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 int
  44 ctdb_control_getvnnmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  45 {
  46         struct ctdb_vnn_map_wire *map;
  47         size_t len;
  48
  49         CHECK_CONTROL_DATA_SIZE(0);
  50
  51         len = offsetof(struct ctdb_vnn_map_wire, map) + sizeof(uint32_t)*ctdb->vnn_map->size;
  52         map = talloc_size(outdata, len);
  53         CTDB_NO_MEMORY(ctdb, map);
  54
  55         map->generation = ctdb->vnn_map->generation;
  56         map->size = ctdb->vnn_map->size;
  57         memcpy(map->map, ctdb->vnn_map->map, sizeof(uint32_t)*map->size);
  58
  59         outdata->dsize = len;
  60         outdata->dptr  = (uint8_t *)map;
  61
  62         return 0;
  63 }
  64
  65 int
  66 ctdb_control_setvnnmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  67 {
  68         struct ctdb_vnn_map_wire *map = (struct ctdb_vnn_map_wire *)indata.dptr;
  69
  70         if (ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
  71                 DEBUG(DEBUG_ERR, ("Attempt to set vnnmap when not in recovery\n"));
  72                 return -1;
  73         }
  74
  75         talloc_free(ctdb->vnn_map);
  76
  77         ctdb->vnn_map = talloc(ctdb, struct ctdb_vnn_map);
  78         CTDB_NO_MEMORY(ctdb, ctdb->vnn_map);
  79
  80         ctdb->vnn_map->generation = map->generation;
  81         ctdb->vnn_map->size       = map->size;
  82         ctdb->vnn_map->map = talloc_array(ctdb->vnn_map, uint32_t, map->size);
  83         CTDB_NO_MEMORY(ctdb, ctdb->vnn_map->map);
  84
  85         memcpy(ctdb->vnn_map->map, map->map, sizeof(uint32_t)*map->size);
  86
  87         return 0;
  88 }
  89
  90 int
  91 ctdb_control_getdbmap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
  92 {
  93         uint32_t i, len;
  94         struct ctdb_db_context *ctdb_db;
  95         struct ctdb_dbid_map_old *dbid_map;
  96
  97         CHECK_CONTROL_DATA_SIZE(0);
  98
  99         len = 0;
 100         for(ctdb_db=ctdb->db_list;ctdb_db;ctdb_db=ctdb_db->next){
 101                 len++;
 102         }
 103
 104
 105         outdata->dsize = offsetof(struct ctdb_dbid_map_old, dbs) + sizeof(dbid_map->dbs[0])*len;
 106         outdata->dptr  = (unsigned char *)talloc_zero_size(outdata, outdata->dsize);
 107         if (!outdata->dptr) {
 108                 DEBUG(DEBUG_ALERT, (__location__ " Failed to allocate dbmap array\n"));
 109                 exit(1);
 110         }
 111
 112         dbid_map = (struct ctdb_dbid_map_old *)outdata->dptr;
 113         dbid_map->num = len;
 114         for (i=0,ctdb_db=ctdb->db_list;ctdb_db;i++,ctdb_db=ctdb_db->next){
 115                 dbid_map->dbs[i].db_id       = ctdb_db->db_id;
 116                 if (ctdb_db->persistent != 0) {
 117                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_PERSISTENT;
 118                 }
 119                 if (ctdb_db->readonly != 0) {
 120                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_READONLY;
 121                 }
 122                 if (ctdb_db->sticky != 0) {
 123                         dbid_map->dbs[i].flags |= CTDB_DB_FLAGS_STICKY;
 124                 }
 125         }
 126
 127         return 0;
 128 }
 129
 130 int
 131 ctdb_control_getnodemap(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata, TDB_DATA *outdata)
 132 {
 133         CHECK_CONTROL_DATA_SIZE(0);
 134
 135         outdata->dptr  = (unsigned char *)ctdb_node_list_to_map(ctdb->nodes,
 136                                                                 ctdb->num_nodes,
 137                                                                 outdata);
 138         if (outdata->dptr == NULL) {
 139                 return -1;
 140         }
 141
 142         outdata->dsize = talloc_get_size(outdata->dptr);
 143
 144         return 0;
 145 }
 146
 147 /*
 148   reload the nodes file
 149 */
 150 int
 151 ctdb_control_reload_nodes_file(struct ctdb_context *ctdb, uint32_t opcode)
 152 {
 153         int i, num_nodes;
 154         TALLOC_CTX *tmp_ctx;
 155         struct ctdb_node **nodes;
 156
 157         tmp_ctx = talloc_new(ctdb);
 158
 159         /* steal the old nodes file for a while */
 160         talloc_steal(tmp_ctx, ctdb->nodes);
 161         nodes = ctdb->nodes;
 162         ctdb->nodes = NULL;
 163         num_nodes = ctdb->num_nodes;
 164         ctdb->num_nodes = 0;
 165
 166         /* load the new nodes file */
 167         ctdb_load_nodes_file(ctdb);
 168
 169         for (i=0; i<ctdb->num_nodes; i++) {
 170                 /* keep any identical pre-existing nodes and connections */
 171                 if ((i < num_nodes) && ctdb_same_address(&ctdb->nodes[i]->address, &nodes[i]->address)) {
 172                         talloc_free(ctdb->nodes[i]);
 173                         ctdb->nodes[i] = talloc_steal(ctdb->nodes, nodes[i]);
 174                         continue;
 175                 }
 176
 177                 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
 178                         continue;
 179                 }
 180
 181                 /* any new or different nodes must be added */
 182                 if (ctdb->methods->add_node(ctdb->nodes[i]) != 0) {
 183                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_node failed at %d\n", i));
 184                         ctdb_fatal(ctdb, "failed to add node. shutting down\n");
 185                 }
 186                 if (ctdb->methods->connect_node(ctdb->nodes[i]) != 0) {
 187                         DEBUG(DEBUG_CRIT, (__location__ " methods->add_connect failed at %d\n", i));
 188                         ctdb_fatal(ctdb, "failed to connect to node. shutting down\n");
 189                 }
 190         }
 191
 192         /* tell the recovery daemon to reaload the nodes file too */
 193         ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELOAD_NODES, tdb_null);
 194
 195         talloc_free(tmp_ctx);
 196
 197         return 0;
 198 }
 199
 200 /*
 201    a traverse function for pulling all relevent records from pulldb
 202  */
 203 struct pulldb_data {
 204         struct ctdb_context *ctdb;
 205         struct ctdb_db_context *ctdb_db;
 206         struct ctdb_marshall_buffer *pulldata;
 207         uint32_t len;
 208         uint32_t allocated_len;
 209         bool failed;
 210 };
 211
 212 static int traverse_pulldb(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *p)
 213 {
 214         struct pulldb_data *params = (struct pulldb_data *)p;
 215         struct ctdb_rec_data_old *rec;
 216         struct ctdb_context *ctdb = params->ctdb;
 217         struct ctdb_db_context *ctdb_db = params->ctdb_db;
 218
 219         /* add the record to the blob */
 220         rec = ctdb_marshall_record(params->pulldata, 0, key, NULL, data);
 221         if (rec == NULL) {
 222                 params->failed = true;
 223                 return -1;
 224         }
 225         if (params->len + rec->length >= params->allocated_len) {
 226                 params->allocated_len = rec->length + params->len + ctdb->tunable.pulldb_preallocation_size;
 227                 params->pulldata = talloc_realloc_size(NULL, params->pulldata, params->allocated_len);
 228         }
 229         if (params->pulldata == NULL) {
 230                 DEBUG(DEBUG_CRIT,(__location__ " Failed to expand pulldb_data to %u\n", rec->length + params->len));
 231                 ctdb_fatal(params->ctdb, "failed to allocate memory for recovery. shutting down\n");
 232         }
 233         params->pulldata->count++;
 234         memcpy(params->len+(uint8_t *)params->pulldata, rec, rec->length);
 235         params->len += rec->length;
 236
 237         if (ctdb->tunable.db_record_size_warn != 0 && rec->length > ctdb->tunable.db_record_size_warn) {
 238                 DEBUG(DEBUG_ERR,("Data record in %s is big. Record size is %d bytes\n", ctdb_db->db_name, (int)rec->length));
 239         }
 240
 241         talloc_free(rec);
 242
 243         return 0;
 244 }
 245
 246 /*
 247   pull a bunch of records from a ltdb, filtering by lmaster
 248  */
 249 int32_t ctdb_control_pull_db(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
 250 {
 251         struct ctdb_pulldb *pull;
 252         struct ctdb_db_context *ctdb_db;
 253         struct pulldb_data params;
 254         struct ctdb_marshall_buffer *reply;
 255
 256         pull = (struct ctdb_pulldb *)indata.dptr;
 257
 258         ctdb_db = find_ctdb_db(ctdb, pull->db_id);
 259         if (!ctdb_db) {
 260                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", pull->db_id));
 261                 return -1;
 262         }
 263
 264         if (!ctdb_db_frozen(ctdb_db)) {
 265                 DEBUG(DEBUG_ERR,
 266                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 267                 return -1;
 268         }
 269
 270         reply = talloc_zero(outdata, struct ctdb_marshall_buffer);
 271         CTDB_NO_MEMORY(ctdb, reply);
 272
 273         reply->db_id = pull->db_id;
 274
 275         params.ctdb = ctdb;
 276         params.ctdb_db = ctdb_db;
 277         params.pulldata = reply;
 278         params.len = offsetof(struct ctdb_marshall_buffer, data);
 279         params.allocated_len = params.len;
 280         params.failed = false;
 281
 282         if (ctdb_db->unhealthy_reason) {
 283                 /* this is just a warning, as the tdb should be empty anyway */
 284                 DEBUG(DEBUG_WARNING,("db(%s) unhealty in ctdb_control_pull_db: %s\n",
 285                                      ctdb_db->db_name, ctdb_db->unhealthy_reason));
 286         }
 287
 288         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 289                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 290                 return -1;
 291         }
 292
 293         if (tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_pulldb, &params) == -1) {
 294                 DEBUG(DEBUG_ERR,(__location__ " Failed to get traverse db '%s'\n", ctdb_db->db_name));
 295                 ctdb_lockdb_unmark(ctdb_db);
 296                 talloc_free(params.pulldata);
 297                 return -1;
 298         }
 299
 300         ctdb_lockdb_unmark(ctdb_db);
 301
 302         outdata->dptr = (uint8_t *)params.pulldata;
 303         outdata->dsize = params.len;
 304
 305         if (ctdb->tunable.db_record_count_warn != 0 && params.pulldata->count > ctdb->tunable.db_record_count_warn) {
 306                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d records\n", ctdb_db->db_name, params.pulldata->count));
 307         }
 308         if (ctdb->tunable.db_size_warn != 0 && outdata->dsize > ctdb->tunable.db_size_warn) {
 309                 DEBUG(DEBUG_ERR,("Database %s is big. Contains %d bytes\n", ctdb_db->db_name, (int)outdata->dsize));
 310         }
 311
 312
 313         return 0;
 314 }
 315
 316 struct db_pull_state {
 317         struct ctdb_context *ctdb;
 318         struct ctdb_db_context *ctdb_db;
 319         struct ctdb_marshall_buffer *recs;
 320         uint32_t pnn;
 321         uint64_t srvid;
 322         uint32_t num_records;
 323 };
 324
 325 static int traverse_db_pull(struct tdb_context *tdb, TDB_DATA key,
 326                             TDB_DATA data, void *private_data)
 327 {
 328         struct db_pull_state *state = (struct db_pull_state *)private_data;
 329         struct ctdb_marshall_buffer *recs;
 330
 331         recs = ctdb_marshall_add(state->ctdb, state->recs,
 332                                  state->ctdb_db->db_id, 0, key, NULL, data);
 333         if (recs == NULL) {
 334                 TALLOC_FREE(state->recs);
 335                 return -1;
 336         }
 337         state->recs = recs;
 338
 339         if (talloc_get_size(state->recs) >=
 340                         state->ctdb->tunable.rec_buffer_size_limit) {
 341                 TDB_DATA buffer;
 342                 int ret;
 343
 344                 buffer = ctdb_marshall_finish(state->recs);
 345                 ret = ctdb_daemon_send_message(state->ctdb, state->pnn,
 346                                                state->srvid, buffer);
 347                 if (ret != 0) {
 348                         TALLOC_FREE(state->recs);
 349                         return -1;
 350                 }
 351
 352                 state->num_records += state->recs->count;
 353                 TALLOC_FREE(state->recs);
 354         }
 355
 356         return 0;
 357 }
 358
 359 int32_t ctdb_control_db_pull(struct ctdb_context *ctdb,
 360                              struct ctdb_req_control_old *c,
 361                              TDB_DATA indata, TDB_DATA *outdata)
 362 {
 363         struct ctdb_pulldb_ext *pulldb_ext;
 364         struct ctdb_db_context *ctdb_db;
 365         struct db_pull_state state;
 366         int ret;
 367
 368         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 369
 370         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 371         if (ctdb_db == NULL) {
 372                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n",
 373                                  pulldb_ext->db_id));
 374                 return -1;
 375         }
 376
 377         if (!ctdb_db_frozen(ctdb_db)) {
 378                 DEBUG(DEBUG_ERR,
 379                       ("rejecting ctdb_control_pull_db when not frozen\n"));
 380                 return -1;
 381         }
 382
 383         if (ctdb_db->unhealthy_reason) {
 384                 /* this is just a warning, as the tdb should be empty anyway */
 385                 DEBUG(DEBUG_WARNING,
 386                       ("db(%s) unhealty in ctdb_control_db_pull: %s\n",
 387                        ctdb_db->db_name, ctdb_db->unhealthy_reason));
 388         }
 389
 390         state.ctdb = ctdb;
 391         state.ctdb_db = ctdb_db;
 392         state.recs = NULL;
 393         state.pnn = c->hdr.srcnode;
 394         state.srvid = pulldb_ext->srvid;
 395         state.num_records = 0;
 396
 397         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 398                 DEBUG(DEBUG_ERR,
 399                       (__location__ " Failed to get lock on entire db - failing\n"));
 400                 return -1;
 401         }
 402
 403         ret = tdb_traverse_read(ctdb_db->ltdb->tdb, traverse_db_pull, &state);
 404         if (ret == -1) {
 405                 DEBUG(DEBUG_ERR,
 406                       (__location__ " Failed to get traverse db '%s'\n",
 407                        ctdb_db->db_name));
 408                 ctdb_lockdb_unmark(ctdb_db);
 409                 return -1;
 410         }
 411
 412         /* Last few records */
 413         if (state.recs != NULL) {
 414                 TDB_DATA buffer;
 415
 416                 buffer = ctdb_marshall_finish(state.recs);
 417                 ret = ctdb_daemon_send_message(state.ctdb, state.pnn,
 418                                                state.srvid, buffer);
 419                 if (ret != 0) {
 420                         TALLOC_FREE(state.recs);
 421                         ctdb_lockdb_unmark(ctdb_db);
 422                         return -1;
 423                 }
 424
 425                 state.num_records += state.recs->count;
 426                 TALLOC_FREE(state.recs);
 427         }
 428
 429         ctdb_lockdb_unmark(ctdb_db);
 430
 431         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 432         if (outdata->dptr == NULL) {
 433                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 434                 return -1;
 435         }
 436
 437         memcpy(outdata->dptr, (uint8_t *)&state.num_records, sizeof(uint32_t));
 438         outdata->dsize = sizeof(uint32_t);
 439
 440         return 0;
 441 }
 442
 443 /*
 444   push a bunch of records into a ltdb, filtering by rsn
 445  */
 446 int32_t ctdb_control_push_db(struct ctdb_context *ctdb, TDB_DATA indata)
 447 {
 448         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
 449         struct ctdb_db_context *ctdb_db;
 450         int i, ret;
 451         struct ctdb_rec_data_old *rec;
 452
 453         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
 454                 DEBUG(DEBUG_ERR,(__location__ " invalid data in pulldb reply\n"));
 455                 return -1;
 456         }
 457
 458         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
 459         if (!ctdb_db) {
 460                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
 461                 return -1;
 462         }
 463
 464         if (!ctdb_db_frozen(ctdb_db)) {
 465                 DEBUG(DEBUG_ERR,
 466                       ("rejecting ctdb_control_push_db when not frozen\n"));
 467                 return -1;
 468         }
 469
 470         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 471                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock on entire db - failing\n"));
 472                 return -1;
 473         }
 474
 475         rec = (struct ctdb_rec_data_old *)&reply->data[0];
 476
 477         DEBUG(DEBUG_INFO,("starting push of %u records for dbid 0x%x\n",
 478                  reply->count, reply->db_id));
 479
 480         for (i=0;i<reply->count;i++) {
 481                 TDB_DATA key, data;
 482                 struct ctdb_ltdb_header *hdr;
 483
 484                 key.dptr = &rec->data[0];
 485                 key.dsize = rec->keylen;
 486                 data.dptr = &rec->data[key.dsize];
 487                 data.dsize = rec->datalen;
 488
 489                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 490                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 491                         goto failed;
 492                 }
 493                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 494                 /* strip off any read only record flags. All readonly records
 495                    are revoked implicitely by a recovery
 496                 */
 497                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 498
 499                 data.dptr += sizeof(*hdr);
 500                 data.dsize -= sizeof(*hdr);
 501
 502                 ret = ctdb_ltdb_store(ctdb_db, key, hdr, data);
 503                 if (ret != 0) {
 504                         DEBUG(DEBUG_CRIT, (__location__ " Unable to store record\n"));
 505                         goto failed;
 506                 }
 507
 508                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 509         }
 510
 511         DEBUG(DEBUG_DEBUG,("finished push of %u records for dbid 0x%x\n",
 512                  reply->count, reply->db_id));
 513
 514         if (ctdb_db->readonly) {
 515                 DEBUG(DEBUG_CRIT,("Clearing the tracking database for dbid 0x%x\n",
 516                                   ctdb_db->db_id));
 517                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 518                         DEBUG(DEBUG_ERR,("Failed to wipe tracking database for 0x%x. Dropping read-only delegation support\n", ctdb_db->db_id));
 519                         ctdb_db->readonly = false;
 520                         tdb_close(ctdb_db->rottdb);
 521                         ctdb_db->rottdb = NULL;
 522                         ctdb_db->readonly = false;
 523                 }
 524                 while (ctdb_db->revokechild_active != NULL) {
 525                         talloc_free(ctdb_db->revokechild_active);
 526                 }
 527         }
 528
 529         ctdb_lockdb_unmark(ctdb_db);
 530         return 0;
 531
 532 failed:
 533         ctdb_lockdb_unmark(ctdb_db);
 534         return -1;
 535 }
 536
 537 struct db_push_state {
 538         struct ctdb_context *ctdb;
 539         struct ctdb_db_context *ctdb_db;
 540         uint64_t srvid;
 541         uint32_t num_records;
 542         bool failed;
 543 };
 544
 545 static void db_push_msg_handler(uint64_t srvid, TDB_DATA indata,
 546                                 void *private_data)
 547 {
 548         struct db_push_state *state = talloc_get_type(
 549                 private_data, struct db_push_state);
 550         struct ctdb_marshall_buffer *recs;
 551         struct ctdb_rec_data_old *rec;
 552         int i, ret;
 553
 554         if (state->failed) {
 555                 return;
 556         }
 557
 558         recs = (struct ctdb_marshall_buffer *)indata.dptr;
 559         rec = (struct ctdb_rec_data_old *)&recs->data[0];
 560
 561         DEBUG(DEBUG_INFO, ("starting push of %u records for dbid 0x%x\n",
 562                            recs->count, recs->db_id));
 563
 564         for (i=0; i<recs->count; i++) {
 565                 TDB_DATA key, data;
 566                 struct ctdb_ltdb_header *hdr;
 567
 568                 key.dptr = &rec->data[0];
 569                 key.dsize = rec->keylen;
 570                 data.dptr = &rec->data[key.dsize];
 571                 data.dsize = rec->datalen;
 572
 573                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
 574                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
 575                         goto failed;
 576                 }
 577
 578                 hdr = (struct ctdb_ltdb_header *)data.dptr;
 579                 /* Strip off any read only record flags.
 580                  * All readonly records are revoked implicitely by a recovery.
 581                  */
 582                 hdr->flags &= ~CTDB_REC_RO_FLAGS;
 583
 584                 data.dptr += sizeof(*hdr);
 585                 data.dsize -= sizeof(*hdr);
 586
 587                 ret = ctdb_ltdb_store(state->ctdb_db, key, hdr, data);
 588                 if (ret != 0) {
 589                         DEBUG(DEBUG_ERR,
 590                               (__location__ " Unable to store record\n"));
 591                         goto failed;
 592                 }
 593
 594                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
 595         }
 596
 597         DEBUG(DEBUG_DEBUG, ("finished push of %u records for dbid 0x%x\n",
 598                             recs->count, recs->db_id));
 599
 600         state->num_records += recs->count;
 601         return;
 602
 603 failed:
 604         state->failed = true;
 605 }
 606
 607 int32_t ctdb_control_db_push_start(struct ctdb_context *ctdb, TDB_DATA indata)
 608 {
 609         struct ctdb_pulldb_ext *pulldb_ext;
 610         struct ctdb_db_context *ctdb_db;
 611         struct db_push_state *state;
 612         int ret;
 613
 614         pulldb_ext = (struct ctdb_pulldb_ext *)indata.dptr;
 615
 616         ctdb_db = find_ctdb_db(ctdb, pulldb_ext->db_id);
 617         if (ctdb_db == NULL) {
 618                 DEBUG(DEBUG_ERR,
 619                       (__location__ " Unknown db 0x%08x\n", pulldb_ext->db_id));
 620                 return -1;
 621         }
 622
 623         if (!ctdb_db_frozen(ctdb_db)) {
 624                 DEBUG(DEBUG_ERR,
 625                       ("rejecting ctdb_control_db_push_start when not frozen\n"));
 626                 return -1;
 627         }
 628
 629         if (ctdb_db->push_started) {
 630                 DEBUG(DEBUG_WARNING,
 631                       (__location__ " DB push already started for %s\n",
 632                        ctdb_db->db_name));
 633
 634                 /* De-register old state */
 635                 state = (struct db_push_state *)ctdb_db->push_state;
 636                 if (state != NULL) {
 637                         srvid_deregister(ctdb->srv, state->srvid, state);
 638                         talloc_free(state);
 639                         ctdb_db->push_state = NULL;
 640                 }
 641         }
 642
 643         state = talloc_zero(ctdb_db, struct db_push_state);
 644         if (state == NULL) {
 645                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 646                 return -1;
 647         }
 648
 649         state->ctdb = ctdb;
 650         state->ctdb_db = ctdb_db;
 651         state->srvid = pulldb_ext->srvid;
 652         state->failed = false;
 653
 654         ret = srvid_register(ctdb->srv, state, state->srvid,
 655                              db_push_msg_handler, state);
 656         if (ret != 0) {
 657                 DEBUG(DEBUG_ERR,
 658                       (__location__ " Failed to register srvid for db push\n"));
 659                 talloc_free(state);
 660                 return -1;
 661         }
 662
 663         if (ctdb_lockdb_mark(ctdb_db) != 0) {
 664                 DEBUG(DEBUG_ERR,
 665                       (__location__ " Failed to get lock on entire db - failing\n"));
 666                 srvid_deregister(ctdb->srv, state->srvid, state);
 667                 talloc_free(state);
 668                 return -1;
 669         }
 670
 671         ctdb_db->push_started = true;
 672         ctdb_db->push_state = state;
 673
 674         return 0;
 675 }
 676
 677 int32_t ctdb_control_db_push_confirm(struct ctdb_context *ctdb,
 678                                      TDB_DATA indata, TDB_DATA *outdata)
 679 {
 680         uint32_t db_id;
 681         struct ctdb_db_context *ctdb_db;
 682         struct db_push_state *state;
 683
 684         db_id = *(uint32_t *)indata.dptr;
 685
 686         ctdb_db = find_ctdb_db(ctdb, db_id);
 687         if (ctdb_db == NULL) {
 688                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", db_id));
 689                 return -1;
 690         }
 691
 692         if (!ctdb_db_frozen(ctdb_db)) {
 693                 DEBUG(DEBUG_ERR,
 694                       ("rejecting ctdb_control_db_push_confirm when not frozen\n"));
 695                 return -1;
 696         }
 697
 698         if (!ctdb_db->push_started) {
 699                 DEBUG(DEBUG_ERR, (__location__ " DB push not started\n"));
 700                 return -1;
 701         }
 702
 703         if (ctdb_db->readonly) {
 704                 DEBUG(DEBUG_ERR,
 705                       ("Clearing the tracking database for dbid 0x%x\n",
 706                        ctdb_db->db_id));
 707                 if (tdb_wipe_all(ctdb_db->rottdb) != 0) {
 708                         DEBUG(DEBUG_ERR,
 709                               ("Failed to wipe tracking database for 0x%x."
 710                                " Dropping read-only delegation support\n",
 711                                ctdb_db->db_id));
 712                         ctdb_db->readonly = false;
 713                         tdb_close(ctdb_db->rottdb);
 714                         ctdb_db->rottdb = NULL;
 715                         ctdb_db->readonly = false;
 716                 }
 717
 718                 while (ctdb_db->revokechild_active != NULL) {
 719                         talloc_free(ctdb_db->revokechild_active);
 720                 }
 721         }
 722
 723         ctdb_lockdb_unmark(ctdb_db);
 724
 725         state = (struct db_push_state *)ctdb_db->push_state;
 726         if (state == NULL) {
 727                 DEBUG(DEBUG_ERR, (__location__ " Missing push db state\n"));
 728                 return -1;
 729         }
 730
 731         srvid_deregister(ctdb->srv, state->srvid, state);
 732
 733         outdata->dptr = talloc_size(outdata, sizeof(uint32_t));
 734         if (outdata->dptr == NULL) {
 735                 DEBUG(DEBUG_ERR, (__location__ " Memory allocation error\n"));
 736                 talloc_free(state);
 737                 ctdb_db->push_state = NULL;
 738                 return -1;
 739         }
 740
 741         memcpy(outdata->dptr, (uint8_t *)&state->num_records, sizeof(uint32_t));
 742         outdata->dsize = sizeof(uint32_t);
 743
 744         talloc_free(state);
 745         ctdb_db->push_state = NULL;
 746
 747         return 0;
 748 }
 749
 750 struct ctdb_cluster_mutex_handle;
 751 typedef void (*cluster_mutex_handler_t) (
 752         struct ctdb_context *ctdb,
 753         char status,
 754         double latency,
 755         struct ctdb_cluster_mutex_handle *h,
 756         void *private_data);
 757
 758 struct ctdb_cluster_mutex_handle {
 759         struct ctdb_context *ctdb;
 760         cluster_mutex_handler_t handler;
 761         void *private_data;
 762         int fd[2];
 763         struct tevent_timer *te;
 764         struct tevent_fd *fde;
 765         pid_t child;
 766         struct timeval start_time;
 767 };
 768
 769 static void set_recmode_handler(struct ctdb_context *ctdb,
 770                                 char status,
 771                                 double latency,
 772                                 struct ctdb_cluster_mutex_handle *h,
 773                                 void *private_data)
 774 {
 775         /* It would be good to use talloc_get_type() here.  However,
 776          * the name of the packet is manually set - not sure why.
 777          * Could use talloc_check_name() but this seems like a lot of
 778          * manual overkill. */
 779         struct ctdb_req_control_old *c =
 780                 (struct ctdb_req_control_old *) private_data;
 781         int s = 0;
 782         const char *err = NULL;
 783
 784         switch (status) {
 785         case '0':
 786                 /* Mutex taken */
 787                 DEBUG(DEBUG_ERR,
 788                       ("ERROR: Daemon able to take recovery lock on \"%s\" during recovery\n",
 789                        ctdb->recovery_lock_file));
 790                 s = -1;
 791                 err = "Took recovery lock from daemon during recovery - probably a cluster filesystem lock coherence problem";
 792                 break;
 793
 794         case '1':
 795                 /* Contention */
 796                 DEBUG(DEBUG_DEBUG, (__location__ " Recovery lock check OK\n"));
 797                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 798                 ctdb_process_deferred_attach(ctdb);
 799
 800                 s = 0;
 801
 802                 CTDB_UPDATE_RECLOCK_LATENCY(ctdb, "daemon reclock",
 803                                             reclock.ctdbd, latency);
 804                 break;
 805
 806         case '2':
 807                 /* Timeout.  Consider this a success, not a failure,
 808                  * as we failed to set the recovery lock which is what
 809                  * we wanted.  This can be caused by the cluster
 810                  * filesystem being very slow to arbitrate locks
 811                  * immediately after a node failure. */
 812                 DEBUG(DEBUG_WARNING,
 813                       (__location__
 814                        "Time out getting recovery lock, allowing recmode set anyway\n"));
 815                 ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 816                 ctdb_process_deferred_attach(ctdb);
 817
 818                 s = 0;
 819                 break;
 820
 821         default:
 822                 DEBUG(DEBUG_ERR,
 823                       ("Unexpected error when testing recovery lock\n"));
 824                 s = -1;
 825                 err = "Unexpected error when testing recovery lock";
 826         }
 827
 828         ctdb_request_control_reply(ctdb, c, NULL, s, err);
 829         talloc_free(h);
 830 }
 831
 832 /*
 833   called if our set_recmode child times out. this would happen if
 834   ctdb_recovery_lock() would block.
 835  */
 836 static void cluster_mutex_timeout(struct tevent_context *ev,
 837                                   struct tevent_timer *te,
 838                                   struct timeval t, void *private_data)
 839 {
 840         struct ctdb_cluster_mutex_handle *h =
 841                 talloc_get_type(private_data, struct ctdb_cluster_mutex_handle);
 842         double latency = timeval_elapsed(&h->start_time);
 843
 844         if (h->handler != NULL) {
 845                 h->handler(h->ctdb, '2', latency, h, h->private_data);
 846         }
 847 }
 848
 849
 850 /* When the handle is freed it causes any child holding the mutex to
 851  * be killed, thus freeing the mutex */
 852 static int cluster_mutex_destructor(struct ctdb_cluster_mutex_handle *h)
 853 {
 854         if (h->fd[0] != -1) {
 855                 h->fd[0] = -1;
 856         }
 857         ctdb_kill(h->ctdb, h->child, SIGKILL);
 858         return 0;
 859 }
 860
 861 /* this is called when the client process has completed ctdb_recovery_lock()
 862    and has written data back to us through the pipe.
 863 */
 864 static void cluster_mutex_handler(struct tevent_context *ev,
 865                                   struct tevent_fd *fde,
 866                                   uint16_t flags, void *private_data)
 867 {
 868         struct ctdb_cluster_mutex_handle *h=
 869                 talloc_get_type(private_data, struct ctdb_cluster_mutex_handle);
 870         double latency = timeval_elapsed(&h->start_time);
 871         char c = '0';
 872         int ret;
 873
 874         /* Got response from child process so abort timeout */
 875         TALLOC_FREE(h->te);
 876
 877         ret = sys_read(h->fd[0], &c, 1);
 878
 879         /* If the child wrote status then just pass it to the handler.
 880          * If no status was written then this is an unexpected error
 881          * so pass generic error code to handler. */
 882         if (h->handler != NULL) {
 883                 h->handler(h->ctdb, ret == 1 ? c : '3', latency,
 884                            h, h->private_data);
 885         }
 886 }
 887
 888 static void
 889 ctdb_drop_all_ips_event(struct tevent_context *ev, struct tevent_timer *te,
 890                         struct timeval t, void *private_data)
 891 {
 892         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 893
 894         DEBUG(DEBUG_ERR,(__location__ " Been in recovery mode for too long. Dropping all IPS\n"));
 895         talloc_free(ctdb->release_ips_ctx);
 896         ctdb->release_ips_ctx = NULL;
 897
 898         ctdb_release_all_ips(ctdb);
 899 }
 900
 901 static char cluster_mutex_helper[PATH_MAX+1] = "";
 902
 903 static bool cluster_mutex_helper_args(TALLOC_CTX *mem_ctx,
 904                                       const char *lockfile, char ***argv)
 905 {
 906         int nargs;
 907         char **args = NULL;
 908
 909         /* Anticipate the size of the array.  Given that lock file is
 910          * really now some arbitrary arguments to a configuration
 911          * helper, it really needs to be parsed... but not yet. */
 912         args = talloc_array(mem_ctx, char *, 3);
 913         if (args == NULL) {
 914                 DEBUG(DEBUG_ERR,(__location__ " out of memory\n"));
 915                 return false;
 916         }
 917
 918         nargs = 0;
 919
 920         if (!ctdb_set_helper("cluster mutex helper",
 921                              cluster_mutex_helper,
 922                              sizeof(cluster_mutex_helper),
 923                              "CTDB_CLUSTER_MUTEX_HELPER",
 924                              CTDB_HELPER_BINDIR, "ctdb_mutex_fcntl_helper")) {
 925                 DEBUG(DEBUG_ERR,("ctdb exiting with error: %s\n",
 926                                  __location__
 927                                  " Unable to set cluster mutex helper\n"));
 928                 exit(1);
 929         }
 930         args[nargs++] = cluster_mutex_helper;
 931
 932         args[nargs] = talloc_strdup(args, lockfile);
 933         if (args[nargs] == NULL) {
 934                 talloc_free(args);
 935                 DEBUG(DEBUG_ERR,(__location__ " out of memory\n"));
 936                 return false;
 937         }
 938         nargs++;
 939
 940         /* Make sure last argument is NULL */
 941         args[nargs] = NULL;
 942
 943         *argv = args;
 944         return true;
 945 }
 946
 947 static struct ctdb_cluster_mutex_handle *
 948 ctdb_cluster_mutex(struct ctdb_context *ctdb, int timeout)
 949 {
 950         struct ctdb_cluster_mutex_handle *h;
 951         char **args;
 952         int ret;
 953
 954         h = talloc(ctdb, struct ctdb_cluster_mutex_handle);
 955         if (h == NULL) {
 956                 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
 957                 return NULL;
 958         }
 959
 960         h->start_time = timeval_current();
 961         h->fd[0] = -1;
 962         h->fd[1] = -1;
 963
 964         ret = pipe(h->fd);
 965         if (ret != 0) {
 966                 talloc_free(h);
 967                 DEBUG(DEBUG_ERR, (__location__ " Failed to open pipe\n"));
 968                 return NULL;
 969         }
 970         set_close_on_exec(h->fd[0]);
 971
 972         /* Create arguments for lock helper */
 973         if (!cluster_mutex_helper_args(h, ctdb->recovery_lock_file, &args)) {
 974                 close(h->fd[0]);
 975                 close(h->fd[1]);
 976                 talloc_free(h);
 977                 return NULL;
 978         }
 979
 980         h->child = ctdb_fork(ctdb);
 981         if (h->child == (pid_t)-1) {
 982                 close(h->fd[0]);
 983                 close(h->fd[1]);
 984                 talloc_free(h);
 985                 return NULL;
 986         }
 987
 988         if (h->child == 0) {
 989                 /* Make stdout point to the pipe */
 990                 close(STDOUT_FILENO);
 991                 dup2(h->fd[1], STDOUT_FILENO);
 992                 close(h->fd[1]);
 993
 994                 execv(args[0], args);
 995
 996                 /* Only happens on error */
 997                 DEBUG(DEBUG_ERR, (__location__ "execv() failed\n"));
 998                 _exit(1);
 999         }
1000
1001         /* Parent */
1002
1003         DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d\n", h->fd[0]));
1004         set_close_on_exec(h->fd[0]);
1005
1006         close(h->fd[1]);
1007         h->fd[1] = -1;
1008
1009         talloc_set_destructor(h, cluster_mutex_destructor);
1010
1011         if (timeout != 0) {
1012                 h->te = tevent_add_timer(ctdb->ev, h,
1013                                          timeval_current_ofs(timeout, 0),
1014                                          cluster_mutex_timeout, h);
1015         } else {
1016                 h->te = NULL;
1017         }
1018
1019         h->fde = tevent_add_fd(ctdb->ev, h, h->fd[0], TEVENT_FD_READ,
1020                                cluster_mutex_handler, (void *)h);
1021
1022         if (h->fde == NULL) {
1023                 talloc_free(h);
1024                 return NULL;
1025         }
1026         tevent_fd_set_auto_close(h->fde);
1027
1028         h->ctdb = ctdb;
1029         h->handler = NULL;
1030         h->private_data = NULL;
1031
1032         return h;
1033 }
1034
1035
1036 /*
1037  * Set up an event to drop all public ips if we remain in recovery for too
1038  * long
1039  */
1040 int ctdb_deferred_drop_all_ips(struct ctdb_context *ctdb)
1041 {
1042         if (ctdb->release_ips_ctx != NULL) {
1043                 talloc_free(ctdb->release_ips_ctx);
1044         }
1045         ctdb->release_ips_ctx = talloc_new(ctdb);
1046         CTDB_NO_MEMORY(ctdb, ctdb->release_ips_ctx);
1047
1048         tevent_add_timer(ctdb->ev, ctdb->release_ips_ctx,
1049                          timeval_current_ofs(ctdb->tunable.recovery_drop_all_ips, 0),
1050                          ctdb_drop_all_ips_event, ctdb);
1051         return 0;
1052 }
1053
1054 /*
1055   set the recovery mode
1056  */
1057 int32_t ctdb_control_set_recmode(struct ctdb_context *ctdb,
1058                                  struct ctdb_req_control_old *c,
1059                                  TDB_DATA indata, bool *async_reply,
1060                                  const char **errormsg)
1061 {
1062         uint32_t recmode = *(uint32_t *)indata.dptr;
1063         int i;
1064         struct ctdb_db_context *ctdb_db;
1065         struct ctdb_cluster_mutex_handle *h;
1066
1067         /* if we enter recovery but stay in recovery for too long
1068            we will eventually drop all our ip addresses
1069         */
1070         if (recmode == CTDB_RECOVERY_NORMAL) {
1071                 talloc_free(ctdb->release_ips_ctx);
1072                 ctdb->release_ips_ctx = NULL;
1073         } else {
1074                 if (ctdb_deferred_drop_all_ips(ctdb) != 0) {
1075                         DEBUG(DEBUG_ERR,("Failed to set up deferred drop all ips\n"));
1076                 }
1077         }
1078
1079         if (recmode != ctdb->recovery_mode) {
1080                 DEBUG(DEBUG_NOTICE,(__location__ " Recovery mode set to %s\n",
1081                          recmode==CTDB_RECOVERY_NORMAL?"NORMAL":"ACTIVE"));
1082         }
1083
1084         if (recmode != CTDB_RECOVERY_NORMAL ||
1085             ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
1086                 ctdb->recovery_mode = recmode;
1087                 return 0;
1088         }
1089
1090         /* From this point: recmode == CTDB_RECOVERY_NORMAL
1091          *
1092          * Therefore, what follows is special handling when setting
1093          * recovery mode back to normal */
1094
1095         for (ctdb_db = ctdb->db_list; ctdb_db != NULL; ctdb_db = ctdb_db->next) {
1096                 if (ctdb_db->generation != ctdb->vnn_map->generation) {
1097                         DEBUG(DEBUG_ERR,
1098                               ("Inconsistent DB generation %u for %s\n",
1099                                ctdb_db->generation, ctdb_db->db_name));
1100                         DEBUG(DEBUG_ERR, ("Recovery mode set to ACTIVE\n"));
1101                         return -1;
1102                 }
1103         }
1104
1105         /* force the databases to thaw */
1106         for (i=1; i<=NUM_DB_PRIORITIES; i++) {
1107                 if (ctdb_db_prio_frozen(ctdb, i)) {
1108                         ctdb_control_thaw(ctdb, i, false);
1109                 }
1110         }
1111
1112         /* release any deferred attach calls from clients */
1113         if (recmode == CTDB_RECOVERY_NORMAL) {
1114                 ctdb_process_deferred_attach(ctdb);
1115         }
1116
1117         if (ctdb->recovery_lock_file == NULL) {
1118                 /* Not using recovery lock file */
1119                 ctdb->recovery_mode = recmode;
1120                 return 0;
1121         }
1122
1123         h = ctdb_cluster_mutex(ctdb, 5);
1124         if (h == NULL) {
1125                 return -1;
1126         }
1127
1128         /* set_recmode_handler() frees h */
1129         h->handler = set_recmode_handler;
1130         h->private_data = talloc_steal(h, c);
1131
1132         *async_reply = true;
1133
1134         return 0;
1135 }
1136
1137
1138 bool ctdb_recovery_have_lock(struct ctdb_context *ctdb)
1139 {
1140         return ctdb->recovery_lock_fd != -1;
1141 }
1142
1143 /*
1144   try and get the recovery lock in shared storage - should only work
1145   on the recovery master recovery daemon. Anywhere else is a bug
1146  */
1147 bool ctdb_recovery_lock(struct ctdb_context *ctdb)
1148 {
1149         struct flock lock;
1150
1151         ctdb->recovery_lock_fd = open(ctdb->recovery_lock_file,
1152                                       O_RDWR|O_CREAT, 0600);
1153         if (ctdb->recovery_lock_fd == -1) {
1154                 DEBUG(DEBUG_ERR,
1155                       ("ctdb_recovery_lock: Unable to open %s - (%s)\n",
1156                        ctdb->recovery_lock_file, strerror(errno)));
1157                 return false;
1158         }
1159
1160         set_close_on_exec(ctdb->recovery_lock_fd);
1161
1162         lock.l_type = F_WRLCK;
1163         lock.l_whence = SEEK_SET;
1164         lock.l_start = 0;
1165         lock.l_len = 1;
1166         lock.l_pid = 0;
1167
1168         if (fcntl(ctdb->recovery_lock_fd, F_SETLK, &lock) != 0) {
1169                 int saved_errno = errno;
1170                 close(ctdb->recovery_lock_fd);
1171                 ctdb->recovery_lock_fd = -1;
1172                 /* Fail silently on these errors, since they indicate
1173                  * lock contention, but log an error for any other
1174                  * failure. */
1175                 if (saved_errno != EACCES &&
1176                     saved_errno != EAGAIN) {
1177                         DEBUG(DEBUG_ERR,("ctdb_recovery_lock: Failed to get "
1178                                          "recovery lock on '%s' - (%s)\n",
1179                                          ctdb->recovery_lock_file,
1180                                          strerror(saved_errno)));
1181                 }
1182                 return false;
1183         }
1184
1185         return true;
1186 }
1187
1188 void ctdb_recovery_unlock(struct ctdb_context *ctdb)
1189 {
1190         if (ctdb->recovery_lock_fd != -1) {
1191                 DEBUG(DEBUG_NOTICE, ("Releasing recovery lock\n"));
1192                 close(ctdb->recovery_lock_fd);
1193                 ctdb->recovery_lock_fd = -1;
1194         }
1195 }
1196
1197 /*
1198   delete a record as part of the vacuum process
1199   only delete if we are not lmaster or dmaster, and our rsn is <= the provided rsn
1200   use non-blocking locks
1201
1202   return 0 if the record was successfully deleted (i.e. it does not exist
1203   when the function returns)
1204   or !0 is the record still exists in the tdb after returning.
1205  */
1206 static int delete_tdb_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, struct ctdb_rec_data_old *rec)
1207 {
1208         TDB_DATA key, data, data2;
1209         struct ctdb_ltdb_header *hdr, *hdr2;
1210
1211         /* these are really internal tdb functions - but we need them here for
1212            non-blocking lock of the freelist */
1213         int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
1214         int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
1215
1216
1217         key.dsize = rec->keylen;
1218         key.dptr  = &rec->data[0];
1219         data.dsize = rec->datalen;
1220         data.dptr = &rec->data[rec->keylen];
1221
1222         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1223                 DEBUG(DEBUG_INFO,(__location__ " Called delete on record where we are lmaster\n"));
1224                 return -1;
1225         }
1226
1227         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1228                 DEBUG(DEBUG_ERR,(__location__ " Bad record size\n"));
1229                 return -1;
1230         }
1231
1232         hdr = (struct ctdb_ltdb_header *)data.dptr;
1233
1234         /* use a non-blocking lock */
1235         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1236                 return -1;
1237         }
1238
1239         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1240         if (data2.dptr == NULL) {
1241                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1242                 return 0;
1243         }
1244
1245         if (data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1246                 if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) == 0) {
1247                         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1248                                 DEBUG(DEBUG_CRIT,(__location__ " Failed to delete corrupt record\n"));
1249                         }
1250                         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1251                         DEBUG(DEBUG_CRIT,(__location__ " Deleted corrupt record\n"));
1252                 }
1253                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1254                 free(data2.dptr);
1255                 return 0;
1256         }
1257
1258         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1259
1260         if (hdr2->rsn > hdr->rsn) {
1261                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1262                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with rsn=%llu - called with rsn=%llu\n",
1263                          (unsigned long long)hdr2->rsn, (unsigned long long)hdr->rsn));
1264                 free(data2.dptr);
1265                 return -1;
1266         }
1267
1268         /* do not allow deleting record that have readonly flags set. */
1269         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1270                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1271                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1272                 free(data2.dptr);
1273                 return -1;
1274         }
1275         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1276                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1277                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1278                 free(data2.dptr);
1279                 return -1;
1280         }
1281
1282         if (hdr2->dmaster == ctdb->pnn) {
1283                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1284                 DEBUG(DEBUG_INFO,(__location__ " Attempted delete record where we are the dmaster\n"));
1285                 free(data2.dptr);
1286                 return -1;
1287         }
1288
1289         if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) != 0) {
1290                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1291                 free(data2.dptr);
1292                 return -1;
1293         }
1294
1295         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1296                 tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1297                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1298                 DEBUG(DEBUG_INFO,(__location__ " Failed to delete record\n"));
1299                 free(data2.dptr);
1300                 return -1;
1301         }
1302
1303         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1304         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1305         free(data2.dptr);
1306         return 0;
1307 }
1308
1309
1310
1311 struct recovery_callback_state {
1312         struct ctdb_req_control_old *c;
1313 };
1314
1315
1316 /*
1317   called when the 'recovered' event script has finished
1318  */
1319 static void ctdb_end_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1320 {
1321         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1322
1323         ctdb_enable_monitoring(ctdb);
1324         CTDB_INCREMENT_STAT(ctdb, num_recoveries);
1325
1326         if (status != 0) {
1327                 DEBUG(DEBUG_ERR,(__location__ " recovered event script failed (status %d)\n", status));
1328                 if (status == -ETIME) {
1329                         ctdb_ban_self(ctdb);
1330                 }
1331         }
1332
1333         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1334         talloc_free(state);
1335
1336         gettimeofday(&ctdb->last_recovery_finished, NULL);
1337
1338         if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
1339                 ctdb_set_runstate(ctdb, CTDB_RUNSTATE_STARTUP);
1340         }
1341 }
1342
1343 /*
1344   recovery has finished
1345  */
1346 int32_t ctdb_control_end_recovery(struct ctdb_context *ctdb,
1347                                 struct ctdb_req_control_old *c,
1348                                 bool *async_reply)
1349 {
1350         int ret;
1351         struct recovery_callback_state *state;
1352
1353         DEBUG(DEBUG_NOTICE,("Recovery has finished\n"));
1354
1355         ctdb_persistent_finish_trans3_commits(ctdb);
1356
1357         state = talloc(ctdb, struct recovery_callback_state);
1358         CTDB_NO_MEMORY(ctdb, state);
1359
1360         state->c    = c;
1361
1362         ctdb_disable_monitoring(ctdb);
1363
1364         ret = ctdb_event_script_callback(ctdb, state,
1365                                          ctdb_end_recovery_callback,
1366                                          state,
1367                                          CTDB_EVENT_RECOVERED, "%s", "");
1368
1369         if (ret != 0) {
1370                 ctdb_enable_monitoring(ctdb);
1371
1372                 DEBUG(DEBUG_ERR,(__location__ " Failed to end recovery\n"));
1373                 talloc_free(state);
1374                 return -1;
1375         }
1376
1377         /* tell the control that we will be reply asynchronously */
1378         state->c    = talloc_steal(state, c);
1379         *async_reply = true;
1380         return 0;
1381 }
1382
1383 /*
1384   called when the 'startrecovery' event script has finished
1385  */
1386 static void ctdb_start_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1387 {
1388         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1389
1390         if (status != 0) {
1391                 DEBUG(DEBUG_ERR,(__location__ " startrecovery event script failed (status %d)\n", status));
1392         }
1393
1394         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1395         talloc_free(state);
1396 }
1397
1398 /*
1399   run the startrecovery eventscript
1400  */
1401 int32_t ctdb_control_start_recovery(struct ctdb_context *ctdb,
1402                                 struct ctdb_req_control_old *c,
1403                                 bool *async_reply)
1404 {
1405         int ret;
1406         struct recovery_callback_state *state;
1407
1408         DEBUG(DEBUG_NOTICE,(__location__ " startrecovery eventscript has been invoked\n"));
1409         gettimeofday(&ctdb->last_recovery_started, NULL);
1410
1411         state = talloc(ctdb, struct recovery_callback_state);
1412         CTDB_NO_MEMORY(ctdb, state);
1413
1414         state->c    = talloc_steal(state, c);
1415
1416         ctdb_disable_monitoring(ctdb);
1417
1418         ret = ctdb_event_script_callback(ctdb, state,
1419                                          ctdb_start_recovery_callback,
1420                                          state,
1421                                          CTDB_EVENT_START_RECOVERY,
1422                                          "%s", "");
1423
1424         if (ret != 0) {
1425                 DEBUG(DEBUG_ERR,(__location__ " Failed to start recovery\n"));
1426                 talloc_free(state);
1427                 return -1;
1428         }
1429
1430         /* tell the control that we will be reply asynchronously */
1431         *async_reply = true;
1432         return 0;
1433 }
1434
1435 /*
1436  try to delete all these records as part of the vacuuming process
1437  and return the records we failed to delete
1438 */
1439 int32_t ctdb_control_try_delete_records(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
1440 {
1441         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1442         struct ctdb_db_context *ctdb_db;
1443         int i;
1444         struct ctdb_rec_data_old *rec;
1445         struct ctdb_marshall_buffer *records;
1446
1447         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1448                 DEBUG(DEBUG_ERR,(__location__ " invalid data in try_delete_records\n"));
1449                 return -1;
1450         }
1451
1452         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1453         if (!ctdb_db) {
1454                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
1455                 return -1;
1456         }
1457
1458
1459         DEBUG(DEBUG_DEBUG,("starting try_delete_records of %u records for dbid 0x%x\n",
1460                  reply->count, reply->db_id));
1461
1462
1463         /* create a blob to send back the records we couldnt delete */
1464         records = (struct ctdb_marshall_buffer *)
1465                         talloc_zero_size(outdata,
1466                                     offsetof(struct ctdb_marshall_buffer, data));
1467         if (records == NULL) {
1468                 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
1469                 return -1;
1470         }
1471         records->db_id = ctdb_db->db_id;
1472
1473
1474         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1475         for (i=0;i<reply->count;i++) {
1476                 TDB_DATA key, data;
1477
1478                 key.dptr = &rec->data[0];
1479                 key.dsize = rec->keylen;
1480                 data.dptr = &rec->data[key.dsize];
1481                 data.dsize = rec->datalen;
1482
1483                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1484                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record in indata\n"));
1485                         return -1;
1486                 }
1487
1488                 /* If we cant delete the record we must add it to the reply
1489                    so the lmaster knows it may not purge this record
1490                 */
1491                 if (delete_tdb_record(ctdb, ctdb_db, rec) != 0) {
1492                         size_t old_size;
1493                         struct ctdb_ltdb_header *hdr;
1494
1495                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1496                         data.dptr += sizeof(*hdr);
1497                         data.dsize -= sizeof(*hdr);
1498
1499                         DEBUG(DEBUG_INFO, (__location__ " Failed to vacuum delete record with hash 0x%08x\n", ctdb_hash(&key)));
1500
1501                         old_size = talloc_get_size(records);
1502                         records = talloc_realloc_size(outdata, records, old_size + rec->length);
1503                         if (records == NULL) {
1504                                 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
1505                                 return -1;
1506                         }
1507                         records->count++;
1508                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1509                 }
1510
1511                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1512         }
1513
1514
1515         *outdata = ctdb_marshall_finish(records);
1516
1517         return 0;
1518 }
1519
1520 /**
1521  * Store a record as part of the vacuum process:
1522  * This is called from the RECEIVE_RECORD control which
1523  * the lmaster uses to send the current empty copy
1524  * to all nodes for storing, before it lets the other
1525  * nodes delete the records in the second phase with
1526  * the TRY_DELETE_RECORDS control.
1527  *
1528  * Only store if we are not lmaster or dmaster, and our
1529  * rsn is <= the provided rsn. Use non-blocking locks.
1530  *
1531  * return 0 if the record was successfully stored.
1532  * return !0 if the record still exists in the tdb after returning.
1533  */
1534 static int store_tdb_record(struct ctdb_context *ctdb,
1535                             struct ctdb_db_context *ctdb_db,
1536                             struct ctdb_rec_data_old *rec)
1537 {
1538         TDB_DATA key, data, data2;
1539         struct ctdb_ltdb_header *hdr, *hdr2;
1540         int ret;
1541
1542         key.dsize = rec->keylen;
1543         key.dptr = &rec->data[0];
1544         data.dsize = rec->datalen;
1545         data.dptr = &rec->data[rec->keylen];
1546
1547         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1548                 DEBUG(DEBUG_INFO, (__location__ " Called store_tdb_record "
1549                                    "where we are lmaster\n"));
1550                 return -1;
1551         }
1552
1553         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1554                 DEBUG(DEBUG_ERR, (__location__ " Bad record size\n"));
1555                 return -1;
1556         }
1557
1558         hdr = (struct ctdb_ltdb_header *)data.dptr;
1559
1560         /* use a non-blocking lock */
1561         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1562                 DEBUG(DEBUG_INFO, (__location__ " Failed to lock chain in non-blocking mode\n"));
1563                 return -1;
1564         }
1565
1566         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1567         if (data2.dptr == NULL || data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1568                 if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) == -1) {
1569                         DEBUG(DEBUG_ERR, (__location__ "Failed to store record\n"));
1570                         ret = -1;
1571                         goto done;
1572                 }
1573                 DEBUG(DEBUG_INFO, (__location__ " Stored record\n"));
1574                 ret = 0;
1575                 goto done;
1576         }
1577
1578         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1579
1580         if (hdr2->rsn > hdr->rsn) {
1581                 DEBUG(DEBUG_INFO, (__location__ " Skipping record with "
1582                                    "rsn=%llu - called with rsn=%llu\n",
1583                                    (unsigned long long)hdr2->rsn,
1584                                    (unsigned long long)hdr->rsn));
1585                 ret = -1;
1586                 goto done;
1587         }
1588
1589         /* do not allow vacuuming of records that have readonly flags set. */
1590         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1591                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1592                                   "flags set\n"));
1593                 ret = -1;
1594                 goto done;
1595         }
1596         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1597                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1598                                   "flags set\n"));
1599                 ret = -1;
1600                 goto done;
1601         }
1602
1603         if (hdr2->dmaster == ctdb->pnn) {
1604                 DEBUG(DEBUG_INFO, (__location__ " Attempted to store record "
1605                                    "where we are the dmaster\n"));
1606                 ret = -1;
1607                 goto done;
1608         }
1609
1610         if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) != 0) {
1611                 DEBUG(DEBUG_INFO,(__location__ " Failed to store record\n"));
1612                 ret = -1;
1613                 goto done;
1614         }
1615
1616         ret = 0;
1617
1618 done:
1619         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1620         free(data2.dptr);
1621         return  ret;
1622 }
1623
1624
1625
1626 /**
1627  * Try to store all these records as part of the vacuuming process
1628  * and return the records we failed to store.
1629  */
1630 int32_t ctdb_control_receive_records(struct ctdb_context *ctdb,
1631                                      TDB_DATA indata, TDB_DATA *outdata)
1632 {
1633         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1634         struct ctdb_db_context *ctdb_db;
1635         int i;
1636         struct ctdb_rec_data_old *rec;
1637         struct ctdb_marshall_buffer *records;
1638
1639         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1640                 DEBUG(DEBUG_ERR,
1641                       (__location__ " invalid data in receive_records\n"));
1642                 return -1;
1643         }
1644
1645         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1646         if (!ctdb_db) {
1647                 DEBUG(DEBUG_ERR, (__location__ " Unknown db 0x%08x\n",
1648                                   reply->db_id));
1649                 return -1;
1650         }
1651
1652         DEBUG(DEBUG_DEBUG, ("starting receive_records of %u records for "
1653                             "dbid 0x%x\n", reply->count, reply->db_id));
1654
1655         /* create a blob to send back the records we could not store */
1656         records = (struct ctdb_marshall_buffer *)
1657                         talloc_zero_size(outdata,
1658                                 offsetof(struct ctdb_marshall_buffer, data));
1659         if (records == NULL) {
1660                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1661                 return -1;
1662         }
1663         records->db_id = ctdb_db->db_id;
1664
1665         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1666         for (i=0; i<reply->count; i++) {
1667                 TDB_DATA key, data;
1668
1669                 key.dptr = &rec->data[0];
1670                 key.dsize = rec->keylen;
1671                 data.dptr = &rec->data[key.dsize];
1672                 data.dsize = rec->datalen;
1673
1674                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1675                         DEBUG(DEBUG_CRIT, (__location__ " bad ltdb record "
1676                                            "in indata\n"));
1677                         return -1;
1678                 }
1679
1680                 /*
1681                  * If we can not store the record we must add it to the reply
1682                  * so the lmaster knows it may not purge this record.
1683                  */
1684                 if (store_tdb_record(ctdb, ctdb_db, rec) != 0) {
1685                         size_t old_size;
1686                         struct ctdb_ltdb_header *hdr;
1687
1688                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1689                         data.dptr += sizeof(*hdr);
1690                         data.dsize -= sizeof(*hdr);
1691
1692                         DEBUG(DEBUG_INFO, (__location__ " Failed to store "
1693                                            "record with hash 0x%08x in vacuum "
1694                                            "via RECEIVE_RECORDS\n",
1695                                            ctdb_hash(&key)));
1696
1697                         old_size = talloc_get_size(records);
1698                         records = talloc_realloc_size(outdata, records,
1699                                                       old_size + rec->length);
1700                         if (records == NULL) {
1701                                 DEBUG(DEBUG_ERR, (__location__ " Failed to "
1702                                                   "expand\n"));
1703                                 return -1;
1704                         }
1705                         records->count++;
1706                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1707                 }
1708
1709                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1710         }
1711
1712         *outdata = ctdb_marshall_finish(records);
1713
1714         return 0;
1715 }
1716
1717
1718 /*
1719   report capabilities
1720  */
1721 int32_t ctdb_control_get_capabilities(struct ctdb_context *ctdb, TDB_DATA *outdata)
1722 {
1723         uint32_t *capabilities = NULL;
1724
1725         capabilities = talloc(outdata, uint32_t);
1726         CTDB_NO_MEMORY(ctdb, capabilities);
1727         *capabilities = ctdb->capabilities;
1728
1729         outdata->dsize = sizeof(uint32_t);
1730         outdata->dptr = (uint8_t *)capabilities;
1731
1732         return 0;
1733 }
1734
1735 /* The recovery daemon will ping us at regular intervals.
1736    If we havent been pinged for a while we assume the recovery
1737    daemon is inoperable and we restart.
1738 */
1739 static void ctdb_recd_ping_timeout(struct tevent_context *ev,
1740                                    struct tevent_timer *te,
1741                                    struct timeval t, void *p)
1742 {
1743         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
1744         uint32_t *count = talloc_get_type(ctdb->recd_ping_count, uint32_t);
1745
1746         DEBUG(DEBUG_ERR, ("Recovery daemon ping timeout. Count : %u\n", *count));
1747
1748         if (*count < ctdb->tunable.recd_ping_failcount) {
1749                 (*count)++;
1750                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1751                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1752                                  ctdb_recd_ping_timeout, ctdb);
1753                 return;
1754         }
1755
1756         DEBUG(DEBUG_ERR, ("Final timeout for recovery daemon ping. Restarting recovery daemon. (This can be caused if the cluster filesystem has hung)\n"));
1757
1758         ctdb_stop_recoverd(ctdb);
1759         ctdb_start_recoverd(ctdb);
1760 }
1761
1762 int32_t ctdb_control_recd_ping(struct ctdb_context *ctdb)
1763 {
1764         talloc_free(ctdb->recd_ping_count);
1765
1766         ctdb->recd_ping_count = talloc_zero(ctdb, uint32_t);
1767         CTDB_NO_MEMORY(ctdb, ctdb->recd_ping_count);
1768
1769         if (ctdb->tunable.recd_ping_timeout != 0) {
1770                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1771                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1772                                  ctdb_recd_ping_timeout, ctdb);
1773         }
1774
1775         return 0;
1776 }
1777
1778
1779
1780 int32_t ctdb_control_set_recmaster(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata)
1781 {
1782         uint32_t new_recmaster;
1783
1784         CHECK_CONTROL_DATA_SIZE(sizeof(uint32_t));
1785         new_recmaster = ((uint32_t *)(&indata.dptr[0]))[0];
1786
1787         if (ctdb->pnn != new_recmaster && ctdb->recovery_master == ctdb->pnn) {
1788                 DEBUG(DEBUG_NOTICE,
1789                       ("This node (%u) is no longer the recovery master\n", ctdb->pnn));
1790         }
1791
1792         if (ctdb->pnn == new_recmaster && ctdb->recovery_master != new_recmaster) {
1793                 DEBUG(DEBUG_NOTICE,
1794                       ("This node (%u) is now the recovery master\n", ctdb->pnn));
1795         }
1796
1797         ctdb->recovery_master = new_recmaster;
1798         return 0;
1799 }
1800
1801
1802 int32_t ctdb_control_stop_node(struct ctdb_context *ctdb)
1803 {
1804         DEBUG(DEBUG_NOTICE, ("Stopping node\n"));
1805         ctdb_disable_monitoring(ctdb);
1806         ctdb->nodes[ctdb->pnn]->flags |= NODE_FLAGS_STOPPED;
1807
1808         return 0;
1809 }
1810
1811 int32_t ctdb_control_continue_node(struct ctdb_context *ctdb)
1812 {
1813         DEBUG(DEBUG_NOTICE, ("Continue node\n"));
1814         ctdb->nodes[ctdb->pnn]->flags &= ~NODE_FLAGS_STOPPED;
1815
1816         return 0;
1817 }
1818