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                 if (ctdb_db->persistent != 0) {
 119                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_PERSISTENT;
 120                 }
 121                 if (ctdb_db->readonly != 0) {
 122                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_READONLY;
 123                 }
 124                 if (ctdb_db->sticky != 0) {
 125                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_STICKY;
 126                 }
 127         }
 128
 129         return 0;
 130 }
 131
 132 int
 133 ctdb_control_getnodemap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
 134 {
 135         CHECK_CONTROL_DATA_SIZE(0);
 136
 137         outdata->dptr  = (unsigned char *)ctdb_node_list_to_map(ctdb->nodes,
 138                                                                 ctdb->num_nodes,
 139                                                                 outdata);
 140         if (outdata->dptr == NULL) {
 141                 return -1;
 142         }
 143
 144         outdata->dsize = talloc_get_size(outdata->dptr);
 145
 146         return 0;
 147 }
 148
 149 /*
 150   reload the nodes file
 151 */
 152 int
 153 ctdb_control_reload_nodes_file(struct ctdb_context *ctdb, uint32_t opcode)
 154 {
 155         int i, num_nodes;
 156         TALLOC_CTX *tmp_ctx;
 157         struct ctdb_node **nodes;
 158
 159         tmp_ctx = talloc_new(ctdb);
 160
 161         /* steal the old nodes file for a while */
 162         talloc_steal(tmp_ctx, ctdb->nodes);
 163         nodes = ctdb->nodes;
 164         ctdb->nodes = NULL;
 165         num_nodes = ctdb->num_nodes;
 166         ctdb->num_nodes = 0;
 167
 168         /* load the new nodes file */
 169         ctdb_load_nodes_file(ctdb);
 170
 171         for (i=0; i<ctdb->num_nodes; i++) {
 172                 /* keep any identical pre-existing nodes and connections */
 173                 if ((i < num_nodes) && ctdb_same_address(&ctdb->nodes[i]->address, &nodes[i]->address)) {
 174                         talloc_free(ctdb->nodes[i]);
 175                         ctdb->nodes[i] = talloc_steal(ctdb->nodes, nodes[i]);
 176                         continue;
 177                 }
 178
 179                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 180                         continue;
 181                 }
 182
 183                 /* any new or different nodes must be added */
 184                 if (ctdb->methods->add_node(ctdb->nodes[i]) != 0) {
 185                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_node failed at %d\n", i));
 186                         ctdb_fatal(ctdb, "failed to add node. shutting down\n");
 187                 }
 188                 if (ctdb->methods->connect_node(ctdb->nodes[i]) != 0) {
 189                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_connect failed at %d\n", i));
 190                         ctdb_fatal(ctdb, "failed to connect to node. shutting down\n");
 191                 }
 192         }
 193
 194         /* tell the recovery daemon to reaload the nodes file too */
 195         ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELOAD_NODES, tdb_null);
 196
 197         talloc_free(tmp_ctx);
 198
 199         return 0;
 200 }
 201
 202 /*
 203    a traverse function for pulling all relevent records from pulldb
 204  */
 205 struct pulldb_data {
 206         struct ctdb_context *ctdb;
 207         struct ctdb_db_context *ctdb_db;
 208         struct ctdb_marshall_buffer *pulldata;
 209         uint32_t len;
 210         uint32_t allocated_len;
 211         bool failed;
 212 };
 213
 214 static int traverse_pulldb(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *p)
 215 {
 216         struct pulldb_data *params = (struct pulldb_data *)p;
 217         struct ctdb_rec_data_old *rec;
 218         struct ctdb_context *ctdb = params->ctdb;
 219         struct ctdb_db_context *ctdb_db = params->ctdb_db;
 220
 221         /* add the record to the blob */
 222         rec = ctdb_marshall_record(params->pulldata, 0, key, NULL, data);
 223         if (rec == NULL) {
 224                 params->failed = true;
 225                 return -1;
 226         }
 227         if (params->len + rec->length >= params->allocated_len) {
 228                 params->allocated_len = rec->length + params->len + ctdb->tunable.pulldb_preallocation_size;
 229                 params->pulldata = talloc_realloc_size(NULL, params->pulldata, params->allocated_len);
 230         }
 231         if (params->pulldata == NULL) {
 232                 DEBUG(DEBUG_CRIT,(__location__ " Failed to expand pulldb_data to %u\n", rec->length + params->len));
 233                 ctdb_fatal(params->ctdb, "failed to allocate memory for recovery. shutting down\n");
 234         }
 235         params->pulldata->count++;
 236         memcpy(params->len+(uint8_t *)params->pulldata, rec, rec->length);
 237         params->len += rec->length;
 238
 239         if (ctdb->tunable.db_record_size_warn != 0 && rec->length > ctdb->tunable.db_record_size_warn) {
 240                 DEBUG(DEBUG_ERR,("Data record in %s is big. Record size is %d bytes\n", ctdb_db->db_name, (int)rec->length));
 241         }
 242
 243         talloc_free(rec);
 244
 245         return 0;
 246 }
 247
 248 /*
 249   pull a bunch of records from a ltdb, filtering by lmaster
 250  */
 251 int32_t ctdb_control_pull_db(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
 252 {
 253         struct ctdb_pulldb *pull;
 254         struct ctdb_db_context *ctdb_db;
 255         struct pulldb_data params;
 256         struct ctdb_marshall_buffer *reply;
 257
 258         pull = (struct ctdb_pulldb *)indata.dptr;
 259
 260         ctdb_db = find_ctdb_db(ctdb, pull->db_id);
 261         if (!ctdb_db) {
 262                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", pull->db_id));
 263                 return -1;
 264         }
 265
 266         if (!ctdb_db_frozen(ctdb_db)) {
 267                 DEBUG(DEBUG_ERR,
 268                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 269                 return -1;
 270         }
 271
 272         reply = talloc_zero(outdata, struct ctdb_marshall_buffer);
 273         CTDB_NO_MEMORY(ctdb, reply);
 274
 275         reply->db_id = pull->db_id;
 276
 277         params.ctdb = ctdb;
 278         params.ctdb_db = ctdb_db;
 279         params.pulldata = reply;
 280         params.len = offsetof(struct ctdb_marshall_buffer, data);
 281         params.allocated_len = params.len;
 282         params.failed = false;
 283
 284         if (ctdb_db->unhealthy_reason) {
 285                 /* this is just a warning, as the tdb should be empty anyway */
 286                 DEBUG(DEBUG_WARNING,("db(%s) unhealty in ctdb_control_pull_db: %s\n",
 287                                      ctdb_db->db_name, ctdb_db->unhealthy_reason));
 288         }
 289
 290         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 291                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 292                 return -1;
 293         }
 294
 295         if (tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_pulldb, &params) == -1) {
 296                 DEBUG(DEBUG_ERR,(__location__ " Failed to get traverse db '%s'\n", ctdb_db->db_name));
 297                 ctdb_lockdb_unmark(ctdb_db);
 298                 talloc_free(params.pulldata);
 299                 return -1;
 300         }
 301
 302         ctdb_lockdb_unmark(ctdb_db);
 303
 304         outdata->dptr = (uint8_t *)params.pulldata;
 305         outdata->dsize = params.len;
 306
 307         if (ctdb->tunable.db_record_count_warn != 0 && params.pulldata->count > ctdb->tunable.db_record_count_warn) {
 308                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d records\n", ctdb_db->db_name, params.pulldata->count));
 309         }
 310         if (ctdb->tunable.db_size_warn != 0 && outdata->dsize > ctdb->tunable.db_size_warn) {
 311                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d bytes\n", ctdb_db->db_name, (int)outdata->dsize));
 312         }
 313
 314
 315         return 0;
 316 }
 317
 318 struct db_pull_state {
 319         struct ctdb_context *ctdb;
 320         struct ctdb_db_context *ctdb_db;
 321         struct ctdb_marshall_buffer *recs;
 322         uint32_t pnn;
 323         uint64_t srvid;
 324         uint32_t num_records;
 325 };
 326
 327 static int traverse_db_pull(struct tdb_context *tdb, TDB_DATA key,
 328                             TDB_DATA data, void *private_data)
 329 {
 330         struct db_pull_state *state = (struct db_pull_state *)private_data;
 331         struct ctdb_marshall_buffer *recs;
 332
 333         recs = ctdb_marshall_add(state->ctdb, state->recs,
 334                                  state->ctdb_db->db_id, 0, key, NULL, data);
 335         if (recs == NULL) {
 336                 TALLOC_FREE(state->recs);
 337                 return -1;
 338         }
 339         state->recs = recs;
 340
 341         if (talloc_get_size(state->recs) >=
 342                         state->ctdb->tunable.rec_buffer_size_limit) {
 343                 TDB_DATA buffer;
 344                 int ret;
 345
 346                 buffer = ctdb_marshall_finish(state->recs);
 347                 ret = ctdb_daemon_send_message(state->ctdb, state->pnn,
 348                                                state->srvid, buffer);
 349                 if (ret != 0) {
 350                         TALLOC_FREE(state->recs);
 351                         return -1;
 352                 }
 353
 354                 state->num_records += state->recs->count;
 355                 TALLOC_FREE(state->recs);
 356         }
 357
 358         return 0;
 359 }
 360
 361 int32_t ctdb_control_db_pull(struct ctdb_context *ctdb,
 362                              struct ctdb_req_control_old *c,
 363                              TDB_DATA indata, TDB_DATA *outdata)
 364 {
 365         struct ctdb_pulldb_ext *pulldb_ext;
 366         struct ctdb_db_context *ctdb_db;
 367         struct db_pull_state state;
 368         int ret;
 369
 370         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 371
 372         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 373         if (ctdb_db == NULL) {
 374                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n",
 375                                  pulldb_ext->db_id));
 376                 return -1;
 377         }
 378
 379         if (!ctdb_db_frozen(ctdb_db)) {
 380                 DEBUG(DEBUG_ERR,
 381                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 382                 return -1;
 383         }
 384
 385         if (ctdb_db->unhealthy_reason) {
 386                 /* this is just a warning, as the tdb should be empty anyway */
 387                 DEBUG(DEBUG_WARNING,
 388                       ("db(%s) unhealty in ctdb_control_db_pull: %s\n",
 389                        ctdb_db->db_name, ctdb_db->unhealthy_reason));
 390         }
 391
 392         state.ctdb = ctdb;
 393         state.ctdb_db = ctdb_db;
 394         state.recs = NULL;
 395         state.pnn = c->hdr.srcnode;
 396         state.srvid = pulldb_ext->srvid;
 397         state.num_records = 0;
 398
 399         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 400                 DEBUG(DEBUG_ERR,
 401                       (__location__ " Failed to get lock on entire db - failing\n"));
 402                 return -1;
 403         }
 404
 405         ret = tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_db_pull, &state);
 406         if (ret == -1) {
 407                 DEBUG(DEBUG_ERR,
 408                       (__location__ " Failed to get traverse db '%s'\n",
 409                        ctdb_db->db_name));
 410                 ctdb_lockdb_unmark(ctdb_db);
 411                 return -1;
 412         }
 413
 414         /* Last few records */
 415         if (state.recs != NULL) {
 416                 TDB_DATA buffer;
 417
 418                 buffer = ctdb_marshall_finish(state.recs);
 419                 ret = ctdb_daemon_send_message(state.ctdb, state.pnn,
 420                                                state.srvid, buffer);
 421                 if (ret != 0) {
 422                         TALLOC_FREE(state.recs);
 423                         ctdb_lockdb_unmark(ctdb_db);
 424                         return -1;
 425                 }
 426
 427                 state.num_records += state.recs->count;
 428                 TALLOC_FREE(state.recs);
 429         }
 430
 431         ctdb_lockdb_unmark(ctdb_db);
 432
 433         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 434         if (outdata->dptr == NULL) {
 435                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 436                 return -1;
 437         }
 438
 439         memcpy(outdata->dptr, (uint8_t *)&state.num_records, sizeof(uint32_t));
 440         outdata->dsize = sizeof(uint32_t);
 441
 442         return 0;
 443 }
 444
 445 /*
 446   push a bunch of records into a ltdb, filtering by rsn
 447  */
 448 int32_t ctdb_control_push_db(struct ctdb_context *ctdb, TDB_DATA indata)
 449 {
 450         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
 451         struct ctdb_db_context *ctdb_db;
 452         int i, ret;
 453         struct ctdb_rec_data_old *rec;
 454
 455         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
 456                 DEBUG(DEBUG_ERR,(__location__ " invalid data in pulldb reply\n"));
 457                 return -1;
 458         }
 459
 460         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
 461         if (!ctdb_db) {
 462                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
 463                 return -1;
 464         }
 465
 466         if (!ctdb_db_frozen(ctdb_db)) {
 467                 DEBUG(DEBUG_ERR,
 468                       ("rejecting ctdb_control_push_db when not frozen\n"));
 469                 return -1;
 470         }
 471
 472         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 473                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 474                 return -1;
 475         }
 476
 477         rec = (struct ctdb_rec_data_old *)&reply->data[0];
 478
 479         DEBUG(DEBUG_INFO,("starting push of %u records for dbid 0x%x\n",
 480                  reply->count, reply->db_id));
 481
 482         for (i=0;i<reply->count;i++) {
 483                 TDB_DATA key, data;
 484                 struct ctdb_ltdb_header *hdr;
 485
 486                 key.dptr = &rec->data[0];
 487                 key.dsize = rec->keylen;
 488                 data.dptr = &rec->data[key.dsize];
 489                 data.dsize = rec->datalen;
 490
 491                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 492                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 493                         goto failed;
 494                 }
 495                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 496                 /* strip off any read only record flags. All readonly records
 497                    are revoked implicitely by a recovery
 498                 */
 499                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 500
 501                 data.dptr += sizeof(*hdr);
 502                 data.dsize -= sizeof(*hdr);
 503
 504                 ret = ctdb_ltdb_store(ctdb_db, key, hdr, data);
 505                 if (ret != 0) {
 506                         DEBUG(DEBUG_CRIT, (__location__ " Unable to store record\n"));
 507                         goto failed;
 508                 }
 509
 510                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 511         }
 512
 513         DEBUG(DEBUG_DEBUG,("finished push of %u records for dbid 0x%x\n",
 514                  reply->count, reply->db_id));
 515
 516         if (ctdb_db->readonly) {
 517                 DEBUG(DEBUG_CRIT,("Clearing the tracking database for dbid 0x%x\n",
 518                                   ctdb_db->db_id));
 519                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 520                         DEBUG(DEBUG_ERR,("Failed to wipe tracking database for 0x%x. Dropping read-only delegation support\n", ctdb_db->db_id));
 521                         ctdb_db->readonly = false;
 522                         tdb_close(ctdb_db->rottdb);
 523                         ctdb_db->rottdb = NULL;
 524                         ctdb_db->readonly = false;
 525                 }
 526                 while (ctdb_db->revokechild_active != NULL) {
 527                         talloc_free(ctdb_db->revokechild_active);
 528                 }
 529         }
 530
 531         ctdb_lockdb_unmark(ctdb_db);
 532         return 0;
 533
 534 failed:
 535         ctdb_lockdb_unmark(ctdb_db);
 536         return -1;
 537 }
 538
 539 struct db_push_state {
 540         struct ctdb_context *ctdb;
 541         struct ctdb_db_context *ctdb_db;
 542         uint64_t srvid;
 543         uint32_t num_records;
 544         bool failed;
 545 };
 546
 547 static void db_push_msg_handler(uint64_t srvid, TDB_DATA indata,
 548                                 void *private_data)
 549 {
 550         struct db_push_state *state = talloc_get_type(
 551                 private_data, struct db_push_state);
 552         struct ctdb_marshall_buffer *recs;
 553         struct ctdb_rec_data_old *rec;
 554         int i, ret;
 555
 556         if (state->failed) {
 557                 return;
 558         }
 559
 560         recs = (struct ctdb_marshall_buffer *)indata.dptr;
 561         rec = (struct ctdb_rec_data_old *)&recs->data[0];
 562
 563         DEBUG(DEBUG_INFO, ("starting push of %u records for dbid 0x%x\n",
 564                            recs->count, recs->db_id));
 565
 566         for (i=0; i<recs->count; i++) {
 567                 TDB_DATA key, data;
 568                 struct ctdb_ltdb_header *hdr;
 569
 570                 key.dptr = &rec->data[0];
 571                 key.dsize = rec->keylen;
 572                 data.dptr = &rec->data[key.dsize];
 573                 data.dsize = rec->datalen;
 574
 575                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 576                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 577                         goto failed;
 578                 }
 579
 580                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 581                 /* Strip off any read only record flags.
 582                  * All readonly records are revoked implicitely by a recovery.
 583                  */
 584                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 585
 586                 data.dptr += sizeof(*hdr);
 587                 data.dsize -= sizeof(*hdr);
 588
 589                 ret = ctdb_ltdb_store(state->ctdb_db, key, hdr, data);
 590                 if (ret != 0) {
 591                         DEBUG(DEBUG_ERR,
 592                               (__location__ " Unable to store record\n"));
 593                         goto failed;
 594                 }
 595
 596                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 597         }
 598
 599         DEBUG(DEBUG_DEBUG, ("finished push of %u records for dbid 0x%x\n",
 600                             recs->count, recs->db_id));
 601
 602         state->num_records += recs->count;
 603         return;
 604
 605 failed:
 606         state->failed = true;
 607 }
 608
 609 int32_t ctdb_control_db_push_start(struct ctdb_context *ctdb, TDB_DATA indata)
 610 {
 611         struct ctdb_pulldb_ext *pulldb_ext;
 612         struct ctdb_db_context *ctdb_db;
 613         struct db_push_state *state;
 614         int ret;
 615
 616         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 617
 618         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 619         if (ctdb_db == NULL) {
 620                 DEBUG(DEBUG_ERR,
 621                       (__location__ " Unknown db 0x%08x\n", pulldb_ext->db_id));
 622                 return -1;
 623         }
 624
 625         if (!ctdb_db_frozen(ctdb_db)) {
 626                 DEBUG(DEBUG_ERR,
 627                       ("rejecting ctdb_control_db_push_start when not frozen\n"));
 628                 return -1;
 629         }
 630
 631         if (ctdb_db->push_started) {
 632                 DEBUG(DEBUG_WARNING,
 633                       (__location__ " DB push already started for %s\n",
 634                        ctdb_db->db_name));
 635
 636                 /* De-register old state */
 637                 state = (struct db_push_state *)ctdb_db->push_state;
 638                 if (state != NULL) {
 639                         srvid_deregister(ctdb->srv, state->srvid, state);
 640                         talloc_free(state);
 641                         ctdb_db->push_state = NULL;
 642                 }
 643         }
 644
 645         state = talloc_zero(ctdb_db, struct db_push_state);
 646         if (state == NULL) {
 647                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 648                 return -1;
 649         }
 650
 651         state->ctdb = ctdb;
 652         state->ctdb_db = ctdb_db;
 653         state->srvid = pulldb_ext->srvid;
 654         state->failed = false;
 655
 656         ret = srvid_register(ctdb->srv, state, state->srvid,
 657                              db_push_msg_handler, state);
 658         if (ret != 0) {
 659                 DEBUG(DEBUG_ERR,
 660                       (__location__ " Failed to register srvid for db push\n"));
 661                 talloc_free(state);
 662                 return -1;
 663         }
 664
 665         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 666                 DEBUG(DEBUG_ERR,
 667                       (__location__ " Failed to get lock on entire db - failing\n"));
 668                 srvid_deregister(ctdb->srv, state->srvid, state);
 669                 talloc_free(state);
 670                 return -1;
 671         }
 672
 673         ctdb_db->push_started = true;
 674         ctdb_db->push_state = state;
 675
 676         return 0;
 677 }
 678
 679 int32_t ctdb_control_db_push_confirm(struct ctdb_context *ctdb,
 680                                      TDB_DATA indata, TDB_DATA *outdata)
 681 {
 682         uint32_t db_id;
 683         struct ctdb_db_context *ctdb_db;
 684         struct db_push_state *state;
 685
 686         db_id = *(uint32_t *)indata.dptr;
 687
 688         ctdb_db = find_ctdb_db(ctdb, db_id);
 689         if (ctdb_db == NULL) {
 690                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", db_id));
 691                 return -1;
 692         }
 693
 694         if (!ctdb_db_frozen(ctdb_db)) {
 695                 DEBUG(DEBUG_ERR,
 696                       ("rejecting ctdb_control_db_push_confirm when not frozen\n"));
 697                 return -1;
 698         }
 699
 700         if (!ctdb_db->push_started) {
 701                 DEBUG(DEBUG_ERR, (__location__ " DB push not started\n"));
 702                 return -1;
 703         }
 704
 705         if (ctdb_db->readonly) {
 706                 DEBUG(DEBUG_ERR,
 707                       ("Clearing the tracking database for dbid 0x%x\n",
 708                        ctdb_db->db_id));
 709                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 710                         DEBUG(DEBUG_ERR,
 711                               ("Failed to wipe tracking database for 0x%x."
 712                                " Dropping read-only delegation support\n",
 713                                ctdb_db->db_id));
 714                         ctdb_db->readonly = false;
 715                         tdb_close(ctdb_db->rottdb);
 716                         ctdb_db->rottdb = NULL;
 717                         ctdb_db->readonly = false;
 718                 }
 719
 720                 while (ctdb_db->revokechild_active != NULL) {
 721                         talloc_free(ctdb_db->revokechild_active);
 722                 }
 723         }
 724
 725         ctdb_lockdb_unmark(ctdb_db);
 726
 727         state = (struct db_push_state *)ctdb_db->push_state;
 728         if (state == NULL) {
 729                 DEBUG(DEBUG_ERR, (__location__ " Missing push db state\n"));
 730                 return -1;
 731         }
 732
 733         srvid_deregister(ctdb->srv, state->srvid, state);
 734
 735         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 736         if (outdata->dptr == NULL) {
 737                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 738                 talloc_free(state);
 739                 ctdb_db->push_state = NULL;
 740                 return -1;
 741         }
 742
 743         memcpy(outdata->dptr, (uint8_t *)&state->num_records, sizeof(uint32_t));
 744         outdata->dsize = sizeof(uint32_t);
 745
 746         talloc_free(state);
 747         ctdb_db->push_state = NULL;
 748
 749         return 0;
 750 }
 751
 752 struct set_recmode_state {
 753         struct ctdb_context *ctdb;
 754         struct ctdb_req_control_old *c;
 755 };
 756
 757 static void set_recmode_handler(char status,
 758                                 double latency,
 759                                 void *private_data)
 760 {
 761         struct set_recmode_state *state = talloc_get_type_abort(
 762                 private_data, struct set_recmode_state);
 763         int s = 0;
 764         const char *err = NULL;
 765
 766         switch (status) {
 767         case '0':
 768                 /* Mutex taken */
 769                 DEBUG(DEBUG_ERR,
 770                       ("ERROR: Daemon able to take recovery lock on \"%s\" during recovery\n",
 771                        state->ctdb->recovery_lock));
 772                 s = -1;
 773                 err = "Took recovery lock from daemon during recovery - probably a cluster filesystem lock coherence problem";
 774                 break;
 775
 776         case '1':
 777                 /* Contention */
 778                 DEBUG(DEBUG_DEBUG, (__location__ " Recovery lock check OK\n"));
 779                 state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 780                 ctdb_process_deferred_attach(state->ctdb);
 781
 782                 s = 0;
 783
 784                 CTDB_UPDATE_RECLOCK_LATENCY(state->ctdb, "daemon reclock",
 785                                             reclock.ctdbd, latency);
 786                 break;
 787
 788         case '2':
 789                 /* Timeout.  Consider this a success, not a failure,
 790                  * as we failed to set the recovery lock which is what
 791                  * we wanted.  This can be caused by the cluster
 792                  * filesystem being very slow to arbitrate locks
 793                  * immediately after a node failure. */
 794                 DEBUG(DEBUG_WARNING,
 795                       (__location__
 796                        "Time out getting recovery lock, allowing recmode set anyway\n"));
 797                 state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 798                 ctdb_process_deferred_attach(state->ctdb);
 799
 800                 s = 0;
 801                 break;
 802
 803         default:
 804                 DEBUG(DEBUG_ERR,
 805                       ("Unexpected error when testing recovery lock\n"));
 806                 s = -1;
 807                 err = "Unexpected error when testing recovery lock";
 808         }
 809
 810         ctdb_request_control_reply(state->ctdb, state->c, NULL, s, err);
 811         talloc_free(state);
 812 }
 813
 814 static void
 815 ctdb_drop_all_ips_event(struct tevent_context *ev, struct tevent_timer *te,
 816                         struct timeval t, void *private_data)
 817 {
 818         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 819
 820         DEBUG(DEBUG_ERR,(__location__ " Been in recovery mode for too long. Dropping all IPS\n"));
 821         talloc_free(ctdb->release_ips_ctx);
 822         ctdb->release_ips_ctx = NULL;
 823
 824         ctdb_release_all_ips(ctdb);
 825 }
 826
 827 /*
 828  * Set up an event to drop all public ips if we remain in recovery for too
 829  * long
 830  */
 831 int ctdb_deferred_drop_all_ips(struct ctdb_context *ctdb)
 832 {
 833         if (ctdb->release_ips_ctx != NULL) {
 834                 talloc_free(ctdb->release_ips_ctx);
 835         }
 836         ctdb->release_ips_ctx = talloc_new(ctdb);
 837         CTDB_NO_MEMORY(ctdb, ctdb->release_ips_ctx);
 838
 839         tevent_add_timer(ctdb->ev, ctdb->release_ips_ctx,
 840                          timeval_current_ofs(ctdb->tunable.recovery_drop_all_ips, 0),
 841                          ctdb_drop_all_ips_event, ctdb);
 842         return 0;
 843 }
 844
 845 /*
 846   set the recovery mode
 847  */
 848 int32_t ctdb_control_set_recmode(struct ctdb_context *ctdb,
 849                                  struct ctdb_req_control_old *c,
 850                                  TDB_DATA indata, bool *async_reply,
 851                                  const char **errormsg)
 852 {
 853         uint32_t recmode = *(uint32_t *)indata.dptr;
 854         int i;
 855         struct ctdb_db_context *ctdb_db;
 856         struct set_recmode_state *state;
 857         struct ctdb_cluster_mutex_handle *h;
 858
 859         /* if we enter recovery but stay in recovery for too long
 860            we will eventually drop all our ip addresses
 861         */
 862         if (recmode == CTDB_RECOVERY_NORMAL) {
 863                 talloc_free(ctdb->release_ips_ctx);
 864                 ctdb->release_ips_ctx = NULL;
 865         } else {
 866                 if (ctdb_deferred_drop_all_ips(ctdb) != 0) {
 867                         DEBUG(DEBUG_ERR,("Failed to set up deferred drop all ips\n"));
 868                 }
 869         }
 870
 871         if (recmode != ctdb->recovery_mode) {
 872                 DEBUG(DEBUG_NOTICE,(__location__ " Recovery mode set to %s\n",
 873                          recmode==CTDB_RECOVERY_NORMAL?"NORMAL":"ACTIVE"));
 874         }
 875
 876         if (recmode != CTDB_RECOVERY_NORMAL ||
 877             ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
 878                 ctdb->recovery_mode = recmode;
 879                 return 0;
 880         }
 881
 882         /* From this point: recmode == CTDB_RECOVERY_NORMAL
 883          *
 884          * Therefore, what follows is special handling when setting
 885          * recovery mode back to normal */
 886
 887         for (ctdb_db = ctdb->db_list; ctdb_db != NULL; ctdb_db = ctdb_db->next) {
 888                 if (ctdb_db->generation != ctdb->vnn_map->generation) {
 889                         DEBUG(DEBUG_ERR,
 890                               ("Inconsistent DB generation %u for %s\n",
 891                                ctdb_db->generation, ctdb_db->db_name));
 892                         DEBUG(DEBUG_ERR, ("Recovery mode set to ACTIVE\n"));
 893                         return -1;
 894                 }
 895         }
 896
 897         /* force the databases to thaw */
 898         for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 899                 if (ctdb_db_prio_frozen(ctdb, i)) {
 900                         ctdb_control_thaw(ctdb, i, false);
 901                 }
 902         }
 903
 904         if (ctdb->recovery_lock == NULL) {
 905                 /* Not using recovery lock file */
 906                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 907                 ctdb_process_deferred_attach(ctdb);
 908                 return 0;
 909         }
 910
 911         state = talloc_zero(ctdb, struct set_recmode_state);
 912         if (state == NULL) {
 913                 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
 914                 return -1;
 915         }
 916         state->ctdb = ctdb;
 917         state->c = NULL;
 918
 919         h = ctdb_cluster_mutex(state, ctdb, ctdb->recovery_lock, 5,
 920                                set_recmode_handler, state, NULL, NULL);
 921         if (h == NULL) {
 922                 talloc_free(state);
 923                 return -1;
 924         }
 925
 926         state->c = talloc_steal(state, c);
 927         *async_reply = true;
 928
 929         return 0;
 930 }
 931
 932
 933 /*
 934   delete a record as part of the vacuum process
 935   only delete if we are not lmaster or dmaster, and our rsn is <= the provided rsn
 936   use non-blocking locks
 937
 938   return 0 if the record was successfully deleted (i.e. it does not exist
 939   when the function returns)
 940   or !0 is the record still exists in the tdb after returning.
 941  */
 942 static int delete_tdb_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, struct ctdb_rec_data_old *rec)
 943 {
 944         TDB_DATA key, data, data2;
 945         struct ctdb_ltdb_header *hdr, *hdr2;
 946
 947         /* these are really internal tdb functions - but we need them here for
 948            non-blocking lock of the freelist */
 949         int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
 950         int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
 951
 952
 953         key.dsize = rec->keylen;
 954         key.dptr  = &rec->data[0];
 955         data.dsize = rec->datalen;
 956         data.dptr = &rec->data[rec->keylen];
 957
 958         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
 959                 DEBUG(DEBUG_INFO,(__location__ " Called delete on record where we are lmaster\n"));
 960                 return -1;
 961         }
 962
 963         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
 964                 DEBUG(DEBUG_ERR,(__location__ " Bad record size\n"));
 965                 return -1;
 966         }
 967
 968         hdr = (struct ctdb_ltdb_header *)data.dptr;
 969
 970         /* use a non-blocking lock */
 971         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
 972                 return -1;
 973         }
 974
 975         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
 976         if (data2.dptr == NULL) {
 977                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 978                 return 0;
 979         }
 980
 981         if (data2.dsize < sizeof(struct ctdb_ltdb_header)) {
 982                 if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) == 0) {
 983                         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
 984                                 DEBUG(DEBUG_CRIT,(__location__ " Failed to delete corrupt record\n"));
 985                         }
 986                         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
 987                         DEBUG(DEBUG_CRIT,(__location__ " Deleted corrupt record\n"));
 988                 }
 989                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 990                 free(data2.dptr);
 991                 return 0;
 992         }
 993
 994         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
 995
 996         if (hdr2->rsn > hdr->rsn) {
 997                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 998                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with rsn=%llu - called with rsn=%llu\n",
 999                          (unsigned long long)hdr2->rsn, (unsigned long long)hdr->rsn));
1000                 free(data2.dptr);
1001                 return -1;
1002         }
1003
1004         /* do not allow deleting record that have readonly flags set. */
1005         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1006                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1007                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1008                 free(data2.dptr);
1009                 return -1;
1010         }
1011         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1012                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1013                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1014                 free(data2.dptr);
1015                 return -1;
1016         }
1017
1018         if (hdr2->dmaster == ctdb->pnn) {
1019                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1020                 DEBUG(DEBUG_INFO,(__location__ " Attempted delete record where we are the dmaster\n"));
1021                 free(data2.dptr);
1022                 return -1;
1023         }
1024
1025         if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) != 0) {
1026                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1027                 free(data2.dptr);
1028                 return -1;
1029         }
1030
1031         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1032                 tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1033                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1034                 DEBUG(DEBUG_INFO,(__location__ " Failed to delete record\n"));
1035                 free(data2.dptr);
1036                 return -1;
1037         }
1038
1039         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1040         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1041         free(data2.dptr);
1042         return 0;
1043 }
1044
1045
1046
1047 struct recovery_callback_state {
1048         struct ctdb_req_control_old *c;
1049 };
1050
1051
1052 /*
1053   called when the 'recovered' event script has finished
1054  */
1055 static void ctdb_end_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1056 {
1057         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1058
1059         ctdb_enable_monitoring(ctdb);
1060         CTDB_INCREMENT_STAT(ctdb, num_recoveries);
1061
1062         if (status != 0) {
1063                 DEBUG(DEBUG_ERR,(__location__ " recovered event script failed (status %d)\n", status));
1064                 if (status == -ETIME) {
1065                         ctdb_ban_self(ctdb);
1066                 }
1067         }
1068
1069         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1070         talloc_free(state);
1071
1072         gettimeofday(&ctdb->last_recovery_finished, NULL);
1073
1074         if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
1075                 ctdb_set_runstate(ctdb, CTDB_RUNSTATE_STARTUP);
1076         }
1077 }
1078
1079 /*
1080   recovery has finished
1081  */
1082 int32_t ctdb_control_end_recovery(struct ctdb_context *ctdb,
1083                                 struct ctdb_req_control_old *c,
1084                                 bool *async_reply)
1085 {
1086         int ret;
1087         struct recovery_callback_state *state;
1088
1089         DEBUG(DEBUG_NOTICE,("Recovery has finished\n"));
1090
1091         ctdb_persistent_finish_trans3_commits(ctdb);
1092
1093         state = talloc(ctdb, struct recovery_callback_state);
1094         CTDB_NO_MEMORY(ctdb, state);
1095
1096         state->c    = c;
1097
1098         ctdb_disable_monitoring(ctdb);
1099
1100         ret = ctdb_event_script_callback(ctdb, state,
1101                                          ctdb_end_recovery_callback,
1102                                          state,
1103                                          CTDB_EVENT_RECOVERED, "%s", "");
1104
1105         if (ret != 0) {
1106                 ctdb_enable_monitoring(ctdb);
1107
1108                 DEBUG(DEBUG_ERR,(__location__ " Failed to end recovery\n"));
1109                 talloc_free(state);
1110                 return -1;
1111         }
1112
1113         /* tell the control that we will be reply asynchronously */
1114         state->c    = talloc_steal(state, c);
1115         *async_reply = true;
1116         return 0;
1117 }
1118
1119 /*
1120   called when the 'startrecovery' event script has finished
1121  */
1122 static void ctdb_start_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1123 {
1124         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1125
1126         if (status != 0) {
1127                 DEBUG(DEBUG_ERR,(__location__ " startrecovery event script failed (status %d)\n", status));
1128         }
1129
1130         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1131         talloc_free(state);
1132 }
1133
1134 static void run_start_recovery_event(struct ctdb_context *ctdb,
1135                                      struct recovery_callback_state *state)
1136 {
1137         int ret;
1138
1139         ctdb_disable_monitoring(ctdb);
1140
1141         ret = ctdb_event_script_callback(ctdb, state,
1142                                          ctdb_start_recovery_callback,
1143                                          state,
1144                                          CTDB_EVENT_START_RECOVERY,
1145                                          "%s", "");
1146
1147         if (ret != 0) {
1148                 DEBUG(DEBUG_ERR,("Unable to run startrecovery event\n"));
1149                 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
1150                 talloc_free(state);
1151                 return;
1152         }
1153
1154         return;
1155 }
1156
1157 static bool reclock_strings_equal(const char *a, const char *b)
1158 {
1159         return (a == NULL && b == NULL) ||
1160                 (a != NULL && b != NULL && strcmp(a, b) == 0);
1161 }
1162
1163 static void start_recovery_reclock_callback(struct ctdb_context *ctdb,
1164                                                 int32_t status,
1165                                                 TDB_DATA data,
1166                                                 const char *errormsg,
1167                                                 void *private_data)
1168 {
1169         struct recovery_callback_state *state = talloc_get_type_abort(
1170                 private_data, struct recovery_callback_state);
1171         const char *local = ctdb->recovery_lock;
1172         const char *remote = NULL;
1173
1174         if (status != 0) {
1175                 DEBUG(DEBUG_ERR, (__location__ " GET_RECLOCK failed\n"));
1176                 ctdb_request_control_reply(ctdb, state->c, NULL,
1177                                            status, errormsg);
1178                 talloc_free(state);
1179                 return;
1180         }
1181
1182         /* Check reclock consistency */
1183         if (data.dsize > 0) {
1184                 /* Ensure NUL-termination */
1185                 data.dptr[data.dsize-1] = '\0';
1186                 remote = (const char *)data.dptr;
1187         }
1188         if (! reclock_strings_equal(local, remote)) {
1189                 /* Inconsistent */
1190                 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
1191                 DEBUG(DEBUG_ERR,
1192                       ("Recovery lock configuration inconsistent: "
1193                        "recmaster has %s, this node has %s, shutting down\n",
1194                        remote == NULL ? "NULL" : remote,
1195                        local == NULL ? "NULL" : local));
1196                 talloc_free(state);
1197                 ctdb_shutdown_sequence(ctdb, 1);
1198         }
1199         DEBUG(DEBUG_INFO,
1200               ("Recovery lock consistency check successful\n"));
1201
1202         run_start_recovery_event(ctdb, state);
1203 }
1204
1205 /* Check recovery lock consistency and run eventscripts for the
1206  * "startrecovery" event */
1207 int32_t ctdb_control_start_recovery(struct ctdb_context *ctdb,
1208                                     struct ctdb_req_control_old *c,
1209                                     bool *async_reply)
1210 {
1211         int ret;
1212         struct recovery_callback_state *state;
1213         uint32_t recmaster = c->hdr.srcnode;
1214
1215         DEBUG(DEBUG_NOTICE, ("Running startrecovery event\n"));
1216         gettimeofday(&ctdb->last_recovery_started, NULL);
1217
1218         state = talloc(ctdb, struct recovery_callback_state);
1219         CTDB_NO_MEMORY(ctdb, state);
1220
1221         state->c = c;
1222
1223         /* Although the recovery master sent this node a start
1224          * recovery control, this node might still think the recovery
1225          * master is disconnected.  In this case defer the recovery
1226          * lock consistency check. */
1227         if (ctdb->nodes[recmaster]->flags & NODE_FLAGS_DISCONNECTED) {
1228                 run_start_recovery_event(ctdb, state);
1229         } else {
1230                 /* Ask the recovery master about its reclock setting */
1231                 ret = ctdb_daemon_send_control(ctdb,
1232                                                recmaster,
1233                                                0,
1234                                                CTDB_CONTROL_GET_RECLOCK_FILE,
1235                                                0, 0,
1236                                                tdb_null,
1237                                                start_recovery_reclock_callback,
1238                                                state);
1239
1240                 if (ret != 0) {
1241                         DEBUG(DEBUG_ERR, (__location__ " GET_RECLOCK failed\n"));
1242                         talloc_free(state);
1243                         return -1;
1244                 }
1245         }
1246
1247         /* tell the control that we will be reply asynchronously */
1248         state->c = talloc_steal(state, c);
1249         *async_reply = true;
1250
1251         return 0;
1252 }
1253
1254 /*
1255  try to delete all these records as part of the vacuuming process
1256  and return the records we failed to delete
1257 */
1258 int32_t ctdb_control_try_delete_records(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
1259 {
1260         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1261         struct ctdb_db_context *ctdb_db;
1262         int i;
1263         struct ctdb_rec_data_old *rec;
1264         struct ctdb_marshall_buffer *records;
1265
1266         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1267                 DEBUG(DEBUG_ERR,(__location__ " invalid data in try_delete_records\n"));
1268                 return -1;
1269         }
1270
1271         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1272         if (!ctdb_db) {
1273                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
1274                 return -1;
1275         }
1276
1277
1278         DEBUG(DEBUG_DEBUG,("starting try_delete_records of %u records for dbid 0x%x\n",
1279                  reply->count, reply->db_id));
1280
1281
1282         /* create a blob to send back the records we couldnt delete */
1283         records = (struct ctdb_marshall_buffer *)
1284                         talloc_zero_size(outdata,
1285                                     offsetof(struct ctdb_marshall_buffer, data));
1286         if (records == NULL) {
1287                 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
1288                 return -1;
1289         }
1290         records->db_id = ctdb_db->db_id;
1291
1292
1293         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1294         for (i=0;i<reply->count;i++) {
1295                 TDB_DATA key, data;
1296
1297                 key.dptr = &rec->data[0];
1298                 key.dsize = rec->keylen;
1299                 data.dptr = &rec->data[key.dsize];
1300                 data.dsize = rec->datalen;
1301
1302                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1303                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record in indata\n"));
1304                         return -1;
1305                 }
1306
1307                 /* If we cant delete the record we must add it to the reply
1308                    so the lmaster knows it may not purge this record
1309                 */
1310                 if (delete_tdb_record(ctdb, ctdb_db, rec) != 0) {
1311                         size_t old_size;
1312                         struct ctdb_ltdb_header *hdr;
1313
1314                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1315                         data.dptr += sizeof(*hdr);
1316                         data.dsize -= sizeof(*hdr);
1317
1318                         DEBUG(DEBUG_INFO, (__location__ " Failed to vacuum delete record with hash 0x%08x\n", ctdb_hash(&key)));
1319
1320                         old_size = talloc_get_size(records);
1321                         records = talloc_realloc_size(outdata, records, old_size + rec->length);
1322                         if (records == NULL) {
1323                                 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
1324                                 return -1;
1325                         }
1326                         records->count++;
1327                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1328                 }
1329
1330                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1331         }
1332
1333
1334         *outdata = ctdb_marshall_finish(records);
1335
1336         return 0;
1337 }
1338
1339 /**
1340  * Store a record as part of the vacuum process:
1341  * This is called from the RECEIVE_RECORD control which
1342  * the lmaster uses to send the current empty copy
1343  * to all nodes for storing, before it lets the other
1344  * nodes delete the records in the second phase with
1345  * the TRY_DELETE_RECORDS control.
1346  *
1347  * Only store if we are not lmaster or dmaster, and our
1348  * rsn is <= the provided rsn. Use non-blocking locks.
1349  *
1350  * return 0 if the record was successfully stored.
1351  * return !0 if the record still exists in the tdb after returning.
1352  */
1353 static int store_tdb_record(struct ctdb_context *ctdb,
1354                             struct ctdb_db_context *ctdb_db,
1355                             struct ctdb_rec_data_old *rec)
1356 {
1357         TDB_DATA key, data, data2;
1358         struct ctdb_ltdb_header *hdr, *hdr2;
1359         int ret;
1360
1361         key.dsize = rec->keylen;
1362         key.dptr = &rec->data[0];
1363         data.dsize = rec->datalen;
1364         data.dptr = &rec->data[rec->keylen];
1365
1366         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1367                 DEBUG(DEBUG_INFO, (__location__ " Called store_tdb_record "
1368                                    "where we are lmaster\n"));
1369                 return -1;
1370         }
1371
1372         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1373                 DEBUG(DEBUG_ERR, (__location__ " Bad record size\n"));
1374                 return -1;
1375         }
1376
1377         hdr = (struct ctdb_ltdb_header *)data.dptr;
1378
1379         /* use a non-blocking lock */
1380         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1381                 DEBUG(DEBUG_INFO, (__location__ " Failed to lock chain in non-blocking mode\n"));
1382                 return -1;
1383         }
1384
1385         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1386         if (data2.dptr == NULL || data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1387                 if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) == -1) {
1388                         DEBUG(DEBUG_ERR, (__location__ "Failed to store record\n"));
1389                         ret = -1;
1390                         goto done;
1391                 }
1392                 DEBUG(DEBUG_INFO, (__location__ " Stored record\n"));
1393                 ret = 0;
1394                 goto done;
1395         }
1396
1397         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1398
1399         if (hdr2->rsn > hdr->rsn) {
1400                 DEBUG(DEBUG_INFO, (__location__ " Skipping record with "
1401                                    "rsn=%llu - called with rsn=%llu\n",
1402                                    (unsigned long long)hdr2->rsn,
1403                                    (unsigned long long)hdr->rsn));
1404                 ret = -1;
1405                 goto done;
1406         }
1407
1408         /* do not allow vacuuming of records that have readonly flags set. */
1409         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1410                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1411                                   "flags set\n"));
1412                 ret = -1;
1413                 goto done;
1414         }
1415         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1416                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1417                                   "flags set\n"));
1418                 ret = -1;
1419                 goto done;
1420         }
1421
1422         if (hdr2->dmaster == ctdb->pnn) {
1423                 DEBUG(DEBUG_INFO, (__location__ " Attempted to store record "
1424                                    "where we are the dmaster\n"));
1425                 ret = -1;
1426                 goto done;
1427         }
1428
1429         if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) != 0) {
1430                 DEBUG(DEBUG_INFO,(__location__ " Failed to store record\n"));
1431                 ret = -1;
1432                 goto done;
1433         }
1434
1435         ret = 0;
1436
1437 done:
1438         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1439         free(data2.dptr);
1440         return  ret;
1441 }
1442
1443
1444
1445 /**
1446  * Try to store all these records as part of the vacuuming process
1447  * and return the records we failed to store.
1448  */
1449 int32_t ctdb_control_receive_records(struct ctdb_context *ctdb,
1450                                      TDB_DATA indata, TDB_DATA *outdata)
1451 {
1452         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1453         struct ctdb_db_context *ctdb_db;
1454         int i;
1455         struct ctdb_rec_data_old *rec;
1456         struct ctdb_marshall_buffer *records;
1457
1458         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1459                 DEBUG(DEBUG_ERR,
1460                       (__location__ " invalid data in receive_records\n"));
1461                 return -1;
1462         }
1463
1464         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1465         if (!ctdb_db) {
1466                 DEBUG(DEBUG_ERR, (__location__ " Unknown db 0x%08x\n",
1467                                   reply->db_id));
1468                 return -1;
1469         }
1470
1471         DEBUG(DEBUG_DEBUG, ("starting receive_records of %u records for "
1472                             "dbid 0x%x\n", reply->count, reply->db_id));
1473
1474         /* create a blob to send back the records we could not store */
1475         records = (struct ctdb_marshall_buffer *)
1476                         talloc_zero_size(outdata,
1477                                 offsetof(struct ctdb_marshall_buffer, data));
1478         if (records == NULL) {
1479                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1480                 return -1;
1481         }
1482         records->db_id = ctdb_db->db_id;
1483
1484         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1485         for (i=0; i<reply->count; i++) {
1486                 TDB_DATA key, data;
1487
1488                 key.dptr = &rec->data[0];
1489                 key.dsize = rec->keylen;
1490                 data.dptr = &rec->data[key.dsize];
1491                 data.dsize = rec->datalen;
1492
1493                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1494                         DEBUG(DEBUG_CRIT, (__location__ " bad ltdb record "
1495                                            "in indata\n"));
1496                         return -1;
1497                 }
1498
1499                 /*
1500                  * If we can not store the record we must add it to the reply
1501                  * so the lmaster knows it may not purge this record.
1502                  */
1503                 if (store_tdb_record(ctdb, ctdb_db, rec) != 0) {
1504                         size_t old_size;
1505                         struct ctdb_ltdb_header *hdr;
1506
1507                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1508                         data.dptr += sizeof(*hdr);
1509                         data.dsize -= sizeof(*hdr);
1510
1511                         DEBUG(DEBUG_INFO, (__location__ " Failed to store "
1512                                            "record with hash 0x%08x in vacuum "
1513                                            "via RECEIVE_RECORDS\n",
1514                                            ctdb_hash(&key)));
1515
1516                         old_size = talloc_get_size(records);
1517                         records = talloc_realloc_size(outdata, records,
1518                                                       old_size + rec->length);
1519                         if (records == NULL) {
1520                                 DEBUG(DEBUG_ERR, (__location__ " Failed to "
1521                                                   "expand\n"));
1522                                 return -1;
1523                         }
1524                         records->count++;
1525                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1526                 }
1527
1528                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1529         }
1530
1531         *outdata = ctdb_marshall_finish(records);
1532
1533         return 0;
1534 }
1535
1536
1537 /*
1538   report capabilities
1539  */
1540 int32_t ctdb_control_get_capabilities(struct ctdb_context *ctdb, TDB_DATA *outdata)
1541 {
1542         uint32_t *capabilities = NULL;
1543
1544         capabilities = talloc(outdata, uint32_t);
1545         CTDB_NO_MEMORY(ctdb, capabilities);
1546         *capabilities = ctdb->capabilities;
1547
1548         outdata->dsize = sizeof(uint32_t);
1549         outdata->dptr = (uint8_t *)capabilities;
1550
1551         return 0;
1552 }
1553
1554 /* The recovery daemon will ping us at regular intervals.
1555    If we havent been pinged for a while we assume the recovery
1556    daemon is inoperable and we restart.
1557 */
1558 static void ctdb_recd_ping_timeout(struct tevent_context *ev,
1559                                    struct tevent_timer *te,
1560                                    struct timeval t, void *p)
1561 {
1562         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
1563         uint32_t *count = talloc_get_type(ctdb->recd_ping_count, uint32_t);
1564
1565         DEBUG(DEBUG_ERR, ("Recovery daemon ping timeout. Count : %u\n", *count));
1566
1567         if (*count < ctdb->tunable.recd_ping_failcount) {
1568                 (*count)++;
1569                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1570                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1571                                  ctdb_recd_ping_timeout, ctdb);
1572                 return;
1573         }
1574
1575         DEBUG(DEBUG_ERR, ("Final timeout for recovery daemon ping. Restarting recovery daemon. (This can be caused if the cluster filesystem has hung)\n"));
1576
1577         ctdb_stop_recoverd(ctdb);
1578         ctdb_start_recoverd(ctdb);
1579 }
1580
1581 int32_t ctdb_control_recd_ping(struct ctdb_context *ctdb)
1582 {
1583         talloc_free(ctdb->recd_ping_count);
1584
1585         ctdb->recd_ping_count = talloc_zero(ctdb, uint32_t);
1586         CTDB_NO_MEMORY(ctdb, ctdb->recd_ping_count);
1587
1588         if (ctdb->tunable.recd_ping_timeout != 0) {
1589                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1590                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1591                                  ctdb_recd_ping_timeout, ctdb);
1592         }
1593
1594         return 0;
1595 }
1596
1597
1598
1599 int32_t ctdb_control_set_recmaster(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata)
1600 {
1601         uint32_t new_recmaster;
1602
1603         CHECK_CONTROL_DATA_SIZE(sizeof(uint32_t));
1604         new_recmaster = ((uint32_t *)(&indata.dptr[0]))[0];
1605
1606         if (ctdb->pnn != new_recmaster && ctdb->recovery_master == ctdb->pnn) {
1607                 DEBUG(DEBUG_NOTICE,
1608                       ("This node (%u) is no longer the recovery master\n", ctdb->pnn));
1609         }
1610
1611         if (ctdb->pnn == new_recmaster && ctdb->recovery_master != new_recmaster) {
1612                 DEBUG(DEBUG_NOTICE,
1613                       ("This node (%u) is now the recovery master\n", ctdb->pnn));
1614         }
1615
1616         ctdb->recovery_master = new_recmaster;
1617         return 0;
1618 }
1619
1620
1621 int32_t ctdb_control_stop_node(struct ctdb_context *ctdb)
1622 {
1623         DEBUG(DEBUG_NOTICE, ("Stopping node\n"));
1624         ctdb_disable_monitoring(ctdb);
1625         ctdb->nodes[ctdb->pnn]->flags |= NODE_FLAGS_STOPPED;
1626
1627         return 0;
1628 }
1629
1630 int32_t ctdb_control_continue_node(struct ctdb_context *ctdb)
1631 {
1632         DEBUG(DEBUG_NOTICE, ("Continue node\n"));
1633         ctdb->nodes[ctdb->pnn]->flags &= ~NODE_FLAGS_STOPPED;
1634
1635         return 0;
1636 }
1637