ctdb/common/ctdb_ltdb.c

   1 /*
   2    ctdb ltdb code
   3
   4    Copyright (C) Andrew Tridgell  2006
   5    Copyright (C) Ronnie sahlberg  2011
   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
  21 #include "replace.h"
  22 #include "system/network.h"
  23 #include "system/filesys.h"
  24
  25 #include <tdb.h>
  26
  27 #include "lib/tdb_wrap/tdb_wrap.h"
  28 #include "lib/util/dlinklist.h"
  29 #include "lib/util/debug.h"
  30
  31 #include "ctdb_private.h"
  32
  33 #include "common/common.h"
  34 #include "common/logging.h"
  35
  36
  37 /*
  38  * Calculate tdb flags based on databse type
  39  */
  40 int ctdb_db_tdb_flags(uint8_t db_flags, bool with_valgrind, bool with_mutex)
  41 {
  42         int tdb_flags = 0;
  43
  44         if (db_flags & CTDB_DB_FLAGS_PERSISTENT) {
  45                 tdb_flags = TDB_DEFAULT;
  46
  47         } else if (db_flags & CTDB_DB_FLAGS_REPLICATED) {
  48                 tdb_flags = TDB_NOSYNC |
  49                             TDB_CLEAR_IF_FIRST |
  50                             TDB_INCOMPATIBLE_HASH;
  51
  52         } else {
  53                 tdb_flags = TDB_NOSYNC |
  54                             TDB_CLEAR_IF_FIRST |
  55                             TDB_INCOMPATIBLE_HASH;
  56
  57 #ifdef TDB_MUTEX_LOCKING
  58                 if (with_mutex && tdb_runtime_check_for_robust_mutexes()) {
  59                         tdb_flags |= TDB_MUTEX_LOCKING;
  60                 }
  61 #endif
  62
  63         }
  64
  65         tdb_flags |= TDB_DISALLOW_NESTING;
  66         if (with_valgrind) {
  67                 tdb_flags |= TDB_NOMMAP;
  68         }
  69
  70         return tdb_flags;
  71 }
  72
  73 /*
  74   find an attached ctdb_db handle given a name
  75  */
  76 struct ctdb_db_context *ctdb_db_handle(struct ctdb_context *ctdb, const char *name)
  77 {
  78         struct ctdb_db_context *tmp_db;
  79         for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
  80                 if (strcmp(name, tmp_db->db_name) == 0) {
  81                         return tmp_db;
  82                 }
  83         }
  84         return NULL;
  85 }
  86
  87 bool ctdb_db_persistent(struct ctdb_db_context *ctdb_db)
  88 {
  89         if (ctdb_db->db_flags & CTDB_DB_FLAGS_PERSISTENT) {
  90                 return true;
  91         }
  92         return false;
  93 }
  94
  95 bool ctdb_db_replicated(struct ctdb_db_context *ctdb_db)
  96 {
  97         if (ctdb_db->db_flags & CTDB_DB_FLAGS_REPLICATED) {
  98                 return true;
  99         }
 100         return false;
 101 }
 102
 103 bool ctdb_db_volatile(struct ctdb_db_context *ctdb_db)
 104 {
 105         if ((ctdb_db->db_flags & CTDB_DB_FLAGS_PERSISTENT) ||
 106             (ctdb_db->db_flags & CTDB_DB_FLAGS_REPLICATED)) {
 107                 return false;
 108         }
 109         return true;
 110 }
 111
 112 bool ctdb_db_readonly(struct ctdb_db_context *ctdb_db)
 113 {
 114         if (ctdb_db->db_flags & CTDB_DB_FLAGS_READONLY) {
 115                 return true;
 116         }
 117         return false;
 118 }
 119
 120 void ctdb_db_set_readonly(struct ctdb_db_context *ctdb_db)
 121 {
 122         ctdb_db->db_flags |= CTDB_DB_FLAGS_READONLY;
 123 }
 124
 125 void ctdb_db_reset_readonly(struct ctdb_db_context *ctdb_db)
 126 {
 127         ctdb_db->db_flags &= ~CTDB_DB_FLAGS_READONLY;
 128 }
 129
 130 bool ctdb_db_sticky(struct ctdb_db_context *ctdb_db)
 131 {
 132         if (ctdb_db->db_flags & CTDB_DB_FLAGS_STICKY) {
 133                 return true;
 134         }
 135         return false;
 136 }
 137
 138 void ctdb_db_set_sticky(struct ctdb_db_context *ctdb_db)
 139 {
 140         ctdb_db->db_flags |= CTDB_DB_FLAGS_STICKY;
 141 }
 142
 143 /*
 144   return the lmaster given a key
 145 */
 146 uint32_t ctdb_lmaster(struct ctdb_context *ctdb, const TDB_DATA *key)
 147 {
 148         uint32_t idx, lmaster;
 149
 150         idx = ctdb_hash(key) % ctdb->vnn_map->size;
 151         lmaster = ctdb->vnn_map->map[idx];
 152
 153         return lmaster;
 154 }
 155
 156
 157 /*
 158   construct an initial header for a record with no ltdb header yet
 159 */
 160 static void ltdb_initial_header(struct ctdb_db_context *ctdb_db,
 161                                 TDB_DATA key,
 162                                 struct ctdb_ltdb_header *header)
 163 {
 164         ZERO_STRUCTP(header);
 165         /* initial dmaster is the lmaster */
 166         header->dmaster = ctdb_lmaster(ctdb_db->ctdb, &key);
 167         header->flags = CTDB_REC_FLAG_AUTOMATIC;
 168 }
 169
 170
 171 /*
 172   fetch a record from the ltdb, separating out the header information
 173   and returning the body of the record. A valid (initial) header is
 174   returned if the record is not present
 175 */
 176 int ctdb_ltdb_fetch(struct ctdb_db_context *ctdb_db,
 177                     TDB_DATA key, struct ctdb_ltdb_header *header,
 178                     TALLOC_CTX *mem_ctx, TDB_DATA *data)
 179 {
 180         TDB_DATA rec;
 181         struct ctdb_context *ctdb = ctdb_db->ctdb;
 182
 183         rec = tdb_fetch(ctdb_db->ltdb->tdb, key);
 184         if (rec.dsize < sizeof(*header)) {
 185                 /* return an initial header */
 186                 if (rec.dptr) free(rec.dptr);
 187                 if (ctdb->vnn_map == NULL) {
 188                         /* called from the client */
 189                         ZERO_STRUCTP(data);
 190                         header->dmaster = (uint32_t)-1;
 191                         return -1;
 192                 }
 193                 ltdb_initial_header(ctdb_db, key, header);
 194                 if (data) {
 195                         *data = tdb_null;
 196                 }
 197                 if (ctdb_db_persistent(ctdb_db) ||
 198                     header->dmaster == ctdb_db->ctdb->pnn) {
 199                         if (ctdb_ltdb_store(ctdb_db, key, header, tdb_null) != 0) {
 200                                 DEBUG(DEBUG_NOTICE,
 201                                       (__location__ "failed to store initial header\n"));
 202                         }
 203                 }
 204                 return 0;
 205         }
 206
 207         *header = *(struct ctdb_ltdb_header *)rec.dptr;
 208
 209         if (data) {
 210                 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
 211                 data->dptr = talloc_memdup(mem_ctx,
 212                                            sizeof(struct ctdb_ltdb_header)+rec.dptr,
 213                                            data->dsize);
 214         }
 215
 216         free(rec.dptr);
 217         if (data) {
 218                 CTDB_NO_MEMORY(ctdb, data->dptr);
 219         }
 220
 221         return 0;
 222 }
 223
 224 /*
 225   fetch a record from the ltdb, separating out the header information
 226   and returning the body of the record.
 227   if the record does not exist, *header will be NULL
 228   and data = {0, NULL}
 229 */
 230 int ctdb_ltdb_fetch_with_header(struct ctdb_db_context *ctdb_db,
 231                     TDB_DATA key, struct ctdb_ltdb_header *header,
 232                     TALLOC_CTX *mem_ctx, TDB_DATA *data)
 233 {
 234         TDB_DATA rec;
 235
 236         rec = tdb_fetch(ctdb_db->ltdb->tdb, key);
 237         if (rec.dsize < sizeof(*header)) {
 238                 free(rec.dptr);
 239
 240                 data->dsize = 0;
 241                 data->dptr = NULL;
 242                 return -1;
 243         }
 244
 245         *header = *(struct ctdb_ltdb_header *)rec.dptr;
 246         if (data) {
 247                 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
 248                 data->dptr = talloc_memdup(mem_ctx,
 249                                            sizeof(struct ctdb_ltdb_header)+rec.dptr,
 250                                            data->dsize);
 251         }
 252
 253         free(rec.dptr);
 254
 255         return 0;
 256 }
 257
 258
 259 /*
 260   write a record to a normal database
 261 */
 262 int ctdb_ltdb_store(struct ctdb_db_context *ctdb_db, TDB_DATA key,
 263                     struct ctdb_ltdb_header *header, TDB_DATA data)
 264 {
 265         struct ctdb_context *ctdb = ctdb_db->ctdb;
 266         TDB_DATA rec[2];
 267         uint32_t hsize = sizeof(struct ctdb_ltdb_header);
 268         int ret;
 269         bool seqnum_suppressed = false;
 270
 271         if (ctdb_db->ctdb_ltdb_store_fn) {
 272                 return ctdb_db->ctdb_ltdb_store_fn(ctdb_db, key, header, data);
 273         }
 274
 275         if (ctdb->flags & CTDB_FLAG_TORTURE) {
 276                 TDB_DATA old;
 277                 struct ctdb_ltdb_header *h2;
 278
 279                 old = tdb_fetch(ctdb_db->ltdb->tdb, key);
 280                 h2 = (struct ctdb_ltdb_header *)old.dptr;
 281                 if (old.dptr != NULL && old.dsize >= hsize &&
 282                     h2->rsn > header->rsn) {
 283                         DEBUG(DEBUG_ERR,
 284                               ("RSN regression! %"PRIu64" %"PRIu64"\n",
 285                                h2->rsn, header->rsn));
 286                 }
 287                 if (old.dptr != NULL) {
 288                         free(old.dptr);
 289                 }
 290         }
 291
 292         rec[0].dsize = hsize;
 293         rec[0].dptr = (uint8_t *)header;
 294
 295         rec[1].dsize = data.dsize;
 296         rec[1].dptr = data.dptr;
 297
 298         /* Databases with seqnum updates enabled only get their seqnum
 299            changes when/if we modify the data */
 300         if (ctdb_db->seqnum_update != NULL) {
 301                 TDB_DATA old;
 302                 old = tdb_fetch(ctdb_db->ltdb->tdb, key);
 303
 304                 if ((old.dsize == hsize + data.dsize) &&
 305                     memcmp(old.dptr+hsize, data.dptr, data.dsize) == 0) {
 306                         tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
 307                         seqnum_suppressed = true;
 308                 }
 309                 if (old.dptr != NULL) {
 310                         free(old.dptr);
 311                 }
 312         }
 313         ret = tdb_storev(ctdb_db->ltdb->tdb, key, rec, 2, TDB_REPLACE);
 314         if (ret != 0) {
 315                 DEBUG(DEBUG_ERR, (__location__ " Failed to store dynamic data\n"));
 316         }
 317         if (seqnum_suppressed) {
 318                 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
 319         }
 320
 321         return ret;
 322 }
 323
 324 /*
 325   lock a record in the ltdb, given a key
 326  */
 327 int ctdb_ltdb_lock(struct ctdb_db_context *ctdb_db, TDB_DATA key)
 328 {
 329         return tdb_chainlock(ctdb_db->ltdb->tdb, key);
 330 }
 331
 332 /*
 333   unlock a record in the ltdb, given a key
 334  */
 335 int ctdb_ltdb_unlock(struct ctdb_db_context *ctdb_db, TDB_DATA key)
 336 {
 337         int ret = tdb_chainunlock(ctdb_db->ltdb->tdb, key);
 338         if (ret != 0) {
 339                 DEBUG(DEBUG_ERR,("tdb_chainunlock failed on db %s [%s]\n", ctdb_db->db_name, tdb_errorstr(ctdb_db->ltdb->tdb)));
 340         }
 341         return ret;
 342 }
 343
 344
 345 /*
 346   delete a record from a normal database
 347 */
 348 int ctdb_ltdb_delete(struct ctdb_db_context *ctdb_db, TDB_DATA key)
 349 {
 350         if (! ctdb_db_volatile(ctdb_db)) {
 351                 DEBUG(DEBUG_WARNING,
 352                       ("Ignored deletion of empty record from "
 353                        "non-volatile database\n"));
 354                 return 0;
 355         }
 356         if (tdb_delete(ctdb_db->ltdb->tdb, key) != 0) {
 357                 DEBUG(DEBUG_ERR,("Failed to delete empty record."));
 358                 return -1;
 359         }
 360         return 0;
 361 }
 362
 363 int ctdb_trackingdb_add_pnn(struct ctdb_context *ctdb, TDB_DATA *data, uint32_t pnn)
 364 {
 365         int byte_pos = pnn / 8;
 366         int bit_mask   = 1 << (pnn % 8);
 367
 368         if (byte_pos + 1 > data->dsize) {
 369                 char *buf;
 370
 371                 buf = malloc(byte_pos + 1);
 372                 memset(buf, 0, byte_pos + 1);
 373                 if (buf == NULL) {
 374                         DEBUG(DEBUG_ERR, ("Out of memory when allocating buffer of %d bytes for trackingdb\n", byte_pos + 1));
 375                         return -1;
 376                 }
 377                 if (data->dptr != NULL) {
 378                         memcpy(buf, data->dptr, data->dsize);
 379                         free(data->dptr);
 380                 }
 381                 data->dptr  = (uint8_t *)buf;
 382                 data->dsize = byte_pos + 1;
 383         }
 384
 385         data->dptr[byte_pos] |= bit_mask;
 386         return 0;
 387 }
 388
 389 void ctdb_trackingdb_traverse(struct ctdb_context *ctdb, TDB_DATA data, ctdb_trackingdb_cb cb, void *private_data)
 390 {
 391         int i;
 392
 393         for(i = 0; i < data.dsize; i++) {
 394                 int j;
 395
 396                 for (j=0; j<8; j++) {
 397                         int mask = 1<<j;
 398
 399                         if (data.dptr[i] & mask) {
 400                                 cb(ctdb, i * 8 + j, private_data);
 401                         }
 402                 }
 403         }
 404 }
 405
 406 /*
 407   this is the dummy null procedure that all databases support
 408 */
 409 int ctdb_null_func(struct ctdb_call_info *call)
 410 {
 411         return 0;
 412 }
 413
 414 /*
 415   this is a plain fetch procedure that all databases support
 416 */
 417 int ctdb_fetch_func(struct ctdb_call_info *call)
 418 {
 419         call->reply_data = &call->record_data;
 420         return 0;
 421 }
 422
 423 /*
 424   this is a plain fetch procedure that all databases support
 425   this returns the full record including the ltdb header
 426 */
 427 int ctdb_fetch_with_header_func(struct ctdb_call_info *call)
 428 {
 429         call->reply_data = talloc(call, TDB_DATA);
 430         if (call->reply_data == NULL) {
 431                 return -1;
 432         }
 433         call->reply_data->dsize = sizeof(struct ctdb_ltdb_header) + call->record_data.dsize;
 434         call->reply_data->dptr  = talloc_size(call->reply_data, call->reply_data->dsize);
 435         if (call->reply_data->dptr == NULL) {
 436                 return -1;
 437         }
 438         memcpy(call->reply_data->dptr, call->header, sizeof(struct ctdb_ltdb_header));
 439         memcpy(&call->reply_data->dptr[sizeof(struct ctdb_ltdb_header)], call->record_data.dptr, call->record_data.dsize);
 440
 441         return 0;
 442 }
 443