ctdb/server/ctdb_recover.c

   1 /*
   2    ctdb recovery code
   3
   4    Copyright (C) Andrew Tridgell  2007
   5    Copyright (C) Ronnie Sahlberg  2007
   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 #include "replace.h"
  21 #include "system/time.h"
  22 #include "system/network.h"
  23 #include "system/filesys.h"
  24 #include "system/wait.h"
  25
  26 #include <talloc.h>
  27 #include <tevent.h>
  28 #include <tdb.h>
  29
  30 #include "lib/tdb_wrap/tdb_wrap.h"
  31 #include "lib/util/dlinklist.h"
  32 #include "lib/util/debug.h"
  33 #include "lib/util/time.h"
  34 #include "lib/util/util_process.h"
  35
  36 #include "ctdb_private.h"
  37 #include "ctdb_client.h"
  38
  39 #include "common/system.h"
  40 #include "common/common.h"
  41 #include "common/logging.h"
  42
  43 #include "ctdb_cluster_mutex.h"
  44
  45 int
  46 ctdb_control_getvnnmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  47 {
  48         struct ctdb_vnn_map_wire *map;
  49         size_t len;
  50
  51         CHECK_CONTROL_DATA_SIZE(0);
  52
  53         len = offsetof(struct ctdb_vnn_map_wire, map) + sizeof(uint32_t)*ctdb->vnn_map->size;
  54         map = talloc_size(outdata, len);
  55         CTDB_NO_MEMORY(ctdb, map);
  56
  57         map->generation = ctdb->vnn_map->generation;
  58         map->size = ctdb->vnn_map->size;
  59         memcpy(map->map, ctdb->vnn_map->map, sizeof(uint32_t)*map->size);
  60
  61         outdata->dsize = len;
  62         outdata->dptr  = (uint8_t *)map;
  63
  64         return 0;
  65 }
  66
  67 int
  68 ctdb_control_setvnnmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  69 {
  70         struct ctdb_vnn_map_wire *map = (struct ctdb_vnn_map_wire *)indata.dptr;
  71
  72         if (ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
  73                 DEBUG(DEBUG_ERR, ("Attempt to set vnnmap when not in recovery\n"));
  74                 return -1;
  75         }
  76
  77         talloc_free(ctdb->vnn_map);
  78
  79         ctdb->vnn_map = talloc(ctdb, struct ctdb_vnn_map);
  80         CTDB_NO_MEMORY(ctdb, ctdb->vnn_map);
  81
  82         ctdb->vnn_map->generation = map->generation;
  83         ctdb->vnn_map->size       = map->size;
  84         ctdb->vnn_map->map = talloc_array(ctdb->vnn_map, uint32_t, map->size);
  85         CTDB_NO_MEMORY(ctdb, ctdb->vnn_map->map);
  86
  87         memcpy(ctdb->vnn_map->map, map->map, sizeof(uint32_t)*map->size);
  88
  89         return 0;
  90 }
  91
  92 int
  93 ctdb_control_getdbmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  94 {
  95         uint32_t i, len;
  96         struct ctdb_db_context *ctdb_db;
  97         struct ctdb_dbid_map_old *dbid_map;
  98
  99         CHECK_CONTROL_DATA_SIZE(0);
 100
 101         len = 0;
 102         for(ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next){
 103                 len++;
 104         }
 105
 106
 107         outdata->dsize = offsetof(struct ctdb_dbid_map_old, dbs) + sizeof(dbid_map->dbs[0])*len;
 108         outdata->dptr  = (unsigned char *)talloc_zero_size(outdata, outdata->dsize);
 109         if (!outdata->dptr) {
 110                 DEBUG(DEBUG_ALERT, (__location__ " Failed to allocate dbmap array\n"));
 111                 exit(1);
 112         }
 113
 114         dbid_map = (struct ctdb_dbid_map_old *)outdata->dptr;
 115         dbid_map->num = len;
 116         for (i=0,ctdb_db=ctdb->db_list;ctdb_db;i++,ctdb_db=ctdb_db->next){
 117                 dbid_map->dbs[i].db_id       = ctdb_db->db_id;
 118                 dbid_map->dbs[i].flags       = ctdb_db->db_flags;
 119         }
 120
 121         return 0;
 122 }
 123
 124 int
 125 ctdb_control_getnodemap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
 126 {
 127         CHECK_CONTROL_DATA_SIZE(0);
 128
 129         outdata->dptr  = (unsigned char *)ctdb_node_list_to_map(ctdb->nodes,
 130                                                                 ctdb->num_nodes,
 131                                                                 outdata);
 132         if (outdata->dptr == NULL) {
 133                 return -1;
 134         }
 135
 136         outdata->dsize = talloc_get_size(outdata->dptr);
 137
 138         return 0;
 139 }
 140
 141 /*
 142   reload the nodes file
 143 */
 144 int
 145 ctdb_control_reload_nodes_file(struct ctdb_context *ctdb, uint32_t opcode)
 146 {
 147         int i, num_nodes;
 148         TALLOC_CTX *tmp_ctx;
 149         struct ctdb_node **nodes;
 150
 151         tmp_ctx = talloc_new(ctdb);
 152
 153         /* steal the old nodes file for a while */
 154         talloc_steal(tmp_ctx, ctdb->nodes);
 155         nodes = ctdb->nodes;
 156         ctdb->nodes = NULL;
 157         num_nodes = ctdb->num_nodes;
 158         ctdb->num_nodes = 0;
 159
 160         /* load the new nodes file */
 161         ctdb_load_nodes_file(ctdb);
 162
 163         for (i=0; i<ctdb->num_nodes; i++) {
 164                 /* keep any identical pre-existing nodes and connections */
 165                 if ((i < num_nodes) && ctdb_same_address(&ctdb->nodes[i]->address, &nodes[i]->address)) {
 166                         talloc_free(ctdb->nodes[i]);
 167                         ctdb->nodes[i] = talloc_steal(ctdb->nodes, nodes[i]);
 168                         continue;
 169                 }
 170
 171                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 172                         continue;
 173                 }
 174
 175                 /* any new or different nodes must be added */
 176                 if (ctdb->methods->add_node(ctdb->nodes[i]) != 0) {
 177                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_node failed at %d\n", i));
 178                         ctdb_fatal(ctdb, "failed to add node. shutting down\n");
 179                 }
 180                 if (ctdb->methods->connect_node(ctdb->nodes[i]) != 0) {
 181                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_connect failed at %d\n", i));
 182                         ctdb_fatal(ctdb, "failed to connect to node. shutting down\n");
 183                 }
 184         }
 185
 186         /* tell the recovery daemon to reaload the nodes file too */
 187         ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELOAD_NODES, tdb_null);
 188
 189         talloc_free(tmp_ctx);
 190
 191         return 0;
 192 }
 193
 194 /*
 195    a traverse function for pulling all relevent records from pulldb
 196  */
 197 struct pulldb_data {
 198         struct ctdb_context *ctdb;
 199         struct ctdb_db_context *ctdb_db;
 200         struct ctdb_marshall_buffer *pulldata;
 201         uint32_t len;
 202         uint32_t allocated_len;
 203         bool failed;
 204 };
 205
 206 static int traverse_pulldb(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *p)
 207 {
 208         struct pulldb_data *params = (struct pulldb_data *)p;
 209         struct ctdb_rec_data_old *rec;
 210         struct ctdb_context *ctdb = params->ctdb;
 211         struct ctdb_db_context *ctdb_db = params->ctdb_db;
 212
 213         /* add the record to the blob */
 214         rec = ctdb_marshall_record(params->pulldata, 0, key, NULL, data);
 215         if (rec == NULL) {
 216                 params->failed = true;
 217                 return -1;
 218         }
 219         if (params->len + rec->length >= params->allocated_len) {
 220                 params->allocated_len = rec->length + params->len + ctdb->tunable.pulldb_preallocation_size;
 221                 params->pulldata = talloc_realloc_size(NULL, params->pulldata, params->allocated_len);
 222         }
 223         if (params->pulldata == NULL) {
 224                 DEBUG(DEBUG_CRIT,(__location__ " Failed to expand pulldb_data to %u\n", rec->length + params->len));
 225                 ctdb_fatal(params->ctdb, "failed to allocate memory for recovery. shutting down\n");
 226         }
 227         params->pulldata->count++;
 228         memcpy(params->len+(uint8_t *)params->pulldata, rec, rec->length);
 229         params->len += rec->length;
 230
 231         if (ctdb->tunable.db_record_size_warn != 0 && rec->length > ctdb->tunable.db_record_size_warn) {
 232                 DEBUG(DEBUG_ERR,("Data record in %s is big. Record size is %d bytes\n", ctdb_db->db_name, (int)rec->length));
 233         }
 234
 235         talloc_free(rec);
 236
 237         return 0;
 238 }
 239
 240 /*
 241   pull a bunch of records from a ltdb, filtering by lmaster
 242  */
 243 int32_t ctdb_control_pull_db(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
 244 {
 245         struct ctdb_pulldb *pull;
 246         struct ctdb_db_context *ctdb_db;
 247         struct pulldb_data params;
 248         struct ctdb_marshall_buffer *reply;
 249
 250         pull = (struct ctdb_pulldb *)indata.dptr;
 251
 252         ctdb_db = find_ctdb_db(ctdb, pull->db_id);
 253         if (!ctdb_db) {
 254                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", pull->db_id));
 255                 return -1;
 256         }
 257
 258         if (!ctdb_db_frozen(ctdb_db)) {
 259                 DEBUG(DEBUG_ERR,
 260                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 261                 return -1;
 262         }
 263
 264         reply = talloc_zero(outdata, struct ctdb_marshall_buffer);
 265         CTDB_NO_MEMORY(ctdb, reply);
 266
 267         reply->db_id = pull->db_id;
 268
 269         params.ctdb = ctdb;
 270         params.ctdb_db = ctdb_db;
 271         params.pulldata = reply;
 272         params.len = offsetof(struct ctdb_marshall_buffer, data);
 273         params.allocated_len = params.len;
 274         params.failed = false;
 275
 276         if (ctdb_db->unhealthy_reason) {
 277                 /* this is just a warning, as the tdb should be empty anyway */
 278                 DEBUG(DEBUG_WARNING,("db(%s) unhealty in ctdb_control_pull_db: %s\n",
 279                                      ctdb_db->db_name, ctdb_db->unhealthy_reason));
 280         }
 281
 282         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 283                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 284                 return -1;
 285         }
 286
 287         if (tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_pulldb, &params) == -1) {
 288                 DEBUG(DEBUG_ERR,(__location__ " Failed to get traverse db '%s'\n", ctdb_db->db_name));
 289                 ctdb_lockdb_unmark(ctdb_db);
 290                 talloc_free(params.pulldata);
 291                 return -1;
 292         }
 293
 294         ctdb_lockdb_unmark(ctdb_db);
 295
 296         outdata->dptr = (uint8_t *)params.pulldata;
 297         outdata->dsize = params.len;
 298
 299         if (ctdb->tunable.db_record_count_warn != 0 && params.pulldata->count > ctdb->tunable.db_record_count_warn) {
 300                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d records\n", ctdb_db->db_name, params.pulldata->count));
 301         }
 302         if (ctdb->tunable.db_size_warn != 0 && outdata->dsize > ctdb->tunable.db_size_warn) {
 303                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d bytes\n", ctdb_db->db_name, (int)outdata->dsize));
 304         }
 305
 306
 307         return 0;
 308 }
 309
 310 struct db_pull_state {
 311         struct ctdb_context *ctdb;
 312         struct ctdb_db_context *ctdb_db;
 313         struct ctdb_marshall_buffer *recs;
 314         uint32_t pnn;
 315         uint64_t srvid;
 316         uint32_t num_records;
 317 };
 318
 319 static int traverse_db_pull(struct tdb_context *tdb, TDB_DATA key,
 320                             TDB_DATA data, void *private_data)
 321 {
 322         struct db_pull_state *state = (struct db_pull_state *)private_data;
 323         struct ctdb_marshall_buffer *recs;
 324
 325         recs = ctdb_marshall_add(state->ctdb, state->recs,
 326                                  state->ctdb_db->db_id, 0, key, NULL, data);
 327         if (recs == NULL) {
 328                 TALLOC_FREE(state->recs);
 329                 return -1;
 330         }
 331         state->recs = recs;
 332
 333         if (talloc_get_size(state->recs) >=
 334                         state->ctdb->tunable.rec_buffer_size_limit) {
 335                 TDB_DATA buffer;
 336                 int ret;
 337
 338                 buffer = ctdb_marshall_finish(state->recs);
 339                 ret = ctdb_daemon_send_message(state->ctdb, state->pnn,
 340                                                state->srvid, buffer);
 341                 if (ret != 0) {
 342                         TALLOC_FREE(state->recs);
 343                         return -1;
 344                 }
 345
 346                 state->num_records += state->recs->count;
 347                 TALLOC_FREE(state->recs);
 348         }
 349
 350         return 0;
 351 }
 352
 353 int32_t ctdb_control_db_pull(struct ctdb_context *ctdb,
 354                              struct ctdb_req_control_old *c,
 355                              TDB_DATA indata, TDB_DATA *outdata)
 356 {
 357         struct ctdb_pulldb_ext *pulldb_ext;
 358         struct ctdb_db_context *ctdb_db;
 359         struct db_pull_state state;
 360         int ret;
 361
 362         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 363
 364         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 365         if (ctdb_db == NULL) {
 366                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n",
 367                                  pulldb_ext->db_id));
 368                 return -1;
 369         }
 370
 371         if (!ctdb_db_frozen(ctdb_db)) {
 372                 DEBUG(DEBUG_ERR,
 373                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 374                 return -1;
 375         }
 376
 377         if (ctdb_db->unhealthy_reason) {
 378                 /* this is just a warning, as the tdb should be empty anyway */
 379                 DEBUG(DEBUG_WARNING,
 380                       ("db(%s) unhealty in ctdb_control_db_pull: %s\n",
 381                        ctdb_db->db_name, ctdb_db->unhealthy_reason));
 382         }
 383
 384         state.ctdb = ctdb;
 385         state.ctdb_db = ctdb_db;
 386         state.recs = NULL;
 387         state.pnn = c->hdr.srcnode;
 388         state.srvid = pulldb_ext->srvid;
 389         state.num_records = 0;
 390
 391         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 392                 DEBUG(DEBUG_ERR,
 393                       (__location__ " Failed to get lock on entire db - failing\n"));
 394                 return -1;
 395         }
 396
 397         ret = tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_db_pull, &state);
 398         if (ret == -1) {
 399                 DEBUG(DEBUG_ERR,
 400                       (__location__ " Failed to get traverse db '%s'\n",
 401                        ctdb_db->db_name));
 402                 ctdb_lockdb_unmark(ctdb_db);
 403                 return -1;
 404         }
 405
 406         /* Last few records */
 407         if (state.recs != NULL) {
 408                 TDB_DATA buffer;
 409
 410                 buffer = ctdb_marshall_finish(state.recs);
 411                 ret = ctdb_daemon_send_message(state.ctdb, state.pnn,
 412                                                state.srvid, buffer);
 413                 if (ret != 0) {
 414                         TALLOC_FREE(state.recs);
 415                         ctdb_lockdb_unmark(ctdb_db);
 416                         return -1;
 417                 }
 418
 419                 state.num_records += state.recs->count;
 420                 TALLOC_FREE(state.recs);
 421         }
 422
 423         ctdb_lockdb_unmark(ctdb_db);
 424
 425         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 426         if (outdata->dptr == NULL) {
 427                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 428                 return -1;
 429         }
 430
 431         memcpy(outdata->dptr, (uint8_t *)&state.num_records, sizeof(uint32_t));
 432         outdata->dsize = sizeof(uint32_t);
 433
 434         return 0;
 435 }
 436
 437 /*
 438   push a bunch of records into a ltdb, filtering by rsn
 439  */
 440 int32_t ctdb_control_push_db(struct ctdb_context *ctdb, TDB_DATA indata)
 441 {
 442         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
 443         struct ctdb_db_context *ctdb_db;
 444         int i, ret;
 445         struct ctdb_rec_data_old *rec;
 446
 447         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
 448                 DEBUG(DEBUG_ERR,(__location__ " invalid data in pulldb reply\n"));
 449                 return -1;
 450         }
 451
 452         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
 453         if (!ctdb_db) {
 454                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
 455                 return -1;
 456         }
 457
 458         if (!ctdb_db_frozen(ctdb_db)) {
 459                 DEBUG(DEBUG_ERR,
 460                       ("rejecting ctdb_control_push_db when not frozen\n"));
 461                 return -1;
 462         }
 463
 464         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 465                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 466                 return -1;
 467         }
 468
 469         rec = (struct ctdb_rec_data_old *)&reply->data[0];
 470
 471         DEBUG(DEBUG_INFO,("starting push of %u records for dbid 0x%x\n",
 472                  reply->count, reply->db_id));
 473
 474         for (i=0;i<reply->count;i++) {
 475                 TDB_DATA key, data;
 476                 struct ctdb_ltdb_header *hdr;
 477
 478                 key.dptr = &rec->data[0];
 479                 key.dsize = rec->keylen;
 480                 data.dptr = &rec->data[key.dsize];
 481                 data.dsize = rec->datalen;
 482
 483                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 484                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 485                         goto failed;
 486                 }
 487                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 488                 /* strip off any read only record flags. All readonly records
 489                    are revoked implicitely by a recovery
 490                 */
 491                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 492
 493                 data.dptr += sizeof(*hdr);
 494                 data.dsize -= sizeof(*hdr);
 495
 496                 ret = ctdb_ltdb_store(ctdb_db, key, hdr, data);
 497                 if (ret != 0) {
 498                         DEBUG(DEBUG_CRIT, (__location__ " Unable to store record\n"));
 499                         goto failed;
 500                 }
 501
 502                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 503         }
 504
 505         DEBUG(DEBUG_DEBUG,("finished push of %u records for dbid 0x%x\n",
 506                  reply->count, reply->db_id));
 507
 508         if (ctdb_db_readonly(ctdb_db)) {
 509                 DEBUG(DEBUG_CRIT,("Clearing the tracking database for dbid 0x%x\n",
 510                                   ctdb_db->db_id));
 511                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 512                         DEBUG(DEBUG_ERR,("Failed to wipe tracking database for 0x%x. Dropping read-only delegation support\n", ctdb_db->db_id));
 513                         tdb_close(ctdb_db->rottdb);
 514                         ctdb_db->rottdb = NULL;
 515                         ctdb_db_reset_readonly(ctdb_db);
 516                 }
 517                 while (ctdb_db->revokechild_active != NULL) {
 518                         talloc_free(ctdb_db->revokechild_active);
 519                 }
 520         }
 521
 522         ctdb_lockdb_unmark(ctdb_db);
 523         return 0;
 524
 525 failed:
 526         ctdb_lockdb_unmark(ctdb_db);
 527         return -1;
 528 }
 529
 530 struct db_push_state {
 531         struct ctdb_context *ctdb;
 532         struct ctdb_db_context *ctdb_db;
 533         uint64_t srvid;
 534         uint32_t num_records;
 535         bool failed;
 536 };
 537
 538 static void db_push_msg_handler(uint64_t srvid, TDB_DATA indata,
 539                                 void *private_data)
 540 {
 541         struct db_push_state *state = talloc_get_type(
 542                 private_data, struct db_push_state);
 543         struct ctdb_marshall_buffer *recs;
 544         struct ctdb_rec_data_old *rec;
 545         int i, ret;
 546
 547         if (state->failed) {
 548                 return;
 549         }
 550
 551         recs = (struct ctdb_marshall_buffer *)indata.dptr;
 552         rec = (struct ctdb_rec_data_old *)&recs->data[0];
 553
 554         DEBUG(DEBUG_INFO, ("starting push of %u records for dbid 0x%x\n",
 555                            recs->count, recs->db_id));
 556
 557         for (i=0; i<recs->count; i++) {
 558                 TDB_DATA key, data;
 559                 struct ctdb_ltdb_header *hdr;
 560
 561                 key.dptr = &rec->data[0];
 562                 key.dsize = rec->keylen;
 563                 data.dptr = &rec->data[key.dsize];
 564                 data.dsize = rec->datalen;
 565
 566                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 567                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 568                         goto failed;
 569                 }
 570
 571                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 572                 /* Strip off any read only record flags.
 573                  * All readonly records are revoked implicitely by a recovery.
 574                  */
 575                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 576
 577                 data.dptr += sizeof(*hdr);
 578                 data.dsize -= sizeof(*hdr);
 579
 580                 ret = ctdb_ltdb_store(state->ctdb_db, key, hdr, data);
 581                 if (ret != 0) {
 582                         DEBUG(DEBUG_ERR,
 583                               (__location__ " Unable to store record\n"));
 584                         goto failed;
 585                 }
 586
 587                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 588         }
 589
 590         DEBUG(DEBUG_DEBUG, ("finished push of %u records for dbid 0x%x\n",
 591                             recs->count, recs->db_id));
 592
 593         state->num_records += recs->count;
 594         return;
 595
 596 failed:
 597         state->failed = true;
 598 }
 599
 600 int32_t ctdb_control_db_push_start(struct ctdb_context *ctdb, TDB_DATA indata)
 601 {
 602         struct ctdb_pulldb_ext *pulldb_ext;
 603         struct ctdb_db_context *ctdb_db;
 604         struct db_push_state *state;
 605         int ret;
 606
 607         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 608
 609         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 610         if (ctdb_db == NULL) {
 611                 DEBUG(DEBUG_ERR,
 612                       (__location__ " Unknown db 0x%08x\n", pulldb_ext->db_id));
 613                 return -1;
 614         }
 615
 616         if (!ctdb_db_frozen(ctdb_db)) {
 617                 DEBUG(DEBUG_ERR,
 618                       ("rejecting ctdb_control_db_push_start when not frozen\n"));
 619                 return -1;
 620         }
 621
 622         if (ctdb_db->push_started) {
 623                 DEBUG(DEBUG_WARNING,
 624                       (__location__ " DB push already started for %s\n",
 625                        ctdb_db->db_name));
 626
 627                 /* De-register old state */
 628                 state = (struct db_push_state *)ctdb_db->push_state;
 629                 if (state != NULL) {
 630                         srvid_deregister(ctdb->srv, state->srvid, state);
 631                         talloc_free(state);
 632                         ctdb_db->push_state = NULL;
 633                 }
 634         }
 635
 636         state = talloc_zero(ctdb_db, struct db_push_state);
 637         if (state == NULL) {
 638                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 639                 return -1;
 640         }
 641
 642         state->ctdb = ctdb;
 643         state->ctdb_db = ctdb_db;
 644         state->srvid = pulldb_ext->srvid;
 645         state->failed = false;
 646
 647         ret = srvid_register(ctdb->srv, state, state->srvid,
 648                              db_push_msg_handler, state);
 649         if (ret != 0) {
 650                 DEBUG(DEBUG_ERR,
 651                       (__location__ " Failed to register srvid for db push\n"));
 652                 talloc_free(state);
 653                 return -1;
 654         }
 655
 656         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 657                 DEBUG(DEBUG_ERR,
 658                       (__location__ " Failed to get lock on entire db - failing\n"));
 659                 srvid_deregister(ctdb->srv, state->srvid, state);
 660                 talloc_free(state);
 661                 return -1;
 662         }
 663
 664         ctdb_db->push_started = true;
 665         ctdb_db->push_state = state;
 666
 667         return 0;
 668 }
 669
 670 int32_t ctdb_control_db_push_confirm(struct ctdb_context *ctdb,
 671                                      TDB_DATA indata, TDB_DATA *outdata)
 672 {
 673         uint32_t db_id;
 674         struct ctdb_db_context *ctdb_db;
 675         struct db_push_state *state;
 676
 677         db_id = *(uint32_t *)indata.dptr;
 678
 679         ctdb_db = find_ctdb_db(ctdb, db_id);
 680         if (ctdb_db == NULL) {
 681                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", db_id));
 682                 return -1;
 683         }
 684
 685         if (!ctdb_db_frozen(ctdb_db)) {
 686                 DEBUG(DEBUG_ERR,
 687                       ("rejecting ctdb_control_db_push_confirm when not frozen\n"));
 688                 return -1;
 689         }
 690
 691         if (!ctdb_db->push_started) {
 692                 DEBUG(DEBUG_ERR, (__location__ " DB push not started\n"));
 693                 return -1;
 694         }
 695
 696         if (ctdb_db_readonly(ctdb_db)) {
 697                 DEBUG(DEBUG_ERR,
 698                       ("Clearing the tracking database for dbid 0x%x\n",
 699                        ctdb_db->db_id));
 700                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 701                         DEBUG(DEBUG_ERR,
 702                               ("Failed to wipe tracking database for 0x%x."
 703                                " Dropping read-only delegation support\n",
 704                                ctdb_db->db_id));
 705                         tdb_close(ctdb_db->rottdb);
 706                         ctdb_db->rottdb = NULL;
 707                         ctdb_db_reset_readonly(ctdb_db);
 708                 }
 709
 710                 while (ctdb_db->revokechild_active != NULL) {
 711                         talloc_free(ctdb_db->revokechild_active);
 712                 }
 713         }
 714
 715         ctdb_lockdb_unmark(ctdb_db);
 716
 717         state = (struct db_push_state *)ctdb_db->push_state;
 718         if (state == NULL) {
 719                 DEBUG(DEBUG_ERR, (__location__ " Missing push db state\n"));
 720                 return -1;
 721         }
 722
 723         srvid_deregister(ctdb->srv, state->srvid, state);
 724
 725         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 726         if (outdata->dptr == NULL) {
 727                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 728                 talloc_free(state);
 729                 ctdb_db->push_state = NULL;
 730                 return -1;
 731         }
 732
 733         memcpy(outdata->dptr, (uint8_t *)&state->num_records, sizeof(uint32_t));
 734         outdata->dsize = sizeof(uint32_t);
 735
 736         talloc_free(state);
 737         ctdb_db->push_started = false;
 738         ctdb_db->push_state = NULL;
 739
 740         return 0;
 741 }
 742
 743 struct set_recmode_state {
 744         struct ctdb_context *ctdb;
 745         struct ctdb_req_control_old *c;
 746 };
 747
 748 static void set_recmode_handler(char status,
 749                                 double latency,
 750                                 void *private_data)
 751 {
 752         struct set_recmode_state *state = talloc_get_type_abort(
 753                 private_data, struct set_recmode_state);
 754         int s = 0;
 755         const char *err = NULL;
 756
 757         switch (status) {
 758         case '0':
 759                 /* Mutex taken */
 760                 DEBUG(DEBUG_ERR,
 761                       ("ERROR: Daemon able to take recovery lock on \"%s\" during recovery\n",
 762                        state->ctdb->recovery_lock));
 763                 s = -1;
 764                 err = "Took recovery lock from daemon during recovery - probably a cluster filesystem lock coherence problem";
 765                 break;
 766
 767         case '1':
 768                 /* Contention */
 769                 DEBUG(DEBUG_DEBUG, (__location__ " Recovery lock check OK\n"));
 770                 state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 771                 ctdb_process_deferred_attach(state->ctdb);
 772
 773                 s = 0;
 774
 775                 CTDB_UPDATE_RECLOCK_LATENCY(state->ctdb, "daemon reclock",
 776                                             reclock.ctdbd, latency);
 777                 break;
 778
 779         case '2':
 780                 /* Timeout.  Consider this a success, not a failure,
 781                  * as we failed to set the recovery lock which is what
 782                  * we wanted.  This can be caused by the cluster
 783                  * filesystem being very slow to arbitrate locks
 784                  * immediately after a node failure. */
 785                 DEBUG(DEBUG_WARNING,
 786                       (__location__
 787                        "Time out getting recovery lock, allowing recmode set anyway\n"));
 788                 state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 789                 ctdb_process_deferred_attach(state->ctdb);
 790
 791                 s = 0;
 792                 break;
 793
 794         default:
 795                 DEBUG(DEBUG_ERR,
 796                       ("Unexpected error when testing recovery lock\n"));
 797                 s = -1;
 798                 err = "Unexpected error when testing recovery lock";
 799         }
 800
 801         ctdb_request_control_reply(state->ctdb, state->c, NULL, s, err);
 802         talloc_free(state);
 803 }
 804
 805 static void
 806 ctdb_drop_all_ips_event(struct tevent_context *ev, struct tevent_timer *te,
 807                         struct timeval t, void *private_data)
 808 {
 809         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 810
 811         DEBUG(DEBUG_ERR,(__location__ " Been in recovery mode for too long. Dropping all IPS\n"));
 812         talloc_free(ctdb->release_ips_ctx);
 813         ctdb->release_ips_ctx = NULL;
 814
 815         ctdb_release_all_ips(ctdb);
 816 }
 817
 818 /*
 819  * Set up an event to drop all public ips if we remain in recovery for too
 820  * long
 821  */
 822 int ctdb_deferred_drop_all_ips(struct ctdb_context *ctdb)
 823 {
 824         if (ctdb->release_ips_ctx != NULL) {
 825                 talloc_free(ctdb->release_ips_ctx);
 826         }
 827         ctdb->release_ips_ctx = talloc_new(ctdb);
 828         CTDB_NO_MEMORY(ctdb, ctdb->release_ips_ctx);
 829
 830         tevent_add_timer(ctdb->ev, ctdb->release_ips_ctx,
 831                          timeval_current_ofs(ctdb->tunable.recovery_drop_all_ips, 0),
 832                          ctdb_drop_all_ips_event, ctdb);
 833         return 0;
 834 }
 835
 836 /*
 837   set the recovery mode
 838  */
 839 int32_t ctdb_control_set_recmode(struct ctdb_context *ctdb,
 840                                  struct ctdb_req_control_old *c,
 841                                  TDB_DATA indata, bool *async_reply,
 842                                  const char **errormsg)
 843 {
 844         uint32_t recmode = *(uint32_t *)indata.dptr;
 845         struct ctdb_db_context *ctdb_db;
 846         struct set_recmode_state *state;
 847         struct ctdb_cluster_mutex_handle *h;
 848
 849         if (recmode == ctdb->recovery_mode) {
 850                 D_INFO("Recovery mode already set to %s\n",
 851                        recmode == CTDB_RECOVERY_NORMAL ? "NORMAL" : "ACTIVE");
 852                 return 0;
 853         }
 854
 855         D_NOTICE("Recovery mode set to %s\n",
 856                  recmode == CTDB_RECOVERY_NORMAL ? "NORMAL" : "ACTIVE");
 857
 858         /* if we enter recovery but stay in recovery for too long
 859            we will eventually drop all our ip addresses
 860         */
 861         if (recmode == CTDB_RECOVERY_ACTIVE) {
 862                 if (ctdb_deferred_drop_all_ips(ctdb) != 0) {
 863                         D_ERR("Failed to set up deferred drop all ips\n");
 864                 }
 865
 866                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 867                 return 0;
 868         }
 869
 870         /* From this point: recmode == CTDB_RECOVERY_NORMAL
 871          *
 872          * Therefore, what follows is special handling when setting
 873          * recovery mode back to normal */
 874
 875         TALLOC_FREE(ctdb->release_ips_ctx);
 876
 877         for (ctdb_db = ctdb->db_list; ctdb_db != NULL; ctdb_db = ctdb_db->next) {
 878                 if (ctdb_db->generation != ctdb->vnn_map->generation) {
 879                         DEBUG(DEBUG_ERR,
 880                               ("Inconsistent DB generation %u for %s\n",
 881                                ctdb_db->generation, ctdb_db->db_name));
 882                         DEBUG(DEBUG_ERR, ("Recovery mode set to ACTIVE\n"));
 883                         return -1;
 884                 }
 885         }
 886
 887         /* force the databases to thaw */
 888         if (ctdb_db_all_frozen(ctdb)) {
 889                 ctdb_control_thaw(ctdb, false);
 890         }
 891
 892         if (ctdb->recovery_lock == NULL) {
 893                 /* Not using recovery lock file */
 894                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 895                 ctdb_process_deferred_attach(ctdb);
 896                 return 0;
 897         }
 898
 899         state = talloc_zero(ctdb, struct set_recmode_state);
 900         if (state == NULL) {
 901                 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
 902                 return -1;
 903         }
 904         state->ctdb = ctdb;
 905         state->c = NULL;
 906
 907         h = ctdb_cluster_mutex(state, ctdb, ctdb->recovery_lock, 5,
 908                                set_recmode_handler, state, NULL, NULL);
 909         if (h == NULL) {
 910                 talloc_free(state);
 911                 return -1;
 912         }
 913
 914         state->c = talloc_steal(state, c);
 915         *async_reply = true;
 916
 917         return 0;
 918 }
 919
 920
 921 /*
 922   delete a record as part of the vacuum process
 923   only delete if we are not lmaster or dmaster, and our rsn is <= the provided rsn
 924   use non-blocking locks
 925
 926   return 0 if the record was successfully deleted (i.e. it does not exist
 927   when the function returns)
 928   or !0 is the record still exists in the tdb after returning.
 929  */
 930 static int delete_tdb_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, struct ctdb_rec_data_old *rec)
 931 {
 932         TDB_DATA key, data, data2;
 933         struct ctdb_ltdb_header *hdr, *hdr2;
 934
 935         /* these are really internal tdb functions - but we need them here for
 936            non-blocking lock of the freelist */
 937         int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
 938         int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
 939
 940
 941         key.dsize = rec->keylen;
 942         key.dptr  = &rec->data[0];
 943         data.dsize = rec->datalen;
 944         data.dptr = &rec->data[rec->keylen];
 945
 946         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
 947                 DEBUG(DEBUG_INFO,(__location__ " Called delete on record where we are lmaster\n"));
 948                 return -1;
 949         }
 950
 951         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
 952                 DEBUG(DEBUG_ERR,(__location__ " Bad record size\n"));
 953                 return -1;
 954         }
 955
 956         hdr = (struct ctdb_ltdb_header *)data.dptr;
 957
 958         /* use a non-blocking lock */
 959         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
 960                 return -1;
 961         }
 962
 963         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
 964         if (data2.dptr == NULL) {
 965                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 966                 return 0;
 967         }
 968
 969         if (data2.dsize < sizeof(struct ctdb_ltdb_header)) {
 970                 if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) == 0) {
 971                         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
 972                                 DEBUG(DEBUG_CRIT,(__location__ " Failed to delete corrupt record\n"));
 973                         }
 974                         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
 975                         DEBUG(DEBUG_CRIT,(__location__ " Deleted corrupt record\n"));
 976                 }
 977                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 978                 free(data2.dptr);
 979                 return 0;
 980         }
 981
 982         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
 983
 984         if (hdr2->rsn > hdr->rsn) {
 985                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 986                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with rsn=%llu - called with rsn=%llu\n",
 987                          (unsigned long long)hdr2->rsn, (unsigned long long)hdr->rsn));
 988                 free(data2.dptr);
 989                 return -1;
 990         }
 991
 992         /* do not allow deleting record that have readonly flags set. */
 993         if (hdr->flags & CTDB_REC_RO_FLAGS) {
 994                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 995                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
 996                 free(data2.dptr);
 997                 return -1;
 998         }
 999         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1000                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1001                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1002                 free(data2.dptr);
1003                 return -1;
1004         }
1005
1006         if (hdr2->dmaster == ctdb->pnn) {
1007                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1008                 DEBUG(DEBUG_INFO,(__location__ " Attempted delete record where we are the dmaster\n"));
1009                 free(data2.dptr);
1010                 return -1;
1011         }
1012
1013         if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) != 0) {
1014                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1015                 free(data2.dptr);
1016                 return -1;
1017         }
1018
1019         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1020                 tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1021                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1022                 DEBUG(DEBUG_INFO,(__location__ " Failed to delete record\n"));
1023                 free(data2.dptr);
1024                 return -1;
1025         }
1026
1027         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1028         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1029         free(data2.dptr);
1030         return 0;
1031 }
1032
1033
1034
1035 struct recovery_callback_state {
1036         struct ctdb_req_control_old *c;
1037 };
1038
1039
1040 /*
1041   called when the 'recovered' event script has finished
1042  */
1043 static void ctdb_end_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1044 {
1045         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1046
1047         ctdb_enable_monitoring(ctdb);
1048         CTDB_INCREMENT_STAT(ctdb, num_recoveries);
1049
1050         if (status != 0) {
1051                 DEBUG(DEBUG_ERR,(__location__ " recovered event script failed (status %d)\n", status));
1052                 if (status == -ETIME) {
1053                         ctdb_ban_self(ctdb);
1054                 }
1055         }
1056
1057         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1058         talloc_free(state);
1059
1060         gettimeofday(&ctdb->last_recovery_finished, NULL);
1061
1062         if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
1063                 ctdb_set_runstate(ctdb, CTDB_RUNSTATE_STARTUP);
1064         }
1065 }
1066
1067 /*
1068   recovery has finished
1069  */
1070 int32_t ctdb_control_end_recovery(struct ctdb_context *ctdb,
1071                                 struct ctdb_req_control_old *c,
1072                                 bool *async_reply)
1073 {
1074         int ret;
1075         struct recovery_callback_state *state;
1076
1077         DEBUG(DEBUG_NOTICE,("Recovery has finished\n"));
1078
1079         ctdb_persistent_finish_trans3_commits(ctdb);
1080
1081         state = talloc(ctdb, struct recovery_callback_state);
1082         CTDB_NO_MEMORY(ctdb, state);
1083
1084         state->c    = c;
1085
1086         ctdb_disable_monitoring(ctdb);
1087
1088         ret = ctdb_event_script_callback(ctdb, state,
1089                                          ctdb_end_recovery_callback,
1090                                          state,
1091                                          CTDB_EVENT_RECOVERED, "%s", "");
1092
1093         if (ret != 0) {
1094                 ctdb_enable_monitoring(ctdb);
1095
1096                 DEBUG(DEBUG_ERR,(__location__ " Failed to end recovery\n"));
1097                 talloc_free(state);
1098                 return -1;
1099         }
1100
1101         /* tell the control that we will be reply asynchronously */
1102         state->c    = talloc_steal(state, c);
1103         *async_reply = true;
1104         return 0;
1105 }
1106
1107 /*
1108   called when the 'startrecovery' event script has finished
1109  */
1110 static void ctdb_start_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1111 {
1112         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1113
1114         if (status != 0) {
1115                 DEBUG(DEBUG_ERR,(__location__ " startrecovery event script failed (status %d)\n", status));
1116         }
1117
1118         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1119         talloc_free(state);
1120 }
1121
1122 static void run_start_recovery_event(struct ctdb_context *ctdb,
1123                                      struct recovery_callback_state *state)
1124 {
1125         int ret;
1126
1127         ctdb_disable_monitoring(ctdb);
1128
1129         ret = ctdb_event_script_callback(ctdb, state,
1130                                          ctdb_start_recovery_callback,
1131                                          state,
1132                                          CTDB_EVENT_START_RECOVERY,
1133                                          "%s", "");
1134
1135         if (ret != 0) {
1136                 DEBUG(DEBUG_ERR,("Unable to run startrecovery event\n"));
1137                 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
1138                 talloc_free(state);
1139                 return;
1140         }
1141
1142         return;
1143 }
1144
1145 static bool reclock_strings_equal(const char *a, const char *b)
1146 {
1147         return (a == NULL && b == NULL) ||
1148                 (a != NULL && b != NULL && strcmp(a, b) == 0);
1149 }
1150
1151 static void start_recovery_reclock_callback(struct ctdb_context *ctdb,
1152                                                 int32_t status,
1153                                                 TDB_DATA data,
1154                                                 const char *errormsg,
1155                                                 void *private_data)
1156 {
1157         struct recovery_callback_state *state = talloc_get_type_abort(
1158                 private_data, struct recovery_callback_state);
1159         const char *local = ctdb->recovery_lock;
1160         const char *remote = NULL;
1161
1162         if (status != 0) {
1163                 DEBUG(DEBUG_ERR, (__location__ " GET_RECLOCK failed\n"));
1164                 ctdb_request_control_reply(ctdb, state->c, NULL,
1165                                            status, errormsg);
1166                 talloc_free(state);
1167                 return;
1168         }
1169
1170         /* Check reclock consistency */
1171         if (data.dsize > 0) {
1172                 /* Ensure NUL-termination */
1173                 data.dptr[data.dsize-1] = '\0';
1174                 remote = (const char *)data.dptr;
1175         }
1176         if (! reclock_strings_equal(local, remote)) {
1177                 /* Inconsistent */
1178                 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
1179                 DEBUG(DEBUG_ERR,
1180                       ("Recovery lock configuration inconsistent: "
1181                        "recmaster has %s, this node has %s, shutting down\n",
1182                        remote == NULL ? "NULL" : remote,
1183                        local == NULL ? "NULL" : local));
1184                 talloc_free(state);
1185                 ctdb_shutdown_sequence(ctdb, 1);
1186         }
1187         DEBUG(DEBUG_INFO,
1188               ("Recovery lock consistency check successful\n"));
1189
1190         run_start_recovery_event(ctdb, state);
1191 }
1192
1193 /* Check recovery lock consistency and run eventscripts for the
1194  * "startrecovery" event */
1195 int32_t ctdb_control_start_recovery(struct ctdb_context *ctdb,
1196                                     struct ctdb_req_control_old *c,
1197                                     bool *async_reply)
1198 {
1199         int ret;
1200         struct recovery_callback_state *state;
1201         uint32_t recmaster = c->hdr.srcnode;
1202
1203         DEBUG(DEBUG_NOTICE, ("Recovery has started\n"));
1204         gettimeofday(&ctdb->last_recovery_started, NULL);
1205
1206         state = talloc(ctdb, struct recovery_callback_state);
1207         CTDB_NO_MEMORY(ctdb, state);
1208
1209         state->c = c;
1210
1211         /* Although the recovery master sent this node a start
1212          * recovery control, this node might still think the recovery
1213          * master is disconnected.  In this case defer the recovery
1214          * lock consistency check. */
1215         if (ctdb->nodes[recmaster]->flags & NODE_FLAGS_DISCONNECTED) {
1216                 run_start_recovery_event(ctdb, state);
1217         } else {
1218                 /* Ask the recovery master about its reclock setting */
1219                 ret = ctdb_daemon_send_control(ctdb,
1220                                                recmaster,
1221                                                0,
1222                                                CTDB_CONTROL_GET_RECLOCK_FILE,
1223                                                0, 0,
1224                                                tdb_null,
1225                                                start_recovery_reclock_callback,
1226                                                state);
1227
1228                 if (ret != 0) {
1229                         DEBUG(DEBUG_ERR, (__location__ " GET_RECLOCK failed\n"));
1230                         talloc_free(state);
1231                         return -1;
1232                 }
1233         }
1234
1235         /* tell the control that we will be reply asynchronously */
1236         state->c = talloc_steal(state, c);
1237         *async_reply = true;
1238
1239         return 0;
1240 }
1241
1242 /*
1243  try to delete all these records as part of the vacuuming process
1244  and return the records we failed to delete
1245 */
1246 int32_t ctdb_control_try_delete_records(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
1247 {
1248         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1249         struct ctdb_db_context *ctdb_db;
1250         int i;
1251         struct ctdb_rec_data_old *rec;
1252         struct ctdb_marshall_buffer *records;
1253
1254         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1255                 DEBUG(DEBUG_ERR,(__location__ " invalid data in try_delete_records\n"));
1256                 return -1;
1257         }
1258
1259         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1260         if (!ctdb_db) {
1261                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
1262                 return -1;
1263         }
1264
1265
1266         DEBUG(DEBUG_DEBUG,("starting try_delete_records of %u records for dbid 0x%x\n",
1267                  reply->count, reply->db_id));
1268
1269
1270         /* create a blob to send back the records we couldnt delete */
1271         records = (struct ctdb_marshall_buffer *)
1272                         talloc_zero_size(outdata,
1273                                     offsetof(struct ctdb_marshall_buffer, data));
1274         if (records == NULL) {
1275                 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
1276                 return -1;
1277         }
1278         records->db_id = ctdb_db->db_id;
1279
1280
1281         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1282         for (i=0;i<reply->count;i++) {
1283                 TDB_DATA key, data;
1284
1285                 key.dptr = &rec->data[0];
1286                 key.dsize = rec->keylen;
1287                 data.dptr = &rec->data[key.dsize];
1288                 data.dsize = rec->datalen;
1289
1290                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1291                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record in indata\n"));
1292                         talloc_free(records);
1293                         return -1;
1294                 }
1295
1296                 /* If we cant delete the record we must add it to the reply
1297                    so the lmaster knows it may not purge this record
1298                 */
1299                 if (delete_tdb_record(ctdb, ctdb_db, rec) != 0) {
1300                         size_t old_size;
1301                         struct ctdb_ltdb_header *hdr;
1302
1303                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1304                         data.dptr += sizeof(*hdr);
1305                         data.dsize -= sizeof(*hdr);
1306
1307                         DEBUG(DEBUG_INFO, (__location__ " Failed to vacuum delete record with hash 0x%08x\n", ctdb_hash(&key)));
1308
1309                         old_size = talloc_get_size(records);
1310                         records = talloc_realloc_size(outdata, records, old_size + rec->length);
1311                         if (records == NULL) {
1312                                 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
1313                                 return -1;
1314                         }
1315                         records->count++;
1316                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1317                 }
1318
1319                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1320         }
1321
1322
1323         *outdata = ctdb_marshall_finish(records);
1324
1325         return 0;
1326 }
1327
1328 /**
1329  * Store a record as part of the vacuum process:
1330  * This is called from the RECEIVE_RECORD control which
1331  * the lmaster uses to send the current empty copy
1332  * to all nodes for storing, before it lets the other
1333  * nodes delete the records in the second phase with
1334  * the TRY_DELETE_RECORDS control.
1335  *
1336  * Only store if we are not lmaster or dmaster, and our
1337  * rsn is <= the provided rsn. Use non-blocking locks.
1338  *
1339  * return 0 if the record was successfully stored.
1340  * return !0 if the record still exists in the tdb after returning.
1341  */
1342 static int store_tdb_record(struct ctdb_context *ctdb,
1343                             struct ctdb_db_context *ctdb_db,
1344                             struct ctdb_rec_data_old *rec)
1345 {
1346         TDB_DATA key, data, data2;
1347         struct ctdb_ltdb_header *hdr, *hdr2;
1348         int ret;
1349
1350         key.dsize = rec->keylen;
1351         key.dptr = &rec->data[0];
1352         data.dsize = rec->datalen;
1353         data.dptr = &rec->data[rec->keylen];
1354
1355         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1356                 DEBUG(DEBUG_INFO, (__location__ " Called store_tdb_record "
1357                                    "where we are lmaster\n"));
1358                 return -1;
1359         }
1360
1361         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1362                 DEBUG(DEBUG_ERR, (__location__ " Bad record size\n"));
1363                 return -1;
1364         }
1365
1366         hdr = (struct ctdb_ltdb_header *)data.dptr;
1367
1368         /* use a non-blocking lock */
1369         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1370                 DEBUG(DEBUG_INFO, (__location__ " Failed to lock chain in non-blocking mode\n"));
1371                 return -1;
1372         }
1373
1374         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1375         if (data2.dptr == NULL || data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1376                 if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) == -1) {
1377                         DEBUG(DEBUG_ERR, (__location__ "Failed to store record\n"));
1378                         ret = -1;
1379                         goto done;
1380                 }
1381                 DEBUG(DEBUG_INFO, (__location__ " Stored record\n"));
1382                 ret = 0;
1383                 goto done;
1384         }
1385
1386         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1387
1388         if (hdr2->rsn > hdr->rsn) {
1389                 DEBUG(DEBUG_INFO, (__location__ " Skipping record with "
1390                                    "rsn=%llu - called with rsn=%llu\n",
1391                                    (unsigned long long)hdr2->rsn,
1392                                    (unsigned long long)hdr->rsn));
1393                 ret = -1;
1394                 goto done;
1395         }
1396
1397         /* do not allow vacuuming of records that have readonly flags set. */
1398         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1399                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1400                                   "flags set\n"));
1401                 ret = -1;
1402                 goto done;
1403         }
1404         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1405                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1406                                   "flags set\n"));
1407                 ret = -1;
1408                 goto done;
1409         }
1410
1411         if (hdr2->dmaster == ctdb->pnn) {
1412                 DEBUG(DEBUG_INFO, (__location__ " Attempted to store record "
1413                                    "where we are the dmaster\n"));
1414                 ret = -1;
1415                 goto done;
1416         }
1417
1418         if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) != 0) {
1419                 DEBUG(DEBUG_INFO,(__location__ " Failed to store record\n"));
1420                 ret = -1;
1421                 goto done;
1422         }
1423
1424         ret = 0;
1425
1426 done:
1427         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1428         free(data2.dptr);
1429         return  ret;
1430 }
1431
1432
1433
1434 /**
1435  * Try to store all these records as part of the vacuuming process
1436  * and return the records we failed to store.
1437  */
1438 int32_t ctdb_control_receive_records(struct ctdb_context *ctdb,
1439                                      TDB_DATA indata, TDB_DATA *outdata)
1440 {
1441         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1442         struct ctdb_db_context *ctdb_db;
1443         int i;
1444         struct ctdb_rec_data_old *rec;
1445         struct ctdb_marshall_buffer *records;
1446
1447         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1448                 DEBUG(DEBUG_ERR,
1449                       (__location__ " invalid data in receive_records\n"));
1450                 return -1;
1451         }
1452
1453         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1454         if (!ctdb_db) {
1455                 DEBUG(DEBUG_ERR, (__location__ " Unknown db 0x%08x\n",
1456                                   reply->db_id));
1457                 return -1;
1458         }
1459
1460         DEBUG(DEBUG_DEBUG, ("starting receive_records of %u records for "
1461                             "dbid 0x%x\n", reply->count, reply->db_id));
1462
1463         /* create a blob to send back the records we could not store */
1464         records = (struct ctdb_marshall_buffer *)
1465                         talloc_zero_size(outdata,
1466                                 offsetof(struct ctdb_marshall_buffer, data));
1467         if (records == NULL) {
1468                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1469                 return -1;
1470         }
1471         records->db_id = ctdb_db->db_id;
1472
1473         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1474         for (i=0; i<reply->count; i++) {
1475                 TDB_DATA key, data;
1476
1477                 key.dptr = &rec->data[0];
1478                 key.dsize = rec->keylen;
1479                 data.dptr = &rec->data[key.dsize];
1480                 data.dsize = rec->datalen;
1481
1482                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1483                         DEBUG(DEBUG_CRIT, (__location__ " bad ltdb record "
1484                                            "in indata\n"));
1485                         talloc_free(records);
1486                         return -1;
1487                 }
1488
1489                 /*
1490                  * If we can not store the record we must add it to the reply
1491                  * so the lmaster knows it may not purge this record.
1492                  */
1493                 if (store_tdb_record(ctdb, ctdb_db, rec) != 0) {
1494                         size_t old_size;
1495                         struct ctdb_ltdb_header *hdr;
1496
1497                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1498                         data.dptr += sizeof(*hdr);
1499                         data.dsize -= sizeof(*hdr);
1500
1501                         DEBUG(DEBUG_INFO, (__location__ " Failed to store "
1502                                            "record with hash 0x%08x in vacuum "
1503                                            "via RECEIVE_RECORDS\n",
1504                                            ctdb_hash(&key)));
1505
1506                         old_size = talloc_get_size(records);
1507                         records = talloc_realloc_size(outdata, records,
1508                                                       old_size + rec->length);
1509                         if (records == NULL) {
1510                                 DEBUG(DEBUG_ERR, (__location__ " Failed to "
1511                                                   "expand\n"));
1512                                 return -1;
1513                         }
1514                         records->count++;
1515                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1516                 }
1517
1518                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1519         }
1520
1521         *outdata = ctdb_marshall_finish(records);
1522
1523         return 0;
1524 }
1525
1526
1527 /*
1528   report capabilities
1529  */
1530 int32_t ctdb_control_get_capabilities(struct ctdb_context *ctdb, TDB_DATA *outdata)
1531 {
1532         uint32_t *capabilities = NULL;
1533
1534         capabilities = talloc(outdata, uint32_t);
1535         CTDB_NO_MEMORY(ctdb, capabilities);
1536         *capabilities = ctdb->capabilities;
1537
1538         outdata->dsize = sizeof(uint32_t);
1539         outdata->dptr = (uint8_t *)capabilities;
1540
1541         return 0;
1542 }
1543
1544 /* The recovery daemon will ping us at regular intervals.
1545    If we havent been pinged for a while we assume the recovery
1546    daemon is inoperable and we restart.
1547 */
1548 static void ctdb_recd_ping_timeout(struct tevent_context *ev,
1549                                    struct tevent_timer *te,
1550                                    struct timeval t, void *p)
1551 {
1552         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
1553         uint32_t *count = talloc_get_type(ctdb->recd_ping_count, uint32_t);
1554
1555         DEBUG(DEBUG_ERR, ("Recovery daemon ping timeout. Count : %u\n", *count));
1556
1557         if (*count < ctdb->tunable.recd_ping_failcount) {
1558                 (*count)++;
1559                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1560                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1561                                  ctdb_recd_ping_timeout, ctdb);
1562                 return;
1563         }
1564
1565         DEBUG(DEBUG_ERR, ("Final timeout for recovery daemon ping. Restarting recovery daemon. (This can be caused if the cluster filesystem has hung)\n"));
1566
1567         ctdb_stop_recoverd(ctdb);
1568         ctdb_start_recoverd(ctdb);
1569 }
1570
1571 int32_t ctdb_control_recd_ping(struct ctdb_context *ctdb)
1572 {
1573         talloc_free(ctdb->recd_ping_count);
1574
1575         ctdb->recd_ping_count = talloc_zero(ctdb, uint32_t);
1576         CTDB_NO_MEMORY(ctdb, ctdb->recd_ping_count);
1577
1578         if (ctdb->tunable.recd_ping_timeout != 0) {
1579                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1580                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1581                                  ctdb_recd_ping_timeout, ctdb);
1582         }
1583
1584         return 0;
1585 }
1586
1587
1588
1589 int32_t ctdb_control_set_recmaster(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata)
1590 {
1591         uint32_t new_recmaster;
1592
1593         CHECK_CONTROL_DATA_SIZE(sizeof(uint32_t));
1594         new_recmaster = ((uint32_t *)(&indata.dptr[0]))[0];
1595
1596         if (ctdb->pnn != new_recmaster && ctdb->recovery_master == ctdb->pnn) {
1597                 DEBUG(DEBUG_NOTICE,
1598                       ("Remote node (%u) is now the recovery master\n",
1599                        new_recmaster));
1600         }
1601
1602         if (ctdb->pnn == new_recmaster && ctdb->recovery_master != new_recmaster) {
1603                 DEBUG(DEBUG_NOTICE,
1604                       ("This node (%u) is now the recovery master\n",
1605                        ctdb->pnn));
1606         }
1607
1608         ctdb->recovery_master = new_recmaster;
1609         return 0;
1610 }
1611
1612
1613 int32_t ctdb_control_stop_node(struct ctdb_context *ctdb)
1614 {
1615         DEBUG(DEBUG_NOTICE, ("Stopping node\n"));
1616         ctdb_disable_monitoring(ctdb);
1617         ctdb->nodes[ctdb->pnn]->flags |= NODE_FLAGS_STOPPED;
1618
1619         return 0;
1620 }
1621
1622 int32_t ctdb_control_continue_node(struct ctdb_context *ctdb)
1623 {
1624         DEBUG(DEBUG_NOTICE, ("Continue node\n"));
1625         ctdb->nodes[ctdb->pnn]->flags &= ~NODE_FLAGS_STOPPED;
1626
1627         return 0;
1628 }
1629