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         set_nonblocking(tnode->fd);
 158         set_close_on_exec(tnode->fd);
 159
 160         DEBUG(DEBUG_DEBUG, (__location__ " Created TCP SOCKET FD:%d\n", tnode->fd));
 161
 162         /* Bind our side of the socketpair to the same address we use to listen
 163          * on incoming CTDB traffic.
 164          * We must specify this address to make sure that the address we expose to
 165          * the remote side is actually routable in case CTDB traffic will run on
 166          * a dedicated non-routeable network.
 167          */
 168         ZERO_STRUCT(sock_in);
 169         if (ctdb_tcp_get_address(ctdb, ctdb->address.address, &sock_in) != 0) {
 170                 DEBUG(DEBUG_ERR, (__location__ " Failed to find our address. Failing bind.\n"));
 171                 close(tnode->fd);
 172                 return;
 173         }
 174
 175         /* AIX libs check to see if the socket address and length
 176            arguments are consistent with each other on calls like
 177            connect().   Can not get by with just sizeof(sock_in),
 178            need sizeof(sock_in.ip).
 179         */
 180         switch (sock_in.sa.sa_family) {
 181         case AF_INET:
 182                 sockin_size = sizeof(sock_in.ip);
 183                 sockout_size = sizeof(sock_out.ip);
 184                 break;
 185         case AF_INET6:
 186                 sockin_size = sizeof(sock_in.ip6);
 187                 sockout_size = sizeof(sock_out.ip6);
 188                 break;
 189         default:
 190                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 191                         sock_in.sa.sa_family));
 192                 close(tnode->fd);
 193                 return;
 194         }
 195 #ifdef HAVE_SOCK_SIN_LEN
 196         sock_in.ip.sin_len = sockin_size;
 197         sock_out.ip.sin_len = sockout_size;
 198 #endif
 199         bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size);
 200
 201         if (connect(tnode->fd, (struct sockaddr *)&sock_out, sockout_size) != 0 &&
 202             errno != EINPROGRESS) {
 203                 ctdb_tcp_stop_connection(node);
 204                 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
 205                                                     timeval_current_ofs(1, 0),
 206                                                     ctdb_tcp_node_connect, node);
 207                 return;
 208         }
 209
 210         /* non-blocking connect - wait for write event */
 211         tnode->connect_fde = event_add_fd(node->ctdb->ev, tnode, tnode->fd,
 212                                           EVENT_FD_WRITE|EVENT_FD_READ,
 213                                           ctdb_node_connect_write, node);
 214
 215         /* don't give it long to connect - retry in one second. This ensures
 216            that we find a node is up quickly (tcp normally backs off a syn reply
 217            delay by quite a lot) */
 218         tnode->connect_te = event_add_timed(ctdb->ev, tnode, timeval_current_ofs(1, 0),
 219                                             ctdb_tcp_node_connect, node);
 220 }
 221
 222 /*
 223   called when we get contacted by another node
 224   currently makes no attempt to check if the connection is really from a ctdb
 225   node in our cluster
 226 */
 227 static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
 228                               uint16_t flags, void *private_data)
 229 {
 230         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 231         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data, struct ctdb_tcp);
 232         ctdb_sock_addr addr;
 233         socklen_t len;
 234         int fd, nodeid;
 235         struct ctdb_incoming *in;
 236         int one = 1;
 237         const char *incoming_node;
 238
 239         memset(&addr, 0, sizeof(addr));
 240         len = sizeof(addr);
 241         fd = accept(ctcp->listen_fd, (struct sockaddr *)&addr, &len);
 242         if (fd == -1) return;
 243
 244         incoming_node = ctdb_addr_to_str(&addr);
 245         nodeid = ctdb_ip_to_nodeid(ctdb, incoming_node);
 246
 247         if (nodeid == -1) {
 248                 DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", incoming_node));
 249                 close(fd);
 250                 return;
 251         }
 252
 253         in = talloc_zero(ctcp, struct ctdb_incoming);
 254         in->fd = fd;
 255         in->ctdb = ctdb;
 256
 257         set_nonblocking(in->fd);
 258         set_close_on_exec(in->fd);
 259
 260         DEBUG(DEBUG_DEBUG, (__location__ " Created SOCKET FD:%d to incoming ctdb connection\n", fd));
 261
 262         setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
 263
 264         in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT,
 265                                      ctdb_tcp_read_cb, in, "ctdbd-%s", incoming_node);
 266 }
 267
 268
 269 /*
 270   automatically find which address to listen on
 271 */
 272 static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
 273 {
 274         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 275                                                 struct ctdb_tcp);
 276         ctdb_sock_addr sock;
 277         int lock_fd, i;
 278         const char *lock_path = "/tmp/.ctdb_socket_lock";
 279         struct flock lock;
 280         int one = 1;
 281         int sock_size;
 282         struct tevent_fd *fde;
 283
 284         /* If there are no nodes, then it won't be possible to find
 285          * the first one.  Log a failure and short circuit the whole
 286          * process.
 287          */
 288         if (ctdb->num_nodes == 0) {
 289                 DEBUG(DEBUG_CRIT,("No nodes available to attempt bind to - is the nodes file empty?\n"));
 290                 return -1;
 291         }
 292
 293         /* in order to ensure that we don't get two nodes with the
 294            same adddress, we must make the bind() and listen() calls
 295            atomic. The SO_REUSEADDR setsockopt only prevents double
 296            binds if the first socket is in LISTEN state  */
 297         lock_fd = open(lock_path, O_RDWR|O_CREAT, 0666);
 298         if (lock_fd == -1) {
 299                 DEBUG(DEBUG_CRIT,("Unable to open %s\n", lock_path));
 300                 return -1;
 301         }
 302
 303         lock.l_type = F_WRLCK;
 304         lock.l_whence = SEEK_SET;
 305         lock.l_start = 0;
 306         lock.l_len = 1;
 307         lock.l_pid = 0;
 308
 309         if (fcntl(lock_fd, F_SETLKW, &lock) != 0) {
 310                 DEBUG(DEBUG_CRIT,("Unable to lock %s\n", lock_path));
 311                 close(lock_fd);
 312                 return -1;
 313         }
 314
 315         for (i=0; i < ctdb->num_nodes; i++) {
 316                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 317                         continue;
 318                 }
 319                 ZERO_STRUCT(sock);
 320                 if (ctdb_tcp_get_address(ctdb,
 321                                 ctdb->nodes[i]->address.address,
 322                                 &sock) != 0) {
 323                         continue;
 324                 }
 325
 326                 switch (sock.sa.sa_family) {
 327                 case AF_INET:
 328                         sock.ip.sin_port = htons(ctdb->nodes[i]->address.port);
 329                         sock_size = sizeof(sock.ip);
 330                         break;
 331                 case AF_INET6:
 332                         sock.ip6.sin6_port = htons(ctdb->nodes[i]->address.port);
 333                         sock_size = sizeof(sock.ip6);
 334                         break;
 335                 default:
 336                         DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 337                                 sock.sa.sa_family));
 338                         continue;
 339                 }
 340 #ifdef HAVE_SOCK_SIN_LEN
 341                 sock.ip.sin_len = sock_size;
 342 #endif
 343
 344                 ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 345                 if (ctcp->listen_fd == -1) {
 346                         ctdb_set_error(ctdb, "socket failed\n");
 347                         continue;
 348                 }
 349
 350                 set_close_on_exec(ctcp->listen_fd);
 351
 352                 setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
 353
 354                 if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) == 0) {
 355                         break;
 356                 }
 357
 358                 if (errno == EADDRNOTAVAIL) {
 359                         DEBUG(DEBUG_DEBUG,(__location__ " Failed to bind() to socket. %s(%d)\n",
 360                                         strerror(errno), errno));
 361                 } else {
 362                         DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n",
 363                                         strerror(errno), errno));
 364                 }
 365         }
 366
 367         if (i == ctdb->num_nodes) {
 368                 DEBUG(DEBUG_CRIT,("Unable to bind to any of the node addresses - giving up\n"));
 369                 goto failed;
 370         }
 371         ctdb->address.address = talloc_strdup(ctdb, ctdb->nodes[i]->address.address);
 372         ctdb->address.port    = ctdb->nodes[i]->address.port;
 373         ctdb->name = talloc_asprintf(ctdb, "%s:%u",
 374                                      ctdb->address.address,
 375                                      ctdb->address.port);
 376         ctdb->pnn = ctdb->nodes[i]->pnn;
 377         DEBUG(DEBUG_INFO,("ctdb chose network address %s:%u pnn %u\n",
 378                  ctdb->address.address,
 379                  ctdb->address.port,
 380                  ctdb->pnn));
 381
 382         if (listen(ctcp->listen_fd, 10) == -1) {
 383                 goto failed;
 384         }
 385
 386         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 387                            ctdb_listen_event, ctdb);
 388         tevent_fd_set_auto_close(fde);
 389
 390         close(lock_fd);
 391
 392         return 0;
 393
 394 failed:
 395         close(lock_fd);
 396         close(ctcp->listen_fd);
 397         ctcp->listen_fd = -1;
 398         return -1;
 399 }
 400
 401
 402 /*
 403   listen on our own address
 404 */
 405 int ctdb_tcp_listen(struct ctdb_context *ctdb)
 406 {
 407         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 408                                                 struct ctdb_tcp);
 409         ctdb_sock_addr sock;
 410         int sock_size;
 411         int one = 1;
 412         struct tevent_fd *fde;
 413
 414         /* we can either auto-bind to the first available address, or we can
 415            use a specified address */
 416         if (!ctdb->address.address) {
 417                 return ctdb_tcp_listen_automatic(ctdb);
 418         }
 419
 420         ZERO_STRUCT(sock);
 421         if (ctdb_tcp_get_address(ctdb, ctdb->address.address,
 422                                  &sock) != 0) {
 423                 goto failed;
 424         }
 425
 426         switch (sock.sa.sa_family) {
 427         case AF_INET:
 428                 sock.ip.sin_port = htons(ctdb->address.port);
 429                 sock_size = sizeof(sock.ip);
 430                 break;
 431         case AF_INET6:
 432                 sock.ip6.sin6_port = htons(ctdb->address.port);
 433                 sock_size = sizeof(sock.ip6);
 434                 break;
 435         default:
 436                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 437                         sock.sa.sa_family));
 438                 goto failed;
 439         }
 440 #ifdef HAVE_SOCK_SIN_LEN
 441         sock.ip.sin_len = sock_size;
 442 #endif
 443
 444         ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 445         if (ctcp->listen_fd == -1) {
 446                 ctdb_set_error(ctdb, "socket failed\n");
 447                 return -1;
 448         }
 449
 450         set_close_on_exec(ctcp->listen_fd);
 451
 452         setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
 453
 454         if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) != 0) {
 455                 DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
 456                 goto failed;
 457         }
 458
 459         if (listen(ctcp->listen_fd, 10) == -1) {
 460                 goto failed;
 461         }
 462
 463         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 464                      ctdb_listen_event, ctdb);
 465         tevent_fd_set_auto_close(fde);
 466
 467         return 0;
 468
 469 failed:
 470         if (ctcp->listen_fd != -1) {
 471                 close(ctcp->listen_fd);
 472         }
 473         ctcp->listen_fd = -1;
 474         return -1;
 475 }
 476