Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Backend variables and functions. |
| 3 | * |
| 4 | * Copyright 2000-2006 Willy Tarreau <w@1wt.eu> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | */ |
| 12 | |
| 13 | #include <errno.h> |
| 14 | #include <fcntl.h> |
| 15 | #include <stdio.h> |
| 16 | #include <stdlib.h> |
| 17 | #include <syslog.h> |
Willy Tarreau | f19cf37 | 2006-11-14 15:40:51 +0100 | [diff] [blame] | 18 | #include <string.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 19 | |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 20 | #include <common/compat.h> |
Willy Tarreau | e3ba5f0 | 2006-06-29 18:54:54 +0200 | [diff] [blame] | 21 | #include <common/config.h> |
Willy Tarreau | 2dd0d47 | 2006-06-29 17:53:05 +0200 | [diff] [blame] | 22 | #include <common/time.h> |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 23 | |
| 24 | #include <types/buffers.h> |
| 25 | #include <types/global.h> |
| 26 | #include <types/polling.h> |
| 27 | #include <types/proxy.h> |
| 28 | #include <types/server.h> |
| 29 | #include <types/session.h> |
| 30 | |
| 31 | #include <proto/backend.h> |
| 32 | #include <proto/fd.h> |
| 33 | #include <proto/log.h> |
| 34 | #include <proto/proto_http.h> |
| 35 | #include <proto/queue.h> |
| 36 | #include <proto/stream_sock.h> |
| 37 | #include <proto/task.h> |
| 38 | |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 39 | #ifdef CONFIG_HAP_CTTPROXY |
| 40 | #include <import/ip_tproxy.h> |
| 41 | #endif |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 42 | |
| 43 | /* |
| 44 | * This function recounts the number of usable active and backup servers for |
| 45 | * proxy <p>. These numbers are returned into the p->srv_act and p->srv_bck. |
| 46 | * This function also recomputes the total active and backup weights. |
| 47 | */ |
| 48 | void recount_servers(struct proxy *px) |
| 49 | { |
| 50 | struct server *srv; |
| 51 | |
| 52 | px->srv_act = 0; px->srv_bck = px->tot_wact = px->tot_wbck = 0; |
| 53 | for (srv = px->srv; srv != NULL; srv = srv->next) { |
| 54 | if (srv->state & SRV_RUNNING) { |
| 55 | if (srv->state & SRV_BACKUP) { |
| 56 | px->srv_bck++; |
| 57 | px->tot_wbck += srv->eweight + 1; |
| 58 | } else { |
| 59 | px->srv_act++; |
| 60 | px->tot_wact += srv->eweight + 1; |
| 61 | } |
| 62 | } |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | /* This function recomputes the server map for proxy px. It |
| 67 | * relies on px->tot_wact and px->tot_wbck, so it must be |
| 68 | * called after recount_servers(). It also expects px->srv_map |
| 69 | * to be initialized to the largest value needed. |
| 70 | */ |
| 71 | void recalc_server_map(struct proxy *px) |
| 72 | { |
| 73 | int o, tot, flag; |
| 74 | struct server *cur, *best; |
| 75 | |
| 76 | if (px->srv_act) { |
| 77 | flag = SRV_RUNNING; |
| 78 | tot = px->tot_wact; |
| 79 | } else if (px->srv_bck) { |
| 80 | flag = SRV_RUNNING | SRV_BACKUP; |
| 81 | if (px->options & PR_O_USE_ALL_BK) |
| 82 | tot = px->tot_wbck; |
| 83 | else |
| 84 | tot = 1; /* the first server is enough */ |
| 85 | } else { |
| 86 | px->srv_map_sz = 0; |
| 87 | return; |
| 88 | } |
| 89 | |
| 90 | /* this algorithm gives priority to the first server, which means that |
| 91 | * it will respect the declaration order for equivalent weights, and |
| 92 | * that whatever the weights, the first server called will always be |
| 93 | * the first declard. This is an important asumption for the backup |
| 94 | * case, where we want the first server only. |
| 95 | */ |
| 96 | for (cur = px->srv; cur; cur = cur->next) |
| 97 | cur->wscore = 0; |
| 98 | |
| 99 | for (o = 0; o < tot; o++) { |
| 100 | int max = 0; |
| 101 | best = NULL; |
| 102 | for (cur = px->srv; cur; cur = cur->next) { |
| 103 | if ((cur->state & (SRV_RUNNING | SRV_BACKUP)) == flag) { |
| 104 | int v; |
| 105 | |
| 106 | /* If we are forced to return only one server, we don't want to |
| 107 | * go further, because we would return the wrong one due to |
| 108 | * divide overflow. |
| 109 | */ |
| 110 | if (tot == 1) { |
| 111 | best = cur; |
| 112 | break; |
| 113 | } |
| 114 | |
| 115 | cur->wscore += cur->eweight + 1; |
| 116 | v = (cur->wscore + tot) / tot; /* result between 0 and 3 */ |
| 117 | if (best == NULL || v > max) { |
| 118 | max = v; |
| 119 | best = cur; |
| 120 | } |
| 121 | } |
| 122 | } |
| 123 | px->srv_map[o] = best; |
| 124 | best->wscore -= tot; |
| 125 | } |
| 126 | px->srv_map_sz = tot; |
| 127 | } |
| 128 | |
| 129 | |
| 130 | /* |
| 131 | * This function marks the session as 'assigned' in direct or dispatch modes, |
| 132 | * or tries to assign one in balance mode, according to the algorithm. It does |
| 133 | * nothing if the session had already been assigned a server. |
| 134 | * |
| 135 | * It may return : |
| 136 | * SRV_STATUS_OK if everything is OK. s->srv will be valid. |
| 137 | * SRV_STATUS_NOSRV if no server is available. s->srv = NULL. |
| 138 | * SRV_STATUS_FULL if all servers are saturated. s->srv = NULL. |
| 139 | * SRV_STATUS_INTERNAL for other unrecoverable errors. |
| 140 | * |
| 141 | * Upon successful return, the session flag SN_ASSIGNED to indicate that it does |
| 142 | * not need to be called anymore. This usually means that s->srv can be trusted |
| 143 | * in balance and direct modes. This flag is not cleared, so it's to the caller |
| 144 | * to clear it if required (eg: redispatch). |
| 145 | * |
| 146 | */ |
| 147 | |
| 148 | int assign_server(struct session *s) |
| 149 | { |
| 150 | #ifdef DEBUG_FULL |
| 151 | fprintf(stderr,"assign_server : s=%p\n",s); |
| 152 | #endif |
| 153 | |
| 154 | if (s->pend_pos) |
| 155 | return SRV_STATUS_INTERNAL; |
| 156 | |
| 157 | if (!(s->flags & SN_ASSIGNED)) { |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 158 | if ((s->be->options & PR_O_BALANCE) && !(s->flags & SN_DIRECT)) { |
| 159 | if (!s->be->srv_act && !s->be->srv_bck) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 160 | return SRV_STATUS_NOSRV; |
| 161 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 162 | if (s->be->options & PR_O_BALANCE_RR) { |
| 163 | s->srv = get_server_rr_with_conns(s->be); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 164 | if (!s->srv) |
| 165 | return SRV_STATUS_FULL; |
| 166 | } |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 167 | else if (s->be->options & PR_O_BALANCE_SH) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 168 | int len; |
| 169 | |
| 170 | if (s->cli_addr.ss_family == AF_INET) |
| 171 | len = 4; |
| 172 | else if (s->cli_addr.ss_family == AF_INET6) |
| 173 | len = 16; |
| 174 | else /* unknown IP family */ |
| 175 | return SRV_STATUS_INTERNAL; |
| 176 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 177 | s->srv = get_server_sh(s->be, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 178 | (void *)&((struct sockaddr_in *)&s->cli_addr)->sin_addr, |
| 179 | len); |
| 180 | } |
| 181 | else /* unknown balancing algorithm */ |
| 182 | return SRV_STATUS_INTERNAL; |
| 183 | } |
| 184 | s->flags |= SN_ASSIGNED; |
| 185 | } |
| 186 | return SRV_STATUS_OK; |
| 187 | } |
| 188 | |
| 189 | |
| 190 | /* |
| 191 | * This function assigns a server address to a session, and sets SN_ADDR_SET. |
| 192 | * The address is taken from the currently assigned server, or from the |
| 193 | * dispatch or transparent address. |
| 194 | * |
| 195 | * It may return : |
| 196 | * SRV_STATUS_OK if everything is OK. |
| 197 | * SRV_STATUS_INTERNAL for other unrecoverable errors. |
| 198 | * |
| 199 | * Upon successful return, the session flag SN_ADDR_SET is set. This flag is |
| 200 | * not cleared, so it's to the caller to clear it if required. |
| 201 | * |
| 202 | */ |
| 203 | int assign_server_address(struct session *s) |
| 204 | { |
| 205 | #ifdef DEBUG_FULL |
| 206 | fprintf(stderr,"assign_server_address : s=%p\n",s); |
| 207 | #endif |
| 208 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 209 | if (s->flags & SN_DIRECT || s->be->options & PR_O_BALANCE) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 210 | /* A server is necessarily known for this session */ |
| 211 | if (!(s->flags & SN_ASSIGNED)) |
| 212 | return SRV_STATUS_INTERNAL; |
| 213 | |
| 214 | s->srv_addr = s->srv->addr; |
| 215 | |
| 216 | /* if this server remaps proxied ports, we'll use |
| 217 | * the port the client connected to with an offset. */ |
| 218 | if (s->srv->state & SRV_MAPPORTS) { |
| 219 | struct sockaddr_in sockname; |
| 220 | socklen_t namelen = sizeof(sockname); |
| 221 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 222 | if (!(s->fe->options & PR_O_TRANSP) || |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 223 | get_original_dst(s->cli_fd, (struct sockaddr_in *)&sockname, &namelen) == -1) |
| 224 | getsockname(s->cli_fd, (struct sockaddr *)&sockname, &namelen); |
| 225 | s->srv_addr.sin_port = htons(ntohs(s->srv_addr.sin_port) + ntohs(sockname.sin_port)); |
| 226 | } |
| 227 | } |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 228 | else if (*(int *)&s->be->dispatch_addr.sin_addr) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 229 | /* connect to the defined dispatch addr */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 230 | s->srv_addr = s->be->dispatch_addr; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 231 | } |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 232 | else if (s->fe->options & PR_O_TRANSP) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 233 | /* in transparent mode, use the original dest addr if no dispatch specified */ |
| 234 | socklen_t salen = sizeof(s->srv_addr); |
| 235 | |
| 236 | if (get_original_dst(s->cli_fd, &s->srv_addr, &salen) == -1) { |
| 237 | qfprintf(stderr, "Cannot get original server address.\n"); |
| 238 | return SRV_STATUS_INTERNAL; |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | s->flags |= SN_ADDR_SET; |
| 243 | return SRV_STATUS_OK; |
| 244 | } |
| 245 | |
| 246 | |
| 247 | /* This function assigns a server to session <s> if required, and can add the |
| 248 | * connection to either the assigned server's queue or to the proxy's queue. |
| 249 | * |
| 250 | * Returns : |
| 251 | * |
| 252 | * SRV_STATUS_OK if everything is OK. |
| 253 | * SRV_STATUS_NOSRV if no server is available. s->srv = NULL. |
| 254 | * SRV_STATUS_QUEUED if the connection has been queued. |
| 255 | * SRV_STATUS_FULL if the server(s) is/are saturated and the |
| 256 | * connection could not be queued. |
| 257 | * SRV_STATUS_INTERNAL for other unrecoverable errors. |
| 258 | * |
| 259 | */ |
| 260 | int assign_server_and_queue(struct session *s) |
| 261 | { |
| 262 | struct pendconn *p; |
| 263 | int err; |
| 264 | |
| 265 | if (s->pend_pos) |
| 266 | return SRV_STATUS_INTERNAL; |
| 267 | |
| 268 | if (s->flags & SN_ASSIGNED) { |
| 269 | /* a server does not need to be assigned, perhaps because we're in |
| 270 | * direct mode, or in dispatch or transparent modes where the server |
| 271 | * is not needed. |
| 272 | */ |
| 273 | if (s->srv && |
| 274 | s->srv->maxconn && s->srv->cur_sess >= srv_dynamic_maxconn(s->srv)) { |
| 275 | p = pendconn_add(s); |
| 276 | if (p) |
| 277 | return SRV_STATUS_QUEUED; |
| 278 | else |
| 279 | return SRV_STATUS_FULL; |
| 280 | } |
| 281 | return SRV_STATUS_OK; |
| 282 | } |
| 283 | |
| 284 | /* a server needs to be assigned */ |
| 285 | err = assign_server(s); |
| 286 | switch (err) { |
| 287 | case SRV_STATUS_OK: |
| 288 | /* in balance mode, we might have servers with connection limits */ |
| 289 | if (s->srv && |
| 290 | s->srv->maxconn && s->srv->cur_sess >= srv_dynamic_maxconn(s->srv)) { |
| 291 | p = pendconn_add(s); |
| 292 | if (p) |
| 293 | return SRV_STATUS_QUEUED; |
| 294 | else |
| 295 | return SRV_STATUS_FULL; |
| 296 | } |
| 297 | return SRV_STATUS_OK; |
| 298 | |
| 299 | case SRV_STATUS_FULL: |
| 300 | /* queue this session into the proxy's queue */ |
| 301 | p = pendconn_add(s); |
| 302 | if (p) |
| 303 | return SRV_STATUS_QUEUED; |
| 304 | else |
| 305 | return SRV_STATUS_FULL; |
| 306 | |
| 307 | case SRV_STATUS_NOSRV: |
| 308 | case SRV_STATUS_INTERNAL: |
| 309 | return err; |
| 310 | default: |
| 311 | return SRV_STATUS_INTERNAL; |
| 312 | } |
| 313 | } |
| 314 | |
| 315 | |
| 316 | /* |
| 317 | * This function initiates a connection to the server assigned to this session |
| 318 | * (s->srv, s->srv_addr). It will assign a server if none is assigned yet. |
| 319 | * It can return one of : |
| 320 | * - SN_ERR_NONE if everything's OK |
| 321 | * - SN_ERR_SRVTO if there are no more servers |
| 322 | * - SN_ERR_SRVCL if the connection was refused by the server |
| 323 | * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn) |
| 324 | * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 325 | * - SN_ERR_INTERNAL for any other purely internal errors |
| 326 | * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted. |
| 327 | */ |
| 328 | int connect_server(struct session *s) |
| 329 | { |
| 330 | int fd, err; |
| 331 | |
| 332 | if (!(s->flags & SN_ADDR_SET)) { |
| 333 | err = assign_server_address(s); |
| 334 | if (err != SRV_STATUS_OK) |
| 335 | return SN_ERR_INTERNAL; |
| 336 | } |
| 337 | |
| 338 | if ((fd = s->srv_fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) == -1) { |
| 339 | qfprintf(stderr, "Cannot get a server socket.\n"); |
| 340 | |
| 341 | if (errno == ENFILE) |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 342 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 343 | "Proxy %s reached system FD limit at %d. Please check system tunables.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 344 | s->be->id, maxfd); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 345 | else if (errno == EMFILE) |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 346 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 347 | "Proxy %s reached process FD limit at %d. Please check 'ulimit-n' and restart.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 348 | s->be->id, maxfd); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 349 | else if (errno == ENOBUFS || errno == ENOMEM) |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 350 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 351 | "Proxy %s reached system memory limit at %d sockets. Please check system tunables.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 352 | s->be->id, maxfd); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 353 | /* this is a resource error */ |
| 354 | return SN_ERR_RESOURCE; |
| 355 | } |
| 356 | |
| 357 | if (fd >= global.maxsock) { |
| 358 | /* do not log anything there, it's a normal condition when this option |
| 359 | * is used to serialize connections to a server ! |
| 360 | */ |
| 361 | Alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n"); |
| 362 | close(fd); |
| 363 | return SN_ERR_PRXCOND; /* it is a configuration limit */ |
| 364 | } |
| 365 | |
| 366 | if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1) || |
| 367 | (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *) &one, sizeof(one)) == -1)) { |
| 368 | qfprintf(stderr,"Cannot set client socket to non blocking mode.\n"); |
| 369 | close(fd); |
| 370 | return SN_ERR_INTERNAL; |
| 371 | } |
| 372 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 373 | if (s->be->options & PR_O_TCP_SRV_KA) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 374 | setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, (char *) &one, sizeof(one)); |
| 375 | |
| 376 | /* allow specific binding : |
| 377 | * - server-specific at first |
| 378 | * - proxy-specific next |
| 379 | */ |
| 380 | if (s->srv != NULL && s->srv->state & SRV_BIND_SRC) { |
| 381 | setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &one, sizeof(one)); |
| 382 | if (bind(fd, (struct sockaddr *)&s->srv->source_addr, sizeof(s->srv->source_addr)) == -1) { |
| 383 | Alert("Cannot bind to source address before connect() for server %s/%s. Aborting.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 384 | s->be->id, s->srv->id); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 385 | close(fd); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 386 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 387 | "Cannot bind to source address before connect() for server %s/%s.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 388 | s->be->id, s->srv->id); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 389 | return SN_ERR_RESOURCE; |
| 390 | } |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 391 | #ifdef CONFIG_HAP_CTTPROXY |
| 392 | if (s->srv->state & SRV_TPROXY_MASK) { |
| 393 | struct in_tproxy itp1, itp2; |
| 394 | memset(&itp1, 0, sizeof(itp1)); |
| 395 | |
| 396 | itp1.op = TPROXY_ASSIGN; |
| 397 | switch (s->srv->state & SRV_TPROXY_MASK) { |
| 398 | case SRV_TPROXY_ADDR: |
| 399 | itp1.v.addr.faddr = s->srv->tproxy_addr.sin_addr; |
| 400 | itp1.v.addr.fport = s->srv->tproxy_addr.sin_port; |
| 401 | break; |
| 402 | case SRV_TPROXY_CLI: |
| 403 | itp1.v.addr.fport = ((struct sockaddr_in *)&s->cli_addr)->sin_port; |
| 404 | /* fall through */ |
| 405 | case SRV_TPROXY_CIP: |
| 406 | /* FIXME: what can we do if the client connects in IPv6 ? */ |
| 407 | itp1.v.addr.faddr = ((struct sockaddr_in *)&s->cli_addr)->sin_addr; |
| 408 | break; |
| 409 | } |
| 410 | |
| 411 | /* set connect flag on socket */ |
| 412 | itp2.op = TPROXY_FLAGS; |
| 413 | itp2.v.flags = ITP_CONNECT | ITP_ONCE; |
| 414 | |
| 415 | if (setsockopt(fd, SOL_IP, IP_TPROXY, &itp1, sizeof(itp1)) == -1 || |
| 416 | setsockopt(fd, SOL_IP, IP_TPROXY, &itp2, sizeof(itp2)) == -1) { |
| 417 | Alert("Cannot bind to tproxy source address before connect() for server %s/%s. Aborting.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 418 | s->be->id, s->srv->id); |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 419 | close(fd); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 420 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 421 | "Cannot bind to tproxy source address before connect() for server %s/%s.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 422 | s->be->id, s->srv->id); |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 423 | return SN_ERR_RESOURCE; |
| 424 | } |
| 425 | } |
| 426 | #endif |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 427 | } |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 428 | else if (s->be->options & PR_O_BIND_SRC) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 429 | setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &one, sizeof(one)); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 430 | if (bind(fd, (struct sockaddr *)&s->be->source_addr, sizeof(s->be->source_addr)) == -1) { |
| 431 | Alert("Cannot bind to source address before connect() for proxy %s. Aborting.\n", s->be->id); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 432 | close(fd); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 433 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 434 | "Cannot bind to source address before connect() for server %s/%s.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 435 | s->be->id, s->srv->id); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 436 | return SN_ERR_RESOURCE; |
| 437 | } |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 438 | #ifdef CONFIG_HAP_CTTPROXY |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 439 | if (s->be->options & PR_O_TPXY_MASK) { |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 440 | struct in_tproxy itp1, itp2; |
| 441 | memset(&itp1, 0, sizeof(itp1)); |
| 442 | |
| 443 | itp1.op = TPROXY_ASSIGN; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 444 | switch (s->be->options & PR_O_TPXY_MASK) { |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 445 | case PR_O_TPXY_ADDR: |
| 446 | itp1.v.addr.faddr = s->srv->tproxy_addr.sin_addr; |
| 447 | itp1.v.addr.fport = s->srv->tproxy_addr.sin_port; |
| 448 | break; |
| 449 | case PR_O_TPXY_CLI: |
| 450 | itp1.v.addr.fport = ((struct sockaddr_in *)&s->cli_addr)->sin_port; |
| 451 | /* fall through */ |
| 452 | case PR_O_TPXY_CIP: |
| 453 | /* FIXME: what can we do if the client connects in IPv6 ? */ |
| 454 | itp1.v.addr.faddr = ((struct sockaddr_in *)&s->cli_addr)->sin_addr; |
| 455 | break; |
| 456 | } |
| 457 | |
| 458 | /* set connect flag on socket */ |
| 459 | itp2.op = TPROXY_FLAGS; |
| 460 | itp2.v.flags = ITP_CONNECT | ITP_ONCE; |
| 461 | |
| 462 | if (setsockopt(fd, SOL_IP, IP_TPROXY, &itp1, sizeof(itp1)) == -1 || |
| 463 | setsockopt(fd, SOL_IP, IP_TPROXY, &itp2, sizeof(itp2)) == -1) { |
| 464 | Alert("Cannot bind to tproxy source address before connect() for proxy %s. Aborting.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 465 | s->be->id); |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 466 | close(fd); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 467 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 468 | "Cannot bind to tproxy source address before connect() for server %s/%s.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 469 | s->be->id, s->srv->id); |
Willy Tarreau | 77074d5 | 2006-11-12 23:57:19 +0100 | [diff] [blame] | 470 | return SN_ERR_RESOURCE; |
| 471 | } |
| 472 | } |
| 473 | #endif |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 474 | } |
| 475 | |
| 476 | if ((connect(fd, (struct sockaddr *)&s->srv_addr, sizeof(s->srv_addr)) == -1) && |
| 477 | (errno != EINPROGRESS) && (errno != EALREADY) && (errno != EISCONN)) { |
| 478 | |
| 479 | if (errno == EAGAIN || errno == EADDRINUSE) { |
| 480 | char *msg; |
| 481 | if (errno == EAGAIN) /* no free ports left, try again later */ |
| 482 | msg = "no free ports"; |
| 483 | else |
| 484 | msg = "local address already in use"; |
| 485 | |
| 486 | qfprintf(stderr,"Cannot connect: %s.\n",msg); |
| 487 | close(fd); |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 488 | send_log(s->be, LOG_EMERG, |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 489 | "Connect() failed for server %s/%s: %s.\n", |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 490 | s->be->id, s->srv->id, msg); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 491 | return SN_ERR_RESOURCE; |
| 492 | } else if (errno == ETIMEDOUT) { |
| 493 | //qfprintf(stderr,"Connect(): ETIMEDOUT"); |
| 494 | close(fd); |
| 495 | return SN_ERR_SRVTO; |
| 496 | } else { |
| 497 | // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM) |
| 498 | //qfprintf(stderr,"Connect(): %d", errno); |
| 499 | close(fd); |
| 500 | return SN_ERR_SRVCL; |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | fdtab[fd].owner = s->task; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 505 | fdtab[fd].state = FD_STCONN; /* connection in progress */ |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 506 | fdtab[fd].cb[DIR_RD].f = &stream_sock_read; |
Willy Tarreau | 5446940 | 2006-07-29 16:59:06 +0200 | [diff] [blame] | 507 | fdtab[fd].cb[DIR_RD].b = s->rep; |
Willy Tarreau | f8306d5 | 2006-07-29 19:01:31 +0200 | [diff] [blame] | 508 | fdtab[fd].cb[DIR_WR].f = &stream_sock_write; |
Willy Tarreau | 5446940 | 2006-07-29 16:59:06 +0200 | [diff] [blame] | 509 | fdtab[fd].cb[DIR_WR].b = s->req; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 510 | |
Willy Tarreau | 2a42950 | 2006-10-15 14:52:29 +0200 | [diff] [blame] | 511 | MY_FD_SET(fd, StaticWriteEvent); /* for connect status */ |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 512 | #if defined(DEBUG_FULL) && defined(ENABLE_EPOLL) |
| 513 | if (PrevReadEvent) { |
Willy Tarreau | 2a42950 | 2006-10-15 14:52:29 +0200 | [diff] [blame] | 514 | assert(!(MY_FD_ISSET(fd, PrevReadEvent))); |
| 515 | assert(!(MY_FD_ISSET(fd, PrevWriteEvent))); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 516 | } |
| 517 | #endif |
| 518 | |
| 519 | fd_insert(fd); |
| 520 | if (s->srv) { |
| 521 | s->srv->cur_sess++; |
| 522 | if (s->srv->cur_sess > s->srv->cur_sess_max) |
| 523 | s->srv->cur_sess_max = s->srv->cur_sess; |
| 524 | } |
| 525 | |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 526 | if (s->be->contimeout) |
| 527 | tv_delayfrom(&s->req->cex, &now, s->be->contimeout); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 528 | else |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 529 | tv_eternity(&s->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 530 | return SN_ERR_NONE; /* connection is OK */ |
| 531 | } |
| 532 | |
| 533 | |
| 534 | /* |
| 535 | * This function checks the retry count during the connect() job. |
| 536 | * It updates the session's srv_state and retries, so that the caller knows |
| 537 | * what it has to do. It uses the last connection error to set the log when |
| 538 | * it expires. It returns 1 when it has expired, and 0 otherwise. |
| 539 | */ |
| 540 | int srv_count_retry_down(struct session *t, int conn_err) |
| 541 | { |
| 542 | /* we are in front of a retryable error */ |
| 543 | t->conn_retries--; |
| 544 | if (t->conn_retries < 0) { |
| 545 | /* if not retryable anymore, let's abort */ |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 546 | tv_eternity(&t->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 547 | srv_close_with_err(t, conn_err, SN_FINST_C, |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 548 | 503, t->fe->errmsg.len503, t->fe->errmsg.msg503); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 549 | if (t->srv) |
| 550 | t->srv->failed_conns++; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 551 | t->be->failed_conns++; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 552 | |
| 553 | /* We used to have a free connection slot. Since we'll never use it, |
| 554 | * we have to inform the server that it may be used by another session. |
| 555 | */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 556 | if (may_dequeue_tasks(t->srv, t->be)) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 557 | task_wakeup(&rq, t->srv->queue_mgt); |
| 558 | return 1; |
| 559 | } |
| 560 | return 0; |
| 561 | } |
| 562 | |
| 563 | |
| 564 | /* |
| 565 | * This function performs the retryable part of the connect() job. |
| 566 | * It updates the session's srv_state and retries, so that the caller knows |
| 567 | * what it has to do. It returns 1 when it breaks out of the loop, or 0 if |
| 568 | * it needs to redispatch. |
| 569 | */ |
| 570 | int srv_retryable_connect(struct session *t) |
| 571 | { |
| 572 | int conn_err; |
| 573 | |
| 574 | /* This loop ensures that we stop before the last retry in case of a |
| 575 | * redispatchable server. |
| 576 | */ |
| 577 | do { |
| 578 | /* initiate a connection to the server */ |
| 579 | conn_err = connect_server(t); |
| 580 | switch (conn_err) { |
| 581 | |
| 582 | case SN_ERR_NONE: |
| 583 | //fprintf(stderr,"0: c=%d, s=%d\n", c, s); |
| 584 | t->srv_state = SV_STCONN; |
| 585 | return 1; |
| 586 | |
| 587 | case SN_ERR_INTERNAL: |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 588 | tv_eternity(&t->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 589 | srv_close_with_err(t, SN_ERR_INTERNAL, SN_FINST_C, |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 590 | 500, t->fe->errmsg.len500, t->fe->errmsg.msg500); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 591 | if (t->srv) |
| 592 | t->srv->failed_conns++; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 593 | t->be->failed_conns++; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 594 | /* release other sessions waiting for this server */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 595 | if (may_dequeue_tasks(t->srv, t->be)) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 596 | task_wakeup(&rq, t->srv->queue_mgt); |
| 597 | return 1; |
| 598 | } |
| 599 | /* ensure that we have enough retries left */ |
| 600 | if (srv_count_retry_down(t, conn_err)) { |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 601 | return 1; |
| 602 | } |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 603 | } while (t->srv == NULL || t->conn_retries > 0 || !(t->be->options & PR_O_REDISP)); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 604 | |
| 605 | /* We're on our last chance, and the REDISP option was specified. |
| 606 | * We will ignore cookie and force to balance or use the dispatcher. |
| 607 | */ |
| 608 | /* let's try to offer this slot to anybody */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 609 | if (may_dequeue_tasks(t->srv, t->be)) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 610 | task_wakeup(&rq, t->srv->queue_mgt); |
| 611 | |
| 612 | if (t->srv) |
| 613 | t->srv->failed_conns++; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 614 | t->be->failed_conns++; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 615 | |
| 616 | t->flags &= ~(SN_DIRECT | SN_ASSIGNED | SN_ADDR_SET); |
| 617 | t->srv = NULL; /* it's left to the dispatcher to choose a server */ |
| 618 | if ((t->flags & SN_CK_MASK) == SN_CK_VALID) { |
| 619 | t->flags &= ~SN_CK_MASK; |
| 620 | t->flags |= SN_CK_DOWN; |
| 621 | } |
| 622 | return 0; |
| 623 | } |
| 624 | |
| 625 | |
| 626 | /* This function performs the "redispatch" part of a connection attempt. It |
| 627 | * will assign a server if required, queue the connection if required, and |
| 628 | * handle errors that might arise at this level. It can change the server |
| 629 | * state. It will return 1 if it encounters an error, switches the server |
| 630 | * state, or has to queue a connection. Otherwise, it will return 0 indicating |
| 631 | * that the connection is ready to use. |
| 632 | */ |
| 633 | |
| 634 | int srv_redispatch_connect(struct session *t) |
| 635 | { |
| 636 | int conn_err; |
| 637 | |
| 638 | /* We know that we don't have any connection pending, so we will |
| 639 | * try to get a new one, and wait in this state if it's queued |
| 640 | */ |
| 641 | conn_err = assign_server_and_queue(t); |
| 642 | switch (conn_err) { |
| 643 | case SRV_STATUS_OK: |
| 644 | break; |
| 645 | |
| 646 | case SRV_STATUS_NOSRV: |
| 647 | /* note: it is guaranteed that t->srv == NULL here */ |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 648 | tv_eternity(&t->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 649 | srv_close_with_err(t, SN_ERR_SRVTO, SN_FINST_C, |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 650 | 503, t->fe->errmsg.len503, t->fe->errmsg.msg503); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 651 | if (t->srv) |
| 652 | t->srv->failed_conns++; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 653 | t->be->failed_conns++; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 654 | |
| 655 | return 1; |
| 656 | |
| 657 | case SRV_STATUS_QUEUED: |
| 658 | /* FIXME-20060503 : we should use the queue timeout instead */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 659 | if (t->be->contimeout) |
| 660 | tv_delayfrom(&t->req->cex, &now, t->be->contimeout); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 661 | else |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 662 | tv_eternity(&t->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 663 | t->srv_state = SV_STIDLE; |
| 664 | /* do nothing else and do not wake any other session up */ |
| 665 | return 1; |
| 666 | |
| 667 | case SRV_STATUS_FULL: |
| 668 | case SRV_STATUS_INTERNAL: |
| 669 | default: |
Willy Tarreau | d797128 | 2006-07-29 18:36:34 +0200 | [diff] [blame] | 670 | tv_eternity(&t->req->cex); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 671 | srv_close_with_err(t, SN_ERR_INTERNAL, SN_FINST_C, |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 672 | 500, t->fe->errmsg.len500, t->fe->errmsg.msg500); |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 673 | if (t->srv) |
| 674 | t->srv->failed_conns++; |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 675 | t->be->failed_conns++; |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 676 | |
| 677 | /* release other sessions waiting for this server */ |
Willy Tarreau | 73de989 | 2006-11-30 11:40:23 +0100 | [diff] [blame] | 678 | if (may_dequeue_tasks(t->srv, t->be)) |
Willy Tarreau | baaee00 | 2006-06-26 02:48:02 +0200 | [diff] [blame] | 679 | task_wakeup(&rq, t->srv->queue_mgt); |
| 680 | return 1; |
| 681 | } |
| 682 | /* if we get here, it's because we got SRV_STATUS_OK, which also |
| 683 | * means that the connection has not been queued. |
| 684 | */ |
| 685 | return 0; |
| 686 | } |
| 687 | |
| 688 | |
| 689 | /* |
| 690 | * Local variables: |
| 691 | * c-indent-level: 8 |
| 692 | * c-basic-offset: 8 |
| 693 | * End: |
| 694 | */ |