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         if (setsockopt(tnode->fd,IPPROTO_TCP,TCP_NODELAY,(char *)&one,sizeof(one)) == -1) {
  99                 DEBUG(DEBUG_WARNING, ("Failed to set TCP_NODELAY on fd - %s\n",
 100                                       strerror(errno)));
 101         }
 102         if (setsockopt(tnode->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one)) == -1) {
 103                 DEBUG(DEBUG_WARNING, ("Failed to set KEEPALIVE on fd - %s\n",
 104                                       strerror(errno)));
 105         }
 106
 107         ctdb_queue_set_fd(tnode->out_queue, tnode->fd);
 108
 109         /* the queue subsystem now owns this fd */
 110         tnode->fd = -1;
 111 }
 112
 113
 114 /*
 115   called when we should try and establish a tcp connection to a node
 116 */
 117 void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
 118                            struct timeval t, void *private_data)
 119 {
 120         struct ctdb_node *node = talloc_get_type(private_data,
 121                                                  struct ctdb_node);
 122         struct ctdb_tcp_node *tnode = talloc_get_type(node->private_data,
 123                                                       struct ctdb_tcp_node);
 124         struct ctdb_context *ctdb = node->ctdb;
 125         ctdb_sock_addr sock_in;
 126         int sockin_size;
 127         int sockout_size;
 128         ctdb_sock_addr sock_out;
 129
 130         ctdb_tcp_stop_connection(node);
 131
 132         sock_out = node->address;
 133
 134         tnode->fd = socket(sock_out.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 135         if (tnode->fd == -1) {
 136                 DEBUG(DEBUG_ERR, (__location__ "Failed to create socket\n"));
 137                 return;
 138         }
 139         set_nonblocking(tnode->fd);
 140         set_close_on_exec(tnode->fd);
 141
 142         DEBUG(DEBUG_DEBUG, (__location__ " Created TCP SOCKET FD:%d\n", tnode->fd));
 143
 144         /* Bind our side of the socketpair to the same address we use to listen
 145          * on incoming CTDB traffic.
 146          * We must specify this address to make sure that the address we expose to
 147          * the remote side is actually routable in case CTDB traffic will run on
 148          * a dedicated non-routeable network.
 149          */
 150         sock_in = *ctdb->address;
 151
 152         /* AIX libs check to see if the socket address and length
 153            arguments are consistent with each other on calls like
 154            connect().   Can not get by with just sizeof(sock_in),
 155            need sizeof(sock_in.ip).
 156         */
 157         switch (sock_in.sa.sa_family) {
 158         case AF_INET:
 159                 sock_in.ip.sin_port = 0 /* Any port */;
 160                 sockin_size = sizeof(sock_in.ip);
 161                 sockout_size = sizeof(sock_out.ip);
 162                 break;
 163         case AF_INET6:
 164                 sock_in.ip6.sin6_port = 0 /* Any port */;
 165                 sockin_size = sizeof(sock_in.ip6);
 166                 sockout_size = sizeof(sock_out.ip6);
 167                 break;
 168         default:
 169                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 170                         sock_in.sa.sa_family));
 171                 close(tnode->fd);
 172                 return;
 173         }
 174
 175         if (bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size) == -1) {
 176                 DEBUG(DEBUG_ERR, (__location__ "Failed to bind socket %s(%d)\n",
 177                                   strerror(errno), errno));
 178                 close(tnode->fd);
 179                 return;
 180         }
 181
 182         if (connect(tnode->fd, (struct sockaddr *)&sock_out, sockout_size) != 0 &&
 183             errno != EINPROGRESS) {
 184                 ctdb_tcp_stop_connection(node);
 185                 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
 186                                                     timeval_current_ofs(1, 0),
 187                                                     ctdb_tcp_node_connect, node);
 188                 return;
 189         }
 190
 191         /* non-blocking connect - wait for write event */
 192         tnode->connect_fde = event_add_fd(node->ctdb->ev, tnode, tnode->fd,
 193                                           EVENT_FD_WRITE|EVENT_FD_READ,
 194                                           ctdb_node_connect_write, node);
 195
 196         /* don't give it long to connect - retry in one second. This ensures
 197            that we find a node is up quickly (tcp normally backs off a syn reply
 198            delay by quite a lot) */
 199         tnode->connect_te = event_add_timed(ctdb->ev, tnode, timeval_current_ofs(1, 0),
 200                                             ctdb_tcp_node_connect, node);
 201 }
 202
 203 /*
 204   called when we get contacted by another node
 205   currently makes no attempt to check if the connection is really from a ctdb
 206   node in our cluster
 207 */
 208 static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
 209                               uint16_t flags, void *private_data)
 210 {
 211         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 212         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data, struct ctdb_tcp);
 213         ctdb_sock_addr addr;
 214         socklen_t len;
 215         int fd, nodeid;
 216         struct ctdb_incoming *in;
 217         int one = 1;
 218
 219         memset(&addr, 0, sizeof(addr));
 220         len = sizeof(addr);
 221         fd = accept(ctcp->listen_fd, (struct sockaddr *)&addr, &len);
 222         if (fd == -1) return;
 223
 224         nodeid = ctdb_ip_to_nodeid(ctdb, &addr);
 225
 226         if (nodeid == -1) {
 227                 DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", ctdb_addr_to_str(&addr)));
 228                 close(fd);
 229                 return;
 230         }
 231
 232         in = talloc_zero(ctcp, struct ctdb_incoming);
 233         in->fd = fd;
 234         in->ctdb = ctdb;
 235
 236         set_nonblocking(in->fd);
 237         set_close_on_exec(in->fd);
 238
 239         DEBUG(DEBUG_DEBUG, (__location__ " Created SOCKET FD:%d to incoming ctdb connection\n", fd));
 240
 241         if (setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one)) == -1) {
 242                 DEBUG(DEBUG_WARNING, ("Failed to set KEEPALIVE on fd - %s\n",
 243                                       strerror(errno)));
 244         }
 245
 246         in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT,
 247                                      ctdb_tcp_read_cb, in, "ctdbd-%s", ctdb_addr_to_str(&addr));
 248 }
 249
 250
 251 /*
 252   automatically find which address to listen on
 253 */
 254 static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
 255 {
 256         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 257                                                 struct ctdb_tcp);
 258         ctdb_sock_addr sock;
 259         int lock_fd, i;
 260         const char *lock_path = CTDB_RUNDIR "/.socket_lock";
 261         struct flock lock;
 262         int one = 1;
 263         int sock_size;
 264         struct tevent_fd *fde;
 265
 266         /* If there are no nodes, then it won't be possible to find
 267          * the first one.  Log a failure and short circuit the whole
 268          * process.
 269          */
 270         if (ctdb->num_nodes == 0) {
 271                 DEBUG(DEBUG_CRIT,("No nodes available to attempt bind to - is the nodes file empty?\n"));
 272                 return -1;
 273         }
 274
 275         /* in order to ensure that we don't get two nodes with the
 276            same adddress, we must make the bind() and listen() calls
 277            atomic. The SO_REUSEADDR setsockopt only prevents double
 278            binds if the first socket is in LISTEN state  */
 279         lock_fd = open(lock_path, O_RDWR|O_CREAT, 0666);
 280         if (lock_fd == -1) {
 281                 DEBUG(DEBUG_CRIT,("Unable to open %s\n", lock_path));
 282                 return -1;
 283         }
 284
 285         lock.l_type = F_WRLCK;
 286         lock.l_whence = SEEK_SET;
 287         lock.l_start = 0;
 288         lock.l_len = 1;
 289         lock.l_pid = 0;
 290
 291         if (fcntl(lock_fd, F_SETLKW, &lock) != 0) {
 292                 DEBUG(DEBUG_CRIT,("Unable to lock %s\n", lock_path));
 293                 close(lock_fd);
 294                 return -1;
 295         }
 296
 297         for (i=0; i < ctdb->num_nodes; i++) {
 298                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 299                         continue;
 300                 }
 301                 sock = ctdb->nodes[i]->address;
 302
 303                 switch (sock.sa.sa_family) {
 304                 case AF_INET:
 305                         sock_size = sizeof(sock.ip);
 306                         break;
 307                 case AF_INET6:
 308                         sock_size = sizeof(sock.ip6);
 309                         break;
 310                 default:
 311                         DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 312                                 sock.sa.sa_family));
 313                         continue;
 314                 }
 315
 316                 ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 317                 if (ctcp->listen_fd == -1) {
 318                         ctdb_set_error(ctdb, "socket failed\n");
 319                         continue;
 320                 }
 321
 322                 set_close_on_exec(ctcp->listen_fd);
 323
 324                 if (setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,
 325                                (char *)&one,sizeof(one)) == -1) {
 326                         DEBUG(DEBUG_WARNING, ("Failed to set REUSEADDR on fd - %s\n",
 327                                               strerror(errno)));
 328                 }
 329
 330                 if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) == 0) {
 331                         break;
 332                 }
 333
 334                 if (errno == EADDRNOTAVAIL) {
 335                         DEBUG(DEBUG_DEBUG,(__location__ " Failed to bind() to socket. %s(%d)\n",
 336                                         strerror(errno), errno));
 337                 } else {
 338                         DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n",
 339                                         strerror(errno), errno));
 340                 }
 341         }
 342
 343         if (i == ctdb->num_nodes) {
 344                 DEBUG(DEBUG_CRIT,("Unable to bind to any of the node addresses - giving up\n"));
 345                 goto failed;
 346         }
 347         ctdb->address = talloc_memdup(ctdb,
 348                                       &ctdb->nodes[i]->address,
 349                                       sizeof(ctdb_sock_addr));
 350         CTDB_NO_MEMORY(ctdb, ctdb->address);
 351         ctdb->name = talloc_asprintf(ctdb, "%s:%u",
 352                                      ctdb_addr_to_str(ctdb->address),
 353                                      ctdb_addr_to_port(ctdb->address));
 354         DEBUG(DEBUG_INFO,("ctdb chose network address %s\n", ctdb->name));
 355
 356         if (listen(ctcp->listen_fd, 10) == -1) {
 357                 goto failed;
 358         }
 359
 360         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 361                            ctdb_listen_event, ctdb);
 362         tevent_fd_set_auto_close(fde);
 363
 364         close(lock_fd);
 365
 366         return 0;
 367
 368 failed:
 369         close(lock_fd);
 370         if (ctcp->listen_fd != -1) {
 371                 close(ctcp->listen_fd);
 372                 ctcp->listen_fd = -1;
 373         }
 374         return -1;
 375 }
 376
 377
 378 /*
 379   listen on our own address
 380 */
 381 int ctdb_tcp_listen(struct ctdb_context *ctdb)
 382 {
 383         struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
 384                                                 struct ctdb_tcp);
 385         ctdb_sock_addr sock;
 386         int sock_size;
 387         int one = 1;
 388         struct tevent_fd *fde;
 389
 390         /* we can either auto-bind to the first available address, or we can
 391            use a specified address */
 392         if (!ctdb->address) {
 393                 return ctdb_tcp_listen_automatic(ctdb);
 394         }
 395
 396         sock = *ctdb->address;
 397
 398         switch (sock.sa.sa_family) {
 399         case AF_INET:
 400                 sock_size = sizeof(sock.ip);
 401                 break;
 402         case AF_INET6:
 403                 sock_size = sizeof(sock.ip6);
 404                 break;
 405         default:
 406                 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
 407                         sock.sa.sa_family));
 408                 goto failed;
 409         }
 410
 411         ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
 412         if (ctcp->listen_fd == -1) {
 413                 ctdb_set_error(ctdb, "socket failed\n");
 414                 return -1;
 415         }
 416
 417         set_close_on_exec(ctcp->listen_fd);
 418
 419         if (setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one)) == -1) {
 420                 DEBUG(DEBUG_WARNING, ("Failed to set REUSEADDR on fd - %s\n",
 421                                       strerror(errno)));
 422         }
 423
 424         if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) != 0) {
 425                 DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
 426                 goto failed;
 427         }
 428
 429         if (listen(ctcp->listen_fd, 10) == -1) {
 430                 goto failed;
 431         }
 432
 433         fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
 434                      ctdb_listen_event, ctdb);
 435         tevent_fd_set_auto_close(fde);
 436
 437         return 0;
 438
 439 failed:
 440         if (ctcp->listen_fd != -1) {
 441                 close(ctcp->listen_fd);
 442         }
 443         ctcp->listen_fd = -1;
 444         return -1;
 445 }
 446