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