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_set_recmode_state {
 751         struct ctdb_context *ctdb;
 752         struct ctdb_req_control_old *c;
 753         int fd[2];
 754         struct tevent_timer *te;
 755         struct tevent_fd *fde;
 756         pid_t child;
 757         struct timeval start_time;
 758 };
 759
 760 /*
 761   called if our set_recmode child times out. this would happen if
 762   ctdb_recovery_lock() would block.
 763  */
 764 static void ctdb_set_recmode_timeout(struct tevent_context *ev,
 765                                      struct tevent_timer *te,
 766                                      struct timeval t, void *private_data)
 767 {
 768         struct ctdb_set_recmode_state *state = talloc_get_type(private_data,
 769                                            struct ctdb_set_recmode_state);
 770
 771         /* we consider this a success, not a failure, as we failed to
 772            set the recovery lock which is what we wanted.  This can be
 773            caused by the cluster filesystem being very slow to
 774            arbitrate locks immediately after a node failure.
 775          */
 776         DEBUG(DEBUG_ERR,(__location__ " set_recmode child process hung/timedout CFS slow to grant locks? (allowing recmode set anyway)\n"));
 777         state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 778         ctdb_request_control_reply(state->ctdb, state->c, NULL, 0, NULL);
 779         talloc_free(state);
 780 }
 781
 782
 783 /* when we free the recmode state we must kill any child process.
 784 */
 785 static int set_recmode_destructor(struct ctdb_set_recmode_state *state)
 786 {
 787         if (state->fd[0] != -1) {
 788                 state->fd[0] = -1;
 789         }
 790         ctdb_kill(state->ctdb, state->child, SIGKILL);
 791         return 0;
 792 }
 793
 794 /* this is called when the client process has completed ctdb_recovery_lock()
 795    and has written data back to us through the pipe.
 796 */
 797 static void set_recmode_handler(struct tevent_context *ev,
 798                                 struct tevent_fd *fde,
 799                                 uint16_t flags, void *private_data)
 800 {
 801         struct ctdb_set_recmode_state *state= talloc_get_type(private_data,
 802                                              struct ctdb_set_recmode_state);
 803         char c = 0;
 804         int ret;
 805         int status = 0;
 806         const char *err = NULL;
 807
 808         /* we got a response from our child process so we can abort the
 809            timeout.
 810         */
 811         talloc_free(state->te);
 812         state->te = NULL;
 813
 814         ret = sys_read(state->fd[0], &c, 1);
 815         if (ret == 1) {
 816                 /* Child wrote status. EACCES indicates that it was unable
 817                  * to take the lock, which is the expected outcome.
 818                  * 0 indicates that it was able to take the
 819                  * lock, which is an error because the recovery daemon
 820                  * should be holding the lock. */
 821                 double l = timeval_elapsed(&state->start_time);
 822
 823                 if (c == EACCES) {
 824                         status = 0;
 825                         err = NULL;
 826
 827                         state->ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
 828
 829                         /* release any deferred attach calls from clients */
 830                         ctdb_process_deferred_attach(state->ctdb);
 831
 832                         CTDB_UPDATE_RECLOCK_LATENCY(state->ctdb, "daemon reclock", reclock.ctdbd, l);
 833                 } else {
 834                         status = -1;
 835                         err = "Took recovery lock from daemon during recovery - probably a cluster filesystem lock coherence problem";
 836                 }
 837         } else {
 838                 /* Child did not write status.  Unexpected error.
 839                  * Child may have received a signal.  */
 840                 status = -1;
 841                 err = "Unexpected error when testing recovery lock";
 842         }
 843
 844         ctdb_request_control_reply(state->ctdb, state->c, NULL, status, err);
 845         talloc_free(state);
 846 }
 847
 848 static void
 849 ctdb_drop_all_ips_event(struct tevent_context *ev, struct tevent_timer *te,
 850                         struct timeval t, void *private_data)
 851 {
 852         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 853
 854         DEBUG(DEBUG_ERR,(__location__ " Been in recovery mode for too long. Dropping all IPS\n"));
 855         talloc_free(ctdb->release_ips_ctx);
 856         ctdb->release_ips_ctx = NULL;
 857
 858         ctdb_release_all_ips(ctdb);
 859 }
 860
 861 /*
 862  * Set up an event to drop all public ips if we remain in recovery for too
 863  * long
 864  */
 865 int ctdb_deferred_drop_all_ips(struct ctdb_context *ctdb)
 866 {
 867         if (ctdb->release_ips_ctx != NULL) {
 868                 talloc_free(ctdb->release_ips_ctx);
 869         }
 870         ctdb->release_ips_ctx = talloc_new(ctdb);
 871         CTDB_NO_MEMORY(ctdb, ctdb->release_ips_ctx);
 872
 873         tevent_add_timer(ctdb->ev, ctdb->release_ips_ctx,
 874                          timeval_current_ofs(ctdb->tunable.recovery_drop_all_ips, 0),
 875                          ctdb_drop_all_ips_event, ctdb);
 876         return 0;
 877 }
 878
 879 /*
 880   set the recovery mode
 881  */
 882 int32_t ctdb_control_set_recmode(struct ctdb_context *ctdb,
 883                                  struct ctdb_req_control_old *c,
 884                                  TDB_DATA indata, bool *async_reply,
 885                                  const char **errormsg)
 886 {
 887         uint32_t recmode = *(uint32_t *)indata.dptr;
 888         int i, ret;
 889         struct ctdb_set_recmode_state *state;
 890         pid_t parent = getpid();
 891         struct ctdb_db_context *ctdb_db;
 892
 893         /* if we enter recovery but stay in recovery for too long
 894            we will eventually drop all our ip addresses
 895         */
 896         if (recmode == CTDB_RECOVERY_NORMAL) {
 897                 talloc_free(ctdb->release_ips_ctx);
 898                 ctdb->release_ips_ctx = NULL;
 899         } else {
 900                 if (ctdb_deferred_drop_all_ips(ctdb) != 0) {
 901                         DEBUG(DEBUG_ERR,("Failed to set up deferred drop all ips\n"));
 902                 }
 903         }
 904
 905         if (recmode != ctdb->recovery_mode) {
 906                 DEBUG(DEBUG_NOTICE,(__location__ " Recovery mode set to %s\n",
 907                          recmode==CTDB_RECOVERY_NORMAL?"NORMAL":"ACTIVE"));
 908         }
 909
 910         if (recmode != CTDB_RECOVERY_NORMAL ||
 911             ctdb->recovery_mode != CTDB_RECOVERY_ACTIVE) {
 912                 ctdb->recovery_mode = recmode;
 913                 return 0;
 914         }
 915
 916         /* From this point: recmode == CTDB_RECOVERY_NORMAL
 917          *
 918          * Therefore, what follows is special handling when setting
 919          * recovery mode back to normal */
 920
 921         for (ctdb_db = ctdb->db_list; ctdb_db != NULL; ctdb_db = ctdb_db->next) {
 922                 if (ctdb_db->generation != ctdb->vnn_map->generation) {
 923                         DEBUG(DEBUG_ERR,
 924                               ("Inconsistent DB generation %u for %s\n",
 925                                ctdb_db->generation, ctdb_db->db_name));
 926                         DEBUG(DEBUG_ERR, ("Recovery mode set to ACTIVE\n"));
 927                         return -1;
 928                 }
 929         }
 930
 931         /* force the databases to thaw */
 932         for (i=1; i<=NUM_DB_PRIORITIES; i++) {
 933                 if (ctdb_db_prio_frozen(ctdb, i)) {
 934                         ctdb_control_thaw(ctdb, i, false);
 935                 }
 936         }
 937
 938         /* release any deferred attach calls from clients */
 939         if (recmode == CTDB_RECOVERY_NORMAL) {
 940                 ctdb_process_deferred_attach(ctdb);
 941         }
 942
 943         if (ctdb->recovery_lock_file == NULL) {
 944                 /* Not using recovery lock file */
 945                 ctdb->recovery_mode = recmode;
 946                 return 0;
 947         }
 948
 949         state = talloc(ctdb, struct ctdb_set_recmode_state);
 950         CTDB_NO_MEMORY(ctdb, state);
 951
 952         state->start_time = timeval_current();
 953         state->fd[0] = -1;
 954         state->fd[1] = -1;
 955
 956         /* For the rest of what needs to be done, we need to do this in
 957            a child process since
 958            1, the call to ctdb_recovery_lock() can block if the cluster
 959               filesystem is in the process of recovery.
 960         */
 961         ret = pipe(state->fd);
 962         if (ret != 0) {
 963                 talloc_free(state);
 964                 DEBUG(DEBUG_CRIT,(__location__ " Failed to open pipe for set_recmode child\n"));
 965                 return -1;
 966         }
 967
 968         state->child = ctdb_fork(ctdb);
 969         if (state->child == (pid_t)-1) {
 970                 close(state->fd[0]);
 971                 close(state->fd[1]);
 972                 talloc_free(state);
 973                 return -1;
 974         }
 975
 976         if (state->child == 0) {
 977                 char cc = EACCES;
 978                 close(state->fd[0]);
 979
 980                 prctl_set_comment("ctdb_recmode");
 981                 debug_extra = talloc_asprintf(NULL, "set_recmode:");
 982                 /* Daemon should not be able to get the recover lock,
 983                  * as it should be held by the recovery master */
 984                 if (ctdb_recovery_lock(ctdb)) {
 985                         DEBUG(DEBUG_ERR,
 986                               ("ERROR: Daemon able to take recovery lock on \"%s\" during recovery\n",
 987                                ctdb->recovery_lock_file));
 988                         ctdb_recovery_unlock(ctdb);
 989                         cc = 0;
 990                 }
 991
 992                 sys_write(state->fd[1], &cc, 1);
 993                 ctdb_wait_for_process_to_exit(parent);
 994                 _exit(0);
 995         }
 996         close(state->fd[1]);
 997         set_close_on_exec(state->fd[0]);
 998
 999         state->fd[1] = -1;
1000
1001         talloc_set_destructor(state, set_recmode_destructor);
1002
1003         DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d for setrecmode\n", state->fd[0]));
1004
1005         state->te = tevent_add_timer(ctdb->ev, state, timeval_current_ofs(5, 0),
1006                                      ctdb_set_recmode_timeout, state);
1007
1008         state->fde = tevent_add_fd(ctdb->ev, state, state->fd[0], TEVENT_FD_READ,
1009                                    set_recmode_handler, (void *)state);
1010
1011         if (state->fde == NULL) {
1012                 talloc_free(state);
1013                 return -1;
1014         }
1015         tevent_fd_set_auto_close(state->fde);
1016
1017         state->ctdb    = ctdb;
1018         state->c       = talloc_steal(state, c);
1019
1020         *async_reply = true;
1021
1022         return 0;
1023 }
1024
1025
1026 bool ctdb_recovery_have_lock(struct ctdb_context *ctdb)
1027 {
1028         return ctdb->recovery_lock_fd != -1;
1029 }
1030
1031 /*
1032   try and get the recovery lock in shared storage - should only work
1033   on the recovery master recovery daemon. Anywhere else is a bug
1034  */
1035 bool ctdb_recovery_lock(struct ctdb_context *ctdb)
1036 {
1037         struct flock lock;
1038
1039         ctdb->recovery_lock_fd = open(ctdb->recovery_lock_file,
1040                                       O_RDWR|O_CREAT, 0600);
1041         if (ctdb->recovery_lock_fd == -1) {
1042                 DEBUG(DEBUG_ERR,
1043                       ("ctdb_recovery_lock: Unable to open %s - (%s)\n",
1044                        ctdb->recovery_lock_file, strerror(errno)));
1045                 return false;
1046         }
1047
1048         set_close_on_exec(ctdb->recovery_lock_fd);
1049
1050         lock.l_type = F_WRLCK;
1051         lock.l_whence = SEEK_SET;
1052         lock.l_start = 0;
1053         lock.l_len = 1;
1054         lock.l_pid = 0;
1055
1056         if (fcntl(ctdb->recovery_lock_fd, F_SETLK, &lock) != 0) {
1057                 int saved_errno = errno;
1058                 close(ctdb->recovery_lock_fd);
1059                 ctdb->recovery_lock_fd = -1;
1060                 /* Fail silently on these errors, since they indicate
1061                  * lock contention, but log an error for any other
1062                  * failure. */
1063                 if (saved_errno != EACCES &&
1064                     saved_errno != EAGAIN) {
1065                         DEBUG(DEBUG_ERR,("ctdb_recovery_lock: Failed to get "
1066                                          "recovery lock on '%s' - (%s)\n",
1067                                          ctdb->recovery_lock_file,
1068                                          strerror(saved_errno)));
1069                 }
1070                 return false;
1071         }
1072
1073         return true;
1074 }
1075
1076 void ctdb_recovery_unlock(struct ctdb_context *ctdb)
1077 {
1078         if (ctdb->recovery_lock_fd != -1) {
1079                 DEBUG(DEBUG_NOTICE, ("Releasing recovery lock\n"));
1080                 close(ctdb->recovery_lock_fd);
1081                 ctdb->recovery_lock_fd = -1;
1082         }
1083 }
1084
1085 /*
1086   delete a record as part of the vacuum process
1087   only delete if we are not lmaster or dmaster, and our rsn is <= the provided rsn
1088   use non-blocking locks
1089
1090   return 0 if the record was successfully deleted (i.e. it does not exist
1091   when the function returns)
1092   or !0 is the record still exists in the tdb after returning.
1093  */
1094 static int delete_tdb_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, struct ctdb_rec_data_old *rec)
1095 {
1096         TDB_DATA key, data, data2;
1097         struct ctdb_ltdb_header *hdr, *hdr2;
1098
1099         /* these are really internal tdb functions - but we need them here for
1100            non-blocking lock of the freelist */
1101         int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
1102         int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
1103
1104
1105         key.dsize = rec->keylen;
1106         key.dptr  = &rec->data[0];
1107         data.dsize = rec->datalen;
1108         data.dptr = &rec->data[rec->keylen];
1109
1110         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1111                 DEBUG(DEBUG_INFO,(__location__ " Called delete on record where we are lmaster\n"));
1112                 return -1;
1113         }
1114
1115         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1116                 DEBUG(DEBUG_ERR,(__location__ " Bad record size\n"));
1117                 return -1;
1118         }
1119
1120         hdr = (struct ctdb_ltdb_header *)data.dptr;
1121
1122         /* use a non-blocking lock */
1123         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1124                 return -1;
1125         }
1126
1127         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1128         if (data2.dptr == NULL) {
1129                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1130                 return 0;
1131         }
1132
1133         if (data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1134                 if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) == 0) {
1135                         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1136                                 DEBUG(DEBUG_CRIT,(__location__ " Failed to delete corrupt record\n"));
1137                         }
1138                         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1139                         DEBUG(DEBUG_CRIT,(__location__ " Deleted corrupt record\n"));
1140                 }
1141                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1142                 free(data2.dptr);
1143                 return 0;
1144         }
1145
1146         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1147
1148         if (hdr2->rsn > hdr->rsn) {
1149                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1150                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with rsn=%llu - called with rsn=%llu\n",
1151                          (unsigned long long)hdr2->rsn, (unsigned long long)hdr->rsn));
1152                 free(data2.dptr);
1153                 return -1;
1154         }
1155
1156         /* do not allow deleting record that have readonly flags set. */
1157         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1158                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1159                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1160                 free(data2.dptr);
1161                 return -1;
1162         }
1163         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1164                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1165                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly flags set\n"));
1166                 free(data2.dptr);
1167                 return -1;
1168         }
1169
1170         if (hdr2->dmaster == ctdb->pnn) {
1171                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1172                 DEBUG(DEBUG_INFO,(__location__ " Attempted delete record where we are the dmaster\n"));
1173                 free(data2.dptr);
1174                 return -1;
1175         }
1176
1177         if (tdb_lock_nonblock(ctdb_db->ltdb->tdb, -1, F_WRLCK) != 0) {
1178                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1179                 free(data2.dptr);
1180                 return -1;
1181         }
1182
1183         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
1184                 tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1185                 tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1186                 DEBUG(DEBUG_INFO,(__location__ " Failed to delete record\n"));
1187                 free(data2.dptr);
1188                 return -1;
1189         }
1190
1191         tdb_unlock(ctdb_db->ltdb->tdb, -1, F_WRLCK);
1192         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1193         free(data2.dptr);
1194         return 0;
1195 }
1196
1197
1198
1199 struct recovery_callback_state {
1200         struct ctdb_req_control_old *c;
1201 };
1202
1203
1204 /*
1205   called when the 'recovered' event script has finished
1206  */
1207 static void ctdb_end_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1208 {
1209         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1210
1211         ctdb_enable_monitoring(ctdb);
1212         CTDB_INCREMENT_STAT(ctdb, num_recoveries);
1213
1214         if (status != 0) {
1215                 DEBUG(DEBUG_ERR,(__location__ " recovered event script failed (status %d)\n", status));
1216                 if (status == -ETIME) {
1217                         ctdb_ban_self(ctdb);
1218                 }
1219         }
1220
1221         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1222         talloc_free(state);
1223
1224         gettimeofday(&ctdb->last_recovery_finished, NULL);
1225
1226         if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
1227                 ctdb_set_runstate(ctdb, CTDB_RUNSTATE_STARTUP);
1228         }
1229 }
1230
1231 /*
1232   recovery has finished
1233  */
1234 int32_t ctdb_control_end_recovery(struct ctdb_context *ctdb,
1235                                 struct ctdb_req_control_old *c,
1236                                 bool *async_reply)
1237 {
1238         int ret;
1239         struct recovery_callback_state *state;
1240
1241         DEBUG(DEBUG_NOTICE,("Recovery has finished\n"));
1242
1243         ctdb_persistent_finish_trans3_commits(ctdb);
1244
1245         state = talloc(ctdb, struct recovery_callback_state);
1246         CTDB_NO_MEMORY(ctdb, state);
1247
1248         state->c    = c;
1249
1250         ctdb_disable_monitoring(ctdb);
1251
1252         ret = ctdb_event_script_callback(ctdb, state,
1253                                          ctdb_end_recovery_callback,
1254                                          state,
1255                                          CTDB_EVENT_RECOVERED, "%s", "");
1256
1257         if (ret != 0) {
1258                 ctdb_enable_monitoring(ctdb);
1259
1260                 DEBUG(DEBUG_ERR,(__location__ " Failed to end recovery\n"));
1261                 talloc_free(state);
1262                 return -1;
1263         }
1264
1265         /* tell the control that we will be reply asynchronously */
1266         state->c    = talloc_steal(state, c);
1267         *async_reply = true;
1268         return 0;
1269 }
1270
1271 /*
1272   called when the 'startrecovery' event script has finished
1273  */
1274 static void ctdb_start_recovery_callback(struct ctdb_context *ctdb, int status, void *p)
1275 {
1276         struct recovery_callback_state *state = talloc_get_type(p, struct recovery_callback_state);
1277
1278         if (status != 0) {
1279                 DEBUG(DEBUG_ERR,(__location__ " startrecovery event script failed (status %d)\n", status));
1280         }
1281
1282         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1283         talloc_free(state);
1284 }
1285
1286 /*
1287   run the startrecovery eventscript
1288  */
1289 int32_t ctdb_control_start_recovery(struct ctdb_context *ctdb,
1290                                 struct ctdb_req_control_old *c,
1291                                 bool *async_reply)
1292 {
1293         int ret;
1294         struct recovery_callback_state *state;
1295
1296         DEBUG(DEBUG_NOTICE,(__location__ " startrecovery eventscript has been invoked\n"));
1297         gettimeofday(&ctdb->last_recovery_started, NULL);
1298
1299         state = talloc(ctdb, struct recovery_callback_state);
1300         CTDB_NO_MEMORY(ctdb, state);
1301
1302         state->c    = talloc_steal(state, c);
1303
1304         ctdb_disable_monitoring(ctdb);
1305
1306         ret = ctdb_event_script_callback(ctdb, state,
1307                                          ctdb_start_recovery_callback,
1308                                          state,
1309                                          CTDB_EVENT_START_RECOVERY,
1310                                          "%s", "");
1311
1312         if (ret != 0) {
1313                 DEBUG(DEBUG_ERR,(__location__ " Failed to start recovery\n"));
1314                 talloc_free(state);
1315                 return -1;
1316         }
1317
1318         /* tell the control that we will be reply asynchronously */
1319         *async_reply = true;
1320         return 0;
1321 }
1322
1323 /*
1324  try to delete all these records as part of the vacuuming process
1325  and return the records we failed to delete
1326 */
1327 int32_t ctdb_control_try_delete_records(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
1328 {
1329         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1330         struct ctdb_db_context *ctdb_db;
1331         int i;
1332         struct ctdb_rec_data_old *rec;
1333         struct ctdb_marshall_buffer *records;
1334
1335         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1336                 DEBUG(DEBUG_ERR,(__location__ " invalid data in try_delete_records\n"));
1337                 return -1;
1338         }
1339
1340         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1341         if (!ctdb_db) {
1342                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%08x\n", reply->db_id));
1343                 return -1;
1344         }
1345
1346
1347         DEBUG(DEBUG_DEBUG,("starting try_delete_records of %u records for dbid 0x%x\n",
1348                  reply->count, reply->db_id));
1349
1350
1351         /* create a blob to send back the records we couldnt delete */
1352         records = (struct ctdb_marshall_buffer *)
1353                         talloc_zero_size(outdata,
1354                                     offsetof(struct ctdb_marshall_buffer, data));
1355         if (records == NULL) {
1356                 DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
1357                 return -1;
1358         }
1359         records->db_id = ctdb_db->db_id;
1360
1361
1362         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1363         for (i=0;i<reply->count;i++) {
1364                 TDB_DATA key, data;
1365
1366                 key.dptr = &rec->data[0];
1367                 key.dsize = rec->keylen;
1368                 data.dptr = &rec->data[key.dsize];
1369                 data.dsize = rec->datalen;
1370
1371                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1372                         DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record in indata\n"));
1373                         return -1;
1374                 }
1375
1376                 /* If we cant delete the record we must add it to the reply
1377                    so the lmaster knows it may not purge this record
1378                 */
1379                 if (delete_tdb_record(ctdb, ctdb_db, rec) != 0) {
1380                         size_t old_size;
1381                         struct ctdb_ltdb_header *hdr;
1382
1383                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1384                         data.dptr += sizeof(*hdr);
1385                         data.dsize -= sizeof(*hdr);
1386
1387                         DEBUG(DEBUG_INFO, (__location__ " Failed to vacuum delete record with hash 0x%08x\n", ctdb_hash(&key)));
1388
1389                         old_size = talloc_get_size(records);
1390                         records = talloc_realloc_size(outdata, records, old_size + rec->length);
1391                         if (records == NULL) {
1392                                 DEBUG(DEBUG_ERR,(__location__ " Failed to expand\n"));
1393                                 return -1;
1394                         }
1395                         records->count++;
1396                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1397                 }
1398
1399                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1400         }
1401
1402
1403         *outdata = ctdb_marshall_finish(records);
1404
1405         return 0;
1406 }
1407
1408 /**
1409  * Store a record as part of the vacuum process:
1410  * This is called from the RECEIVE_RECORD control which
1411  * the lmaster uses to send the current empty copy
1412  * to all nodes for storing, before it lets the other
1413  * nodes delete the records in the second phase with
1414  * the TRY_DELETE_RECORDS control.
1415  *
1416  * Only store if we are not lmaster or dmaster, and our
1417  * rsn is <= the provided rsn. Use non-blocking locks.
1418  *
1419  * return 0 if the record was successfully stored.
1420  * return !0 if the record still exists in the tdb after returning.
1421  */
1422 static int store_tdb_record(struct ctdb_context *ctdb,
1423                             struct ctdb_db_context *ctdb_db,
1424                             struct ctdb_rec_data_old *rec)
1425 {
1426         TDB_DATA key, data, data2;
1427         struct ctdb_ltdb_header *hdr, *hdr2;
1428         int ret;
1429
1430         key.dsize = rec->keylen;
1431         key.dptr = &rec->data[0];
1432         data.dsize = rec->datalen;
1433         data.dptr = &rec->data[rec->keylen];
1434
1435         if (ctdb_lmaster(ctdb, &key) == ctdb->pnn) {
1436                 DEBUG(DEBUG_INFO, (__location__ " Called store_tdb_record "
1437                                    "where we are lmaster\n"));
1438                 return -1;
1439         }
1440
1441         if (data.dsize != sizeof(struct ctdb_ltdb_header)) {
1442                 DEBUG(DEBUG_ERR, (__location__ " Bad record size\n"));
1443                 return -1;
1444         }
1445
1446         hdr = (struct ctdb_ltdb_header *)data.dptr;
1447
1448         /* use a non-blocking lock */
1449         if (tdb_chainlock_nonblock(ctdb_db->ltdb->tdb, key) != 0) {
1450                 DEBUG(DEBUG_INFO, (__location__ " Failed to lock chain in non-blocking mode\n"));
1451                 return -1;
1452         }
1453
1454         data2 = tdb_fetch(ctdb_db->ltdb->tdb, key);
1455         if (data2.dptr == NULL || data2.dsize < sizeof(struct ctdb_ltdb_header)) {
1456                 if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) == -1) {
1457                         DEBUG(DEBUG_ERR, (__location__ "Failed to store record\n"));
1458                         ret = -1;
1459                         goto done;
1460                 }
1461                 DEBUG(DEBUG_INFO, (__location__ " Stored record\n"));
1462                 ret = 0;
1463                 goto done;
1464         }
1465
1466         hdr2 = (struct ctdb_ltdb_header *)data2.dptr;
1467
1468         if (hdr2->rsn > hdr->rsn) {
1469                 DEBUG(DEBUG_INFO, (__location__ " Skipping record with "
1470                                    "rsn=%llu - called with rsn=%llu\n",
1471                                    (unsigned long long)hdr2->rsn,
1472                                    (unsigned long long)hdr->rsn));
1473                 ret = -1;
1474                 goto done;
1475         }
1476
1477         /* do not allow vacuuming of records that have readonly flags set. */
1478         if (hdr->flags & CTDB_REC_RO_FLAGS) {
1479                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1480                                   "flags set\n"));
1481                 ret = -1;
1482                 goto done;
1483         }
1484         if (hdr2->flags & CTDB_REC_RO_FLAGS) {
1485                 DEBUG(DEBUG_INFO,(__location__ " Skipping record with readonly "
1486                                   "flags set\n"));
1487                 ret = -1;
1488                 goto done;
1489         }
1490
1491         if (hdr2->dmaster == ctdb->pnn) {
1492                 DEBUG(DEBUG_INFO, (__location__ " Attempted to store record "
1493                                    "where we are the dmaster\n"));
1494                 ret = -1;
1495                 goto done;
1496         }
1497
1498         if (tdb_store(ctdb_db->ltdb->tdb, key, data, 0) != 0) {
1499                 DEBUG(DEBUG_INFO,(__location__ " Failed to store record\n"));
1500                 ret = -1;
1501                 goto done;
1502         }
1503
1504         ret = 0;
1505
1506 done:
1507         tdb_chainunlock(ctdb_db->ltdb->tdb, key);
1508         free(data2.dptr);
1509         return  ret;
1510 }
1511
1512
1513
1514 /**
1515  * Try to store all these records as part of the vacuuming process
1516  * and return the records we failed to store.
1517  */
1518 int32_t ctdb_control_receive_records(struct ctdb_context *ctdb,
1519                                      TDB_DATA indata, TDB_DATA *outdata)
1520 {
1521         struct ctdb_marshall_buffer *reply = (struct ctdb_marshall_buffer *)indata.dptr;
1522         struct ctdb_db_context *ctdb_db;
1523         int i;
1524         struct ctdb_rec_data_old *rec;
1525         struct ctdb_marshall_buffer *records;
1526
1527         if (indata.dsize < offsetof(struct ctdb_marshall_buffer, data)) {
1528                 DEBUG(DEBUG_ERR,
1529                       (__location__ " invalid data in receive_records\n"));
1530                 return -1;
1531         }
1532
1533         ctdb_db = find_ctdb_db(ctdb, reply->db_id);
1534         if (!ctdb_db) {
1535                 DEBUG(DEBUG_ERR, (__location__ " Unknown db 0x%08x\n",
1536                                   reply->db_id));
1537                 return -1;
1538         }
1539
1540         DEBUG(DEBUG_DEBUG, ("starting receive_records of %u records for "
1541                             "dbid 0x%x\n", reply->count, reply->db_id));
1542
1543         /* create a blob to send back the records we could not store */
1544         records = (struct ctdb_marshall_buffer *)
1545                         talloc_zero_size(outdata,
1546                                 offsetof(struct ctdb_marshall_buffer, data));
1547         if (records == NULL) {
1548                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1549                 return -1;
1550         }
1551         records->db_id = ctdb_db->db_id;
1552
1553         rec = (struct ctdb_rec_data_old *)&reply->data[0];
1554         for (i=0; i<reply->count; i++) {
1555                 TDB_DATA key, data;
1556
1557                 key.dptr = &rec->data[0];
1558                 key.dsize = rec->keylen;
1559                 data.dptr = &rec->data[key.dsize];
1560                 data.dsize = rec->datalen;
1561
1562                 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1563                         DEBUG(DEBUG_CRIT, (__location__ " bad ltdb record "
1564                                            "in indata\n"));
1565                         return -1;
1566                 }
1567
1568                 /*
1569                  * If we can not store the record we must add it to the reply
1570                  * so the lmaster knows it may not purge this record.
1571                  */
1572                 if (store_tdb_record(ctdb, ctdb_db, rec) != 0) {
1573                         size_t old_size;
1574                         struct ctdb_ltdb_header *hdr;
1575
1576                         hdr = (struct ctdb_ltdb_header *)data.dptr;
1577                         data.dptr += sizeof(*hdr);
1578                         data.dsize -= sizeof(*hdr);
1579
1580                         DEBUG(DEBUG_INFO, (__location__ " Failed to store "
1581                                            "record with hash 0x%08x in vacuum "
1582                                            "via RECEIVE_RECORDS\n",
1583                                            ctdb_hash(&key)));
1584
1585                         old_size = talloc_get_size(records);
1586                         records = talloc_realloc_size(outdata, records,
1587                                                       old_size + rec->length);
1588                         if (records == NULL) {
1589                                 DEBUG(DEBUG_ERR, (__location__ " Failed to "
1590                                                   "expand\n"));
1591                                 return -1;
1592                         }
1593                         records->count++;
1594                         memcpy(old_size+(uint8_t *)records, rec, rec->length);
1595                 }
1596
1597                 rec = (struct ctdb_rec_data_old *)(rec->length + (uint8_t *)rec);
1598         }
1599
1600         *outdata = ctdb_marshall_finish(records);
1601
1602         return 0;
1603 }
1604
1605
1606 /*
1607   report capabilities
1608  */
1609 int32_t ctdb_control_get_capabilities(struct ctdb_context *ctdb, TDB_DATA *outdata)
1610 {
1611         uint32_t *capabilities = NULL;
1612
1613         capabilities = talloc(outdata, uint32_t);
1614         CTDB_NO_MEMORY(ctdb, capabilities);
1615         *capabilities = ctdb->capabilities;
1616
1617         outdata->dsize = sizeof(uint32_t);
1618         outdata->dptr = (uint8_t *)capabilities;
1619
1620         return 0;
1621 }
1622
1623 /* The recovery daemon will ping us at regular intervals.
1624    If we havent been pinged for a while we assume the recovery
1625    daemon is inoperable and we restart.
1626 */
1627 static void ctdb_recd_ping_timeout(struct tevent_context *ev,
1628                                    struct tevent_timer *te,
1629                                    struct timeval t, void *p)
1630 {
1631         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
1632         uint32_t *count = talloc_get_type(ctdb->recd_ping_count, uint32_t);
1633
1634         DEBUG(DEBUG_ERR, ("Recovery daemon ping timeout. Count : %u\n", *count));
1635
1636         if (*count < ctdb->tunable.recd_ping_failcount) {
1637                 (*count)++;
1638                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1639                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1640                                  ctdb_recd_ping_timeout, ctdb);
1641                 return;
1642         }
1643
1644         DEBUG(DEBUG_ERR, ("Final timeout for recovery daemon ping. Restarting recovery daemon. (This can be caused if the cluster filesystem has hung)\n"));
1645
1646         ctdb_stop_recoverd(ctdb);
1647         ctdb_start_recoverd(ctdb);
1648 }
1649
1650 int32_t ctdb_control_recd_ping(struct ctdb_context *ctdb)
1651 {
1652         talloc_free(ctdb->recd_ping_count);
1653
1654         ctdb->recd_ping_count = talloc_zero(ctdb, uint32_t);
1655         CTDB_NO_MEMORY(ctdb, ctdb->recd_ping_count);
1656
1657         if (ctdb->tunable.recd_ping_timeout != 0) {
1658                 tevent_add_timer(ctdb->ev, ctdb->recd_ping_count,
1659                                  timeval_current_ofs(ctdb->tunable.recd_ping_timeout, 0),
1660                                  ctdb_recd_ping_timeout, ctdb);
1661         }
1662
1663         return 0;
1664 }
1665
1666
1667
1668 int32_t ctdb_control_set_recmaster(struct ctdb_context *ctdb, uint32_t opcode, TDB_DATA indata)
1669 {
1670         uint32_t new_recmaster;
1671
1672         CHECK_CONTROL_DATA_SIZE(sizeof(uint32_t));
1673         new_recmaster = ((uint32_t *)(&indata.dptr[0]))[0];
1674
1675         if (ctdb->pnn != new_recmaster && ctdb->recovery_master == ctdb->pnn) {
1676                 DEBUG(DEBUG_NOTICE,
1677                       ("This node (%u) is no longer the recovery master\n", ctdb->pnn));
1678         }
1679
1680         if (ctdb->pnn == new_recmaster && ctdb->recovery_master != new_recmaster) {
1681                 DEBUG(DEBUG_NOTICE,
1682                       ("This node (%u) is now the recovery master\n", ctdb->pnn));
1683         }
1684
1685         ctdb->recovery_master = new_recmaster;
1686         return 0;
1687 }
1688
1689
1690 int32_t ctdb_control_stop_node(struct ctdb_context *ctdb)
1691 {
1692         DEBUG(DEBUG_NOTICE, ("Stopping node\n"));
1693         ctdb_disable_monitoring(ctdb);
1694         ctdb->nodes[ctdb->pnn]->flags |= NODE_FLAGS_STOPPED;
1695
1696         return 0;
1697 }
1698
1699 int32_t ctdb_control_continue_node(struct ctdb_context *ctdb)
1700 {
1701         DEBUG(DEBUG_NOTICE, ("Continue node\n"));
1702         ctdb->nodes[ctdb->pnn]->flags &= ~NODE_FLAGS_STOPPED;
1703
1704         return 0;
1705 }
1706