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> |
| 25 | #include <sys/socket.h> |
Dmitry Sivachenko | caf5898 | 2009-08-24 15:11:06 +0400 | [diff] [blame] | 26 | #include <sys/types.h> |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 27 | #include <sys/wait.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 28 | #include <netinet/in.h> |
Willy Tarreau | 1274bc4 | 2009-07-15 07:16:31 +0200 | [diff] [blame] | 29 | #include <netinet/tcp.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 30 | #include <arpa/inet.h> |
| 31 | |
Willy Tarreau | c7e4238 | 2012-08-24 19:22:53 +0200 | [diff] [blame] | 32 | #include <common/chunk.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 33 | #include <common/compat.h> |
| 34 | #include <common/config.h> |
| 35 | #include <common/mini-clist.h> |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 36 | #include <common/standard.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 37 | #include <common/time.h> |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 38 | #include <common/hathreads.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 39 | |
| 40 | #include <types/global.h> |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 41 | #include <types/dns.h> |
William Lallemand | 9ed6203 | 2016-11-21 17:49:11 +0100 | [diff] [blame] | 42 | #include <types/stats.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 43 | |
| 44 | #include <proto/backend.h> |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 45 | #include <proto/checks.h> |
William Lallemand | 9ed6203 | 2016-11-21 17:49:11 +0100 | [diff] [blame] | 46 | #include <proto/stats.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 47 | #include <proto/fd.h> |
| 48 | #include <proto/log.h> |
Willy Tarreau | 53a4766 | 2017-08-28 10:53:00 +0200 | [diff] [blame] | 49 | #include <proto/mux_pt.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 50 | #include <proto/queue.h> |
Willy Tarreau | c6f4ce8 | 2009-06-10 11:09:37 +0200 | [diff] [blame] | 51 | #include <proto/port_range.h> |
Willy Tarreau | e8c66af | 2008-01-13 18:40:14 +0100 | [diff] [blame] | 52 | #include <proto/proto_tcp.h> |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 53 | #include <proto/protocol.h> |
Willy Tarreau | 2b5652f | 2006-12-31 17:46:05 +0100 | [diff] [blame] | 54 | #include <proto/proxy.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 55 | #include <proto/server.h> |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 56 | #include <proto/signal.h> |
Willy Tarreau | 9e000c6 | 2011-03-10 14:03:36 +0100 | [diff] [blame] | 57 | #include <proto/stream_interface.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 58 | #include <proto/task.h> |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 59 | #include <proto/log.h> |
| 60 | #include <proto/dns.h> |
| 61 | #include <proto/proto_udp.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 62 | |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 63 | #ifdef USE_OPENSSL |
| 64 | #include <proto/ssl_sock.h> |
| 65 | #endif /* USE_OPENSSL */ |
| 66 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 67 | static int httpchk_expect(struct server *s, int done); |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 68 | static int tcpcheck_get_step_id(struct check *); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 69 | static char * tcpcheck_get_step_comment(struct check *, int); |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 70 | static int tcpcheck_main(struct check *); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 71 | static void __event_srv_chk_w(struct conn_stream *cs); |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 72 | static int wake_srv_chk(struct conn_stream *cs); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 73 | static void __event_srv_chk_r(struct conn_stream *cs); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 74 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 75 | static struct pool_head *pool_head_email_alert = NULL; |
| 76 | static struct pool_head *pool_head_tcpcheck_rule = NULL; |
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 }, |
| 120 | [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_EVAL_INIT }, |
| 121 | [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_EVAL_INIT }, |
| 122 | [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT }, |
| 123 | [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG }, |
| 124 | }; |
| 125 | |
Simon 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 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 142 | /* |
| 143 | * Convert check_status code to description |
| 144 | */ |
| 145 | const char *get_check_status_description(short check_status) { |
| 146 | |
| 147 | const char *desc; |
| 148 | |
| 149 | if (check_status < HCHK_STATUS_SIZE) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 150 | desc = check_statuses[check_status].desc; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 151 | else |
| 152 | desc = NULL; |
| 153 | |
| 154 | if (desc && *desc) |
| 155 | return desc; |
| 156 | else |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 157 | return check_statuses[HCHK_STATUS_UNKNOWN].desc; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 158 | } |
| 159 | |
| 160 | /* |
| 161 | * Convert check_status code to short info |
| 162 | */ |
| 163 | const char *get_check_status_info(short check_status) { |
| 164 | |
| 165 | const char *info; |
| 166 | |
| 167 | if (check_status < HCHK_STATUS_SIZE) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 168 | info = check_statuses[check_status].info; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 169 | else |
| 170 | info = NULL; |
| 171 | |
| 172 | if (info && *info) |
| 173 | return info; |
| 174 | else |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 175 | return check_statuses[HCHK_STATUS_UNKNOWN].info; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 176 | } |
| 177 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 178 | const char *get_analyze_status(short analyze_status) { |
| 179 | |
| 180 | const char *desc; |
| 181 | |
| 182 | if (analyze_status < HANA_STATUS_SIZE) |
| 183 | desc = analyze_statuses[analyze_status].desc; |
| 184 | else |
| 185 | desc = NULL; |
| 186 | |
| 187 | if (desc && *desc) |
| 188 | return desc; |
| 189 | else |
| 190 | return analyze_statuses[HANA_STATUS_UNKNOWN].desc; |
| 191 | } |
| 192 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 193 | /* |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 194 | * Set check->status, update check->duration and fill check->result with |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 195 | * an adequate CHK_RES_* value. The new check->health is computed based |
| 196 | * on the result. |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 197 | * |
| 198 | * Show information in logs about failed health check if server is UP |
| 199 | * or succeeded health checks if server is DOWN. |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 200 | */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 201 | 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] | 202 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 203 | struct server *s = check->server; |
Willy Tarreau | bef1b32 | 2014-05-13 21:01:39 +0200 | [diff] [blame] | 204 | short prev_status = check->status; |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 205 | int report = 0; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 206 | |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 207 | if (status == HCHK_STATUS_START) { |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 208 | check->result = CHK_RES_UNKNOWN; /* no result yet */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 209 | check->desc[0] = '\0'; |
| 210 | check->start = now; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 211 | return; |
| 212 | } |
| 213 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 214 | if (!check->status) |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 215 | return; |
| 216 | |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 217 | if (desc && *desc) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 218 | strncpy(check->desc, desc, HCHK_DESC_LEN-1); |
| 219 | check->desc[HCHK_DESC_LEN-1] = '\0'; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 220 | } else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 221 | check->desc[0] = '\0'; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 222 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 223 | check->status = status; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 224 | if (check_statuses[status].result) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 225 | check->result = check_statuses[status].result; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 226 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 227 | if (status == HCHK_STATUS_HANA) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 228 | check->duration = -1; |
| 229 | else if (!tv_iszero(&check->start)) { |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 230 | /* set_server_check_status() may be called more than once */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 231 | check->duration = tv_ms_elapsed(&check->start, &now); |
| 232 | tv_zero(&check->start); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 233 | } |
| 234 | |
Willy Tarreau | 2396418 | 2014-05-20 20:56:30 +0200 | [diff] [blame] | 235 | /* no change is expected if no state change occurred */ |
| 236 | if (check->result == CHK_RES_NEUTRAL) |
| 237 | return; |
| 238 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 239 | report = 0; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 240 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 241 | switch (check->result) { |
| 242 | case CHK_RES_FAILED: |
Willy Tarreau | 12634e1 | 2014-05-23 11:32:36 +0200 | [diff] [blame] | 243 | /* Failure to connect to the agent as a secondary check should not |
| 244 | * cause the server to be marked down. |
| 245 | */ |
| 246 | if ((!(check->state & CHK_ST_AGENT) || |
Simon Horman | eaabd52 | 2015-02-26 11:26:17 +0900 | [diff] [blame] | 247 | (check->status >= HCHK_STATUS_L57DATA)) && |
Christopher Faulet | b119a79 | 2018-05-02 12:12:45 +0200 | [diff] [blame] | 248 | (check->health > 0)) { |
Christopher Faulet | 29f77e8 | 2017-06-08 14:04:45 +0200 | [diff] [blame] | 249 | HA_ATOMIC_ADD(&s->counters.failed_checks, 1); |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 250 | report = 1; |
| 251 | check->health--; |
| 252 | if (check->health < check->rise) |
| 253 | check->health = 0; |
| 254 | } |
| 255 | break; |
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 | case CHK_RES_PASSED: |
| 258 | case CHK_RES_CONDPASS: /* "condpass" cannot make the first step but it OK after a "passed" */ |
| 259 | if ((check->health < check->rise + check->fall - 1) && |
| 260 | (check->result == CHK_RES_PASSED || check->health > 0)) { |
| 261 | report = 1; |
| 262 | check->health++; |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 263 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 264 | if (check->health >= check->rise) |
| 265 | check->health = check->rise + check->fall - 1; /* OK now */ |
| 266 | } |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 267 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 268 | /* clear consecutive_errors if observing is enabled */ |
| 269 | if (s->onerror) |
| 270 | s->consecutive_errors = 0; |
| 271 | break; |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 272 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 273 | default: |
| 274 | break; |
| 275 | } |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 276 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 277 | if (s->proxy->options2 & PR_O2_LOGHCHKS && |
| 278 | (status != prev_status || report)) { |
| 279 | chunk_printf(&trash, |
Willy Tarreau | 12634e1 | 2014-05-23 11:32:36 +0200 | [diff] [blame] | 280 | "%s check for %sserver %s/%s %s%s", |
| 281 | (check->state & CHK_ST_AGENT) ? "Agent" : "Health", |
Willy Tarreau | c93cd16 | 2014-05-13 15:54:22 +0200 | [diff] [blame] | 282 | s->flags & SRV_F_BACKUP ? "backup " : "", |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 283 | s->proxy->id, s->id, |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 284 | (check->result == CHK_RES_CONDPASS) ? "conditionally ":"", |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 285 | (check->result >= CHK_RES_PASSED) ? "succeeded" : "failed"); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 286 | |
Emeric Brun | 5a13351 | 2017-10-19 14:42:30 +0200 | [diff] [blame] | 287 | srv_append_status(&trash, s, check, -1, 0); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 288 | |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 289 | chunk_appendf(&trash, ", status: %d/%d %s", |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 290 | (check->health >= check->rise) ? check->health - check->rise + 1 : check->health, |
| 291 | (check->health >= check->rise) ? check->fall : check->rise, |
| 292 | (check->health >= check->rise) ? (s->uweight ? "UP" : "DRAIN") : "DOWN"); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 293 | |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 294 | ha_warning("%s.\n", trash.area); |
| 295 | send_log(s->proxy, LOG_NOTICE, "%s.\n", trash.area); |
| 296 | send_email_alert(s, LOG_INFO, "%s", trash.area); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 297 | } |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 298 | } |
| 299 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 300 | /* Marks the check <check>'s server down if the current check is already failed |
| 301 | * and the server is not down yet nor in maintenance. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 302 | */ |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 303 | static void check_notify_failure(struct check *check) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 304 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 305 | struct server *s = check->server; |
Simon Horman | e0d1bfb | 2011-06-21 14:34:58 +0900 | [diff] [blame] | 306 | |
Willy Tarreau | 7b1d47c | 2014-05-20 14:55:13 +0200 | [diff] [blame] | 307 | /* The agent secondary check should only cause a server to be marked |
| 308 | * as down if check->status is HCHK_STATUS_L7STS, which indicates |
| 309 | * that the agent returned "fail", "stopped" or "down". |
| 310 | * The implication here is that failure to connect to the agent |
| 311 | * as a secondary check should not cause the server to be marked |
| 312 | * down. */ |
| 313 | if ((check->state & CHK_ST_AGENT) && check->status != HCHK_STATUS_L7STS) |
| 314 | return; |
| 315 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 316 | if (check->health > 0) |
| 317 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 318 | |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 319 | /* 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] | 320 | 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] | 321 | } |
| 322 | |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 323 | /* 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] | 324 | * it isn't in maintenance, it is not tracking a down server and other checks |
| 325 | * comply. The rule is simple : by default, a server is up, unless any of the |
| 326 | * following conditions is true : |
| 327 | * - health check failed (check->health < rise) |
| 328 | * - agent check failed (agent->health < rise) |
| 329 | * - the server tracks a down server (track && track->state == STOPPED) |
| 330 | * Note that if the server has a slowstart, it will switch to STARTING instead |
| 331 | * of RUNNING. Also, only the health checks support the nolb mode, so the |
| 332 | * agent's success may not take the server out of this mode. |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 333 | */ |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 334 | static void check_notify_success(struct check *check) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 335 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 336 | struct server *s = check->server; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 337 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 338 | if (s->next_admin & SRV_ADMF_MAINT) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 339 | return; |
Cyril Bonté | cd19e51 | 2010-01-31 22:34:03 +0100 | [diff] [blame] | 340 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 341 | if (s->track && s->track->next_state == SRV_ST_STOPPED) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 342 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 343 | |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 344 | if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise)) |
| 345 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 346 | |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 347 | if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise)) |
| 348 | return; |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 349 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 350 | if ((check->state & CHK_ST_AGENT) && s->next_state == SRV_ST_STOPPING) |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 351 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 352 | |
Emeric Brun | 5a13351 | 2017-10-19 14:42:30 +0200 | [diff] [blame] | 353 | 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] | 354 | } |
| 355 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 356 | /* Marks the check <check> as valid and tries to set its server into stopping mode |
| 357 | * if it was running or starting, and provided it isn't in maintenance and other |
| 358 | * checks comply. The conditions for the server to be marked in stopping mode are |
| 359 | * the same as for it to be turned up. Also, only the health checks support the |
| 360 | * nolb mode. |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 361 | */ |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 362 | static void check_notify_stopping(struct check *check) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 363 | { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 364 | struct server *s = check->server; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 365 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 366 | if (s->next_admin & SRV_ADMF_MAINT) |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 367 | return; |
| 368 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 369 | if (check->state & CHK_ST_AGENT) |
| 370 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 371 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 372 | if (s->track && s->track->next_state == SRV_ST_STOPPED) |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 373 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 374 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 375 | if ((s->check.state & CHK_ST_ENABLED) && (s->check.health < s->check.rise)) |
| 376 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 377 | |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 378 | if ((s->agent.state & CHK_ST_ENABLED) && (s->agent.health < s->agent.rise)) |
| 379 | return; |
Krzysztof Piotr Oledzki | c8b16fc | 2008-02-18 01:26:35 +0100 | [diff] [blame] | 380 | |
Willy Tarreau | b26881a | 2017-12-23 11:16:49 +0100 | [diff] [blame] | 381 | 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] | 382 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 383 | |
Willy Tarreau | 9fe7aae | 2013-12-31 23:47:37 +0100 | [diff] [blame] | 384 | /* note: use health_adjust() only, which first checks that the observe mode is |
| 385 | * enabled. |
| 386 | */ |
| 387 | void __health_adjust(struct server *s, short status) |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 388 | { |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 389 | int failed; |
| 390 | int expire; |
| 391 | |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 392 | if (s->observe >= HANA_OBS_SIZE) |
| 393 | return; |
| 394 | |
Willy Tarreau | bb95666 | 2013-01-24 00:37:39 +0100 | [diff] [blame] | 395 | if (status >= HANA_STATUS_SIZE || !analyze_statuses[status].desc) |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 396 | return; |
| 397 | |
| 398 | switch (analyze_statuses[status].lr[s->observe - 1]) { |
| 399 | case 1: |
| 400 | failed = 1; |
| 401 | break; |
| 402 | |
| 403 | case 2: |
| 404 | failed = 0; |
| 405 | break; |
| 406 | |
| 407 | default: |
| 408 | return; |
| 409 | } |
| 410 | |
| 411 | if (!failed) { |
| 412 | /* good: clear consecutive_errors */ |
| 413 | s->consecutive_errors = 0; |
| 414 | return; |
| 415 | } |
| 416 | |
Christopher Faulet | 29f77e8 | 2017-06-08 14:04:45 +0200 | [diff] [blame] | 417 | HA_ATOMIC_ADD(&s->consecutive_errors, 1); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 418 | |
| 419 | if (s->consecutive_errors < s->consecutive_errors_limit) |
| 420 | return; |
| 421 | |
Willy Tarreau | 19d14ef | 2012-10-29 16:51:55 +0100 | [diff] [blame] | 422 | chunk_printf(&trash, "Detected %d consecutive errors, last one was: %s", |
| 423 | s->consecutive_errors, get_analyze_status(status)); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 424 | |
| 425 | switch (s->onerror) { |
| 426 | case HANA_ONERR_FASTINTER: |
| 427 | /* force fastinter - nothing to do here as all modes force it */ |
| 428 | break; |
| 429 | |
| 430 | case HANA_ONERR_SUDDTH: |
| 431 | /* simulate a pre-fatal failed health check */ |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 432 | if (s->check.health > s->check.rise) |
| 433 | s->check.health = s->check.rise + 1; |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 434 | |
| 435 | /* no break - fall through */ |
| 436 | |
| 437 | case HANA_ONERR_FAILCHK: |
| 438 | /* simulate a failed health check */ |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 439 | set_server_check_status(&s->check, HCHK_STATUS_HANA, |
| 440 | trash.area); |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 441 | check_notify_failure(&s->check); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 442 | break; |
| 443 | |
| 444 | case HANA_ONERR_MARKDWN: |
| 445 | /* mark server down */ |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 446 | s->check.health = s->check.rise; |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 447 | set_server_check_status(&s->check, HCHK_STATUS_HANA, |
| 448 | trash.area); |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 449 | check_notify_failure(&s->check); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 450 | break; |
| 451 | |
| 452 | default: |
| 453 | /* write a warning? */ |
| 454 | break; |
| 455 | } |
| 456 | |
| 457 | s->consecutive_errors = 0; |
Christopher Faulet | 29f77e8 | 2017-06-08 14:04:45 +0200 | [diff] [blame] | 458 | HA_ATOMIC_ADD(&s->counters.failed_hana, 1); |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 459 | |
Simon Horman | 6618300 | 2013-02-23 10:16:43 +0900 | [diff] [blame] | 460 | if (s->check.fastinter) { |
| 461 | expire = tick_add(now_ms, MS_TO_TICKS(s->check.fastinter)); |
Sergiy Prykhodko | 1d57e50 | 2013-09-21 12:05:00 +0300 | [diff] [blame] | 462 | if (s->check.task->expire > expire) { |
Willy Tarreau | 5b3a202 | 2012-09-28 15:01:02 +0200 | [diff] [blame] | 463 | s->check.task->expire = expire; |
Sergiy Prykhodko | 1d57e50 | 2013-09-21 12:05:00 +0300 | [diff] [blame] | 464 | /* requeue check task with new expire */ |
| 465 | task_queue(s->check.task); |
| 466 | } |
Krzysztof Piotr Oledzki | 97f07b8 | 2009-12-15 22:31:24 +0100 | [diff] [blame] | 467 | } |
| 468 | } |
| 469 | |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 470 | 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] | 471 | { |
| 472 | int sv_state; |
| 473 | int ratio; |
| 474 | int hlen = 0; |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 475 | char addr[46]; |
| 476 | char port[6]; |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 477 | const char *srv_hlt_st[7] = { "DOWN", "DOWN %d/%d", |
| 478 | "UP %d/%d", "UP", |
| 479 | "NOLB %d/%d", "NOLB", |
| 480 | "no check" }; |
| 481 | |
| 482 | memcpy(buffer + hlen, "X-Haproxy-Server-State: ", 24); |
| 483 | hlen += 24; |
| 484 | |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 485 | if (!(s->check.state & CHK_ST_ENABLED)) |
| 486 | sv_state = 6; |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 487 | else if (s->cur_state != SRV_ST_STOPPED) { |
Simon Horman | 58c3297 | 2013-11-25 10:46:38 +0900 | [diff] [blame] | 488 | if (s->check.health == s->check.rise + s->check.fall - 1) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 489 | sv_state = 3; /* UP */ |
| 490 | else |
| 491 | sv_state = 2; /* going down */ |
| 492 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 493 | if (s->cur_state == SRV_ST_STOPPING) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 494 | sv_state += 2; |
| 495 | } else { |
Simon Horman | 125d099 | 2013-02-24 17:23:38 +0900 | [diff] [blame] | 496 | if (s->check.health) |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 497 | sv_state = 1; /* going up */ |
| 498 | else |
| 499 | sv_state = 0; /* DOWN */ |
| 500 | } |
| 501 | |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 502 | hlen += snprintf(buffer + hlen, size - hlen, |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 503 | srv_hlt_st[sv_state], |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 504 | (s->cur_state != SRV_ST_STOPPED) ? (s->check.health - s->check.rise + 1) : (s->check.health), |
| 505 | (s->cur_state != SRV_ST_STOPPED) ? (s->check.fall) : (s->check.rise)); |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 506 | |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 507 | addr_to_str(&s->addr, addr, sizeof(addr)); |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 508 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 509 | snprintf(port, sizeof(port), "%u", s->svc_port); |
| 510 | else |
| 511 | *port = 0; |
Joseph Lynch | 514061c | 2015-01-15 17:52:59 -0800 | [diff] [blame] | 512 | |
| 513 | hlen += snprintf(buffer + hlen, size - hlen, "; address=%s; port=%s; name=%s/%s; node=%s; weight=%d/%d; scur=%d/%d; qcur=%d", |
| 514 | addr, port, s->proxy->id, s->id, |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 515 | global.node, |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 516 | (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] | 517 | (s->proxy->lbprm.tot_weight * s->proxy->lbprm.wmult + s->proxy->lbprm.wdiv - 1) / s->proxy->lbprm.wdiv, |
| 518 | s->cur_sess, s->proxy->beconn - s->proxy->nbpend, |
| 519 | s->nbpend); |
| 520 | |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 521 | if ((s->cur_state == SRV_ST_STARTING) && |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 522 | now.tv_sec < s->last_change + s->slowstart && |
| 523 | now.tv_sec >= s->last_change) { |
| 524 | ratio = MAX(1, 100 * (now.tv_sec - s->last_change) / s->slowstart); |
Willy Tarreau | a1dab55 | 2014-04-14 15:04:54 +0200 | [diff] [blame] | 525 | hlen += snprintf(buffer + hlen, size - hlen, "; throttle=%d%%", ratio); |
Willy Tarreau | ef78104 | 2010-01-27 11:53:01 +0100 | [diff] [blame] | 526 | } |
| 527 | |
| 528 | buffer[hlen++] = '\r'; |
| 529 | buffer[hlen++] = '\n'; |
| 530 | |
| 531 | return hlen; |
| 532 | } |
| 533 | |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 534 | /* Check the connection. If an error has already been reported or the socket is |
| 535 | * closed, keep errno intact as it is supposed to contain the valid error code. |
| 536 | * If no error is reported, check the socket's error queue using getsockopt(). |
| 537 | * Warning, this must be done only once when returning from poll, and never |
| 538 | * after an I/O error was attempted, otherwise the error queue might contain |
| 539 | * inconsistent errors. If an error is detected, the CO_FL_ERROR is set on the |
| 540 | * socket. Returns non-zero if an error was reported, zero if everything is |
| 541 | * clean (including a properly closed socket). |
| 542 | */ |
| 543 | static int retrieve_errno_from_socket(struct connection *conn) |
| 544 | { |
| 545 | int skerr; |
| 546 | socklen_t lskerr = sizeof(skerr); |
| 547 | |
| 548 | if (conn->flags & CO_FL_ERROR && ((errno && errno != EAGAIN) || !conn->ctrl)) |
| 549 | return 1; |
| 550 | |
Willy Tarreau | 3c72872 | 2014-01-23 13:50:42 +0100 | [diff] [blame] | 551 | if (!conn_ctrl_ready(conn)) |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 552 | return 0; |
| 553 | |
Willy Tarreau | 585744b | 2017-08-24 14:31:19 +0200 | [diff] [blame] | 554 | if (getsockopt(conn->handle.fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr) == 0) |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 555 | errno = skerr; |
| 556 | |
| 557 | if (errno == EAGAIN) |
| 558 | errno = 0; |
| 559 | |
| 560 | if (!errno) { |
| 561 | /* we could not retrieve an error, that does not mean there is |
| 562 | * none. Just don't change anything and only report the prior |
| 563 | * error if any. |
| 564 | */ |
| 565 | if (conn->flags & CO_FL_ERROR) |
| 566 | return 1; |
| 567 | else |
| 568 | return 0; |
| 569 | } |
| 570 | |
| 571 | conn->flags |= CO_FL_ERROR | CO_FL_SOCK_WR_SH | CO_FL_SOCK_RD_SH; |
| 572 | return 1; |
| 573 | } |
| 574 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 575 | /* Try to collect as much information as possible on the connection status, |
| 576 | * and adjust the server status accordingly. It may make use of <errno_bck> |
| 577 | * if non-null when the caller is absolutely certain of its validity (eg: |
| 578 | * checked just after a syscall). If the caller doesn't have a valid errno, |
| 579 | * it can pass zero, and retrieve_errno_from_socket() will be called to try |
| 580 | * to extract errno from the socket. If no error is reported, it will consider |
| 581 | * the <expired> flag. This is intended to be used when a connection error was |
| 582 | * reported in conn->flags or when a timeout was reported in <expired>. The |
| 583 | * function takes care of not updating a server status which was already set. |
| 584 | * All situations where at least one of <expired> or CO_FL_ERROR are set |
| 585 | * produce a status. |
| 586 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 587 | 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] | 588 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 589 | struct conn_stream *cs = check->cs; |
| 590 | struct connection *conn = cs_conn(cs); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 591 | const char *err_msg; |
Willy Tarreau | 83061a8 | 2018-07-13 11:56:34 +0200 | [diff] [blame] | 592 | struct buffer *chk; |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 593 | int step; |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 594 | char *comment; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 595 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 596 | if (check->result != CHK_RES_UNKNOWN) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 597 | return; |
| 598 | |
| 599 | errno = errno_bck; |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 600 | if (conn && (!errno || errno == EAGAIN)) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 601 | retrieve_errno_from_socket(conn); |
| 602 | |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 603 | if (conn && !(conn->flags & CO_FL_ERROR) && |
| 604 | !(cs->flags & CS_FL_ERROR) && !expired) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 605 | return; |
| 606 | |
| 607 | /* we'll try to build a meaningful error message depending on the |
| 608 | * context of the error possibly present in conn->err_code, and the |
| 609 | * socket error possibly collected above. This is useful to know the |
| 610 | * exact step of the L6 layer (eg: SSL handshake). |
| 611 | */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 612 | chk = get_trash_chunk(); |
| 613 | |
| 614 | if (check->type == PR_O2_TCPCHK_CHK) { |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 615 | step = tcpcheck_get_step_id(check); |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 616 | if (!step) |
| 617 | chunk_printf(chk, " at initial connection step of tcp-check"); |
| 618 | else { |
| 619 | chunk_printf(chk, " at step %d of tcp-check", step); |
| 620 | /* we were looking for a string */ |
| 621 | if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_CONNECT) { |
| 622 | if (check->last_started_step->port) |
| 623 | chunk_appendf(chk, " (connect port %d)" ,check->last_started_step->port); |
| 624 | else |
| 625 | chunk_appendf(chk, " (connect)"); |
| 626 | } |
| 627 | else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_EXPECT) { |
| 628 | if (check->last_started_step->string) |
Baptiste Assmann | 96a5c9b | 2015-05-01 08:09:29 +0200 | [diff] [blame] | 629 | chunk_appendf(chk, " (expect string '%s')", check->last_started_step->string); |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 630 | else if (check->last_started_step->expect_regex) |
| 631 | chunk_appendf(chk, " (expect regex)"); |
| 632 | } |
| 633 | else if (check->last_started_step && check->last_started_step->action == TCPCHK_ACT_SEND) { |
| 634 | chunk_appendf(chk, " (send)"); |
| 635 | } |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 636 | |
| 637 | comment = tcpcheck_get_step_comment(check, step); |
| 638 | if (comment) |
| 639 | chunk_appendf(chk, " comment: '%s'", comment); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 640 | } |
| 641 | } |
| 642 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 643 | if (conn && conn->err_code) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 644 | if (errno && errno != EAGAIN) |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 645 | chunk_printf(&trash, "%s (%s)%s", conn_err_code_str(conn), strerror(errno), |
| 646 | chk->area); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 647 | else |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 648 | chunk_printf(&trash, "%s%s", conn_err_code_str(conn), |
| 649 | chk->area); |
| 650 | err_msg = trash.area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 651 | } |
| 652 | else { |
| 653 | if (errno && errno != EAGAIN) { |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 654 | chunk_printf(&trash, "%s%s", strerror(errno), |
| 655 | chk->area); |
| 656 | err_msg = trash.area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 657 | } |
| 658 | else { |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 659 | err_msg = chk->area; |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 660 | } |
| 661 | } |
| 662 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 663 | if (check->state & CHK_ST_PORT_MISS) { |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 664 | /* NOTE: this is reported after <fall> tries */ |
| 665 | chunk_printf(chk, "No port available for the TCP connection"); |
| 666 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 667 | } |
| 668 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 669 | if (!conn) { |
| 670 | /* connection allocation error before the connection was established */ |
| 671 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 672 | } |
| 673 | else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L4_CONN)) == CO_FL_WAIT_L4_CONN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 674 | /* L4 not established (yet) */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 675 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 676 | set_server_check_status(check, HCHK_STATUS_L4CON, err_msg); |
| 677 | else if (expired) |
| 678 | set_server_check_status(check, HCHK_STATUS_L4TOUT, err_msg); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 679 | |
| 680 | /* |
| 681 | * might be due to a server IP change. |
| 682 | * Let's trigger a DNS resolution if none are currently running. |
| 683 | */ |
Christopher Faulet | 67957bd | 2017-09-27 11:00:59 +0200 | [diff] [blame] | 684 | dns_trigger_resolution(check->server->dns_requester); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 685 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 686 | } |
| 687 | else if ((conn->flags & (CO_FL_CONNECTED|CO_FL_WAIT_L6_CONN)) == CO_FL_WAIT_L6_CONN) { |
| 688 | /* L6 not established (yet) */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 689 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 690 | set_server_check_status(check, HCHK_STATUS_L6RSP, err_msg); |
| 691 | else if (expired) |
| 692 | set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg); |
| 693 | } |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 694 | else if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 695 | /* I/O error after connection was established and before we could diagnose */ |
| 696 | set_server_check_status(check, HCHK_STATUS_SOCKERR, err_msg); |
| 697 | } |
| 698 | else if (expired) { |
| 699 | /* connection established but expired check */ |
| 700 | if (check->type == PR_O2_SSL3_CHK) |
| 701 | set_server_check_status(check, HCHK_STATUS_L6TOUT, err_msg); |
| 702 | else /* HTTP, SMTP, ... */ |
| 703 | set_server_check_status(check, HCHK_STATUS_L7TOUT, err_msg); |
| 704 | } |
| 705 | |
| 706 | return; |
| 707 | } |
| 708 | |
Olivier Houchard | 5c110b9 | 2018-08-14 17:04:58 +0200 | [diff] [blame] | 709 | /* This function checks if any I/O is wanted, and if so, attempts to do so */ |
| 710 | 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] | 711 | { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 712 | struct check *check = ctx; |
| 713 | struct conn_stream *cs = check->cs; |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 714 | |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 715 | if (!(check->wait_list.wait_reason & SUB_CAN_SEND)) |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 716 | wake_srv_chk(cs); |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 717 | if (!(check->wait_list.wait_reason & SUB_CAN_RECV)) { |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 718 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
| 719 | __event_srv_chk_r(cs); |
| 720 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
| 721 | } |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 722 | return NULL; |
| 723 | } |
| 724 | |
| 725 | /* same as above but protected by the server lock. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 726 | * |
| 727 | * 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] | 728 | * via the out label. NOTE THAT THIS FUNCTION DOESN'T LOCK, YOU PROBABLY WANT |
| 729 | * TO USE event_srv_chk_w() instead. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 730 | */ |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 731 | static void __event_srv_chk_w(struct conn_stream *cs) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 732 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 733 | struct connection *conn = cs->conn; |
| 734 | struct check *check = cs->data; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 735 | struct server *s = check->server; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 736 | struct task *t = check->task; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 737 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 738 | if (unlikely(check->result == CHK_RES_FAILED)) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 739 | goto out_wakeup; |
| 740 | |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 741 | if (conn->flags & CO_FL_HANDSHAKE) { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 742 | cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 743 | goto out; |
| 744 | } |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 745 | |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 746 | if (retrieve_errno_from_socket(conn)) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 747 | chk_report_conn_err(check, errno, 0); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 748 | __cs_stop_both(cs); |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 749 | goto out_wakeup; |
| 750 | } |
Krzysztof Piotr Oledzki | 6492db5 | 2010-01-02 22:03:01 +0100 | [diff] [blame] | 751 | |
Willy Tarreau | bbae3f0 | 2017-08-30 09:59:52 +0200 | [diff] [blame] | 752 | if (conn->flags & CO_FL_SOCK_WR_SH) { |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 753 | /* if the output is closed, we can't do anything */ |
| 754 | conn->flags |= CO_FL_ERROR; |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 755 | chk_report_conn_err(check, 0, 0); |
Willy Tarreau | 20a1834 | 2013-12-05 00:31:46 +0100 | [diff] [blame] | 756 | goto out_wakeup; |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 757 | } |
Willy Tarreau | 6996e15 | 2007-04-30 14:37:43 +0200 | [diff] [blame] | 758 | |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 759 | /* here, we know that the connection is established. That's enough for |
| 760 | * a pure TCP check. |
| 761 | */ |
| 762 | if (!check->type) |
| 763 | goto out_wakeup; |
| 764 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 765 | /* wake() will take care of calling tcpcheck_main() */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 766 | if (check->type == PR_O2_TCPCHK_CHK) |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 767 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 768 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 769 | if (b_data(&check->bo)) { |
Olivier Houchard | ed0f207 | 2018-08-16 15:41:52 +0200 | [diff] [blame] | 770 | 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] | 771 | b_realign_if_empty(&check->bo); |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 772 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 773 | chk_report_conn_err(check, errno, 0); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 774 | __cs_stop_both(cs); |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 775 | goto out_wakeup; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 776 | } |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 777 | if (b_data(&check->bo)) { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 778 | conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list); |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 779 | goto out; |
| 780 | } |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 781 | } |
Willy Tarreau | 6996e15 | 2007-04-30 14:37:43 +0200 | [diff] [blame] | 782 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 783 | /* full request sent, we allow up to <timeout.check> if nonzero for a response */ |
| 784 | if (s->proxy->timeout.check) { |
| 785 | t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check); |
| 786 | task_queue(t); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 787 | } |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 788 | goto out_nowake; |
| 789 | |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 790 | out_wakeup: |
Willy Tarreau | fdccded | 2008-08-29 18:19:04 +0200 | [diff] [blame] | 791 | task_wakeup(t, TASK_WOKEN_IO); |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 792 | out_nowake: |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 793 | __cs_stop_send(cs); /* nothing more to write */ |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 794 | out: |
| 795 | return; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 796 | } |
| 797 | |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 798 | /* |
Willy Tarreau | f3c6920 | 2006-07-09 16:42:34 +0200 | [diff] [blame] | 799 | * 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] | 800 | * 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] | 801 | * set_server_check_status() to update check->status, check->duration |
| 802 | * and check->result. |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 803 | |
| 804 | * The set_server_check_status function is called with HCHK_STATUS_L7OKD if |
| 805 | * an HTTP server replies HTTP 2xx or 3xx (valid responses), if an SMTP server |
| 806 | * returns 2xx, HCHK_STATUS_L6OK if an SSL server returns at least 5 bytes in |
| 807 | * response to an SSL HELLO (the principle is that this is enough to |
| 808 | * distinguish between an SSL server and a pure TCP relay). All other cases will |
| 809 | * call it with a proper error status like HCHK_STATUS_L7STS, HCHK_STATUS_L6RSP, |
| 810 | * etc. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 811 | * |
| 812 | * 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] | 813 | * via the out label. |
| 814 | * |
| 815 | * This must be called with the server lock held. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 816 | */ |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 817 | static void __event_srv_chk_r(struct conn_stream *cs) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 818 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 819 | struct connection *conn = cs->conn; |
| 820 | struct check *check = cs->data; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 821 | struct server *s = check->server; |
| 822 | struct task *t = check->task; |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 823 | char *desc; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 824 | int done; |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 825 | unsigned short msglen; |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 826 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 827 | if (unlikely(check->result == CHK_RES_FAILED)) |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 828 | goto out_wakeup; |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 829 | |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 830 | if (conn->flags & CO_FL_HANDSHAKE) { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 831 | cs->conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 832 | goto out; |
| 833 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 834 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 835 | /* wake() will take care of calling tcpcheck_main() */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 836 | if (check->type == PR_O2_TCPCHK_CHK) |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 837 | goto out; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 838 | |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 839 | /* Warning! Linux returns EAGAIN on SO_ERROR if data are still available |
| 840 | * but the connection was closed on the remote end. Fortunately, recv still |
| 841 | * works correctly and we don't need to do the getsockopt() on linux. |
| 842 | */ |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 843 | |
| 844 | /* Set buffer to point to the end of the data already read, and check |
| 845 | * that there is free space remaining. If the buffer is full, proceed |
| 846 | * with running the checks without attempting another socket read. |
| 847 | */ |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 848 | |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 849 | done = 0; |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 850 | |
Olivier Houchard | 511efea | 2018-08-16 15:30:32 +0200 | [diff] [blame] | 851 | 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] | 852 | 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] | 853 | done = 1; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 854 | 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] | 855 | /* Report network errors only if we got no other data. Otherwise |
| 856 | * we'll let the upper layers decide whether the response is OK |
| 857 | * or not. It is very common that an RST sent by the server is |
| 858 | * reported as an error just after the last data chunk. |
| 859 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 860 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | c1a0796 | 2010-03-16 20:55:43 +0100 | [diff] [blame] | 861 | goto out_wakeup; |
| 862 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 863 | } |
| 864 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 865 | |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 866 | /* Intermediate or complete response received. |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 867 | * Terminate string in b_head(&check->bi) buffer. |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 868 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 869 | if (b_data(&check->bi) < b_size(&check->bi)) |
| 870 | b_head(&check->bi)[b_data(&check->bi)] = '\0'; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 871 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 872 | b_head(&check->bi)[b_data(&check->bi) - 1] = '\0'; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 873 | done = 1; /* buffer full, don't wait for more data */ |
| 874 | } |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 875 | |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 876 | /* Run the checks... */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 877 | switch (check->type) { |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 878 | case PR_O2_HTTP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 879 | if (!done && b_data(&check->bi) < strlen("HTTP/1.0 000\r")) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 880 | goto wait_more_data; |
| 881 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 882 | /* Check if the server speaks HTTP 1.X */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 883 | if ((b_data(&check->bi) < strlen("HTTP/1.0 000\r")) || |
| 884 | (memcmp(b_head(&check->bi), "HTTP/1.", 7) != 0 || |
| 885 | (*(b_head(&check->bi) + 12) != ' ' && *(b_head(&check->bi) + 12) != '\r')) || |
| 886 | !isdigit((unsigned char) *(b_head(&check->bi) + 9)) || !isdigit((unsigned char) *(b_head(&check->bi) + 10)) || |
| 887 | !isdigit((unsigned char) *(b_head(&check->bi) + 11))) { |
| 888 | cut_crlf(b_head(&check->bi)); |
| 889 | 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] | 890 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 891 | goto out_wakeup; |
| 892 | } |
| 893 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 894 | check->code = str2uic(b_head(&check->bi) + 9); |
| 895 | desc = ltrim(b_head(&check->bi) + 12, ' '); |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 896 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 897 | if ((s->proxy->options & PR_O_DISABLE404) && |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 898 | (s->next_state != SRV_ST_STOPPED) && (check->code == 404)) { |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 899 | /* 404 may be accepted as "stopping" only if the server was up */ |
| 900 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 901 | set_server_check_status(check, HCHK_STATUS_L7OKCD, desc); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 902 | } |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 903 | else if (s->proxy->options2 & PR_O2_EXP_TYPE) { |
| 904 | /* Run content verification check... We know we have at least 13 chars */ |
| 905 | if (!httpchk_expect(s, done)) |
| 906 | goto wait_more_data; |
| 907 | } |
| 908 | /* check the reply : HTTP/1.X 2xx and 3xx are OK */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 909 | 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] | 910 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 911 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 912 | } |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 913 | else { |
| 914 | cut_crlf(desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 915 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 916 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 917 | break; |
| 918 | |
| 919 | case PR_O2_SSL3_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 920 | if (!done && b_data(&check->bi) < 5) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 921 | goto wait_more_data; |
| 922 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 923 | /* Check for SSLv3 alert or handshake */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 924 | 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] | 925 | set_server_check_status(check, HCHK_STATUS_L6OK, NULL); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 926 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 927 | set_server_check_status(check, HCHK_STATUS_L6RSP, NULL); |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 928 | break; |
| 929 | |
| 930 | case PR_O2_SMTP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 931 | if (!done && b_data(&check->bi) < strlen("000\r")) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 932 | goto wait_more_data; |
| 933 | |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 934 | /* Check if the server speaks SMTP */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 935 | if ((b_data(&check->bi) < strlen("000\r")) || |
| 936 | (*(b_head(&check->bi) + 3) != ' ' && *(b_head(&check->bi) + 3) != '\r') || |
| 937 | !isdigit((unsigned char) *b_head(&check->bi)) || !isdigit((unsigned char) *(b_head(&check->bi) + 1)) || |
| 938 | !isdigit((unsigned char) *(b_head(&check->bi) + 2))) { |
| 939 | cut_crlf(b_head(&check->bi)); |
| 940 | 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] | 941 | goto out_wakeup; |
| 942 | } |
| 943 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 944 | check->code = str2uic(b_head(&check->bi)); |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 945 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 946 | desc = ltrim(b_head(&check->bi) + 3, ' '); |
Krzysztof Piotr Oledzki | f7089f5 | 2009-10-10 21:06:49 +0200 | [diff] [blame] | 947 | cut_crlf(desc); |
| 948 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 949 | /* Check for SMTP code 2xx (should be 250) */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 950 | if (*b_head(&check->bi) == '2') |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 951 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Krzysztof Piotr Oledzki | 213014e | 2009-09-27 15:50:02 +0200 | [diff] [blame] | 952 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 953 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 954 | break; |
| 955 | |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 956 | case PR_O2_LB_AGENT_CHK: { |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 957 | int status = HCHK_STATUS_CHECKED; |
| 958 | const char *hs = NULL; /* health status */ |
| 959 | const char *as = NULL; /* admin status */ |
| 960 | const char *ps = NULL; /* performance status */ |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 961 | const char *cs = NULL; /* maxconn */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 962 | const char *err = NULL; /* first error to report */ |
| 963 | const char *wrn = NULL; /* first warning to report */ |
| 964 | char *cmd, *p; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 965 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 966 | /* We're getting an agent check response. The agent could |
| 967 | * have been disabled in the mean time with a long check |
| 968 | * still pending. It is important that we ignore the whole |
| 969 | * response. |
| 970 | */ |
| 971 | if (!(check->server->agent.state & CHK_ST_ENABLED)) |
| 972 | break; |
| 973 | |
| 974 | /* The agent supports strings made of a single line ended by the |
| 975 | * first CR ('\r') or LF ('\n'). This line is composed of words |
| 976 | * delimited by spaces (' '), tabs ('\t'), or commas (','). The |
| 977 | * line may optionally contained a description of a state change |
| 978 | * after a sharp ('#'), which is only considered if a health state |
| 979 | * is announced. |
| 980 | * |
| 981 | * Words may be composed of : |
| 982 | * - a numeric weight suffixed by the percent character ('%'). |
| 983 | * - a health status among "up", "down", "stopped", and "fail". |
| 984 | * - an admin status among "ready", "drain", "maint". |
| 985 | * |
| 986 | * These words may appear in any order. If multiple words of the |
| 987 | * same category appear, the last one wins. |
| 988 | */ |
| 989 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 990 | p = b_head(&check->bi); |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 991 | while (*p && *p != '\n' && *p != '\r') |
| 992 | p++; |
| 993 | |
| 994 | if (!*p) { |
| 995 | if (!done) |
| 996 | goto wait_more_data; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 997 | |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 998 | /* at least inform the admin that the agent is mis-behaving */ |
| 999 | set_server_check_status(check, check->status, "Ignoring incomplete line from agent"); |
| 1000 | break; |
| 1001 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1002 | |
Willy Tarreau | 9809b78 | 2013-12-11 21:40:11 +0100 | [diff] [blame] | 1003 | *p = 0; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1004 | cmd = b_head(&check->bi); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1005 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1006 | while (*cmd) { |
| 1007 | /* look for next word */ |
| 1008 | if (*cmd == ' ' || *cmd == '\t' || *cmd == ',') { |
| 1009 | cmd++; |
| 1010 | continue; |
| 1011 | } |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 1012 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1013 | if (*cmd == '#') { |
| 1014 | /* this is the beginning of a health status description, |
| 1015 | * skip the sharp and blanks. |
| 1016 | */ |
| 1017 | cmd++; |
| 1018 | while (*cmd == '\t' || *cmd == ' ') |
| 1019 | cmd++; |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 1020 | break; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1021 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1022 | |
| 1023 | /* find the end of the word so that we have a null-terminated |
| 1024 | * word between <cmd> and <p>. |
| 1025 | */ |
| 1026 | p = cmd + 1; |
| 1027 | while (*p && *p != '\t' && *p != ' ' && *p != '\n' && *p != ',') |
| 1028 | p++; |
| 1029 | if (*p) |
| 1030 | *p++ = 0; |
| 1031 | |
| 1032 | /* first, health statuses */ |
| 1033 | if (strcasecmp(cmd, "up") == 0) { |
| 1034 | check->health = check->rise + check->fall - 1; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1035 | status = HCHK_STATUS_L7OKD; |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1036 | hs = cmd; |
| 1037 | } |
| 1038 | else if (strcasecmp(cmd, "down") == 0) { |
| 1039 | check->health = 0; |
| 1040 | status = HCHK_STATUS_L7STS; |
| 1041 | hs = cmd; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1042 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1043 | else if (strcasecmp(cmd, "stopped") == 0) { |
| 1044 | check->health = 0; |
| 1045 | status = HCHK_STATUS_L7STS; |
| 1046 | hs = cmd; |
| 1047 | } |
| 1048 | else if (strcasecmp(cmd, "fail") == 0) { |
| 1049 | check->health = 0; |
| 1050 | status = HCHK_STATUS_L7STS; |
| 1051 | hs = cmd; |
| 1052 | } |
| 1053 | /* admin statuses */ |
| 1054 | else if (strcasecmp(cmd, "ready") == 0) { |
| 1055 | as = cmd; |
| 1056 | } |
| 1057 | else if (strcasecmp(cmd, "drain") == 0) { |
| 1058 | as = cmd; |
| 1059 | } |
| 1060 | else if (strcasecmp(cmd, "maint") == 0) { |
| 1061 | as = cmd; |
| 1062 | } |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1063 | /* try to parse a weight here and keep the last one */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1064 | else if (isdigit((unsigned char)*cmd) && strchr(cmd, '%') != NULL) { |
| 1065 | ps = cmd; |
| 1066 | } |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1067 | /* try to parse a maxconn here */ |
| 1068 | else if (strncasecmp(cmd, "maxconn:", strlen("maxconn:")) == 0) { |
| 1069 | cs = cmd; |
| 1070 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1071 | else { |
| 1072 | /* keep a copy of the first error */ |
| 1073 | if (!err) |
| 1074 | err = cmd; |
| 1075 | } |
| 1076 | /* skip to next word */ |
| 1077 | cmd = p; |
| 1078 | } |
| 1079 | /* here, cmd points either to \0 or to the beginning of a |
| 1080 | * description. Skip possible leading spaces. |
| 1081 | */ |
| 1082 | while (*cmd == ' ' || *cmd == '\n') |
| 1083 | cmd++; |
| 1084 | |
| 1085 | /* First, update the admin status so that we avoid sending other |
| 1086 | * possibly useless warnings and can also update the health if |
| 1087 | * present after going back up. |
| 1088 | */ |
| 1089 | if (as) { |
| 1090 | if (strcasecmp(as, "drain") == 0) |
| 1091 | srv_adm_set_drain(check->server); |
| 1092 | else if (strcasecmp(as, "maint") == 0) |
| 1093 | srv_adm_set_maint(check->server); |
| 1094 | else |
| 1095 | srv_adm_set_ready(check->server); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1096 | } |
| 1097 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1098 | /* now change weights */ |
| 1099 | if (ps) { |
| 1100 | const char *msg; |
| 1101 | |
| 1102 | msg = server_parse_weight_change_request(s, ps); |
| 1103 | if (!wrn || !*wrn) |
| 1104 | wrn = msg; |
| 1105 | } |
| 1106 | |
Nenad Merdanovic | 174dd37 | 2016-04-24 23:10:06 +0200 | [diff] [blame] | 1107 | if (cs) { |
| 1108 | const char *msg; |
| 1109 | |
| 1110 | cs += strlen("maxconn:"); |
| 1111 | |
| 1112 | msg = server_parse_maxconn_change_request(s, cs); |
| 1113 | if (!wrn || !*wrn) |
| 1114 | wrn = msg; |
| 1115 | } |
| 1116 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1117 | /* and finally health status */ |
| 1118 | if (hs) { |
| 1119 | /* We'll report some of the warnings and errors we have |
| 1120 | * here. Down reports are critical, we leave them untouched. |
| 1121 | * Lack of report, or report of 'UP' leaves the room for |
| 1122 | * ERR first, then WARN. |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1123 | */ |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1124 | const char *msg = cmd; |
Willy Tarreau | 83061a8 | 2018-07-13 11:56:34 +0200 | [diff] [blame] | 1125 | struct buffer *t; |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1126 | |
| 1127 | if (!*msg || status == HCHK_STATUS_L7OKD) { |
| 1128 | if (err && *err) |
| 1129 | msg = err; |
| 1130 | else if (wrn && *wrn) |
| 1131 | msg = wrn; |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1132 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1133 | |
| 1134 | t = get_trash_chunk(); |
| 1135 | chunk_printf(t, "via agent : %s%s%s%s", |
| 1136 | hs, *msg ? " (" : "", |
| 1137 | msg, *msg ? ")" : ""); |
| 1138 | |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1139 | set_server_check_status(check, status, t->area); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1140 | } |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1141 | else if (err && *err) { |
| 1142 | /* No status change but we'd like to report something odd. |
| 1143 | * Just report the current state and copy the message. |
| 1144 | */ |
| 1145 | chunk_printf(&trash, "agent reports an error : %s", err); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1146 | set_server_check_status(check, status/*check->status*/, |
| 1147 | trash.area); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1148 | |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1149 | } |
| 1150 | else if (wrn && *wrn) { |
| 1151 | /* No status change but we'd like to report something odd. |
| 1152 | * Just report the current state and copy the message. |
| 1153 | */ |
| 1154 | chunk_printf(&trash, "agent warns : %s", wrn); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1155 | set_server_check_status(check, status/*check->status*/, |
| 1156 | trash.area); |
Willy Tarreau | 81f5d94 | 2013-12-09 20:51:51 +0100 | [diff] [blame] | 1157 | } |
| 1158 | else |
| 1159 | set_server_check_status(check, status, NULL); |
Simon Horman | a2b9dad | 2013-02-12 10:45:54 +0900 | [diff] [blame] | 1160 | break; |
| 1161 | } |
| 1162 | |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1163 | case PR_O2_PGSQL_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1164 | if (!done && b_data(&check->bi) < 9) |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1165 | goto wait_more_data; |
| 1166 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1167 | if (b_head(&check->bi)[0] == 'R') { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1168 | set_server_check_status(check, HCHK_STATUS_L7OKD, "PostgreSQL server is ok"); |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1169 | } |
| 1170 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1171 | if ((b_head(&check->bi)[0] == 'E') && (b_head(&check->bi)[5]!=0) && (b_head(&check->bi)[6]!=0)) |
| 1172 | desc = &b_head(&check->bi)[6]; |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1173 | else |
| 1174 | desc = "PostgreSQL unknown error"; |
| 1175 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1176 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Rauf Kuliyev | 38b4156 | 2011-01-04 15:14:13 +0100 | [diff] [blame] | 1177 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1178 | break; |
| 1179 | |
| 1180 | case PR_O2_REDIS_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1181 | if (!done && b_data(&check->bi) < 7) |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1182 | goto wait_more_data; |
| 1183 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1184 | if (strcmp(b_head(&check->bi), "+PONG\r\n") == 0) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1185 | set_server_check_status(check, HCHK_STATUS_L7OKD, "Redis server is ok"); |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1186 | } |
| 1187 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1188 | set_server_check_status(check, HCHK_STATUS_L7STS, b_head(&check->bi)); |
Hervé COMMOWICK | ec032d6 | 2011-08-05 16:23:48 +0200 | [diff] [blame] | 1189 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1190 | break; |
| 1191 | |
| 1192 | case PR_O2_MYSQL_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1193 | if (!done && b_data(&check->bi) < 5) |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1194 | goto wait_more_data; |
| 1195 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1196 | if (s->proxy->check_len == 0) { // old mode |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1197 | if (*(b_head(&check->bi) + 4) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1198 | /* We set the MySQL Version in description for information purpose |
| 1199 | * FIXME : it can be cool to use MySQL Version for other purpose, |
| 1200 | * like mark as down old MySQL server. |
| 1201 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1202 | if (b_data(&check->bi) > 51) { |
| 1203 | desc = ltrim(b_head(&check->bi) + 5, ' '); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1204 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1205 | } |
| 1206 | else { |
| 1207 | if (!done) |
| 1208 | goto wait_more_data; |
Emeric Brun | 9f0b458 | 2017-10-23 14:39:51 +0200 | [diff] [blame] | 1209 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1210 | /* it seems we have a OK packet but without a valid length, |
| 1211 | * it must be a protocol error |
| 1212 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1213 | set_server_check_status(check, HCHK_STATUS_L7RSP, b_head(&check->bi)); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1214 | } |
Hervé COMMOWICK | 698ae00 | 2010-01-12 09:25:13 +0100 | [diff] [blame] | 1215 | } |
| 1216 | else { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1217 | /* An error message is attached in the Error packet */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1218 | desc = ltrim(b_head(&check->bi) + 7, ' '); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1219 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1220 | } |
| 1221 | } else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1222 | unsigned int first_packet_len = ((unsigned int) *b_head(&check->bi)) + |
| 1223 | (((unsigned int) *(b_head(&check->bi) + 1)) << 8) + |
| 1224 | (((unsigned int) *(b_head(&check->bi) + 2)) << 16); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1225 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1226 | if (b_data(&check->bi) == first_packet_len + 4) { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1227 | /* MySQL Error packet always begin with field_count = 0xff */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1228 | if (*(b_head(&check->bi) + 4) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1229 | /* We have only one MySQL packet and it is a Handshake Initialization packet |
| 1230 | * but we need to have a second packet to know if it is alright |
| 1231 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1232 | if (!done && b_data(&check->bi) < first_packet_len + 5) |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1233 | goto wait_more_data; |
| 1234 | } |
| 1235 | else { |
| 1236 | /* We have only one packet and it is an Error packet, |
| 1237 | * an error message is attached, so we can display it |
| 1238 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1239 | desc = &b_head(&check->bi)[7]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1240 | //ha_warning("onlyoneERR: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1241 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1242 | } |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1243 | } else if (b_data(&check->bi) > first_packet_len + 4) { |
| 1244 | unsigned int second_packet_len = ((unsigned int) *(b_head(&check->bi) + first_packet_len + 4)) + |
| 1245 | (((unsigned int) *(b_head(&check->bi) + first_packet_len + 5)) << 8) + |
| 1246 | (((unsigned int) *(b_head(&check->bi) + first_packet_len + 6)) << 16); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1247 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1248 | 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] | 1249 | /* We have 2 packets and that's good */ |
| 1250 | /* Check if the second packet is a MySQL Error packet or not */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1251 | if (*(b_head(&check->bi) + first_packet_len + 8) != '\xff') { |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1252 | /* No error packet */ |
| 1253 | /* We set the MySQL Version in description for information purpose */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1254 | desc = &b_head(&check->bi)[5]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1255 | //ha_warning("2packetOK: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1256 | set_server_check_status(check, HCHK_STATUS_L7OKD, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1257 | } |
| 1258 | else { |
| 1259 | /* An error message is attached in the Error packet |
| 1260 | * so we can display it ! :) |
| 1261 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1262 | desc = &b_head(&check->bi)[first_packet_len+11]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1263 | //ha_warning("2packetERR: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1264 | set_server_check_status(check, HCHK_STATUS_L7STS, desc); |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1265 | } |
| 1266 | } |
| 1267 | } |
| 1268 | else { |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1269 | if (!done) |
| 1270 | goto wait_more_data; |
Emeric Brun | 9f0b458 | 2017-10-23 14:39:51 +0200 | [diff] [blame] | 1271 | |
Hervé COMMOWICK | 8776f1b | 2010-10-18 15:58:36 +0200 | [diff] [blame] | 1272 | /* 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] | 1273 | * it must be a protocol error |
| 1274 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1275 | desc = &b_head(&check->bi)[5]; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1276 | //ha_warning("protoerr: %s\n", desc); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1277 | set_server_check_status(check, HCHK_STATUS_L7RSP, desc); |
Hervé COMMOWICK | 698ae00 | 2010-01-12 09:25:13 +0100 | [diff] [blame] | 1278 | } |
| 1279 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1280 | break; |
| 1281 | |
| 1282 | case PR_O2_LDAP_CHK: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1283 | if (!done && b_data(&check->bi) < 14) |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1284 | goto wait_more_data; |
| 1285 | |
| 1286 | /* Check if the server speaks LDAP (ASN.1/BER) |
| 1287 | * http://en.wikipedia.org/wiki/Basic_Encoding_Rules |
| 1288 | * http://tools.ietf.org/html/rfc4511 |
| 1289 | */ |
| 1290 | |
| 1291 | /* http://tools.ietf.org/html/rfc4511#section-4.1.1 |
| 1292 | * LDAPMessage: 0x30: SEQUENCE |
| 1293 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1294 | if ((b_data(&check->bi) < 14) || (*(b_head(&check->bi)) != '\x30')) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1295 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1296 | } |
| 1297 | else { |
| 1298 | /* size of LDAPMessage */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1299 | 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] | 1300 | |
| 1301 | /* http://tools.ietf.org/html/rfc4511#section-4.2.2 |
| 1302 | * messageID: 0x02 0x01 0x01: INTEGER 1 |
| 1303 | * protocolOp: 0x61: bindResponse |
| 1304 | */ |
| 1305 | if ((msglen > 2) || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1306 | (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] | 1307 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1308 | goto out_wakeup; |
| 1309 | } |
| 1310 | |
| 1311 | /* size of bindResponse */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1312 | 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] | 1313 | |
| 1314 | /* http://tools.ietf.org/html/rfc4511#section-4.1.9 |
| 1315 | * ldapResult: 0x0a 0x01: ENUMERATION |
| 1316 | */ |
| 1317 | if ((msglen > 4) || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1318 | (memcmp(b_head(&check->bi) + 7 + msglen, "\x0a\x01", 2) != 0)) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1319 | set_server_check_status(check, HCHK_STATUS_L7RSP, "Not LDAPv3 protocol"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1320 | goto out_wakeup; |
| 1321 | } |
| 1322 | |
| 1323 | /* http://tools.ietf.org/html/rfc4511#section-4.1.9 |
| 1324 | * resultCode |
| 1325 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1326 | check->code = *(b_head(&check->bi) + msglen + 9); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1327 | if (check->code) { |
| 1328 | 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] | 1329 | } else { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 1330 | set_server_check_status(check, HCHK_STATUS_L7OKD, "Success"); |
Gabor Lekeny | b4c81e4 | 2010-09-29 18:17:05 +0200 | [diff] [blame] | 1331 | } |
| 1332 | } |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1333 | break; |
| 1334 | |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1335 | case PR_O2_SPOP_CHK: { |
| 1336 | unsigned int framesz; |
| 1337 | char err[HCHK_DESC_LEN]; |
| 1338 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1339 | if (!done && b_data(&check->bi) < 4) |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1340 | goto wait_more_data; |
| 1341 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1342 | memcpy(&framesz, b_head(&check->bi), 4); |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1343 | framesz = ntohl(framesz); |
| 1344 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1345 | if (!done && b_data(&check->bi) < (4+framesz)) |
Christopher Faulet | ba7bc16 | 2016-11-07 21:07:38 +0100 | [diff] [blame] | 1346 | goto wait_more_data; |
| 1347 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1348 | 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] | 1349 | set_server_check_status(check, HCHK_STATUS_L7OKD, "SPOA server is ok"); |
| 1350 | else |
| 1351 | set_server_check_status(check, HCHK_STATUS_L7STS, err); |
| 1352 | break; |
| 1353 | } |
| 1354 | |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1355 | default: |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 1356 | /* for other checks (eg: pure TCP), delegate to the main task */ |
Willy Tarreau | 1620ec3 | 2011-08-06 17:05:02 +0200 | [diff] [blame] | 1357 | break; |
| 1358 | } /* switch */ |
Willy Tarreau | 8374918 | 2007-04-15 20:56:27 +0200 | [diff] [blame] | 1359 | |
Willy Tarreau | c7dd71a | 2007-11-30 08:33:21 +0100 | [diff] [blame] | 1360 | out_wakeup: |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1361 | /* collect possible new errors */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 1362 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 1363 | chk_report_conn_err(check, 0, 0); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1364 | |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 1365 | /* Reset the check buffer... */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1366 | *b_head(&check->bi) = '\0'; |
| 1367 | b_reset(&check->bi); |
Nick Chalk | 57b1bf7 | 2010-03-16 15:50:46 +0000 | [diff] [blame] | 1368 | |
Steven Davidovitz | 544d481 | 2017-03-08 11:06:20 -0800 | [diff] [blame] | 1369 | /* Close the connection... We still attempt to nicely close if, |
| 1370 | * for instance, SSL needs to send a "close notify." Later, we perform |
| 1371 | * a hard close and reset the connection if some data are pending, |
| 1372 | * otherwise we end up with many TIME_WAITs and eat all the source port |
| 1373 | * range quickly. To avoid sending RSTs all the time, we first try to |
| 1374 | * drain pending data. |
Willy Tarreau | fd29cc5 | 2012-11-23 09:18:20 +0100 | [diff] [blame] | 1375 | */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1376 | __cs_stop_both(cs); |
Willy Tarreau | ecdb3fe | 2017-10-05 15:25:48 +0200 | [diff] [blame] | 1377 | cs_shutw(cs, CS_SHW_NORMAL); |
Willy Tarreau | 2b57cb8 | 2013-06-10 19:56:38 +0200 | [diff] [blame] | 1378 | |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1379 | /* OK, let's not stay here forever */ |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 1380 | if (check->result == CHK_RES_FAILED) |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1381 | conn->flags |= CO_FL_ERROR; |
| 1382 | |
Willy Tarreau | fdccded | 2008-08-29 18:19:04 +0200 | [diff] [blame] | 1383 | task_wakeup(t, TASK_WOKEN_IO); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 1384 | out: |
Willy Tarreau | 3267d36 | 2012-08-17 23:53:56 +0200 | [diff] [blame] | 1385 | return; |
Willy Tarreau | 0393818 | 2010-03-17 21:52:07 +0100 | [diff] [blame] | 1386 | |
| 1387 | wait_more_data: |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1388 | __cs_want_recv(cs); |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 1389 | cs->conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 1390 | goto out; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1391 | } |
| 1392 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1393 | /* |
| 1394 | * This function is used only for server health-checks. It handles connection |
| 1395 | * status updates including errors. If necessary, it wakes the check task up. |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1396 | * It returns 0 on normal cases, <0 if at least one close() has happened on the |
| 1397 | * connection (eg: reconnect). |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1398 | */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1399 | static int wake_srv_chk(struct conn_stream *cs) |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1400 | { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1401 | struct connection *conn = cs->conn; |
| 1402 | struct check *check = cs->data; |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1403 | int ret = 0; |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1404 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1405 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
Emeric Brun | f6ba17d | 2017-11-02 14:35:27 +0100 | [diff] [blame] | 1406 | |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 1407 | /* we may have to make progress on the TCP checks */ |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1408 | if (check->type == PR_O2_TCPCHK_CHK) { |
| 1409 | ret = tcpcheck_main(check); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1410 | cs = check->cs; |
Willy Tarreau | 543abd4 | 2018-09-20 11:25:12 +0200 | [diff] [blame] | 1411 | conn = cs->conn; |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 1412 | } else if (!(check->wait_list.wait_reason & SUB_CAN_SEND)) |
Olivier Houchard | 910b2bc | 2018-07-17 18:49:38 +0200 | [diff] [blame] | 1413 | __event_srv_chk_w(cs); |
Willy Tarreau | c09572f | 2017-10-04 11:58:22 +0200 | [diff] [blame] | 1414 | |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 1415 | if (unlikely(conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR)) { |
Willy Tarreau | 02b0f58 | 2013-12-03 15:42:33 +0100 | [diff] [blame] | 1416 | /* We may get error reports bypassing the I/O handlers, typically |
| 1417 | * the case when sending a pure TCP check which fails, then the I/O |
| 1418 | * handlers above are not called. This is completely handled by the |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1419 | * main processing task so let's simply wake it up. If we get here, |
| 1420 | * we expect errno to still be valid. |
| 1421 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 1422 | chk_report_conn_err(check, errno, 0); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1423 | __cs_stop_both(cs); |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1424 | task_wakeup(check->task, TASK_WOKEN_IO); |
| 1425 | } |
Willy Tarreau | 2500fc2 | 2018-04-03 19:31:38 +0200 | [diff] [blame] | 1426 | else if (!(conn->flags & CO_FL_HANDSHAKE) && !(cs->flags & (CS_FL_DATA_RD_ENA|CS_FL_DATA_WR_ENA))) { |
Willy Tarreau | 3be293f | 2014-02-05 18:31:24 +0100 | [diff] [blame] | 1427 | /* we may get here if only a connection probe was required : we |
| 1428 | * don't have any data to send nor anything expected in response, |
| 1429 | * so the completion of the connection establishment is enough. |
| 1430 | */ |
| 1431 | task_wakeup(check->task, TASK_WOKEN_IO); |
| 1432 | } |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1433 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 1434 | if (check->result != CHK_RES_UNKNOWN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 1435 | /* We're here because nobody wants to handle the error, so we |
| 1436 | * sure want to abort the hard way. |
Willy Tarreau | 02b0f58 | 2013-12-03 15:42:33 +0100 | [diff] [blame] | 1437 | */ |
Willy Tarreau | d85c485 | 2015-03-13 00:40:28 +0100 | [diff] [blame] | 1438 | conn_sock_drain(conn); |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 1439 | cs_close(cs); |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1440 | ret = -1; |
Willy Tarreau | 2d351b6 | 2013-12-05 02:36:25 +0100 | [diff] [blame] | 1441 | } |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1442 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1443 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
Emeric Brun | f6ba17d | 2017-11-02 14:35:27 +0100 | [diff] [blame] | 1444 | |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 1445 | /* if a connection got replaced, we must absolutely prevent the connection |
| 1446 | * handler from touching its fd, and perform the FD polling updates ourselves |
| 1447 | */ |
| 1448 | if (ret < 0) |
| 1449 | conn_cond_update_polling(conn); |
| 1450 | |
| 1451 | return ret; |
Willy Tarreau | 20bea42 | 2012-07-06 12:00:49 +0200 | [diff] [blame] | 1452 | } |
| 1453 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1454 | struct data_cb check_conn_cb = { |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1455 | .wake = wake_srv_chk, |
Willy Tarreau | 8e0bb0a | 2016-11-24 16:58:12 +0100 | [diff] [blame] | 1456 | .name = "CHCK", |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 1457 | }; |
| 1458 | |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1459 | /* |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1460 | * updates the server's weight during a warmup stage. Once the final weight is |
| 1461 | * reached, the task automatically stops. Note that any server status change |
| 1462 | * must have updated s->last_change accordingly. |
| 1463 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 1464 | 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] | 1465 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 1466 | struct server *s = context; |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1467 | |
| 1468 | /* by default, plan on stopping the task */ |
| 1469 | t->expire = TICK_ETERNITY; |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 1470 | if ((s->next_admin & SRV_ADMF_MAINT) || |
| 1471 | (s->next_state != SRV_ST_STARTING)) |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1472 | return t; |
| 1473 | |
Willy Tarreau | 892337c | 2014-05-13 23:41:20 +0200 | [diff] [blame] | 1474 | /* recalculate the weights and update the state */ |
Willy Tarreau | 3ff577e | 2018-08-02 11:48:52 +0200 | [diff] [blame] | 1475 | server_recalc_eweight(s, 1); |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1476 | |
| 1477 | /* probably that we can refill this server with a bit more connections */ |
Willy Tarreau | 4aac7db | 2014-05-16 11:48:10 +0200 | [diff] [blame] | 1478 | pendconn_grab_from_px(s); |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1479 | |
| 1480 | /* get back there in 1 second or 1/20th of the slowstart interval, |
| 1481 | * whichever is greater, resulting in small 5% steps. |
| 1482 | */ |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 1483 | if (s->next_state == SRV_ST_STARTING) |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1484 | t->expire = tick_add(now_ms, MS_TO_TICKS(MAX(1000, s->slowstart / 20))); |
| 1485 | return t; |
| 1486 | } |
| 1487 | |
Willy Tarreau | 894c642 | 2017-10-04 15:58:52 +0200 | [diff] [blame] | 1488 | /* returns the first NON-COMMENT tcp-check rule from list <list> or NULL if |
| 1489 | * none was found. |
| 1490 | */ |
| 1491 | static struct tcpcheck_rule *get_first_tcpcheck_rule(struct list *list) |
| 1492 | { |
| 1493 | struct tcpcheck_rule *r; |
| 1494 | |
| 1495 | list_for_each_entry(r, list, list) { |
| 1496 | if (r->action != TCPCHK_ACT_COMMENT) |
| 1497 | return r; |
| 1498 | } |
| 1499 | return NULL; |
| 1500 | } |
| 1501 | |
Willy Tarreau | 2e99390 | 2011-10-31 11:53:20 +0100 | [diff] [blame] | 1502 | /* |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1503 | * establish a server health-check that makes use of a connection. |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1504 | * |
| 1505 | * It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1506 | * - SF_ERR_NONE if everything's OK and tcpcheck_main() was not called |
| 1507 | * - SF_ERR_UP if if everything's OK and tcpcheck_main() was called |
| 1508 | * - SF_ERR_SRVTO if there are no more servers |
| 1509 | * - SF_ERR_SRVCL if the connection was refused by the server |
| 1510 | * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn) |
| 1511 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 1512 | * - SF_ERR_INTERNAL for any other purely internal errors |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1513 | * - 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] | 1514 | * 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] | 1515 | * Note that we try to prevent the network stack from sending the ACK during the |
| 1516 | * connect() when a pure TCP check is used (without PROXY protocol). |
| 1517 | */ |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1518 | static int connect_conn_chk(struct task *t) |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1519 | { |
| 1520 | struct check *check = t->context; |
| 1521 | struct server *s = check->server; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1522 | struct conn_stream *cs = check->cs; |
| 1523 | struct connection *conn = cs_conn(cs); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1524 | struct protocol *proto; |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1525 | struct tcpcheck_rule *tcp_rule = NULL; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1526 | int ret; |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 1527 | int quickack; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1528 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 1529 | /* we cannot have a connection here */ |
| 1530 | if (conn) |
| 1531 | return SF_ERR_INTERNAL; |
| 1532 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1533 | /* tcpcheck send/expect initialisation */ |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1534 | if (check->type == PR_O2_TCPCHK_CHK) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1535 | check->current_step = NULL; |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1536 | tcp_rule = get_first_tcpcheck_rule(check->tcpcheck_rules); |
| 1537 | } |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1538 | |
| 1539 | /* prepare the check buffer. |
| 1540 | * This should not be used if check is the secondary agent check |
| 1541 | * of a server as s->proxy->check_req will relate to the |
| 1542 | * configuration of the primary check. Similarly, tcp-check uses |
| 1543 | * its own strings. |
| 1544 | */ |
| 1545 | 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] | 1546 | b_putblk(&check->bo, s->proxy->check_req, s->proxy->check_len); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1547 | |
| 1548 | /* we want to check if this host replies to HTTP or SSLv3 requests |
| 1549 | * so we'll send the request, and won't wake the checker up now. |
| 1550 | */ |
| 1551 | if ((check->type) == PR_O2_SSL3_CHK) { |
| 1552 | /* SSL requires that we put Unix time in the request */ |
| 1553 | int gmt_time = htonl(date.tv_sec); |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1554 | memcpy(b_head(&check->bo) + 11, &gmt_time, 4); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1555 | } |
| 1556 | else if ((check->type) == PR_O2_HTTP_CHK) { |
| 1557 | if (s->proxy->options2 & PR_O2_CHK_SNDST) |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 1558 | b_putblk(&check->bo, trash.area, |
| 1559 | httpchk_build_status_header(s, trash.area, trash.size)); |
Cyril Bonté | 32602d2 | 2015-01-30 00:07:07 +0100 | [diff] [blame] | 1560 | /* prevent HTTP keep-alive when "http-check expect" is used */ |
| 1561 | if (s->proxy->options2 & PR_O2_EXP_TYPE) |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1562 | b_putist(&check->bo, ist("Connection: close\r\n")); |
| 1563 | b_putist(&check->bo, ist("\r\n")); |
| 1564 | *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */ |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1565 | } |
| 1566 | } |
| 1567 | |
James Brown | 55f9ff1 | 2015-10-21 18:19:05 -0700 | [diff] [blame] | 1568 | if ((check->type & PR_O2_LB_AGENT_CHK) && check->send_string_len) { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 1569 | b_putblk(&check->bo, check->send_string, check->send_string_len); |
James Brown | 55f9ff1 | 2015-10-21 18:19:05 -0700 | [diff] [blame] | 1570 | } |
| 1571 | |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1572 | /* for tcp-checks, the initial connection setup is handled separately as |
| 1573 | * it may be sent to a specific port and not to the server's. |
| 1574 | */ |
| 1575 | if (tcp_rule && tcp_rule->action == TCPCHK_ACT_CONNECT) { |
| 1576 | tcpcheck_main(check); |
| 1577 | return SF_ERR_UP; |
| 1578 | } |
| 1579 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1580 | /* prepare a new connection */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1581 | cs = check->cs = cs_new(NULL); |
| 1582 | if (!check->cs) |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 1583 | return SF_ERR_RESOURCE; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 1584 | conn = cs->conn; |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 1585 | /* Maybe there were an older connection we were waiting on */ |
| 1586 | check->wait_list.wait_reason = 0; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1587 | |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 1588 | if (is_addr(&check->addr)) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1589 | /* we'll connect to the check addr specified on the server */ |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 1590 | conn->addr.to = check->addr; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1591 | } |
| 1592 | else { |
| 1593 | /* we'll connect to the addr on the server */ |
| 1594 | conn->addr.to = s->addr; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1595 | } |
| 1596 | |
Olivier Houchard | 6377a00 | 2017-12-01 22:04:05 +0100 | [diff] [blame] | 1597 | proto = protocol_by_family(conn->addr.to.ss_family); |
Olivier Houchard | 6377a00 | 2017-12-01 22:04:05 +0100 | [diff] [blame] | 1598 | conn->target = &s->obj_type; |
| 1599 | |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 1600 | if ((conn->addr.to.ss_family == AF_INET) || (conn->addr.to.ss_family == AF_INET6)) { |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1601 | int i = 0; |
| 1602 | |
| 1603 | i = srv_check_healthcheck_port(check); |
Olivier Houchard | 6377a00 | 2017-12-01 22:04:05 +0100 | [diff] [blame] | 1604 | if (i == 0) |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1605 | return SF_ERR_CHK_PORT; |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 1606 | |
| 1607 | set_host_port(&conn->addr.to, i); |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1608 | } |
| 1609 | |
Thierry FOURNIER | bb2ae64 | 2015-01-14 11:31:49 +0100 | [diff] [blame] | 1610 | /* no client address */ |
| 1611 | clear_addr(&conn->addr.from); |
| 1612 | |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 1613 | conn_prepare(conn, proto, check->xprt); |
Willy Tarreau | 175a2bb | 2018-09-12 12:02:05 +0200 | [diff] [blame] | 1614 | conn_install_mux(conn, &mux_pt_ops, cs, s->proxy); |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 1615 | cs_attach(cs, check, &check_conn_cb); |
| 1616 | |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 1617 | /* only plain tcp-check supports quick ACK */ |
| 1618 | quickack = check->type == 0 || check->type == PR_O2_TCPCHK_CHK; |
| 1619 | |
Willy Tarreau | f411cce | 2017-10-04 16:21:19 +0200 | [diff] [blame] | 1620 | if (tcp_rule && tcp_rule->action == TCPCHK_ACT_EXPECT) |
| 1621 | quickack = 0; |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1622 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1623 | ret = SF_ERR_INTERNAL; |
Olivier Houchard | b68fda4 | 2017-08-04 18:39:01 +0200 | [diff] [blame] | 1624 | if (proto && proto->connect) |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 1625 | ret = proto->connect(conn, check->type, quickack ? 2 : 0); |
Willy Tarreau | 16257f6 | 2017-11-02 15:45:00 +0100 | [diff] [blame] | 1626 | |
| 1627 | if (check->type) |
| 1628 | cs_want_send(cs); |
| 1629 | |
Olivier Houchard | 9130a96 | 2017-10-17 17:33:43 +0200 | [diff] [blame] | 1630 | #ifdef USE_OPENSSL |
| 1631 | if (s->check.sni) |
| 1632 | ssl_sock_set_servername(conn, s->check.sni); |
| 1633 | #endif |
Willy Tarreau | f494977 | 2017-05-06 08:45:28 +0200 | [diff] [blame] | 1634 | if (s->check.send_proxy && !(check->state & CHK_ST_AGENT)) { |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1635 | conn->send_proxy_ofs = 1; |
| 1636 | conn->flags |= CO_FL_SEND_PROXY; |
| 1637 | } |
| 1638 | |
| 1639 | return ret; |
| 1640 | } |
| 1641 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1642 | static struct list pid_list = LIST_HEAD_INIT(pid_list); |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1643 | static struct pool_head *pool_head_pid_list; |
Christopher Faulet | 9dcf9b6 | 2017-11-13 10:34:01 +0100 | [diff] [blame] | 1644 | __decl_hathreads(HA_SPINLOCK_T pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1645 | |
| 1646 | void block_sigchld(void) |
| 1647 | { |
| 1648 | sigset_t set; |
| 1649 | sigemptyset(&set); |
| 1650 | sigaddset(&set, SIGCHLD); |
William Lallemand | 6e1796e | 2018-06-07 11:23:40 +0200 | [diff] [blame] | 1651 | assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1652 | } |
| 1653 | |
| 1654 | void unblock_sigchld(void) |
| 1655 | { |
| 1656 | sigset_t set; |
| 1657 | sigemptyset(&set); |
Willy Tarreau | ebc9244 | 2016-06-21 17:29:46 +0200 | [diff] [blame] | 1658 | sigaddset(&set, SIGCHLD); |
William Lallemand | 6e1796e | 2018-06-07 11:23:40 +0200 | [diff] [blame] | 1659 | assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1660 | } |
| 1661 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1662 | static struct pid_list *pid_list_add(pid_t pid, struct task *t) |
| 1663 | { |
| 1664 | struct pid_list *elem; |
| 1665 | struct check *check = t->context; |
| 1666 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1667 | elem = pool_alloc(pool_head_pid_list); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1668 | if (!elem) |
| 1669 | return NULL; |
| 1670 | elem->pid = pid; |
| 1671 | elem->t = t; |
| 1672 | elem->exited = 0; |
| 1673 | check->curpid = elem; |
| 1674 | LIST_INIT(&elem->list); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1675 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1676 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1677 | LIST_ADD(&pid_list, &elem->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1678 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1679 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1680 | return elem; |
| 1681 | } |
| 1682 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1683 | static void pid_list_del(struct pid_list *elem) |
| 1684 | { |
| 1685 | struct check *check; |
| 1686 | |
| 1687 | if (!elem) |
| 1688 | return; |
| 1689 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1690 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1691 | LIST_DEL(&elem->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1692 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1693 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1694 | if (!elem->exited) |
| 1695 | kill(elem->pid, SIGTERM); |
| 1696 | |
| 1697 | check = elem->t->context; |
| 1698 | check->curpid = NULL; |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1699 | pool_free(pool_head_pid_list, elem); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1700 | } |
| 1701 | |
| 1702 | /* Called from inside SIGCHLD handler, SIGCHLD is blocked */ |
| 1703 | static void pid_list_expire(pid_t pid, int status) |
| 1704 | { |
| 1705 | struct pid_list *elem; |
| 1706 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1707 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1708 | list_for_each_entry(elem, &pid_list, list) { |
| 1709 | if (elem->pid == pid) { |
| 1710 | elem->t->expire = now_ms; |
| 1711 | elem->status = status; |
| 1712 | elem->exited = 1; |
Cyril Bonté | 9dbcfab | 2014-08-07 01:55:39 +0200 | [diff] [blame] | 1713 | task_wakeup(elem->t, TASK_WOKEN_IO); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1714 | break; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1715 | } |
| 1716 | } |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1717 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1718 | } |
| 1719 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1720 | static void sigchld_handler(struct sig_handler *sh) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1721 | { |
| 1722 | pid_t pid; |
| 1723 | int status; |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1724 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1725 | while ((pid = waitpid(0, &status, WNOHANG)) > 0) |
| 1726 | pid_list_expire(pid, status); |
| 1727 | } |
| 1728 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1729 | static int init_pid_list(void) |
| 1730 | { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1731 | if (pool_head_pid_list != NULL) |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1732 | /* Nothing to do */ |
| 1733 | return 0; |
| 1734 | |
Willy Tarreau | 48d6bf2 | 2016-06-21 16:27:34 +0200 | [diff] [blame] | 1735 | if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1736 | ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n", |
| 1737 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1738 | return 1; |
| 1739 | } |
| 1740 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 1741 | pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED); |
| 1742 | if (pool_head_pid_list == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1743 | ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n", |
| 1744 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1745 | return 1; |
| 1746 | } |
| 1747 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 1748 | HA_SPIN_INIT(&pid_list_lock); |
Christopher Faulet | cfda847 | 2017-10-20 15:40:23 +0200 | [diff] [blame] | 1749 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1750 | return 0; |
| 1751 | } |
| 1752 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1753 | /* helper macro to set an environment variable and jump to a specific label on failure. */ |
| 1754 | #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] | 1755 | |
| 1756 | /* |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1757 | * helper function to allocate enough memory to store an environment variable. |
| 1758 | * It will also check that the environment variable is updatable, and silently |
| 1759 | * fail if not. |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1760 | */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1761 | static int extchk_setenv(struct check *check, int idx, const char *value) |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1762 | { |
| 1763 | int len, ret; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1764 | char *envname; |
| 1765 | int vmaxlen; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1766 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1767 | if (idx < 0 || idx >= EXTCHK_SIZE) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1768 | ha_alert("Illegal environment variable index %d. Aborting.\n", idx); |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1769 | return 1; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1770 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1771 | |
| 1772 | envname = extcheck_envs[idx].name; |
| 1773 | vmaxlen = extcheck_envs[idx].vmaxlen; |
| 1774 | |
| 1775 | /* Check if the environment variable is already set, and silently reject |
| 1776 | * the update if this one is not updatable. */ |
| 1777 | if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx])) |
| 1778 | return 0; |
| 1779 | |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1780 | /* Instead of sending NOT_USED, sending an empty value is preferable */ |
| 1781 | if (strcmp(value, "NOT_USED") == 0) { |
| 1782 | value = ""; |
| 1783 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1784 | |
| 1785 | len = strlen(envname) + 1; |
| 1786 | if (vmaxlen == EXTCHK_SIZE_EVAL_INIT) |
| 1787 | len += strlen(value); |
| 1788 | else |
| 1789 | len += vmaxlen; |
| 1790 | |
| 1791 | if (!check->envp[idx]) |
| 1792 | check->envp[idx] = malloc(len + 1); |
| 1793 | |
| 1794 | if (!check->envp[idx]) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1795 | 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] | 1796 | return 1; |
| 1797 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1798 | ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1799 | if (ret < 0) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1800 | 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] | 1801 | return 1; |
| 1802 | } |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1803 | else if (ret > len) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1804 | ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1805 | return 1; |
| 1806 | } |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1807 | return 0; |
| 1808 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1809 | |
| 1810 | static int prepare_external_check(struct check *check) |
| 1811 | { |
| 1812 | struct server *s = check->server; |
| 1813 | struct proxy *px = s->proxy; |
| 1814 | struct listener *listener = NULL, *l; |
| 1815 | int i; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1816 | const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1817 | char buf[256]; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1818 | |
| 1819 | list_for_each_entry(l, &px->conf.listeners, by_fe) |
| 1820 | /* Use the first INET, INET6 or UNIX listener */ |
| 1821 | if (l->addr.ss_family == AF_INET || |
| 1822 | l->addr.ss_family == AF_INET6 || |
| 1823 | l->addr.ss_family == AF_UNIX) { |
| 1824 | listener = l; |
| 1825 | break; |
| 1826 | } |
| 1827 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1828 | check->curpid = NULL; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1829 | check->envp = calloc((EXTCHK_SIZE + 1), sizeof(char *)); |
| 1830 | if (!check->envp) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1831 | ha_alert("Failed to allocate memory for environment variables. Aborting\n"); |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1832 | goto err; |
| 1833 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1834 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1835 | check->argv = calloc(6, sizeof(char *)); |
| 1836 | if (!check->argv) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1837 | 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] | 1838 | goto err; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1839 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1840 | |
| 1841 | check->argv[0] = px->check_command; |
| 1842 | |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1843 | if (!listener) { |
| 1844 | check->argv[1] = strdup("NOT_USED"); |
| 1845 | check->argv[2] = strdup("NOT_USED"); |
| 1846 | } |
| 1847 | else if (listener->addr.ss_family == AF_INET || |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1848 | listener->addr.ss_family == AF_INET6) { |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1849 | addr_to_str(&listener->addr, buf, sizeof(buf)); |
| 1850 | check->argv[1] = strdup(buf); |
| 1851 | port_to_str(&listener->addr, buf, sizeof(buf)); |
| 1852 | check->argv[2] = strdup(buf); |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1853 | } |
| 1854 | else if (listener->addr.ss_family == AF_UNIX) { |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1855 | const struct sockaddr_un *un; |
| 1856 | |
| 1857 | un = (struct sockaddr_un *)&listener->addr; |
| 1858 | check->argv[1] = strdup(un->sun_path); |
| 1859 | check->argv[2] = strdup("NOT_USED"); |
Cyril Bonté | 777be86 | 2014-12-02 21:21:35 +0100 | [diff] [blame] | 1860 | } |
| 1861 | else { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1862 | 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] | 1863 | goto err; |
| 1864 | } |
| 1865 | |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1866 | addr_to_str(&s->addr, buf, sizeof(buf)); |
| 1867 | check->argv[3] = strdup(buf); |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 1868 | |
| 1869 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 1870 | snprintf(buf, sizeof(buf), "%u", s->svc_port); |
| 1871 | else |
| 1872 | *buf = 0; |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1873 | check->argv[4] = strdup(buf); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1874 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1875 | for (i = 0; i < 5; i++) { |
| 1876 | if (!check->argv[i]) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1877 | 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] | 1878 | goto err; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1879 | } |
| 1880 | } |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1881 | |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1882 | EXTCHK_SETENV(check, EXTCHK_PATH, path, err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1883 | /* Add proxy environment variables */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1884 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err); |
| 1885 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err); |
| 1886 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err); |
| 1887 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1888 | /* Add server environment variables */ |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1889 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err); |
| 1890 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err); |
| 1891 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err); |
| 1892 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err); |
| 1893 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err); |
| 1894 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err); |
| 1895 | |
| 1896 | /* Ensure that we don't leave any hole in check->envp */ |
| 1897 | for (i = 0; i < EXTCHK_SIZE; i++) |
| 1898 | if (!check->envp[i]) |
| 1899 | EXTCHK_SETENV(check, i, "", err); |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1900 | |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 1901 | return 1; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1902 | err: |
| 1903 | if (check->envp) { |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1904 | for (i = 0; i < EXTCHK_SIZE; i++) |
Cyril Bonté | 9ede66b | 2014-12-02 21:21:36 +0100 | [diff] [blame] | 1905 | free(check->envp[i]); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1906 | free(check->envp); |
| 1907 | check->envp = NULL; |
| 1908 | } |
| 1909 | |
| 1910 | if (check->argv) { |
| 1911 | for (i = 1; i < 5; i++) |
| 1912 | free(check->argv[i]); |
| 1913 | free(check->argv); |
| 1914 | check->argv = NULL; |
| 1915 | } |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 1916 | return 0; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1917 | } |
| 1918 | |
Simon Horman | b00d17a | 2014-06-13 16:18:16 +0900 | [diff] [blame] | 1919 | /* |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1920 | * establish a server health-check that makes use of a process. |
| 1921 | * |
| 1922 | * It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1923 | * - SF_ERR_NONE if everything's OK |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1924 | * - 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] | 1925 | * 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] | 1926 | * |
| 1927 | * Blocks and then unblocks SIGCHLD |
| 1928 | */ |
| 1929 | static int connect_proc_chk(struct task *t) |
| 1930 | { |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1931 | char buf[256]; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1932 | struct check *check = t->context; |
| 1933 | struct server *s = check->server; |
| 1934 | struct proxy *px = s->proxy; |
| 1935 | int status; |
| 1936 | pid_t pid; |
| 1937 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1938 | status = SF_ERR_RESOURCE; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1939 | |
| 1940 | block_sigchld(); |
| 1941 | |
| 1942 | pid = fork(); |
| 1943 | if (pid < 0) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1944 | ha_alert("Failed to fork process for external health check: %s. Aborting.\n", |
| 1945 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1946 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 1947 | goto out; |
| 1948 | } |
| 1949 | if (pid == 0) { |
| 1950 | /* Child */ |
| 1951 | extern char **environ; |
Willy Tarreau | b7b2478 | 2016-06-21 15:32:29 +0200 | [diff] [blame] | 1952 | int fd; |
| 1953 | |
| 1954 | /* close all FDs. Keep stdin/stdout/stderr in verbose mode */ |
| 1955 | fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3; |
| 1956 | |
| 1957 | while (fd < global.rlimit_nofile) |
| 1958 | close(fd++); |
| 1959 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1960 | environ = check->envp; |
Cyril Bonté | ac92a06 | 2014-12-27 22:28:38 +0100 | [diff] [blame] | 1961 | extchk_setenv(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf))); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1962 | execvp(px->check_command, check->argv); |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 1963 | ha_alert("Failed to exec process for external health check: %s. Aborting.\n", |
| 1964 | strerror(errno)); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1965 | exit(-1); |
| 1966 | } |
| 1967 | |
| 1968 | /* Parent */ |
| 1969 | if (check->result == CHK_RES_UNKNOWN) { |
| 1970 | if (pid_list_add(pid, t) != NULL) { |
| 1971 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 1972 | |
| 1973 | if (px->timeout.check && px->timeout.connect) { |
| 1974 | int t_con = tick_add(now_ms, px->timeout.connect); |
| 1975 | t->expire = tick_first(t->expire, t_con); |
| 1976 | } |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 1977 | status = SF_ERR_NONE; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 1978 | goto out; |
| 1979 | } |
| 1980 | else { |
| 1981 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 1982 | } |
| 1983 | kill(pid, SIGTERM); /* process creation error */ |
| 1984 | } |
| 1985 | else |
| 1986 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 1987 | |
| 1988 | out: |
| 1989 | unblock_sigchld(); |
| 1990 | return status; |
| 1991 | } |
| 1992 | |
| 1993 | /* |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 1994 | * manages a server health-check that uses an external process. Returns |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1995 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 1996 | * |
| 1997 | * Please do NOT place any return statement in this function and only leave |
| 1998 | * via the out_unlock label. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1999 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2000 | 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] | 2001 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2002 | struct check *check = context; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2003 | struct server *s = check->server; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2004 | int rv; |
| 2005 | int ret; |
| 2006 | int expired = tick_is_expired(t->expire, now_ms); |
| 2007 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2008 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2009 | if (!(check->state & CHK_ST_INPROGRESS)) { |
| 2010 | /* no check currently running */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2011 | if (!expired) /* woke up too early */ |
| 2012 | goto out_unlock; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2013 | |
| 2014 | /* we don't send any health-checks when the proxy is |
| 2015 | * stopped, the server should not be checked or the check |
| 2016 | * is disabled. |
| 2017 | */ |
| 2018 | if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) || |
| 2019 | s->proxy->state == PR_STSTOPPED) |
| 2020 | goto reschedule; |
| 2021 | |
| 2022 | /* we'll initiate a new check */ |
| 2023 | set_server_check_status(check, HCHK_STATUS_START, NULL); |
| 2024 | |
| 2025 | check->state |= CHK_ST_INPROGRESS; |
| 2026 | |
Simon Horman | dbf7019 | 2015-01-30 11:22:53 +0900 | [diff] [blame] | 2027 | ret = connect_proc_chk(t); |
Willy Tarreau | d7c3fbd | 2017-10-04 15:19:26 +0200 | [diff] [blame] | 2028 | if (ret == SF_ERR_NONE) { |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 2029 | /* the process was forked, we allow up to min(inter, |
| 2030 | * timeout.connect) for it to report its status, but |
| 2031 | * only when timeout.check is set as it may be to short |
| 2032 | * for a full check otherwise. |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2033 | */ |
| 2034 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 2035 | |
| 2036 | if (s->proxy->timeout.check && s->proxy->timeout.connect) { |
| 2037 | int t_con = tick_add(now_ms, s->proxy->timeout.connect); |
| 2038 | t->expire = tick_first(t->expire, t_con); |
| 2039 | } |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2040 | task_set_affinity(t, tid_bit); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2041 | goto reschedule; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2042 | } |
| 2043 | |
Willy Tarreau | 1e62e2a | 2017-10-04 15:07:02 +0200 | [diff] [blame] | 2044 | /* here, we failed to start the check */ |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2045 | |
| 2046 | check->state &= ~CHK_ST_INPROGRESS; |
| 2047 | check_notify_failure(check); |
| 2048 | |
| 2049 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2050 | * to establish but only when timeout.check is set |
| 2051 | * as it may be to short for a full check otherwise |
| 2052 | */ |
| 2053 | while (tick_is_expired(t->expire, now_ms)) { |
| 2054 | int t_con; |
| 2055 | |
| 2056 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
| 2057 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 2058 | |
| 2059 | if (s->proxy->timeout.check) |
| 2060 | t->expire = tick_first(t->expire, t_con); |
| 2061 | } |
| 2062 | } |
| 2063 | else { |
| 2064 | /* there was a test running. |
| 2065 | * First, let's check whether there was an uncaught error, |
| 2066 | * which can happen on connect timeout or error. |
| 2067 | */ |
| 2068 | if (check->result == CHK_RES_UNKNOWN) { |
| 2069 | /* good connection is enough for pure TCP check */ |
| 2070 | struct pid_list *elem = check->curpid; |
| 2071 | int status = HCHK_STATUS_UNKNOWN; |
| 2072 | |
| 2073 | if (elem->exited) { |
| 2074 | status = elem->status; /* Save in case the process exits between use below */ |
| 2075 | if (!WIFEXITED(status)) |
| 2076 | check->code = -1; |
| 2077 | else |
| 2078 | check->code = WEXITSTATUS(status); |
| 2079 | if (!WIFEXITED(status) || WEXITSTATUS(status)) |
| 2080 | status = HCHK_STATUS_PROCERR; |
| 2081 | else |
| 2082 | status = HCHK_STATUS_PROCOK; |
| 2083 | } else if (expired) { |
| 2084 | status = HCHK_STATUS_PROCTOUT; |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2085 | ha_warning("kill %d\n", (int)elem->pid); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2086 | kill(elem->pid, SIGTERM); |
| 2087 | } |
| 2088 | set_server_check_status(check, status, NULL); |
| 2089 | } |
| 2090 | |
| 2091 | if (check->result == CHK_RES_FAILED) { |
| 2092 | /* a failure or timeout detected */ |
| 2093 | check_notify_failure(check); |
| 2094 | } |
| 2095 | else if (check->result == CHK_RES_CONDPASS) { |
| 2096 | /* check is OK but asks for stopping mode */ |
| 2097 | check_notify_stopping(check); |
| 2098 | } |
| 2099 | else if (check->result == CHK_RES_PASSED) { |
| 2100 | /* a success was detected */ |
| 2101 | check_notify_success(check); |
| 2102 | } |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2103 | task_set_affinity(t, MAX_THREADS_MASK); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2104 | check->state &= ~CHK_ST_INPROGRESS; |
| 2105 | |
| 2106 | pid_list_del(check->curpid); |
| 2107 | |
| 2108 | rv = 0; |
| 2109 | if (global.spread_checks > 0) { |
| 2110 | rv = srv_getinter(check) * global.spread_checks / 100; |
| 2111 | rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0))); |
| 2112 | } |
| 2113 | t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv)); |
| 2114 | } |
| 2115 | |
| 2116 | reschedule: |
| 2117 | while (tick_is_expired(t->expire, now_ms)) |
| 2118 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2119 | |
| 2120 | out_unlock: |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2121 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2122 | return t; |
| 2123 | } |
| 2124 | |
| 2125 | /* |
| 2126 | * manages a server health-check that uses a connection. Returns |
| 2127 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2128 | * |
| 2129 | * Please do NOT place any return statement in this function and only leave |
| 2130 | * via the out_unlock label. |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2131 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2132 | 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] | 2133 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2134 | struct check *check = context; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2135 | struct server *s = check->server; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2136 | struct conn_stream *cs = check->cs; |
| 2137 | struct connection *conn = cs_conn(cs); |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2138 | int rv; |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2139 | int ret; |
Willy Tarreau | acbdc7a | 2012-11-23 14:02:10 +0100 | [diff] [blame] | 2140 | int expired = tick_is_expired(t->expire, now_ms); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2141 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2142 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2143 | if (!(check->state & CHK_ST_INPROGRESS)) { |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2144 | /* no check currently running */ |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2145 | if (!expired) /* woke up too early */ |
| 2146 | goto out_unlock; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2147 | |
Simon Horman | 671b6f0 | 2013-11-25 10:46:39 +0900 | [diff] [blame] | 2148 | /* we don't send any health-checks when the proxy is |
| 2149 | * stopped, the server should not be checked or the check |
| 2150 | * is disabled. |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2151 | */ |
Willy Tarreau | 0d924cc | 2013-12-11 21:26:24 +0100 | [diff] [blame] | 2152 | if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) || |
Willy Tarreau | 33a08db | 2013-12-11 21:03:31 +0100 | [diff] [blame] | 2153 | s->proxy->state == PR_STSTOPPED) |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2154 | goto reschedule; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2155 | |
| 2156 | /* we'll initiate a new check */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2157 | set_server_check_status(check, HCHK_STATUS_START, NULL); |
Willy Tarreau | 1ae1b7b | 2012-09-28 15:28:30 +0200 | [diff] [blame] | 2158 | |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2159 | check->state |= CHK_ST_INPROGRESS; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2160 | b_reset(&check->bi); |
| 2161 | b_reset(&check->bo); |
Willy Tarreau | 1ae1b7b | 2012-09-28 15:28:30 +0200 | [diff] [blame] | 2162 | |
Simon Horman | dbf7019 | 2015-01-30 11:22:53 +0900 | [diff] [blame] | 2163 | ret = connect_conn_chk(t); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2164 | cs = check->cs; |
| 2165 | conn = cs_conn(cs); |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2166 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2167 | switch (ret) { |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2168 | case SF_ERR_UP: |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2169 | goto out_unlock; |
| 2170 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2171 | case SF_ERR_NONE: |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2172 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2173 | * to establish but only when timeout.check is set |
| 2174 | * as it may be to short for a full check otherwise |
| 2175 | */ |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2176 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
Krzysztof Piotr Oledzki | 0960541 | 2009-09-23 22:09:24 +0200 | [diff] [blame] | 2177 | |
Willy Tarreau | fb56aab | 2012-09-28 14:40:02 +0200 | [diff] [blame] | 2178 | if (s->proxy->timeout.check && s->proxy->timeout.connect) { |
| 2179 | int t_con = tick_add(now_ms, s->proxy->timeout.connect); |
| 2180 | t->expire = tick_first(t->expire, t_con); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2181 | } |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 2182 | |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 2183 | if (check->type) { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2184 | cs_want_recv(cs); /* prepare for reading a possible reply */ |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 2185 | __event_srv_chk_r(cs); |
| 2186 | } |
Willy Tarreau | 06559ac | 2013-12-05 01:53:08 +0100 | [diff] [blame] | 2187 | |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2188 | task_set_affinity(t, tid_bit); |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2189 | goto reschedule; |
| 2190 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2191 | case SF_ERR_SRVTO: /* ETIMEDOUT */ |
| 2192 | case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2193 | if (conn) |
| 2194 | conn->flags |= CO_FL_ERROR; |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2195 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2196 | break; |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 2197 | /* should share same code than cases below */ |
| 2198 | case SF_ERR_CHK_PORT: |
| 2199 | check->state |= CHK_ST_PORT_MISS; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2200 | case SF_ERR_PRXCOND: |
| 2201 | case SF_ERR_RESOURCE: |
| 2202 | case SF_ERR_INTERNAL: |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2203 | if (conn) |
| 2204 | conn->flags |= CO_FL_ERROR; |
| 2205 | chk_report_conn_err(check, conn ? 0 : ENOMEM, 0); |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2206 | break; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2207 | } |
| 2208 | |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2209 | /* here, we have seen a synchronous error, no fd was allocated */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2210 | if (cs) { |
| 2211 | cs_destroy(cs); |
| 2212 | cs = check->cs = NULL; |
| 2213 | conn = NULL; |
Olivier Houchard | 390485a | 2017-10-24 19:03:30 +0200 | [diff] [blame] | 2214 | } |
Willy Tarreau | 6b0a850 | 2012-11-23 08:51:32 +0100 | [diff] [blame] | 2215 | |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2216 | check->state &= ~CHK_ST_INPROGRESS; |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 2217 | check_notify_failure(check); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2218 | |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2219 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2220 | * to establish but only when timeout.check is set |
| 2221 | * as it may be to short for a full check otherwise |
| 2222 | */ |
Willy Tarreau | 0c303ee | 2008-07-07 00:09:58 +0200 | [diff] [blame] | 2223 | while (tick_is_expired(t->expire, now_ms)) { |
| 2224 | int t_con; |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2225 | |
Willy Tarreau | 0c303ee | 2008-07-07 00:09:58 +0200 | [diff] [blame] | 2226 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2227 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2228 | |
Willy Tarreau | 0c303ee | 2008-07-07 00:09:58 +0200 | [diff] [blame] | 2229 | if (s->proxy->timeout.check) |
| 2230 | t->expire = tick_first(t->expire, t_con); |
Krzysztof Piotr Oledzki | 5259dfe | 2008-01-21 01:54:06 +0100 | [diff] [blame] | 2231 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2232 | } |
| 2233 | else { |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2234 | /* there was a test running. |
| 2235 | * First, let's check whether there was an uncaught error, |
| 2236 | * which can happen on connect timeout or error. |
| 2237 | */ |
Simon Horman | ccaabcd | 2014-06-20 12:29:47 +0900 | [diff] [blame] | 2238 | if (check->result == CHK_RES_UNKNOWN) { |
Willy Tarreau | 25e2ab5 | 2013-12-04 11:17:05 +0100 | [diff] [blame] | 2239 | /* good connection is enough for pure TCP check */ |
| 2240 | if ((conn->flags & CO_FL_CONNECTED) && !check->type) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2241 | if (check->use_ssl) |
| 2242 | set_server_check_status(check, HCHK_STATUS_L6OK, NULL); |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2243 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2244 | set_server_check_status(check, HCHK_STATUS_L4OK, NULL); |
Willy Tarreau | acbdc7a | 2012-11-23 14:02:10 +0100 | [diff] [blame] | 2245 | } |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 2246 | 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] | 2247 | chk_report_conn_err(check, 0, expired); |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2248 | } |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2249 | else |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2250 | goto out_unlock; /* timeout not reached, wait again */ |
Willy Tarreau | f150317 | 2012-09-28 19:39:36 +0200 | [diff] [blame] | 2251 | } |
| 2252 | |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2253 | /* check complete or aborted */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2254 | if (conn && conn->xprt) { |
Willy Tarreau | 5ba04f6 | 2013-02-12 15:23:12 +0100 | [diff] [blame] | 2255 | /* The check was aborted and the connection was not yet closed. |
| 2256 | * This can happen upon timeout, or when an external event such |
| 2257 | * as a failed response coupled with "observe layer7" caused the |
| 2258 | * server state to be suddenly changed. |
| 2259 | */ |
Willy Tarreau | d85c485 | 2015-03-13 00:40:28 +0100 | [diff] [blame] | 2260 | conn_sock_drain(conn); |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 2261 | cs_close(cs); |
Willy Tarreau | 5ba04f6 | 2013-02-12 15:23:12 +0100 | [diff] [blame] | 2262 | } |
| 2263 | |
Willy Tarreau | ac59f36 | 2017-10-08 11:10:19 +0200 | [diff] [blame] | 2264 | if (cs) { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2265 | cs_destroy(cs); |
| 2266 | cs = check->cs = NULL; |
| 2267 | conn = NULL; |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2268 | } |
| 2269 | |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 2270 | if (check->result == CHK_RES_FAILED) { |
| 2271 | /* a failure or timeout detected */ |
Willy Tarreau | 4eec547 | 2014-05-20 22:32:27 +0200 | [diff] [blame] | 2272 | check_notify_failure(check); |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 2273 | } |
Willy Tarreau | db58b79 | 2014-05-21 13:57:23 +0200 | [diff] [blame] | 2274 | else if (check->result == CHK_RES_CONDPASS) { |
| 2275 | /* check is OK but asks for stopping mode */ |
| 2276 | check_notify_stopping(check); |
Willy Tarreau | af54958 | 2014-05-16 17:37:50 +0200 | [diff] [blame] | 2277 | } |
Willy Tarreau | 3e04838 | 2014-05-21 10:30:54 +0200 | [diff] [blame] | 2278 | else if (check->result == CHK_RES_PASSED) { |
| 2279 | /* a success was detected */ |
| 2280 | check_notify_success(check); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2281 | } |
Christopher Faulet | 88ce5d1 | 2017-10-20 15:41:18 +0200 | [diff] [blame] | 2282 | task_set_affinity(t, MAX_THREADS_MASK); |
Willy Tarreau | 2c115e5 | 2013-12-11 19:41:16 +0100 | [diff] [blame] | 2283 | check->state &= ~CHK_ST_INPROGRESS; |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2284 | |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2285 | rv = 0; |
| 2286 | if (global.spread_checks > 0) { |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2287 | rv = srv_getinter(check) * global.spread_checks / 100; |
Willy Tarreau | 74fa7fb | 2012-11-23 14:43:49 +0100 | [diff] [blame] | 2288 | rv -= (int) (2 * rv * (rand() / (RAND_MAX + 1.0))); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2289 | } |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2290 | 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] | 2291 | } |
Willy Tarreau | 5a78f36 | 2012-11-23 12:47:05 +0100 | [diff] [blame] | 2292 | |
| 2293 | reschedule: |
| 2294 | while (tick_is_expired(t->expire, now_ms)) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2295 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2296 | out_unlock: |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 2297 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
Willy Tarreau | 26c2506 | 2009-03-08 09:38:41 +0100 | [diff] [blame] | 2298 | return t; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2299 | } |
| 2300 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2301 | /* |
| 2302 | * manages a server health-check. Returns |
| 2303 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
| 2304 | */ |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2305 | 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] | 2306 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2307 | struct check *check = context; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2308 | |
| 2309 | if (check->type == PR_O2_EXT_CHK) |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 2310 | return process_chk_proc(t, context, state); |
| 2311 | return process_chk_conn(t, context, state); |
Baptiste Assmann | a68ca96 | 2015-04-14 01:15:08 +0200 | [diff] [blame] | 2312 | |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2313 | } |
| 2314 | |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2315 | static int start_check_task(struct check *check, int mininter, |
| 2316 | int nbcheck, int srvpos) |
| 2317 | { |
| 2318 | struct task *t; |
| 2319 | /* task for the check */ |
Emeric Brun | c60def8 | 2017-09-27 14:59:38 +0200 | [diff] [blame] | 2320 | if ((t = task_new(MAX_THREADS_MASK)) == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2321 | ha_alert("Starting [%s:%s] check: out of memory.\n", |
| 2322 | check->server->proxy->id, check->server->id); |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2323 | return 0; |
| 2324 | } |
| 2325 | |
| 2326 | check->task = t; |
| 2327 | t->process = process_chk; |
| 2328 | t->context = check; |
| 2329 | |
Willy Tarreau | 1746eec | 2014-04-25 10:46:47 +0200 | [diff] [blame] | 2330 | if (mininter < srv_getinter(check)) |
| 2331 | mininter = srv_getinter(check); |
| 2332 | |
| 2333 | if (global.max_spread_checks && mininter > global.max_spread_checks) |
| 2334 | mininter = global.max_spread_checks; |
| 2335 | |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2336 | /* check this every ms */ |
Willy Tarreau | 1746eec | 2014-04-25 10:46:47 +0200 | [diff] [blame] | 2337 | t->expire = tick_add(now_ms, MS_TO_TICKS(mininter * srvpos / nbcheck)); |
Simon Horman | 5c94242 | 2013-11-25 10:46:32 +0900 | [diff] [blame] | 2338 | check->start = now; |
| 2339 | task_queue(t); |
| 2340 | |
| 2341 | return 1; |
| 2342 | } |
| 2343 | |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2344 | /* |
| 2345 | * Start health-check. |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2346 | * 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] | 2347 | */ |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2348 | static int start_checks() |
| 2349 | { |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2350 | |
| 2351 | struct proxy *px; |
| 2352 | struct server *s; |
| 2353 | struct task *t; |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2354 | int nbcheck=0, mininter=0, srvpos=0; |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2355 | |
Willy Tarreau | 2c43a1e | 2007-10-14 23:05:39 +0200 | [diff] [blame] | 2356 | /* 1- count the checkers to run simultaneously. |
| 2357 | * We also determine the minimum interval among all of those which |
| 2358 | * have an interval larger than SRV_CHK_INTER_THRES. This interval |
| 2359 | * will be used to spread their start-up date. Those which have |
Jamie Gloudon | 801a0a3 | 2012-08-25 00:18:33 -0400 | [diff] [blame] | 2360 | * a shorter interval will start independently and will not dictate |
Willy Tarreau | 2c43a1e | 2007-10-14 23:05:39 +0200 | [diff] [blame] | 2361 | * too short an interval for all others. |
| 2362 | */ |
Olivier Houchard | fbc74e8 | 2017-11-24 16:54:05 +0100 | [diff] [blame] | 2363 | for (px = proxies_list; px; px = px->next) { |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2364 | for (s = px->srv; s; s = s->next) { |
Willy Tarreau | e7b7348 | 2013-11-21 11:50:50 +0100 | [diff] [blame] | 2365 | if (s->slowstart) { |
Emeric Brun | c60def8 | 2017-09-27 14:59:38 +0200 | [diff] [blame] | 2366 | if ((t = task_new(MAX_THREADS_MASK)) == NULL) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2367 | 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] | 2368 | return ERR_ALERT | ERR_FATAL; |
Willy Tarreau | e7b7348 | 2013-11-21 11:50:50 +0100 | [diff] [blame] | 2369 | } |
| 2370 | /* We need a warmup task that will be called when the server |
| 2371 | * state switches from down to up. |
| 2372 | */ |
| 2373 | s->warmup = t; |
| 2374 | t->process = server_warmup; |
| 2375 | t->context = s; |
Baptiste Assmann | 6076d1c | 2015-09-17 22:53:59 +0200 | [diff] [blame] | 2376 | /* server can be in this state only because of */ |
Emeric Brun | 52a91d3 | 2017-08-31 14:41:55 +0200 | [diff] [blame] | 2377 | if (s->next_state == SRV_ST_STARTING) |
Baptiste Assmann | 6076d1c | 2015-09-17 22:53:59 +0200 | [diff] [blame] | 2378 | 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] | 2379 | } |
| 2380 | |
Willy Tarreau | d8514a2 | 2013-12-11 21:10:14 +0100 | [diff] [blame] | 2381 | if (s->check.state & CHK_ST_CONFIGURED) { |
| 2382 | nbcheck++; |
| 2383 | if ((srv_getinter(&s->check) >= SRV_CHK_INTER_THRES) && |
| 2384 | (!mininter || mininter > srv_getinter(&s->check))) |
| 2385 | mininter = srv_getinter(&s->check); |
| 2386 | } |
Willy Tarreau | 15f3910 | 2013-12-11 20:41:18 +0100 | [diff] [blame] | 2387 | |
Willy Tarreau | d8514a2 | 2013-12-11 21:10:14 +0100 | [diff] [blame] | 2388 | if (s->agent.state & CHK_ST_CONFIGURED) { |
| 2389 | nbcheck++; |
| 2390 | if ((srv_getinter(&s->agent) >= SRV_CHK_INTER_THRES) && |
| 2391 | (!mininter || mininter > srv_getinter(&s->agent))) |
| 2392 | mininter = srv_getinter(&s->agent); |
| 2393 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2394 | } |
| 2395 | } |
| 2396 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2397 | if (!nbcheck) |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2398 | return 0; |
| 2399 | |
| 2400 | srand((unsigned)time(NULL)); |
| 2401 | |
| 2402 | /* |
| 2403 | * 2- start them as far as possible from each others. For this, we will |
| 2404 | * start them after their interval set to the min interval divided by |
| 2405 | * the number of servers, weighted by the server's position in the list. |
| 2406 | */ |
Olivier Houchard | fbc74e8 | 2017-11-24 16:54:05 +0100 | [diff] [blame] | 2407 | for (px = proxies_list; px; px = px->next) { |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2408 | if ((px->options2 & PR_O2_CHK_ANY) == PR_O2_EXT_CHK) { |
| 2409 | if (init_pid_list()) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 2410 | ha_alert("Starting [%s] check: out of memory.\n", px->id); |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2411 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | 98637e5 | 2014-06-20 12:30:16 +0900 | [diff] [blame] | 2412 | } |
| 2413 | } |
| 2414 | |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2415 | for (s = px->srv; s; s = s->next) { |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2416 | /* A task for the main check */ |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 2417 | if (s->check.state & CHK_ST_CONFIGURED) { |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2418 | if (s->check.type == PR_O2_EXT_CHK) { |
| 2419 | if (!prepare_external_check(&s->check)) |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2420 | return ERR_ALERT | ERR_FATAL; |
Cyril Bonté | 99c5bf5 | 2014-08-07 01:55:38 +0200 | [diff] [blame] | 2421 | } |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2422 | if (!start_check_task(&s->check, mininter, nbcheck, srvpos)) |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2423 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2424 | srvpos++; |
| 2425 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2426 | |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2427 | /* A task for a auxiliary agent check */ |
Willy Tarreau | ff5ae35 | 2013-12-11 20:36:34 +0100 | [diff] [blame] | 2428 | if (s->agent.state & CHK_ST_CONFIGURED) { |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2429 | if (!start_check_task(&s->agent, mininter, nbcheck, srvpos)) { |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 2430 | return ERR_ALERT | ERR_FATAL; |
Simon Horman | d60d691 | 2013-11-25 10:46:36 +0900 | [diff] [blame] | 2431 | } |
| 2432 | srvpos++; |
| 2433 | } |
Krzysztof Oledzki | b304dc7 | 2007-10-14 23:40:01 +0200 | [diff] [blame] | 2434 | } |
| 2435 | } |
| 2436 | return 0; |
| 2437 | } |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 2438 | |
| 2439 | /* |
Willy Tarreau | 5b3a202 | 2012-09-28 15:01:02 +0200 | [diff] [blame] | 2440 | * Perform content verification check on data in s->check.buffer buffer. |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2441 | * The buffer MUST be terminated by a null byte before calling this function. |
| 2442 | * Sets server status appropriately. The caller is responsible for ensuring |
| 2443 | * that the buffer contains at least 13 characters. If <done> is zero, we may |
| 2444 | * return 0 to indicate that data is required to decide of a match. |
| 2445 | */ |
| 2446 | static int httpchk_expect(struct server *s, int done) |
| 2447 | { |
Christopher Faulet | 1bc04c7 | 2017-10-29 20:14:08 +0100 | [diff] [blame] | 2448 | static THREAD_LOCAL char status_msg[] = "HTTP status check returned code <000>"; |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2449 | char status_code[] = "000"; |
| 2450 | char *contentptr; |
| 2451 | int crlf; |
| 2452 | int ret; |
| 2453 | |
| 2454 | switch (s->proxy->options2 & PR_O2_EXP_TYPE) { |
| 2455 | case PR_O2_EXP_STS: |
| 2456 | case PR_O2_EXP_RSTS: |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2457 | memcpy(status_code, b_head(&s->check.bi) + 9, 3); |
| 2458 | 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] | 2459 | |
| 2460 | if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STS) |
| 2461 | ret = strncmp(s->proxy->expect_str, status_code, 3) == 0; |
| 2462 | else |
Thierry FOURNIER | 09af0d6 | 2014-06-18 11:35:54 +0200 | [diff] [blame] | 2463 | ret = regex_exec(s->proxy->expect_regex, status_code); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2464 | |
| 2465 | /* we necessarily have the response, so there are no partial failures */ |
| 2466 | if (s->proxy->options2 & PR_O2_EXP_INV) |
| 2467 | ret = !ret; |
| 2468 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2469 | 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] | 2470 | break; |
| 2471 | |
| 2472 | case PR_O2_EXP_STR: |
| 2473 | case PR_O2_EXP_RSTR: |
| 2474 | /* very simple response parser: ignore CR and only count consecutive LFs, |
| 2475 | * stop with contentptr pointing to first char after the double CRLF or |
| 2476 | * to '\0' if crlf < 2. |
| 2477 | */ |
| 2478 | crlf = 0; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2479 | for (contentptr = b_head(&s->check.bi); *contentptr; contentptr++) { |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2480 | if (crlf >= 2) |
| 2481 | break; |
| 2482 | if (*contentptr == '\r') |
| 2483 | continue; |
| 2484 | else if (*contentptr == '\n') |
| 2485 | crlf++; |
| 2486 | else |
| 2487 | crlf = 0; |
| 2488 | } |
| 2489 | |
| 2490 | /* Check that response contains a body... */ |
| 2491 | if (crlf < 2) { |
| 2492 | if (!done) |
| 2493 | return 0; |
| 2494 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2495 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2496 | "HTTP content check could not find a response body"); |
| 2497 | return 1; |
| 2498 | } |
| 2499 | |
| 2500 | /* Check that response body is not empty... */ |
| 2501 | if (*contentptr == '\0') { |
Willy Tarreau | a164fb5 | 2011-04-13 09:32:41 +0200 | [diff] [blame] | 2502 | if (!done) |
| 2503 | return 0; |
| 2504 | |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2505 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2506 | "HTTP content check found empty response body"); |
| 2507 | return 1; |
| 2508 | } |
| 2509 | |
| 2510 | /* Check the response content against the supplied string |
| 2511 | * or regex... */ |
| 2512 | if ((s->proxy->options2 & PR_O2_EXP_TYPE) == PR_O2_EXP_STR) |
| 2513 | ret = strstr(contentptr, s->proxy->expect_str) != NULL; |
| 2514 | else |
Thierry FOURNIER | 09af0d6 | 2014-06-18 11:35:54 +0200 | [diff] [blame] | 2515 | ret = regex_exec(s->proxy->expect_regex, contentptr); |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2516 | |
| 2517 | /* if we don't match, we may need to wait more */ |
| 2518 | if (!ret && !done) |
| 2519 | return 0; |
| 2520 | |
| 2521 | if (ret) { |
| 2522 | /* content matched */ |
| 2523 | if (s->proxy->options2 & PR_O2_EXP_INV) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2524 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2525 | "HTTP check matched unwanted content"); |
| 2526 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2527 | set_server_check_status(&s->check, HCHK_STATUS_L7OKD, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2528 | "HTTP content check matched"); |
| 2529 | } |
| 2530 | else { |
| 2531 | if (s->proxy->options2 & PR_O2_EXP_INV) |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2532 | set_server_check_status(&s->check, HCHK_STATUS_L7OKD, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2533 | "HTTP check did not match unwanted content"); |
| 2534 | else |
Simon Horman | 4a74143 | 2013-02-23 15:35:38 +0900 | [diff] [blame] | 2535 | set_server_check_status(&s->check, HCHK_STATUS_L7RSP, |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 2536 | "HTTP content check did not match"); |
| 2537 | } |
| 2538 | break; |
| 2539 | } |
| 2540 | return 1; |
| 2541 | } |
| 2542 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2543 | /* |
| 2544 | * return the id of a step in a send/expect session |
| 2545 | */ |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2546 | static int tcpcheck_get_step_id(struct check *check) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2547 | { |
| 2548 | struct tcpcheck_rule *cur = NULL, *next = NULL; |
| 2549 | int i = 0; |
| 2550 | |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2551 | /* not even started anything yet => step 0 = initial connect */ |
Baptiste Assmann | f95bc8e | 2015-04-25 16:03:06 +0200 | [diff] [blame] | 2552 | if (!check->current_step) |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2553 | return 0; |
| 2554 | |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2555 | cur = check->last_started_step; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2556 | |
| 2557 | /* no step => first step */ |
| 2558 | if (cur == NULL) |
| 2559 | return 1; |
| 2560 | |
| 2561 | /* increment i until current step */ |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2562 | list_for_each_entry(next, check->tcpcheck_rules, list) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2563 | if (next->list.p == &cur->list) |
| 2564 | break; |
| 2565 | ++i; |
| 2566 | } |
| 2567 | |
| 2568 | return i; |
| 2569 | } |
| 2570 | |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2571 | /* |
| 2572 | * return the latest known comment before (including) the given stepid |
| 2573 | * returns NULL if no comment found |
| 2574 | */ |
| 2575 | static char * tcpcheck_get_step_comment(struct check *check, int stepid) |
| 2576 | { |
| 2577 | struct tcpcheck_rule *cur = NULL; |
| 2578 | char *ret = NULL; |
| 2579 | int i = 0; |
| 2580 | |
| 2581 | /* not even started anything yet, return latest comment found before any action */ |
| 2582 | if (!check->current_step) { |
| 2583 | list_for_each_entry(cur, check->tcpcheck_rules, list) { |
| 2584 | if (cur->action == TCPCHK_ACT_COMMENT) |
| 2585 | ret = cur->comment; |
| 2586 | else |
| 2587 | goto return_comment; |
| 2588 | } |
| 2589 | } |
| 2590 | |
| 2591 | i = 1; |
| 2592 | list_for_each_entry(cur, check->tcpcheck_rules, list) { |
| 2593 | if (cur->comment) |
| 2594 | ret = cur->comment; |
| 2595 | |
| 2596 | if (i >= stepid) |
| 2597 | goto return_comment; |
| 2598 | |
| 2599 | ++i; |
| 2600 | } |
| 2601 | |
| 2602 | return_comment: |
| 2603 | return ret; |
| 2604 | } |
| 2605 | |
Willy Tarreau | be74b88 | 2017-10-04 16:22:49 +0200 | [diff] [blame] | 2606 | /* proceed with next steps for the TCP checks <check>. Note that this is called |
| 2607 | * 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] | 2608 | * It returns 0 on normal cases, or <0 if a close() has happened on an existing |
| 2609 | * connection, presenting the risk of an fd replacement. |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 2610 | * |
| 2611 | * Please do NOT place any return statement in this function and only leave |
| 2612 | * via the out_unlock label after setting retcode. |
Willy Tarreau | be74b88 | 2017-10-04 16:22:49 +0200 | [diff] [blame] | 2613 | */ |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 2614 | static int tcpcheck_main(struct check *check) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2615 | { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2616 | char *contentptr, *comment; |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2617 | struct tcpcheck_rule *next; |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2618 | int done = 0, ret = 0, step = 0; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2619 | struct conn_stream *cs = check->cs; |
| 2620 | struct connection *conn = cs_conn(cs); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2621 | struct server *s = check->server; |
| 2622 | struct task *t = check->task; |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2623 | struct list *head = check->tcpcheck_rules; |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 2624 | int retcode = 0; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2625 | |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2626 | /* here, we know that the check is complete or that it failed */ |
| 2627 | if (check->result != CHK_RES_UNKNOWN) |
| 2628 | goto out_end_tcpcheck; |
| 2629 | |
| 2630 | /* We have 4 possibilities here : |
| 2631 | * 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] | 2632 | * connection attempt was made yet ; conn==NULL;current_step==NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2633 | * 2. we've not yet attempted step 1, and step 1 is a not connect or |
| 2634 | * does not exist (no rule), so a connection attempt was made |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2635 | * before coming here, conn!=NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2636 | * 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] | 2637 | * be waiting for a connection attempt to complete. conn!=NULL. |
| 2638 | * 4. the connection + handshake are complete. conn!=NULL. |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2639 | * |
| 2640 | * #2 and #3 are quite similar, we want both the connection and the |
| 2641 | * handshake to complete before going any further. Thus we must always |
| 2642 | * wait for a connection to complete unless we're before and existing |
| 2643 | * step 1. |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2644 | */ |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2645 | |
| 2646 | /* find first rule and skip comments */ |
| 2647 | next = LIST_NEXT(head, struct tcpcheck_rule *, list); |
| 2648 | while (&next->list != head && next->action == TCPCHK_ACT_COMMENT) |
| 2649 | next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list); |
| 2650 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2651 | if ((check->current_step || &next->list == head) && |
| 2652 | (!(conn->flags & CO_FL_CONNECTED) || (conn->flags & CO_FL_HANDSHAKE))) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2653 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2654 | * to establish but only when timeout.check is set |
| 2655 | * as it may be to short for a full check otherwise |
| 2656 | */ |
| 2657 | while (tick_is_expired(t->expire, now_ms)) { |
| 2658 | int t_con; |
| 2659 | |
| 2660 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
| 2661 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 2662 | |
| 2663 | if (s->proxy->timeout.check) |
| 2664 | t->expire = tick_first(t->expire, t_con); |
| 2665 | } |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 2666 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2667 | } |
| 2668 | |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2669 | /* special case: option tcp-check with no rule, a connect is enough */ |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2670 | if (&next->list == head) { |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2671 | set_server_check_status(check, HCHK_STATUS_L4OK, NULL); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2672 | goto out_end_tcpcheck; |
Willy Tarreau | ef95395 | 2014-10-02 14:30:14 +0200 | [diff] [blame] | 2673 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2674 | |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2675 | /* no step means first step initialisation */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2676 | if (check->current_step == NULL) { |
Willy Tarreau | 213c678 | 2014-10-02 14:51:02 +0200 | [diff] [blame] | 2677 | check->last_started_step = NULL; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2678 | b_reset(&check->bo); |
| 2679 | b_reset(&check->bi); |
Willy Tarreau | 449f952 | 2015-05-13 15:39:48 +0200 | [diff] [blame] | 2680 | check->current_step = next; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2681 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 2682 | if (s->proxy->timeout.check) |
| 2683 | t->expire = tick_add_ifset(now_ms, s->proxy->timeout.check); |
| 2684 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2685 | |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2686 | /* It's only the rules which will enable send/recv */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2687 | if (cs) |
| 2688 | cs_stop_both(cs); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2689 | |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2690 | while (1) { |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2691 | /* We have to try to flush the output buffer before reading, at |
| 2692 | * the end, or if we're about to send a string that does not fit |
| 2693 | * in the remaining space. That explains why we break out of the |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2694 | * loop after this control. If we have data, conn is valid. |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2695 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2696 | if (b_data(&check->bo) && |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2697 | (&check->current_step->list == head || |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2698 | check->current_step->action != TCPCHK_ACT_SEND || |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2699 | check->current_step->string_len >= b_room(&check->bo))) { |
Willy Tarreau | deccd11 | 2018-06-14 18:38:55 +0200 | [diff] [blame] | 2700 | int ret; |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2701 | |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2702 | __cs_want_send(cs); |
Olivier Houchard | ed0f207 | 2018-08-16 15:41:52 +0200 | [diff] [blame] | 2703 | 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] | 2704 | b_realign_if_empty(&check->bo); |
Willy Tarreau | deccd11 | 2018-06-14 18:38:55 +0200 | [diff] [blame] | 2705 | |
| 2706 | if (ret <= 0) { |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 2707 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) { |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2708 | chk_report_conn_err(check, errno, 0); |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2709 | __cs_stop_both(cs); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2710 | goto out_end_tcpcheck; |
| 2711 | } |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2712 | break; |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2713 | } |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 2714 | if (b_data(&check->bo)) { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 2715 | cs->conn->mux->subscribe(cs, SUB_CAN_SEND, &check->wait_list); |
Olivier Houchard | 4501c3e | 2018-08-28 19:36:18 +0200 | [diff] [blame] | 2716 | goto out; |
| 2717 | } |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2718 | } |
| 2719 | |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2720 | if (&check->current_step->list == head) |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2721 | break; |
| 2722 | |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2723 | /* have 'next' point to the next rule or NULL if we're on the |
| 2724 | * last one, connect() needs this. |
| 2725 | */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2726 | next = LIST_NEXT(&check->current_step->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2727 | |
| 2728 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 2729 | while (&next->list != head && next->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2730 | next = LIST_NEXT(&next->list, struct tcpcheck_rule *, list); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2731 | |
| 2732 | /* NULL if we're on the last rule */ |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 2733 | if (&next->list == head) |
| 2734 | next = NULL; |
| 2735 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2736 | if (check->current_step->action == TCPCHK_ACT_CONNECT) { |
| 2737 | struct protocol *proto; |
| 2738 | struct xprt_ops *xprt; |
| 2739 | |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2740 | /* For a connect action we'll create a new connection. |
| 2741 | * We may also have to kill a previous one. But we don't |
| 2742 | * want to leave *without* a connection if we came here |
| 2743 | * from the connection layer, hence with a connection. |
| 2744 | * Thus we'll proceed in the following order : |
| 2745 | * 1: close but not release previous connection |
| 2746 | * 2: try to get a new connection |
| 2747 | * 3: release and replace the old one on success |
| 2748 | */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2749 | if (check->cs) { |
Willy Tarreau | a553ae9 | 2017-10-05 18:52:17 +0200 | [diff] [blame] | 2750 | cs_close(check->cs); |
| 2751 | retcode = -1; /* do not reuse the fd in the caller! */ |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2752 | } |
| 2753 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2754 | /* mark the step as started */ |
| 2755 | check->last_started_step = check->current_step; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2756 | |
| 2757 | /* prepare new connection */ |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2758 | cs = cs_new(NULL); |
| 2759 | if (!cs) { |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2760 | step = tcpcheck_get_step_id(check); |
| 2761 | chunk_printf(&trash, "TCPCHK error allocating connection at step %d", step); |
| 2762 | comment = tcpcheck_get_step_comment(check, step); |
| 2763 | if (comment) |
| 2764 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2765 | set_server_check_status(check, HCHK_STATUS_SOCKERR, |
| 2766 | trash.area); |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2767 | check->current_step = NULL; |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 2768 | goto out; |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2769 | } |
| 2770 | |
Willy Tarreau | ac59f36 | 2017-10-08 11:10:19 +0200 | [diff] [blame] | 2771 | if (check->cs) |
| 2772 | cs_destroy(check->cs); |
Willy Tarreau | 0014912 | 2017-10-04 18:05:01 +0200 | [diff] [blame] | 2773 | |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2774 | check->cs = cs; |
| 2775 | conn = cs->conn; |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 2776 | /* Maybe there were an older connection we were waiting on */ |
| 2777 | check->wait_list.wait_reason = 0; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2778 | conn->target = &s->obj_type; |
| 2779 | |
| 2780 | /* no client address */ |
| 2781 | clear_addr(&conn->addr.from); |
| 2782 | |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 2783 | if (is_addr(&check->addr)) { |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2784 | /* we'll connect to the check addr specified on the server */ |
Simon Horman | 41f5876 | 2015-01-30 11:22:56 +0900 | [diff] [blame] | 2785 | conn->addr.to = check->addr; |
Willy Tarreau | 640556c | 2014-05-09 23:38:15 +0200 | [diff] [blame] | 2786 | } |
| 2787 | else { |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2788 | /* we'll connect to the addr on the server */ |
| 2789 | conn->addr.to = s->addr; |
Willy Tarreau | 640556c | 2014-05-09 23:38:15 +0200 | [diff] [blame] | 2790 | } |
Thierry FOURNIER | bb2ae64 | 2015-01-14 11:31:49 +0100 | [diff] [blame] | 2791 | proto = protocol_by_family(conn->addr.to.ss_family); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2792 | |
| 2793 | /* port */ |
| 2794 | if (check->current_step->port) |
| 2795 | set_host_port(&conn->addr.to, check->current_step->port); |
| 2796 | else if (check->port) |
| 2797 | set_host_port(&conn->addr.to, check->port); |
Baptiste Assmann | 2f3a56b | 2018-03-19 12:22:41 +0100 | [diff] [blame] | 2798 | else if (s->svc_port) |
| 2799 | set_host_port(&conn->addr.to, s->svc_port); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2800 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2801 | if (check->current_step->conn_opts & TCPCHK_OPT_SSL) { |
Willy Tarreau | a261e9b | 2016-12-22 20:44:00 +0100 | [diff] [blame] | 2802 | xprt = xprt_get(XPRT_SSL); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2803 | } |
| 2804 | else { |
Willy Tarreau | a261e9b | 2016-12-22 20:44:00 +0100 | [diff] [blame] | 2805 | xprt = xprt_get(XPRT_RAW); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2806 | } |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 2807 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2808 | conn_prepare(conn, proto, xprt); |
Willy Tarreau | 175a2bb | 2018-09-12 12:02:05 +0200 | [diff] [blame] | 2809 | conn_install_mux(conn, &mux_pt_ops, cs, s->proxy); |
Willy Tarreau | be37315 | 2018-09-06 11:45:30 +0200 | [diff] [blame] | 2810 | cs_attach(cs, check, &check_conn_cb); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2811 | |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2812 | ret = SF_ERR_INTERNAL; |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2813 | if (proto->connect) |
Willy Tarreau | f3d3482 | 2014-12-08 12:11:28 +0100 | [diff] [blame] | 2814 | ret = proto->connect(conn, |
| 2815 | 1 /* I/O polling is always needed */, |
| 2816 | (next && next->action == TCPCHK_ACT_EXPECT) ? 0 : 2); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2817 | if (check->current_step->conn_opts & TCPCHK_OPT_SEND_PROXY) { |
| 2818 | conn->send_proxy_ofs = 1; |
| 2819 | conn->flags |= CO_FL_SEND_PROXY; |
| 2820 | } |
| 2821 | |
| 2822 | /* It can return one of : |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2823 | * - SF_ERR_NONE if everything's OK |
| 2824 | * - SF_ERR_SRVTO if there are no more servers |
| 2825 | * - SF_ERR_SRVCL if the connection was refused by the server |
| 2826 | * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn) |
| 2827 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 2828 | * - SF_ERR_INTERNAL for any other purely internal errors |
Tim Düsterhus | 4896c44 | 2016-11-29 02:15:19 +0100 | [diff] [blame] | 2829 | * 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] | 2830 | * Note that we try to prevent the network stack from sending the ACK during the |
| 2831 | * connect() when a pure TCP check is used (without PROXY protocol). |
| 2832 | */ |
| 2833 | switch (ret) { |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2834 | case SF_ERR_NONE: |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2835 | /* we allow up to min(inter, timeout.connect) for a connection |
| 2836 | * to establish but only when timeout.check is set |
| 2837 | * as it may be to short for a full check otherwise |
| 2838 | */ |
| 2839 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 2840 | |
| 2841 | if (s->proxy->timeout.check && s->proxy->timeout.connect) { |
| 2842 | int t_con = tick_add(now_ms, s->proxy->timeout.connect); |
| 2843 | t->expire = tick_first(t->expire, t_con); |
| 2844 | } |
| 2845 | break; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2846 | case SF_ERR_SRVTO: /* ETIMEDOUT */ |
| 2847 | case SF_ERR_SRVCL: /* ECONNREFUSED, ENETUNREACH, ... */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2848 | step = tcpcheck_get_step_id(check); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2849 | chunk_printf(&trash, "TCPCHK error establishing connection at step %d: %s", |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2850 | step, strerror(errno)); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2851 | comment = tcpcheck_get_step_comment(check, step); |
| 2852 | if (comment) |
| 2853 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2854 | set_server_check_status(check, HCHK_STATUS_L4CON, |
| 2855 | trash.area); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2856 | goto out_end_tcpcheck; |
Willy Tarreau | e7dff02 | 2015-04-03 01:14:29 +0200 | [diff] [blame] | 2857 | case SF_ERR_PRXCOND: |
| 2858 | case SF_ERR_RESOURCE: |
| 2859 | case SF_ERR_INTERNAL: |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2860 | step = tcpcheck_get_step_id(check); |
| 2861 | chunk_printf(&trash, "TCPCHK error establishing connection at step %d", step); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2862 | comment = tcpcheck_get_step_comment(check, step); |
| 2863 | if (comment) |
| 2864 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2865 | set_server_check_status(check, HCHK_STATUS_SOCKERR, |
| 2866 | trash.area); |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2867 | goto out_end_tcpcheck; |
| 2868 | } |
| 2869 | |
| 2870 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2871 | 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] | 2872 | |
| 2873 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 2874 | while (&check->current_step->list != head && |
| 2875 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2876 | 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] | 2877 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 2878 | if (&check->current_step->list == head) |
| 2879 | break; |
| 2880 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2881 | } /* end 'connect' */ |
| 2882 | else if (check->current_step->action == TCPCHK_ACT_SEND) { |
| 2883 | /* mark the step as started */ |
| 2884 | check->last_started_step = check->current_step; |
| 2885 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2886 | /* reset the read buffer */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2887 | if (*b_head(&check->bi) != '\0') { |
| 2888 | *b_head(&check->bi) = '\0'; |
| 2889 | b_reset(&check->bi); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2890 | } |
| 2891 | |
Willy Tarreau | bbae3f0 | 2017-08-30 09:59:52 +0200 | [diff] [blame] | 2892 | if (conn->flags & CO_FL_SOCK_WR_SH) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2893 | conn->flags |= CO_FL_ERROR; |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2894 | chk_report_conn_err(check, 0, 0); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2895 | goto out_end_tcpcheck; |
| 2896 | } |
| 2897 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2898 | if (check->current_step->string_len >= b_size(&check->bo)) { |
Willy Tarreau | 506a29a | 2018-07-18 10:07:58 +0200 | [diff] [blame] | 2899 | 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] | 2900 | check->current_step->string_len, (unsigned int)b_size(&check->bo), |
Simon Horman | e16c1b3 | 2015-01-30 11:22:57 +0900 | [diff] [blame] | 2901 | tcpcheck_get_step_id(check)); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2902 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 2903 | trash.area); |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2904 | goto out_end_tcpcheck; |
| 2905 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2906 | |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2907 | /* do not try to send if there is no space */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2908 | if (check->current_step->string_len >= b_room(&check->bo)) |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2909 | continue; |
| 2910 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2911 | b_putblk(&check->bo, check->current_step->string, check->current_step->string_len); |
| 2912 | *b_tail(&check->bo) = '\0'; /* to make gdb output easier to read */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2913 | |
Willy Tarreau | abca5b6 | 2013-12-06 14:19:25 +0100 | [diff] [blame] | 2914 | /* go to next rule and try to send */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2915 | 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] | 2916 | |
| 2917 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 2918 | while (&check->current_step->list != head && |
| 2919 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 2920 | 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] | 2921 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2922 | } /* end 'send' */ |
Willy Tarreau | 98aec9f | 2013-12-06 16:16:41 +0100 | [diff] [blame] | 2923 | else if (check->current_step->action == TCPCHK_ACT_EXPECT) { |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 2924 | if (unlikely(check->result == CHK_RES_FAILED)) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2925 | goto out_end_tcpcheck; |
| 2926 | |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 2927 | __cs_want_recv(cs); |
Olivier Houchard | 511efea | 2018-08-16 15:30:32 +0200 | [diff] [blame] | 2928 | 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] | 2929 | 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] | 2930 | done = 1; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2931 | 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] | 2932 | /* Report network errors only if we got no other data. Otherwise |
| 2933 | * we'll let the upper layers decide whether the response is OK |
| 2934 | * or not. It is very common that an RST sent by the server is |
| 2935 | * reported as an error just after the last data chunk. |
| 2936 | */ |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 2937 | chk_report_conn_err(check, errno, 0); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2938 | goto out_end_tcpcheck; |
| 2939 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2940 | } |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 2941 | else { |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 2942 | conn->mux->subscribe(cs, SUB_CAN_RECV, &check->wait_list); |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 2943 | break; |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 2944 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2945 | } |
| 2946 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 2947 | /* mark the step as started */ |
| 2948 | check->last_started_step = check->current_step; |
| 2949 | |
| 2950 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2951 | /* Intermediate or complete response received. |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2952 | * Terminate string in b_head(&check->bi) buffer. |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2953 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2954 | if (b_data(&check->bi) < b_size(&check->bi)) { |
| 2955 | b_head(&check->bi)[b_data(&check->bi)] = '\0'; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2956 | } |
| 2957 | else { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2958 | b_head(&check->bi)[b_data(&check->bi) - 1] = '\0'; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2959 | done = 1; /* buffer full, don't wait for more data */ |
| 2960 | } |
| 2961 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2962 | contentptr = b_head(&check->bi); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2963 | |
| 2964 | /* Check that response body is not empty... */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2965 | if (!b_data(&check->bi)) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2966 | if (!done) |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 2967 | continue; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2968 | |
| 2969 | /* empty response */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2970 | step = tcpcheck_get_step_id(check); |
| 2971 | chunk_printf(&trash, "TCPCHK got an empty response at step %d", step); |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 2972 | comment = tcpcheck_get_step_comment(check, step); |
| 2973 | if (comment) |
| 2974 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 2975 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 2976 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2977 | |
| 2978 | goto out_end_tcpcheck; |
| 2979 | } |
| 2980 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2981 | if (!done && (check->current_step->string != NULL) && (b_data(&check->bi) < check->current_step->string_len) ) |
Willy Tarreau | a970c28 | 2013-12-06 12:47:19 +0100 | [diff] [blame] | 2982 | continue; /* try to read more */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2983 | |
Willy Tarreau | a970c28 | 2013-12-06 12:47:19 +0100 | [diff] [blame] | 2984 | tcpcheck_expect: |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2985 | if (check->current_step->string != NULL) |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 2986 | ret = my_memmem(contentptr, b_data(&check->bi), check->current_step->string, check->current_step->string_len) != NULL; |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2987 | else if (check->current_step->expect_regex != NULL) |
| 2988 | ret = regex_exec(check->current_step->expect_regex, contentptr); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2989 | |
| 2990 | if (!ret && !done) |
Willy Tarreau | a970c28 | 2013-12-06 12:47:19 +0100 | [diff] [blame] | 2991 | continue; /* try to read more */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2992 | |
| 2993 | /* matched */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 2994 | step = tcpcheck_get_step_id(check); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2995 | if (ret) { |
| 2996 | /* matched but we did not want to => ERROR */ |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2997 | if (check->current_step->inverse) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 2998 | /* we were looking for a string */ |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 2999 | if (check->current_step->string != NULL) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3000 | chunk_printf(&trash, "TCPCHK matched unwanted content '%s' at step %d", |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 3001 | check->current_step->string, step); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3002 | } |
| 3003 | else { |
| 3004 | /* we were looking for a regex */ |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3005 | chunk_printf(&trash, "TCPCHK matched unwanted content (regex) at step %d", step); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3006 | } |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3007 | comment = tcpcheck_get_step_comment(check, step); |
| 3008 | if (comment) |
| 3009 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3010 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3011 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3012 | goto out_end_tcpcheck; |
| 3013 | } |
| 3014 | /* matched and was supposed to => OK, next step */ |
| 3015 | else { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3016 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3017 | 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] | 3018 | |
| 3019 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3020 | while (&check->current_step->list != head && |
| 3021 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3022 | 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] | 3023 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3024 | if (&check->current_step->list == head) |
| 3025 | break; |
| 3026 | |
Willy Tarreau | 98aec9f | 2013-12-06 16:16:41 +0100 | [diff] [blame] | 3027 | if (check->current_step->action == TCPCHK_ACT_EXPECT) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3028 | goto tcpcheck_expect; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 3029 | __cs_stop_recv(cs); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3030 | } |
| 3031 | } |
| 3032 | else { |
| 3033 | /* not matched */ |
| 3034 | /* not matched and was not supposed to => OK, next step */ |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 3035 | if (check->current_step->inverse) { |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3036 | /* allow next rule */ |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3037 | 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] | 3038 | |
| 3039 | /* bypass all comment rules */ |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3040 | while (&check->current_step->list != head && |
| 3041 | check->current_step->action == TCPCHK_ACT_COMMENT) |
Willy Tarreau | 5581c27 | 2015-05-13 12:24:53 +0200 | [diff] [blame] | 3042 | 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] | 3043 | |
Willy Tarreau | f2c8735 | 2015-05-13 12:08:21 +0200 | [diff] [blame] | 3044 | if (&check->current_step->list == head) |
| 3045 | break; |
| 3046 | |
Willy Tarreau | 98aec9f | 2013-12-06 16:16:41 +0100 | [diff] [blame] | 3047 | if (check->current_step->action == TCPCHK_ACT_EXPECT) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3048 | goto tcpcheck_expect; |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 3049 | __cs_stop_recv(cs); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3050 | } |
| 3051 | /* not matched but was supposed to => ERROR */ |
| 3052 | else { |
| 3053 | /* we were looking for a string */ |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 3054 | if (check->current_step->string != NULL) { |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3055 | chunk_printf(&trash, "TCPCHK did not match content '%s' at step %d", |
Willy Tarreau | ce8c42a | 2015-05-13 11:23:01 +0200 | [diff] [blame] | 3056 | check->current_step->string, step); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3057 | } |
| 3058 | else { |
| 3059 | /* we were looking for a regex */ |
| 3060 | chunk_printf(&trash, "TCPCHK did not match content (regex) at step %d", |
Baptiste Assmann | cfbd1b8 | 2015-05-02 09:00:23 +0200 | [diff] [blame] | 3061 | step); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3062 | } |
Baptiste Assmann | 22b09d2 | 2015-05-01 08:03:04 +0200 | [diff] [blame] | 3063 | comment = tcpcheck_get_step_comment(check, step); |
| 3064 | if (comment) |
| 3065 | chunk_appendf(&trash, " comment: '%s'", comment); |
Willy Tarreau | 843b7cb | 2018-07-13 10:54:26 +0200 | [diff] [blame] | 3066 | set_server_check_status(check, HCHK_STATUS_L7RSP, |
| 3067 | trash.area); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3068 | goto out_end_tcpcheck; |
| 3069 | } |
| 3070 | } |
| 3071 | } /* end expect */ |
| 3072 | } /* end loop over double chained step list */ |
| 3073 | |
Baptiste Assmann | 248f117 | 2018-03-01 21:49:01 +0100 | [diff] [blame] | 3074 | /* don't do anything until the connection is established */ |
| 3075 | if (!(conn->flags & CO_FL_CONNECTED)) { |
| 3076 | /* update expire time, should be done by process_chk */ |
| 3077 | /* we allow up to min(inter, timeout.connect) for a connection |
| 3078 | * to establish but only when timeout.check is set |
| 3079 | * as it may be to short for a full check otherwise |
| 3080 | */ |
| 3081 | while (tick_is_expired(t->expire, now_ms)) { |
| 3082 | int t_con; |
| 3083 | |
| 3084 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
| 3085 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 3086 | |
| 3087 | if (s->proxy->timeout.check) |
| 3088 | t->expire = tick_first(t->expire, t_con); |
| 3089 | } |
| 3090 | goto out; |
| 3091 | } |
| 3092 | |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 3093 | /* We're waiting for some I/O to complete, we've reached the end of the |
| 3094 | * rules, or both. Do what we have to do, otherwise we're done. |
| 3095 | */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3096 | if (&check->current_step->list == head && !b_data(&check->bo)) { |
Willy Tarreau | 263013d | 2015-05-13 11:59:14 +0200 | [diff] [blame] | 3097 | set_server_check_status(check, HCHK_STATUS_L7OKD, "(tcp-check)"); |
| 3098 | goto out_end_tcpcheck; |
| 3099 | } |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3100 | |
Willy Tarreau | 53c5a04 | 2015-05-13 11:38:17 +0200 | [diff] [blame] | 3101 | /* warning, current_step may now point to the head */ |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3102 | if (b_data(&check->bo)) |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 3103 | __cs_want_send(cs); |
Willy Tarreau | fbe0edf | 2013-12-06 16:54:31 +0100 | [diff] [blame] | 3104 | |
Willy Tarreau | 53c5a04 | 2015-05-13 11:38:17 +0200 | [diff] [blame] | 3105 | if (&check->current_step->list != head && |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 3106 | check->current_step->action == TCPCHK_ACT_EXPECT) { |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 3107 | __cs_want_recv(cs); |
Olivier Houchard | af4021e | 2018-08-09 13:06:55 +0200 | [diff] [blame] | 3108 | __event_srv_chk_r(cs); |
| 3109 | } |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 3110 | goto out; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3111 | |
| 3112 | out_end_tcpcheck: |
| 3113 | /* collect possible new errors */ |
Willy Tarreau | 4ff3b89 | 2017-10-16 15:17:17 +0200 | [diff] [blame] | 3114 | if (conn->flags & CO_FL_ERROR || cs->flags & CS_FL_ERROR) |
Willy Tarreau | b5259bf | 2017-10-04 14:47:29 +0200 | [diff] [blame] | 3115 | chk_report_conn_err(check, 0, 0); |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3116 | |
Baptiste Assmann | 69e273f | 2013-12-11 00:52:19 +0100 | [diff] [blame] | 3117 | /* cleanup before leaving */ |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3118 | check->current_step = NULL; |
| 3119 | |
Willy Tarreau | 6aaa1b8 | 2013-12-11 17:09:34 +0100 | [diff] [blame] | 3120 | if (check->result == CHK_RES_FAILED) |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3121 | conn->flags |= CO_FL_ERROR; |
| 3122 | |
Olivier Houchard | 9aaf778 | 2017-09-13 18:30:23 +0200 | [diff] [blame] | 3123 | __cs_stop_both(cs); |
Willy Tarreau | 62ac84f | 2017-11-05 10:11:13 +0100 | [diff] [blame] | 3124 | |
Christopher Faulet | b610285 | 2017-11-28 10:06:29 +0100 | [diff] [blame] | 3125 | out: |
Willy Tarreau | 6bdcab0 | 2017-10-04 18:41:00 +0200 | [diff] [blame] | 3126 | return retcode; |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3127 | } |
| 3128 | |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3129 | const char *init_check(struct check *check, int type) |
| 3130 | { |
| 3131 | check->type = type; |
| 3132 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3133 | b_reset(&check->bi); check->bi.size = global.tune.chksize; |
| 3134 | b_reset(&check->bo); check->bo.size = global.tune.chksize; |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3135 | |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3136 | check->bi.area = calloc(check->bi.size, sizeof(char)); |
| 3137 | check->bo.area = calloc(check->bo.size, sizeof(char)); |
| 3138 | |
| 3139 | if (!check->bi.area || !check->bo.area) |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3140 | return "out of memory while allocating check buffer"; |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3141 | |
Olivier Houchard | 26e1a8f | 2018-09-12 15:15:12 +0200 | [diff] [blame] | 3142 | check->wait_list.task = tasklet_new(); |
| 3143 | if (!check->wait_list.task) |
| 3144 | return "out of memroy while allocating check tasklet"; |
| 3145 | LIST_INIT(&check->wait_list.list); |
| 3146 | check->wait_list.wait_reason = 0; |
| 3147 | check->wait_list.task->process = event_srv_chk_io; |
| 3148 | check->wait_list.task->context = check; |
Simon Horman | b1900d5 | 2015-01-30 11:22:54 +0900 | [diff] [blame] | 3149 | return NULL; |
| 3150 | } |
| 3151 | |
Simon Horman | bfb5d33 | 2015-01-30 11:22:55 +0900 | [diff] [blame] | 3152 | void free_check(struct check *check) |
| 3153 | { |
Willy Tarreau | c9fa048 | 2018-07-10 17:43:27 +0200 | [diff] [blame] | 3154 | free(check->bi.area); |
| 3155 | free(check->bo.area); |
Christopher Faulet | 23d86d1 | 2018-01-25 11:36:35 +0100 | [diff] [blame] | 3156 | if (check->cs) { |
| 3157 | free(check->cs->conn); |
| 3158 | check->cs->conn = NULL; |
| 3159 | cs_free(check->cs); |
| 3160 | check->cs = NULL; |
| 3161 | } |
Simon Horman | bfb5d33 | 2015-01-30 11:22:55 +0900 | [diff] [blame] | 3162 | } |
| 3163 | |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3164 | void email_alert_free(struct email_alert *alert) |
| 3165 | { |
| 3166 | struct tcpcheck_rule *rule, *back; |
| 3167 | |
| 3168 | if (!alert) |
| 3169 | return; |
| 3170 | |
Christopher Faulet | de1a75b | 2017-10-23 15:38:19 +0200 | [diff] [blame] | 3171 | list_for_each_entry_safe(rule, back, &alert->tcpcheck_rules, list) { |
| 3172 | LIST_DEL(&rule->list); |
| 3173 | free(rule->comment); |
| 3174 | free(rule->string); |
| 3175 | if (rule->expect_regex) |
| 3176 | regex_free(rule->expect_regex); |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3177 | pool_free(pool_head_tcpcheck_rule, rule); |
Christopher Faulet | de1a75b | 2017-10-23 15:38:19 +0200 | [diff] [blame] | 3178 | } |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3179 | pool_free(pool_head_email_alert, alert); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3180 | } |
| 3181 | |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3182 | 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] | 3183 | { |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3184 | struct check *check = context; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3185 | struct email_alertq *q; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3186 | struct email_alert *alert; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3187 | |
| 3188 | q = container_of(check, typeof(*q), check); |
| 3189 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3190 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3191 | while (1) { |
| 3192 | if (!(check->state & CHK_ST_ENABLED)) { |
| 3193 | if (LIST_ISEMPTY(&q->email_alerts)) { |
| 3194 | /* All alerts processed, queue the task */ |
| 3195 | t->expire = TICK_ETERNITY; |
| 3196 | task_queue(t); |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 3197 | goto end; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3198 | } |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3199 | |
| 3200 | alert = LIST_NEXT(&q->email_alerts, typeof(alert), list); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3201 | LIST_DEL(&alert->list); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3202 | t->expire = now_ms; |
| 3203 | check->server = alert->srv; |
| 3204 | check->tcpcheck_rules = &alert->tcpcheck_rules; |
PiBa-NL | 1714b9f | 2017-12-06 01:35:43 +0100 | [diff] [blame] | 3205 | check->status = HCHK_STATUS_UNKNOWN; // the UNKNOWN status is used to exit set_server_check_status(.) early |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3206 | check->state |= CHK_ST_ENABLED; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3207 | } |
| 3208 | |
Olivier Houchard | 9f6af33 | 2018-05-25 14:04:04 +0200 | [diff] [blame] | 3209 | process_chk(t, context, state); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3210 | if (check->state & CHK_ST_INPROGRESS) |
| 3211 | break; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3212 | |
| 3213 | alert = container_of(check->tcpcheck_rules, typeof(*alert), tcpcheck_rules); |
| 3214 | email_alert_free(alert); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3215 | check->tcpcheck_rules = NULL; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3216 | check->server = NULL; |
| 3217 | check->state &= ~CHK_ST_ENABLED; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3218 | } |
Christopher Faulet | c2a89a6 | 2017-10-23 15:54:24 +0200 | [diff] [blame] | 3219 | end: |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3220 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3221 | return t; |
| 3222 | } |
| 3223 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3224 | /* Initializes mailer alerts for the proxy <p> using <mls> parameters. |
| 3225 | * |
| 3226 | * The function returns 1 in success case, otherwise, it returns 0 and err is |
| 3227 | * filled. |
| 3228 | */ |
| 3229 | int init_email_alert(struct mailers *mls, struct proxy *p, char **err) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3230 | { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3231 | struct mailer *mailer; |
| 3232 | struct email_alertq *queues; |
| 3233 | const char *err_str; |
| 3234 | int i = 0; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3235 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3236 | if ((queues = calloc(mls->count, sizeof(*queues))) == NULL) { |
| 3237 | memprintf(err, "out of memory while allocating mailer alerts queues"); |
| 3238 | goto error; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3239 | } |
| 3240 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3241 | for (mailer = mls->mailer_list; mailer; i++, mailer = mailer->next) { |
| 3242 | struct email_alertq *q = &queues[i]; |
| 3243 | struct check *check = &q->check; |
| 3244 | struct task *t; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3245 | |
| 3246 | LIST_INIT(&q->email_alerts); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3247 | HA_SPIN_INIT(&q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3248 | check->inter = mls->timeout.mail; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3249 | check->rise = DEF_AGENT_RISETIME; |
| 3250 | check->fall = DEF_AGENT_FALLTIME; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3251 | if ((err_str = init_check(check, PR_O2_TCPCHK_CHK))) { |
| 3252 | memprintf(err, "%s", err_str); |
| 3253 | goto error; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3254 | } |
| 3255 | |
| 3256 | check->xprt = mailer->xprt; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3257 | check->addr = mailer->addr; |
Christopher Faulet | b797ae1 | 2018-03-27 15:35:35 +0200 | [diff] [blame] | 3258 | check->port = get_host_port(&mailer->addr); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3259 | //check->server = s; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3260 | |
Emeric Brun | c60def8 | 2017-09-27 14:59:38 +0200 | [diff] [blame] | 3261 | if ((t = task_new(MAX_THREADS_MASK)) == NULL) { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3262 | memprintf(err, "out of memory while allocating mailer alerts task"); |
| 3263 | goto error; |
| 3264 | } |
| 3265 | |
| 3266 | check->task = t; |
| 3267 | t->process = process_email_alert; |
| 3268 | t->context = check; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3269 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3270 | /* check this in one ms */ |
| 3271 | t->expire = TICK_ETERNITY; |
| 3272 | check->start = now; |
| 3273 | task_queue(t); |
| 3274 | } |
| 3275 | |
| 3276 | mls->users++; |
| 3277 | free(p->email_alert.mailers.name); |
| 3278 | p->email_alert.mailers.m = mls; |
| 3279 | p->email_alert.queues = queues; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3280 | return 0; |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3281 | |
| 3282 | error: |
| 3283 | for (i = 0; i < mls->count; i++) { |
| 3284 | struct email_alertq *q = &queues[i]; |
| 3285 | struct check *check = &q->check; |
| 3286 | |
| 3287 | if (check->task) { |
| 3288 | task_delete(check->task); |
| 3289 | task_free(check->task); |
| 3290 | check->task = NULL; |
| 3291 | } |
| 3292 | free_check(check); |
| 3293 | } |
| 3294 | free(queues); |
| 3295 | return 1; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3296 | } |
| 3297 | |
| 3298 | |
| 3299 | static int add_tcpcheck_expect_str(struct list *list, const char *str) |
| 3300 | { |
| 3301 | struct tcpcheck_rule *tcpcheck; |
| 3302 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3303 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3304 | return 0; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3305 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
| 3306 | tcpcheck->action = TCPCHK_ACT_EXPECT; |
| 3307 | tcpcheck->string = strdup(str); |
| 3308 | tcpcheck->expect_regex = NULL; |
| 3309 | tcpcheck->comment = NULL; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3310 | if (!tcpcheck->string) { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3311 | pool_free(pool_head_tcpcheck_rule, tcpcheck); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3312 | return 0; |
| 3313 | } |
| 3314 | |
| 3315 | LIST_ADDQ(list, &tcpcheck->list); |
| 3316 | return 1; |
| 3317 | } |
| 3318 | |
| 3319 | static int add_tcpcheck_send_strs(struct list *list, const char * const *strs) |
| 3320 | { |
| 3321 | struct tcpcheck_rule *tcpcheck; |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3322 | const char *in; |
| 3323 | char *dst; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3324 | int i; |
| 3325 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3326 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3327 | return 0; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3328 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
| 3329 | tcpcheck->action = TCPCHK_ACT_SEND; |
| 3330 | tcpcheck->expect_regex = NULL; |
| 3331 | tcpcheck->comment = NULL; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3332 | tcpcheck->string_len = 0; |
| 3333 | for (i = 0; strs[i]; i++) |
| 3334 | tcpcheck->string_len += strlen(strs[i]); |
| 3335 | |
| 3336 | tcpcheck->string = malloc(tcpcheck->string_len + 1); |
| 3337 | if (!tcpcheck->string) { |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3338 | pool_free(pool_head_tcpcheck_rule, tcpcheck); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3339 | return 0; |
| 3340 | } |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3341 | |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3342 | dst = tcpcheck->string; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3343 | for (i = 0; strs[i]; i++) |
Willy Tarreau | 64345aa | 2016-08-10 19:29:09 +0200 | [diff] [blame] | 3344 | for (in = strs[i]; (*dst = *in++); dst++); |
| 3345 | *dst = 0; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3346 | |
| 3347 | LIST_ADDQ(list, &tcpcheck->list); |
| 3348 | return 1; |
| 3349 | } |
| 3350 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3351 | static int enqueue_one_email_alert(struct proxy *p, struct server *s, |
| 3352 | struct email_alertq *q, const char *msg) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3353 | { |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3354 | struct email_alert *alert; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3355 | struct tcpcheck_rule *tcpcheck; |
| 3356 | struct check *check = &q->check; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3357 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3358 | if ((alert = pool_alloc(pool_head_email_alert)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3359 | goto error; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3360 | LIST_INIT(&alert->list); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3361 | LIST_INIT(&alert->tcpcheck_rules); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3362 | alert->srv = s; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3363 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3364 | if ((tcpcheck = pool_alloc(pool_head_tcpcheck_rule)) == NULL) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3365 | goto error; |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3366 | memset(tcpcheck, 0, sizeof(*tcpcheck)); |
| 3367 | tcpcheck->action = TCPCHK_ACT_CONNECT; |
| 3368 | tcpcheck->comment = NULL; |
| 3369 | tcpcheck->string = NULL; |
| 3370 | tcpcheck->expect_regex = NULL; |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3371 | LIST_ADDQ(&alert->tcpcheck_rules, &tcpcheck->list); |
| 3372 | |
| 3373 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "220 ")) |
| 3374 | goto error; |
| 3375 | |
| 3376 | { |
| 3377 | const char * const strs[4] = { "EHLO ", p->email_alert.myhostname, "\r\n" }; |
| 3378 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3379 | goto error; |
| 3380 | } |
| 3381 | |
| 3382 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3383 | goto error; |
| 3384 | |
| 3385 | { |
| 3386 | const char * const strs[4] = { "MAIL FROM:<", p->email_alert.from, ">\r\n" }; |
| 3387 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3388 | goto error; |
| 3389 | } |
| 3390 | |
| 3391 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3392 | goto error; |
| 3393 | |
| 3394 | { |
| 3395 | const char * const strs[4] = { "RCPT TO:<", p->email_alert.to, ">\r\n" }; |
| 3396 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3397 | goto error; |
| 3398 | } |
| 3399 | |
| 3400 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3401 | goto error; |
| 3402 | |
| 3403 | { |
| 3404 | const char * const strs[2] = { "DATA\r\n" }; |
| 3405 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3406 | goto error; |
| 3407 | } |
| 3408 | |
| 3409 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "354 ")) |
| 3410 | goto error; |
| 3411 | |
| 3412 | { |
| 3413 | struct tm tm; |
| 3414 | char datestr[48]; |
| 3415 | const char * const strs[18] = { |
Pieter Baauw | 5e0964e | 2016-02-13 16:27:35 +0100 | [diff] [blame] | 3416 | "From: ", p->email_alert.from, "\r\n", |
| 3417 | "To: ", p->email_alert.to, "\r\n", |
| 3418 | "Date: ", datestr, "\r\n", |
| 3419 | "Subject: [HAproxy Alert] ", msg, "\r\n", |
| 3420 | "\r\n", |
| 3421 | msg, "\r\n", |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3422 | "\r\n", |
Pieter Baauw | ed35c37 | 2015-07-22 19:51:54 +0200 | [diff] [blame] | 3423 | ".\r\n", |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3424 | NULL |
| 3425 | }; |
| 3426 | |
| 3427 | get_localtime(date.tv_sec, &tm); |
| 3428 | |
| 3429 | if (strftime(datestr, sizeof(datestr), "%a, %d %b %Y %T %z (%Z)", &tm) == 0) { |
| 3430 | goto error; |
| 3431 | } |
| 3432 | |
| 3433 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3434 | goto error; |
| 3435 | } |
| 3436 | |
| 3437 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "250 ")) |
| 3438 | goto error; |
| 3439 | |
| 3440 | { |
| 3441 | const char * const strs[2] = { "QUIT\r\n" }; |
| 3442 | if (!add_tcpcheck_send_strs(&alert->tcpcheck_rules, strs)) |
| 3443 | goto error; |
| 3444 | } |
| 3445 | |
| 3446 | if (!add_tcpcheck_expect_str(&alert->tcpcheck_rules, "221 ")) |
| 3447 | goto error; |
| 3448 | |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3449 | HA_SPIN_LOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3450 | task_wakeup(check->task, TASK_WOKEN_MSG); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3451 | LIST_ADDQ(&q->email_alerts, &alert->list); |
Christopher Faulet | 2a944ee | 2017-11-07 10:42:54 +0100 | [diff] [blame] | 3452 | HA_SPIN_UNLOCK(EMAIL_ALERTS_LOCK, &q->lock); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3453 | return 1; |
| 3454 | |
| 3455 | error: |
| 3456 | email_alert_free(alert); |
| 3457 | return 0; |
| 3458 | } |
| 3459 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3460 | 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] | 3461 | { |
| 3462 | int i; |
| 3463 | struct mailer *mailer; |
| 3464 | |
| 3465 | for (i = 0, mailer = p->email_alert.mailers.m->mailer_list; |
| 3466 | i < p->email_alert.mailers.m->count; i++, mailer = mailer->next) { |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3467 | 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] | 3468 | 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] | 3469 | return; |
| 3470 | } |
| 3471 | } |
| 3472 | |
| 3473 | return; |
| 3474 | } |
| 3475 | |
| 3476 | /* |
| 3477 | * Send email alert if configured. |
| 3478 | */ |
Simon Horman | 64e3416 | 2015-02-06 11:11:57 +0900 | [diff] [blame] | 3479 | void send_email_alert(struct server *s, int level, const char *format, ...) |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3480 | { |
| 3481 | va_list argp; |
| 3482 | char buf[1024]; |
| 3483 | int len; |
| 3484 | struct proxy *p = s->proxy; |
| 3485 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3486 | 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] | 3487 | return; |
| 3488 | |
| 3489 | va_start(argp, format); |
| 3490 | len = vsnprintf(buf, sizeof(buf), format, argp); |
| 3491 | va_end(argp); |
| 3492 | |
Thierry FOURNIER | 62c8a21 | 2017-02-09 12:19:27 +0100 | [diff] [blame] | 3493 | if (len < 0 || len >= sizeof(buf)) { |
Christopher Faulet | 767a84b | 2017-11-24 16:50:31 +0100 | [diff] [blame] | 3494 | ha_alert("Email alert [%s] could not format message\n", p->id); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3495 | return; |
| 3496 | } |
| 3497 | |
Christopher Faulet | 0108bb3 | 2017-10-20 21:34:32 +0200 | [diff] [blame] | 3498 | enqueue_email_alert(p, s, buf); |
Simon Horman | 0ba0e4a | 2015-01-30 11:23:00 +0900 | [diff] [blame] | 3499 | } |
| 3500 | |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3501 | /* |
| 3502 | * Return value: |
| 3503 | * the port to be used for the health check |
| 3504 | * 0 in case no port could be found for the check |
| 3505 | */ |
| 3506 | int srv_check_healthcheck_port(struct check *chk) |
| 3507 | { |
| 3508 | int i = 0; |
| 3509 | struct server *srv = NULL; |
| 3510 | |
| 3511 | srv = chk->server; |
| 3512 | |
| 3513 | /* If neither a port nor an addr was specified and no check transport |
| 3514 | * layer is forced, then the transport layer used by the checks is the |
| 3515 | * same as for the production traffic. Otherwise we use raw_sock by |
| 3516 | * default, unless one is specified. |
| 3517 | */ |
| 3518 | if (!chk->port && !is_addr(&chk->addr)) { |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3519 | chk->use_ssl |= (srv->use_ssl || (srv->proxy->options & PR_O_TCPCHK_SSL)); |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3520 | chk->send_proxy |= (srv->pp_opts); |
| 3521 | } |
| 3522 | |
| 3523 | /* by default, we use the health check port ocnfigured */ |
| 3524 | if (chk->port > 0) |
| 3525 | return chk->port; |
| 3526 | |
| 3527 | /* try to get the port from check_core.addr if check.port not set */ |
| 3528 | i = get_host_port(&chk->addr); |
| 3529 | if (i > 0) |
| 3530 | return i; |
| 3531 | |
| 3532 | /* try to get the port from server address */ |
| 3533 | /* prevent MAPPORTS from working at this point, since checks could |
| 3534 | * not be performed in such case (MAPPORTS impose a relative ports |
| 3535 | * based on live traffic) |
| 3536 | */ |
| 3537 | if (srv->flags & SRV_F_MAPPORTS) |
| 3538 | return 0; |
Willy Tarreau | 04276f3 | 2017-01-06 17:41:29 +0100 | [diff] [blame] | 3539 | |
| 3540 | i = srv->svc_port; /* by default */ |
Baptiste Assmann | 95db2bc | 2016-06-13 14:15:41 +0200 | [diff] [blame] | 3541 | if (i > 0) |
| 3542 | return i; |
| 3543 | |
| 3544 | return 0; |
| 3545 | } |
| 3546 | |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 3547 | __attribute__((constructor)) |
| 3548 | static void __check_init(void) |
| 3549 | { |
| 3550 | hap_register_post_check(start_checks); |
Christopher Faulet | 31dff9b | 2017-10-23 15:45:20 +0200 | [diff] [blame] | 3551 | |
Willy Tarreau | bafbe01 | 2017-11-24 17:34:44 +0100 | [diff] [blame] | 3552 | pool_head_email_alert = create_pool("email_alert", sizeof(struct email_alert), MEM_F_SHARED); |
| 3553 | pool_head_tcpcheck_rule = create_pool("tcpcheck_rule", sizeof(struct tcpcheck_rule), MEM_F_SHARED); |
Willy Tarreau | 865c514 | 2016-12-21 20:04:48 +0100 | [diff] [blame] | 3554 | } |
| 3555 | |
Baptiste Assmann | 5ecb77f | 2013-10-06 23:24:13 +0200 | [diff] [blame] | 3556 | |
Willy Tarreau | bd74154 | 2010-03-16 18:46:54 +0100 | [diff] [blame] | 3557 | /* |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 3558 | * Local variables: |
| 3559 | * c-indent-level: 8 |
| 3560 | * c-basic-offset: 8 |
| 3561 | * End: |
| 3562 | */ |