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 #include "lib/util/strv.h"
  36 #include "lib/util/strv_util.h"
  37
  38 #include "ctdb_private.h"
  39 #include "ctdb_client.h"
  40
  41 #include "common/system.h"
  42 #include "common/common.h"
  43 #include "common/logging.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 ctdb_cluster_mutex_handle;
 753 typedef void (*cluster_mutex_handler_t) (
 754         struct ctdb_context *ctdb,
 755         char status,
 756         double latency,
 757         struct ctdb_cluster_mutex_handle *h,
 758         void *private_data);
 759
 760 struct ctdb_cluster_mutex_handle {
 761         struct ctdb_context *ctdb;
 762         cluster_mutex_handler_t handler;
 763         void *private_data;
 764         int fd[2];
 765         struct tevent_timer *te;
 766         struct tevent_fd *fde;
 767         pid_t child;
 768         struct timeval start_time;
 769 };
 770
 771 static void set_recmode_handler(struct ctdb_context *ctdb,
 772                                 char status,
 773                                 double latency,
 774                                 struct ctdb_cluster_mutex_handle *h,
 775                                 void *private_data)
 776 {
 777         /* It would be good to use talloc_get_type() here.  However,
 778          * the name of the packet is manually set - not sure why.
 779          * Could use talloc_check_name() but this seems like a lot of
 780          * manual overkill. */
 781         struct ctdb_req_control_old *c =
 782                 (struct ctdb_req_control_old *) private_data;
 783         int s = 0;
 784         const char *err = NULL;
 785
 786         switch (status) {
 787         case '0':
 788                 /* Mutex taken */
 789                 DEBUG(DEBUG_ERR,
 790                       ("ERROR: Daemon able to take recovery lock on \"%s\" during recovery\n",
 791                        ctdb->recovery_lock_file));
 792                 s = -1;
 793                 err = "Took recovery lock from daemon during recovery - probably a cluster filesystem lock coherence problem";
 794                 break;
 795
 796         case '1':
 797                 /* Contention */
 798                 DEBUG(DEBUG_DEBUG, (__location__ " Recovery lock check OK\n"));
 799                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 800                 ctdb_process_deferred_attach(ctdb);
 801
 802                 s = 0;
 803
 804                 CTDB_UPDATE_RECLOCK_LATENCY(ctdb, "daemon reclock",
 805                                             reclock.ctdbd, latency);
 806                 break;
 807
 808         case '2':
 809                 /* Timeout.  Consider this a success, not a failure,
 810                  * as we failed to set the recovery lock which is what
 811                  * we wanted.  This can be caused by the cluster
 812                  * filesystem being very slow to arbitrate locks
 813                  * immediately after a node failure. */
 814                 DEBUG(DEBUG_WARNING,
 815                       (__location__
 816                        "Time out getting recovery lock, allowing recmode set anyway\n"));
 817                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 818                 ctdb_process_deferred_attach(ctdb);
 819
 820                 s = 0;
 821                 break;
 822
 823         default:
 824                 DEBUG(DEBUG_ERR,
 825                       ("Unexpected error when testing recovery lock\n"));
 826                 s = -1;
 827                 err = "Unexpected error when testing recovery lock";
 828         }
 829
 830         ctdb_request_control_reply(ctdb, c, NULL, s, err);
 831         talloc_free(h);
 832 }
 833
 834 /*
 835   called if our set_recmode child times out. this would happen if
 836   ctdb_recovery_lock() would block.
 837  */
 838 static void cluster_mutex_timeout(struct tevent_context *ev,
 839                                   struct tevent_timer *te,
 840                                   struct timeval t, void *private_data)
 841 {
 842         struct ctdb_cluster_mutex_handle *h =
 843                 talloc_get_type(private_data, struct ctdb_cluster_mutex_handle);
 844         double latency = timeval_elapsed(&h->start_time);
 845
 846         if (h->handler != NULL) {
 847                 h->handler(h->ctdb, '2', latency, h, h->private_data);
 848         }
 849 }
 850
 851
 852 /* When the handle is freed it causes any child holding the mutex to
 853  * be killed, thus freeing the mutex */
 854 static int cluster_mutex_destructor(struct ctdb_cluster_mutex_handle *h)
 855 {
 856         if (h->fd[0] != -1) {
 857                 h->fd[0] = -1;
 858         }
 859         ctdb_kill(h->ctdb, h->child, SIGTERM);
 860         return 0;
 861 }
 862
 863 /* this is called when the client process has completed ctdb_recovery_lock()
 864    and has written data back to us through the pipe.
 865 */
 866 static void cluster_mutex_handler(struct tevent_context *ev,
 867                                   struct tevent_fd *fde,
 868                                   uint16_t flags, void *private_data)
 869 {
 870         struct ctdb_cluster_mutex_handle *h=
 871                 talloc_get_type(private_data, struct ctdb_cluster_mutex_handle);
 872         double latency = timeval_elapsed(&h->start_time);
 873         char c = '0';
 874         int ret;
 875
 876         /* Got response from child process so abort timeout */
 877         TALLOC_FREE(h->te);
 878
 879         ret = sys_read(h->fd[0], &c, 1);
 880
 881         /* If the child wrote status then just pass it to the handler.
 882          * If no status was written then this is an unexpected error
 883          * so pass generic error code to handler. */
 884         if (h->handler != NULL) {
 885                 h->handler(h->ctdb, ret == 1 ? c : '3', latency,
 886                            h, h->private_data);
 887         }
 888 }
 889
 890 static void
 891 ctdb_drop_all_ips_event(struct tevent_context *ev, struct tevent_timer *te,
 892                         struct timeval t, void *private_data)
 893 {
 894         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 895
 896         DEBUG(DEBUG_ERR,(__location__ " Been in recovery mode for too long. Dropping all IPS\n"));
 897         talloc_free(ctdb->release_ips_ctx);
 898         ctdb->release_ips_ctx = NULL;
 899
 900         ctdb_release_all_ips(ctdb);
 901 }
 902
 903 static char cluster_mutex_helper[PATH_MAX+1] = "";
 904
 905 static bool cluster_mutex_helper_args(TALLOC_CTX *mem_ctx,
 906                                       const char *argstring, char ***argv)
 907 {
 908         int nargs, i, ret, n;
 909         char **args = NULL;
 910         char *strv = NULL;
 911         char *t = NULL;
 912
 913         ret = strv_split(mem_ctx, &strv, argstring, " \t");
 914         if (ret != 0) {
 915                 DEBUG(DEBUG_ERR,
 916                       ("Unable to parse mutex helper string \"%s\" (%s)\n",
 917                        argstring, strerror(ret)));
 918                 return false;
 919         }
 920         n = strv_count(strv);
 921
 922         args = talloc_array(mem_ctx, char *, n + 2);
 923         if (args == NULL) {
 924                 DEBUG(DEBUG_ERR,(__location__ " out of memory\n"));
 925                 return false;
 926         }
 927
 928         nargs = 0;
 929
 930         if (!ctdb_set_helper("cluster mutex helper",
 931                              cluster_mutex_helper,
 932                              sizeof(cluster_mutex_helper),
 933                              "CTDB_CLUSTER_MUTEX_HELPER",
 934                              CTDB_HELPER_BINDIR, "ctdb_mutex_fcntl_helper")) {
 935                 DEBUG(DEBUG_ERR,("ctdb exiting with error: %s\n",
 936                                  __location__
 937                                  " Unable to set cluster mutex helper\n"));
 938                 exit(1);
 939         }
 940         args[nargs++] = cluster_mutex_helper;
 941
 942         t = NULL;
 943         for (i = 0; i < n; i++) {
 944                 /* Don't copy, just keep cmd_args around */
 945                 t = strv_next(strv, t);
 946                 args[nargs++] = t;
 947         }
 948
 949         /* Make sure last argument is NULL */
 950         args[nargs] = NULL;
 951
 952         *argv = args;
 953         return true;
 954 }
 955
 956 static struct ctdb_cluster_mutex_handle *
 957 ctdb_cluster_mutex(struct ctdb_context *ctdb, int timeout)
 958 {
 959         struct ctdb_cluster_mutex_handle *h;
 960         char **args;
 961         int ret;
 962
 963         h = talloc(ctdb, struct ctdb_cluster_mutex_handle);
 964         if (h == NULL) {
 965                 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
 966                 return NULL;
 967         }
 968
 969         h->start_time = timeval_current();
 970         h->fd[0] = -1;
 971         h->fd[1] = -1;
 972
 973         ret = pipe(h->fd);
 974         if (ret != 0) {
 975                 talloc_free(h);
 976                 DEBUG(DEBUG_ERR, (__location__ " Failed to open pipe\n"));
 977                 return NULL;
 978         }
 979         set_close_on_exec(h->fd[0]);
 980
 981         /* Create arguments for lock helper */
 982         if (!cluster_mutex_helper_args(h, ctdb->recovery_lock_file, &args)) {
 983                 close(h->fd[0]);
 984                 close(h->fd[1]);
 985                 talloc_free(h);
 986                 return NULL;
 987         }
 988
 989         h->child = ctdb_fork(ctdb);
 990         if (h->child == (pid_t)-1) {
 991                 close(h->fd[0]);
 992                 close(h->fd[1]);
 993                 talloc_free(h);
 994                 return NULL;
 995         }
 996
 997         if (h->child == 0) {
 998                 /* Make stdout point to the pipe */
 999                 close(STDOUT_FILENO);
1000                 dup2(h->fd[1], STDOUT_FILENO);
1001                 close(h->fd[1]);
1002
1003                 execv(args[0], args);
1004
1005                 /* Only happens on error */
1006                 DEBUG(DEBUG_ERR, (__location__ "execv() failed\n"));
1007                 _exit(1);
1008         }
1009
1010         /* Parent */
1011
1012         DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d\n", h->fd[0]));
1013         set_close_on_exec(h->fd[0]);
1014
1015         close(h->fd[1]);
1016         h->fd[1] = -1;
1017
1018         talloc_set_destructor(h, cluster_mutex_destructor);
1019
1020         if (timeout != 0) {
1021                 h->te = tevent_add_timer(ctdb->ev, h,
1022                                          timeval_current_ofs(timeout, 0),
1023                                          cluster_mutex_timeout, h);
1024         } else {
1025                 h->te = NULL;
1026         }
1027
1028         h->fde = tevent_add_fd(ctdb->ev, h, h->fd[0], TEVENT_FD_READ,
1029                                cluster_mutex_handler, (void *)h);
1030
1031         if (h->fde == NULL) {
1032                 talloc_free(h);
1033                 return NULL;
1034         }
1035         tevent_fd_set_auto_close(h->fde);
1036
1037         h->ctdb = ctdb;
1038         h->handler = NULL;
1039         h->private_data = NULL;
1040
1041         return h;
1042 }
1043
1044
1045 /*
1046  * Set up an event to drop all public ips if we remain in recovery for too
1047  * long
1048  */
1049 int ctdb_deferred_drop_all_ips(struct ctdb_context *ctdb)
1050 {
1051         if (ctdb->release_ips_ctx != NULL) {
1052                 talloc_free(ctdb->release_ips_ctx);
1053         }
1054         ctdb->release_ips_ctx = talloc_new(ctdb);
1055         CTDB_NO_MEMORY(ctdb, ctdb->release_ips_ctx);
1056
1057         tevent_add_timer(ctdb->ev, ctdb->release_ips_ctx,
1058                          timeval_current_ofs(ctdb->tunable.recovery_drop_all_ips, 0),
1059                          ctdb_drop_all_ips_event, ctdb);
1060         return 0;
1061 }
1062
1063 /*
1064   set the recovery mode
1065  */
1066 int32_t ctdb_control_set_recmode(struct ctdb_context *ctdb,
1067                                  struct ctdb_req_control_old *c,
1068                                  TDB_DATA indata, bool *async_reply,
1069                                  const char **errormsg)
1070 {
1071         uint32_t recmode = *(uint32_t *)indata.dptr;
1072         int i;
1073         struct ctdb_db_context *ctdb_db;
1074         struct ctdb_cluster_mutex_handle *h;
1075
1076         /* if we enter recovery but stay in recovery for too long
1077            we will eventually drop all our ip addresses
1078         */
1079         if (recmode == CTDB_RECOVERY_NORMAL) {
1080                 talloc_free(ctdb->release_ips_ctx);
1081                 ctdb->release_ips_ctx = NULL;
1082         } else {
1083                 if (ctdb_deferred_drop_all_ips(ctdb) != 0) {
1084                         DEBUG(DEBUG_ERR,("Failed to set up deferred drop all ips\n"));
1085                 }
1086         }
1087
1088         if (recmode != ctdb->recovery_mode) {
1089                 DEBUG(DEBUG_NOTICE,(__location__ " Recovery mode set to %s\n",
1090                          recmode==CTDB_RECOVERY_NORMAL?"NORMAL":"ACTIVE"));
1091         }
1092
1093         if (recmode != CTDB_RECOVERY_NORMAL ||
1094             ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
1095                 ctdb->recovery_mode = recmode;
1096                 return 0;
1097         }
1098
1099         /* From this point: recmode == CTDB_RECOVERY_NORMAL
1100          *
1101          * Therefore, what follows is special handling when setting
1102          * recovery mode back to normal */
1103
1104         for (ctdb_db = ctdb->db_list; ctdb_db != NULL; ctdb_db = ctdb_db->next) {
1105                 if (ctdb_db->generation != ctdb->vnn_map->generation) {
1106                         DEBUG(DEBUG_ERR,
1107                               ("Inconsistent DB generation %u for %s\n",
1108                                ctdb_db->generation, ctdb_db->db_name));
1109                         DEBUG(DEBUG_ERR, ("Recovery mode set to ACTIVE\n"));
1110                         return -1;
1111                 }
1112         }
1113
1114         /* force the databases to thaw */
1115         for (i=1; i<=NUM_DB_PRIORITIES; i++) {
1116                 if (ctdb_db_prio_frozen(ctdb, i)) {
1117                         ctdb_control_thaw(ctdb, i, false);
1118                 }
1119         }
1120
1121         /* release any deferred attach calls from clients */
1122         if (recmode == CTDB_RECOVERY_NORMAL) {
1123                 ctdb_process_deferred_attach(ctdb);
1124         }
1125
1126         if (ctdb->recovery_lock_file == NULL) {
1127                 /* Not using recovery lock file */
1128                 ctdb->recovery_mode = recmode;
1129                 return 0;
1130         }
1131
1132         h = ctdb_cluster_mutex(ctdb, 5);
1133         if (h == NULL) {
1134                 return -1;
1135         }
1136
1137         /* set_recmode_handler() frees h */
1138         h->handler = set_recmode_handler;
1139         h->private_data = talloc_steal(h, c);
1140
1141         *async_reply = true;
1142
1143         return 0;
1144 }
1145
1146
1147 bool ctdb_recovery_have_lock(struct ctdb_context *ctdb)
1148 {
1149         return (ctdb->recovery_lock_handle != NULL);
1150 }
1151
1152 struct hold_reclock_state {
1153         bool done;
1154         char status;
1155 };
1156
1157 static void hold_reclock_handler(struct ctdb_context *ctdb,
1158                                  char status,
1159                                  double latency,
1160                                  struct ctdb_cluster_mutex_handle *h,
1161                                  void *private_data)
1162 {
1163         struct hold_reclock_state *s =
1164                 (struct hold_reclock_state *) private_data;
1165
1166         switch (status) {
1167         case '0':
1168                 ctdb->recovery_lock_handle = h;
1169                 break;
1170
1171         case '1':
1172                 DEBUG(DEBUG_ERR,
1173                       ("Unable to take recovery lock - contention\n"));
1174                 talloc_free(h);
1175                 break;
1176
1177         default:
1178                 DEBUG(DEBUG_ERR, ("ERROR: when taking recovery lock\n"));
1179                 talloc_free(h);
1180         }
1181
1182         s->done = true;
1183         s->status = status;
1184 }
1185
1186 bool ctdb_recovery_lock(struct ctdb_context *ctdb)
1187 {
1188         struct ctdb_cluster_mutex_handle *h;
1189         struct hold_reclock_state s = {
1190                 .done = false,
1191                 .status = '0',
1192         };
1193
1194         h = ctdb_cluster_mutex(ctdb, 0);
1195         if (h == NULL) {
1196                 return -1;
1197         }
1198
1199         h->handler = hold_reclock_handler;
1200         h->private_data = &s;
1201
1202         while (!s.done) {
1203                 tevent_loop_once(ctdb->ev);
1204         }
1205
1206         h->private_data = NULL;
1207
1208         return (s.status == '0');
1209 }
1210
1211 void ctdb_recovery_unlock(struct ctdb_context *ctdb)
1212 {
1213         if (ctdb->recovery_lock_handle != NULL) {
1214                 DEBUG(DEBUG_NOTICE, ("Releasing recovery lock\n"));
1215                 TALLOC_FREE(ctdb->recovery_lock_handle);
1216         }
1217 }
1218
1219 /*
1220   delete a record as part of the vacuum process
1221   only delete if we are not lmaster or dmaster, and our rsn is <= the provided rsn
1222   use non-blocking locks
1223
1224   return 0 if the record was successfully deleted (i.e. it does not exist
1225   when the function returns)
1226   or !0 is the record still exists in the tdb after returning.
1227  */
1228 static int delete_tdb_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, struct ctdb_rec_data_old *rec)
1229 {
1230         TDB_DATA key, data, data2;
1231         struct ctdb_ltdb_header *hdr, *hdr2;
1232
1233         /* these are really internal tdb functions - but we need them here for
1234            non-blocking lock of the freelist */
1235         int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
1236         int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
1237
1238
1239         key.dsize = rec->keylen;
1240         key.dptr  = &rec->data[0];
1241         data.dsize = rec->datalen;
1242         data.dptr = &rec->data[rec->keylen];
1243
1244         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1245                 DEBUG(DEBUG_INFO,(__location__ " Called delete on record where we are lmaster\n"));
1246                 return -1;
1247         }
1248
1249         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1250                 DEBUG(DEBUG_ERR,(__location__ " Bad record size\n"));
1251                 return -1;
1252         }
1253
1254         hdr = (struct ctdb_ltdb_header *)data.dptr;
1255
1256         /* use a non-blocking lock */
1257         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1258                 return -1;
1259         }
1260
1261         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1262         if (data2.dptr == NULL) {
1263                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1264                 return 0;
1265         }
1266
1267         if (data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1268                 if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) == 0) {
1269                         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1270                                 DEBUG(DEBUG_CRIT,(__location__ " Failed to delete corrupt record\n"));
1271                         }
1272                         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1273                         DEBUG(DEBUG_CRIT,(__location__ " Deleted corrupt record\n"));
1274                 }
1275                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1276                 free(data2.dptr);
1277                 return 0;
1278         }
1279
1280         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1281
1282         if (hdr2->rsn > hdr->rsn) {
1283                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1284                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with rsn=%llu - called with rsn=%llu\n",
1285                          (unsigned long long)hdr2->rsn, (unsigned long long)hdr->rsn));
1286                 free(data2.dptr);
1287                 return -1;
1288         }
1289
1290         /* do not allow deleting record that have readonly flags set. */
1291         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1292                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1293                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1294                 free(data2.dptr);
1295                 return -1;
1296         }
1297         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1298                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1299                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1300                 free(data2.dptr);
1301                 return -1;
1302         }
1303
1304         if (hdr2->dmaster == ctdb->pnn) {
1305                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1306                 DEBUG(DEBUG_INFO,(__location__ " Attempted delete record where we are the dmaster\n"));
1307                 free(data2.dptr);
1308                 return -1;
1309         }
1310
1311         if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) != 0) {
1312                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1313                 free(data2.dptr);
1314                 return -1;
1315         }
1316
1317         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1318                 tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1319                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1320                 DEBUG(DEBUG_INFO,(__location__ " Failed to delete record\n"));
1321                 free(data2.dptr);
1322                 return -1;
1323         }
1324
1325         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1326         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1327         free(data2.dptr);
1328         return 0;
1329 }
1330
1331
1332
1333 struct recovery_callback_state {
1334         struct ctdb_req_control_old *c;
1335 };
1336
1337
1338 /*
1339   called when the 'recovered' event script has finished
1340  */
1341 static void ctdb_end_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1342 {
1343         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1344
1345         ctdb_enable_monitoring(ctdb);
1346         CTDB_INCREMENT_STAT(ctdb, num_recoveries);
1347
1348         if (status != 0) {
1349                 DEBUG(DEBUG_ERR,(__location__ " recovered event script failed (status %d)\n", status));
1350                 if (status == -ETIME) {
1351                         ctdb_ban_self(ctdb);
1352                 }
1353         }
1354
1355         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1356         talloc_free(state);
1357
1358         gettimeofday(&ctdb->last_recovery_finished, NULL);
1359
1360         if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
1361                 ctdb_set_runstate(ctdb, CTDB_RUNSTATE_STARTUP);
1362         }
1363 }
1364
1365 /*
1366   recovery has finished
1367  */
1368 int32_t ctdb_control_end_recovery(struct ctdb_context *ctdb,
1369                                 struct ctdb_req_control_old *c,
1370                                 bool *async_reply)
1371 {
1372         int ret;
1373         struct recovery_callback_state *state;
1374
1375         DEBUG(DEBUG_NOTICE,("Recovery has finished\n"));
1376
1377         ctdb_persistent_finish_trans3_commits(ctdb);
1378
1379         state = talloc(ctdb, struct recovery_callback_state);
1380         CTDB_NO_MEMORY(ctdb, state);
1381
1382         state->c    = c;
1383
1384         ctdb_disable_monitoring(ctdb);
1385
1386         ret = ctdb_event_script_callback(ctdb, state,
1387                                          ctdb_end_recovery_callback,
1388                                          state,
1389                                          CTDB_EVENT_RECOVERED, "%s", "");
1390
1391         if (ret != 0) {
1392                 ctdb_enable_monitoring(ctdb);
1393
1394                 DEBUG(DEBUG_ERR,(__location__ " Failed to end recovery\n"));
1395                 talloc_free(state);
1396                 return -1;
1397         }
1398
1399         /* tell the control that we will be reply asynchronously */
1400         state->c    = talloc_steal(state, c);
1401         *async_reply = true;
1402         return 0;
1403 }
1404
1405 /*
1406   called when the 'startrecovery' event script has finished
1407  */
1408 static void ctdb_start_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1409 {
1410         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1411
1412         if (status != 0) {
1413                 DEBUG(DEBUG_ERR,(__location__ " startrecovery event script failed (status %d)\n", status));
1414         }
1415
1416         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1417         talloc_free(state);
1418 }
1419
1420 /*
1421   run the startrecovery eventscript
1422  */
1423 int32_t ctdb_control_start_recovery(struct ctdb_context *ctdb,
1424                                 struct ctdb_req_control_old *c,
1425                                 bool *async_reply)
1426 {
1427         int ret;
1428         struct recovery_callback_state *state;
1429
1430         DEBUG(DEBUG_NOTICE,(__location__ " startrecovery eventscript has been invoked\n"));
1431         gettimeofday(&ctdb->last_recovery_started, NULL);
1432
1433         state = talloc(ctdb, struct recovery_callback_state);
1434         CTDB_NO_MEMORY(ctdb, state);
1435
1436         state->c    = talloc_steal(state, c);
1437
1438         ctdb_disable_monitoring(ctdb);
1439
1440         ret = ctdb_event_script_callback(ctdb, state,
1441                                          ctdb_start_recovery_callback,
1442                                          state,
1443                                          CTDB_EVENT_START_RECOVERY,
1444                                          "%s", "");
1445
1446         if (ret != 0) {
1447                 DEBUG(DEBUG_ERR,(__location__ " Failed to start recovery\n"));
1448                 talloc_free(state);
1449                 return -1;
1450         }
1451
1452         /* tell the control that we will be reply asynchronously */
1453         *async_reply = true;
1454         return 0;
1455 }
1456
1457 /*
1458  try to delete all these records as part of the vacuuming process
1459  and return the records we failed to delete
1460 */
1461 int32_t ctdb_control_try_delete_records(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
1462 {
1463         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1464         struct ctdb_db_context *ctdb_db;
1465         int i;
1466         struct ctdb_rec_data_old *rec;
1467         struct ctdb_marshall_buffer *records;
1468
1469         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1470                 DEBUG(DEBUG_ERR,(__location__ " invalid data in try_delete_records\n"));
1471                 return -1;
1472         }
1473
1474         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1475         if (!ctdb_db) {
1476                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
1477                 return -1;
1478         }
1479
1480
1481         DEBUG(DEBUG_DEBUG,("starting try_delete_records of %u records for dbid 0x%x\n",
1482                  reply->count, reply->db_id));
1483
1484
1485         /* create a blob to send back the records we couldnt delete */
1486         records = (struct ctdb_marshall_buffer *)
1487                         talloc_zero_size(outdata,
1488                                     offsetof(struct ctdb_marshall_buffer, data));
1489         if (records == NULL) {
1490                 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
1491                 return -1;
1492         }
1493         records->db_id = ctdb_db->db_id;
1494
1495
1496         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1497         for (i=0;i<reply->count;i++) {
1498                 TDB_DATA key, data;
1499
1500                 key.dptr = &rec->data[0];
1501                 key.dsize = rec->keylen;
1502                 data.dptr = &rec->data[key.dsize];
1503                 data.dsize = rec->datalen;
1504
1505                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1506                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record in indata\n"));
1507                         return -1;
1508                 }
1509
1510                 /* If we cant delete the record we must add it to the reply
1511                    so the lmaster knows it may not purge this record
1512                 */
1513                 if (delete_tdb_record(ctdb, ctdb_db, rec) != 0) {
1514                         size_t old_size;
1515                         struct ctdb_ltdb_header *hdr;
1516
1517                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1518                         data.dptr += sizeof(*hdr);
1519                         data.dsize -= sizeof(*hdr);
1520
1521                         DEBUG(DEBUG_INFO, (__location__ " Failed to vacuum delete record with hash 0x%08x\n", ctdb_hash(&key)));
1522
1523                         old_size = talloc_get_size(records);
1524                         records = talloc_realloc_size(outdata, records, old_size + rec->length);
1525                         if (records == NULL) {
1526                                 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
1527                                 return -1;
1528                         }
1529                         records->count++;
1530                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1531                 }
1532
1533                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1534         }
1535
1536
1537         *outdata = ctdb_marshall_finish(records);
1538
1539         return 0;
1540 }
1541
1542 /**
1543  * Store a record as part of the vacuum process:
1544  * This is called from the RECEIVE_RECORD control which
1545  * the lmaster uses to send the current empty copy
1546  * to all nodes for storing, before it lets the other
1547  * nodes delete the records in the second phase with
1548  * the TRY_DELETE_RECORDS control.
1549  *
1550  * Only store if we are not lmaster or dmaster, and our
1551  * rsn is <= the provided rsn. Use non-blocking locks.
1552  *
1553  * return 0 if the record was successfully stored.
1554  * return !0 if the record still exists in the tdb after returning.
1555  */
1556 static int store_tdb_record(struct ctdb_context *ctdb,
1557                             struct ctdb_db_context *ctdb_db,
1558                             struct ctdb_rec_data_old *rec)
1559 {
1560         TDB_DATA key, data, data2;
1561         struct ctdb_ltdb_header *hdr, *hdr2;
1562         int ret;
1563
1564         key.dsize = rec->keylen;
1565         key.dptr = &rec->data[0];
1566         data.dsize = rec->datalen;
1567         data.dptr = &rec->data[rec->keylen];
1568
1569         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1570                 DEBUG(DEBUG_INFO, (__location__ " Called store_tdb_record "
1571                                    "where we are lmaster\n"));
1572                 return -1;
1573         }
1574
1575         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1576                 DEBUG(DEBUG_ERR, (__location__ " Bad record size\n"));
1577                 return -1;
1578         }
1579
1580         hdr = (struct ctdb_ltdb_header *)data.dptr;
1581
1582         /* use a non-blocking lock */
1583         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1584                 DEBUG(DEBUG_INFO, (__location__ " Failed to lock chain in non-blocking mode\n"));
1585                 return -1;
1586         }
1587
1588         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1589         if (data2.dptr == NULL || data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1590                 if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) == -1) {
1591                         DEBUG(DEBUG_ERR, (__location__ "Failed to store record\n"));
1592                         ret = -1;
1593                         goto done;
1594                 }
1595                 DEBUG(DEBUG_INFO, (__location__ " Stored record\n"));
1596                 ret = 0;
1597                 goto done;
1598         }
1599
1600         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1601
1602         if (hdr2->rsn > hdr->rsn) {
1603                 DEBUG(DEBUG_INFO, (__location__ " Skipping record with "
1604                                    "rsn=%llu - called with rsn=%llu\n",
1605                                    (unsigned long long)hdr2->rsn,
1606                                    (unsigned long long)hdr->rsn));
1607                 ret = -1;
1608                 goto done;
1609         }
1610
1611         /* do not allow vacuuming of records that have readonly flags set. */
1612         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1613                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1614                                   "flags set\n"));
1615                 ret = -1;
1616                 goto done;
1617         }
1618         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1619                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1620                                   "flags set\n"));
1621                 ret = -1;
1622                 goto done;
1623         }
1624
1625         if (hdr2->dmaster == ctdb->pnn) {
1626                 DEBUG(DEBUG_INFO, (__location__ " Attempted to store record "
1627                                    "where we are the dmaster\n"));
1628                 ret = -1;
1629                 goto done;
1630         }
1631
1632         if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) != 0) {
1633                 DEBUG(DEBUG_INFO,(__location__ " Failed to store record\n"));
1634                 ret = -1;
1635                 goto done;
1636         }
1637
1638         ret = 0;
1639
1640 done:
1641         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1642         free(data2.dptr);
1643         return  ret;
1644 }
1645
1646
1647
1648 /**
1649  * Try to store all these records as part of the vacuuming process
1650  * and return the records we failed to store.
1651  */
1652 int32_t ctdb_control_receive_records(struct ctdb_context *ctdb,
1653                                      TDB_DATA indata, TDB_DATA *outdata)
1654 {
1655         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1656         struct ctdb_db_context *ctdb_db;
1657         int i;
1658         struct ctdb_rec_data_old *rec;
1659         struct ctdb_marshall_buffer *records;
1660
1661         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1662                 DEBUG(DEBUG_ERR,
1663                       (__location__ " invalid data in receive_records\n"));
1664                 return -1;
1665         }
1666
1667         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1668         if (!ctdb_db) {
1669                 DEBUG(DEBUG_ERR, (__location__ " Unknown db 0x%08x\n",
1670                                   reply->db_id));
1671                 return -1;
1672         }
1673
1674         DEBUG(DEBUG_DEBUG, ("starting receive_records of %u records for "
1675                             "dbid 0x%x\n", reply->count, reply->db_id));
1676
1677         /* create a blob to send back the records we could not store */
1678         records = (struct ctdb_marshall_buffer *)
1679                         talloc_zero_size(outdata,
1680                                 offsetof(struct ctdb_marshall_buffer, data));
1681         if (records == NULL) {
1682                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1683                 return -1;
1684         }
1685         records->db_id = ctdb_db->db_id;
1686
1687         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1688         for (i=0; i<reply->count; i++) {
1689                 TDB_DATA key, data;
1690
1691                 key.dptr = &rec->data[0];
1692                 key.dsize = rec->keylen;
1693                 data.dptr = &rec->data[key.dsize];
1694                 data.dsize = rec->datalen;
1695
1696                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1697                         DEBUG(DEBUG_CRIT, (__location__ " bad ltdb record "
1698                                            "in indata\n"));
1699                         return -1;
1700                 }
1701
1702                 /*
1703                  * If we can not store the record we must add it to the reply
1704                  * so the lmaster knows it may not purge this record.
1705                  */
1706                 if (store_tdb_record(ctdb, ctdb_db, rec) != 0) {
1707                         size_t old_size;
1708                         struct ctdb_ltdb_header *hdr;
1709
1710                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1711                         data.dptr += sizeof(*hdr);
1712                         data.dsize -= sizeof(*hdr);
1713
1714                         DEBUG(DEBUG_INFO, (__location__ " Failed to store "
1715                                            "record with hash 0x%08x in vacuum "
1716                                            "via RECEIVE_RECORDS\n",
1717                                            ctdb_hash(&key)));
1718
1719                         old_size = talloc_get_size(records);
1720                         records = talloc_realloc_size(outdata, records,
1721                                                       old_size + rec->length);
1722                         if (records == NULL) {
1723                                 DEBUG(DEBUG_ERR, (__location__ " Failed to "
1724                                                   "expand\n"));
1725                                 return -1;
1726                         }
1727                         records->count++;
1728                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1729                 }
1730
1731                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1732         }
1733
1734         *outdata = ctdb_marshall_finish(records);
1735
1736         return 0;
1737 }
1738
1739
1740 /*
1741   report capabilities
1742  */
1743 int32_t ctdb_control_get_capabilities(struct ctdb_context *ctdb, TDB_DATA *outdata)
1744 {
1745         uint32_t *capabilities = NULL;
1746
1747         capabilities = talloc(outdata, uint32_t);
1748         CTDB_NO_MEMORY(ctdb, capabilities);
1749         *capabilities = ctdb->capabilities;
1750
1751         outdata->dsize = sizeof(uint32_t);
1752         outdata->dptr = (uint8_t *)capabilities;
1753
1754         return 0;
1755 }
1756
1757 /* The recovery daemon will ping us at regular intervals.
1758    If we havent been pinged for a while we assume the recovery
1759    daemon is inoperable and we restart.
1760 */
1761 static void ctdb_recd_ping_timeout(struct tevent_context *ev,
1762                                    struct tevent_timer *te,
1763                                    struct timeval t, void *p)
1764 {
1765         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
1766         uint32_t *count = talloc_get_type(ctdb->recd_ping_count, uint32_t);
1767
1768         DEBUG(DEBUG_ERR, ("Recovery daemon ping timeout. Count : %u\n", *count));
1769
1770         if (*count < ctdb->tunable.recd_ping_failcount) {
1771                 (*count)++;
1772                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1773                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1774                                  ctdb_recd_ping_timeout, ctdb);
1775                 return;
1776         }
1777
1778         DEBUG(DEBUG_ERR, ("Final timeout for recovery daemon ping. Restarting recovery daemon. (This can be caused if the cluster filesystem has hung)\n"));
1779
1780         ctdb_stop_recoverd(ctdb);
1781         ctdb_start_recoverd(ctdb);
1782 }
1783
1784 int32_t ctdb_control_recd_ping(struct ctdb_context *ctdb)
1785 {
1786         talloc_free(ctdb->recd_ping_count);
1787
1788         ctdb->recd_ping_count = talloc_zero(ctdb, uint32_t);
1789         CTDB_NO_MEMORY(ctdb, ctdb->recd_ping_count);
1790
1791         if (ctdb->tunable.recd_ping_timeout != 0) {
1792                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1793                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1794                                  ctdb_recd_ping_timeout, ctdb);
1795         }
1796
1797         return 0;
1798 }
1799
1800
1801
1802 int32_t ctdb_control_set_recmaster(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata)
1803 {
1804         uint32_t new_recmaster;
1805
1806         CHECK_CONTROL_DATA_SIZE(sizeof(uint32_t));
1807         new_recmaster = ((uint32_t *)(&indata.dptr[0]))[0];
1808
1809         if (ctdb->pnn != new_recmaster && ctdb->recovery_master == ctdb->pnn) {
1810                 DEBUG(DEBUG_NOTICE,
1811                       ("This node (%u) is no longer the recovery master\n", ctdb->pnn));
1812         }
1813
1814         if (ctdb->pnn == new_recmaster && ctdb->recovery_master != new_recmaster) {
1815                 DEBUG(DEBUG_NOTICE,
1816                       ("This node (%u) is now the recovery master\n", ctdb->pnn));
1817         }
1818
1819         ctdb->recovery_master = new_recmaster;
1820         return 0;
1821 }
1822
1823
1824 int32_t ctdb_control_stop_node(struct ctdb_context *ctdb)
1825 {
1826         DEBUG(DEBUG_NOTICE, ("Stopping node\n"));
1827         ctdb_disable_monitoring(ctdb);
1828         ctdb->nodes[ctdb->pnn]->flags |= NODE_FLAGS_STOPPED;
1829
1830         return 0;
1831 }
1832
1833 int32_t ctdb_control_continue_node(struct ctdb_context *ctdb)
1834 {
1835         DEBUG(DEBUG_NOTICE, ("Continue node\n"));
1836         ctdb->nodes[ctdb->pnn]->flags &= ~NODE_FLAGS_STOPPED;
1837
1838         return 0;
1839 }
1840