Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Health-checks functions. |
| 3 | * |
Willy Tarreau | 26c2506 | 2009-03-08 09:38:41 +0100 | [diff] [blame] | 4 | * Copyright 2000-2009 Willy Tarreau <w@1wt.eu> |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 5 | * Copyright 2007-2009 Krzysztof Piotr Oledzki <ole@ans.pl> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 6 | * |
| 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 Tarreau | b881608 | 2008-01-18 12:18:15 +0100 | [diff] [blame] | 14 | #include <assert.h> |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 15 | #include <ctype.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 16 | #include <errno.h> |
| 17 | #include <fcntl.h> |
Willy Tarreau | 9b39dc5 | 2014-07-08 00:54:10 +0200 | [diff] [blame] | 18 | #include <signal.h> |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 19 | #include <stdarg.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 20 | #include <stdio.h> |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 21 | #include <stdlib.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 22 | #include <string.h> |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 23 | #include <time.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 24 | #include <unistd.h> |
Willy Tarreau | 9f6dc72 | 2019-03-01 11:15:10 +0100 | [diff] [blame] | 25 | #include <sys/resource.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 26 | #include <sys/socket.h> |
Dmitry Sivachenko | caf5898 | 2009-08-24 15:11:06 +0400 | [diff] [blame] | 27 | #include <sys/types.h> |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 28 | #include <sys/wait.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 29 | #include <netinet/in.h> |
Willy Tarreau | 1274bc4 | 2009-07-15 07:16:31 +0200 | [diff] [blame] | 30 | #include <netinet/tcp.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 31 | #include <arpa/inet.h> |
| 32 | |
Willy Tarreau | c7e4238 | 2012-08-24 19:22:53 +0200 | [diff] [blame] | 33 | #include <common/chunk.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 34 | #include <common/compat.h> |
| 35 | #include <common/config.h> |
| 36 | #include <common/mini-clist.h> |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 37 | #include <common/standard.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 38 | #include <common/time.h> |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 39 | #include <common/hathreads.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 40 | |
| 41 | #include <types/global.h> |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 42 | #include <types/dns.h> |
William Lallemand | 9ed6203 | 2016-11-21 17:49:11 +0100 | [diff] [blame] | 43 | #include <types/stats.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 44 | |
| 45 | #include <proto/backend.h> |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 46 | #include <proto/checks.h> |
William Lallemand | 9ed6203 | 2016-11-21 17:49:11 +0100 | [diff] [blame] | 47 | #include <proto/stats.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 48 | #include <proto/fd.h> |
| 49 | #include <proto/log.h> |
Willy Tarreau | 53a4766 | 2017-08-28 10:53:00 +0200 | [diff] [blame] | 50 | #include <proto/mux_pt.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 51 | #include <proto/queue.h> |
Willy Tarreau | c6f4ce8 | 2009-06-10 11:09:37 +0200 | [diff] [blame] | 52 | #include <proto/port_range.h> |
Willy Tarreau | e8c66af | 2008-01-13 18:40:14 +0100 | [diff] [blame] | 53 | #include <proto/proto_tcp.h> |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 54 | #include <proto/protocol.h> |
Willy Tarreau | 2b5652f | 2006-12-31 17:46:05 +0100 | [diff] [blame] | 55 | #include <proto/proxy.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 56 | #include <proto/server.h> |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 57 | #include <proto/signal.h> |
Willy Tarreau | 9e000c6 | 2011-03-10 14:03:36 +0100 | [diff] [blame] | 58 | #include <proto/stream_interface.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 59 | #include <proto/task.h> |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 60 | #include <proto/log.h> |
| 61 | #include <proto/dns.h> |
| 62 | #include <proto/proto_udp.h> |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 63 | #include <proto/ssl_sock.h> |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 64 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 65 | static int httpchk_expect(struct server *s, int done); |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 66 | static int tcpcheck_get_step_id(struct check *); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 67 | static char * tcpcheck_get_step_comment(struct check *, int); |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 68 | static int tcpcheck_main(struct check *); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 69 | static void __event_srv_chk_w(struct conn_stream *cs); |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 70 | static int wake_srv_chk(struct conn_stream *cs); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 71 | static void __event_srv_chk_r(struct conn_stream *cs); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 72 | |
Christopher Faulet | 31c30fd | 2020-03-26 21:10:03 +0100 | [diff] [blame] | 73 | static int srv_check_healthcheck_port(struct check *chk); |
| 74 | |
Willy Tarreau | 8ceae72 | 2018-11-26 11:58:30 +0100 | [diff] [blame] | 75 | DECLARE_STATIC_POOL(pool_head_email_alert, "email_alert", sizeof(struct email_alert)); |
| 76 | DECLARE_STATIC_POOL(pool_head_tcpcheck_rule, "tcpcheck_rule", sizeof(struct tcpcheck_rule)); |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 77 | |
| 78 | |
Simon Horman | 63a4a82 | 2012-03-19 07:24:41 +0900 | [diff] [blame] | 79 | static const struct check_status check_statuses[HCHK_STATUS_SIZE] = { |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 80 | [HCHK_STATUS_UNKNOWN] = { CHK_RES_UNKNOWN, "UNK", "Unknown" }, |
| 81 | [HCHK_STATUS_INI] = { CHK_RES_UNKNOWN, "INI", "Initializing" }, |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 82 | [HCHK_STATUS_START] = { /* SPECIAL STATUS*/ }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 83 | |
Willy Tarreau | 2396418 | 2014-05-20 20:56:30 +0200 | [diff] [blame] | 84 | /* Below we have finished checks */ |
| 85 | [HCHK_STATUS_CHECKED] = { CHK_RES_NEUTRAL, "CHECKED", "No status change" }, |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 86 | [HCHK_STATUS_HANA] = { CHK_RES_FAILED, "HANA", "Health analyze" }, |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 87 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 88 | [HCHK_STATUS_SOCKERR] = { CHK_RES_FAILED, "SOCKERR", "Socket error" }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 89 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 90 | [HCHK_STATUS_L4OK] = { CHK_RES_PASSED, "L4OK", "Layer4 check passed" }, |
| 91 | [HCHK_STATUS_L4TOUT] = { CHK_RES_FAILED, "L4TOUT", "Layer4 timeout" }, |
| 92 | [HCHK_STATUS_L4CON] = { CHK_RES_FAILED, "L4CON", "Layer4 connection problem" }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 93 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 94 | [HCHK_STATUS_L6OK] = { CHK_RES_PASSED, "L6OK", "Layer6 check passed" }, |
| 95 | [HCHK_STATUS_L6TOUT] = { CHK_RES_FAILED, "L6TOUT", "Layer6 timeout" }, |
| 96 | [HCHK_STATUS_L6RSP] = { CHK_RES_FAILED, "L6RSP", "Layer6 invalid response" }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 97 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 98 | [HCHK_STATUS_L7TOUT] = { CHK_RES_FAILED, "L7TOUT", "Layer7 timeout" }, |
| 99 | [HCHK_STATUS_L7RSP] = { CHK_RES_FAILED, "L7RSP", "Layer7 invalid response" }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 100 | |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 101 | [HCHK_STATUS_L57DATA] = { /* DUMMY STATUS */ }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 102 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 103 | [HCHK_STATUS_L7OKD] = { CHK_RES_PASSED, "L7OK", "Layer7 check passed" }, |
| 104 | [HCHK_STATUS_L7OKCD] = { CHK_RES_CONDPASS, "L7OKC", "Layer7 check conditionally passed" }, |
| 105 | [HCHK_STATUS_L7STS] = { CHK_RES_FAILED, "L7STS", "Layer7 wrong status" }, |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 106 | |
| 107 | [HCHK_STATUS_PROCERR] = { CHK_RES_FAILED, "PROCERR", "External check error" }, |
| 108 | [HCHK_STATUS_PROCTOUT] = { CHK_RES_FAILED, "PROCTOUT", "External check timeout" }, |
Cyril Bonté | 77010d8 | 2014-08-07 01:55:37 +0200 | [diff] [blame] | 109 | [HCHK_STATUS_PROCOK] = { CHK_RES_PASSED, "PROCOK", "External check passed" }, |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 110 | }; |
| 111 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 112 | const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = { |
| 113 | [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT }, |
| 114 | [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT }, |
| 115 | [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT }, |
| 116 | [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT }, |
| 117 | [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT }, |
| 118 | [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT }, |
| 119 | [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT }, |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 120 | [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_ADDR }, |
| 121 | [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_UINT }, |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 122 | [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT }, |
| 123 | [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG }, |
| 124 | }; |
| 125 | |
Simon Horman | 63a4a82 | 2012-03-19 07:24:41 +0900 | [diff] [blame] | 126 | static const struct analyze_status analyze_statuses[HANA_STATUS_SIZE] = { /* 0: ignore, 1: error, 2: OK */ |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 127 | [HANA_STATUS_UNKNOWN] = { "Unknown", { 0, 0 }}, |
| 128 | |
| 129 | [HANA_STATUS_L4_OK] = { "L4 successful connection", { 2, 0 }}, |
| 130 | [HANA_STATUS_L4_ERR] = { "L4 unsuccessful connection", { 1, 1 }}, |
| 131 | |
| 132 | [HANA_STATUS_HTTP_OK] = { "Correct http response", { 0, 2 }}, |
| 133 | [HANA_STATUS_HTTP_STS] = { "Wrong http response", { 0, 1 }}, |
| 134 | [HANA_STATUS_HTTP_HDRRSP] = { "Invalid http response (headers)", { 0, 1 }}, |
| 135 | [HANA_STATUS_HTTP_RSP] = { "Invalid http response", { 0, 1 }}, |
| 136 | |
| 137 | [HANA_STATUS_HTTP_READ_ERROR] = { "Read error (http)", { 0, 1 }}, |
| 138 | [HANA_STATUS_HTTP_READ_TIMEOUT] = { "Read timeout (http)", { 0, 1 }}, |
| 139 | [HANA_STATUS_HTTP_BROKEN_PIPE] = { "Close from server (http)", { 0, 1 }}, |
| 140 | }; |
| 141 | |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 142 | /* checks if <err> is a real error for errno or one that can be ignored, and |
| 143 | * return 0 for these ones or <err> for real ones. |
| 144 | */ |
| 145 | static inline int unclean_errno(int err) |
| 146 | { |
| 147 | if (err == EAGAIN || err == EINPROGRESS || |
| 148 | err == EISCONN || err == EALREADY) |
| 149 | return 0; |
| 150 | return err; |
| 151 | } |
| 152 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 153 | /* |
| 154 | * Convert check_status code to description |
| 155 | */ |
| 156 | const char *get_check_status_description(short check_status) { |
| 157 | |
| 158 | const char *desc; |
| 159 | |
| 160 | if (check_status < HCHK_STATUS_SIZE) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 161 | desc = check_statuses[check_status].desc; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 162 | else |
| 163 | desc = NULL; |
| 164 | |
| 165 | if (desc && *desc) |
| 166 | return desc; |
| 167 | else |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 168 | return check_statuses[HCHK_STATUS_UNKNOWN].desc; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 169 | } |
| 170 | |
| 171 | /* |
| 172 | * Convert check_status code to short info |
| 173 | */ |
| 174 | const char *get_check_status_info(short check_status) { |
| 175 | |
| 176 | const char *info; |
| 177 | |
| 178 | if (check_status < HCHK_STATUS_SIZE) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 179 | info = check_statuses[check_status].info; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 180 | else |
| 181 | info = NULL; |
| 182 | |
| 183 | if (info && *info) |
| 184 | return info; |
| 185 | else |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 186 | return check_statuses[HCHK_STATUS_UNKNOWN].info; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 187 | } |
| 188 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 189 | const char *get_analyze_status(short analyze_status) { |
| 190 | |
| 191 | const char *desc; |
| 192 | |
| 193 | if (analyze_status < HANA_STATUS_SIZE) |
| 194 | desc = analyze_statuses[analyze_status].desc; |
| 195 | else |
| 196 | desc = NULL; |
| 197 | |
| 198 | if (desc && *desc) |
| 199 | return desc; |
| 200 | else |
| 201 | return analyze_statuses[HANA_STATUS_UNKNOWN].desc; |
| 202 | } |
| 203 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 204 | /* |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 205 | * Set check->status, update check->duration and fill check->result with |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 206 | * an adequate CHK_RES_* value. The new check->health is computed based |
| 207 | * on the result. |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 208 | * |
| 209 | * Show information in logs about failed health check if server is UP |
| 210 | * or succeeded health checks if server is DOWN. |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 211 | */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 212 | static void set_server_check_status(struct check *check, short status, const char *desc) |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 213 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 214 | struct server *s = check->server; |
Willy Tarreau | bef1b32 | 2014-05-13 21:01:39 +0200 | [diff] [blame] | 215 | short prev_status = check->status; |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 216 | int report = 0; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 217 | |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 218 | if (status == HCHK_STATUS_START) { |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 219 | check->result = CHK_RES_UNKNOWN; /* no result yet */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 220 | check->desc[0] = '\0'; |
| 221 | check->start = now; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 222 | return; |
| 223 | } |
| 224 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 225 | if (!check->status) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 226 | return; |
| 227 | |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 228 | if (desc && *desc) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 229 | strncpy(check->desc, desc, HCHK_DESC_LEN-1); |
| 230 | check->desc[HCHK_DESC_LEN-1] = '\0'; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 231 | } else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 232 | check->desc[0] = '\0'; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 233 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 234 | check->status = status; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 235 | if (check_statuses[status].result) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 236 | check->result = check_statuses[status].result; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 237 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 238 | if (status == HCHK_STATUS_HANA) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 239 | check->duration = -1; |
| 240 | else if (!tv_iszero(&check->start)) { |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 241 | /* set_server_check_status() may be called more than once */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 242 | check->duration = tv_ms_elapsed(&check->start, &now); |
| 243 | tv_zero(&check->start); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 244 | } |
| 245 | |
Willy Tarreau | 2396418 | 2014-05-20 20:56:30 +0200 | [diff] [blame] | 246 | /* no change is expected if no state change occurred */ |
| 247 | if (check->result == CHK_RES_NEUTRAL) |
| 248 | return; |
| 249 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 250 | /* If the check was really just sending a mail, it won't have an |
| 251 | * associated server, so we're done now. |
| 252 | */ |
| 253 | if (!s) |
| 254 | return; |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 255 | report = 0; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 256 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 257 | switch (check->result) { |
| 258 | case CHK_RES_FAILED: |
Willy Tarreau | 12634e1 | 2014-05-23 11:32:36 +0200 | [diff] [blame] | 259 | /* Failure to connect to the agent as a secondary check should not |
| 260 | * cause the server to be marked down. |
| 261 | */ |
| 262 | if ((!(check->state & CHK_ST_AGENT) || |
Simon Horman | eaabd52 | 2015-02-26 11:26:17 +0900 | [diff] [blame] | 263 | (check->status >= HCHK_STATUS_L57DATA)) && |
Christopher Faulet | b119a79 | 2018-05-02 12:12:45 +0200 | [diff] [blame] | 264 | (check->health > 0)) { |
Olivier Houchard | 7059c55 | 2019-03-08 18:49:32 +0100 | [diff] [blame] | 265 | _HA_ATOMIC_ADD(&s->counters.failed_checks, 1); |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 266 | report = 1; |
| 267 | check->health--; |
| 268 | if (check->health < check->rise) |
| 269 | check->health = 0; |
| 270 | } |
| 271 | break; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 272 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 273 | case CHK_RES_PASSED: |
| 274 | case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */ |
| 275 | if ((check->health < check->rise + check->fall - 1) && |
| 276 | (check->result == CHK_RES_PASSED || check->health > 0)) { |
| 277 | report = 1; |
| 278 | check->health++; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 279 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 280 | if (check->health >= check->rise) |
| 281 | check->health = check->rise + check->fall - 1; /* OK now */ |
| 282 | } |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 283 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 284 | /* clear consecutive_errors if observing is enabled */ |
| 285 | if (s->onerror) |
| 286 | s->consecutive_errors = 0; |
| 287 | break; |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 288 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 289 | default: |
| 290 | break; |
| 291 | } |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 292 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 293 | if (s->proxy->options2 & PR_O2_LOGHCHKS && |
| 294 | (status != prev_status || report)) { |
| 295 | chunk_printf(&trash, |
Willy Tarreau | 12634e1 | 2014-05-23 11:32:36 +0200 | [diff] [blame] | 296 | "%s check for %sserver %s/%s %s%s", |
| 297 | (check->state & CHK_ST_AGENT) ? "Agent" : "Health", |
Willy Tarreau | c93cd16 | 2014-05-13 15:54:22 +0200 | [diff] [blame] | 298 | s->flags & SRV_F_BACKUP ? "backup " : "", |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 299 | s->proxy->id, s->id, |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 300 | (check->result == CHK_RES_CONDPASS) ? "conditionally ":"", |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 301 | (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed"); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 302 | |
Emeric Brun | 5a13351 | 2017-10-19 14:42:30 +0200 | [diff] [blame] | 303 | srv_append_status(&trash, s, check, -1, 0); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 304 | |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 305 | chunk_appendf(&trash, ", status: %d/%d %s", |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 306 | (check->health >= check->rise) ? check->health - check->rise + 1 : check->health, |
| 307 | (check->health >= check->rise) ? check->fall : check->rise, |
| 308 | (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN"); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 309 | |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 310 | ha_warning("%s.\n", trash.area); |
| 311 | send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area); |
| 312 | send_email_alert(s, LOG_INFO, "%s", trash.area); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 313 | } |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 314 | } |
| 315 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 316 | /* Marks the check <check>'s server down if the current check is already failed |
| 317 | * and the server is not down yet nor in maintenance. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 318 | */ |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 319 | static void check_notify_failure(struct check *check) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 320 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 321 | struct server *s = check->server; |
Simon Horman | e0d1bfb | 2011-06-21 14:34:58 +0900 | [diff] [blame] | 322 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 323 | /* The agent secondary check should only cause a server to be marked |
| 324 | * as down if check->status is HCHK_STATUS_L7STS, which indicates |
| 325 | * that the agent returned "fail", "stopped" or "down". |
| 326 | * The implication here is that failure to connect to the agent |
| 327 | * as a secondary check should not cause the server to be marked |
| 328 | * down. */ |
| 329 | if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS) |
| 330 | return; |
| 331 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 332 | if (check->health > 0) |
| 333 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 334 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 335 | /* We only report a reason for the check if we did not do so previously */ |
Emeric Brun | 5a13351 | 2017-10-19 14:42:30 +0200 | [diff] [blame] | 336 | srv_set_stopped(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 337 | } |
| 338 | |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 339 | /* Marks the check <check> as valid and tries to set its server up, provided |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 340 | * it isn't in maintenance, it is not tracking a down server and other checks |
| 341 | * comply. The rule is simple : by default, a server is up, unless any of the |
| 342 | * following conditions is true : |
| 343 | * - health check failed (check->health < rise) |
| 344 | * - agent check failed (agent->health < rise) |
| 345 | * - the server tracks a down server (track && track->state == STOPPED) |
| 346 | * Note that if the server has a slowstart, it will switch to STARTING instead |
| 347 | * of RUNNING. Also, only the health checks support the nolb mode, so the |
| 348 | * agent's success may not take the server out of this mode. |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 349 | */ |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 350 | static void check_notify_success(struct check *check) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 351 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 352 | struct server *s = check->server; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 353 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 354 | if (s->next_admin & SRV_ADMF_MAINT) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 355 | return; |
Cyril Bonté | cd19e51 | 2010-01-31 22:34:03 +0100 | [diff] [blame] | 356 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 357 | if (s->track && s->track->next_state == SRV_ST_STOPPED) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 358 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 359 | |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 360 | if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise)) |
| 361 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 362 | |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 363 | if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise)) |
| 364 | return; |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 365 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 366 | if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING) |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 367 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 368 | |
Emeric Brun | 5a13351 | 2017-10-19 14:42:30 +0200 | [diff] [blame] | 369 | srv_set_running(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL); |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 370 | } |
| 371 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 372 | /* Marks the check <check> as valid and tries to set its server into stopping mode |
| 373 | * if it was running or starting, and provided it isn't in maintenance and other |
| 374 | * checks comply. The conditions for the server to be marked in stopping mode are |
| 375 | * the same as for it to be turned up. Also, only the health checks support the |
| 376 | * nolb mode. |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 377 | */ |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 378 | static void check_notify_stopping(struct check *check) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 379 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 380 | struct server *s = check->server; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 381 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 382 | if (s->next_admin & SRV_ADMF_MAINT) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 383 | return; |
| 384 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 385 | if (check->state & CHK_ST_AGENT) |
| 386 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 387 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 388 | if (s->track && s->track->next_state == SRV_ST_STOPPED) |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 389 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 390 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 391 | if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise)) |
| 392 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 393 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 394 | if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise)) |
| 395 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 396 | |
Willy Tarreau | b26881a | 2017-12-23 11:16:49 +0100 | [diff] [blame] | 397 | srv_set_stopping(s, NULL, (!s->track && !(s->proxy->options2 & PR_O2_LOGHCHKS)) ? check : NULL); |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 398 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 399 | |
Willy Tarreau | 9fe7aae | 2013-12-31 23:47:37 +0100 | [diff] [blame] | 400 | /* note: use health_adjust() only, which first checks that the observe mode is |
| 401 | * enabled. |
| 402 | */ |
| 403 | void __health_adjust(struct server *s, short status) |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 404 | { |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 405 | int failed; |
| 406 | int expire; |
| 407 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 408 | if (s->observe >= HANA_OBS_SIZE) |
| 409 | return; |
| 410 | |
Willy Tarreau | bb95666 | 2013-01-24 00:37:39 +0100 | [diff] [blame] | 411 | if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc) |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 412 | return; |
| 413 | |
| 414 | switch (analyze_statuses[status].lr[s->observe - 1]) { |
| 415 | case 1: |
| 416 | failed = 1; |
| 417 | break; |
| 418 | |
| 419 | case 2: |
| 420 | failed = 0; |
| 421 | break; |
| 422 | |
| 423 | default: |
| 424 | return; |
| 425 | } |
| 426 | |
| 427 | if (!failed) { |
| 428 | /* good: clear consecutive_errors */ |
| 429 | s->consecutive_errors = 0; |
| 430 | return; |
| 431 | } |
| 432 | |
Olivier Houchard | 7059c55 | 2019-03-08 18:49:32 +0100 | [diff] [blame] | 433 | _HA_ATOMIC_ADD(&s->consecutive_errors, 1); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 434 | |
| 435 | if (s->consecutive_errors < s->consecutive_errors_limit) |
| 436 | return; |
| 437 | |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 438 | chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s", |
| 439 | s->consecutive_errors, get_analyze_status(status)); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 440 | |
| 441 | switch (s->onerror) { |
| 442 | case HANA_ONERR_FASTINTER: |
| 443 | /* force fastinter - nothing to do here as all modes force it */ |
| 444 | break; |
| 445 | |
| 446 | case HANA_ONERR_SUDDTH: |
| 447 | /* simulate a pre-fatal failed health check */ |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 448 | if (s->check.health > s->check.rise) |
| 449 | s->check.health = s->check.rise + 1; |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 450 | |
| 451 | /* no break - fall through */ |
| 452 | |
| 453 | case HANA_ONERR_FAILCHK: |
| 454 | /* simulate a failed health check */ |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 455 | set_server_check_status(&s->check, HCHK_STATUS_HANA, |
| 456 | trash.area); |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 457 | check_notify_failure(&s->check); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 458 | break; |
| 459 | |
| 460 | case HANA_ONERR_MARKDWN: |
| 461 | /* mark server down */ |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 462 | s->check.health = s->check.rise; |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 463 | set_server_check_status(&s->check, HCHK_STATUS_HANA, |
| 464 | trash.area); |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 465 | check_notify_failure(&s->check); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 466 | break; |
| 467 | |
| 468 | default: |
| 469 | /* write a warning? */ |
| 470 | break; |
| 471 | } |
| 472 | |
| 473 | s->consecutive_errors = 0; |
Olivier Houchard | 7059c55 | 2019-03-08 18:49:32 +0100 | [diff] [blame] | 474 | _HA_ATOMIC_ADD(&s->counters.failed_hana, 1); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 475 | |
Simon Horman | 6618300 | 2013-02-23 10:16:43 +0900 | [diff] [blame] | 476 | if (s->check.fastinter) { |
| 477 | expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter)); |
Sergiy Prykhodko | 1d57e50 | 2013-09-21 12:05:00 +0300 | [diff] [blame] | 478 | if (s->check.task->expire > expire) { |
Willy Tarreau | 5b3a202 | 2012-09-28 15:01:02 +0200 | [diff] [blame] | 479 | s->check.task->expire = expire; |
Sergiy Prykhodko | 1d57e50 | 2013-09-21 12:05:00 +0300 | [diff] [blame] | 480 | /* requeue check task with new expire */ |
| 481 | task_queue(s->check.task); |
| 482 | } |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 483 | } |
| 484 | } |
| 485 | |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 486 | static int httpchk_build_status_header(struct server *s, char *buffer, int size) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 487 | { |
| 488 | int sv_state; |
| 489 | int ratio; |
| 490 | int hlen = 0; |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 491 | char addr[46]; |
| 492 | char port[6]; |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 493 | const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d", |
| 494 | "UP %d/%d", "UP", |
| 495 | "NOLB %d/%d", "NOLB", |
| 496 | "no check" }; |
| 497 | |
| 498 | memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24); |
| 499 | hlen += 24; |
| 500 | |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 501 | if (!(s->check.state & CHK_ST_ENABLED)) |
| 502 | sv_state = 6; |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 503 | else if (s->cur_state != SRV_ST_STOPPED) { |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 504 | if (s->check.health == s->check.rise + s->check.fall - 1) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 505 | sv_state = 3; /* UP */ |
| 506 | else |
| 507 | sv_state = 2; /* going down */ |
| 508 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 509 | if (s->cur_state == SRV_ST_STOPPING) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 510 | sv_state += 2; |
| 511 | } else { |
Simon Horman | 125d099 | 2013-02-24 17:23:38 +0900 | [diff] [blame] | 512 | if (s->check.health) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 513 | sv_state = 1; /* going up */ |
| 514 | else |
| 515 | sv_state = 0; /* DOWN */ |
| 516 | } |
| 517 | |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 518 | hlen += snprintf(buffer + hlen, size - hlen, |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 519 | srv_hlt_st[sv_state], |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 520 | (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health), |
| 521 | (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise)); |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 522 | |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 523 | addr_to_str(&s->addr, addr, sizeof(addr)); |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 524 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 525 | snprintf(port, sizeof(port), "%u", s->svc_port); |
| 526 | else |
| 527 | *port = 0; |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 528 | |
| 529 | hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d", |
| 530 | addr, port, s->proxy->id, s->id, |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 531 | global.node, |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 532 | (s->cur_eweight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv, |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 533 | (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv, |
| 534 | s->cur_sess, s->proxy->beconn - s->proxy->nbpend, |
| 535 | s->nbpend); |
| 536 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 537 | if ((s->cur_state == SRV_ST_STARTING) && |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 538 | now.tv_sec < s->last_change + s->slowstart && |
| 539 | now.tv_sec >= s->last_change) { |
| 540 | ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart); |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 541 | hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio); |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 542 | } |
| 543 | |
| 544 | buffer[hlen++] = '\r'; |
| 545 | buffer[hlen++] = '\n'; |
| 546 | |
| 547 | return hlen; |
| 548 | } |
| 549 | |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 550 | /* Check the connection. If an error has already been reported or the socket is |
| 551 | * closed, keep errno intact as it is supposed to contain the valid error code. |
| 552 | * If no error is reported, check the socket's error queue using getsockopt(). |
| 553 | * Warning, this must be done only once when returning from poll, and never |
| 554 | * after an I/O error was attempted, otherwise the error queue might contain |
| 555 | * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the |
| 556 | * socket. Returns non-zero if an error was reported, zero if everything is |
| 557 | * clean (including a properly closed socket). |
| 558 | */ |
| 559 | static int retrieve_errno_from_socket(struct connection *conn) |
| 560 | { |
| 561 | int skerr; |
| 562 | socklen_t lskerr = sizeof(skerr); |
| 563 | |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 564 | if (conn->flags & CO_FL_ERROR && (unclean_errno(errno) || !conn->ctrl)) |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 565 | return 1; |
| 566 | |
Willy Tarreau | 3c72872 | 2014-01-23 13:50:42 +0100 | [diff] [blame] | 567 | if (!conn_ctrl_ready(conn)) |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 568 | return 0; |
| 569 | |
Willy Tarreau | 585744b | 2017-08-24 14:31:19 +0200 | [diff] [blame] | 570 | if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0) |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 571 | errno = skerr; |
| 572 | |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 573 | errno = unclean_errno(errno); |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 574 | |
| 575 | if (!errno) { |
| 576 | /* we could not retrieve an error, that does not mean there is |
| 577 | * none. Just don't change anything and only report the prior |
| 578 | * error if any. |
| 579 | */ |
| 580 | if (conn->flags & CO_FL_ERROR) |
| 581 | return 1; |
| 582 | else |
| 583 | return 0; |
| 584 | } |
| 585 | |
| 586 | conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH; |
| 587 | return 1; |
| 588 | } |
| 589 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 590 | /* Try to collect as much information as possible on the connection status, |
| 591 | * and adjust the server status accordingly. It may make use of <errno_bck> |
| 592 | * if non-null when the caller is absolutely certain of its validity (eg: |
| 593 | * checked just after a syscall). If the caller doesn't have a valid errno, |
| 594 | * it can pass zero, and retrieve_errno_from_socket() will be called to try |
| 595 | * to extract errno from the socket. If no error is reported, it will consider |
| 596 | * the <expired> flag. This is intended to be used when a connection error was |
| 597 | * reported in conn->flags or when a timeout was reported in <expired>. The |
| 598 | * function takes care of not updating a server status which was already set. |
| 599 | * All situations where at least one of <expired> or CO_FL_ERROR are set |
| 600 | * produce a status. |
| 601 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 602 | static void chk_report_conn_err(struct check *check, int errno_bck, int expired) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 603 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 604 | struct conn_stream *cs = check->cs; |
| 605 | struct connection *conn = cs_conn(cs); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 606 | const char *err_msg; |
Willy Tarreau | 83061a8 | 2018-07-13 11:56:34 +0200 | [diff] [blame] | 607 | struct buffer *chk; |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 608 | int step; |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 609 | char *comment; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 610 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 611 | if (check->result != CHK_RES_UNKNOWN) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 612 | return; |
| 613 | |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 614 | errno = unclean_errno(errno_bck); |
| 615 | if (conn && errno) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 616 | retrieve_errno_from_socket(conn); |
| 617 | |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 618 | if (conn && !(conn->flags & CO_FL_ERROR) && |
| 619 | !(cs->flags & CS_FL_ERROR) && !expired) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 620 | return; |
| 621 | |
| 622 | /* we'll try to build a meaningful error message depending on the |
| 623 | * context of the error possibly present in conn->err_code, and the |
| 624 | * socket error possibly collected above. This is useful to know the |
| 625 | * exact step of the L6 layer (eg: SSL handshake). |
| 626 | */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 627 | chk = get_trash_chunk(); |
| 628 | |
| 629 | if (check->type == PR_O2_TCPCHK_CHK) { |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 630 | step = tcpcheck_get_step_id(check); |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 631 | if (!step) |
| 632 | chunk_printf(chk, " at initial connection step of tcp-check"); |
| 633 | else { |
| 634 | chunk_printf(chk, " at step %d of tcp-check", step); |
| 635 | /* we were looking for a string */ |
| 636 | if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) { |
| 637 | if (check->last_started_step->port) |
| 638 | chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port); |
| 639 | else |
| 640 | chunk_appendf(chk, " (connect)"); |
| 641 | } |
| 642 | else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) { |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 643 | struct tcpcheck_expect *expect = &check->last_started_step->expect; |
| 644 | |
| 645 | switch (expect->type) { |
| 646 | case TCPCHK_EXPECT_STRING: |
| 647 | chunk_appendf(chk, " (expect string '%s')", expect->string); |
| 648 | break; |
| 649 | case TCPCHK_EXPECT_BINARY: |
| 650 | chunk_appendf(chk, " (expect binary '%s')", expect->string); |
| 651 | break; |
| 652 | case TCPCHK_EXPECT_REGEX: |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 653 | chunk_appendf(chk, " (expect regex)"); |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 654 | break; |
| 655 | case TCPCHK_EXPECT_UNDEF: |
| 656 | chunk_appendf(chk, " (undefined expect!)"); |
| 657 | break; |
| 658 | } |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 659 | } |
| 660 | else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) { |
| 661 | chunk_appendf(chk, " (send)"); |
| 662 | } |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 663 | |
| 664 | comment = tcpcheck_get_step_comment(check, step); |
| 665 | if (comment) |
| 666 | chunk_appendf(chk, " comment: '%s'", comment); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 667 | } |
| 668 | } |
| 669 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 670 | if (conn && conn->err_code) { |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 671 | if (unclean_errno(errno)) |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 672 | chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), |
| 673 | chk->area); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 674 | else |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 675 | chunk_printf(&trash, "%s%s", conn_err_code_str(conn), |
| 676 | chk->area); |
| 677 | err_msg = trash.area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 678 | } |
| 679 | else { |
Willy Tarreau | c8dc20a | 2019-12-27 12:03:27 +0100 | [diff] [blame] | 680 | if (unclean_errno(errno)) { |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 681 | chunk_printf(&trash, "%s%s", strerror(errno), |
| 682 | chk->area); |
| 683 | err_msg = trash.area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 684 | } |
| 685 | else { |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 686 | err_msg = chk->area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 687 | } |
| 688 | } |
| 689 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 690 | if (check->state & CHK_ST_PORT_MISS) { |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 691 | /* NOTE: this is reported after <fall> tries */ |
| 692 | chunk_printf(chk, "No port available for the TCP connection"); |
| 693 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 694 | } |
| 695 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 696 | if (!conn) { |
| 697 | /* connection allocation error before the connection was established */ |
| 698 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 699 | } |
Willy Tarreau | c192b0a | 2020-01-23 09:11:58 +0100 | [diff] [blame] | 700 | else if (conn->flags & CO_FL_WAIT_L4_CONN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 701 | /* L4 not established (yet) */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 702 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 703 | set_server_check_status(check, HCHK_STATUS_L4CON, err_msg); |
| 704 | else if (expired) |
| 705 | set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 706 | |
| 707 | /* |
| 708 | * might be due to a server IP change. |
| 709 | * Let's trigger a DNS resolution if none are currently running. |
| 710 | */ |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 711 | if (check->server) |
| 712 | dns_trigger_resolution(check->server->dns_requester); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 713 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 714 | } |
Willy Tarreau | c192b0a | 2020-01-23 09:11:58 +0100 | [diff] [blame] | 715 | else if (conn->flags & CO_FL_WAIT_L6_CONN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 716 | /* L6 not established (yet) */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 717 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 718 | set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg); |
| 719 | else if (expired) |
| 720 | set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg); |
| 721 | } |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 722 | else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 723 | /* I/O error after connection was established and before we could diagnose */ |
| 724 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 725 | } |
| 726 | else if (expired) { |
| 727 | /* connection established but expired check */ |
| 728 | if (check->type == PR_O2_SSL3_CHK) |
| 729 | set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg); |
| 730 | else /* HTTP, SMTP, ... */ |
| 731 | set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg); |
| 732 | } |
| 733 | |
| 734 | return; |
| 735 | } |
| 736 | |
Olivier Houchard | 5c110b9 | 2018-08-14 17:04:58 +0200 | [diff] [blame] | 737 | /* This function checks if any I/O is wanted, and if so, attempts to do so */ |
| 738 | static struct task *event_srv_chk_io(struct task *t, void *ctx, unsigned short state) |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 739 | { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 740 | struct check *check = ctx; |
| 741 | struct conn_stream *cs = check->cs; |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 742 | struct email_alertq *q = container_of(check, typeof(*q), check); |
Olivier Houchard | bc89ad8 | 2019-07-09 17:28:51 +0200 | [diff] [blame] | 743 | int ret = 0; |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 744 | |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 745 | if (!(check->wait_list.events & SUB_RETRY_SEND)) |
Olivier Houchard | bc89ad8 | 2019-07-09 17:28:51 +0200 | [diff] [blame] | 746 | ret = wake_srv_chk(cs); |
| 747 | if (ret == 0 && !(check->wait_list.events & SUB_RETRY_RECV)) { |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 748 | if (check->server) |
| 749 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
| 750 | else |
| 751 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 752 | __event_srv_chk_r(cs); |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 753 | if (check->server) |
| 754 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
| 755 | else |
| 756 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 757 | } |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 758 | return NULL; |
| 759 | } |
| 760 | |
| 761 | /* same as above but protected by the server lock. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 762 | * |
| 763 | * Please do NOT place any return statement in this function and only leave |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 764 | * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT |
| 765 | * TO USE event_srv_chk_w() instead. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 766 | */ |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 767 | static void __event_srv_chk_w(struct conn_stream *cs) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 768 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 769 | struct connection *conn = cs->conn; |
| 770 | struct check *check = cs->data; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 771 | struct server *s = check->server; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 772 | struct task *t = check->task; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 773 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 774 | if (unlikely(check->result == CHK_RES_FAILED)) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 775 | goto out_wakeup; |
| 776 | |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 777 | if (retrieve_errno_from_socket(conn)) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 778 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 779 | goto out_wakeup; |
| 780 | } |
Krzysztof Piotr Oledzki | 6492db5 | 2010-01-02 22:03:01 +0100 | [diff] [blame] | 781 | |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 782 | /* here, we know that the connection is established. That's enough for |
| 783 | * a pure TCP check. |
| 784 | */ |
| 785 | if (!check->type) |
| 786 | goto out_wakeup; |
| 787 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 788 | /* wake() will take care of calling tcpcheck_main() */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 789 | if (check->type == PR_O2_TCPCHK_CHK) |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 790 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 791 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 792 | if (b_data(&check->bo)) { |
Olivier Houchard | ed0f207 | 2018-08-16 15:41:52 +0200 | [diff] [blame] | 793 | cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0); |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 794 | b_realign_if_empty(&check->bo); |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 795 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 796 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 797 | goto out_wakeup; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 798 | } |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 799 | if (b_data(&check->bo)) { |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 800 | conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 801 | goto out; |
| 802 | } |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 803 | } |
Willy Tarreau | 6996e15 | 2007-04-30 14:37:43 +0200 | [diff] [blame] | 804 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 805 | /* full request sent, we allow up to <timeout.check> if nonzero for a response */ |
| 806 | if (s->proxy->timeout.check) { |
| 807 | t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check); |
| 808 | task_queue(t); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 809 | } |
Olivier Houchard | 53216e7 | 2018-10-10 15:46:36 +0200 | [diff] [blame] | 810 | goto out; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 811 | |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 812 | out_wakeup: |
Willy Tarreau | fdccded | 2008-08-29 18:19:04 +0200 | [diff] [blame] | 813 | task_wakeup(t, TASK_WOKEN_IO); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 814 | out: |
| 815 | return; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 816 | } |
| 817 | |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 818 | /* |
Willy Tarreau | f3c6920 | 2006-07-09 16:42:34 +0200 | [diff] [blame] | 819 | * This function is used only for server health-checks. It handles the server's |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 820 | * reply to an HTTP request, SSL HELLO or MySQL client Auth. It calls |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 821 | * set_server_check_status() to update check->status, check->duration |
| 822 | * and check->result. |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 823 | |
| 824 | * The set_server_check_status function is called with HCHK_STATUS_L7OKD if |
| 825 | * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server |
| 826 | * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in |
| 827 | * response to an SSL HELLO (the principle is that this is enough to |
| 828 | * distinguish between an SSL server and a pure TCP relay). All other cases will |
| 829 | * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP, |
| 830 | * etc. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 831 | * |
| 832 | * Please do NOT place any return statement in this function and only leave |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 833 | * via the out label. |
| 834 | * |
| 835 | * This must be called with the server lock held. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 836 | */ |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 837 | static void __event_srv_chk_r(struct conn_stream *cs) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 838 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 839 | struct connection *conn = cs->conn; |
| 840 | struct check *check = cs->data; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 841 | struct server *s = check->server; |
| 842 | struct task *t = check->task; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 843 | char *desc; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 844 | int done; |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 845 | unsigned short msglen; |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 846 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 847 | if (unlikely(check->result == CHK_RES_FAILED)) |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 848 | goto out_wakeup; |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 849 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 850 | /* wake() will take care of calling tcpcheck_main() */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 851 | if (check->type == PR_O2_TCPCHK_CHK) |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 852 | goto out; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 853 | |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 854 | /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available |
| 855 | * but the connection was closed on the remote end. Fortunately, recv still |
| 856 | * works correctly and we don't need to do the getsockopt() on linux. |
| 857 | */ |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 858 | |
| 859 | /* Set buffer to point to the end of the data already read, and check |
| 860 | * that there is free space remaining. If the buffer is full, proceed |
| 861 | * with running the checks without attempting another socket read. |
| 862 | */ |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 863 | |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 864 | done = 0; |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 865 | |
Olivier Houchard | 511efea | 2018-08-16 15:30:32 +0200 | [diff] [blame] | 866 | cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0); |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 867 | if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 868 | done = 1; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 869 | if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) { |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 870 | /* Report network errors only if we got no other data. Otherwise |
| 871 | * we'll let the upper layers decide whether the response is OK |
| 872 | * or not. It is very common that an RST sent by the server is |
| 873 | * reported as an error just after the last data chunk. |
| 874 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 875 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | c1a0796 | 2010-03-16 20:55:43 +0100 | [diff] [blame] | 876 | goto out_wakeup; |
| 877 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 878 | } |
| 879 | |
Willy Tarreau | 0f0393f | 2019-09-24 10:43:03 +0200 | [diff] [blame] | 880 | /* the rest of the code below expects the connection to be ready! */ |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 881 | if (conn->flags & CO_FL_WAIT_XPRT && !done) |
Willy Tarreau | 0f0393f | 2019-09-24 10:43:03 +0200 | [diff] [blame] | 882 | goto wait_more_data; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 883 | |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 884 | /* Intermediate or complete response received. |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 885 | * Terminate string in b_head(&check->bi) buffer. |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 886 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 887 | if (b_data(&check->bi) < b_size(&check->bi)) |
| 888 | b_head(&check->bi)[b_data(&check->bi)] = '\0'; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 889 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 890 | b_head(&check->bi)[b_data(&check->bi) - 1] = '\0'; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 891 | done = 1; /* buffer full, don't wait for more data */ |
| 892 | } |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 893 | |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 894 | /* Run the checks... */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 895 | switch (check->type) { |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 896 | case PR_O2_HTTP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 897 | if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r")) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 898 | goto wait_more_data; |
| 899 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 900 | /* Check if the server speaks HTTP 1.X */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 901 | if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) || |
| 902 | (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 || |
| 903 | (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) || |
| 904 | !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) || |
| 905 | !isdigit((unsigned char) *(b_head(&check->bi) + 11))) { |
| 906 | cut_crlf(b_head(&check->bi)); |
| 907 | set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi)); |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 908 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 909 | goto out_wakeup; |
| 910 | } |
| 911 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 912 | check->code = str2uic(b_head(&check->bi) + 9); |
| 913 | desc = ltrim(b_head(&check->bi) + 12, ' '); |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 914 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 915 | if ((s->proxy->options & PR_O_DISABLE404) && |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 916 | (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) { |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 917 | /* 404 may be accepted as "stopping" only if the server was up */ |
| 918 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 919 | set_server_check_status(check, HCHK_STATUS_L7OKCD, desc); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 920 | } |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 921 | else if (s->proxy->options2 & PR_O2_EXP_TYPE) { |
| 922 | /* Run content verification check... We know we have at least 13 chars */ |
| 923 | if (!httpchk_expect(s, done)) |
| 924 | goto wait_more_data; |
| 925 | } |
| 926 | /* check the reply : HTTP/1.X 2xx and 3xx are OK */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 927 | else if (*(b_head(&check->bi) + 9) == '2' || *(b_head(&check->bi) + 9) == '3') { |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 928 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 929 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 930 | } |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 931 | else { |
| 932 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 933 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 934 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 935 | break; |
| 936 | |
| 937 | case PR_O2_SSL3_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 938 | if (!done && b_data(&check->bi) < 5) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 939 | goto wait_more_data; |
| 940 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 941 | /* Check for SSLv3 alert or handshake */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 942 | if ((b_data(&check->bi) >= 5) && (*b_head(&check->bi) == 0x15 || *b_head(&check->bi) == 0x16)) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 943 | set_server_check_status(check, HCHK_STATUS_L6OK, NULL); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 944 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 945 | set_server_check_status(check, HCHK_STATUS_L6RSP, NULL); |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 946 | break; |
| 947 | |
| 948 | case PR_O2_SMTP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 949 | if (!done && b_data(&check->bi) < strlen("000\r")) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 950 | goto wait_more_data; |
| 951 | |
Willy Tarreau | 5488a62 | 2019-08-06 16:26:31 +0200 | [diff] [blame] | 952 | /* do not reset when closing, servers don't like this */ |
| 953 | if (conn_ctrl_ready(cs->conn)) |
| 954 | fdtab[cs->conn->handle.fd].linger_risk = 0; |
| 955 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 956 | /* Check if the server speaks SMTP */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 957 | if ((b_data(&check->bi) < strlen("000\r")) || |
| 958 | (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') || |
| 959 | !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) || |
| 960 | !isdigit((unsigned char) *(b_head(&check->bi) + 2))) { |
| 961 | cut_crlf(b_head(&check->bi)); |
| 962 | set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi)); |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 963 | goto out_wakeup; |
| 964 | } |
| 965 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 966 | check->code = str2uic(b_head(&check->bi)); |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 967 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 968 | desc = ltrim(b_head(&check->bi) + 3, ' '); |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 969 | cut_crlf(desc); |
| 970 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 971 | /* Check for SMTP code 2xx (should be 250) */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 972 | if (*b_head(&check->bi) == '2') |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 973 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 974 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 975 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 976 | break; |
| 977 | |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 978 | case PR_O2_LB_AGENT_CHK: { |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 979 | int status = HCHK_STATUS_CHECKED; |
| 980 | const char *hs = NULL; /* health status */ |
| 981 | const char *as = NULL; /* admin status */ |
| 982 | const char *ps = NULL; /* performance status */ |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 983 | const char *cs = NULL; /* maxconn */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 984 | const char *err = NULL; /* first error to report */ |
| 985 | const char *wrn = NULL; /* first warning to report */ |
| 986 | char *cmd, *p; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 987 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 988 | /* We're getting an agent check response. The agent could |
| 989 | * have been disabled in the mean time with a long check |
| 990 | * still pending. It is important that we ignore the whole |
| 991 | * response. |
| 992 | */ |
| 993 | if (!(check->server->agent.state & CHK_ST_ENABLED)) |
| 994 | break; |
| 995 | |
| 996 | /* The agent supports strings made of a single line ended by the |
| 997 | * first CR ('\r') or LF ('\n'). This line is composed of words |
| 998 | * delimited by spaces (' '), tabs ('\t'), or commas (','). The |
| 999 | * line may optionally contained a description of a state change |
| 1000 | * after a sharp ('#'), which is only considered if a health state |
| 1001 | * is announced. |
| 1002 | * |
| 1003 | * Words may be composed of : |
| 1004 | * - a numeric weight suffixed by the percent character ('%'). |
| 1005 | * - a health status among "up", "down", "stopped", and "fail". |
| 1006 | * - an admin status among "ready", "drain", "maint". |
| 1007 | * |
| 1008 | * These words may appear in any order. If multiple words of the |
| 1009 | * same category appear, the last one wins. |
| 1010 | */ |
| 1011 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1012 | p = b_head(&check->bi); |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 1013 | while (*p && *p != '\n' && *p != '\r') |
| 1014 | p++; |
| 1015 | |
| 1016 | if (!*p) { |
| 1017 | if (!done) |
| 1018 | goto wait_more_data; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1019 | |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 1020 | /* at least inform the admin that the agent is mis-behaving */ |
| 1021 | set_server_check_status(check, check->status, "Ignoring incomplete line from agent"); |
| 1022 | break; |
| 1023 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1024 | |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 1025 | *p = 0; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1026 | cmd = b_head(&check->bi); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1027 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1028 | while (*cmd) { |
| 1029 | /* look for next word */ |
| 1030 | if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') { |
| 1031 | cmd++; |
| 1032 | continue; |
| 1033 | } |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 1034 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1035 | if (*cmd == '#') { |
| 1036 | /* this is the beginning of a health status description, |
| 1037 | * skip the sharp and blanks. |
| 1038 | */ |
| 1039 | cmd++; |
| 1040 | while (*cmd == '\t' || *cmd == ' ') |
| 1041 | cmd++; |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 1042 | break; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1043 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1044 | |
| 1045 | /* find the end of the word so that we have a null-terminated |
| 1046 | * word between <cmd> and <p>. |
| 1047 | */ |
| 1048 | p = cmd + 1; |
| 1049 | while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',') |
| 1050 | p++; |
| 1051 | if (*p) |
| 1052 | *p++ = 0; |
| 1053 | |
| 1054 | /* first, health statuses */ |
| 1055 | if (strcasecmp(cmd, "up") == 0) { |
| 1056 | check->health = check->rise + check->fall - 1; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1057 | status = HCHK_STATUS_L7OKD; |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1058 | hs = cmd; |
| 1059 | } |
| 1060 | else if (strcasecmp(cmd, "down") == 0) { |
| 1061 | check->health = 0; |
| 1062 | status = HCHK_STATUS_L7STS; |
| 1063 | hs = cmd; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1064 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1065 | else if (strcasecmp(cmd, "stopped") == 0) { |
| 1066 | check->health = 0; |
| 1067 | status = HCHK_STATUS_L7STS; |
| 1068 | hs = cmd; |
| 1069 | } |
| 1070 | else if (strcasecmp(cmd, "fail") == 0) { |
| 1071 | check->health = 0; |
| 1072 | status = HCHK_STATUS_L7STS; |
| 1073 | hs = cmd; |
| 1074 | } |
| 1075 | /* admin statuses */ |
| 1076 | else if (strcasecmp(cmd, "ready") == 0) { |
| 1077 | as = cmd; |
| 1078 | } |
| 1079 | else if (strcasecmp(cmd, "drain") == 0) { |
| 1080 | as = cmd; |
| 1081 | } |
| 1082 | else if (strcasecmp(cmd, "maint") == 0) { |
| 1083 | as = cmd; |
| 1084 | } |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1085 | /* try to parse a weight here and keep the last one */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1086 | else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) { |
| 1087 | ps = cmd; |
| 1088 | } |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1089 | /* try to parse a maxconn here */ |
| 1090 | else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) { |
| 1091 | cs = cmd; |
| 1092 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1093 | else { |
| 1094 | /* keep a copy of the first error */ |
| 1095 | if (!err) |
| 1096 | err = cmd; |
| 1097 | } |
| 1098 | /* skip to next word */ |
| 1099 | cmd = p; |
| 1100 | } |
| 1101 | /* here, cmd points either to \0 or to the beginning of a |
| 1102 | * description. Skip possible leading spaces. |
| 1103 | */ |
| 1104 | while (*cmd == ' ' || *cmd == '\n') |
| 1105 | cmd++; |
| 1106 | |
| 1107 | /* First, update the admin status so that we avoid sending other |
| 1108 | * possibly useless warnings and can also update the health if |
| 1109 | * present after going back up. |
| 1110 | */ |
| 1111 | if (as) { |
| 1112 | if (strcasecmp(as, "drain") == 0) |
| 1113 | srv_adm_set_drain(check->server); |
| 1114 | else if (strcasecmp(as, "maint") == 0) |
| 1115 | srv_adm_set_maint(check->server); |
| 1116 | else |
| 1117 | srv_adm_set_ready(check->server); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1118 | } |
| 1119 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1120 | /* now change weights */ |
| 1121 | if (ps) { |
| 1122 | const char *msg; |
| 1123 | |
| 1124 | msg = server_parse_weight_change_request(s, ps); |
| 1125 | if (!wrn || !*wrn) |
| 1126 | wrn = msg; |
| 1127 | } |
| 1128 | |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1129 | if (cs) { |
| 1130 | const char *msg; |
| 1131 | |
| 1132 | cs += strlen("maxconn:"); |
| 1133 | |
| 1134 | msg = server_parse_maxconn_change_request(s, cs); |
| 1135 | if (!wrn || !*wrn) |
| 1136 | wrn = msg; |
| 1137 | } |
| 1138 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1139 | /* and finally health status */ |
| 1140 | if (hs) { |
| 1141 | /* We'll report some of the warnings and errors we have |
| 1142 | * here. Down reports are critical, we leave them untouched. |
| 1143 | * Lack of report, or report of 'UP' leaves the room for |
| 1144 | * ERR first, then WARN. |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1145 | */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1146 | const char *msg = cmd; |
Willy Tarreau | 83061a8 | 2018-07-13 11:56:34 +0200 | [diff] [blame] | 1147 | struct buffer *t; |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1148 | |
| 1149 | if (!*msg || status == HCHK_STATUS_L7OKD) { |
| 1150 | if (err && *err) |
| 1151 | msg = err; |
| 1152 | else if (wrn && *wrn) |
| 1153 | msg = wrn; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1154 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1155 | |
| 1156 | t = get_trash_chunk(); |
| 1157 | chunk_printf(t, "via agent : %s%s%s%s", |
| 1158 | hs, *msg ? " (" : "", |
| 1159 | msg, *msg ? ")" : ""); |
| 1160 | |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1161 | set_server_check_status(check, status, t->area); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1162 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1163 | else if (err && *err) { |
| 1164 | /* No status change but we'd like to report something odd. |
| 1165 | * Just report the current state and copy the message. |
| 1166 | */ |
| 1167 | chunk_printf(&trash, "agent reports an error : %s", err); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1168 | set_server_check_status(check, status/*check->status*/, |
| 1169 | trash.area); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1170 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1171 | } |
| 1172 | else if (wrn && *wrn) { |
| 1173 | /* No status change but we'd like to report something odd. |
| 1174 | * Just report the current state and copy the message. |
| 1175 | */ |
| 1176 | chunk_printf(&trash, "agent warns : %s", wrn); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1177 | set_server_check_status(check, status/*check->status*/, |
| 1178 | trash.area); |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1179 | } |
| 1180 | else |
| 1181 | set_server_check_status(check, status, NULL); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1182 | break; |
| 1183 | } |
| 1184 | |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1185 | case PR_O2_PGSQL_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1186 | if (!done && b_data(&check->bi) < 9) |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1187 | goto wait_more_data; |
| 1188 | |
Willy Tarreau | 5488a62 | 2019-08-06 16:26:31 +0200 | [diff] [blame] | 1189 | /* do not reset when closing, servers don't like this */ |
| 1190 | if (conn_ctrl_ready(cs->conn)) |
| 1191 | fdtab[cs->conn->handle.fd].linger_risk = 0; |
| 1192 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1193 | if (b_head(&check->bi)[0] == 'R') { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1194 | set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok"); |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1195 | } |
| 1196 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1197 | if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0)) |
| 1198 | desc = &b_head(&check->bi)[6]; |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1199 | else |
| 1200 | desc = "PostgreSQL unknown error"; |
| 1201 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1202 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1203 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1204 | break; |
| 1205 | |
| 1206 | case PR_O2_REDIS_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1207 | if (!done && b_data(&check->bi) < 7) |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1208 | goto wait_more_data; |
| 1209 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1210 | if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1211 | set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok"); |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1212 | } |
| 1213 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1214 | set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi)); |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1215 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1216 | break; |
| 1217 | |
| 1218 | case PR_O2_MYSQL_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1219 | if (!done && b_data(&check->bi) < 5) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1220 | goto wait_more_data; |
| 1221 | |
Willy Tarreau | 5488a62 | 2019-08-06 16:26:31 +0200 | [diff] [blame] | 1222 | /* do not reset when closing, servers don't like this */ |
| 1223 | if (conn_ctrl_ready(cs->conn)) |
| 1224 | fdtab[cs->conn->handle.fd].linger_risk = 0; |
| 1225 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1226 | if (s->proxy->check_len == 0) { // old mode |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1227 | if (*(b_head(&check->bi) + 4) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1228 | /* We set the MySQL Version in description for information purpose |
| 1229 | * FIXME : it can be cool to use MySQL Version for other purpose, |
| 1230 | * like mark as down old MySQL server. |
| 1231 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1232 | if (b_data(&check->bi) > 51) { |
| 1233 | desc = ltrim(b_head(&check->bi) + 5, ' '); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1234 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1235 | } |
| 1236 | else { |
| 1237 | if (!done) |
| 1238 | goto wait_more_data; |
Emeric Brun | 9f0b458 | 2017-10-23 14:39:51 +0200 | [diff] [blame] | 1239 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1240 | /* it seems we have a OK packet but without a valid length, |
| 1241 | * it must be a protocol error |
| 1242 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1243 | set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi)); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1244 | } |
Hervé COMMOWICK | 698ae00 | 2010-01-12 09:25:13 +0100 | [diff] [blame] | 1245 | } |
| 1246 | else { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1247 | /* An error message is attached in the Error packet */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1248 | desc = ltrim(b_head(&check->bi) + 7, ' '); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1249 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1250 | } |
| 1251 | } else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1252 | unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) + |
| 1253 | (((unsigned int) *(b_head(&check->bi) + 1)) << 8) + |
| 1254 | (((unsigned int) *(b_head(&check->bi) + 2)) << 16); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1255 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1256 | if (b_data(&check->bi) == first_packet_len + 4) { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1257 | /* MySQL Error packet always begin with field_count = 0xff */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1258 | if (*(b_head(&check->bi) + 4) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1259 | /* We have only one MySQL packet and it is a Handshake Initialization packet |
| 1260 | * but we need to have a second packet to know if it is alright |
| 1261 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1262 | if (!done && b_data(&check->bi) < first_packet_len + 5) |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1263 | goto wait_more_data; |
| 1264 | } |
| 1265 | else { |
| 1266 | /* We have only one packet and it is an Error packet, |
| 1267 | * an error message is attached, so we can display it |
| 1268 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1269 | desc = &b_head(&check->bi)[7]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1270 | //ha_warning("onlyoneERR: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1271 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1272 | } |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1273 | } else if (b_data(&check->bi) > first_packet_len + 4) { |
| 1274 | unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) + |
| 1275 | (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) + |
| 1276 | (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1277 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1278 | if (b_data(&check->bi) == first_packet_len + 4 + second_packet_len + 4 ) { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1279 | /* We have 2 packets and that's good */ |
| 1280 | /* Check if the second packet is a MySQL Error packet or not */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1281 | if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1282 | /* No error packet */ |
| 1283 | /* We set the MySQL Version in description for information purpose */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1284 | desc = &b_head(&check->bi)[5]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1285 | //ha_warning("2packetOK: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1286 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1287 | } |
| 1288 | else { |
| 1289 | /* An error message is attached in the Error packet |
| 1290 | * so we can display it ! :) |
| 1291 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1292 | desc = &b_head(&check->bi)[first_packet_len+11]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1293 | //ha_warning("2packetERR: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1294 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1295 | } |
| 1296 | } |
| 1297 | } |
| 1298 | else { |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1299 | if (!done) |
| 1300 | goto wait_more_data; |
Emeric Brun | 9f0b458 | 2017-10-23 14:39:51 +0200 | [diff] [blame] | 1301 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1302 | /* it seems we have a Handshake Initialization packet but without a valid length, |
Hervé COMMOWICK | 698ae00 | 2010-01-12 09:25:13 +0100 | [diff] [blame] | 1303 | * it must be a protocol error |
| 1304 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1305 | desc = &b_head(&check->bi)[5]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1306 | //ha_warning("protoerr: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1307 | set_server_check_status(check, HCHK_STATUS_L7RSP, desc); |
Hervé COMMOWICK | 698ae00 | 2010-01-12 09:25:13 +0100 | [diff] [blame] | 1308 | } |
| 1309 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1310 | break; |
| 1311 | |
| 1312 | case PR_O2_LDAP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1313 | if (!done && b_data(&check->bi) < 14) |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1314 | goto wait_more_data; |
| 1315 | |
| 1316 | /* Check if the server speaks LDAP (ASN.1/BER) |
| 1317 | * http://en.wikipedia.org/wiki/Basic_Encoding_Rules |
| 1318 | * http://tools.ietf.org/html/rfc4511 |
| 1319 | */ |
| 1320 | |
| 1321 | /* http://tools.ietf.org/html/rfc4511#section-4.1.1 |
| 1322 | * LDAPMessage: 0x30: SEQUENCE |
| 1323 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1324 | if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1325 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1326 | } |
| 1327 | else { |
| 1328 | /* size of LDAPMessage */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1329 | msglen = (*(b_head(&check->bi) + 1) & 0x80) ? (*(b_head(&check->bi) + 1) & 0x7f) : 0; |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1330 | |
| 1331 | /* http://tools.ietf.org/html/rfc4511#section-4.2.2 |
| 1332 | * messageID: 0x02 0x01 0x01: INTEGER 1 |
| 1333 | * protocolOp: 0x61: bindResponse |
| 1334 | */ |
| 1335 | if ((msglen > 2) || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1336 | (memcmp(b_head(&check->bi) + 2 + msglen, "\x02\x01\x01\x61", 4) != 0)) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1337 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1338 | goto out_wakeup; |
| 1339 | } |
| 1340 | |
| 1341 | /* size of bindResponse */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1342 | msglen += (*(b_head(&check->bi) + msglen + 6) & 0x80) ? (*(b_head(&check->bi) + msglen + 6) & 0x7f) : 0; |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1343 | |
| 1344 | /* http://tools.ietf.org/html/rfc4511#section-4.1.9 |
| 1345 | * ldapResult: 0x0a 0x01: ENUMERATION |
| 1346 | */ |
| 1347 | if ((msglen > 4) || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1348 | (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1349 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1350 | goto out_wakeup; |
| 1351 | } |
| 1352 | |
| 1353 | /* http://tools.ietf.org/html/rfc4511#section-4.1.9 |
| 1354 | * resultCode |
| 1355 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1356 | check->code = *(b_head(&check->bi) + msglen + 9); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1357 | if (check->code) { |
| 1358 | set_server_check_status(check, HCHK_STATUS_L7STS, "See RFC: http://tools.ietf.org/html/rfc4511#section-4.1.9"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1359 | } else { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1360 | set_server_check_status(check, HCHK_STATUS_L7OKD, "Success"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1361 | } |
| 1362 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1363 | break; |
| 1364 | |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1365 | case PR_O2_SPOP_CHK: { |
| 1366 | unsigned int framesz; |
| 1367 | char err[HCHK_DESC_LEN]; |
| 1368 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1369 | if (!done && b_data(&check->bi) < 4) |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1370 | goto wait_more_data; |
| 1371 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1372 | memcpy(&framesz, b_head(&check->bi), 4); |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1373 | framesz = ntohl(framesz); |
| 1374 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1375 | if (!done && b_data(&check->bi) < (4+framesz)) |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1376 | goto wait_more_data; |
| 1377 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1378 | if (!spoe_handle_healthcheck_response(b_head(&check->bi)+4, framesz, err, HCHK_DESC_LEN-1)) |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1379 | set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok"); |
| 1380 | else |
| 1381 | set_server_check_status(check, HCHK_STATUS_L7STS, err); |
| 1382 | break; |
| 1383 | } |
| 1384 | |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1385 | default: |
Willy Tarreau | 4c1a2b3 | 2019-09-05 18:43:22 +0200 | [diff] [blame] | 1386 | /* good connection is enough for pure TCP check */ |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 1387 | if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) { |
Christopher Faulet | f61f33a | 2020-03-27 18:55:49 +0100 | [diff] [blame] | 1388 | if (check->use_ssl == 1) |
Willy Tarreau | 4c1a2b3 | 2019-09-05 18:43:22 +0200 | [diff] [blame] | 1389 | set_server_check_status(check, HCHK_STATUS_L6OK, NULL); |
| 1390 | else |
| 1391 | set_server_check_status(check, HCHK_STATUS_L4OK, NULL); |
| 1392 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1393 | break; |
| 1394 | } /* switch */ |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 1395 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 1396 | out_wakeup: |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1397 | /* collect possible new errors */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 1398 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 1399 | chk_report_conn_err(check, 0, 0); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1400 | |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 1401 | /* Reset the check buffer... */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1402 | *b_head(&check->bi) = '\0'; |
| 1403 | b_reset(&check->bi); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 1404 | |
Steven Davidovitz | 544d481 | 2017-03-08 11:06:20 -0800 | [diff] [blame] | 1405 | /* Close the connection... We still attempt to nicely close if, |
| 1406 | * for instance, SSL needs to send a "close notify." Later, we perform |
| 1407 | * a hard close and reset the connection if some data are pending, |
| 1408 | * otherwise we end up with many TIME_WAITs and eat all the source port |
| 1409 | * range quickly. To avoid sending RSTs all the time, we first try to |
| 1410 | * drain pending data. |
Willy Tarreau | fd29cc5 | 2012-11-23 09:18:20 +0100 | [diff] [blame] | 1411 | */ |
Olivier Houchard | 6c7e96a | 2019-07-02 16:35:18 +0200 | [diff] [blame] | 1412 | /* Call cs_shutr() first, to add the CO_FL_SOCK_RD_SH flag on the |
| 1413 | * connection, to make sure cs_shutw() will not lead to a shutdown() |
| 1414 | * that would provoke TIME_WAITs. |
| 1415 | */ |
| 1416 | cs_shutr(cs, CS_SHR_DRAIN); |
Willy Tarreau | ecdb3fe | 2017-10-05 15:25:48 +0200 | [diff] [blame] | 1417 | cs_shutw(cs, CS_SHW_NORMAL); |
Willy Tarreau | 2b57cb8 | 2013-06-10 19:56:38 +0200 | [diff] [blame] | 1418 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1419 | /* OK, let's not stay here forever */ |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 1420 | if (check->result == CHK_RES_FAILED) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1421 | conn->flags |= CO_FL_ERROR; |
| 1422 | |
Willy Tarreau | fdccded | 2008-08-29 18:19:04 +0200 | [diff] [blame] | 1423 | task_wakeup(t, TASK_WOKEN_IO); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 1424 | out: |
Willy Tarreau | 3267d36 | 2012-08-17 23:53:56 +0200 | [diff] [blame] | 1425 | return; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1426 | |
| 1427 | wait_more_data: |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 1428 | cs->conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 1429 | goto out; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1430 | } |
| 1431 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1432 | /* |
| 1433 | * This function is used only for server health-checks. It handles connection |
| 1434 | * status updates including errors. If necessary, it wakes the check task up. |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1435 | * It returns 0 on normal cases, <0 if at least one close() has happened on the |
| 1436 | * connection (eg: reconnect). |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1437 | */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1438 | static int wake_srv_chk(struct conn_stream *cs) |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1439 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1440 | struct connection *conn = cs->conn; |
| 1441 | struct check *check = cs->data; |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 1442 | struct email_alertq *q = container_of(check, typeof(*q), check); |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1443 | int ret = 0; |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1444 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 1445 | if (check->server) |
| 1446 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
| 1447 | else |
| 1448 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Emeric Brun | f6ba17d | 2017-11-02 14:35:27 +0100 | [diff] [blame] | 1449 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 1450 | /* we may have to make progress on the TCP checks */ |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1451 | if (check->type == PR_O2_TCPCHK_CHK) { |
| 1452 | ret = tcpcheck_main(check); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1453 | cs = check->cs; |
Willy Tarreau | 543abd4 | 2018-09-20 11:25:12 +0200 | [diff] [blame] | 1454 | conn = cs->conn; |
Willy Tarreau | c594039 | 2019-09-05 17:38:40 +0200 | [diff] [blame] | 1455 | } else { |
| 1456 | if (!(check->wait_list.events & SUB_RETRY_SEND)) |
| 1457 | __event_srv_chk_w(cs); |
| 1458 | if (!(check->wait_list.events & SUB_RETRY_RECV)) |
| 1459 | __event_srv_chk_r(cs); |
| 1460 | } |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 1461 | |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 1462 | if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) { |
Willy Tarreau | 02b0f58 | 2013-12-03 15:42:33 +0100 | [diff] [blame] | 1463 | /* We may get error reports bypassing the I/O handlers, typically |
| 1464 | * the case when sending a pure TCP check which fails, then the I/O |
| 1465 | * handlers above are not called. This is completely handled by the |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1466 | * main processing task so let's simply wake it up. If we get here, |
| 1467 | * we expect errno to still be valid. |
| 1468 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 1469 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1470 | task_wakeup(check->task, TASK_WOKEN_IO); |
| 1471 | } |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 1472 | else if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) { |
Willy Tarreau | 3be293f | 2014-02-05 18:31:24 +0100 | [diff] [blame] | 1473 | /* we may get here if only a connection probe was required : we |
| 1474 | * don't have any data to send nor anything expected in response, |
| 1475 | * so the completion of the connection establishment is enough. |
| 1476 | */ |
| 1477 | task_wakeup(check->task, TASK_WOKEN_IO); |
| 1478 | } |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1479 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 1480 | if (check->result != CHK_RES_UNKNOWN) { |
Christopher Faulet | 774c486 | 2019-01-21 14:15:50 +0100 | [diff] [blame] | 1481 | /* Check complete or aborted. If connection not yet closed do it |
| 1482 | * now and wake the check task up to be sure the result is |
| 1483 | * handled ASAP. */ |
Willy Tarreau | d85c485 | 2015-03-13 00:40:28 +0100 | [diff] [blame] | 1484 | conn_sock_drain(conn); |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 1485 | cs_close(cs); |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1486 | ret = -1; |
Olivier Houchard | f494957 | 2019-07-02 17:42:22 +0200 | [diff] [blame] | 1487 | /* We may have been scheduled to run, and the |
| 1488 | * I/O handler expects to have a cs, so remove |
| 1489 | * the tasklet |
| 1490 | */ |
| 1491 | tasklet_remove_from_tasklet_list(check->wait_list.tasklet); |
Christopher Faulet | 774c486 | 2019-01-21 14:15:50 +0100 | [diff] [blame] | 1492 | task_wakeup(check->task, TASK_WOKEN_IO); |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1493 | } |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1494 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 1495 | if (check->server) |
| 1496 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
| 1497 | else |
| 1498 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Emeric Brun | f6ba17d | 2017-11-02 14:35:27 +0100 | [diff] [blame] | 1499 | |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1500 | /* if a connection got replaced, we must absolutely prevent the connection |
| 1501 | * handler from touching its fd, and perform the FD polling updates ourselves |
| 1502 | */ |
| 1503 | if (ret < 0) |
| 1504 | conn_cond_update_polling(conn); |
| 1505 | |
| 1506 | return ret; |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1507 | } |
| 1508 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1509 | struct data_cb check_conn_cb = { |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1510 | .wake = wake_srv_chk, |
Willy Tarreau | 8e0bb0a | 2016-11-24 16:58:12 +0100 | [diff] [blame] | 1511 | .name = "CHCK", |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1512 | }; |
| 1513 | |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1514 | /* |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1515 | * updates the server's weight during a warmup stage. Once the final weight is |
| 1516 | * reached, the task automatically stops. Note that any server status change |
| 1517 | * must have updated s->last_change accordingly. |
| 1518 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 1519 | static struct task *server_warmup(struct task *t, void *context, unsigned short state) |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1520 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 1521 | struct server *s = context; |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1522 | |
| 1523 | /* by default, plan on stopping the task */ |
| 1524 | t->expire = TICK_ETERNITY; |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 1525 | if ((s->next_admin & SRV_ADMF_MAINT) || |
| 1526 | (s->next_state != SRV_ST_STARTING)) |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1527 | return t; |
| 1528 | |
Willy Tarreau | 4fc49a9 | 2019-05-05 06:54:22 +0200 | [diff] [blame] | 1529 | HA_SPIN_LOCK(SERVER_LOCK, &s->lock); |
| 1530 | |
Willy Tarreau | 892337c | 2014-05-13 23:41:20 +0200 | [diff] [blame] | 1531 | /* recalculate the weights and update the state */ |
Willy Tarreau | 3ff577e | 2018-08-02 11:48:52 +0200 | [diff] [blame] | 1532 | server_recalc_eweight(s, 1); |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1533 | |
| 1534 | /* probably that we can refill this server with a bit more connections */ |
Willy Tarreau | 4aac7db | 2014-05-16 11:48:10 +0200 | [diff] [blame] | 1535 | pendconn_grab_from_px(s); |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1536 | |
Willy Tarreau | 4fc49a9 | 2019-05-05 06:54:22 +0200 | [diff] [blame] | 1537 | HA_SPIN_UNLOCK(SERVER_LOCK, &s->lock); |
| 1538 | |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1539 | /* get back there in 1 second or 1/20th of the slowstart interval, |
| 1540 | * whichever is greater, resulting in small 5% steps. |
| 1541 | */ |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 1542 | if (s->next_state == SRV_ST_STARTING) |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1543 | t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20))); |
| 1544 | return t; |
| 1545 | } |
| 1546 | |
Willy Tarreau | 894c642 | 2017-10-04 15:58:52 +0200 | [diff] [blame] | 1547 | /* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if |
| 1548 | * none was found. |
| 1549 | */ |
| 1550 | static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list) |
| 1551 | { |
| 1552 | struct tcpcheck_rule *r; |
| 1553 | |
| 1554 | list_for_each_entry(r, list, list) { |
| 1555 | if (r->action != TCPCHK_ACT_COMMENT) |
| 1556 | return r; |
| 1557 | } |
| 1558 | return NULL; |
| 1559 | } |
| 1560 | |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1561 | /* |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1562 | * establish a server health-check that makes use of a connection. |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1563 | * |
| 1564 | * It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1565 | * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called |
| 1566 | * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called |
| 1567 | * - SF_ERR_SRVTO if there are no more servers |
| 1568 | * - SF_ERR_SRVCL if the connection was refused by the server |
| 1569 | * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn) |
| 1570 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 1571 | * - SF_ERR_INTERNAL for any other purely internal errors |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1572 | * - SF_ERR_CHK_PORT if no port could be found to run a health check on an AF_INET* socket |
Tim Düsterhus | 4896c44 | 2016-11-29 02:15:19 +0100 | [diff] [blame] | 1573 | * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted. |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1574 | * Note that we try to prevent the network stack from sending the ACK during the |
| 1575 | * connect() when a pure TCP check is used (without PROXY protocol). |
| 1576 | */ |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1577 | static int connect_conn_chk(struct task *t) |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1578 | { |
| 1579 | struct check *check = t->context; |
| 1580 | struct server *s = check->server; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1581 | struct conn_stream *cs = check->cs; |
| 1582 | struct connection *conn = cs_conn(cs); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1583 | struct protocol *proto; |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1584 | struct tcpcheck_rule *tcp_rule = NULL; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1585 | int ret; |
Olivier Houchard | fdcb007 | 2019-05-06 18:32:29 +0200 | [diff] [blame] | 1586 | int connflags = 0; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1587 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 1588 | /* we cannot have a connection here */ |
| 1589 | if (conn) |
| 1590 | return SF_ERR_INTERNAL; |
| 1591 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1592 | /* tcpcheck send/expect initialisation */ |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1593 | if (check->type == PR_O2_TCPCHK_CHK) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1594 | check->current_step = NULL; |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1595 | tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules); |
| 1596 | } |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1597 | |
| 1598 | /* prepare the check buffer. |
| 1599 | * This should not be used if check is the secondary agent check |
| 1600 | * of a server as s->proxy->check_req will relate to the |
| 1601 | * configuration of the primary check. Similarly, tcp-check uses |
| 1602 | * its own strings. |
| 1603 | */ |
| 1604 | if (check->type && check->type != PR_O2_TCPCHK_CHK && !(check->state & CHK_ST_AGENT)) { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1605 | b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1606 | |
| 1607 | /* we want to check if this host replies to HTTP or SSLv3 requests |
| 1608 | * so we'll send the request, and won't wake the checker up now. |
| 1609 | */ |
| 1610 | if ((check->type) == PR_O2_SSL3_CHK) { |
| 1611 | /* SSL requires that we put Unix time in the request */ |
| 1612 | int gmt_time = htonl(date.tv_sec); |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1613 | memcpy(b_head(&check->bo) + 11, &gmt_time, 4); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1614 | } |
| 1615 | else if ((check->type) == PR_O2_HTTP_CHK) { |
Cyril Bonté | 32602d2 | 2015-01-30 00:07:07 +0100 | [diff] [blame] | 1616 | /* prevent HTTP keep-alive when "http-check expect" is used */ |
| 1617 | if (s->proxy->options2 & PR_O2_EXP_TYPE) |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1618 | b_putist(&check->bo, ist("Connection: close\r\n")); |
Christopher Faulet | 8acb128 | 2020-04-09 08:44:06 +0200 | [diff] [blame] | 1619 | |
| 1620 | /* If there is a body, add its content-length */ |
| 1621 | if (s->proxy->check_body_len) |
| 1622 | chunk_appendf(&check->bo, "Content-Length: %s\r\n", ultoa(s->proxy->check_body_len)); |
| 1623 | |
| 1624 | /* Add configured headers */ |
| 1625 | if (s->proxy->check_hdrs) |
| 1626 | b_putblk(&check->bo, s->proxy->check_hdrs, s->proxy->check_hdrs_len); |
| 1627 | |
| 1628 | /* Add send-state header */ |
| 1629 | if (s->proxy->options2 & PR_O2_CHK_SNDST) |
| 1630 | b_putblk(&check->bo, trash.area, |
| 1631 | httpchk_build_status_header(s, trash.area, trash.size)); |
| 1632 | |
| 1633 | /* end-of-header */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1634 | b_putist(&check->bo, ist("\r\n")); |
Christopher Faulet | 8acb128 | 2020-04-09 08:44:06 +0200 | [diff] [blame] | 1635 | |
| 1636 | /* Add the body */ |
| 1637 | if (s->proxy->check_body) |
| 1638 | b_putblk(&check->bo, s->proxy->check_body, s->proxy->check_body_len); |
| 1639 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1640 | *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */ |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1641 | } |
| 1642 | } |
| 1643 | |
James Brown | 55f9ff1 | 2015-10-21 18:19:05 -0700 | [diff] [blame] | 1644 | if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1645 | b_putblk(&check->bo, check->send_string, check->send_string_len); |
James Brown | 55f9ff1 | 2015-10-21 18:19:05 -0700 | [diff] [blame] | 1646 | } |
| 1647 | |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1648 | /* for tcp-checks, the initial connection setup is handled separately as |
| 1649 | * it may be sent to a specific port and not to the server's. |
| 1650 | */ |
| 1651 | if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) { |
| 1652 | tcpcheck_main(check); |
| 1653 | return SF_ERR_UP; |
| 1654 | } |
| 1655 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1656 | /* prepare a new connection */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1657 | cs = check->cs = cs_new(NULL); |
| 1658 | if (!check->cs) |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 1659 | return SF_ERR_RESOURCE; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1660 | conn = cs->conn; |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 1661 | /* Maybe there were an older connection we were waiting on */ |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 1662 | check->wait_list.events = 0; |
Olivier Houchard | ff1e9f3 | 2019-09-20 17:18:35 +0200 | [diff] [blame] | 1663 | tasklet_set_tid(check->wait_list.tasklet, tid); |
| 1664 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1665 | |
Willy Tarreau | ca79f59 | 2019-07-17 19:04:47 +0200 | [diff] [blame] | 1666 | if (!sockaddr_alloc(&conn->dst)) |
| 1667 | return SF_ERR_RESOURCE; |
| 1668 | |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 1669 | if (is_addr(&check->addr)) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1670 | /* we'll connect to the check addr specified on the server */ |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 1671 | *conn->dst = check->addr; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1672 | } |
| 1673 | else { |
| 1674 | /* we'll connect to the addr on the server */ |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 1675 | *conn->dst = s->addr; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1676 | } |
| 1677 | |
Alexander Liu | 2a54bb7 | 2019-05-22 19:44:48 +0800 | [diff] [blame] | 1678 | if (s->check.via_socks4 && (s->flags & SRV_F_SOCKS4_PROXY)) { |
| 1679 | conn->send_proxy_ofs = 1; |
| 1680 | conn->flags |= CO_FL_SOCKS4; |
| 1681 | } |
| 1682 | |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 1683 | proto = protocol_by_family(conn->dst->ss_family); |
Olivier Houchard | 6377a00 | 2017-12-01 22:04:05 +0100 | [diff] [blame] | 1684 | conn->target = &s->obj_type; |
| 1685 | |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 1686 | if ((conn->dst->ss_family == AF_INET) || (conn->dst->ss_family == AF_INET6)) { |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1687 | int i = 0; |
| 1688 | |
| 1689 | i = srv_check_healthcheck_port(check); |
Olivier Houchard | 6377a00 | 2017-12-01 22:04:05 +0100 | [diff] [blame] | 1690 | if (i == 0) |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1691 | return SF_ERR_CHK_PORT; |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1692 | |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 1693 | set_host_port(conn->dst, i); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1694 | } |
| 1695 | |
Thierry FOURNIER | bb2ae64 | 2015-01-14 11:31:49 +0100 | [diff] [blame] | 1696 | /* no client address */ |
Thierry FOURNIER | bb2ae64 | 2015-01-14 11:31:49 +0100 | [diff] [blame] | 1697 | |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 1698 | conn_prepare(conn, proto, check->xprt); |
Olivier Houchard | f67be93 | 2019-01-29 15:47:43 +0100 | [diff] [blame] | 1699 | if (conn_install_mux(conn, &mux_pt_ops, cs, s->proxy, NULL) < 0) |
| 1700 | return SF_ERR_RESOURCE; |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 1701 | cs_attach(cs, check, &check_conn_cb); |
| 1702 | |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 1703 | /* only plain tcp-check supports quick ACK */ |
Olivier Houchard | fdcb007 | 2019-05-06 18:32:29 +0200 | [diff] [blame] | 1704 | if (check->type != 0) |
| 1705 | connflags |= CONNECT_HAS_DATA; |
| 1706 | if ((check->type == 0 || check->type == PR_O2_TCPCHK_CHK) && |
| 1707 | (!tcp_rule || tcp_rule->action != TCPCHK_ACT_EXPECT)) |
| 1708 | connflags |= CONNECT_DELACK_ALWAYS; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1709 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1710 | ret = SF_ERR_INTERNAL; |
Olivier Houchard | b68fda4 | 2017-08-04 18:39:01 +0200 | [diff] [blame] | 1711 | if (proto && proto->connect) |
Olivier Houchard | fdcb007 | 2019-05-06 18:32:29 +0200 | [diff] [blame] | 1712 | ret = proto->connect(conn, connflags); |
Willy Tarreau | 16257f6 | 2017-11-02 15:45:00 +0100 | [diff] [blame] | 1713 | |
Willy Tarreau | 16257f6 | 2017-11-02 15:45:00 +0100 | [diff] [blame] | 1714 | |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 1715 | #ifdef USE_OPENSSL |
Olivier Houchard | a48437b | 2019-01-29 16:37:52 +0100 | [diff] [blame] | 1716 | if (ret == SF_ERR_NONE) { |
| 1717 | if (s->check.sni) |
| 1718 | ssl_sock_set_servername(conn, s->check.sni); |
| 1719 | if (s->check.alpn_str) |
| 1720 | ssl_sock_set_alpn(conn, (unsigned char *)s->check.alpn_str, |
| 1721 | s->check.alpn_len); |
| 1722 | } |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 1723 | #endif |
Willy Tarreau | f494977 | 2017-05-06 08:45:28 +0200 | [diff] [blame] | 1724 | if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1725 | conn->send_proxy_ofs = 1; |
| 1726 | conn->flags |= CO_FL_SEND_PROXY; |
Olivier Houchard | 37d7897 | 2019-12-30 15:13:42 +0100 | [diff] [blame] | 1727 | } |
| 1728 | if (conn->flags & (CO_FL_SEND_PROXY | CO_FL_SOCKS4) && |
| 1729 | conn_ctrl_ready(conn)) { |
Olivier Houchard | fe50bfb | 2019-05-27 12:09:19 +0200 | [diff] [blame] | 1730 | if (xprt_add_hs(conn) < 0) |
| 1731 | ret = SF_ERR_RESOURCE; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1732 | } |
| 1733 | |
| 1734 | return ret; |
| 1735 | } |
| 1736 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1737 | static struct list pid_list = LIST_HEAD_INIT(pid_list); |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1738 | static struct pool_head *pool_head_pid_list; |
Willy Tarreau | 86abe44 | 2018-11-25 20:12:18 +0100 | [diff] [blame] | 1739 | __decl_spinlock(pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1740 | |
| 1741 | void block_sigchld(void) |
| 1742 | { |
| 1743 | sigset_t set; |
| 1744 | sigemptyset(&set); |
| 1745 | sigaddset(&set, SIGCHLD); |
William Lallemand | 6e1796e | 2018-06-07 11:23:40 +0200 | [diff] [blame] | 1746 | assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1747 | } |
| 1748 | |
| 1749 | void unblock_sigchld(void) |
| 1750 | { |
| 1751 | sigset_t set; |
| 1752 | sigemptyset(&set); |
Willy Tarreau | ebc9244 | 2016-06-21 17:29:46 +0200 | [diff] [blame] | 1753 | sigaddset(&set, SIGCHLD); |
William Lallemand | 6e1796e | 2018-06-07 11:23:40 +0200 | [diff] [blame] | 1754 | assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1755 | } |
| 1756 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1757 | static struct pid_list *pid_list_add(pid_t pid, struct task *t) |
| 1758 | { |
| 1759 | struct pid_list *elem; |
| 1760 | struct check *check = t->context; |
| 1761 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1762 | elem = pool_alloc(pool_head_pid_list); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1763 | if (!elem) |
| 1764 | return NULL; |
| 1765 | elem->pid = pid; |
| 1766 | elem->t = t; |
| 1767 | elem->exited = 0; |
| 1768 | check->curpid = elem; |
| 1769 | LIST_INIT(&elem->list); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1770 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1771 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1772 | LIST_ADD(&pid_list, &elem->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1773 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1774 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1775 | return elem; |
| 1776 | } |
| 1777 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1778 | static void pid_list_del(struct pid_list *elem) |
| 1779 | { |
| 1780 | struct check *check; |
| 1781 | |
| 1782 | if (!elem) |
| 1783 | return; |
| 1784 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1785 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1786 | LIST_DEL(&elem->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1787 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1788 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1789 | if (!elem->exited) |
| 1790 | kill(elem->pid, SIGTERM); |
| 1791 | |
| 1792 | check = elem->t->context; |
| 1793 | check->curpid = NULL; |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1794 | pool_free(pool_head_pid_list, elem); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1795 | } |
| 1796 | |
| 1797 | /* Called from inside SIGCHLD handler, SIGCHLD is blocked */ |
| 1798 | static void pid_list_expire(pid_t pid, int status) |
| 1799 | { |
| 1800 | struct pid_list *elem; |
| 1801 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1802 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1803 | list_for_each_entry(elem, &pid_list, list) { |
| 1804 | if (elem->pid == pid) { |
| 1805 | elem->t->expire = now_ms; |
| 1806 | elem->status = status; |
| 1807 | elem->exited = 1; |
Cyril Bonté | 9dbcfab | 2014-08-07 01:55:39 +0200 | [diff] [blame] | 1808 | task_wakeup(elem->t, TASK_WOKEN_IO); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1809 | break; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1810 | } |
| 1811 | } |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1812 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1813 | } |
| 1814 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1815 | static void sigchld_handler(struct sig_handler *sh) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1816 | { |
| 1817 | pid_t pid; |
| 1818 | int status; |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1819 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1820 | while ((pid = waitpid(0, &status, WNOHANG)) > 0) |
| 1821 | pid_list_expire(pid, status); |
| 1822 | } |
| 1823 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1824 | static int init_pid_list(void) |
| 1825 | { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1826 | if (pool_head_pid_list != NULL) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1827 | /* Nothing to do */ |
| 1828 | return 0; |
| 1829 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1830 | if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1831 | ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n", |
| 1832 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1833 | return 1; |
| 1834 | } |
| 1835 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1836 | pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED); |
| 1837 | if (pool_head_pid_list == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1838 | ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n", |
| 1839 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1840 | return 1; |
| 1841 | } |
| 1842 | |
| 1843 | return 0; |
| 1844 | } |
| 1845 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1846 | /* helper macro to set an environment variable and jump to a specific label on failure. */ |
| 1847 | #define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; } |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1848 | |
| 1849 | /* |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1850 | * helper function to allocate enough memory to store an environment variable. |
| 1851 | * It will also check that the environment variable is updatable, and silently |
| 1852 | * fail if not. |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1853 | */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1854 | static int extchk_setenv(struct check *check, int idx, const char *value) |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1855 | { |
| 1856 | int len, ret; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1857 | char *envname; |
| 1858 | int vmaxlen; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1859 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1860 | if (idx < 0 || idx >= EXTCHK_SIZE) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1861 | ha_alert("Illegal environment variable index %d. Aborting.\n", idx); |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1862 | return 1; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1863 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1864 | |
| 1865 | envname = extcheck_envs[idx].name; |
| 1866 | vmaxlen = extcheck_envs[idx].vmaxlen; |
| 1867 | |
| 1868 | /* Check if the environment variable is already set, and silently reject |
| 1869 | * the update if this one is not updatable. */ |
| 1870 | if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx])) |
| 1871 | return 0; |
| 1872 | |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1873 | /* Instead of sending NOT_USED, sending an empty value is preferable */ |
| 1874 | if (strcmp(value, "NOT_USED") == 0) { |
| 1875 | value = ""; |
| 1876 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1877 | |
| 1878 | len = strlen(envname) + 1; |
| 1879 | if (vmaxlen == EXTCHK_SIZE_EVAL_INIT) |
| 1880 | len += strlen(value); |
| 1881 | else |
| 1882 | len += vmaxlen; |
| 1883 | |
| 1884 | if (!check->envp[idx]) |
| 1885 | check->envp[idx] = malloc(len + 1); |
| 1886 | |
| 1887 | if (!check->envp[idx]) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1888 | ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1889 | return 1; |
| 1890 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1891 | ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1892 | if (ret < 0) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1893 | ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno)); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1894 | return 1; |
| 1895 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1896 | else if (ret > len) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1897 | ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1898 | return 1; |
| 1899 | } |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1900 | return 0; |
| 1901 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1902 | |
| 1903 | static int prepare_external_check(struct check *check) |
| 1904 | { |
| 1905 | struct server *s = check->server; |
| 1906 | struct proxy *px = s->proxy; |
| 1907 | struct listener *listener = NULL, *l; |
| 1908 | int i; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1909 | const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1910 | char buf[256]; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1911 | |
| 1912 | list_for_each_entry(l, &px->conf.listeners, by_fe) |
| 1913 | /* Use the first INET, INET6 or UNIX listener */ |
| 1914 | if (l->addr.ss_family == AF_INET || |
| 1915 | l->addr.ss_family == AF_INET6 || |
| 1916 | l->addr.ss_family == AF_UNIX) { |
| 1917 | listener = l; |
| 1918 | break; |
| 1919 | } |
| 1920 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1921 | check->curpid = NULL; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1922 | check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *)); |
| 1923 | if (!check->envp) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1924 | ha_alert("Failed to allocate memory for environment variables. Aborting\n"); |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1925 | goto err; |
| 1926 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1927 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1928 | check->argv = calloc(6, sizeof(char *)); |
| 1929 | if (!check->argv) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1930 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1931 | goto err; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1932 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1933 | |
| 1934 | check->argv[0] = px->check_command; |
| 1935 | |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1936 | if (!listener) { |
| 1937 | check->argv[1] = strdup("NOT_USED"); |
| 1938 | check->argv[2] = strdup("NOT_USED"); |
| 1939 | } |
| 1940 | else if (listener->addr.ss_family == AF_INET || |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1941 | listener->addr.ss_family == AF_INET6) { |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1942 | addr_to_str(&listener->addr, buf, sizeof(buf)); |
| 1943 | check->argv[1] = strdup(buf); |
| 1944 | port_to_str(&listener->addr, buf, sizeof(buf)); |
| 1945 | check->argv[2] = strdup(buf); |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1946 | } |
| 1947 | else if (listener->addr.ss_family == AF_UNIX) { |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1948 | const struct sockaddr_un *un; |
| 1949 | |
| 1950 | un = (struct sockaddr_un *)&listener->addr; |
| 1951 | check->argv[1] = strdup(un->sun_path); |
| 1952 | check->argv[2] = strdup("NOT_USED"); |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1953 | } |
| 1954 | else { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1955 | ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1956 | goto err; |
| 1957 | } |
| 1958 | |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 1959 | if (!check->argv[1] || !check->argv[2]) { |
| 1960 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 1961 | goto err; |
| 1962 | } |
| 1963 | |
| 1964 | check->argv[3] = calloc(EXTCHK_SIZE_ADDR, sizeof(*check->argv[3])); |
| 1965 | check->argv[4] = calloc(EXTCHK_SIZE_UINT, sizeof(*check->argv[4])); |
| 1966 | if (!check->argv[3] || !check->argv[4]) { |
| 1967 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 1968 | goto err; |
| 1969 | } |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 1970 | |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 1971 | addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR); |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 1972 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 1973 | snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1974 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1975 | for (i = 0; i < 5; i++) { |
| 1976 | if (!check->argv[i]) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1977 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1978 | goto err; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1979 | } |
| 1980 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1981 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1982 | EXTCHK_SETENV(check, EXTCHK_PATH, path, err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1983 | /* Add proxy environment variables */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1984 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err); |
| 1985 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err); |
| 1986 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err); |
| 1987 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1988 | /* Add server environment variables */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1989 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err); |
| 1990 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err); |
| 1991 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err); |
| 1992 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err); |
| 1993 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err); |
| 1994 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err); |
| 1995 | |
| 1996 | /* Ensure that we don't leave any hole in check->envp */ |
| 1997 | for (i = 0; i < EXTCHK_SIZE; i++) |
| 1998 | if (!check->envp[i]) |
| 1999 | EXTCHK_SETENV(check, i, "", err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 2000 | |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2001 | return 1; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2002 | err: |
| 2003 | if (check->envp) { |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 2004 | for (i = 0; i < EXTCHK_SIZE; i++) |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 2005 | free(check->envp[i]); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2006 | free(check->envp); |
| 2007 | check->envp = NULL; |
| 2008 | } |
| 2009 | |
| 2010 | if (check->argv) { |
| 2011 | for (i = 1; i < 5; i++) |
| 2012 | free(check->argv[i]); |
| 2013 | free(check->argv); |
| 2014 | check->argv = NULL; |
| 2015 | } |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2016 | return 0; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2017 | } |
| 2018 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 2019 | /* |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2020 | * establish a server health-check that makes use of a process. |
| 2021 | * |
| 2022 | * It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2023 | * - SF_ERR_NONE if everything's OK |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2024 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
Tim Düsterhus | 4896c44 | 2016-11-29 02:15:19 +0100 | [diff] [blame] | 2025 | * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted. |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2026 | * |
| 2027 | * Blocks and then unblocks SIGCHLD |
| 2028 | */ |
| 2029 | static int connect_proc_chk(struct task *t) |
| 2030 | { |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 2031 | char buf[256]; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2032 | struct check *check = t->context; |
| 2033 | struct server *s = check->server; |
| 2034 | struct proxy *px = s->proxy; |
| 2035 | int status; |
| 2036 | pid_t pid; |
| 2037 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2038 | status = SF_ERR_RESOURCE; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2039 | |
| 2040 | block_sigchld(); |
| 2041 | |
| 2042 | pid = fork(); |
| 2043 | if (pid < 0) { |
Willy Tarreau | d96f112 | 2019-12-03 07:07:36 +0100 | [diff] [blame] | 2044 | ha_alert("Failed to fork process for external health check%s: %s. Aborting.\n", |
| 2045 | (global.tune.options & GTUNE_INSECURE_FORK) ? |
| 2046 | "" : " (likely caused by missing 'insecure-fork-wanted')", |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2047 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2048 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 2049 | goto out; |
| 2050 | } |
| 2051 | if (pid == 0) { |
| 2052 | /* Child */ |
| 2053 | extern char **environ; |
Willy Tarreau | 9f6dc72 | 2019-03-01 11:15:10 +0100 | [diff] [blame] | 2054 | struct rlimit limit; |
Willy Tarreau | b7b2478 | 2016-06-21 15:32:29 +0200 | [diff] [blame] | 2055 | int fd; |
| 2056 | |
| 2057 | /* close all FDs. Keep stdin/stdout/stderr in verbose mode */ |
| 2058 | fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3; |
| 2059 | |
Willy Tarreau | 2555ccf | 2019-02-21 22:22:06 +0100 | [diff] [blame] | 2060 | my_closefrom(fd); |
Willy Tarreau | b7b2478 | 2016-06-21 15:32:29 +0200 | [diff] [blame] | 2061 | |
Willy Tarreau | 9f6dc72 | 2019-03-01 11:15:10 +0100 | [diff] [blame] | 2062 | /* restore the initial FD limits */ |
| 2063 | limit.rlim_cur = rlim_fd_cur_at_boot; |
| 2064 | limit.rlim_max = rlim_fd_max_at_boot; |
| 2065 | if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { |
| 2066 | getrlimit(RLIMIT_NOFILE, &limit); |
| 2067 | ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n", |
| 2068 | rlim_fd_cur_at_boot, rlim_fd_max_at_boot, |
| 2069 | (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max); |
| 2070 | } |
| 2071 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2072 | environ = check->envp; |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 2073 | |
| 2074 | /* Update some environment variables and command args: curconn, server addr and server port */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 2075 | extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf))); |
Christopher Faulet | aaae9a0 | 2020-04-26 09:50:31 +0200 | [diff] [blame] | 2076 | |
| 2077 | addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR); |
| 2078 | extchk_setenv(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3]); |
| 2079 | |
| 2080 | *check->argv[4] = 0; |
| 2081 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 2082 | snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port); |
| 2083 | extchk_setenv(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4]); |
| 2084 | |
Willy Tarreau | 2df8cad | 2019-07-01 07:51:29 +0200 | [diff] [blame] | 2085 | haproxy_unblock_signals(); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2086 | execvp(px->check_command, check->argv); |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2087 | ha_alert("Failed to exec process for external health check: %s. Aborting.\n", |
| 2088 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2089 | exit(-1); |
| 2090 | } |
| 2091 | |
| 2092 | /* Parent */ |
| 2093 | if (check->result == CHK_RES_UNKNOWN) { |
| 2094 | if (pid_list_add(pid, t) != NULL) { |
| 2095 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 2096 | |
| 2097 | if (px->timeout.check && px->timeout.connect) { |
| 2098 | int t_con = tick_add(now_ms, px->timeout.connect); |
| 2099 | t->expire = tick_first(t->expire, t_con); |
| 2100 | } |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2101 | status = SF_ERR_NONE; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2102 | goto out; |
| 2103 | } |
| 2104 | else { |
| 2105 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 2106 | } |
| 2107 | kill(pid, SIGTERM); /* process creation error */ |
| 2108 | } |
| 2109 | else |
| 2110 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 2111 | |
| 2112 | out: |
| 2113 | unblock_sigchld(); |
| 2114 | return status; |
| 2115 | } |
| 2116 | |
| 2117 | /* |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 2118 | * manages a server health-check that uses an external process. Returns |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2119 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2120 | * |
| 2121 | * Please do NOT place any return statement in this function and only leave |
| 2122 | * via the out_unlock label. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2123 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2124 | static struct task *process_chk_proc(struct task *t, void *context, unsigned short state) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2125 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2126 | struct check *check = context; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2127 | struct server *s = check->server; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2128 | int rv; |
| 2129 | int ret; |
| 2130 | int expired = tick_is_expired(t->expire, now_ms); |
| 2131 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2132 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2133 | if (!(check->state & CHK_ST_INPROGRESS)) { |
| 2134 | /* no check currently running */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2135 | if (!expired) /* woke up too early */ |
| 2136 | goto out_unlock; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2137 | |
| 2138 | /* we don't send any health-checks when the proxy is |
| 2139 | * stopped, the server should not be checked or the check |
| 2140 | * is disabled. |
| 2141 | */ |
| 2142 | if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) || |
| 2143 | s->proxy->state == PR_STSTOPPED) |
| 2144 | goto reschedule; |
| 2145 | |
| 2146 | /* we'll initiate a new check */ |
| 2147 | set_server_check_status(check, HCHK_STATUS_START, NULL); |
| 2148 | |
| 2149 | check->state |= CHK_ST_INPROGRESS; |
| 2150 | |
Simon Horman | dbf7019 | 2015-01-30 11:22:53 +0900 | [diff] [blame] | 2151 | ret = connect_proc_chk(t); |
Willy Tarreau | d7c3fbd | 2017-10-04 15:19:26 +0200 | [diff] [blame] | 2152 | if (ret == SF_ERR_NONE) { |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 2153 | /* the process was forked, we allow up to min(inter, |
| 2154 | * timeout.connect) for it to report its status, but |
| 2155 | * only when timeout.check is set as it may be to short |
| 2156 | * for a full check otherwise. |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2157 | */ |
| 2158 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 2159 | |
| 2160 | if (s->proxy->timeout.check && s->proxy->timeout.connect) { |
| 2161 | int t_con = tick_add(now_ms, s->proxy->timeout.connect); |
| 2162 | t->expire = tick_first(t->expire, t_con); |
| 2163 | } |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2164 | task_set_affinity(t, tid_bit); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2165 | goto reschedule; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2166 | } |
| 2167 | |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 2168 | /* here, we failed to start the check */ |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2169 | |
| 2170 | check->state &= ~CHK_ST_INPROGRESS; |
| 2171 | check_notify_failure(check); |
| 2172 | |
| 2173 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2174 | * to establish but only when timeout.check is set |
| 2175 | * as it may be to short for a full check otherwise |
| 2176 | */ |
| 2177 | while (tick_is_expired(t->expire, now_ms)) { |
| 2178 | int t_con; |
| 2179 | |
| 2180 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
| 2181 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 2182 | |
| 2183 | if (s->proxy->timeout.check) |
| 2184 | t->expire = tick_first(t->expire, t_con); |
| 2185 | } |
| 2186 | } |
| 2187 | else { |
| 2188 | /* there was a test running. |
| 2189 | * First, let's check whether there was an uncaught error, |
| 2190 | * which can happen on connect timeout or error. |
| 2191 | */ |
| 2192 | if (check->result == CHK_RES_UNKNOWN) { |
| 2193 | /* good connection is enough for pure TCP check */ |
| 2194 | struct pid_list *elem = check->curpid; |
| 2195 | int status = HCHK_STATUS_UNKNOWN; |
| 2196 | |
| 2197 | if (elem->exited) { |
| 2198 | status = elem->status; /* Save in case the process exits between use below */ |
| 2199 | if (!WIFEXITED(status)) |
| 2200 | check->code = -1; |
| 2201 | else |
| 2202 | check->code = WEXITSTATUS(status); |
| 2203 | if (!WIFEXITED(status) || WEXITSTATUS(status)) |
| 2204 | status = HCHK_STATUS_PROCERR; |
| 2205 | else |
| 2206 | status = HCHK_STATUS_PROCOK; |
| 2207 | } else if (expired) { |
| 2208 | status = HCHK_STATUS_PROCTOUT; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2209 | ha_warning("kill %d\n", (int)elem->pid); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2210 | kill(elem->pid, SIGTERM); |
| 2211 | } |
| 2212 | set_server_check_status(check, status, NULL); |
| 2213 | } |
| 2214 | |
| 2215 | if (check->result == CHK_RES_FAILED) { |
| 2216 | /* a failure or timeout detected */ |
| 2217 | check_notify_failure(check); |
| 2218 | } |
| 2219 | else if (check->result == CHK_RES_CONDPASS) { |
| 2220 | /* check is OK but asks for stopping mode */ |
| 2221 | check_notify_stopping(check); |
| 2222 | } |
| 2223 | else if (check->result == CHK_RES_PASSED) { |
| 2224 | /* a success was detected */ |
| 2225 | check_notify_success(check); |
| 2226 | } |
Willy Tarreau | 6dd4ac8 | 2019-09-03 18:55:02 +0200 | [diff] [blame] | 2227 | task_set_affinity(t, 1); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2228 | check->state &= ~CHK_ST_INPROGRESS; |
| 2229 | |
| 2230 | pid_list_del(check->curpid); |
| 2231 | |
| 2232 | rv = 0; |
| 2233 | if (global.spread_checks > 0) { |
| 2234 | rv = srv_getinter(check) * global.spread_checks / 100; |
Willy Tarreau | 5a6d3e7 | 2020-03-08 17:53:53 +0100 | [diff] [blame] | 2235 | rv -= (int) (2 * rv * (ha_random32() / 4294967295.0)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2236 | } |
| 2237 | t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv)); |
| 2238 | } |
| 2239 | |
| 2240 | reschedule: |
| 2241 | while (tick_is_expired(t->expire, now_ms)) |
| 2242 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2243 | |
| 2244 | out_unlock: |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2245 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2246 | return t; |
| 2247 | } |
| 2248 | |
| 2249 | /* |
| 2250 | * manages a server health-check that uses a connection. Returns |
| 2251 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2252 | * |
| 2253 | * Please do NOT place any return statement in this function and only leave |
| 2254 | * via the out_unlock label. |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2255 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2256 | static struct task *process_chk_conn(struct task *t, void *context, unsigned short state) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2257 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2258 | struct check *check = context; |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2259 | struct proxy *proxy = check->proxy; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2260 | struct conn_stream *cs = check->cs; |
| 2261 | struct connection *conn = cs_conn(cs); |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2262 | int rv; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2263 | int ret; |
Willy Tarreau | acbdc7a | 2012-11-23 14:02:10 +0100 | [diff] [blame] | 2264 | int expired = tick_is_expired(t->expire, now_ms); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2265 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2266 | if (check->server) |
| 2267 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2268 | if (!(check->state & CHK_ST_INPROGRESS)) { |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2269 | /* no check currently running */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2270 | if (!expired) /* woke up too early */ |
| 2271 | goto out_unlock; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2272 | |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 2273 | /* we don't send any health-checks when the proxy is |
| 2274 | * stopped, the server should not be checked or the check |
| 2275 | * is disabled. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2276 | */ |
Willy Tarreau | 0d924cc | 2013-12-11 21:26:24 +0100 | [diff] [blame] | 2277 | if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) || |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2278 | proxy->state == PR_STSTOPPED) |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2279 | goto reschedule; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2280 | |
| 2281 | /* we'll initiate a new check */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2282 | set_server_check_status(check, HCHK_STATUS_START, NULL); |
Willy Tarreau | 1ae1b7b | 2012-09-28 15:28:30 +0200 | [diff] [blame] | 2283 | |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2284 | check->state |= CHK_ST_INPROGRESS; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2285 | b_reset(&check->bi); |
| 2286 | b_reset(&check->bo); |
Willy Tarreau | 1ae1b7b | 2012-09-28 15:28:30 +0200 | [diff] [blame] | 2287 | |
Olivier Houchard | aebeff7 | 2019-11-29 16:18:51 +0100 | [diff] [blame] | 2288 | task_set_affinity(t, tid_bit); |
Simon Horman | dbf7019 | 2015-01-30 11:22:53 +0900 | [diff] [blame] | 2289 | ret = connect_conn_chk(t); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2290 | cs = check->cs; |
| 2291 | conn = cs_conn(cs); |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2292 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2293 | switch (ret) { |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2294 | case SF_ERR_UP: |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2295 | goto out_unlock; |
| 2296 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2297 | case SF_ERR_NONE: |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2298 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2299 | * to establish but only when timeout.check is set |
| 2300 | * as it may be to short for a full check otherwise |
| 2301 | */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2302 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 2303 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2304 | if (proxy->timeout.check && proxy->timeout.connect) { |
| 2305 | int t_con = tick_add(now_ms, proxy->timeout.connect); |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2306 | t->expire = tick_first(t->expire, t_con); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2307 | } |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 2308 | |
Willy Tarreau | cc705a6 | 2019-09-05 17:51:30 +0200 | [diff] [blame] | 2309 | if (check->type) { |
| 2310 | /* send the request if we have one. We avoid receiving |
| 2311 | * if not connected, unless we didn't subscribe for |
| 2312 | * sending since otherwise we won't be woken up. |
| 2313 | */ |
| 2314 | __event_srv_chk_w(cs); |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 2315 | if (!(conn->flags & CO_FL_WAIT_XPRT) || |
Willy Tarreau | c594039 | 2019-09-05 17:38:40 +0200 | [diff] [blame] | 2316 | !(check->wait_list.events & SUB_RETRY_SEND)) |
| 2317 | __event_srv_chk_r(cs); |
Willy Tarreau | cc705a6 | 2019-09-05 17:51:30 +0200 | [diff] [blame] | 2318 | } |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 2319 | |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2320 | goto reschedule; |
| 2321 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2322 | case SF_ERR_SRVTO: /* ETIMEDOUT */ |
| 2323 | case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2324 | if (conn) |
| 2325 | conn->flags |= CO_FL_ERROR; |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2326 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2327 | break; |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 2328 | /* should share same code than cases below */ |
| 2329 | case SF_ERR_CHK_PORT: |
| 2330 | check->state |= CHK_ST_PORT_MISS; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2331 | case SF_ERR_PRXCOND: |
| 2332 | case SF_ERR_RESOURCE: |
| 2333 | case SF_ERR_INTERNAL: |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2334 | if (conn) |
| 2335 | conn->flags |= CO_FL_ERROR; |
| 2336 | chk_report_conn_err(check, conn ? 0 : ENOMEM, 0); |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2337 | break; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2338 | } |
| 2339 | |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2340 | /* here, we have seen a synchronous error, no fd was allocated */ |
Olivier Houchard | aebeff7 | 2019-11-29 16:18:51 +0100 | [diff] [blame] | 2341 | task_set_affinity(t, MAX_THREADS_MASK); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2342 | if (cs) { |
Olivier Houchard | 4906554 | 2019-05-31 19:20:36 +0200 | [diff] [blame] | 2343 | if (check->wait_list.events) |
| 2344 | cs->conn->xprt->unsubscribe(cs->conn, |
| 2345 | cs->conn->xprt_ctx, |
| 2346 | check->wait_list.events, |
| 2347 | &check->wait_list); |
| 2348 | /* We may have been scheduled to run, and the |
| 2349 | * I/O handler expects to have a cs, so remove |
| 2350 | * the tasklet |
| 2351 | */ |
Willy Tarreau | 86eded6 | 2019-06-14 14:47:49 +0200 | [diff] [blame] | 2352 | tasklet_remove_from_tasklet_list(check->wait_list.tasklet); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2353 | cs_destroy(cs); |
| 2354 | cs = check->cs = NULL; |
| 2355 | conn = NULL; |
Olivier Houchard | 390485a | 2017-10-24 19:03:30 +0200 | [diff] [blame] | 2356 | } |
Willy Tarreau | 6b0a850 | 2012-11-23 08:51:32 +0100 | [diff] [blame] | 2357 | |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2358 | check->state &= ~CHK_ST_INPROGRESS; |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 2359 | check_notify_failure(check); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2360 | |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2361 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2362 | * to establish but only when timeout.check is set |
| 2363 | * as it may be to short for a full check otherwise |
| 2364 | */ |
Willy Tarreau | 0c303ee | 2008-07-07 00:09:58 +0200 | [diff] [blame] | 2365 | while (tick_is_expired(t->expire, now_ms)) { |
| 2366 | int t_con; |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2367 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2368 | t_con = tick_add(t->expire, proxy->timeout.connect); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2369 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2370 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2371 | if (proxy->timeout.check) |
Willy Tarreau | 0c303ee | 2008-07-07 00:09:58 +0200 | [diff] [blame] | 2372 | t->expire = tick_first(t->expire, t_con); |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2373 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2374 | } |
| 2375 | else { |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2376 | /* there was a test running. |
| 2377 | * First, let's check whether there was an uncaught error, |
| 2378 | * which can happen on connect timeout or error. |
| 2379 | */ |
Simon Horman | ccaabcd | 2014-06-20 12:29:47 +0900 | [diff] [blame] | 2380 | if (check->result == CHK_RES_UNKNOWN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 2381 | /* good connection is enough for pure TCP check */ |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 2382 | if (!(conn->flags & CO_FL_WAIT_XPRT) && !check->type) { |
Christopher Faulet | f61f33a | 2020-03-27 18:55:49 +0100 | [diff] [blame] | 2383 | if (check->use_ssl == 1) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2384 | set_server_check_status(check, HCHK_STATUS_L6OK, NULL); |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2385 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2386 | set_server_check_status(check, HCHK_STATUS_L4OK, NULL); |
Willy Tarreau | acbdc7a | 2012-11-23 14:02:10 +0100 | [diff] [blame] | 2387 | } |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 2388 | else if ((conn->flags & CO_FL_ERROR) || cs->flags & CS_FL_ERROR || expired) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2389 | chk_report_conn_err(check, 0, expired); |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2390 | } |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2391 | else |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2392 | goto out_unlock; /* timeout not reached, wait again */ |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2393 | } |
| 2394 | |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2395 | /* check complete or aborted */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2396 | if (conn && conn->xprt) { |
Willy Tarreau | 5ba04f6 | 2013-02-12 15:23:12 +0100 | [diff] [blame] | 2397 | /* The check was aborted and the connection was not yet closed. |
| 2398 | * This can happen upon timeout, or when an external event such |
| 2399 | * as a failed response coupled with "observe layer7" caused the |
| 2400 | * server state to be suddenly changed. |
| 2401 | */ |
Willy Tarreau | d85c485 | 2015-03-13 00:40:28 +0100 | [diff] [blame] | 2402 | conn_sock_drain(conn); |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 2403 | cs_close(cs); |
Willy Tarreau | 5ba04f6 | 2013-02-12 15:23:12 +0100 | [diff] [blame] | 2404 | } |
| 2405 | |
Willy Tarreau | ac59f36 | 2017-10-08 11:10:19 +0200 | [diff] [blame] | 2406 | if (cs) { |
Olivier Houchard | 4906554 | 2019-05-31 19:20:36 +0200 | [diff] [blame] | 2407 | if (check->wait_list.events) |
| 2408 | cs->conn->xprt->unsubscribe(cs->conn, |
| 2409 | cs->conn->xprt_ctx, |
| 2410 | check->wait_list.events, |
| 2411 | &check->wait_list); |
| 2412 | /* We may have been scheduled to run, and the |
Willy Tarreau | 86eded6 | 2019-06-14 14:47:49 +0200 | [diff] [blame] | 2413 | * I/O handler expects to have a cs, so remove |
| 2414 | * the tasklet |
| 2415 | */ |
| 2416 | tasklet_remove_from_tasklet_list(check->wait_list.tasklet); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2417 | cs_destroy(cs); |
| 2418 | cs = check->cs = NULL; |
| 2419 | conn = NULL; |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2420 | } |
| 2421 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2422 | if (check->server) { |
| 2423 | if (check->result == CHK_RES_FAILED) { |
| 2424 | /* a failure or timeout detected */ |
| 2425 | check_notify_failure(check); |
| 2426 | } |
| 2427 | else if (check->result == CHK_RES_CONDPASS) { |
| 2428 | /* check is OK but asks for stopping mode */ |
| 2429 | check_notify_stopping(check); |
| 2430 | } |
| 2431 | else if (check->result == CHK_RES_PASSED) { |
| 2432 | /* a success was detected */ |
| 2433 | check_notify_success(check); |
| 2434 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2435 | } |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2436 | task_set_affinity(t, MAX_THREADS_MASK); |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2437 | check->state &= ~CHK_ST_INPROGRESS; |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2438 | |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2439 | if (check->server) { |
| 2440 | rv = 0; |
| 2441 | if (global.spread_checks > 0) { |
| 2442 | rv = srv_getinter(check) * global.spread_checks / 100; |
Willy Tarreau | 5a6d3e7 | 2020-03-08 17:53:53 +0100 | [diff] [blame] | 2443 | rv -= (int) (2 * rv * (ha_random32() / 4294967295.0)); |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2444 | } |
| 2445 | t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv)); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2446 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2447 | } |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2448 | |
| 2449 | reschedule: |
| 2450 | while (tick_is_expired(t->expire, now_ms)) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2451 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2452 | out_unlock: |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2453 | if (check->server) |
| 2454 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
Willy Tarreau | 26c2506 | 2009-03-08 09:38:41 +0100 | [diff] [blame] | 2455 | return t; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2456 | } |
| 2457 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2458 | /* |
| 2459 | * manages a server health-check. Returns |
| 2460 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
| 2461 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2462 | static struct task *process_chk(struct task *t, void *context, unsigned short state) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2463 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2464 | struct check *check = context; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2465 | |
| 2466 | if (check->type == PR_O2_EXT_CHK) |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2467 | return process_chk_proc(t, context, state); |
| 2468 | return process_chk_conn(t, context, state); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 2469 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2470 | } |
| 2471 | |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2472 | static int start_check_task(struct check *check, int mininter, |
| 2473 | int nbcheck, int srvpos) |
| 2474 | { |
| 2475 | struct task *t; |
Willy Tarreau | 6dd4ac8 | 2019-09-03 18:55:02 +0200 | [diff] [blame] | 2476 | unsigned long thread_mask = MAX_THREADS_MASK; |
| 2477 | |
| 2478 | if (check->type == PR_O2_EXT_CHK) |
| 2479 | thread_mask = 1; |
| 2480 | |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2481 | /* task for the check */ |
Willy Tarreau | 6dd4ac8 | 2019-09-03 18:55:02 +0200 | [diff] [blame] | 2482 | if ((t = task_new(thread_mask)) == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2483 | ha_alert("Starting [%s:%s] check: out of memory.\n", |
| 2484 | check->server->proxy->id, check->server->id); |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2485 | return 0; |
| 2486 | } |
| 2487 | |
| 2488 | check->task = t; |
| 2489 | t->process = process_chk; |
| 2490 | t->context = check; |
| 2491 | |
Willy Tarreau | 1746eec | 2014-04-25 10:46:47 +0200 | [diff] [blame] | 2492 | if (mininter < srv_getinter(check)) |
| 2493 | mininter = srv_getinter(check); |
| 2494 | |
| 2495 | if (global.max_spread_checks && mininter > global.max_spread_checks) |
| 2496 | mininter = global.max_spread_checks; |
| 2497 | |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2498 | /* check this every ms */ |
Willy Tarreau | 1746eec | 2014-04-25 10:46:47 +0200 | [diff] [blame] | 2499 | t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck)); |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2500 | check->start = now; |
| 2501 | task_queue(t); |
| 2502 | |
| 2503 | return 1; |
| 2504 | } |
| 2505 | |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2506 | /* |
| 2507 | * Start health-check. |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2508 | * Returns 0 if OK, ERR_FATAL on error, and prints the error in this case. |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2509 | */ |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2510 | static int start_checks() |
| 2511 | { |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2512 | |
| 2513 | struct proxy *px; |
| 2514 | struct server *s; |
| 2515 | struct task *t; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2516 | int nbcheck=0, mininter=0, srvpos=0; |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2517 | |
Willy Tarreau | 2c43a1e | 2007-10-14 23:05:39 +0200 | [diff] [blame] | 2518 | /* 1- count the checkers to run simultaneously. |
| 2519 | * We also determine the minimum interval among all of those which |
| 2520 | * have an interval larger than SRV_CHK_INTER_THRES. This interval |
| 2521 | * will be used to spread their start-up date. Those which have |
Jamie Gloudon | 801a0a3 | 2012-08-25 00:18:33 -0400 | [diff] [blame] | 2522 | * a shorter interval will start independently and will not dictate |
Willy Tarreau | 2c43a1e | 2007-10-14 23:05:39 +0200 | [diff] [blame] | 2523 | * too short an interval for all others. |
| 2524 | */ |
Olivier Houchard | fbc74e8 | 2017-11-24 16:54:05 +0100 | [diff] [blame] | 2525 | for (px = proxies_list; px; px = px->next) { |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2526 | for (s = px->srv; s; s = s->next) { |
Willy Tarreau | e7b7348 | 2013-11-21 11:50:50 +0100 | [diff] [blame] | 2527 | if (s->slowstart) { |
Emeric Brun | c60def8 | 2017-09-27 14:59:38 +0200 | [diff] [blame] | 2528 | if ((t = task_new(MAX_THREADS_MASK)) == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2529 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2530 | return ERR_ALERT | ERR_FATAL; |
Willy Tarreau | e7b7348 | 2013-11-21 11:50:50 +0100 | [diff] [blame] | 2531 | } |
| 2532 | /* We need a warmup task that will be called when the server |
| 2533 | * state switches from down to up. |
| 2534 | */ |
| 2535 | s->warmup = t; |
| 2536 | t->process = server_warmup; |
| 2537 | t->context = s; |
Baptiste Assmann | 6076d1c | 2015-09-17 22:53:59 +0200 | [diff] [blame] | 2538 | /* server can be in this state only because of */ |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 2539 | if (s->next_state == SRV_ST_STARTING) |
Baptiste Assmann | 6076d1c | 2015-09-17 22:53:59 +0200 | [diff] [blame] | 2540 | task_schedule(s->warmup, tick_add(now_ms, MS_TO_TICKS(MAX(1000, (now.tv_sec - s->last_change)) / 20))); |
Willy Tarreau | e7b7348 | 2013-11-21 11:50:50 +0100 | [diff] [blame] | 2541 | } |
| 2542 | |
Willy Tarreau | d8514a2 | 2013-12-11 21:10:14 +0100 | [diff] [blame] | 2543 | if (s->check.state & CHK_ST_CONFIGURED) { |
| 2544 | nbcheck++; |
| 2545 | if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) && |
| 2546 | (!mininter || mininter > srv_getinter(&s->check))) |
| 2547 | mininter = srv_getinter(&s->check); |
| 2548 | } |
Willy Tarreau | 15f3910 | 2013-12-11 20:41:18 +0100 | [diff] [blame] | 2549 | |
Willy Tarreau | d8514a2 | 2013-12-11 21:10:14 +0100 | [diff] [blame] | 2550 | if (s->agent.state & CHK_ST_CONFIGURED) { |
| 2551 | nbcheck++; |
| 2552 | if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) && |
| 2553 | (!mininter || mininter > srv_getinter(&s->agent))) |
| 2554 | mininter = srv_getinter(&s->agent); |
| 2555 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2556 | } |
| 2557 | } |
| 2558 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2559 | if (!nbcheck) |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2560 | return 0; |
| 2561 | |
| 2562 | srand((unsigned)time(NULL)); |
| 2563 | |
| 2564 | /* |
| 2565 | * 2- start them as far as possible from each others. For this, we will |
| 2566 | * start them after their interval set to the min interval divided by |
| 2567 | * the number of servers, weighted by the server's position in the list. |
| 2568 | */ |
Olivier Houchard | fbc74e8 | 2017-11-24 16:54:05 +0100 | [diff] [blame] | 2569 | for (px = proxies_list; px; px = px->next) { |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2570 | if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) { |
| 2571 | if (init_pid_list()) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2572 | ha_alert("Starting [%s] check: out of memory.\n", px->id); |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2573 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2574 | } |
| 2575 | } |
| 2576 | |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2577 | for (s = px->srv; s; s = s->next) { |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2578 | /* A task for the main check */ |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 2579 | if (s->check.state & CHK_ST_CONFIGURED) { |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2580 | if (s->check.type == PR_O2_EXT_CHK) { |
| 2581 | if (!prepare_external_check(&s->check)) |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2582 | return ERR_ALERT | ERR_FATAL; |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2583 | } |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2584 | if (!start_check_task(&s->check, mininter, nbcheck, srvpos)) |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2585 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2586 | srvpos++; |
| 2587 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2588 | |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2589 | /* A task for a auxiliary agent check */ |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 2590 | if (s->agent.state & CHK_ST_CONFIGURED) { |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2591 | if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) { |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2592 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2593 | } |
| 2594 | srvpos++; |
| 2595 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2596 | } |
| 2597 | } |
| 2598 | return 0; |
| 2599 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2600 | |
| 2601 | /* |
Willy Tarreau | 5b3a202 | 2012-09-28 15:01:02 +0200 | [diff] [blame] | 2602 | * Perform content verification check on data in s->check.buffer buffer. |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2603 | * The buffer MUST be terminated by a null byte before calling this function. |
| 2604 | * Sets server status appropriately. The caller is responsible for ensuring |
| 2605 | * that the buffer contains at least 13 characters. If <done> is zero, we may |
| 2606 | * return 0 to indicate that data is required to decide of a match. |
| 2607 | */ |
| 2608 | static int httpchk_expect(struct server *s, int done) |
| 2609 | { |
Christopher Faulet | 1bc04c7 | 2017-10-29 20:14:08 +0100 | [diff] [blame] | 2610 | static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>"; |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2611 | char status_code[] = "000"; |
| 2612 | char *contentptr; |
| 2613 | int crlf; |
| 2614 | int ret; |
| 2615 | |
| 2616 | switch (s->proxy->options2 & PR_O2_EXP_TYPE) { |
| 2617 | case PR_O2_EXP_STS: |
| 2618 | case PR_O2_EXP_RSTS: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2619 | memcpy(status_code, b_head(&s->check.bi) + 9, 3); |
| 2620 | memcpy(status_msg + strlen(status_msg) - 4, b_head(&s->check.bi) + 9, 3); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2621 | |
| 2622 | if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS) |
| 2623 | ret = strncmp(s->proxy->expect_str, status_code, 3) == 0; |
| 2624 | else |
Thierry FOURNIER | 09af0d6 | 2014-06-18 11:35:54 +0200 | [diff] [blame] | 2625 | ret = regex_exec(s->proxy->expect_regex, status_code); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2626 | |
| 2627 | /* we necessarily have the response, so there are no partial failures */ |
| 2628 | if (s->proxy->options2 & PR_O2_EXP_INV) |
| 2629 | ret = !ret; |
| 2630 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2631 | set_server_check_status(&s->check, ret ? HCHK_STATUS_L7OKD : HCHK_STATUS_L7STS, status_msg); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2632 | break; |
| 2633 | |
| 2634 | case PR_O2_EXP_STR: |
| 2635 | case PR_O2_EXP_RSTR: |
| 2636 | /* very simple response parser: ignore CR and only count consecutive LFs, |
| 2637 | * stop with contentptr pointing to first char after the double CRLF or |
| 2638 | * to '\0' if crlf < 2. |
| 2639 | */ |
| 2640 | crlf = 0; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2641 | for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) { |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2642 | if (crlf >= 2) |
| 2643 | break; |
| 2644 | if (*contentptr == '\r') |
| 2645 | continue; |
| 2646 | else if (*contentptr == '\n') |
| 2647 | crlf++; |
| 2648 | else |
| 2649 | crlf = 0; |
| 2650 | } |
| 2651 | |
| 2652 | /* Check that response contains a body... */ |
| 2653 | if (crlf < 2) { |
| 2654 | if (!done) |
| 2655 | return 0; |
| 2656 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2657 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2658 | "HTTP content check could not find a response body"); |
| 2659 | return 1; |
| 2660 | } |
| 2661 | |
| 2662 | /* Check that response body is not empty... */ |
| 2663 | if (*contentptr == '\0') { |
Willy Tarreau | a164fb5 | 2011-04-13 09:32:41 +0200 | [diff] [blame] | 2664 | if (!done) |
| 2665 | return 0; |
| 2666 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2667 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2668 | "HTTP content check found empty response body"); |
| 2669 | return 1; |
| 2670 | } |
| 2671 | |
| 2672 | /* Check the response content against the supplied string |
| 2673 | * or regex... */ |
| 2674 | if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR) |
| 2675 | ret = strstr(contentptr, s->proxy->expect_str) != NULL; |
| 2676 | else |
Thierry FOURNIER | 09af0d6 | 2014-06-18 11:35:54 +0200 | [diff] [blame] | 2677 | ret = regex_exec(s->proxy->expect_regex, contentptr); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2678 | |
| 2679 | /* if we don't match, we may need to wait more */ |
| 2680 | if (!ret && !done) |
| 2681 | return 0; |
| 2682 | |
| 2683 | if (ret) { |
| 2684 | /* content matched */ |
| 2685 | if (s->proxy->options2 & PR_O2_EXP_INV) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2686 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2687 | "HTTP check matched unwanted content"); |
| 2688 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2689 | set_server_check_status(&s->check, HCHK_STATUS_L7OKD, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2690 | "HTTP content check matched"); |
| 2691 | } |
| 2692 | else { |
| 2693 | if (s->proxy->options2 & PR_O2_EXP_INV) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2694 | set_server_check_status(&s->check, HCHK_STATUS_L7OKD, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2695 | "HTTP check did not match unwanted content"); |
| 2696 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2697 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2698 | "HTTP content check did not match"); |
| 2699 | } |
| 2700 | break; |
| 2701 | } |
| 2702 | return 1; |
| 2703 | } |
| 2704 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2705 | /* |
| 2706 | * return the id of a step in a send/expect session |
| 2707 | */ |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2708 | static int tcpcheck_get_step_id(struct check *check) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2709 | { |
| 2710 | struct tcpcheck_rule *cur = NULL, *next = NULL; |
| 2711 | int i = 0; |
| 2712 | |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2713 | /* not even started anything yet => step 0 = initial connect */ |
Baptiste Assmann | f95bc8e | 2015-04-25 16:03:06 +0200 | [diff] [blame] | 2714 | if (!check->current_step) |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2715 | return 0; |
| 2716 | |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2717 | cur = check->last_started_step; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2718 | |
| 2719 | /* no step => first step */ |
| 2720 | if (cur == NULL) |
| 2721 | return 1; |
| 2722 | |
| 2723 | /* increment i until current step */ |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2724 | list_for_each_entry(next, check->tcpcheck_rules, list) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2725 | if (next->list.p == &cur->list) |
| 2726 | break; |
| 2727 | ++i; |
| 2728 | } |
| 2729 | |
| 2730 | return i; |
| 2731 | } |
| 2732 | |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2733 | /* |
| 2734 | * return the latest known comment before (including) the given stepid |
| 2735 | * returns NULL if no comment found |
| 2736 | */ |
| 2737 | static char * tcpcheck_get_step_comment(struct check *check, int stepid) |
| 2738 | { |
| 2739 | struct tcpcheck_rule *cur = NULL; |
| 2740 | char *ret = NULL; |
| 2741 | int i = 0; |
| 2742 | |
| 2743 | /* not even started anything yet, return latest comment found before any action */ |
| 2744 | if (!check->current_step) { |
| 2745 | list_for_each_entry(cur, check->tcpcheck_rules, list) { |
| 2746 | if (cur->action == TCPCHK_ACT_COMMENT) |
| 2747 | ret = cur->comment; |
| 2748 | else |
| 2749 | goto return_comment; |
| 2750 | } |
| 2751 | } |
| 2752 | |
| 2753 | i = 1; |
| 2754 | list_for_each_entry(cur, check->tcpcheck_rules, list) { |
| 2755 | if (cur->comment) |
| 2756 | ret = cur->comment; |
| 2757 | |
| 2758 | if (i >= stepid) |
| 2759 | goto return_comment; |
| 2760 | |
| 2761 | ++i; |
| 2762 | } |
| 2763 | |
| 2764 | return_comment: |
| 2765 | return ret; |
| 2766 | } |
| 2767 | |
Willy Tarreau | be74b88 | 2017-10-04 16:22:49 +0200 | [diff] [blame] | 2768 | /* proceed with next steps for the TCP checks <check>. Note that this is called |
| 2769 | * both from the connection's wake() callback and from the check scheduling task. |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 2770 | * It returns 0 on normal cases, or <0 if a close() has happened on an existing |
| 2771 | * connection, presenting the risk of an fd replacement. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2772 | * |
| 2773 | * Please do NOT place any return statement in this function and only leave |
Willy Tarreau | 2ab5c38 | 2019-07-17 18:48:07 +0200 | [diff] [blame] | 2774 | * via the out_end_tcpcheck label after setting retcode. |
Willy Tarreau | be74b88 | 2017-10-04 16:22:49 +0200 | [diff] [blame] | 2775 | */ |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 2776 | static int tcpcheck_main(struct check *check) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2777 | { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2778 | char *contentptr, *comment; |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2779 | struct tcpcheck_rule *next; |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2780 | int done = 0, ret = 0, step = 0; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2781 | struct conn_stream *cs = check->cs; |
| 2782 | struct connection *conn = cs_conn(cs); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2783 | struct server *s = check->server; |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2784 | struct proxy *proxy = check->proxy; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2785 | struct task *t = check->task; |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2786 | struct list *head = check->tcpcheck_rules; |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 2787 | int retcode = 0; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2788 | |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2789 | /* here, we know that the check is complete or that it failed */ |
| 2790 | if (check->result != CHK_RES_UNKNOWN) |
| 2791 | goto out_end_tcpcheck; |
| 2792 | |
| 2793 | /* We have 4 possibilities here : |
| 2794 | * 1. we've not yet attempted step 1, and step 1 is a connect, so no |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2795 | * connection attempt was made yet ; conn==NULL;current_step==NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2796 | * 2. we've not yet attempted step 1, and step 1 is a not connect or |
| 2797 | * does not exist (no rule), so a connection attempt was made |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2798 | * before coming here, conn!=NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2799 | * 3. we're coming back after having started with step 1, so we may |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2800 | * be waiting for a connection attempt to complete. conn!=NULL. |
| 2801 | * 4. the connection + handshake are complete. conn!=NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2802 | * |
| 2803 | * #2 and #3 are quite similar, we want both the connection and the |
| 2804 | * handshake to complete before going any further. Thus we must always |
| 2805 | * wait for a connection to complete unless we're before and existing |
| 2806 | * step 1. |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2807 | */ |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2808 | |
| 2809 | /* find first rule and skip comments */ |
| 2810 | next = LIST_NEXT(head, struct tcpcheck_rule *, list); |
| 2811 | while (&next->list != head && next->action == TCPCHK_ACT_COMMENT) |
| 2812 | next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list); |
| 2813 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2814 | if ((check->current_step || &next->list == head) && |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 2815 | (conn->flags & CO_FL_WAIT_XPRT)) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2816 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2817 | * to establish but only when timeout.check is set |
| 2818 | * as it may be to short for a full check otherwise |
| 2819 | */ |
| 2820 | while (tick_is_expired(t->expire, now_ms)) { |
| 2821 | int t_con; |
| 2822 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2823 | t_con = tick_add(t->expire, proxy->timeout.connect); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2824 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 2825 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2826 | if (proxy->timeout.check) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2827 | t->expire = tick_first(t->expire, t_con); |
| 2828 | } |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 2829 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2830 | } |
| 2831 | |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2832 | /* special case: option tcp-check with no rule, a connect is enough */ |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2833 | if (&next->list == head) { |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2834 | set_server_check_status(check, HCHK_STATUS_L4OK, NULL); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2835 | goto out_end_tcpcheck; |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2836 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2837 | |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2838 | /* no step means first step initialisation */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2839 | if (check->current_step == NULL) { |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2840 | check->last_started_step = NULL; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2841 | b_reset(&check->bo); |
| 2842 | b_reset(&check->bi); |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2843 | check->current_step = next; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2844 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 2845 | if (proxy->timeout.check) |
| 2846 | t->expire = tick_add_ifset(now_ms, proxy->timeout.check); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2847 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2848 | |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2849 | while (1) { |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2850 | /* We have to try to flush the output buffer before reading, at |
| 2851 | * the end, or if we're about to send a string that does not fit |
| 2852 | * in the remaining space. That explains why we break out of the |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2853 | * loop after this control. If we have data, conn is valid. |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2854 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2855 | if (b_data(&check->bo) && |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2856 | (&check->current_step->list == head || |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2857 | check->current_step->action != TCPCHK_ACT_SEND || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2858 | check->current_step->string_len >= b_room(&check->bo))) { |
Willy Tarreau | deccd11 | 2018-06-14 18:38:55 +0200 | [diff] [blame] | 2859 | int ret; |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2860 | |
Olivier Houchard | ed0f207 | 2018-08-16 15:41:52 +0200 | [diff] [blame] | 2861 | ret = cs->conn->mux->snd_buf(cs, &check->bo, b_data(&check->bo), 0); |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2862 | b_realign_if_empty(&check->bo); |
Willy Tarreau | deccd11 | 2018-06-14 18:38:55 +0200 | [diff] [blame] | 2863 | |
| 2864 | if (ret <= 0) { |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 2865 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2866 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2867 | goto out_end_tcpcheck; |
| 2868 | } |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2869 | break; |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2870 | } |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 2871 | if (b_data(&check->bo)) { |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 2872 | cs->conn->mux->subscribe(cs, SUB_RETRY_SEND, &check->wait_list); |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 2873 | goto out; |
| 2874 | } |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2875 | } |
| 2876 | |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2877 | if (&check->current_step->list == head) |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2878 | break; |
| 2879 | |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2880 | /* have 'next' point to the next rule or NULL if we're on the |
| 2881 | * last one, connect() needs this. |
| 2882 | */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2883 | next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2884 | |
| 2885 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 2886 | while (&next->list != head && next->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2887 | next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2888 | |
| 2889 | /* NULL if we're on the last rule */ |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 2890 | if (&next->list == head) |
| 2891 | next = NULL; |
| 2892 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2893 | if (check->current_step->action == TCPCHK_ACT_CONNECT) { |
| 2894 | struct protocol *proto; |
| 2895 | struct xprt_ops *xprt; |
| 2896 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2897 | /* For a connect action we'll create a new connection. |
| 2898 | * We may also have to kill a previous one. But we don't |
| 2899 | * want to leave *without* a connection if we came here |
| 2900 | * from the connection layer, hence with a connection. |
| 2901 | * Thus we'll proceed in the following order : |
| 2902 | * 1: close but not release previous connection |
| 2903 | * 2: try to get a new connection |
| 2904 | * 3: release and replace the old one on success |
| 2905 | */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2906 | if (check->cs) { |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 2907 | cs_close(check->cs); |
| 2908 | retcode = -1; /* do not reuse the fd in the caller! */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2909 | } |
| 2910 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2911 | /* mark the step as started */ |
| 2912 | check->last_started_step = check->current_step; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2913 | |
| 2914 | /* prepare new connection */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2915 | cs = cs_new(NULL); |
| 2916 | if (!cs) { |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2917 | step = tcpcheck_get_step_id(check); |
| 2918 | chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step); |
| 2919 | comment = tcpcheck_get_step_comment(check, step); |
| 2920 | if (comment) |
| 2921 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2922 | set_server_check_status(check, HCHK_STATUS_SOCKERR, |
| 2923 | trash.area); |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2924 | check->current_step = NULL; |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 2925 | goto out; |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2926 | } |
| 2927 | |
Olivier Houchard | 4906554 | 2019-05-31 19:20:36 +0200 | [diff] [blame] | 2928 | if (check->cs) { |
| 2929 | if (check->wait_list.events) |
| 2930 | cs->conn->xprt->unsubscribe(cs->conn, |
| 2931 | cs->conn->xprt_ctx, |
| 2932 | check->wait_list.events, |
| 2933 | &check->wait_list); |
Willy Tarreau | 86eded6 | 2019-06-14 14:47:49 +0200 | [diff] [blame] | 2934 | /* We may have been scheduled to run, and the |
| 2935 | * I/O handler expects to have a cs, so remove |
| 2936 | * the tasklet |
| 2937 | */ |
| 2938 | tasklet_remove_from_tasklet_list(check->wait_list.tasklet); |
Willy Tarreau | ac59f36 | 2017-10-08 11:10:19 +0200 | [diff] [blame] | 2939 | cs_destroy(check->cs); |
Olivier Houchard | 4906554 | 2019-05-31 19:20:36 +0200 | [diff] [blame] | 2940 | } |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2941 | |
Olivier Houchard | ff1e9f3 | 2019-09-20 17:18:35 +0200 | [diff] [blame] | 2942 | tasklet_set_tid(check->wait_list.tasklet, tid); |
| 2943 | |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2944 | check->cs = cs; |
| 2945 | conn = cs->conn; |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 2946 | /* Maybe there were an older connection we were waiting on */ |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 2947 | check->wait_list.events = 0; |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 2948 | conn->target = s ? &s->obj_type : &proxy->obj_type; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2949 | |
| 2950 | /* no client address */ |
Willy Tarreau | ca79f59 | 2019-07-17 19:04:47 +0200 | [diff] [blame] | 2951 | |
| 2952 | if (!sockaddr_alloc(&conn->dst)) { |
| 2953 | ret = SF_ERR_RESOURCE; |
| 2954 | goto fail_check; |
| 2955 | } |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2956 | |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 2957 | if (is_addr(&check->addr)) { |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2958 | /* we'll connect to the check addr specified on the server */ |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2959 | *conn->dst = check->addr; |
Willy Tarreau | 640556c | 2014-05-09 23:38:15 +0200 | [diff] [blame] | 2960 | } |
| 2961 | else { |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2962 | /* we'll connect to the addr on the server */ |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2963 | *conn->dst = s->addr; |
Willy Tarreau | 640556c | 2014-05-09 23:38:15 +0200 | [diff] [blame] | 2964 | } |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2965 | proto = protocol_by_family(conn->dst->ss_family); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2966 | |
| 2967 | /* port */ |
| 2968 | if (check->current_step->port) |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2969 | set_host_port(conn->dst, check->current_step->port); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2970 | else if (check->port) |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2971 | set_host_port(conn->dst, check->port); |
Baptiste Assmann | 2f3a56b | 2018-03-19 12:22:41 +0100 | [diff] [blame] | 2972 | else if (s->svc_port) |
Willy Tarreau | b3c81cb | 2019-07-17 16:54:52 +0200 | [diff] [blame] | 2973 | set_host_port(conn->dst, s->svc_port); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2974 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2975 | if (check->current_step->conn_opts & TCPCHK_OPT_SSL) { |
Willy Tarreau | a261e9b | 2016-12-22 20:44:00 +0100 | [diff] [blame] | 2976 | xprt = xprt_get(XPRT_SSL); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2977 | } |
| 2978 | else { |
Willy Tarreau | a261e9b | 2016-12-22 20:44:00 +0100 | [diff] [blame] | 2979 | xprt = xprt_get(XPRT_RAW); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2980 | } |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 2981 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2982 | conn_prepare(conn, proto, xprt); |
Willy Tarreau | 2ab5c38 | 2019-07-17 18:48:07 +0200 | [diff] [blame] | 2983 | |
| 2984 | if (conn_install_mux(conn, &mux_pt_ops, cs, proxy, NULL) < 0) { |
| 2985 | ret = SF_ERR_RESOURCE; |
| 2986 | goto fail_check; |
| 2987 | } |
| 2988 | |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 2989 | cs_attach(cs, check, &check_conn_cb); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2990 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2991 | ret = SF_ERR_INTERNAL; |
Gaetan Rivet | 08fdcb3 | 2020-02-28 11:04:21 +0100 | [diff] [blame] | 2992 | if (proto && proto->connect) { |
| 2993 | int flags; |
| 2994 | |
| 2995 | flags = CONNECT_HAS_DATA; |
| 2996 | if (next && next->action != TCPCHK_ACT_EXPECT) |
| 2997 | flags |= CONNECT_DELACK_ALWAYS; |
| 2998 | ret = proto->connect(conn, flags); |
| 2999 | } |
Olivier Houchard | 37d7897 | 2019-12-30 15:13:42 +0100 | [diff] [blame] | 3000 | if (conn_ctrl_ready(conn) && |
| 3001 | check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) { |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3002 | conn->send_proxy_ofs = 1; |
| 3003 | conn->flags |= CO_FL_SEND_PROXY; |
Olivier Houchard | fe50bfb | 2019-05-27 12:09:19 +0200 | [diff] [blame] | 3004 | if (xprt_add_hs(conn) < 0) |
| 3005 | ret = SF_ERR_RESOURCE; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3006 | } |
| 3007 | |
Gaetan Rivet | f8ba677 | 2020-02-07 15:37:17 +0100 | [diff] [blame] | 3008 | if (conn_ctrl_ready(conn) && |
| 3009 | check->current_step->conn_opts & TCPCHK_OPT_LINGER) { |
| 3010 | /* Some servers don't like reset on close */ |
| 3011 | fdtab[cs->conn->handle.fd].linger_risk = 0; |
| 3012 | } |
| 3013 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3014 | /* It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 3015 | * - SF_ERR_NONE if everything's OK |
| 3016 | * - SF_ERR_SRVTO if there are no more servers |
| 3017 | * - SF_ERR_SRVCL if the connection was refused by the server |
| 3018 | * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn) |
| 3019 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 3020 | * - SF_ERR_INTERNAL for any other purely internal errors |
Tim Düsterhus | 4896c44 | 2016-11-29 02:15:19 +0100 | [diff] [blame] | 3021 | * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted. |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3022 | * Note that we try to prevent the network stack from sending the ACK during the |
| 3023 | * connect() when a pure TCP check is used (without PROXY protocol). |
| 3024 | */ |
Willy Tarreau | 2ab5c38 | 2019-07-17 18:48:07 +0200 | [diff] [blame] | 3025 | fail_check: |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3026 | switch (ret) { |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 3027 | case SF_ERR_NONE: |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3028 | /* we allow up to min(inter, timeout.connect) for a connection |
| 3029 | * to establish but only when timeout.check is set |
| 3030 | * as it may be to short for a full check otherwise |
| 3031 | */ |
| 3032 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 3033 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 3034 | if (proxy->timeout.check && proxy->timeout.connect) { |
| 3035 | int t_con = tick_add(now_ms, proxy->timeout.connect); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3036 | t->expire = tick_first(t->expire, t_con); |
| 3037 | } |
| 3038 | break; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 3039 | case SF_ERR_SRVTO: /* ETIMEDOUT */ |
| 3040 | case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3041 | step = tcpcheck_get_step_id(check); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3042 | chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s", |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3043 | step, strerror(errno)); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3044 | comment = tcpcheck_get_step_comment(check, step); |
| 3045 | if (comment) |
| 3046 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3047 | set_server_check_status(check, HCHK_STATUS_L4CON, |
| 3048 | trash.area); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3049 | goto out_end_tcpcheck; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 3050 | case SF_ERR_PRXCOND: |
| 3051 | case SF_ERR_RESOURCE: |
| 3052 | case SF_ERR_INTERNAL: |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3053 | step = tcpcheck_get_step_id(check); |
| 3054 | chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3055 | comment = tcpcheck_get_step_comment(check, step); |
| 3056 | if (comment) |
| 3057 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3058 | set_server_check_status(check, HCHK_STATUS_SOCKERR, |
| 3059 | trash.area); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3060 | goto out_end_tcpcheck; |
| 3061 | } |
| 3062 | |
| 3063 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3064 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3065 | |
| 3066 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3067 | while (&check->current_step->list != head && |
| 3068 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3069 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3070 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3071 | if (&check->current_step->list == head) |
| 3072 | break; |
| 3073 | |
Willy Tarreau | 7df8ca6 | 2019-07-15 10:57:51 +0200 | [diff] [blame] | 3074 | /* don't do anything until the connection is established */ |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 3075 | if (conn->flags & CO_FL_WAIT_XPRT) |
Willy Tarreau | 7df8ca6 | 2019-07-15 10:57:51 +0200 | [diff] [blame] | 3076 | break; |
| 3077 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3078 | } /* end 'connect' */ |
| 3079 | else if (check->current_step->action == TCPCHK_ACT_SEND) { |
| 3080 | /* mark the step as started */ |
| 3081 | check->last_started_step = check->current_step; |
| 3082 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3083 | /* reset the read buffer */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3084 | if (*b_head(&check->bi) != '\0') { |
| 3085 | *b_head(&check->bi) = '\0'; |
| 3086 | b_reset(&check->bi); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3087 | } |
| 3088 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3089 | if (check->current_step->string_len >= b_size(&check->bo)) { |
Willy Tarreau | 506a29a | 2018-07-18 10:07:58 +0200 | [diff] [blame] | 3090 | chunk_printf(&trash, "tcp-check send : string too large (%d) for buffer size (%u) at step %d", |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3091 | check->current_step->string_len, (unsigned int)b_size(&check->bo), |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 3092 | tcpcheck_get_step_id(check)); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3093 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3094 | trash.area); |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 3095 | goto out_end_tcpcheck; |
| 3096 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3097 | |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3098 | /* do not try to send if there is no space */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3099 | if (check->current_step->string_len >= b_room(&check->bo)) |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3100 | continue; |
| 3101 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3102 | b_putblk(&check->bo, check->current_step->string, check->current_step->string_len); |
| 3103 | *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3104 | |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 3105 | /* go to next rule and try to send */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3106 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3107 | |
| 3108 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3109 | while (&check->current_step->list != head && |
| 3110 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3111 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3112 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3113 | } /* end 'send' */ |
Willy Tarreau | 98aec9f | 2013-12-06 16:16:41 +0100 | [diff] [blame] | 3114 | else if (check->current_step->action == TCPCHK_ACT_EXPECT) { |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3115 | struct tcpcheck_expect *expect = &check->current_step->expect; |
| 3116 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 3117 | if (unlikely(check->result == CHK_RES_FAILED)) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3118 | goto out_end_tcpcheck; |
| 3119 | |
Olivier Houchard | 0ba6c85 | 2019-07-18 15:09:08 +0200 | [diff] [blame] | 3120 | /* If we already subscribed, then we tried to received |
| 3121 | * and failed, so there's no point trying again. |
| 3122 | */ |
| 3123 | if (check->wait_list.events & SUB_RETRY_RECV) |
| 3124 | break; |
Olivier Houchard | 511efea | 2018-08-16 15:30:32 +0200 | [diff] [blame] | 3125 | if (cs->conn->mux->rcv_buf(cs, &check->bi, b_size(&check->bi), 0) <= 0) { |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 3126 | if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH) || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3127 | done = 1; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3128 | if ((conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) && !b_data(&check->bi)) { |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3129 | /* Report network errors only if we got no other data. Otherwise |
| 3130 | * we'll let the upper layers decide whether the response is OK |
| 3131 | * or not. It is very common that an RST sent by the server is |
| 3132 | * reported as an error just after the last data chunk. |
| 3133 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 3134 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3135 | goto out_end_tcpcheck; |
| 3136 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3137 | } |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 3138 | else { |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 3139 | conn->mux->subscribe(cs, SUB_RETRY_RECV, &check->wait_list); |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 3140 | break; |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 3141 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3142 | } |
| 3143 | |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3144 | /* Having received new data, reset the expect chain to its head. */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3145 | check->current_step = expect->head; |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3146 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3147 | /* mark the step as started */ |
| 3148 | check->last_started_step = check->current_step; |
| 3149 | |
| 3150 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3151 | /* Intermediate or complete response received. |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3152 | * Terminate string in b_head(&check->bi) buffer. |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3153 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3154 | if (b_data(&check->bi) < b_size(&check->bi)) { |
| 3155 | b_head(&check->bi)[b_data(&check->bi)] = '\0'; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3156 | } |
| 3157 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3158 | b_head(&check->bi)[b_data(&check->bi) - 1] = '\0'; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3159 | done = 1; /* buffer full, don't wait for more data */ |
| 3160 | } |
| 3161 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3162 | contentptr = b_head(&check->bi); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3163 | |
| 3164 | /* Check that response body is not empty... */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3165 | if (!b_data(&check->bi)) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3166 | if (!done) |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3167 | continue; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3168 | |
| 3169 | /* empty response */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3170 | step = tcpcheck_get_step_id(check); |
| 3171 | chunk_printf(&trash, "TCPCHK got an empty response at step %d", step); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3172 | comment = tcpcheck_get_step_comment(check, step); |
| 3173 | if (comment) |
| 3174 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3175 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3176 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3177 | |
| 3178 | goto out_end_tcpcheck; |
| 3179 | } |
| 3180 | |
Gaetan Rivet | 738ee76 | 2020-02-13 10:30:01 +0100 | [diff] [blame] | 3181 | tcpcheck_expect: |
Gaetan Rivet | 1afd826 | 2020-02-07 15:37:17 +0100 | [diff] [blame] | 3182 | /* The current expect might need more data than the previous one, check again |
| 3183 | * that the minimum amount data required to match is respected. |
| 3184 | */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3185 | if (!done) { |
| 3186 | if ((expect->type == TCPCHK_EXPECT_STRING || expect->type == TCPCHK_EXPECT_BINARY) && |
| 3187 | (b_data(&check->bi) < expect->length)) |
| 3188 | continue; /* try to read more */ |
| 3189 | if (expect->min_recv > 0 && (b_data(&check->bi) < expect->min_recv)) |
| 3190 | continue; /* try to read more */ |
| 3191 | } |
| 3192 | |
| 3193 | switch (expect->type) { |
| 3194 | case TCPCHK_EXPECT_STRING: |
| 3195 | case TCPCHK_EXPECT_BINARY: |
| 3196 | ret = my_memmem(contentptr, b_data(&check->bi), expect->string, expect->length) != NULL; |
| 3197 | break; |
| 3198 | case TCPCHK_EXPECT_REGEX: |
| 3199 | ret = regex_exec(expect->regex, contentptr); |
| 3200 | break; |
| 3201 | case TCPCHK_EXPECT_UNDEF: |
| 3202 | /* Should never happen. */ |
| 3203 | retcode = -1; |
| 3204 | goto out; |
| 3205 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3206 | |
Gaetan Rivet | 1afd826 | 2020-02-07 15:37:17 +0100 | [diff] [blame] | 3207 | /* Wait for more data on mismatch only if no minimum is defined (-1), |
| 3208 | * otherwise the absence of match is already conclusive. |
| 3209 | */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3210 | if (!ret && !done && (expect->min_recv == -1)) |
Willy Tarreau | a970c28 | 2013-12-06 12:47:19 +0100 | [diff] [blame] | 3211 | continue; /* try to read more */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3212 | |
| 3213 | /* matched */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3214 | step = tcpcheck_get_step_id(check); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3215 | if (ret) { |
| 3216 | /* matched but we did not want to => ERROR */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3217 | if (expect->inverse) { |
| 3218 | switch (expect->type) { |
| 3219 | case TCPCHK_EXPECT_STRING: |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3220 | chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d", |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3221 | expect->string, step); |
| 3222 | break; |
| 3223 | case TCPCHK_EXPECT_BINARY: |
| 3224 | chunk_printf(&trash, "TCPCHK matched unwanted content (binary) at step %d", |
| 3225 | step); |
| 3226 | break; |
| 3227 | case TCPCHK_EXPECT_REGEX: |
| 3228 | chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", |
| 3229 | step); |
| 3230 | break; |
| 3231 | case TCPCHK_EXPECT_UNDEF: |
| 3232 | /* Should never happen. */ |
| 3233 | retcode = -1; |
| 3234 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3235 | } |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3236 | |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3237 | comment = tcpcheck_get_step_comment(check, step); |
| 3238 | if (comment) |
| 3239 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3240 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3241 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3242 | goto out_end_tcpcheck; |
| 3243 | } |
| 3244 | /* matched and was supposed to => OK, next step */ |
| 3245 | else { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3246 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3247 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3248 | |
| 3249 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3250 | while (&check->current_step->list != head && |
| 3251 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3252 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3253 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3254 | if (&check->current_step->list == head) |
| 3255 | break; |
| 3256 | |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3257 | if (check->current_step->action == TCPCHK_ACT_EXPECT) { |
| 3258 | expect = &check->current_step->expect; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3259 | goto tcpcheck_expect; |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3260 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3261 | } |
| 3262 | } |
| 3263 | else { |
| 3264 | /* not matched */ |
| 3265 | /* not matched and was not supposed to => OK, next step */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3266 | if (expect->inverse) { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3267 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3268 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3269 | |
| 3270 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3271 | while (&check->current_step->list != head && |
| 3272 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3273 | check->current_step = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3274 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3275 | if (&check->current_step->list == head) |
| 3276 | break; |
| 3277 | |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3278 | if (check->current_step->action == TCPCHK_ACT_EXPECT) { |
| 3279 | expect = &check->current_step->expect; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3280 | goto tcpcheck_expect; |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3281 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3282 | } |
| 3283 | /* not matched but was supposed to => ERROR */ |
| 3284 | else { |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3285 | switch (expect->type) { |
| 3286 | case TCPCHK_EXPECT_STRING: |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3287 | chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d", |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 3288 | check->current_step->string, step); |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3289 | break; |
| 3290 | case TCPCHK_EXPECT_BINARY: |
| 3291 | chunk_printf(&trash, "TCPCHK did not match content (binary) at step %d", |
| 3292 | step); |
| 3293 | break; |
| 3294 | case TCPCHK_EXPECT_REGEX: |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3295 | chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d", |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3296 | step); |
| 3297 | break; |
| 3298 | case TCPCHK_EXPECT_UNDEF: |
| 3299 | /* Should never happen. */ |
| 3300 | retcode = -1; |
| 3301 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3302 | } |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3303 | |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3304 | comment = tcpcheck_get_step_comment(check, step); |
| 3305 | if (comment) |
| 3306 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3307 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3308 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3309 | goto out_end_tcpcheck; |
| 3310 | } |
| 3311 | } |
| 3312 | } /* end expect */ |
| 3313 | } /* end loop over double chained step list */ |
| 3314 | |
Baptiste Assmann | 248f117 | 2018-03-01 21:49:01 +0100 | [diff] [blame] | 3315 | /* don't do anything until the connection is established */ |
Willy Tarreau | 911db9b | 2020-01-23 16:27:54 +0100 | [diff] [blame] | 3316 | if (conn->flags & CO_FL_WAIT_XPRT) { |
Baptiste Assmann | 248f117 | 2018-03-01 21:49:01 +0100 | [diff] [blame] | 3317 | /* update expire time, should be done by process_chk */ |
| 3318 | /* we allow up to min(inter, timeout.connect) for a connection |
| 3319 | * to establish but only when timeout.check is set |
| 3320 | * as it may be to short for a full check otherwise |
| 3321 | */ |
| 3322 | while (tick_is_expired(t->expire, now_ms)) { |
| 3323 | int t_con; |
| 3324 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 3325 | t_con = tick_add(t->expire, proxy->timeout.connect); |
Baptiste Assmann | 248f117 | 2018-03-01 21:49:01 +0100 | [diff] [blame] | 3326 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 3327 | |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 3328 | if (proxy->timeout.check) |
Baptiste Assmann | 248f117 | 2018-03-01 21:49:01 +0100 | [diff] [blame] | 3329 | t->expire = tick_first(t->expire, t_con); |
| 3330 | } |
| 3331 | goto out; |
| 3332 | } |
| 3333 | |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 3334 | /* We're waiting for some I/O to complete, we've reached the end of the |
| 3335 | * rules, or both. Do what we have to do, otherwise we're done. |
| 3336 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3337 | if (&check->current_step->list == head && !b_data(&check->bo)) { |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 3338 | set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)"); |
| 3339 | goto out_end_tcpcheck; |
| 3340 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3341 | |
Willy Tarreau | 53c5a04 | 2015-05-13 11:38:17 +0200 | [diff] [blame] | 3342 | if (&check->current_step->list != head && |
Olivier Houchard | 53216e7 | 2018-10-10 15:46:36 +0200 | [diff] [blame] | 3343 | check->current_step->action == TCPCHK_ACT_EXPECT) |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 3344 | __event_srv_chk_r(cs); |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 3345 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3346 | |
| 3347 | out_end_tcpcheck: |
| 3348 | /* collect possible new errors */ |
Willy Tarreau | ef91c93 | 2019-07-23 14:37:47 +0200 | [diff] [blame] | 3349 | if ((conn && conn->flags & CO_FL_ERROR) || (cs && cs->flags & CS_FL_ERROR)) |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 3350 | chk_report_conn_err(check, 0, 0); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3351 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3352 | /* cleanup before leaving */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3353 | check->current_step = NULL; |
| 3354 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 3355 | if (check->result == CHK_RES_FAILED) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3356 | conn->flags |= CO_FL_ERROR; |
| 3357 | |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 3358 | out: |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 3359 | return retcode; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3360 | } |
| 3361 | |
Christopher Faulet | 31c30fd | 2020-03-26 21:10:03 +0100 | [diff] [blame] | 3362 | static const char *init_check(struct check *check, int type) |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3363 | { |
| 3364 | check->type = type; |
| 3365 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3366 | b_reset(&check->bi); check->bi.size = global.tune.chksize; |
| 3367 | b_reset(&check->bo); check->bo.size = global.tune.chksize; |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3368 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3369 | check->bi.area = calloc(check->bi.size, sizeof(char)); |
| 3370 | check->bo.area = calloc(check->bo.size, sizeof(char)); |
| 3371 | |
| 3372 | if (!check->bi.area || !check->bo.area) |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3373 | return "out of memory while allocating check buffer"; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3374 | |
Willy Tarreau | 3c39a7d | 2019-06-14 14:42:29 +0200 | [diff] [blame] | 3375 | check->wait_list.tasklet = tasklet_new(); |
| 3376 | if (!check->wait_list.tasklet) |
Ilya Shipitsin | d425950 | 2020-04-08 01:07:56 +0500 | [diff] [blame] | 3377 | return "out of memory while allocating check tasklet"; |
Willy Tarreau | 4f6516d | 2018-12-19 13:59:17 +0100 | [diff] [blame] | 3378 | check->wait_list.events = 0; |
Willy Tarreau | 3c39a7d | 2019-06-14 14:42:29 +0200 | [diff] [blame] | 3379 | check->wait_list.tasklet->process = event_srv_chk_io; |
| 3380 | check->wait_list.tasklet->context = check; |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3381 | return NULL; |
| 3382 | } |
| 3383 | |
Simon Horman | bfb5d33 | 2015-01-30 11:22:55 +0900 | [diff] [blame] | 3384 | void free_check(struct check *check) |
| 3385 | { |
Christopher Faulet | 8892e5d | 2020-03-26 19:48:20 +0100 | [diff] [blame] | 3386 | task_destroy(check->task); |
| 3387 | if (check->wait_list.tasklet) |
| 3388 | tasklet_free(check->wait_list.tasklet); |
| 3389 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3390 | free(check->bi.area); |
| 3391 | free(check->bo.area); |
Christopher Faulet | 23d86d1 | 2018-01-25 11:36:35 +0100 | [diff] [blame] | 3392 | if (check->cs) { |
| 3393 | free(check->cs->conn); |
| 3394 | check->cs->conn = NULL; |
| 3395 | cs_free(check->cs); |
| 3396 | check->cs = NULL; |
| 3397 | } |
Simon Horman | bfb5d33 | 2015-01-30 11:22:55 +0900 | [diff] [blame] | 3398 | } |
| 3399 | |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3400 | void email_alert_free(struct email_alert *alert) |
| 3401 | { |
| 3402 | struct tcpcheck_rule *rule, *back; |
| 3403 | |
| 3404 | if (!alert) |
| 3405 | return; |
| 3406 | |
Christopher Faulet | de1a75b | 2017-10-23 15:38:19 +0200 | [diff] [blame] | 3407 | list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) { |
| 3408 | LIST_DEL(&rule->list); |
| 3409 | free(rule->comment); |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3410 | switch (rule->expect.type) { |
| 3411 | case TCPCHK_EXPECT_STRING: |
| 3412 | case TCPCHK_EXPECT_BINARY: |
| 3413 | free(rule->expect.string); |
| 3414 | break; |
| 3415 | case TCPCHK_EXPECT_REGEX: |
| 3416 | regex_free(rule->expect.regex); |
| 3417 | break; |
| 3418 | case TCPCHK_EXPECT_UNDEF: |
| 3419 | break; |
| 3420 | } |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3421 | pool_free(pool_head_tcpcheck_rule, rule); |
Christopher Faulet | de1a75b | 2017-10-23 15:38:19 +0200 | [diff] [blame] | 3422 | } |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3423 | pool_free(pool_head_email_alert, alert); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3424 | } |
| 3425 | |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3426 | static struct task *process_email_alert(struct task *t, void *context, unsigned short state) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3427 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3428 | struct check *check = context; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3429 | struct email_alertq *q; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3430 | struct email_alert *alert; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3431 | |
| 3432 | q = container_of(check, typeof(*q), check); |
| 3433 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3434 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3435 | while (1) { |
| 3436 | if (!(check->state & CHK_ST_ENABLED)) { |
| 3437 | if (LIST_ISEMPTY(&q->email_alerts)) { |
| 3438 | /* All alerts processed, queue the task */ |
| 3439 | t->expire = TICK_ETERNITY; |
| 3440 | task_queue(t); |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 3441 | goto end; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3442 | } |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3443 | |
| 3444 | alert = LIST_NEXT(&q->email_alerts, typeof(alert), list); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3445 | LIST_DEL(&alert->list); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3446 | t->expire = now_ms; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3447 | check->tcpcheck_rules = &alert->tcpcheck_rules; |
Olivier Houchard | 0923fa4 | 2019-01-11 18:43:04 +0100 | [diff] [blame] | 3448 | check->status = HCHK_STATUS_INI; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3449 | check->state |= CHK_ST_ENABLED; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3450 | } |
| 3451 | |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3452 | process_chk(t, context, state); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3453 | if (check->state & CHK_ST_INPROGRESS) |
| 3454 | break; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3455 | |
| 3456 | alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules); |
| 3457 | email_alert_free(alert); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3458 | check->tcpcheck_rules = NULL; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3459 | check->server = NULL; |
| 3460 | check->state &= ~CHK_ST_ENABLED; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3461 | } |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 3462 | end: |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3463 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3464 | return t; |
| 3465 | } |
| 3466 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3467 | /* Initializes mailer alerts for the proxy <p> using <mls> parameters. |
| 3468 | * |
| 3469 | * The function returns 1 in success case, otherwise, it returns 0 and err is |
| 3470 | * filled. |
| 3471 | */ |
| 3472 | int init_email_alert(struct mailers *mls, struct proxy *p, char **err) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3473 | { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3474 | struct mailer *mailer; |
| 3475 | struct email_alertq *queues; |
| 3476 | const char *err_str; |
| 3477 | int i = 0; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3478 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3479 | if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) { |
| 3480 | memprintf(err, "out of memory while allocating mailer alerts queues"); |
mildis | 5ab01cb | 2018-10-02 16:46:34 +0200 | [diff] [blame] | 3481 | goto fail_no_queue; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3482 | } |
| 3483 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3484 | for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) { |
| 3485 | struct email_alertq *q = &queues[i]; |
| 3486 | struct check *check = &q->check; |
| 3487 | struct task *t; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3488 | |
| 3489 | LIST_INIT(&q->email_alerts); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3490 | HA_SPIN_INIT(&q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3491 | check->inter = mls->timeout.mail; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3492 | check->rise = DEF_AGENT_RISETIME; |
Olivier Houchard | c98aa1f | 2019-01-11 18:17:17 +0100 | [diff] [blame] | 3493 | check->proxy = p; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3494 | check->fall = DEF_AGENT_FALLTIME; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3495 | if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) { |
| 3496 | memprintf(err, "%s", err_str); |
| 3497 | goto error; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3498 | } |
| 3499 | |
| 3500 | check->xprt = mailer->xprt; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3501 | check->addr = mailer->addr; |
Christopher Faulet | b797ae1 | 2018-03-27 15:35:35 +0200 | [diff] [blame] | 3502 | check->port = get_host_port(&mailer->addr); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3503 | |
Emeric Brun | c60def8 | 2017-09-27 14:59:38 +0200 | [diff] [blame] | 3504 | if ((t = task_new(MAX_THREADS_MASK)) == NULL) { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3505 | memprintf(err, "out of memory while allocating mailer alerts task"); |
| 3506 | goto error; |
| 3507 | } |
| 3508 | |
| 3509 | check->task = t; |
| 3510 | t->process = process_email_alert; |
| 3511 | t->context = check; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3512 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3513 | /* check this in one ms */ |
| 3514 | t->expire = TICK_ETERNITY; |
| 3515 | check->start = now; |
| 3516 | task_queue(t); |
| 3517 | } |
| 3518 | |
| 3519 | mls->users++; |
| 3520 | free(p->email_alert.mailers.name); |
| 3521 | p->email_alert.mailers.m = mls; |
| 3522 | p->email_alert.queues = queues; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3523 | return 0; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3524 | |
| 3525 | error: |
| 3526 | for (i = 0; i < mls->count; i++) { |
| 3527 | struct email_alertq *q = &queues[i]; |
| 3528 | struct check *check = &q->check; |
| 3529 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3530 | free_check(check); |
| 3531 | } |
| 3532 | free(queues); |
mildis | 5ab01cb | 2018-10-02 16:46:34 +0200 | [diff] [blame] | 3533 | fail_no_queue: |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3534 | return 1; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3535 | } |
| 3536 | |
| 3537 | |
| 3538 | static int add_tcpcheck_expect_str(struct list *list, const char *str) |
| 3539 | { |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3540 | struct tcpcheck_rule *tcpcheck, *prev_check; |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3541 | struct tcpcheck_expect *expect; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3542 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3543 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3544 | return 0; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3545 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3546 | tcpcheck->action = TCPCHK_ACT_EXPECT; |
| 3547 | |
| 3548 | expect = &tcpcheck->expect; |
| 3549 | expect->type = TCPCHK_EXPECT_STRING; |
| 3550 | expect->string = strdup(str); |
| 3551 | if (!expect->string) { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3552 | pool_free(pool_head_tcpcheck_rule, tcpcheck); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3553 | return 0; |
| 3554 | } |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3555 | expect->length = strlen(expect->string); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3556 | |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3557 | /* All tcp-check expect points back to the first inverse expect rule |
| 3558 | * in a chain of one or more expect rule, potentially itself. |
| 3559 | */ |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3560 | tcpcheck->expect.head = tcpcheck; |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3561 | list_for_each_entry_rev(prev_check, list, list) { |
| 3562 | if (prev_check->action == TCPCHK_ACT_EXPECT) { |
Gaetan Rivet | b616add | 2020-02-07 15:37:17 +0100 | [diff] [blame^] | 3563 | if (prev_check->expect.inverse) |
| 3564 | tcpcheck->expect.head = prev_check; |
Gaetan Rivet | 4038b94 | 2020-02-26 16:19:40 +0100 | [diff] [blame] | 3565 | continue; |
| 3566 | } |
| 3567 | if (prev_check->action != TCPCHK_ACT_COMMENT) |
| 3568 | break; |
| 3569 | } |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3570 | LIST_ADDQ(list, &tcpcheck->list); |
| 3571 | return 1; |
| 3572 | } |
| 3573 | |
| 3574 | static int add_tcpcheck_send_strs(struct list *list, const char * const *strs) |
| 3575 | { |
| 3576 | struct tcpcheck_rule *tcpcheck; |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3577 | const char *in; |
| 3578 | char *dst; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3579 | int i; |
| 3580 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3581 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3582 | return 0; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3583 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
| 3584 | tcpcheck->action = TCPCHK_ACT_SEND; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3585 | tcpcheck->comment = NULL; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3586 | tcpcheck->string_len = 0; |
| 3587 | for (i = 0; strs[i]; i++) |
| 3588 | tcpcheck->string_len += strlen(strs[i]); |
| 3589 | |
| 3590 | tcpcheck->string = malloc(tcpcheck->string_len + 1); |
| 3591 | if (!tcpcheck->string) { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3592 | pool_free(pool_head_tcpcheck_rule, tcpcheck); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3593 | return 0; |
| 3594 | } |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3595 | |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3596 | dst = tcpcheck->string; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3597 | for (i = 0; strs[i]; i++) |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3598 | for (in = strs[i]; (*dst = *in++); dst++); |
| 3599 | *dst = 0; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3600 | |
| 3601 | LIST_ADDQ(list, &tcpcheck->list); |
| 3602 | return 1; |
| 3603 | } |
| 3604 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3605 | static int enqueue_one_email_alert(struct proxy *p, struct server *s, |
| 3606 | struct email_alertq *q, const char *msg) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3607 | { |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3608 | struct email_alert *alert; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3609 | struct tcpcheck_rule *tcpcheck; |
| 3610 | struct check *check = &q->check; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3611 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3612 | if ((alert = pool_alloc(pool_head_email_alert)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3613 | goto error; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3614 | LIST_INIT(&alert->list); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3615 | LIST_INIT(&alert->tcpcheck_rules); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3616 | alert->srv = s; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3617 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3618 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3619 | goto error; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3620 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
| 3621 | tcpcheck->action = TCPCHK_ACT_CONNECT; |
| 3622 | tcpcheck->comment = NULL; |
| 3623 | tcpcheck->string = NULL; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3624 | LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list); |
| 3625 | |
| 3626 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 ")) |
| 3627 | goto error; |
| 3628 | |
| 3629 | { |
| 3630 | const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" }; |
| 3631 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3632 | goto error; |
| 3633 | } |
| 3634 | |
| 3635 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3636 | goto error; |
| 3637 | |
| 3638 | { |
| 3639 | const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" }; |
| 3640 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3641 | goto error; |
| 3642 | } |
| 3643 | |
| 3644 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3645 | goto error; |
| 3646 | |
| 3647 | { |
| 3648 | const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" }; |
| 3649 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3650 | goto error; |
| 3651 | } |
| 3652 | |
| 3653 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3654 | goto error; |
| 3655 | |
| 3656 | { |
| 3657 | const char * const strs[2] = { "DATA\r\n" }; |
| 3658 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3659 | goto error; |
| 3660 | } |
| 3661 | |
| 3662 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 ")) |
| 3663 | goto error; |
| 3664 | |
| 3665 | { |
| 3666 | struct tm tm; |
| 3667 | char datestr[48]; |
| 3668 | const char * const strs[18] = { |
Pieter Baauw | 5e0964e | 2016-02-13 16:27:35 +0100 | [diff] [blame] | 3669 | "From: ", p->email_alert.from, "\r\n", |
| 3670 | "To: ", p->email_alert.to, "\r\n", |
| 3671 | "Date: ", datestr, "\r\n", |
| 3672 | "Subject: [HAproxy Alert] ", msg, "\r\n", |
| 3673 | "\r\n", |
| 3674 | msg, "\r\n", |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3675 | "\r\n", |
Pieter Baauw | ed35c37 | 2015-07-22 19:51:54 +0200 | [diff] [blame] | 3676 | ".\r\n", |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3677 | NULL |
| 3678 | }; |
| 3679 | |
| 3680 | get_localtime(date.tv_sec, &tm); |
| 3681 | |
| 3682 | if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) { |
| 3683 | goto error; |
| 3684 | } |
| 3685 | |
| 3686 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3687 | goto error; |
| 3688 | } |
| 3689 | |
| 3690 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3691 | goto error; |
| 3692 | |
| 3693 | { |
| 3694 | const char * const strs[2] = { "QUIT\r\n" }; |
| 3695 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3696 | goto error; |
| 3697 | } |
| 3698 | |
| 3699 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 ")) |
| 3700 | goto error; |
| 3701 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3702 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3703 | task_wakeup(check->task, TASK_WOKEN_MSG); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3704 | LIST_ADDQ(&q->email_alerts, &alert->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3705 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3706 | return 1; |
| 3707 | |
| 3708 | error: |
| 3709 | email_alert_free(alert); |
| 3710 | return 0; |
| 3711 | } |
| 3712 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3713 | static void enqueue_email_alert(struct proxy *p, struct server *s, const char *msg) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3714 | { |
| 3715 | int i; |
| 3716 | struct mailer *mailer; |
| 3717 | |
| 3718 | for (i = 0, mailer = p->email_alert.mailers.m->mailer_list; |
| 3719 | i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3720 | if (!enqueue_one_email_alert(p, s, &p->email_alert.queues[i], msg)) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 3721 | ha_alert("Email alert [%s] could not be enqueued: out of memory\n", p->id); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3722 | return; |
| 3723 | } |
| 3724 | } |
| 3725 | |
| 3726 | return; |
| 3727 | } |
| 3728 | |
| 3729 | /* |
| 3730 | * Send email alert if configured. |
| 3731 | */ |
Simon Horman | 64e3416 | 2015-02-06 11:11:57 +0900 | [diff] [blame] | 3732 | void send_email_alert(struct server *s, int level, const char *format, ...) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3733 | { |
| 3734 | va_list argp; |
| 3735 | char buf[1024]; |
| 3736 | int len; |
| 3737 | struct proxy *p = s->proxy; |
| 3738 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3739 | if (!p->email_alert.mailers.m || level > p->email_alert.level || format == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3740 | return; |
| 3741 | |
| 3742 | va_start(argp, format); |
| 3743 | len = vsnprintf(buf, sizeof(buf), format, argp); |
| 3744 | va_end(argp); |
| 3745 | |
Thierry FOURNIER | 62c8a21 | 2017-02-09 12:19:27 +0100 | [diff] [blame] | 3746 | if (len < 0 || len >= sizeof(buf)) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 3747 | ha_alert("Email alert [%s] could not format message\n", p->id); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3748 | return; |
| 3749 | } |
| 3750 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3751 | enqueue_email_alert(p, s, buf); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3752 | } |
| 3753 | |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3754 | /* |
| 3755 | * Return value: |
| 3756 | * the port to be used for the health check |
| 3757 | * 0 in case no port could be found for the check |
| 3758 | */ |
Christopher Faulet | 31c30fd | 2020-03-26 21:10:03 +0100 | [diff] [blame] | 3759 | static int srv_check_healthcheck_port(struct check *chk) |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3760 | { |
| 3761 | int i = 0; |
| 3762 | struct server *srv = NULL; |
| 3763 | |
| 3764 | srv = chk->server; |
| 3765 | |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3766 | /* by default, we use the health check port ocnfigured */ |
| 3767 | if (chk->port > 0) |
| 3768 | return chk->port; |
| 3769 | |
| 3770 | /* try to get the port from check_core.addr if check.port not set */ |
| 3771 | i = get_host_port(&chk->addr); |
| 3772 | if (i > 0) |
| 3773 | return i; |
| 3774 | |
| 3775 | /* try to get the port from server address */ |
| 3776 | /* prevent MAPPORTS from working at this point, since checks could |
| 3777 | * not be performed in such case (MAPPORTS impose a relative ports |
| 3778 | * based on live traffic) |
| 3779 | */ |
| 3780 | if (srv->flags & SRV_F_MAPPORTS) |
| 3781 | return 0; |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 3782 | |
| 3783 | i = srv->svc_port; /* by default */ |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3784 | if (i > 0) |
| 3785 | return i; |
| 3786 | |
| 3787 | return 0; |
| 3788 | } |
| 3789 | |
Willy Tarreau | 172f5ce | 2018-11-26 11:21:50 +0100 | [diff] [blame] | 3790 | REGISTER_POST_CHECK(start_checks); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3791 | |
Christopher Faulet | 8892e5d | 2020-03-26 19:48:20 +0100 | [diff] [blame] | 3792 | static int init_srv_check(struct server *srv) |
| 3793 | { |
| 3794 | const char *err; |
| 3795 | struct tcpcheck_rule *r; |
| 3796 | int ret = 0; |
| 3797 | |
| 3798 | if (!srv->do_check) |
| 3799 | goto out; |
| 3800 | |
| 3801 | |
| 3802 | /* If neither a port nor an addr was specified and no check transport |
| 3803 | * layer is forced, then the transport layer used by the checks is the |
| 3804 | * same as for the production traffic. Otherwise we use raw_sock by |
| 3805 | * default, unless one is specified. |
| 3806 | */ |
| 3807 | if (!srv->check.port && !is_addr(&srv->check.addr)) { |
| 3808 | if (!srv->check.use_ssl && srv->use_ssl != -1) { |
| 3809 | srv->check.use_ssl = srv->use_ssl; |
| 3810 | srv->check.xprt = srv->xprt; |
| 3811 | } |
| 3812 | else if (srv->check.use_ssl == 1) |
| 3813 | srv->check.xprt = xprt_get(XPRT_SSL); |
| 3814 | |
| 3815 | srv->check.send_proxy |= (srv->pp_opts); |
| 3816 | } |
| 3817 | |
| 3818 | /* validate <srv> server health-check settings */ |
| 3819 | |
| 3820 | /* We need at least a service port, a check port or the first tcp-check |
| 3821 | * rule must be a 'connect' one when checking an IPv4/IPv6 server. |
| 3822 | */ |
| 3823 | if ((srv_check_healthcheck_port(&srv->check) != 0) || |
| 3824 | (!is_inet_addr(&srv->check.addr) && (is_addr(&srv->check.addr) || !is_inet_addr(&srv->addr)))) |
| 3825 | goto init; |
| 3826 | |
| 3827 | if (!LIST_ISEMPTY(&srv->proxy->tcpcheck_rules)) { |
| 3828 | ha_alert("config: %s '%s': server '%s' has neither service port nor check port.\n", |
| 3829 | proxy_type_str(srv->proxy), srv->proxy->id, srv->id); |
| 3830 | ret |= ERR_ALERT | ERR_ABORT; |
| 3831 | goto out; |
| 3832 | } |
| 3833 | |
| 3834 | /* search the first action (connect / send / expect) in the list */ |
| 3835 | r = get_first_tcpcheck_rule(&srv->proxy->tcpcheck_rules); |
| 3836 | if (!r || (r->action != TCPCHK_ACT_CONNECT) || !r->port) { |
| 3837 | ha_alert("config: %s '%s': server '%s' has neither service port nor check port " |
| 3838 | "nor tcp_check rule 'connect' with port information.\n", |
| 3839 | proxy_type_str(srv->proxy), srv->proxy->id, srv->id); |
| 3840 | ret |= ERR_ALERT | ERR_ABORT; |
| 3841 | goto out; |
| 3842 | } |
| 3843 | |
| 3844 | /* scan the tcp-check ruleset to ensure a port has been configured */ |
| 3845 | list_for_each_entry(r, &srv->proxy->tcpcheck_rules, list) { |
| 3846 | if ((r->action == TCPCHK_ACT_CONNECT) && (!r->port)) { |
| 3847 | ha_alert("config: %s '%s': server '%s' has neither service port nor check port, " |
| 3848 | "and a tcp_check rule 'connect' with no port information.\n", |
| 3849 | proxy_type_str(srv->proxy), srv->proxy->id, srv->id); |
| 3850 | ret |= ERR_ALERT | ERR_ABORT; |
| 3851 | goto out; |
| 3852 | } |
| 3853 | } |
| 3854 | |
| 3855 | init: |
| 3856 | err = init_check(&srv->check, srv->proxy->options2 & PR_O2_CHK_ANY); |
| 3857 | if (err) { |
| 3858 | ha_alert("config: %s '%s': unable to init check for server '%s' (%s).\n", |
| 3859 | proxy_type_str(srv->proxy), srv->proxy->id, srv->id, err); |
| 3860 | ret |= ERR_ALERT | ERR_ABORT; |
| 3861 | goto out; |
| 3862 | } |
| 3863 | srv->check.state |= CHK_ST_CONFIGURED | CHK_ST_ENABLED; |
| 3864 | global.maxsock++; |
| 3865 | |
| 3866 | out: |
| 3867 | return ret; |
| 3868 | } |
| 3869 | |
| 3870 | static int init_srv_agent_check(struct server *srv) |
| 3871 | { |
| 3872 | const char *err; |
| 3873 | int ret = 0; |
| 3874 | |
| 3875 | if (!srv->do_agent) |
| 3876 | goto out; |
| 3877 | |
| 3878 | err = init_check(&srv->agent, PR_O2_LB_AGENT_CHK); |
| 3879 | if (err) { |
| 3880 | ha_alert("config: %s '%s': unable to init agent-check for server '%s' (%s).\n", |
| 3881 | proxy_type_str(srv->proxy), srv->proxy->id, srv->id, err); |
| 3882 | ret |= ERR_ALERT | ERR_ABORT; |
| 3883 | goto out; |
| 3884 | } |
| 3885 | |
| 3886 | if (!srv->agent.inter) |
| 3887 | srv->agent.inter = srv->check.inter; |
| 3888 | |
| 3889 | srv->agent.state |= CHK_ST_CONFIGURED | CHK_ST_ENABLED | CHK_ST_AGENT; |
| 3890 | global.maxsock++; |
| 3891 | |
| 3892 | out: |
| 3893 | return ret; |
| 3894 | } |
| 3895 | |
| 3896 | static void deinit_srv_check(struct server *srv) |
| 3897 | { |
| 3898 | if (srv->do_check) |
| 3899 | free_check(&srv->check); |
| 3900 | } |
| 3901 | |
| 3902 | |
| 3903 | static void deinit_srv_agent_check(struct server *srv) |
| 3904 | { |
| 3905 | if (srv->do_agent) |
| 3906 | free_check(&srv->agent); |
| 3907 | free(srv->agent.send_string); |
| 3908 | } |
| 3909 | |
| 3910 | REGISTER_POST_SERVER_CHECK(init_srv_check); |
| 3911 | REGISTER_POST_SERVER_CHECK(init_srv_agent_check); |
| 3912 | |
| 3913 | REGISTER_SERVER_DEINIT(deinit_srv_check); |
| 3914 | REGISTER_SERVER_DEINIT(deinit_srv_agent_check); |
| 3915 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 3916 | /* |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 3917 | * Local variables: |
| 3918 | * c-indent-level: 8 |
| 3919 | * c-basic-offset: 8 |
| 3920 | * End: |
| 3921 | */ |