ctdb/server/ctdb_server.c

   1 /*
   2    ctdb main protocol code
   3
   4    Copyright (C) Andrew Tridgell  2006
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, see <http://www.gnu.org/licenses/>.
  18 */
  19
  20 #include "includes.h"
  21 #include "tdb.h"
  22 #include "lib/util/dlinklist.h"
  23 #include "system/network.h"
  24 #include "system/filesys.h"
  25 #include "../include/ctdb_private.h"
  26
  27 /*
  28   choose the transport we will use
  29 */
  30 int ctdb_set_transport(struct ctdb_context *ctdb, const char *transport)
  31 {
  32         ctdb->transport = talloc_strdup(ctdb, transport);
  33         CTDB_NO_MEMORY(ctdb, ctdb->transport);
  34
  35         return 0;
  36 }
  37
  38 /*
  39   Check whether an ip is a valid node ip
  40   Returns the node id for this ip address or -1
  41 */
  42 int ctdb_ip_to_nodeid(struct ctdb_context *ctdb, const ctdb_sock_addr *nodeip)
  43 {
  44         int nodeid;
  45
  46         for (nodeid=0;nodeid<ctdb->num_nodes;nodeid++) {
  47                 if (ctdb->nodes[nodeid]->flags & NODE_FLAGS_DELETED) {
  48                         continue;
  49                 }
  50                 if (ctdb_same_ip(&ctdb->nodes[nodeid]->address, nodeip)) {
  51                         return nodeid;
  52                 }
  53         }
  54
  55         return -1;
  56 }
  57
  58 /*
  59   choose the recovery lock file
  60 */
  61 int ctdb_set_recovery_lock_file(struct ctdb_context *ctdb, const char *file)
  62 {
  63         if (ctdb->recovery_lock_file != NULL) {
  64                 talloc_free(ctdb->recovery_lock_file);
  65                 ctdb->recovery_lock_file = NULL;
  66         }
  67
  68         if (file == NULL) {
  69                 DEBUG(DEBUG_ALERT,("Recovery lock file set to \"\". Disabling recovery lock checking\n"));
  70                 return 0;
  71         }
  72
  73         ctdb->recovery_lock_file = talloc_strdup(ctdb, file);
  74         CTDB_NO_MEMORY(ctdb, ctdb->recovery_lock_file);
  75
  76         return 0;
  77 }
  78
  79 /* Load a nodes list file into a nodes array */
  80 static int convert_node_map_to_list(struct ctdb_context *ctdb,
  81                                     TALLOC_CTX *mem_ctx,
  82                                     struct ctdb_node_map *node_map,
  83                                     struct ctdb_node ***nodes,
  84                                     uint32_t *num_nodes)
  85 {
  86         int i;
  87
  88         *nodes = talloc_zero_array(mem_ctx,
  89                                         struct ctdb_node *, node_map->num);
  90         CTDB_NO_MEMORY(ctdb, *nodes);
  91         *num_nodes = node_map->num;
  92
  93         for (i = 0; i < node_map->num; i++) {
  94                 struct ctdb_node *node;
  95
  96                 node = talloc_zero(*nodes, struct ctdb_node);
  97                 CTDB_NO_MEMORY(ctdb, node);
  98                 (*nodes)[i] = node;
  99
 100                 node->address = node_map->nodes[i].addr;
 101                 node->name = talloc_asprintf(node, "%s:%u",
 102                                              ctdb_addr_to_str(&node->address),
 103                                              ctdb_addr_to_port(&node->address));
 104
 105                 node->flags = node_map->nodes[i].flags;
 106                 if (!(node->flags & NODE_FLAGS_DELETED)) {
 107                         node->flags = NODE_FLAGS_UNHEALTHY;
 108                 }
 109                 node->flags |= NODE_FLAGS_DISCONNECTED;
 110
 111                 node->pnn = i;
 112                 node->ctdb = ctdb;
 113                 node->dead_count = 0;
 114         }
 115
 116         return 0;
 117 }
 118
 119 /* Load the nodes list from a file */
 120 void ctdb_load_nodes_file(struct ctdb_context *ctdb)
 121 {
 122         struct ctdb_node_map *node_map;
 123         int ret;
 124
 125         node_map = ctdb_read_nodes_file(ctdb, ctdb->nodes_file);
 126         if (node_map == NULL) {
 127                 goto fail;
 128         }
 129
 130         TALLOC_FREE(ctdb->nodes);
 131         ret = convert_node_map_to_list(ctdb, ctdb, node_map,
 132                                        &ctdb->nodes, &ctdb->num_nodes);
 133         if (ret == -1) {
 134                 goto fail;
 135         }
 136
 137         talloc_free(node_map);
 138         return;
 139
 140 fail:
 141         DEBUG(DEBUG_ERR, ("Failed to load nodes file \"%s\"\n",
 142                           ctdb->nodes_file));
 143         talloc_free(node_map);
 144         exit(1);
 145 }
 146
 147 /*
 148   setup the local node address
 149 */
 150 int ctdb_set_address(struct ctdb_context *ctdb, const char *address)
 151 {
 152         ctdb->address = talloc(ctdb, ctdb_sock_addr);
 153         CTDB_NO_MEMORY(ctdb, ctdb->address);
 154
 155         if (ctdb_parse_address(ctdb, address, ctdb->address) != 0) {
 156                 return -1;
 157         }
 158
 159         ctdb->name = talloc_asprintf(ctdb, "%s:%u",
 160                                      ctdb_addr_to_str(ctdb->address),
 161                                      ctdb_addr_to_port(ctdb->address));
 162         return 0;
 163 }
 164
 165
 166 /*
 167   return the number of active nodes
 168 */
 169 uint32_t ctdb_get_num_active_nodes(struct ctdb_context *ctdb)
 170 {
 171         int i;
 172         uint32_t count=0;
 173         for (i=0; i < ctdb->num_nodes; i++) {
 174                 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_INACTIVE)) {
 175                         count++;
 176                 }
 177         }
 178         return count;
 179 }
 180
 181
 182 /*
 183   called when we need to process a packet. This can be a requeued packet
 184   after a lockwait, or a real packet from another node
 185 */
 186 void ctdb_input_pkt(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 187 {
 188         TALLOC_CTX *tmp_ctx;
 189
 190         /* place the packet as a child of the tmp_ctx. We then use
 191            talloc_free() below to free it. If any of the calls want
 192            to keep it, then they will steal it somewhere else, and the
 193            talloc_free() will only free the tmp_ctx */
 194         tmp_ctx = talloc_new(ctdb);
 195         talloc_steal(tmp_ctx, hdr);
 196
 197         DEBUG(DEBUG_DEBUG,(__location__ " ctdb request %u of type %u length %u from "
 198                  "node %u to %u\n", hdr->reqid, hdr->operation, hdr->length,
 199                  hdr->srcnode, hdr->destnode));
 200
 201         switch (hdr->operation) {
 202         case CTDB_REQ_CALL:
 203         case CTDB_REPLY_CALL:
 204         case CTDB_REQ_DMASTER:
 205         case CTDB_REPLY_DMASTER:
 206                 /* we dont allow these calls when banned */
 207                 if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_BANNED) {
 208                         DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
 209                                 " request %u"
 210                                 " length %u from node %u to %u while node"
 211                                 " is banned\n",
 212                                  hdr->operation, hdr->reqid,
 213                                  hdr->length,
 214                                  hdr->srcnode, hdr->destnode));
 215                         goto done;
 216                 }
 217
 218                 /* for ctdb_call inter-node operations verify that the
 219                    remote node that sent us the call is running in the
 220                    same generation instance as this node
 221                 */
 222                 if (ctdb->vnn_map->generation != hdr->generation) {
 223                         DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
 224                                 " request %u"
 225                                 " length %u from node %u to %u had an"
 226                                 " invalid generation id:%u while our"
 227                                 " generation id is:%u\n",
 228                                  hdr->operation, hdr->reqid,
 229                                  hdr->length,
 230                                  hdr->srcnode, hdr->destnode,
 231                                  hdr->generation, ctdb->vnn_map->generation));
 232                         goto done;
 233                 }
 234         }
 235
 236         switch (hdr->operation) {
 237         case CTDB_REQ_CALL:
 238                 CTDB_INCREMENT_STAT(ctdb, node.req_call);
 239                 ctdb_request_call(ctdb, hdr);
 240                 break;
 241
 242         case CTDB_REPLY_CALL:
 243                 CTDB_INCREMENT_STAT(ctdb, node.reply_call);
 244                 ctdb_reply_call(ctdb, hdr);
 245                 break;
 246
 247         case CTDB_REPLY_ERROR:
 248                 CTDB_INCREMENT_STAT(ctdb, node.reply_error);
 249                 ctdb_reply_error(ctdb, hdr);
 250                 break;
 251
 252         case CTDB_REQ_DMASTER:
 253                 CTDB_INCREMENT_STAT(ctdb, node.req_dmaster);
 254                 ctdb_request_dmaster(ctdb, hdr);
 255                 break;
 256
 257         case CTDB_REPLY_DMASTER:
 258                 CTDB_INCREMENT_STAT(ctdb, node.reply_dmaster);
 259                 ctdb_reply_dmaster(ctdb, hdr);
 260                 break;
 261
 262         case CTDB_REQ_MESSAGE:
 263                 CTDB_INCREMENT_STAT(ctdb, node.req_message);
 264                 ctdb_request_message(ctdb, hdr);
 265                 break;
 266
 267         case CTDB_REQ_CONTROL:
 268                 CTDB_INCREMENT_STAT(ctdb, node.req_control);
 269                 ctdb_request_control(ctdb, hdr);
 270                 break;
 271
 272         case CTDB_REPLY_CONTROL:
 273                 CTDB_INCREMENT_STAT(ctdb, node.reply_control);
 274                 ctdb_reply_control(ctdb, hdr);
 275                 break;
 276
 277         case CTDB_REQ_KEEPALIVE:
 278                 CTDB_INCREMENT_STAT(ctdb, keepalive_packets_recv);
 279                 break;
 280
 281         default:
 282                 DEBUG(DEBUG_CRIT,("%s: Packet with unknown operation %u\n",
 283                          __location__, hdr->operation));
 284                 break;
 285         }
 286
 287 done:
 288         talloc_free(tmp_ctx);
 289 }
 290
 291
 292 /*
 293   called by the transport layer when a node is dead
 294 */
 295 void ctdb_node_dead(struct ctdb_node *node)
 296 {
 297         if (node->flags & NODE_FLAGS_DISCONNECTED) {
 298                 DEBUG(DEBUG_INFO,("%s: node %s is already marked disconnected: %u connected\n",
 299                          node->ctdb->name, node->name,
 300                          node->ctdb->num_connected));
 301                 return;
 302         }
 303         node->ctdb->num_connected--;
 304         node->flags |= NODE_FLAGS_DISCONNECTED | NODE_FLAGS_UNHEALTHY;
 305         node->rx_cnt = 0;
 306         node->dead_count = 0;
 307
 308         DEBUG(DEBUG_NOTICE,("%s: node %s is dead: %u connected\n",
 309                  node->ctdb->name, node->name, node->ctdb->num_connected));
 310         ctdb_daemon_cancel_controls(node->ctdb, node);
 311
 312         if (node->ctdb->methods == NULL) {
 313                 DEBUG(DEBUG_ERR,(__location__ " Can not restart transport while shutting down daemon.\n"));
 314                 return;
 315         }
 316
 317         node->ctdb->methods->restart(node);
 318 }
 319
 320 /*
 321   called by the transport layer when a node is connected
 322 */
 323 void ctdb_node_connected(struct ctdb_node *node)
 324 {
 325         if (!(node->flags & NODE_FLAGS_DISCONNECTED)) {
 326                 DEBUG(DEBUG_INFO,("%s: node %s is already marked connected: %u connected\n",
 327                          node->ctdb->name, node->name,
 328                          node->ctdb->num_connected));
 329                 return;
 330         }
 331         node->ctdb->num_connected++;
 332         node->dead_count = 0;
 333         node->flags &= ~NODE_FLAGS_DISCONNECTED;
 334         node->flags |= NODE_FLAGS_UNHEALTHY;
 335         DEBUG(DEBUG_NOTICE,
 336               ("%s: connected to %s - %u connected\n",
 337                node->ctdb->name, node->name, node->ctdb->num_connected));
 338 }
 339
 340 struct queue_next {
 341         struct ctdb_context *ctdb;
 342         struct ctdb_req_header *hdr;
 343 };
 344
 345
 346 /*
 347   triggered when a deferred packet is due
 348  */
 349 static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
 350                                struct timeval t, void *private_data)
 351 {
 352         struct queue_next *q = talloc_get_type(private_data, struct queue_next);
 353         ctdb_input_pkt(q->ctdb, q->hdr);
 354         talloc_free(q);
 355 }
 356
 357 /*
 358   defer a packet, so it is processed on the next event loop
 359   this is used for sending packets to ourselves
 360  */
 361 static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 362 {
 363         struct queue_next *q;
 364         q = talloc(ctdb, struct queue_next);
 365         if (q == NULL) {
 366                 DEBUG(DEBUG_ERR,(__location__ " Failed to allocate deferred packet\n"));
 367                 return;
 368         }
 369         q->ctdb = ctdb;
 370         q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
 371         if (q->hdr == NULL) {
 372                 DEBUG(DEBUG_ERR,("Error copying deferred packet to self\n"));
 373                 return;
 374         }
 375 #if 0
 376         /* use this to put packets directly into our recv function */
 377         ctdb_input_pkt(q->ctdb, q->hdr);
 378 #else
 379         event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
 380 #endif
 381 }
 382
 383
 384 /*
 385   broadcast a packet to all nodes
 386 */
 387 static void ctdb_broadcast_packet_all(struct ctdb_context *ctdb,
 388                                       struct ctdb_req_header *hdr)
 389 {
 390         int i;
 391         for (i=0; i < ctdb->num_nodes; i++) {
 392                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 393                         continue;
 394                 }
 395                 hdr->destnode = ctdb->nodes[i]->pnn;
 396                 ctdb_queue_packet(ctdb, hdr);
 397         }
 398 }
 399
 400 /*
 401   broadcast a packet to all nodes in the current vnnmap
 402 */
 403 static void ctdb_broadcast_packet_vnnmap(struct ctdb_context *ctdb,
 404                                          struct ctdb_req_header *hdr)
 405 {
 406         int i;
 407         for (i=0;i<ctdb->vnn_map->size;i++) {
 408                 hdr->destnode = ctdb->vnn_map->map[i];
 409                 ctdb_queue_packet(ctdb, hdr);
 410         }
 411 }
 412
 413 /*
 414   broadcast a packet to all connected nodes
 415 */
 416 static void ctdb_broadcast_packet_connected(struct ctdb_context *ctdb,
 417                                             struct ctdb_req_header *hdr)
 418 {
 419         int i;
 420         for (i=0; i < ctdb->num_nodes; i++) {
 421                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 422                         continue;
 423                 }
 424                 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_DISCONNECTED)) {
 425                         hdr->destnode = ctdb->nodes[i]->pnn;
 426                         ctdb_queue_packet(ctdb, hdr);
 427                 }
 428         }
 429 }
 430
 431 /*
 432   queue a packet or die
 433 */
 434 void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 435 {
 436         struct ctdb_node *node;
 437
 438         switch (hdr->destnode) {
 439         case CTDB_BROADCAST_ALL:
 440                 ctdb_broadcast_packet_all(ctdb, hdr);
 441                 return;
 442         case CTDB_BROADCAST_VNNMAP:
 443                 ctdb_broadcast_packet_vnnmap(ctdb, hdr);
 444                 return;
 445         case CTDB_BROADCAST_CONNECTED:
 446                 ctdb_broadcast_packet_connected(ctdb, hdr);
 447                 return;
 448         }
 449
 450         CTDB_INCREMENT_STAT(ctdb, node_packets_sent);
 451
 452         if (!ctdb_validate_pnn(ctdb, hdr->destnode)) {
 453                 DEBUG(DEBUG_CRIT,(__location__ " cant send to node %u that does not exist\n",
 454                          hdr->destnode));
 455                 return;
 456         }
 457
 458         node = ctdb->nodes[hdr->destnode];
 459
 460         if (node->flags & NODE_FLAGS_DELETED) {
 461                 DEBUG(DEBUG_ERR, (__location__ " Can not queue packet to DELETED node %d\n", hdr->destnode));
 462                 return;
 463         }
 464
 465         if (node->pnn == ctdb->pnn) {
 466                 ctdb_defer_packet(ctdb, hdr);
 467                 return;
 468         }
 469
 470         if (ctdb->methods == NULL) {
 471                 DEBUG(DEBUG_ALERT, (__location__ " Can not queue packet. "
 472                                     "Transport is DOWN\n"));
 473                 return;
 474         }
 475
 476         node->tx_cnt++;
 477         if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
 478                 ctdb_fatal(ctdb, "Unable to queue packet\n");
 479         }
 480 }
 481
 482
 483
 484
 485 /*
 486   a valgrind hack to allow us to get opcode specific backtraces
 487   very ugly, and relies on no compiler optimisation!
 488 */
 489 void ctdb_queue_packet_opcode(struct ctdb_context *ctdb, struct ctdb_req_header *hdr, unsigned opcode)
 490 {
 491         switch (opcode) {
 492 #define DO_OP(x) case x: ctdb_queue_packet(ctdb, hdr); break
 493                 DO_OP(1);
 494                 DO_OP(2);
 495                 DO_OP(3);
 496                 DO_OP(4);
 497                 DO_OP(5);
 498                 DO_OP(6);
 499                 DO_OP(7);
 500                 DO_OP(8);
 501                 DO_OP(9);
 502                 DO_OP(10);
 503                 DO_OP(11);
 504                 DO_OP(12);
 505                 DO_OP(13);
 506                 DO_OP(14);
 507                 DO_OP(15);
 508                 DO_OP(16);
 509                 DO_OP(17);
 510                 DO_OP(18);
 511                 DO_OP(19);
 512                 DO_OP(20);
 513                 DO_OP(21);
 514                 DO_OP(22);
 515                 DO_OP(23);
 516                 DO_OP(24);
 517                 DO_OP(25);
 518                 DO_OP(26);
 519                 DO_OP(27);
 520                 DO_OP(28);
 521                 DO_OP(29);
 522                 DO_OP(30);
 523                 DO_OP(31);
 524                 DO_OP(32);
 525                 DO_OP(33);
 526                 DO_OP(34);
 527                 DO_OP(35);
 528                 DO_OP(36);
 529                 DO_OP(37);
 530                 DO_OP(38);
 531                 DO_OP(39);
 532                 DO_OP(40);
 533                 DO_OP(41);
 534                 DO_OP(42);
 535                 DO_OP(43);
 536                 DO_OP(44);
 537                 DO_OP(45);
 538                 DO_OP(46);
 539                 DO_OP(47);
 540                 DO_OP(48);
 541                 DO_OP(49);
 542                 DO_OP(50);
 543                 DO_OP(51);
 544                 DO_OP(52);
 545                 DO_OP(53);
 546                 DO_OP(54);
 547                 DO_OP(55);
 548                 DO_OP(56);
 549                 DO_OP(57);
 550                 DO_OP(58);
 551                 DO_OP(59);
 552                 DO_OP(60);
 553                 DO_OP(61);
 554                 DO_OP(62);
 555                 DO_OP(63);
 556                 DO_OP(64);
 557                 DO_OP(65);
 558                 DO_OP(66);
 559                 DO_OP(67);
 560                 DO_OP(68);
 561                 DO_OP(69);
 562                 DO_OP(70);
 563                 DO_OP(71);
 564                 DO_OP(72);
 565                 DO_OP(73);
 566                 DO_OP(74);
 567                 DO_OP(75);
 568                 DO_OP(76);
 569                 DO_OP(77);
 570                 DO_OP(78);
 571                 DO_OP(79);
 572                 DO_OP(80);
 573                 DO_OP(81);
 574                 DO_OP(82);
 575                 DO_OP(83);
 576                 DO_OP(84);
 577                 DO_OP(85);
 578                 DO_OP(86);
 579                 DO_OP(87);
 580                 DO_OP(88);
 581                 DO_OP(89);
 582                 DO_OP(90);
 583                 DO_OP(91);
 584                 DO_OP(92);
 585                 DO_OP(93);
 586                 DO_OP(94);
 587                 DO_OP(95);
 588                 DO_OP(96);
 589                 DO_OP(97);
 590                 DO_OP(98);
 591                 DO_OP(99);
 592                 DO_OP(100);
 593         default:
 594                 ctdb_queue_packet(ctdb, hdr);
 595                 break;
 596         }
 597 }