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