lib/tdb/common/open.c

   1  /*
   2    Unix SMB/CIFS implementation.
   3
   4    trivial database library
   5
   6    Copyright (C) Andrew Tridgell              1999-2005
   7    Copyright (C) Paul `Rusty' Russell              2000
   8    Copyright (C) Jeremy Allison                    2000-2003
   9
  10      ** NOTE! The following LGPL license applies to the tdb
  11      ** library. This does NOT imply that all of Samba is released
  12      ** under the LGPL
  13
  14    This library is free software; you can redistribute it and/or
  15    modify it under the terms of the GNU Lesser General Public
  16    License as published by the Free Software Foundation; either
  17    version 3 of the License, or (at your option) any later version.
  18
  19    This library is distributed in the hope that it will be useful,
  20    but WITHOUT ANY WARRANTY; without even the implied warranty of
  21    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22    Lesser General Public License for more details.
  23
  24    You should have received a copy of the GNU Lesser General Public
  25    License along with this library; if not, see <http://www.gnu.org/licenses/>.
  26 */
  27
  28 #include "tdb_private.h"
  29
  30 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
  31 static struct tdb_context *tdbs = NULL;
  32
  33
  34 /* This is based on the hash algorithm from gdbm */
  35 static unsigned int default_tdb_hash(TDB_DATA *key)
  36 {
  37         uint32_t value; /* Used to compute the hash value.  */
  38         uint32_t   i;   /* Used to cycle through random values. */
  39
  40         /* Set the initial value from the key size. */
  41         for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
  42                 value = (value + (key->dptr[i] << (i*5 % 24)));
  43
  44         return (1103515243 * value + 12345);
  45 }
  46
  47
  48 /* initialise a new database with a specified hash size */
  49 static int tdb_new_database(struct tdb_context *tdb, int hash_size)
  50 {
  51         struct tdb_header *newdb;
  52         size_t size;
  53         int ret = -1;
  54         ssize_t written;
  55
  56         /* We make it up in memory, then write it out if not internal */
  57         size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
  58         if (!(newdb = (struct tdb_header *)calloc(size, 1)))
  59                 return TDB_ERRCODE(TDB_ERR_OOM, -1);
  60
  61         /* Fill in the header */
  62         newdb->version = TDB_VERSION;
  63         newdb->hash_size = hash_size;
  64         if (tdb->flags & TDB_INTERNAL) {
  65                 tdb->map_size = size;
  66                 tdb->map_ptr = (char *)newdb;
  67                 memcpy(&tdb->header, newdb, sizeof(tdb->header));
  68                 /* Convert the `ondisk' version if asked. */
  69                 CONVERT(*newdb);
  70                 return 0;
  71         }
  72         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
  73                 goto fail;
  74
  75         if (ftruncate(tdb->fd, 0) == -1)
  76                 goto fail;
  77
  78         /* This creates an endian-converted header, as if read from disk */
  79         CONVERT(*newdb);
  80         memcpy(&tdb->header, newdb, sizeof(tdb->header));
  81         /* Don't endian-convert the magic food! */
  82         memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
  83         /* we still have "ret == -1" here */
  84         written = write(tdb->fd, newdb, size);
  85         if (written == size) {
  86                 ret = 0;
  87         } else if (written != -1) {
  88                 /* call write once again, this usually should return -1 and
  89                  * set errno appropriately */
  90                 size -= written;
  91                 written = write(tdb->fd, newdb+written, size);
  92                 if (written == size) {
  93                 ret = 0;
  94                 } else if (written >= 0) {
  95                         /* a second incomplete write - we give up.
  96                          * guessing the errno... */
  97                         errno = ENOSPC;
  98                 }
  99         }
 100
 101   fail:
 102         SAFE_FREE(newdb);
 103         return ret;
 104 }
 105
 106
 107
 108 static int tdb_already_open(dev_t device,
 109                             ino_t ino)
 110 {
 111         struct tdb_context *i;
 112
 113         for (i = tdbs; i; i = i->next) {
 114                 if (i->device == device && i->inode == ino) {
 115                         return 1;
 116                 }
 117         }
 118
 119         return 0;
 120 }
 121
 122 /* open the database, creating it if necessary
 123
 124    The open_flags and mode are passed straight to the open call on the
 125    database file. A flags value of O_WRONLY is invalid. The hash size
 126    is advisory, use zero for a default value.
 127
 128    Return is NULL on error, in which case errno is also set.  Don't
 129    try to call tdb_error or tdb_errname, just do strerror(errno).
 130
 131    @param name may be NULL for internal databases. */
 132 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
 133                       int open_flags, mode_t mode)
 134 {
 135         return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
 136 }
 137
 138 /* a default logging function */
 139 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
 140 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
 141 {
 142 }
 143
 144
 145 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 146                                 int open_flags, mode_t mode,
 147                                 const struct tdb_logging_context *log_ctx,
 148                                 tdb_hash_func hash_fn)
 149 {
 150         struct tdb_context *tdb;
 151         struct stat st;
 152         int rev = 0, locked = 0;
 153         unsigned char *vp;
 154         uint32_t vertest;
 155         unsigned v;
 156
 157         if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
 158                 /* Can't log this */
 159                 errno = ENOMEM;
 160                 goto fail;
 161         }
 162         tdb_io_init(tdb);
 163         tdb->fd = -1;
 164         tdb->name = NULL;
 165         tdb->map_ptr = NULL;
 166         tdb->flags = tdb_flags;
 167         tdb->open_flags = open_flags;
 168         if (log_ctx) {
 169                 tdb->log = *log_ctx;
 170         } else {
 171                 tdb->log.log_fn = null_log_fn;
 172                 tdb->log.log_private = NULL;
 173         }
 174         tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
 175
 176         /* cache the page size */
 177         tdb->page_size = getpagesize();
 178         if (tdb->page_size <= 0) {
 179                 tdb->page_size = 0x2000;
 180         }
 181
 182         tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
 183
 184         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 185                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
 186                          name));
 187                 errno = EINVAL;
 188                 goto fail;
 189         }
 190
 191         if (hash_size == 0)
 192                 hash_size = DEFAULT_HASH_SIZE;
 193         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 194                 tdb->read_only = 1;
 195                 /* read only databases don't do locking or clear if first */
 196                 tdb->flags |= TDB_NOLOCK;
 197                 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
 198         }
 199
 200         /* internal databases don't mmap or lock, and start off cleared */
 201         if (tdb->flags & TDB_INTERNAL) {
 202                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 203                 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
 204                 if (tdb_new_database(tdb, hash_size) != 0) {
 205                         TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
 206                         goto fail;
 207                 }
 208                 goto internal;
 209         }
 210
 211         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 212                 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
 213                          name, strerror(errno)));
 214                 goto fail;      /* errno set by open(2) */
 215         }
 216
 217         /* on exec, don't inherit the fd */
 218         v = fcntl(tdb->fd, F_GETFD, 0);
 219         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 220
 221         /* ensure there is only one process initialising at once */
 222         if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
 223                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
 224                          name, strerror(errno)));
 225                 goto fail;      /* errno set by tdb_brlock */
 226         }
 227
 228         /* we need to zero database if we are the only one with it open */
 229         if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
 230             (!tdb->read_only) &&
 231             (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
 232                 open_flags |= O_CREAT;
 233                 if (ftruncate(tdb->fd, 0) == -1) {
 234                         TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
 235                                  "failed to truncate %s: %s\n",
 236                                  name, strerror(errno)));
 237                         goto fail; /* errno set by ftruncate */
 238                 }
 239         }
 240
 241         errno = 0;
 242         if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
 243             || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
 244             || (tdb->header.version != TDB_VERSION
 245                 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
 246                 /* its not a valid database - possibly initialise it */
 247                 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
 248                         if (errno == 0) {
 249                         errno = EIO; /* ie bad format or something */
 250                         }
 251                         goto fail;
 252                 }
 253                 rev = (tdb->flags & TDB_CONVERT);
 254         }
 255         vp = (unsigned char *)&tdb->header.version;
 256         vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
 257                   (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
 258         tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
 259         if (!rev)
 260                 tdb->flags &= ~TDB_CONVERT;
 261         else {
 262                 tdb->flags |= TDB_CONVERT;
 263                 tdb_convert(&tdb->header, sizeof(tdb->header));
 264         }
 265         if (fstat(tdb->fd, &st) == -1)
 266                 goto fail;
 267
 268         if (tdb->header.rwlocks != 0) {
 269                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
 270                 goto fail;
 271         }
 272
 273         /* Is it already in the open list?  If so, fail. */
 274         if (tdb_already_open(st.st_dev, st.st_ino)) {
 275                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
 276                          "%s (%d,%d) is already open in this process\n",
 277                          name, (int)st.st_dev, (int)st.st_ino));
 278                 errno = EBUSY;
 279                 goto fail;
 280         }
 281
 282         if (!(tdb->name = (char *)strdup(name))) {
 283                 errno = ENOMEM;
 284                 goto fail;
 285         }
 286
 287         tdb->map_size = st.st_size;
 288         tdb->device = st.st_dev;
 289         tdb->inode = st.st_ino;
 290         tdb_mmap(tdb);
 291         if (locked) {
 292                 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
 293                         TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
 294                                  "failed to take ACTIVE_LOCK on %s: %s\n",
 295                                  name, strerror(errno)));
 296                         goto fail;
 297                 }
 298
 299         }
 300
 301         /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
 302            we didn't get the initial exclusive lock as we need to let all other
 303            users know we're using it. */
 304
 305         if (tdb_flags & TDB_CLEAR_IF_FIRST) {
 306                 /* leave this lock in place to indicate it's in use */
 307                 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
 308                         goto fail;
 309         }
 310
 311         /* if needed, run recovery */
 312         if (tdb_transaction_recover(tdb) == -1) {
 313                 goto fail;
 314         }
 315
 316  internal:
 317         /* Internal (memory-only) databases skip all the code above to
 318          * do with disk files, and resume here by releasing their
 319          * global lock and hooking into the active list. */
 320         if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
 321                 goto fail;
 322         tdb->next = tdbs;
 323         tdbs = tdb;
 324         return tdb;
 325
 326  fail:
 327         { int save_errno = errno;
 328
 329         if (!tdb)
 330                 return NULL;
 331
 332         if (tdb->map_ptr) {
 333                 if (tdb->flags & TDB_INTERNAL)
 334                         SAFE_FREE(tdb->map_ptr);
 335                 else
 336                         tdb_munmap(tdb);
 337         }
 338         SAFE_FREE(tdb->name);
 339         if (tdb->fd != -1)
 340                 if (close(tdb->fd) != 0)
 341                         TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
 342         SAFE_FREE(tdb);
 343         errno = save_errno;
 344         return NULL;
 345         }
 346 }
 347
 348 /*
 349  * Set the maximum number of dead records per hash chain
 350  */
 351
 352 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
 353 {
 354         tdb->max_dead_records = max_dead;
 355 }
 356
 357 /**
 358  * Close a database.
 359  *
 360  * @returns -1 for error; 0 for success.
 361  **/
 362 int tdb_close(struct tdb_context *tdb)
 363 {
 364         struct tdb_context **i;
 365         int ret = 0;
 366
 367         if (tdb->transaction) {
 368                 tdb_transaction_cancel(tdb);
 369         }
 370
 371         if (tdb->map_ptr) {
 372                 if (tdb->flags & TDB_INTERNAL)
 373                         SAFE_FREE(tdb->map_ptr);
 374                 else
 375                         tdb_munmap(tdb);
 376         }
 377         SAFE_FREE(tdb->name);
 378         if (tdb->fd != -1)
 379                 ret = close(tdb->fd);
 380         SAFE_FREE(tdb->lockrecs);
 381
 382         /* Remove from contexts list */
 383         for (i = &tdbs; *i; i = &(*i)->next) {
 384                 if (*i == tdb) {
 385                         *i = tdb->next;
 386                         break;
 387                 }
 388         }
 389
 390         memset(tdb, 0, sizeof(*tdb));
 391         SAFE_FREE(tdb);
 392
 393         return ret;
 394 }
 395
 396 /* register a loging function */
 397 void tdb_set_logging_function(struct tdb_context *tdb,
 398                               const struct tdb_logging_context *log_ctx)
 399 {
 400         tdb->log = *log_ctx;
 401 }
 402
 403 void *tdb_get_logging_private(struct tdb_context *tdb)
 404 {
 405         return tdb->log.log_private;
 406 }
 407
 408 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
 409    seek pointer from our parent and to re-establish locks */
 410 int tdb_reopen(struct tdb_context *tdb)
 411 {
 412         struct stat st;
 413
 414         if (tdb->flags & TDB_INTERNAL) {
 415                 return 0; /* Nothing to do. */
 416         }
 417
 418         if (tdb->num_locks != 0 || tdb->global_lock.count) {
 419                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
 420                 goto fail;
 421         }
 422
 423         if (tdb->transaction != 0) {
 424                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
 425                 goto fail;
 426         }
 427
 428 /* If we have real pread & pwrite, we can skip reopen. */
 429 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
 430         !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
 431         if (tdb_munmap(tdb) != 0) {
 432                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
 433                 goto fail;
 434         }
 435         if (close(tdb->fd) != 0)
 436                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
 437         tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
 438         if (tdb->fd == -1) {
 439                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
 440                 goto fail;
 441         }
 442         if (fstat(tdb->fd, &st) != 0) {
 443                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
 444                 goto fail;
 445         }
 446         if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
 447                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
 448                 goto fail;
 449         }
 450         tdb_mmap(tdb);
 451 #endif /* fake pread or pwrite */
 452
 453         if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
 454             (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
 455                 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
 456                 goto fail;
 457         }
 458
 459         return 0;
 460
 461 fail:
 462         tdb_close(tdb);
 463         return -1;
 464 }
 465
 466 /* reopen all tdb's */
 467 int tdb_reopen_all(int parent_longlived)
 468 {
 469         struct tdb_context *tdb;
 470
 471         for (tdb=tdbs; tdb; tdb = tdb->next) {
 472                 /*
 473                  * If the parent is longlived (ie. a
 474                  * parent daemon architecture), we know
 475                  * it will keep it's active lock on a
 476                  * tdb opened with CLEAR_IF_FIRST. Thus
 477                  * for child processes we don't have to
 478                  * add an active lock. This is essential
 479                  * to improve performance on systems that
 480                  * keep POSIX locks as a non-scalable data
 481                  * structure in the kernel.
 482                  */
 483                 if (parent_longlived) {
 484                         /* Ensure no clear-if-first. */
 485                         tdb->flags &= ~TDB_CLEAR_IF_FIRST;
 486                 }
 487
 488                 if (tdb_reopen(tdb) != 0)
 489                         return -1;
 490         }
 491
 492         return 0;
 493 }