step && check->current_step->action == TCPCHK_ACT_SEND) {
614 chunk_appendf(chk, " (send)");
615 }
616 }
617
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100618 if (conn->err_code) {
619 if (errno && errno != EAGAIN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200620 chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100621 else
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200622 chunk_printf(&trash, "%s%s", conn_err_code_str(conn), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100623 err_msg = trash.str;
624 }
625 else {
626 if (errno && errno != EAGAIN) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200627 chunk_printf(&trash, "%s%s", strerror(errno), chk->str);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100628 err_msg = trash.str;
629 }
630 else {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200631 err_msg = chk->str;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100632 }
633 }
634
635 if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) {
636 /* L4 not established (yet) */
637 if (conn->flags & CO_FL_ERROR)
638 set_server_check_status(check, HCHK_STATUS_L4CON, err_msg);
639 else if (expired)
640 set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg);
641 }
642 else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) {
643 /* L6 not established (yet) */
644 if (conn->flags & CO_FL_ERROR)
645 set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg);
646 else if (expired)
647 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
648 }
649 else if (conn->flags & CO_FL_ERROR) {
650 /* I/O error after connection was established and before we could diagnose */
651 set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg);
652 }
653 else if (expired) {
654 /* connection established but expired check */
655 if (check->type == PR_O2_SSL3_CHK)
656 set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg);
657 else /* HTTP, SMTP, ... */
658 set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg);
659 }
660
661 return;
662}
663
Willy Tarreaubaaee002006-06-26 02:48:02 +0200664/*
665 * This function is used only for server health-checks. It handles
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200666 * the connection acknowledgement. If the proxy requires L7 health-checks,
667 * it sends the request. In other cases, it calls set_server_check_status()
Simon Horman4a741432013-02-23 15:35:38 +0900668 * to set check->status, check->duration and check->result.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200669 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200670static void event_srv_chk_w(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200671{
Simon Horman4a741432013-02-23 15:35:38 +0900672 struct check *check = conn->owner;
673 struct server *s = check->server;
Simon Horman4a741432013-02-23 15:35:38 +0900674 struct task *t = check->task;
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200675
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100676 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100677 goto out_wakeup;
678
Willy Tarreau310987a2014-01-22 19:46:33 +0100679 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100680 return;
681
Willy Tarreau20a18342013-12-05 00:31:46 +0100682 if (retrieve_errno_from_socket(conn)) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100683 chk_report_conn_err(conn, errno, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100684 __conn_data_stop_both(conn);
685 goto out_wakeup;
686 }
Krzysztof Piotr Oledzki6492db52010-01-02 22:03:01 +0100687
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100688 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
Willy Tarreau20a18342013-12-05 00:31:46 +0100689 /* if the output is closed, we can't do anything */
690 conn->flags |= CO_FL_ERROR;
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100691 chk_report_conn_err(conn, 0, 0);
Willy Tarreau20a18342013-12-05 00:31:46 +0100692 goto out_wakeup;
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200693 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200694
Willy Tarreau06559ac2013-12-05 01:53:08 +0100695 /* here, we know that the connection is established. That's enough for
696 * a pure TCP check.
697 */
698 if (!check->type)
699 goto out_wakeup;
700
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200701 if (check->type == PR_O2_TCPCHK_CHK) {
702 tcpcheck_main(conn);
703 return;
704 }
705
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100706 if (check->bo->o) {
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100707 conn->xprt->snd_buf(conn, check->bo, 0);
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100708 if (conn->flags & CO_FL_ERROR) {
709 chk_report_conn_err(conn, errno, 0);
710 __conn_data_stop_both(conn);
711 goto out_wakeup;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200712 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100713 if (check->bo->o)
714 return;
715 }
Willy Tarreau6996e152007-04-30 14:37:43 +0200716
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100717 /* full request sent, we allow up to <timeout.check> if nonzero for a response */
718 if (s->proxy->timeout.check) {
719 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
720 task_queue(t);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200721 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100722 goto out_nowake;
723
Willy Tarreau83749182007-04-15 20:56:27 +0200724 out_wakeup:
Willy Tarreaufdccded2008-08-29 18:19:04 +0200725 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau83749182007-04-15 20:56:27 +0200726 out_nowake:
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200727 __conn_data_stop_send(conn); /* nothing more to write */
Willy Tarreaubaaee002006-06-26 02:48:02 +0200728}
729
Willy Tarreaubaaee002006-06-26 02:48:02 +0200730/*
Willy Tarreauf3c69202006-07-09 16:42:34 +0200731 * This function is used only for server health-checks. It handles the server's
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +0200732 * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls
Simon Horman4a741432013-02-23 15:35:38 +0900733 * set_server_check_status() to update check->status, check->duration
734 * and check->result.
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200735
736 * The set_server_check_status function is called with HCHK_STATUS_L7OKD if
737 * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server
738 * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in
739 * response to an SSL HELLO (the principle is that this is enough to
740 * distinguish between an SSL server and a pure TCP relay). All other cases will
741 * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP,
742 * etc.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200743 */
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200744static void event_srv_chk_r(struct connection *conn)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200745{
Simon Horman4a741432013-02-23 15:35:38 +0900746 struct check *check = conn->owner;
747 struct server *s = check->server;
748 struct task *t = check->task;
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200749 char *desc;
Willy Tarreau03938182010-03-17 21:52:07 +0100750 int done;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +0200751 unsigned short msglen;
Willy Tarreau83749182007-04-15 20:56:27 +0200752
Willy Tarreau6aaa1b82013-12-11 17:09:34 +0100753 if (unlikely(check->result == CHK_RES_FAILED))
Willy Tarreau83749182007-04-15 20:56:27 +0200754 goto out_wakeup;
Willy Tarreau83749182007-04-15 20:56:27 +0200755
Willy Tarreau310987a2014-01-22 19:46:33 +0100756 if (conn->flags & CO_FL_HANDSHAKE)
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200757 return;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +0200758
759 if (check->type == PR_O2_TCPCHK_CHK) {
760 tcpcheck_main(conn);
761 return;
762 }
Willy Tarreaufb56aab2012-09-28 14:40:02 +0200763
Willy Tarreau83749182007-04-15 20:56:27 +0200764 /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available
765 * but the connection was closed on the remote end. Fortunately, recv still
766 * works correctly and we don't need to do the getsockopt() on linux.
767 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000768
769 /* Set buffer to point to the end of the data already read, and check
770 * that there is free space remaining. If the buffer is full, proceed
771 * with running the checks without attempting another socket read.
772 */
Nick Chalk57b1bf72010-03-16 15:50:46 +0000773
Willy Tarreau03938182010-03-17 21:52:07 +0100774 done = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +0000775
Simon Horman4a741432013-02-23 15:35:38 +0900776 conn->xprt->rcv_buf(conn, check->bi, check->bi->size);
Willy Tarreauf1503172012-09-28 19:39:36 +0200777 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
Willy Tarreau03938182010-03-17 21:52:07 +0100778 done = 1;
Simon Horman4a741432013-02-23 15:35:38 +0900779 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
Willy Tarreauf1503172012-09-28 19:39:36 +0200780 /* Report network errors only if we got no other data. Otherwise
781 * we'll let the upper layers decide whether the response is OK
782 * or not. It is very common that an RST sent by the server is
783 * reported as an error just after the last data chunk.
784 */
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100785 chk_report_conn_err(conn, errno, 0);
Willy Tarreauc1a07962010-03-16 20:55:43 +0100786 goto out_wakeup;
787 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200788 }
789
Willy Tarreau25e2ab52013-12-04 11:17:05 +0100790
Willy Tarreau03938182010-03-17 21:52:07 +0100791 /* Intermediate or complete response received.
Simon Horman4a741432013-02-23 15:35:38 +0900792 * Terminate string in check->bi->data buffer.
Willy Tarreau03938182010-03-17 21:52:07 +0100793 */
Simon Horman4a741432013-02-23 15:35:38 +0900794 if (check->bi->i < check->bi->size)
795 check->bi->data[check->bi->i] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100796 else {
Simon Horman4a741432013-02-23 15:35:38 +0900797 check->bi->data[check->bi->i - 1] = '\0';
Willy Tarreau03938182010-03-17 21:52:07 +0100798 done = 1; /* buffer full, don't wait for more data */
799 }
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200800
Nick Chalk57b1bf72010-03-16 15:50:46 +0000801 /* Run the checks... */
Simon Horman4a741432013-02-23 15:35:38 +0900802 switch (check->type) {
Willy Tarreau1620ec32011-08-06 17:05:02 +0200803 case PR_O2_HTTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900804 if (!done && check->bi->i < strlen("HTTP/1.0 000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100805 goto wait_more_data;
806
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100807 /* Check if the server speaks HTTP 1.X */
Simon Horman4a741432013-02-23 15:35:38 +0900808 if ((check->bi->i < strlen("HTTP/1.0 000\r")) ||
809 (memcmp(check->bi->data, "HTTP/1.", 7) != 0 ||
810 (*(check->bi->data + 12) != ' ' && *(check->bi->data + 12) != '\r')) ||
811 !isdigit((unsigned char) *(check->bi->data + 9)) || !isdigit((unsigned char) *(check->bi->data + 10)) ||
812 !isdigit((unsigned char) *(check->bi->data + 11))) {
813 cut_crlf(check->bi->data);
814 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200815
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100816 goto out_wakeup;
817 }
818
Simon Horman4a741432013-02-23 15:35:38 +0900819 check->code = str2uic(check->bi->data + 9);
820 desc = ltrim(check->bi->data + 12, ' ');
Nick Chalk57b1bf72010-03-16 15:50:46 +0000821
Willy Tarreaubd741542010-03-16 18:46:54 +0100822 if ((s->proxy->options & PR_O_DISABLE404) &&
Willy Tarreau892337c2014-05-13 23:41:20 +0200823 (s->state != SRV_ST_STOPPED) && (check->code == 404)) {
Nick Chalk57b1bf72010-03-16 15:50:46 +0000824 /* 404 may be accepted as "stopping" only if the server was up */
825 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900826 set_server_check_status(check, HCHK_STATUS_L7OKCD, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000827 }
Willy Tarreaubd741542010-03-16 18:46:54 +0100828 else if (s->proxy->options2 & PR_O2_EXP_TYPE) {
829 /* Run content verification check... We know we have at least 13 chars */
830 if (!httpchk_expect(s, done))
831 goto wait_more_data;
832 }
833 /* check the reply : HTTP/1.X 2xx and 3xx are OK */
Simon Horman4a741432013-02-23 15:35:38 +0900834 else if (*(check->bi->data + 9) == '2' || *(check->bi->data + 9) == '3') {
Willy Tarreaubd741542010-03-16 18:46:54 +0100835 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900836 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Willy Tarreaubd741542010-03-16 18:46:54 +0100837 }
Nick Chalk57b1bf72010-03-16 15:50:46 +0000838 else {
839 cut_crlf(desc);
Simon Horman4a741432013-02-23 15:35:38 +0900840 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Nick Chalk57b1bf72010-03-16 15:50:46 +0000841 }
Willy Tarreau1620ec32011-08-06 17:05:02 +0200842 break;
843
844 case PR_O2_SSL3_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900845 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +0100846 goto wait_more_data;
847
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100848 /* Check for SSLv3 alert or handshake */
Simon Horman4a741432013-02-23 15:35:38 +0900849 if ((check->bi->i >= 5) && (*check->bi->data == 0x15 || *check->bi->data == 0x16))
850 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200851 else
Simon Horman4a741432013-02-23 15:35:38 +0900852 set_server_check_status(check, HCHK_STATUS_L6RSP, NULL);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200853 break;
854
855 case PR_O2_SMTP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +0900856 if (!done && check->bi->i < strlen("000\r"))
Willy Tarreau03938182010-03-17 21:52:07 +0100857 goto wait_more_data;
858
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200859 /* Check if the server speaks SMTP */
Simon Horman4a741432013-02-23 15:35:38 +0900860 if ((check->bi->i < strlen("000\r")) ||
861 (*(check->bi->data + 3) != ' ' && *(check->bi->data + 3) != '\r') ||
862 !isdigit((unsigned char) *check->bi->data) || !isdigit((unsigned char) *(check->bi->data + 1)) ||
863 !isdigit((unsigned char) *(check->bi->data + 2))) {
864 cut_crlf(check->bi->data);
865 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200866
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200867 goto out_wakeup;
868 }
869
Simon Horman4a741432013-02-23 15:35:38 +0900870 check->code = str2uic(check->bi->data);
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +0200871
Simon Horman4a741432013-02-23 15:35:38 +0900872 desc = ltrim(check->bi->data + 3, ' ');
Krzysztof Piotr Oledzkif7089f52009-10-10 21:06:49 +0200873 cut_crlf(desc);
874
Willy Tarreauc7dd71a2007-11-30 08:33:21 +0100875 /* Check for SMTP code 2xx (should be 250) */
Simon Horman4a741432013-02-23 15:35:38 +0900876 if (*check->bi->data == '2')
877 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Krzysztof Piotr Oledzki213014e2009-09-27 15:50:02 +0200878 else
Simon Horman4a741432013-02-23 15:35:38 +0900879 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Willy Tarreau1620ec32011-08-06 17:05:02 +0200880 break;
881
Simon Hormana2b9dad2013-02-12 10:45:54 +0900882 case PR_O2_LB_AGENT_CHK: {
Willy Tarreau81f5d942013-12-09 20:51:51 +0100883 int status = HCHK_STATUS_CHECKED;
884 const char *hs = NULL; /* health status */
885 const char *as = NULL; /* admin status */
886 const char *ps = NULL; /* performance status */
887 const char *err = NULL; /* first error to report */
888 const char *wrn = NULL; /* first warning to report */
889 char *cmd, *p;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900890
Willy Tarreau81f5d942013-12-09 20:51:51 +0100891 /* We're getting an agent check response. The agent could
892 * have been disabled in the mean time with a long check
893 * still pending. It is important that we ignore the whole
894 * response.
895 */
896 if (!(check->server->agent.state & CHK_ST_ENABLED))
897 break;
898
899 /* The agent supports strings made of a single line ended by the
900 * first CR ('\r') or LF ('\n'). This line is composed of words
901 * delimited by spaces (' '), tabs ('\t'), or commas (','). The
902 * line may optionally contained a description of a state change
903 * after a sharp ('#'), which is only considered if a health state
904 * is announced.
905 *
906 * Words may be composed of :
907 * - a numeric weight suffixed by the percent character ('%').
908 * - a health status among "up", "down", "stopped", and "fail".
909 * - an admin status among "ready", "drain", "maint".
910 *
911 * These words may appear in any order. If multiple words of the
912 * same category appear, the last one wins.
913 */
914
Willy Tarreau9809b782013-12-11 21:40:11 +0100915 p = check->bi->data;
916 while (*p && *p != '\n' && *p != '\r')
917 p++;
918
919 if (!*p) {
920 if (!done)
921 goto wait_more_data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900922
Willy Tarreau9809b782013-12-11 21:40:11 +0100923 /* at least inform the admin that the agent is mis-behaving */
924 set_server_check_status(check, check->status, "Ignoring incomplete line from agent");
925 break;
926 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100927
Willy Tarreau9809b782013-12-11 21:40:11 +0100928 *p = 0;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100929 cmd = check->bi->data;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900930
Willy Tarreau81f5d942013-12-09 20:51:51 +0100931 while (*cmd) {
932 /* look for next word */
933 if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') {
934 cmd++;
935 continue;
936 }
Simon Horman671b6f02013-11-25 10:46:39 +0900937
Willy Tarreau81f5d942013-12-09 20:51:51 +0100938 if (*cmd == '#') {
939 /* this is the beginning of a health status description,
940 * skip the sharp and blanks.
941 */
942 cmd++;
943 while (*cmd == '\t' || *cmd == ' ')
944 cmd++;
Simon Horman671b6f02013-11-25 10:46:39 +0900945 break;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900946 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100947
948 /* find the end of the word so that we have a null-terminated
949 * word between <cmd> and <p>.
950 */
951 p = cmd + 1;
952 while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',')
953 p++;
954 if (*p)
955 *p++ = 0;
956
957 /* first, health statuses */
958 if (strcasecmp(cmd, "up") == 0) {
959 check->health = check->rise + check->fall - 1;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900960 status = HCHK_STATUS_L7OKD;
Willy Tarreau81f5d942013-12-09 20:51:51 +0100961 hs = cmd;
962 }
963 else if (strcasecmp(cmd, "down") == 0) {
964 check->health = 0;
965 status = HCHK_STATUS_L7STS;
966 hs = cmd;
Simon Hormana2b9dad2013-02-12 10:45:54 +0900967 }
Willy Tarreau81f5d942013-12-09 20:51:51 +0100968 else if (strcasecmp(cmd, "stopped") == 0) {
969 check->health = 0;
970 status = HCHK_STATUS_L7STS;
971 hs = cmd;
972 }
973 else if (strcasecmp(cmd, "fail") == 0) {
974 check->health = 0;
975 status = HCHK_STATUS_L7STS;
976 hs = cmd;
977 }
978 /* admin statuses */
979 else if (strcasecmp(cmd, "ready") == 0) {
980 as = cmd;
981 }
982 else if (strcasecmp(cmd, "drain") == 0) {
983 as = cmd;
984 }
985 else if (strcasecmp(cmd, "maint") == 0) {
986 as = cmd;
987 }
988 /* else try to parse a weight here and keep the last one */
989 else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) {
990 ps = cmd;
991 }
992 else {
993 /* keep a copy of the first error */
994 if (!err)
995 err = cmd;
996 }
997 /* skip to next word */
998 cmd = p;
999 }
1000 /* here, cmd points either to \0 or to the beginning of a
1001 * description. Skip possible leading spaces.
1002 */
1003 while (*cmd == ' ' || *cmd == '\n')
1004 cmd++;
1005
1006 /* First, update the admin status so that we avoid sending other
1007 * possibly useless warnings and can also update the health if
1008 * present after going back up.
1009 */
1010 if (as) {
1011 if (strcasecmp(as, "drain") == 0)
1012 srv_adm_set_drain(check->server);
1013 else if (strcasecmp(as, "maint") == 0)
1014 srv_adm_set_maint(check->server);
1015 else
1016 srv_adm_set_ready(check->server);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001017 }
1018
Willy Tarreau81f5d942013-12-09 20:51:51 +01001019 /* now change weights */
1020 if (ps) {
1021 const char *msg;
1022
1023 msg = server_parse_weight_change_request(s, ps);
1024 if (!wrn || !*wrn)
1025 wrn = msg;
1026 }
1027
1028 /* and finally health status */
1029 if (hs) {
1030 /* We'll report some of the warnings and errors we have
1031 * here. Down reports are critical, we leave them untouched.
1032 * Lack of report, or report of 'UP' leaves the room for
1033 * ERR first, then WARN.
Simon Hormana2b9dad2013-02-12 10:45:54 +09001034 */
Willy Tarreau81f5d942013-12-09 20:51:51 +01001035 const char *msg = cmd;
1036 struct chunk *t;
1037
1038 if (!*msg || status == HCHK_STATUS_L7OKD) {
1039 if (err && *err)
1040 msg = err;
1041 else if (wrn && *wrn)
1042 msg = wrn;
Simon Hormana2b9dad2013-02-12 10:45:54 +09001043 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001044
1045 t = get_trash_chunk();
1046 chunk_printf(t, "via agent : %s%s%s%s",
1047 hs, *msg ? " (" : "",
1048 msg, *msg ? ")" : "");
1049
1050 set_server_check_status(check, status, t->str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001051 }
Willy Tarreau81f5d942013-12-09 20:51:51 +01001052 else if (err && *err) {
1053 /* No status change but we'd like to report something odd.
1054 * Just report the current state and copy the message.
1055 */
1056 chunk_printf(&trash, "agent reports an error : %s", err);
1057 set_server_check_status(check, status/*check->status*/, trash.str);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001058
Willy Tarreau81f5d942013-12-09 20:51:51 +01001059 }
1060 else if (wrn && *wrn) {
1061 /* No status change but we'd like to report something odd.
1062 * Just report the current state and copy the message.
1063 */
1064 chunk_printf(&trash, "agent warns : %s", wrn);
1065 set_server_check_status(check, status/*check->status*/, trash.str);
1066 }
1067 else
1068 set_server_check_status(check, status, NULL);
Simon Hormana2b9dad2013-02-12 10:45:54 +09001069 break;
1070 }
1071
Willy Tarreau1620ec32011-08-06 17:05:02 +02001072 case PR_O2_PGSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001073 if (!done && check->bi->i < 9)
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001074 goto wait_more_data;
1075
Simon Horman4a741432013-02-23 15:35:38 +09001076 if (check->bi->data[0] == 'R') {
1077 set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok");
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001078 }
1079 else {
Simon Horman4a741432013-02-23 15:35:38 +09001080 if ((check->bi->data[0] == 'E') && (check->bi->data[5]!=0) && (check->bi->data[6]!=0))
1081 desc = &check->bi->data[6];
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001082 else
1083 desc = "PostgreSQL unknown error";
1084
Simon Horman4a741432013-02-23 15:35:38 +09001085 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Rauf Kuliyev38b41562011-01-04 15:14:13 +01001086 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001087 break;
1088
1089 case PR_O2_REDIS_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001090 if (!done && check->bi->i < 7)
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001091 goto wait_more_data;
1092
Simon Horman4a741432013-02-23 15:35:38 +09001093 if (strcmp(check->bi->data, "+PONG\r\n") == 0) {
1094 set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok");
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001095 }
1096 else {
Simon Horman4a741432013-02-23 15:35:38 +09001097 set_server_check_status(check, HCHK_STATUS_L7STS, check->bi->data);
Hervé COMMOWICKec032d62011-08-05 16:23:48 +02001098 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001099 break;
1100
1101 case PR_O2_MYSQL_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001102 if (!done && check->bi->i < 5)
Willy Tarreau03938182010-03-17 21:52:07 +01001103 goto wait_more_data;
1104
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001105 if (s->proxy->check_len == 0) { // old mode
Simon Horman4a741432013-02-23 15:35:38 +09001106 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001107 /* We set the MySQL Version in description for information purpose
1108 * FIXME : it can be cool to use MySQL Version for other purpose,
1109 * like mark as down old MySQL server.
1110 */
Simon Horman4a741432013-02-23 15:35:38 +09001111 if (check->bi->i > 51) {
1112 desc = ltrim(check->bi->data + 5, ' ');
1113 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001114 }
1115 else {
1116 if (!done)
1117 goto wait_more_data;
1118 /* it seems we have a OK packet but without a valid length,
1119 * it must be a protocol error
1120 */
Simon Horman4a741432013-02-23 15:35:38 +09001121 set_server_check_status(check, HCHK_STATUS_L7RSP, check->bi->data);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001122 }
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001123 }
1124 else {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001125 /* An error message is attached in the Error packet */
Simon Horman4a741432013-02-23 15:35:38 +09001126 desc = ltrim(check->bi->data + 7, ' ');
1127 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001128 }
1129 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001130 unsigned int first_packet_len = ((unsigned int) *check->bi->data) +
1131 (((unsigned int) *(check->bi->data + 1)) << 8) +
1132 (((unsigned int) *(check->bi->data + 2)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001133
Simon Horman4a741432013-02-23 15:35:38 +09001134 if (check->bi->i == first_packet_len + 4) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001135 /* MySQL Error packet always begin with field_count = 0xff */
Simon Horman4a741432013-02-23 15:35:38 +09001136 if (*(check->bi->data + 4) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001137 /* We have only one MySQL packet and it is a Handshake Initialization packet
1138 * but we need to have a second packet to know if it is alright
1139 */
Simon Horman4a741432013-02-23 15:35:38 +09001140 if (!done && check->bi->i < first_packet_len + 5)
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001141 goto wait_more_data;
1142 }
1143 else {
1144 /* We have only one packet and it is an Error packet,
1145 * an error message is attached, so we can display it
1146 */
Simon Horman4a741432013-02-23 15:35:38 +09001147 desc = &check->bi->data[7];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001148 //Warning("onlyoneERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001149 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001150 }
Simon Horman4a741432013-02-23 15:35:38 +09001151 } else if (check->bi->i > first_packet_len + 4) {
1152 unsigned int second_packet_len = ((unsigned int) *(check->bi->data + first_packet_len + 4)) +
1153 (((unsigned int) *(check->bi->data + first_packet_len + 5)) << 8) +
1154 (((unsigned int) *(check->bi->data + first_packet_len + 6)) << 16);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001155
Simon Horman4a741432013-02-23 15:35:38 +09001156 if (check->bi->i == first_packet_len + 4 + second_packet_len + 4 ) {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001157 /* We have 2 packets and that's good */
1158 /* Check if the second packet is a MySQL Error packet or not */
Simon Horman4a741432013-02-23 15:35:38 +09001159 if (*(check->bi->data + first_packet_len + 8) != '\xff') {
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001160 /* No error packet */
1161 /* We set the MySQL Version in description for information purpose */
Simon Horman4a741432013-02-23 15:35:38 +09001162 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001163 //Warning("2packetOK: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001164 set_server_check_status(check, HCHK_STATUS_L7OKD, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001165 }
1166 else {
1167 /* An error message is attached in the Error packet
1168 * so we can display it ! :)
1169 */
Simon Horman4a741432013-02-23 15:35:38 +09001170 desc = &check->bi->data[first_packet_len+11];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001171 //Warning("2packetERR: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001172 set_server_check_status(check, HCHK_STATUS_L7STS, desc);
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001173 }
1174 }
1175 }
1176 else {
Willy Tarreau03938182010-03-17 21:52:07 +01001177 if (!done)
1178 goto wait_more_data;
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001179 /* it seems we have a Handshake Initialization packet but without a valid length,
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001180 * it must be a protocol error
1181 */
Simon Horman4a741432013-02-23 15:35:38 +09001182 desc = &check->bi->data[5];
Hervé COMMOWICK8776f1b2010-10-18 15:58:36 +02001183 //Warning("protoerr: %s\n", desc);
Simon Horman4a741432013-02-23 15:35:38 +09001184 set_server_check_status(check, HCHK_STATUS_L7RSP, desc);
Hervé COMMOWICK698ae002010-01-12 09:25:13 +01001185 }
1186 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001187 break;
1188
1189 case PR_O2_LDAP_CHK:
Simon Horman4a741432013-02-23 15:35:38 +09001190 if (!done && check->bi->i < 14)
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001191 goto wait_more_data;
1192
1193 /* Check if the server speaks LDAP (ASN.1/BER)
1194 * http://en.wikipedia.org/wiki/Basic_Encoding_Rules
1195 * http://tools.ietf.org/html/rfc4511
1196 */
1197
1198 /* http://tools.ietf.org/html/rfc4511#section-4.1.1
1199 * LDAPMessage: 0x30: SEQUENCE
1200 */
Simon Horman4a741432013-02-23 15:35:38 +09001201 if ((check->bi->i < 14) || (*(check->bi->data) != '\x30')) {
1202 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001203 }
1204 else {
1205 /* size of LDAPMessage */
Simon Horman4a741432013-02-23 15:35:38 +09001206 msglen = (*(check->bi->data + 1) & 0x80) ? (*(check->bi->data + 1) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001207
1208 /* http://tools.ietf.org/html/rfc4511#section-4.2.2
1209 * messageID: 0x02 0x01 0x01: INTEGER 1
1210 * protocolOp: 0x61: bindResponse
1211 */
1212 if ((msglen > 2) ||
Simon Horman4a741432013-02-23 15:35:38 +09001213 (memcmp(check->bi->data + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) {
1214 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001215
1216 goto out_wakeup;
1217 }
1218
1219 /* size of bindResponse */
Simon Horman4a741432013-02-23 15:35:38 +09001220 msglen += (*(check->bi->data + msglen + 6) & 0x80) ? (*(check->bi->data + msglen + 6) & 0x7f) : 0;
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001221
1222 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1223 * ldapResult: 0x0a 0x01: ENUMERATION
1224 */
1225 if ((msglen > 4) ||
Simon Horman4a741432013-02-23 15:35:38 +09001226 (memcmp(check->bi->data + 7 + msglen, "\x0a\x01", 2) != 0)) {
1227 set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001228
1229 goto out_wakeup;
1230 }
1231
1232 /* http://tools.ietf.org/html/rfc4511#section-4.1.9
1233 * resultCode
1234 */
Simon Horman4a741432013-02-23 15:35:38 +09001235 check->code = *(check->bi->data + msglen + 9);
1236 if (check->code) {
1237 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 +02001238 } else {
Simon Horman4a741432013-02-23 15:35:38 +09001239 set_server_check_status(check, HCHK_STATUS_L7OKD, "Success");
Gabor Lekenyb4c81e42010-09-29 18:17:05 +02001240 }
1241 }
Willy Tarreau1620ec32011-08-06 17:05:02 +02001242 break;
1243
1244 default:
Willy Tarreau06559ac2013-12-05 01:53:08 +01001245 /* for other checks (eg: pure TCP), delegate to the main task */
Willy Tarreau1620ec32011-08-06 17:05:02 +02001246 break;
1247 } /* switch */
Willy Tarreau83749182007-04-15 20:56:27 +02001248
Willy Tarreauc7dd71a2007-11-30 08:33:21 +01001249 out_wakeup:
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001250 /* collect possible new errors */
1251 if (conn->flags & CO_FL_ERROR)
1252 chk_report_conn_err(conn, 0, 0);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001253
Nick Chalk57b1bf72010-03-16 15:50:46 +00001254 /* Reset the check buffer... */
Simon Horman4a741432013-02-23 15:35:38 +09001255 *check->bi->data = '\0';
1256 check->bi->i = 0;
Nick Chalk57b1bf72010-03-16 15:50:46 +00001257
Willy Tarreaufd29cc52012-11-23 09:18:20 +01001258 /* Close the connection... We absolutely want to perform a hard close
1259 * and reset the connection if some data are pending, otherwise we end
1260 * up with many TIME_WAITs and eat all the source port range quickly.
1261 * To avoid sending RSTs all the time, we first try to drain pending
1262 * data.
1263 */
Willy Tarreauf1503172012-09-28 19:39:36 +02001264 if (conn->xprt && conn->xprt->shutw)
1265 conn->xprt->shutw(conn, 0);
Willy Tarreau2b57cb82013-06-10 19:56:38 +02001266
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001267 /* OK, let's not stay here forever */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001268 if (check->result == CHK_RES_FAILED)
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001269 conn->flags |= CO_FL_ERROR;
1270
Willy Tarreaua522f802012-11-23 08:56:35 +01001271 __conn_data_stop_both(conn);
Willy Tarreaufdccded2008-08-29 18:19:04 +02001272 task_wakeup(t, TASK_WOKEN_IO);
Willy Tarreau3267d362012-08-17 23:53:56 +02001273 return;
Willy Tarreau03938182010-03-17 21:52:07 +01001274
1275 wait_more_data:
Willy Tarreauf817e9f2014-01-10 16:58:45 +01001276 __conn_data_want_recv(conn);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001277}
1278
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001279/*
1280 * This function is used only for server health-checks. It handles connection
1281 * status updates including errors. If necessary, it wakes the check task up.
1282 * It always returns 0.
1283 */
1284static int wake_srv_chk(struct connection *conn)
Willy Tarreau20bea422012-07-06 12:00:49 +02001285{
Simon Horman4a741432013-02-23 15:35:38 +09001286 struct check *check = conn->owner;
Willy Tarreau20bea422012-07-06 12:00:49 +02001287
Willy Tarreau6c560da2012-11-24 11:14:45 +01001288 if (unlikely(conn->flags & CO_FL_ERROR)) {
Willy Tarreau02b0f582013-12-03 15:42:33 +01001289 /* We may get error reports bypassing the I/O handlers, typically
1290 * the case when sending a pure TCP check which fails, then the I/O
1291 * handlers above are not called. This is completely handled by the
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001292 * main processing task so let's simply wake it up. If we get here,
1293 * we expect errno to still be valid.
1294 */
1295 chk_report_conn_err(conn, errno, 0);
1296
Willy Tarreau2d351b62013-12-05 02:36:25 +01001297 __conn_data_stop_both(conn);
1298 task_wakeup(check->task, TASK_WOKEN_IO);
1299 }
Willy Tarreau3be293f2014-02-05 18:31:24 +01001300 else if (!(conn->flags & (CO_FL_DATA_RD_ENA|CO_FL_DATA_WR_ENA|CO_FL_HANDSHAKE))) {
1301 /* we may get here if only a connection probe was required : we
1302 * don't have any data to send nor anything expected in response,
1303 * so the completion of the connection establishment is enough.
1304 */
1305 task_wakeup(check->task, TASK_WOKEN_IO);
1306 }
Willy Tarreau2d351b62013-12-05 02:36:25 +01001307
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001308 if (check->result != CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001309 /* We're here because nobody wants to handle the error, so we
1310 * sure want to abort the hard way.
Willy Tarreau02b0f582013-12-03 15:42:33 +01001311 */
Willy Tarreau46be2e52014-01-20 12:10:52 +01001312 conn_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001313 conn_force_close(conn);
Willy Tarreau2d351b62013-12-05 02:36:25 +01001314 }
Willy Tarreau3267d362012-08-17 23:53:56 +02001315 return 0;
Willy Tarreau20bea422012-07-06 12:00:49 +02001316}
1317
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001318struct data_cb check_conn_cb = {
1319 .recv = event_srv_chk_r,
1320 .send = event_srv_chk_w,
1321 .wake = wake_srv_chk,
1322};
1323
Willy Tarreaubaaee002006-06-26 02:48:02 +02001324/*
Willy Tarreau2e993902011-10-31 11:53:20 +01001325 * updates the server's weight during a warmup stage. Once the final weight is
1326 * reached, the task automatically stops. Note that any server status change
1327 * must have updated s->last_change accordingly.
1328 */
1329static struct task *server_warmup(struct task *t)
1330{
1331 struct server *s = t->context;
1332
1333 /* by default, plan on stopping the task */
1334 t->expire = TICK_ETERNITY;
Willy Tarreau20125212014-05-13 19:44:56 +02001335 if ((s->admin & SRV_ADMF_MAINT) ||
Willy Tarreau892337c2014-05-13 23:41:20 +02001336 (s->state != SRV_ST_STARTING))
Willy Tarreau2e993902011-10-31 11:53:20 +01001337 return t;
1338
Willy Tarreau892337c2014-05-13 23:41:20 +02001339 /* recalculate the weights and update the state */
Willy Tarreau004e0452013-11-21 11:22:01 +01001340 server_recalc_eweight(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001341
1342 /* probably that we can refill this server with a bit more connections */
Willy Tarreau4aac7db2014-05-16 11:48:10 +02001343 pendconn_grab_from_px(s);
Willy Tarreau2e993902011-10-31 11:53:20 +01001344
1345 /* get back there in 1 second or 1/20th of the slowstart interval,
1346 * whichever is greater, resulting in small 5% steps.
1347 */
Willy Tarreau892337c2014-05-13 23:41:20 +02001348 if (s->state == SRV_ST_STARTING)
Willy Tarreau2e993902011-10-31 11:53:20 +01001349 t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20)));
1350 return t;
1351}
1352
1353/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02001354 * manages a server health-check. Returns
1355 * the time the task accepts to wait, or TIME_ETERNITY for infinity.
1356 */
Simon Horman63a4a822012-03-19 07:24:41 +09001357static struct task *process_chk(struct task *t)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001358{
Simon Horman4a741432013-02-23 15:35:38 +09001359 struct check *check = t->context;
1360 struct server *s = check->server;
1361 struct connection *conn = check->conn;
Willy Tarreau640556c2014-05-09 23:38:15 +02001362 struct protocol *proto;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001363 int rv;
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001364 int ret;
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001365 int expired = tick_is_expired(t->expire, now_ms);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001366
Willy Tarreau2c115e52013-12-11 19:41:16 +01001367 if (!(check->state & CHK_ST_INPROGRESS)) {
Willy Tarreau5a78f362012-11-23 12:47:05 +01001368 /* no check currently running */
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001369 if (!expired) /* woke up too early */
Willy Tarreau26c25062009-03-08 09:38:41 +01001370 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001371
Simon Horman671b6f02013-11-25 10:46:39 +09001372 /* we don't send any health-checks when the proxy is
1373 * stopped, the server should not be checked or the check
1374 * is disabled.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001375 */
Willy Tarreau0d924cc2013-12-11 21:26:24 +01001376 if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) ||
Willy Tarreau33a08db2013-12-11 21:03:31 +01001377 s->proxy->state == PR_STSTOPPED)
Willy Tarreau5a78f362012-11-23 12:47:05 +01001378 goto reschedule;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001379
1380 /* we'll initiate a new check */
Simon Horman4a741432013-02-23 15:35:38 +09001381 set_server_check_status(check, HCHK_STATUS_START, NULL);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001382
Willy Tarreau2c115e52013-12-11 19:41:16 +01001383 check->state |= CHK_ST_INPROGRESS;
Simon Horman4a741432013-02-23 15:35:38 +09001384 check->bi->p = check->bi->data;
1385 check->bi->i = 0;
1386 check->bo->p = check->bo->data;
1387 check->bo->o = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001388
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001389 /* tcpcheck send/expect initialisation */
1390 if (check->type == PR_O2_TCPCHK_CHK)
1391 check->current_step = NULL;
1392
1393 /* prepare the check buffer.
1394 * This should not be used if check is the secondary agent check
1395 * of a server as s->proxy->check_req will relate to the
1396 * configuration of the primary check. Similarly, tcp-check uses
1397 * its own strings.
1398 */
Willy Tarreau33434322013-12-11 21:15:19 +01001399 if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) {
Simon Horman4a741432013-02-23 15:35:38 +09001400 bo_putblk(check->bo, s->proxy->check_req, s->proxy->check_len);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001401
1402 /* we want to check if this host replies to HTTP or SSLv3 requests
1403 * so we'll send the request, and won't wake the checker up now.
1404 */
Simon Horman4a741432013-02-23 15:35:38 +09001405 if ((check->type) == PR_O2_SSL3_CHK) {
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001406 /* SSL requires that we put Unix time in the request */
1407 int gmt_time = htonl(date.tv_sec);
Simon Horman4a741432013-02-23 15:35:38 +09001408 memcpy(check->bo->data + 11, &gmt_time, 4);
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001409 }
Simon Horman4a741432013-02-23 15:35:38 +09001410 else if ((check->type) == PR_O2_HTTP_CHK) {
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001411 if (s->proxy->options2 & PR_O2_CHK_SNDST)
Willy Tarreaua1dab552014-04-14 15:04:54 +02001412 bo_putblk(check->bo, trash.str, httpchk_build_status_header(s, trash.str, trash.size));
Simon Horman4a741432013-02-23 15:35:38 +09001413 bo_putstr(check->bo, "\r\n");
1414 *check->bo->p = '\0'; /* to make gdb output easier to read */
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001415 }
1416 }
1417
1418 /* prepare a new connection */
Willy Tarreau4bdae8a2013-10-14 17:29:15 +02001419 conn_init(conn);
Willy Tarreau910c6aa2013-10-24 15:08:37 +02001420 conn_prepare(conn, s->check_common.proto, s->check_common.xprt);
Willy Tarreau7abddb52013-10-24 15:31:04 +02001421 conn_attach(conn, check, &check_conn_cb);
Willy Tarreau4bdae8a2013-10-14 17:29:15 +02001422 conn->target = &s->obj_type;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001423
Willy Tarreau5f2877a2012-10-26 19:57:58 +02001424 /* no client address */
1425 clear_addr(&conn->addr.from);
1426
Willy Tarreau640556c2014-05-09 23:38:15 +02001427 if (is_addr(&s->check_common.addr)) {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001428 /* we'll connect to the check addr specified on the server */
Simon Horman66183002013-02-23 10:16:43 +09001429 conn->addr.to = s->check_common.addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001430 proto = s->check_common.proto;
1431 }
1432 else {
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001433 /* we'll connect to the addr on the server */
1434 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001435 proto = s->proto;
1436 }
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001437
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001438 if (check->port) {
1439 set_host_port(&conn->addr.to, check->port);
1440 }
1441
1442 if (check->type == PR_O2_TCPCHK_CHK) {
Baptiste Assmannf621bea2014-02-03 22:38:15 +01001443 struct tcpcheck_rule *r = (struct tcpcheck_rule *) s->proxy->tcpcheck_rules.n;
1444 /* if first step is a 'connect', then tcpcheck_main must run it */
1445 if (r->action == TCPCHK_ACT_CONNECT) {
1446 tcpcheck_main(conn);
1447 return t;
1448 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001449 }
1450
Willy Tarreaue8c66af2008-01-13 18:40:14 +01001451
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001452 /* It can return one of :
1453 * - SN_ERR_NONE if everything's OK
1454 * - SN_ERR_SRVTO if there are no more servers
1455 * - SN_ERR_SRVCL if the connection was refused by the server
1456 * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1457 * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1458 * - SN_ERR_INTERNAL for any other purely internal errors
1459 * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau24db47e2012-11-23 14:16:39 +01001460 * Note that we try to prevent the network stack from sending the ACK during the
Willy Tarreauf0837b22012-11-24 10:24:27 +01001461 * connect() when a pure TCP check is used (without PROXY protocol).
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001462 */
Willy Tarreau8f46cca2013-03-04 20:07:44 +01001463 ret = SN_ERR_INTERNAL;
Willy Tarreau640556c2014-05-09 23:38:15 +02001464 if (proto->connect)
1465 ret = proto->connect(conn, check->type, (check->type) ? 0 : 2);
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001466 conn->flags |= CO_FL_WAKE_DATA;
Willy Tarreau57cd3e42013-10-24 22:01:26 +02001467 if (s->check.send_proxy) {
1468 conn->send_proxy_ofs = 1;
1469 conn->flags |= CO_FL_SEND_PROXY;
1470 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001471
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001472 switch (ret) {
1473 case SN_ERR_NONE:
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001474 /* we allow up to min(inter, timeout.connect) for a connection
1475 * to establish but only when timeout.check is set
1476 * as it may be to short for a full check otherwise
1477 */
Simon Horman4a741432013-02-23 15:35:38 +09001478 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki09605412009-09-23 22:09:24 +02001479
Willy Tarreaufb56aab2012-09-28 14:40:02 +02001480 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
1481 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
1482 t->expire = tick_first(t->expire, t_con);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001483 }
Willy Tarreau06559ac2013-12-05 01:53:08 +01001484
1485 if (check->type)
1486 conn_data_want_recv(conn); /* prepare for reading a possible reply */
1487
Willy Tarreau5a78f362012-11-23 12:47:05 +01001488 goto reschedule;
1489
1490 case SN_ERR_SRVTO: /* ETIMEDOUT */
1491 case SN_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
Willy Tarreau4bd07de2014-01-24 16:10:57 +01001492 conn->flags |= CO_FL_ERROR;
1493 chk_report_conn_err(conn, errno, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01001494 break;
1495 case SN_ERR_PRXCOND:
1496 case SN_ERR_RESOURCE:
1497 case SN_ERR_INTERNAL:
Willy Tarreau4bd07de2014-01-24 16:10:57 +01001498 conn->flags |= CO_FL_ERROR;
1499 chk_report_conn_err(conn, 0, 0);
Willy Tarreau5a78f362012-11-23 12:47:05 +01001500 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001501 }
1502
Willy Tarreau5a78f362012-11-23 12:47:05 +01001503 /* here, we have seen a synchronous error, no fd was allocated */
Willy Tarreau6b0a8502012-11-23 08:51:32 +01001504
Willy Tarreau2c115e52013-12-11 19:41:16 +01001505 check->state &= ~CHK_ST_INPROGRESS;
Willy Tarreau4eec5472014-05-20 22:32:27 +02001506 check_notify_failure(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001507
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001508 /* we allow up to min(inter, timeout.connect) for a connection
1509 * to establish but only when timeout.check is set
1510 * as it may be to short for a full check otherwise
1511 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001512 while (tick_is_expired(t->expire, now_ms)) {
1513 int t_con;
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001514
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001515 t_con = tick_add(t->expire, s->proxy->timeout.connect);
Simon Horman4a741432013-02-23 15:35:38 +09001516 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001517
Willy Tarreau0c303ee2008-07-07 00:09:58 +02001518 if (s->proxy->timeout.check)
1519 t->expire = tick_first(t->expire, t_con);
Krzysztof Piotr Oledzki5259dfe2008-01-21 01:54:06 +01001520 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001521 }
1522 else {
Willy Tarreauf1503172012-09-28 19:39:36 +02001523 /* there was a test running.
1524 * First, let's check whether there was an uncaught error,
1525 * which can happen on connect timeout or error.
1526 */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001527 if (s->check.result == CHK_RES_UNKNOWN) {
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001528 /* good connection is enough for pure TCP check */
1529 if ((conn->flags & CO_FL_CONNECTED) && !check->type) {
Simon Horman4a741432013-02-23 15:35:38 +09001530 if (check->use_ssl)
1531 set_server_check_status(check, HCHK_STATUS_L6OK, NULL);
Willy Tarreauf1503172012-09-28 19:39:36 +02001532 else
Simon Horman4a741432013-02-23 15:35:38 +09001533 set_server_check_status(check, HCHK_STATUS_L4OK, NULL);
Willy Tarreauacbdc7a2012-11-23 14:02:10 +01001534 }
Willy Tarreau25e2ab52013-12-04 11:17:05 +01001535 else if ((conn->flags & CO_FL_ERROR) || expired) {
1536 chk_report_conn_err(conn, 0, expired);
Willy Tarreauf1503172012-09-28 19:39:36 +02001537 }
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001538 else
1539 goto out_wait; /* timeout not reached, wait again */
Willy Tarreauf1503172012-09-28 19:39:36 +02001540 }
1541
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001542 /* check complete or aborted */
Willy Tarreau5ba04f62013-02-12 15:23:12 +01001543 if (conn->xprt) {
1544 /* The check was aborted and the connection was not yet closed.
1545 * This can happen upon timeout, or when an external event such
1546 * as a failed response coupled with "observe layer7" caused the
1547 * server state to be suddenly changed.
1548 */
Willy Tarreau46be2e52014-01-20 12:10:52 +01001549 conn_drain(conn);
Willy Tarreauf79c8172013-10-21 16:30:56 +02001550 conn_force_close(conn);
Willy Tarreau5ba04f62013-02-12 15:23:12 +01001551 }
1552
Willy Tarreauaf549582014-05-16 17:37:50 +02001553 if (check->result == CHK_RES_FAILED) {
1554 /* a failure or timeout detected */
Willy Tarreau4eec5472014-05-20 22:32:27 +02001555 check_notify_failure(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02001556 }
Willy Tarreaudb58b792014-05-21 13:57:23 +02001557 else if (check->result == CHK_RES_CONDPASS) {
1558 /* check is OK but asks for stopping mode */
1559 check_notify_stopping(check);
Willy Tarreauaf549582014-05-16 17:37:50 +02001560 }
Willy Tarreau3e048382014-05-21 10:30:54 +02001561 else if (check->result == CHK_RES_PASSED) {
1562 /* a success was detected */
1563 check_notify_success(check);
Willy Tarreaubaaee002006-06-26 02:48:02 +02001564 }
Willy Tarreau2c115e52013-12-11 19:41:16 +01001565 check->state &= ~CHK_ST_INPROGRESS;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001566
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001567 rv = 0;
1568 if (global.spread_checks > 0) {
Simon Horman4a741432013-02-23 15:35:38 +09001569 rv = srv_getinter(check) * global.spread_checks / 100;
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001570 rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0)));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001571 }
Simon Horman4a741432013-02-23 15:35:38 +09001572 t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001573 }
Willy Tarreau5a78f362012-11-23 12:47:05 +01001574
1575 reschedule:
1576 while (tick_is_expired(t->expire, now_ms))
Simon Horman4a741432013-02-23 15:35:38 +09001577 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
Willy Tarreau74fa7fb2012-11-23 14:43:49 +01001578 out_wait:
Willy Tarreau26c25062009-03-08 09:38:41 +01001579 return t;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001580}
1581
Simon Horman5c942422013-11-25 10:46:32 +09001582static int start_check_task(struct check *check, int mininter,
1583 int nbcheck, int srvpos)
1584{
1585 struct task *t;
1586 /* task for the check */
1587 if ((t = task_new()) == NULL) {
1588 Alert("Starting [%s:%s] check: out of memory.\n",
1589 check->server->proxy->id, check->server->id);
1590 return 0;
1591 }
1592
1593 check->task = t;
1594 t->process = process_chk;
1595 t->context = check;
1596
Willy Tarreau1746eec2014-04-25 10:46:47 +02001597 if (mininter < srv_getinter(check))
1598 mininter = srv_getinter(check);
1599
1600 if (global.max_spread_checks && mininter > global.max_spread_checks)
1601 mininter = global.max_spread_checks;
1602
Simon Horman5c942422013-11-25 10:46:32 +09001603 /* check this every ms */
Willy Tarreau1746eec2014-04-25 10:46:47 +02001604 t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck));
Simon Horman5c942422013-11-25 10:46:32 +09001605 check->start = now;
1606 task_queue(t);
1607
1608 return 1;
1609}
1610
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001611/*
1612 * Start health-check.
1613 * Returns 0 if OK, -1 if error, and prints the error in this case.
1614 */
1615int start_checks() {
1616
1617 struct proxy *px;
1618 struct server *s;
1619 struct task *t;
Simon Horman4a741432013-02-23 15:35:38 +09001620 int nbcheck=0, mininter=0, srvpos=0;
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001621
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02001622 /* 1- count the checkers to run simultaneously.
1623 * We also determine the minimum interval among all of those which
1624 * have an interval larger than SRV_CHK_INTER_THRES. This interval
1625 * will be used to spread their start-up date. Those which have
Jamie Gloudon801a0a32012-08-25 00:18:33 -04001626 * a shorter interval will start independently and will not dictate
Willy Tarreau2c43a1e2007-10-14 23:05:39 +02001627 * too short an interval for all others.
1628 */
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001629 for (px = proxy; px; px = px->next) {
1630 for (s = px->srv; s; s = s->next) {
Willy Tarreaue7b73482013-11-21 11:50:50 +01001631 if (s->slowstart) {
1632 if ((t = task_new()) == NULL) {
1633 Alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id);
1634 return -1;
1635 }
1636 /* We need a warmup task that will be called when the server
1637 * state switches from down to up.
1638 */
1639 s->warmup = t;
1640 t->process = server_warmup;
1641 t->context = s;
1642 t->expire = TICK_ETERNITY;
1643 }
1644
Willy Tarreaud8514a22013-12-11 21:10:14 +01001645 if (s->check.state & CHK_ST_CONFIGURED) {
1646 nbcheck++;
1647 if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) &&
1648 (!mininter || mininter > srv_getinter(&s->check)))
1649 mininter = srv_getinter(&s->check);
1650 }
Willy Tarreau15f39102013-12-11 20:41:18 +01001651
Willy Tarreaud8514a22013-12-11 21:10:14 +01001652 if (s->agent.state & CHK_ST_CONFIGURED) {
1653 nbcheck++;
1654 if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) &&
1655 (!mininter || mininter > srv_getinter(&s->agent)))
1656 mininter = srv_getinter(&s->agent);
1657 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001658 }
1659 }
1660
Simon Horman4a741432013-02-23 15:35:38 +09001661 if (!nbcheck)
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001662 return 0;
1663
1664 srand((unsigned)time(NULL));
1665
1666 /*
1667 * 2- start them as far as possible from each others. For this, we will
1668 * start them after their interval set to the min interval divided by
1669 * the number of servers, weighted by the server's position in the list.
1670 */
1671 for (px = proxy; px; px = px->next) {
1672 for (s = px->srv; s; s = s->next) {
Simon Hormand60d6912013-11-25 10:46:36 +09001673 /* A task for the main check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01001674 if (s->check.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09001675 if (!start_check_task(&s->check, mininter, nbcheck, srvpos))
1676 return -1;
1677 srvpos++;
1678 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001679
Simon Hormand60d6912013-11-25 10:46:36 +09001680 /* A task for a auxiliary agent check */
Willy Tarreauff5ae352013-12-11 20:36:34 +01001681 if (s->agent.state & CHK_ST_CONFIGURED) {
Simon Hormand60d6912013-11-25 10:46:36 +09001682 if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) {
1683 return -1;
1684 }
1685 srvpos++;
1686 }
Krzysztof Oledzkib304dc72007-10-14 23:40:01 +02001687 }
1688 }
1689 return 0;
1690}
Willy Tarreaubaaee002006-06-26 02:48:02 +02001691
1692/*
Willy Tarreau5b3a2022012-09-28 15:01:02 +02001693 * Perform content verification check on data in s->check.buffer buffer.
Willy Tarreaubd741542010-03-16 18:46:54 +01001694 * The buffer MUST be terminated by a null byte before calling this function.
1695 * Sets server status appropriately. The caller is responsible for ensuring
1696 * that the buffer contains at least 13 characters. If <done> is zero, we may
1697 * return 0 to indicate that data is required to decide of a match.
1698 */
1699static int httpchk_expect(struct server *s, int done)
1700{
1701 static char status_msg[] = "HTTP status check returned code <000>";
1702 char status_code[] = "000";
1703 char *contentptr;
1704 int crlf;
1705 int ret;
1706
1707 switch (s->proxy->options2 & PR_O2_EXP_TYPE) {
1708 case PR_O2_EXP_STS:
1709 case PR_O2_EXP_RSTS:
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001710 memcpy(status_code, s->check.bi->data + 9, 3);
1711 memcpy(status_msg + strlen(status_msg) - 4, s->check.bi->data + 9, 3);
Willy Tarreaubd741542010-03-16 18:46:54 +01001712
1713 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS)
1714 ret = strncmp(s->proxy->expect_str, status_code, 3) == 0;
1715 else
1716 ret = regexec(s->proxy->expect_regex, status_code, MAX_MATCH, pmatch, 0) == 0;
1717
1718 /* we necessarily have the response, so there are no partial failures */
1719 if (s->proxy->options2 & PR_O2_EXP_INV)
1720 ret = !ret;
1721
Simon Horman4a741432013-02-23 15:35:38 +09001722 set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg);
Willy Tarreaubd741542010-03-16 18:46:54 +01001723 break;
1724
1725 case PR_O2_EXP_STR:
1726 case PR_O2_EXP_RSTR:
1727 /* very simple response parser: ignore CR and only count consecutive LFs,
1728 * stop with contentptr pointing to first char after the double CRLF or
1729 * to '\0' if crlf < 2.
1730 */
1731 crlf = 0;
Willy Tarreau1ae1b7b2012-09-28 15:28:30 +02001732 for (contentptr = s->check.bi->data; *contentptr; contentptr++) {
Willy Tarreaubd741542010-03-16 18:46:54 +01001733 if (crlf >= 2)
1734 break;
1735 if (*contentptr == '\r')
1736 continue;
1737 else if (*contentptr == '\n')
1738 crlf++;
1739 else
1740 crlf = 0;
1741 }
1742
1743 /* Check that response contains a body... */
1744 if (crlf < 2) {
1745 if (!done)
1746 return 0;
1747
Simon Horman4a741432013-02-23 15:35:38 +09001748 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001749 "HTTP content check could not find a response body");
1750 return 1;
1751 }
1752
1753 /* Check that response body is not empty... */
1754 if (*contentptr == '\0') {
Willy Tarreaua164fb52011-04-13 09:32:41 +02001755 if (!done)
1756 return 0;
1757
Simon Horman4a741432013-02-23 15:35:38 +09001758 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001759 "HTTP content check found empty response body");
1760 return 1;
1761 }
1762
1763 /* Check the response content against the supplied string
1764 * or regex... */
1765 if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR)
1766 ret = strstr(contentptr, s->proxy->expect_str) != NULL;
1767 else
1768 ret = regexec(s->proxy->expect_regex, contentptr, MAX_MATCH, pmatch, 0) == 0;
1769
1770 /* if we don't match, we may need to wait more */
1771 if (!ret && !done)
1772 return 0;
1773
1774 if (ret) {
1775 /* content matched */
1776 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09001777 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001778 "HTTP check matched unwanted content");
1779 else
Simon Horman4a741432013-02-23 15:35:38 +09001780 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01001781 "HTTP content check matched");
1782 }
1783 else {
1784 if (s->proxy->options2 & PR_O2_EXP_INV)
Simon Horman4a741432013-02-23 15:35:38 +09001785 set_server_check_status(&s->check, HCHK_STATUS_L7OKD,
Willy Tarreaubd741542010-03-16 18:46:54 +01001786 "HTTP check did not match unwanted content");
1787 else
Simon Horman4a741432013-02-23 15:35:38 +09001788 set_server_check_status(&s->check, HCHK_STATUS_L7RSP,
Willy Tarreaubd741542010-03-16 18:46:54 +01001789 "HTTP content check did not match");
1790 }
1791 break;
1792 }
1793 return 1;
1794}
1795
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001796/*
1797 * return the id of a step in a send/expect session
1798 */
1799static int tcpcheck_get_step_id(struct server *s)
1800{
1801 struct tcpcheck_rule *cur = NULL, *next = NULL;
1802 int i = 0;
1803
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001804 cur = s->check.last_started_step;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001805
1806 /* no step => first step */
1807 if (cur == NULL)
1808 return 1;
1809
1810 /* increment i until current step */
1811 list_for_each_entry(next, &s->proxy->tcpcheck_rules, list) {
1812 if (next->list.p == &cur->list)
1813 break;
1814 ++i;
1815 }
1816
1817 return i;
1818}
1819
1820static void tcpcheck_main(struct connection *conn)
1821{
1822 char *contentptr;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001823 struct list *head = NULL;
1824 struct tcpcheck_rule *cur = NULL;
1825 int done = 0, ret = 0;
1826
1827 struct check *check = conn->owner;
1828 struct server *s = check->server;
1829 struct task *t = check->task;
1830
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001831 /*
1832 * don't do anything until the connection is established but if we're running
1833 * first step which must be a connect
1834 */
1835 if (check->current_step && (!(conn->flags & CO_FL_CONNECTED))) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001836 /* update expire time, should be done by process_chk */
1837 /* we allow up to min(inter, timeout.connect) for a connection
1838 * to establish but only when timeout.check is set
1839 * as it may be to short for a full check otherwise
1840 */
1841 while (tick_is_expired(t->expire, now_ms)) {
1842 int t_con;
1843
1844 t_con = tick_add(t->expire, s->proxy->timeout.connect);
1845 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
1846
1847 if (s->proxy->timeout.check)
1848 t->expire = tick_first(t->expire, t_con);
1849 }
1850 return;
1851 }
1852
1853 /* here, we know that the connection is established */
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01001854 if (check->result != CHK_RES_UNKNOWN)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001855 goto out_end_tcpcheck;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02001856
1857 /* head is be the first element of the double chained list */
1858 head = &s->proxy->tcpcheck_rules;
1859
1860 /* no step means first step
1861 * initialisation */
1862 if (check->current_step == NULL) {
1863 check->bo->p = check->bo->data;
1864 check->bo->o = 0;
1865 check->bi->p = check->bi->data;
1866 check->bi->i = 0;
1867 cur = check->current_step = LIST_ELEM(head->n, struct tcpcheck_rule *, list);
1868 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1869 if (s->proxy->timeout.check)
1870 t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check);
1871 }
1872 /* keep on processing step */
1873 else {
1874 cur = check->current_step;
1875 }
1876
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001877 if (conn->flags & CO_FL_HANDSHAKE)
1878 return;
1879
1880 /* It's only the rules which will enable send/recv */
1881 __conn_data_stop_both(conn);
1882
Willy Tarreauabca5b62013-12-06 14:19:25 +01001883 while (1) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001884 /* we have to try to flush the output buffer before reading, at the end,
1885 * or if we're about to send a string that does not fit in the remaining space.
1886 */
1887 if (check->bo->o &&
1888 (&cur->list == head ||
1889 check->current_step->action != TCPCHK_ACT_SEND ||
1890 check->current_step->string_len >= buffer_total_space(check->bo))) {
1891
Willy Tarreau1049b1f2014-02-02 01:51:17 +01001892 if (conn->xprt->snd_buf(conn, check->bo, 0) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001893 if (conn->flags & CO_FL_ERROR) {
1894 chk_report_conn_err(conn, errno, 0);
1895 __conn_data_stop_both(conn);
1896 goto out_end_tcpcheck;
1897 }
1898 goto out_need_io;
Willy Tarreauabca5b62013-12-06 14:19:25 +01001899 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01001900 }
1901
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001902 /* did we reach the end ? If so, let's check that everything was sent */
1903 if (&cur->list == head) {
1904 if (check->bo->o)
1905 goto out_need_io;
Willy Tarreauabca5b62013-12-06 14:19:25 +01001906 break;
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01001907 }
Willy Tarreauabca5b62013-12-06 14:19:25 +01001908
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001909 if (check->current_step->action == TCPCHK_ACT_CONNECT) {
1910 struct protocol *proto;
1911 struct xprt_ops *xprt;
1912
1913 /* mark the step as started */
1914 check->last_started_step = check->current_step;
1915 /* first, shut existing connection */
1916 conn_force_close(conn);
1917
1918 /* prepare new connection */
1919 /* initialization */
1920 conn_init(conn);
1921 conn_attach(conn, check, &check_conn_cb);
1922 conn->target = &s->obj_type;
1923
1924 /* no client address */
1925 clear_addr(&conn->addr.from);
1926
Willy Tarreau640556c2014-05-09 23:38:15 +02001927 if (is_addr(&s->check_common.addr)) {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001928 /* we'll connect to the check addr specified on the server */
1929 conn->addr.to = s->check_common.addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001930 proto = s->check_common.proto;
1931 }
1932 else {
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001933 /* we'll connect to the addr on the server */
1934 conn->addr.to = s->addr;
Willy Tarreau640556c2014-05-09 23:38:15 +02001935 proto = s->proto;
1936 }
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001937
1938 /* port */
1939 if (check->current_step->port)
1940 set_host_port(&conn->addr.to, check->current_step->port);
1941 else if (check->port)
1942 set_host_port(&conn->addr.to, check->port);
1943
1944#ifdef USE_OPENSSL
1945 if (check->current_step->conn_opts & TCPCHK_OPT_SSL) {
1946 xprt = &ssl_sock;
Baptiste Assmann69e273f2013-12-11 00:52:19 +01001947 }
1948 else {
1949 xprt = &raw_sock;
1950 }
1951#else /* USE_OPENSSL */
1952 xprt = &raw_sock;
1953#endif /* USE_OPENSSL */
1954 conn_prepare(conn, proto, xprt);
1955
1956 ret = SN_ERR_INTERNAL;
1957 if (proto->connect)
1958 ret = proto->connect(conn, check->type, (check->type) ? 0 : 2);
1959 conn->flags |= CO_FL_WAKE_DATA;
1960 if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) {
1961 conn->send_proxy_ofs = 1;
1962 conn->flags |= CO_FL_SEND_PROXY;
1963 }
1964
1965 /* It can return one of :
1966 * - SN_ERR_NONE if everything's OK
1967 * - SN_ERR_SRVTO if there are no more servers
1968 * - SN_ERR_SRVCL if the connection was refused by the server
1969 * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
1970 * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
1971 * - SN_ERR_INTERNAL for any other purely internal errors
1972 * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted.
1973 * Note that we try to prevent the network stack from sending the ACK during the
1974 * connect() when a pure TCP check is used (without PROXY protocol).
1975 */
1976 switch (ret) {
1977 case SN_ERR_NONE:
1978 /* we allow up to min(inter, timeout.connect) for a connection
1979 * to establish but only when timeout.check is set
1980 * as it may be to short for a full check otherwise
1981 */
1982 t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter));
1983
1984 if (s->proxy->timeout.check && s->proxy->timeout.connect) {
1985 int t_con = tick_add(now_ms, s->proxy->timeout.connect);
1986 t->expire = tick_first(t->expire, t_con);
1987 }
1988 break;
1989 case SN_ERR_SRVTO: /* ETIMEDOUT */
1990 case SN_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */
1991 chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s",
1992 tcpcheck_get_step_id(s), strerror(errno));
1993 set_server_check_status(check, HCHK_STATUS_L4CON, trash.str);
1994 goto out_end_tcpcheck;
1995 case SN_ERR_PRXCOND:
1996 case SN_ERR_RESOURCE:
1997 case SN_ERR_INTERNAL:
1998 chunk_printf(&trash, "TCPCHK error establishing connection at step %d",
1999 tcpcheck_get_step_id(s));
2000 set_server_check_status(check, HCHK_STATUS_SOCKERR, trash.str);
2001 goto out_end_tcpcheck;
2002 }
2003
2004 /* allow next rule */
2005 cur = (struct tcpcheck_rule *)cur->list.n;
2006 check->current_step = cur;
2007
2008 /* don't do anything until the connection is established */
2009 if (!(conn->flags & CO_FL_CONNECTED)) {
2010 /* update expire time, should be done by process_chk */
2011 /* we allow up to min(inter, timeout.connect) for a connection
2012 * to establish but only when timeout.check is set
2013 * as it may be to short for a full check otherwise
2014 */
2015 while (tick_is_expired(t->expire, now_ms)) {
2016 int t_con;
2017
2018 t_con = tick_add(t->expire, s->proxy->timeout.connect);
2019 t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter));
2020
2021 if (s->proxy->timeout.check)
2022 t->expire = tick_first(t->expire, t_con);
2023 }
2024 return;
2025 }
2026
2027 } /* end 'connect' */
2028 else if (check->current_step->action == TCPCHK_ACT_SEND) {
2029 /* mark the step as started */
2030 check->last_started_step = check->current_step;
2031
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002032 /* reset the read buffer */
2033 if (*check->bi->data != '\0') {
2034 *check->bi->data = '\0';
2035 check->bi->i = 0;
2036 }
2037
2038 if (conn->flags & (CO_FL_SOCK_WR_SH | CO_FL_DATA_WR_SH)) {
2039 conn->flags |= CO_FL_ERROR;
2040 chk_report_conn_err(conn, 0, 0);
2041 goto out_end_tcpcheck;
2042 }
2043
Willy Tarreauabca5b62013-12-06 14:19:25 +01002044 if (check->current_step->string_len >= check->bo->size) {
2045 chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%d) at step %d",
2046 check->current_step->string_len, check->bo->size,
2047 tcpcheck_get_step_id(s));
2048 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2049 goto out_end_tcpcheck;
2050 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002051
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002052 /* do not try to send if there is no space */
2053 if (check->current_step->string_len >= buffer_total_space(check->bo))
2054 continue;
2055
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002056 bo_putblk(check->bo, check->current_step->string, check->current_step->string_len);
2057 *check->bo->p = '\0'; /* to make gdb output easier to read */
2058
Willy Tarreauabca5b62013-12-06 14:19:25 +01002059 /* go to next rule and try to send */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002060 cur = (struct tcpcheck_rule *)cur->list.n;
2061 check->current_step = cur;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002062 } /* end 'send' */
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002063 else if (check->current_step->action == TCPCHK_ACT_EXPECT) {
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002064 if (unlikely(check->result == CHK_RES_FAILED))
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002065 goto out_end_tcpcheck;
2066
Willy Tarreau310987a2014-01-22 19:46:33 +01002067 if (conn->xprt->rcv_buf(conn, check->bi, check->bi->size) <= 0) {
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002068 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH)) {
2069 done = 1;
2070 if ((conn->flags & CO_FL_ERROR) && !check->bi->i) {
2071 /* Report network errors only if we got no other data. Otherwise
2072 * we'll let the upper layers decide whether the response is OK
2073 * or not. It is very common that an RST sent by the server is
2074 * reported as an error just after the last data chunk.
2075 */
2076 chk_report_conn_err(conn, errno, 0);
2077 goto out_end_tcpcheck;
2078 }
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002079 }
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002080 else
2081 goto out_need_io;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002082 }
2083
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002084 /* mark the step as started */
2085 check->last_started_step = check->current_step;
2086
2087
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002088 /* Intermediate or complete response received.
2089 * Terminate string in check->bi->data buffer.
2090 */
2091 if (check->bi->i < check->bi->size) {
2092 check->bi->data[check->bi->i] = '\0';
2093 }
2094 else {
2095 check->bi->data[check->bi->i - 1] = '\0';
2096 done = 1; /* buffer full, don't wait for more data */
2097 }
2098
2099 contentptr = check->bi->data;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002100
2101 /* Check that response body is not empty... */
Willy Tarreauec6b0122014-05-13 17:57:29 +02002102 if (!check->bi->i) {
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002103 if (!done)
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002104 continue;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002105
2106 /* empty response */
2107 chunk_printf(&trash, "TCPCHK got an empty response at step %d",
2108 tcpcheck_get_step_id(s));
2109 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2110
2111 goto out_end_tcpcheck;
2112 }
2113
2114 if (!done && (cur->string != NULL) && (check->bi->i < cur->string_len) )
Willy Tarreaua970c282013-12-06 12:47:19 +01002115 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002116
Willy Tarreaua970c282013-12-06 12:47:19 +01002117 tcpcheck_expect:
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002118 if (cur->string != NULL)
Willy Tarreauec6b0122014-05-13 17:57:29 +02002119 ret = my_memmem(contentptr, check->bi->i, cur->string, cur->string_len) != NULL;
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002120 else if (cur->expect_regex != NULL)
2121 ret = regexec(cur->expect_regex, contentptr, MAX_MATCH, pmatch, 0) == 0;
2122
2123 if (!ret && !done)
Willy Tarreaua970c282013-12-06 12:47:19 +01002124 continue; /* try to read more */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002125
2126 /* matched */
2127 if (ret) {
2128 /* matched but we did not want to => ERROR */
2129 if (cur->inverse) {
2130 /* we were looking for a string */
2131 if (cur->string != NULL) {
2132 chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d",
2133 cur->string, tcpcheck_get_step_id(s));
2134 }
2135 else {
2136 /* we were looking for a regex */
2137 chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d",
2138 tcpcheck_get_step_id(s));
2139 }
2140 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2141 goto out_end_tcpcheck;
2142 }
2143 /* matched and was supposed to => OK, next step */
2144 else {
2145 cur = (struct tcpcheck_rule*)cur->list.n;
2146 check->current_step = cur;
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002147 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002148 goto tcpcheck_expect;
2149 __conn_data_stop_recv(conn);
2150 }
2151 }
2152 else {
2153 /* not matched */
2154 /* not matched and was not supposed to => OK, next step */
2155 if (cur->inverse) {
2156 cur = (struct tcpcheck_rule*)cur->list.n;
2157 check->current_step = cur;
Willy Tarreau98aec9f2013-12-06 16:16:41 +01002158 if (check->current_step->action == TCPCHK_ACT_EXPECT)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002159 goto tcpcheck_expect;
2160 __conn_data_stop_recv(conn);
2161 }
2162 /* not matched but was supposed to => ERROR */
2163 else {
2164 /* we were looking for a string */
2165 if (cur->string != NULL) {
2166 chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d",
2167 cur->string, tcpcheck_get_step_id(s));
2168 }
2169 else {
2170 /* we were looking for a regex */
2171 chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d",
2172 tcpcheck_get_step_id(s));
2173 }
2174 set_server_check_status(check, HCHK_STATUS_L7RSP, trash.str);
2175 goto out_end_tcpcheck;
2176 }
2177 }
2178 } /* end expect */
2179 } /* end loop over double chained step list */
2180
2181 set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)");
2182 goto out_end_tcpcheck;
2183
Willy Tarreaufbe0edf2013-12-06 16:54:31 +01002184 out_need_io:
2185 if (check->bo->o)
2186 __conn_data_want_send(conn);
2187
2188 if (check->current_step->action == TCPCHK_ACT_EXPECT)
2189 __conn_data_want_recv(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002190 return;
2191
2192 out_end_tcpcheck:
2193 /* collect possible new errors */
2194 if (conn->flags & CO_FL_ERROR)
2195 chk_report_conn_err(conn, 0, 0);
2196
Baptiste Assmann69e273f2013-12-11 00:52:19 +01002197 /* cleanup before leaving */
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002198 check->current_step = NULL;
2199
Willy Tarreau6aaa1b82013-12-11 17:09:34 +01002200 if (check->result == CHK_RES_FAILED)
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002201 conn->flags |= CO_FL_ERROR;
2202
2203 __conn_data_stop_both(conn);
Baptiste Assmann5ecb77f2013-10-06 23:24:13 +02002204
2205 return;
2206}
2207
2208
Willy Tarreaubd741542010-03-16 18:46:54 +01002209/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02002210 * Local variables:
2211 * c-indent-level: 8
2212 * c-basic-offset: 8
2213 * End:
2214 */