ctdb/tcp/tcp_connect.c

   1 /*
   2    ctdb over TCP
   3
   4    Copyright (C) Andrew Tridgell  2006
   5    Copyright (C) Ronnie Sahlberg  2008
   6
   7    This program is free software; you can redistribute it and/or modify
   8    it under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3 of the License, or
  10    (at your option) any later version.
  11
  12    This program is distributed in the hope that it will be useful,
  13    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15    GNU General Public License for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "includes.h"
  22 #include "tdb.h"
  23 #include "system/network.h"
  24 #include "system/filesys.h"
  25 #include "../include/ctdb_private.h"
  26 #include "ctdb_tcp.h"
  27
  28 /*
  29   stop any connecting (established or pending) to a node
  30  */
  31 void ctdb_tcp_stop_connection(struct ctdb_node *node)
  32 {
  33         struct ctdb_tcp_node *tnode = talloc_get_type(
  34                 node->private_data, struct ctdb_tcp_node);
  35
  36         ctdb_queue_set_fd(tnode->out_queue, -1);
  37         talloc_free(tnode->connect_te);
  38         talloc_free(tnode->connect_fde);
  39         tnode->connect_fde = NULL;
  40         tnode->connect_te = NULL;
  41         if (tnode->fd != -1) {
  42                 close(tnode->fd);
  43                 tnode->fd = -1;
  44         }
  45 }
  46
  47
  48 /*
  49   called when a complete packet has come in - should not happen on this socket
  50   unless the other side closes the connection with RST or FIN
  51  */
  52 void ctdb_tcp_tnode_cb(uint8_t *data, size_t cnt, void *private_data)
  53 {
  54         struct ctdb_node *node = talloc_get_type(private_data, struct ctdb_node);
  55         struct ctdb_tcp_node *tnode = talloc_get_type(
  56                 node->private_data, struct ctdb_tcp_node);
  57
  58         if (data == NULL) {
  59                 node->ctdb->upcalls->node_dead(node);
  60         }
  61
  62         ctdb_tcp_stop_connection(node);
  63         tnode->connect_te = event_add_timed(node->ctdb->ev, tnode,
  64                                             timeval_current_ofs(3, 0),
  65                                             ctdb_tcp_node_connect, node);
  66 }
  67
  68 /*
  69   called when socket becomes writeable on connect
  70 */
  71 static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *fde,
  72                                     uint16_t flags, void *private_data)
  73 {
  74         struct ctdb_node *node = talloc_get_type(private_data,
  75                                                  struct ctdb_node);
  76         struct ctdb_tcp_node *tnode = talloc_get_type(node->private_data,
  77                                                       struct ctdb_tcp_node);
  78         struct ctdb_context *ctdb = node->ctdb;
  79         int error = 0;
  80         socklen_t len = sizeof(error);
  81         int one = 1;
  82
  83         talloc_free(tnode->connect_te);
  84         tnode->connect_te = NULL;
  85
  86         if (getsockopt(tnode->fd, SOL_SOCKET, SO_ERROR, &error, &len) != 0 ||
  87             error != 0) {
  88                 ctdb_tcp_stop_connection(node);
  89                 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
  90                                                     timeval_current_ofs(1, 0),
  91                                                     ctdb_tcp_node_connect, node);
  92                 return;
  93         }
  94
  95         talloc_free(tnode->connect_fde);
  96         tnode->connect_fde = NULL;
  97
  98         setsockopt(tnode->fd,IPPROTO_TCP,TCP_NODELAY,(char *)&one,sizeof(one));
  99         setsockopt(tnode->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
 100
 101         ctdb_queue_set_fd(tnode->out_queue, tnode->fd);
 102
 103         /* the queue subsystem now owns this fd */
 104         tnode->fd = -1;
 105 }
 106
 107
 108 static int ctdb_tcp_get_address(struct ctdb_context *ctdb,
 109                                 const char *address, ctdb_sock_addr *addr)
 110 {
 111         if (parse_ip(address, NULL, 0, addr) == 0) {
 112                 DEBUG(DEBUG_CRIT, (__location__ " Unparsable address : %s.\n", address));
 113                 return -1;
 114         }
 115         return 0;
 116 }
 117
 118 /*
 119   called when we should try and establish a tcp connection to a node
 120 */
 121 void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
 122                            struct timeval t, void *private_data)
 123 {
 124         struct ctdb_node *node = talloc_get_type(private_data,
 125                                                  struct ctdb_node);
 126         struct ctdb_tcp_node *tnode = talloc_get_type(node->private_data,
 127                                                       struct ctdb_tcp_node);
 128         struct ctdb_context *ctdb = node->ctdb;
 129         ctdb_sock_addr sock_in;
 130         int sockin_size;
 131         int sockout_size;
 132         ctdb_sock_addr sock_out;
 133
 134         ctdb_tcp_stop_connection(node);
 135
 136         ZERO_STRUCT(sock_out);
 137 #ifdef HAVE_SOCK_SIN_LEN
 138         sock_out.ip.sin_len = sizeof(sock_out);
 139 #endif
 140         if (ctdb_tcp_get_address(ctdb, node->address.address, &sock_out) != 0) {
 141                 return;
 142         }
 143         switch (sock_out.sa.sa_family) {
 144         case AF_INET:
 145                 sock_out.ip.sin_port = htons(node->address.port);
 146                 break;
 147         case AF_INET6:
 148                 sock_out.ip6.sin6_port = htons(node->address.port);
 149                 break;
 150         default:
 151                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 152                         sock_out.sa.sa_family));
 153                 return;
 154         }
 155
 156         tnode->fd = socket(sock_out.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 157         if (tnode->fd == -1) {
 158                 DEBUG(DEBUG_ERR, (__location__ "Failed to create socket\n"));
 159                 return;
 160         }
 161         set_nonblocking(tnode->fd);
 162         set_close_on_exec(tnode->fd);
 163
 164         DEBUG(DEBUG_DEBUG, (__location__ " Created TCP SOCKET FD:%d\n", tnode->fd));
 165
 166         /* Bind our side of the socketpair to the same address we use to listen
 167          * on incoming CTDB traffic.
 168          * We must specify this address to make sure that the address we expose to
 169          * the remote side is actually routable in case CTDB traffic will run on
 170          * a dedicated non-routeable network.
 171          */
 172         ZERO_STRUCT(sock_in);
 173         if (ctdb_tcp_get_address(ctdb, ctdb->address.address, &sock_in) != 0) {
 174                 DEBUG(DEBUG_ERR, (__location__ " Failed to find our address. Failing bind.\n"));
 175                 close(tnode->fd);
 176                 return;
 177         }
 178
 179         /* AIX libs check to see if the socket address and length
 180            arguments are consistent with each other on calls like
 181            connect().   Can not get by with just sizeof(sock_in),
 182            need sizeof(sock_in.ip).
 183         */
 184         switch (sock_in.sa.sa_family) {
 185         case AF_INET:
 186                 sockin_size = sizeof(sock_in.ip);
 187                 sockout_size = sizeof(sock_out.ip);
 188                 break;
 189         case AF_INET6:
 190                 sockin_size = sizeof(sock_in.ip6);
 191                 sockout_size = sizeof(sock_out.ip6);
 192                 break;
 193         default:
 194                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 195                         sock_in.sa.sa_family));
 196                 close(tnode->fd);
 197                 return;
 198         }
 199 #ifdef HAVE_SOCK_SIN_LEN
 200         sock_in.ip.sin_len = sockin_size;
 201         sock_out.ip.sin_len = sockout_size;
 202 #endif
 203         if (bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size) == -1) {
 204                 DEBUG(DEBUG_ERR, (__location__ "Failed to bind socket %s(%d)\n",
 205                                   strerror(errno), errno));
 206                 close(tnode->fd);
 207                 return;
 208         }
 209
 210         if (connect(tnode->fd, (struct sockaddr *)&sock_out, sockout_size) != 0 &&
 211             errno != EINPROGRESS) {
 212                 ctdb_tcp_stop_connection(node);
 213                 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
 214                                                     timeval_current_ofs(1, 0),
 215                                                     ctdb_tcp_node_connect, node);
 216                 return;
 217         }
 218
 219         /* non-blocking connect - wait for write event */
 220         tnode->connect_fde = event_add_fd(node->ctdb->ev, tnode, tnode->fd,
 221                                           EVENT_FD_WRITE|EVENT_FD_READ,
 222                                           ctdb_node_connect_write, node);
 223
 224         /* don't give it long to connect - retry in one second. This ensures
 225            that we find a node is up quickly (tcp normally backs off a syn reply
 226            delay by quite a lot) */
 227         tnode->connect_te = event_add_timed(ctdb->ev, tnode, timeval_current_ofs(1, 0),
 228                                             ctdb_tcp_node_connect, node);
 229 }
 230
 231 /*
 232   called when we get contacted by another node
 233   currently makes no attempt to check if the connection is really from a ctdb
 234   node in our cluster
 235 */
 236 static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
 237                               uint16_t flags, void *private_data)
 238 {
 239         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 240         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data, struct ctdb_tcp);
 241         ctdb_sock_addr addr;
 242         socklen_t len;
 243         int fd, nodeid;
 244         struct ctdb_incoming *in;
 245         int one = 1;
 246         const char *incoming_node;
 247
 248         memset(&addr, 0, sizeof(addr));
 249         len = sizeof(addr);
 250         fd = accept(ctcp->listen_fd, (struct sockaddr *)&addr, &len);
 251         if (fd == -1) return;
 252
 253         incoming_node = ctdb_addr_to_str(&addr);
 254         nodeid = ctdb_ip_to_nodeid(ctdb, incoming_node);
 255
 256         if (nodeid == -1) {
 257                 DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", incoming_node));
 258                 close(fd);
 259                 return;
 260         }
 261
 262         in = talloc_zero(ctcp, struct ctdb_incoming);
 263         in->fd = fd;
 264         in->ctdb = ctdb;
 265
 266         set_nonblocking(in->fd);
 267         set_close_on_exec(in->fd);
 268
 269         DEBUG(DEBUG_DEBUG, (__location__ " Created SOCKET FD:%d to incoming ctdb connection\n", fd));
 270
 271         setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
 272
 273         in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT,
 274                                      ctdb_tcp_read_cb, in, "ctdbd-%s", incoming_node);
 275 }
 276
 277
 278 /*
 279   automatically find which address to listen on
 280 */
 281 static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
 282 {
 283         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 284                                                 struct ctdb_tcp);
 285         ctdb_sock_addr sock;
 286         int lock_fd, i;
 287         const char *lock_path = "/tmp/.ctdb_socket_lock";
 288         struct flock lock;
 289         int one = 1;
 290         int sock_size;
 291         struct tevent_fd *fde;
 292
 293         /* If there are no nodes, then it won't be possible to find
 294          * the first one.  Log a failure and short circuit the whole
 295          * process.
 296          */
 297         if (ctdb->num_nodes == 0) {
 298                 DEBUG(DEBUG_CRIT,("No nodes available to attempt bind to - is the nodes file empty?\n"));
 299                 return -1;
 300         }
 301
 302         /* in order to ensure that we don't get two nodes with the
 303            same adddress, we must make the bind() and listen() calls
 304            atomic. The SO_REUSEADDR setsockopt only prevents double
 305            binds if the first socket is in LISTEN state  */
 306         lock_fd = open(lock_path, O_RDWR|O_CREAT, 0666);
 307         if (lock_fd == -1) {
 308                 DEBUG(DEBUG_CRIT,("Unable to open %s\n", lock_path));
 309                 return -1;
 310         }
 311
 312         lock.l_type = F_WRLCK;
 313         lock.l_whence = SEEK_SET;
 314         lock.l_start = 0;
 315         lock.l_len = 1;
 316         lock.l_pid = 0;
 317
 318         if (fcntl(lock_fd, F_SETLKW, &lock) != 0) {
 319                 DEBUG(DEBUG_CRIT,("Unable to lock %s\n", lock_path));
 320                 close(lock_fd);
 321                 return -1;
 322         }
 323
 324         for (i=0; i < ctdb->num_nodes; i++) {
 325                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 326                         continue;
 327                 }
 328                 ZERO_STRUCT(sock);
 329                 if (ctdb_tcp_get_address(ctdb,
 330                                 ctdb->nodes[i]->address.address,
 331                                 &sock) != 0) {
 332                         continue;
 333                 }
 334
 335                 switch (sock.sa.sa_family) {
 336                 case AF_INET:
 337                         sock.ip.sin_port = htons(ctdb->nodes[i]->address.port);
 338                         sock_size = sizeof(sock.ip);
 339                         break;
 340                 case AF_INET6:
 341                         sock.ip6.sin6_port = htons(ctdb->nodes[i]->address.port);
 342                         sock_size = sizeof(sock.ip6);
 343                         break;
 344                 default:
 345                         DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 346                                 sock.sa.sa_family));
 347                         continue;
 348                 }
 349 #ifdef HAVE_SOCK_SIN_LEN
 350                 sock.ip.sin_len = sock_size;
 351 #endif
 352
 353                 ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 354                 if (ctcp->listen_fd == -1) {
 355                         ctdb_set_error(ctdb, "socket failed\n");
 356                         continue;
 357                 }
 358
 359                 set_close_on_exec(ctcp->listen_fd);
 360
 361                 setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
 362
 363                 if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) == 0) {
 364                         break;
 365                 }
 366
 367                 if (errno == EADDRNOTAVAIL) {
 368                         DEBUG(DEBUG_DEBUG,(__location__ " Failed to bind() to socket. %s(%d)\n",
 369                                         strerror(errno), errno));
 370                 } else {
 371                         DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n",
 372                                         strerror(errno), errno));
 373                 }
 374         }
 375
 376         if (i == ctdb->num_nodes) {
 377                 DEBUG(DEBUG_CRIT,("Unable to bind to any of the node addresses - giving up\n"));
 378                 goto failed;
 379         }
 380         ctdb->address.address = talloc_strdup(ctdb, ctdb->nodes[i]->address.address);
 381         ctdb->address.port    = ctdb->nodes[i]->address.port;
 382         ctdb->name = talloc_asprintf(ctdb, "%s:%u",
 383                                      ctdb->address.address,
 384                                      ctdb->address.port);
 385         ctdb->pnn = ctdb->nodes[i]->pnn;
 386         DEBUG(DEBUG_INFO,("ctdb chose network address %s:%u pnn %u\n",
 387                  ctdb->address.address,
 388                  ctdb->address.port,
 389                  ctdb->pnn));
 390
 391         if (listen(ctcp->listen_fd, 10) == -1) {
 392                 goto failed;
 393         }
 394
 395         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 396                            ctdb_listen_event, ctdb);
 397         tevent_fd_set_auto_close(fde);
 398
 399         close(lock_fd);
 400
 401         return 0;
 402
 403 failed:
 404         close(lock_fd);
 405         close(ctcp->listen_fd);
 406         ctcp->listen_fd = -1;
 407         return -1;
 408 }
 409
 410
 411 /*
 412   listen on our own address
 413 */
 414 int ctdb_tcp_listen(struct ctdb_context *ctdb)
 415 {
 416         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 417                                                 struct ctdb_tcp);
 418         ctdb_sock_addr sock;
 419         int sock_size;
 420         int one = 1;
 421         struct tevent_fd *fde;
 422
 423         /* we can either auto-bind to the first available address, or we can
 424            use a specified address */
 425         if (!ctdb->address.address) {
 426                 return ctdb_tcp_listen_automatic(ctdb);
 427         }
 428
 429         ZERO_STRUCT(sock);
 430         if (ctdb_tcp_get_address(ctdb, ctdb->address.address,
 431                                  &sock) != 0) {
 432                 goto failed;
 433         }
 434
 435         switch (sock.sa.sa_family) {
 436         case AF_INET:
 437                 sock.ip.sin_port = htons(ctdb->address.port);
 438                 sock_size = sizeof(sock.ip);
 439                 break;
 440         case AF_INET6:
 441                 sock.ip6.sin6_port = htons(ctdb->address.port);
 442                 sock_size = sizeof(sock.ip6);
 443                 break;
 444         default:
 445                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 446                         sock.sa.sa_family));
 447                 goto failed;
 448         }
 449 #ifdef HAVE_SOCK_SIN_LEN
 450         sock.ip.sin_len = sock_size;
 451 #endif
 452
 453         ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 454         if (ctcp->listen_fd == -1) {
 455                 ctdb_set_error(ctdb, "socket failed\n");
 456                 return -1;
 457         }
 458
 459         set_close_on_exec(ctcp->listen_fd);
 460
 461         setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
 462
 463         if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) != 0) {
 464                 DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
 465                 goto failed;
 466         }
 467
 468         if (listen(ctcp->listen_fd, 10) == -1) {
 469                 goto failed;
 470         }
 471
 472         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 473                      ctdb_listen_event, ctdb);
 474         tevent_fd_set_auto_close(fde);
 475
 476         return 0;
 477
 478 failed:
 479         if (ctcp->listen_fd != -1) {
 480                 close(ctcp->listen_fd);
 481         }
 482         ctcp->listen_fd = -1;
 483         return -1;
 484 }
 485