git.ozlabs.org Git - ccan/blob - ccan/tdb2/tdb1_io.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
  29 #include "tdb1_private.h"
  30 #ifndef MAX
  31 #define MAX(a,b) ((a) > (b) ? (a) : (b))
  32 #endif
  33
  34 /* check for an out of bounds access - if it is out of bounds then
  35    see if the database has been expanded by someone else and expand
  36    if necessary
  37    note that "len" is the minimum length needed for the db
  38 */
  39 static int tdb1_oob(struct tdb_context *tdb, tdb1_off_t off, tdb1_len_t len,
  40                     int probe)
  41 {
  42         struct stat st;
  43         if (len + off < len) {
  44                 if (!probe) {
  45                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
  46                                                      "tdb1_oob off %d len %d wrap\n",
  47                                                      (int)off, (int)len);
  48                 }
  49                 return -1;
  50         }
  51
  52         if (off + len <= tdb->file->map_size)
  53                 return 0;
  54         if (tdb->flags & TDB_INTERNAL) {
  55                 if (!probe) {
  56                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
  57                                                      "tdb1_oob len %d beyond internal malloc size %u",
  58                                                      (int)(off + len), (int)tdb->file->map_size);
  59                 }
  60                 return -1;
  61         }
  62
  63         if (fstat(tdb->file->fd, &st) == -1) {
  64                 tdb->last_error = TDB_ERR_IO;
  65                 return -1;
  66         }
  67
  68         if (st.st_size < (size_t)off + len) {
  69                 if (!probe) {
  70                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
  71                                                      "tdb1_oob len %u beyond eof at %u",
  72                                                      (int)(off + len), (int)st.st_size);
  73                 }
  74                 return -1;
  75         }
  76
  77         /* Beware >4G files! */
  78         if ((tdb1_off_t)st.st_size != st.st_size) {
  79                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
  80                                              "tdb1_oob len %llu too large!\n",
  81                                              (long long)st.st_size);
  82                 return -1;
  83         }
  84
  85         /* Unmap, update size, remap */
  86         if (tdb1_munmap(tdb) == -1) {
  87                 tdb->last_error = TDB_ERR_IO;
  88                 return -1;
  89         }
  90         tdb->file->map_size = st.st_size;
  91         tdb1_mmap(tdb);
  92         return 0;
  93 }
  94
  95 /* write a lump of data at a specified offset */
  96 static int tdb1_write(struct tdb_context *tdb, tdb1_off_t off,
  97                      const void *buf, tdb1_len_t len)
  98 {
  99         if (len == 0) {
 100                 return 0;
 101         }
 102
 103         if ((tdb->flags & TDB_RDONLY) || tdb->tdb1.traverse_read) {
 104                 tdb->last_error = TDB_ERR_RDONLY;
 105                 return -1;
 106         }
 107
 108         if (tdb->tdb1.io->tdb1_oob(tdb, off, len, 0) != 0)
 109                 return -1;
 110
 111         if (tdb->file->map_ptr) {
 112                 memcpy(off + (char *)tdb->file->map_ptr, buf, len);
 113         } else {
 114                 ssize_t written = pwrite(tdb->file->fd, buf, len, off);
 115                 if ((written != (ssize_t)len) && (written != -1)) {
 116                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_WARNING,
 117                                    "tdb1_write: wrote only "
 118                                    "%d of %d bytes at %d, trying once more",
 119                                    (int)written, len, off);
 120                         written = pwrite(tdb->file->fd,
 121                                          (const char *)buf+written,
 122                                          len-written,
 123                                          off+written);
 124                 }
 125                 if (written == -1) {
 126                         /* Ensure ecode is set for log fn. */
 127                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 128                                                 "tdb1_write failed at %d "
 129                                                 "len=%d (%s)",
 130                                                 off, len, strerror(errno));
 131                         return -1;
 132                 } else if (written != (ssize_t)len) {
 133                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 134                                                 "tdb1_write: failed to "
 135                                                 "write %d bytes at %d in two attempts",
 136                                                 len, off);
 137                         return -1;
 138                 }
 139         }
 140         return 0;
 141 }
 142
 143 /* Endian conversion: we only ever deal with 4 byte quantities */
 144 void *tdb1_convert(void *buf, uint32_t size)
 145 {
 146         uint32_t i, *p = (uint32_t *)buf;
 147         for (i = 0; i < size / 4; i++)
 148                 p[i] = TDB1_BYTEREV(p[i]);
 149         return buf;
 150 }
 151
 152
 153 /* read a lump of data at a specified offset, maybe convert */
 154 static int tdb1_read(struct tdb_context *tdb, tdb1_off_t off, void *buf,
 155                     tdb1_len_t len, int cv)
 156 {
 157         if (tdb->tdb1.io->tdb1_oob(tdb, off, len, 0) != 0) {
 158                 return -1;
 159         }
 160
 161         if (tdb->file->map_ptr) {
 162                 memcpy(buf, off + (char *)tdb->file->map_ptr, len);
 163         } else {
 164                 ssize_t ret = pread(tdb->file->fd, buf, len, off);
 165                 if (ret != (ssize_t)len) {
 166                         /* Ensure ecode is set for log fn. */
 167                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 168                                                 "tdb1_read failed at %d "
 169                                                 "len=%d ret=%d (%s) map_size=%d",
 170                                                 (int)off, (int)len, (int)ret,
 171                                                 strerror(errno),
 172                                                 (int)tdb->file->map_size);
 173                         return -1;
 174                 }
 175         }
 176         if (cv) {
 177                 tdb1_convert(buf, len);
 178         }
 179         return 0;
 180 }
 181
 182
 183
 184 /*
 185   do an unlocked scan of the hash table heads to find the next non-zero head. The value
 186   will then be confirmed with the lock held
 187 */
 188 static void tdb1_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
 189 {
 190         uint32_t h = *chain;
 191         if (tdb->file->map_ptr) {
 192                 for (;h < tdb->tdb1.header.hash_size;h++) {
 193                         if (0 != *(uint32_t *)(TDB1_HASH_TOP(h) + (unsigned char *)tdb->file->map_ptr)) {
 194                                 break;
 195                         }
 196                 }
 197         } else {
 198                 uint32_t off=0;
 199                 for (;h < tdb->tdb1.header.hash_size;h++) {
 200                         if (tdb1_ofs_read(tdb, TDB1_HASH_TOP(h), &off) != 0 || off != 0) {
 201                                 break;
 202                         }
 203                 }
 204         }
 205         (*chain) = h;
 206 }
 207
 208
 209 int tdb1_munmap(struct tdb_context *tdb)
 210 {
 211         if (tdb->flags & TDB_INTERNAL)
 212                 return 0;
 213
 214 #if HAVE_MMAP
 215         if (tdb->file->map_ptr) {
 216                 int ret;
 217
 218                 ret = munmap(tdb->file->map_ptr, tdb->file->map_size);
 219                 if (ret != 0)
 220                         return ret;
 221         }
 222 #endif
 223         tdb->file->map_ptr = NULL;
 224         return 0;
 225 }
 226
 227 void tdb1_mmap(struct tdb_context *tdb)
 228 {
 229         if (tdb->flags & TDB_INTERNAL)
 230                 return;
 231
 232 #if HAVE_MMAP
 233         if (!(tdb->flags & TDB_NOMMAP)) {
 234                 int mmap_flags;
 235                 if ((tdb->open_flags & O_ACCMODE) == O_RDONLY)
 236                         mmap_flags = PROT_READ;
 237                 else
 238                         mmap_flags = PROT_READ | PROT_WRITE;
 239
 240                 tdb->file->map_ptr = mmap(NULL, tdb->file->map_size,
 241                                     mmap_flags,
 242                                     MAP_SHARED|MAP_FILE, tdb->file->fd, 0);
 243
 244                 /*
 245                  * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
 246                  */
 247
 248                 if (tdb->file->map_ptr == MAP_FAILED) {
 249                         tdb->file->map_ptr = NULL;
 250                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_WARNING,
 251                                    "tdb1_mmap failed for size %d (%s)",
 252                                    tdb->file->map_size, strerror(errno));
 253                 }
 254         } else {
 255                 tdb->file->map_ptr = NULL;
 256         }
 257 #else
 258         tdb->file->map_ptr = NULL;
 259 #endif
 260 }
 261
 262 /* expand a file.  we prefer to use ftruncate, as that is what posix
 263   says to use for mmap expansion */
 264 static int tdb1_expand_file(struct tdb_context *tdb, tdb1_off_t size, tdb1_off_t addition)
 265 {
 266         char buf[8192];
 267
 268         if ((tdb->flags & TDB_RDONLY) || tdb->tdb1.traverse_read) {
 269                 tdb->last_error = TDB_ERR_RDONLY;
 270                 return -1;
 271         }
 272
 273         if (ftruncate(tdb->file->fd, size+addition) == -1) {
 274                 char b = 0;
 275                 ssize_t written = pwrite(tdb->file->fd, &b, 1,
 276                                          (size+addition) - 1);
 277                 if (written == 0) {
 278                         /* try once more, potentially revealing errno */
 279                         written = pwrite(tdb->file->fd, &b, 1,
 280                                          (size+addition) - 1);
 281                 }
 282                 if (written == 0) {
 283                         /* again - give up, guessing errno */
 284                         errno = ENOSPC;
 285                 }
 286                 if (written != 1) {
 287                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 288                                                 "expand_file to %d failed (%s)",
 289                                                 size+addition,
 290                                                 strerror(errno));
 291                         return -1;
 292                 }
 293         }
 294
 295         /* now fill the file with something. This ensures that the
 296            file isn't sparse, which would be very bad if we ran out of
 297            disk. This must be done with write, not via mmap */
 298         memset(buf, TDB1_PAD_BYTE, sizeof(buf));
 299         while (addition) {
 300                 size_t n = addition>sizeof(buf)?sizeof(buf):addition;
 301                 ssize_t written = pwrite(tdb->file->fd, buf, n, size);
 302                 if (written == 0) {
 303                         /* prevent infinite loops: try _once_ more */
 304                         written = pwrite(tdb->file->fd, buf, n, size);
 305                 }
 306                 if (written == 0) {
 307                         /* give up, trying to provide a useful errno */
 308                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 309                                                 "expand_file write "
 310                                                 "returned 0 twice: giving up!");
 311                         errno = ENOSPC;
 312                         return -1;
 313                 } else if (written == -1) {
 314                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 315                                                 "expand_file write of "
 316                                                 "%d bytes failed (%s)", (int)n,
 317                                                 strerror(errno));
 318                         return -1;
 319                 } else if (written != n) {
 320                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_WARNING,
 321                                    "expand_file: wrote "
 322                                    "only %d of %d bytes - retrying",
 323                                    (int)written, (int)n);
 324                 }
 325                 addition -= written;
 326                 size += written;
 327         }
 328         tdb->stats.expands++;
 329         return 0;
 330 }
 331
 332
 333 /* expand the database at least size bytes by expanding the underlying
 334    file and doing the mmap again if necessary */
 335 int tdb1_expand(struct tdb_context *tdb, tdb1_off_t size)
 336 {
 337         struct tdb1_record rec;
 338         tdb1_off_t offset, new_size, top_size, map_size;
 339
 340         if (tdb1_lock(tdb, -1, F_WRLCK) == -1) {
 341                 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
 342                            "lock failed in tdb1_expand");
 343                 return -1;
 344         }
 345
 346         /* must know about any previous expansions by another process */
 347         tdb->tdb1.io->tdb1_oob(tdb, tdb->file->map_size, 1, 1);
 348
 349         /* limit size in order to avoid using up huge amounts of memory for
 350          * in memory tdbs if an oddball huge record creeps in */
 351         if (size > 100 * 1024) {
 352                 top_size = tdb->file->map_size + size * 2;
 353         } else {
 354                 top_size = tdb->file->map_size + size * 100;
 355         }
 356
 357         /* always make room for at least top_size more records, and at
 358            least 25% more space. if the DB is smaller than 100MiB,
 359            otherwise grow it by 10% only. */
 360         if (tdb->file->map_size > 100 * 1024 * 1024) {
 361                 map_size = tdb->file->map_size * 1.10;
 362         } else {
 363                 map_size = tdb->file->map_size * 1.25;
 364         }
 365
 366         /* Round the database up to a multiple of the page size */
 367         new_size = MAX(top_size, map_size);
 368         size = TDB1_ALIGN(new_size, tdb->tdb1.page_size) - tdb->file->map_size;
 369
 370         if (!(tdb->flags & TDB_INTERNAL))
 371                 tdb1_munmap(tdb);
 372
 373         /*
 374          * We must ensure the file is unmapped before doing this
 375          * to ensure consistency with systems like OpenBSD where
 376          * writes and mmaps are not consistent.
 377          */
 378
 379         /* expand the file itself */
 380         if (!(tdb->flags & TDB_INTERNAL)) {
 381                 if (tdb->tdb1.io->tdb1_expand_file(tdb, tdb->file->map_size, size) != 0)
 382                         goto fail;
 383         }
 384
 385         tdb->file->map_size += size;
 386
 387         if (tdb->flags & TDB_INTERNAL) {
 388                 char *new_map_ptr = (char *)realloc(tdb->file->map_ptr,
 389                                                     tdb->file->map_size);
 390                 if (!new_map_ptr) {
 391                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM,
 392                                                      TDB_LOG_ERROR,
 393                                                      "tdb1_expand: no memory");
 394                         tdb->file->map_size -= size;
 395                         goto fail;
 396                 }
 397                 tdb->file->map_ptr = new_map_ptr;
 398         } else {
 399                 /*
 400                  * We must ensure the file is remapped before adding the space
 401                  * to ensure consistency with systems like OpenBSD where
 402                  * writes and mmaps are not consistent.
 403                  */
 404
 405                 /* We're ok if the mmap fails as we'll fallback to read/write */
 406                 tdb1_mmap(tdb);
 407         }
 408
 409         /* form a new freelist record */
 410         memset(&rec,'\0',sizeof(rec));
 411         rec.rec_len = size - sizeof(rec);
 412
 413         /* link it into the free list */
 414         offset = tdb->file->map_size - size;
 415         if (tdb1_free(tdb, offset, &rec) == -1)
 416                 goto fail;
 417
 418         tdb1_unlock(tdb, -1, F_WRLCK);
 419         return 0;
 420  fail:
 421         tdb1_unlock(tdb, -1, F_WRLCK);
 422         return -1;
 423 }
 424
 425 /* read/write a tdb1_off_t */
 426 int tdb1_ofs_read(struct tdb_context *tdb, tdb1_off_t offset, tdb1_off_t *d)
 427 {
 428         return tdb->tdb1.io->tdb1_read(tdb, offset, (char*)d, sizeof(*d), TDB1_DOCONV());
 429 }
 430
 431 int tdb1_ofs_write(struct tdb_context *tdb, tdb1_off_t offset, tdb1_off_t *d)
 432 {
 433         tdb1_off_t off = *d;
 434         return tdb->tdb1.io->tdb1_write(tdb, offset, TDB1_CONV(off), sizeof(*d));
 435 }
 436
 437
 438 /* read a lump of data, allocating the space for it */
 439 unsigned char *tdb1_alloc_read(struct tdb_context *tdb, tdb1_off_t offset, tdb1_len_t len)
 440 {
 441         unsigned char *buf;
 442
 443         /* some systems don't like zero length malloc */
 444
 445         if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
 446                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 447                                              "tdb1_alloc_read malloc failed"
 448                                              " len=%d (%s)",
 449                                              len, strerror(errno));
 450                 return NULL;
 451         }
 452         if (tdb->tdb1.io->tdb1_read(tdb, offset, buf, len, 0) == -1) {
 453                 SAFE_FREE(buf);
 454                 return NULL;
 455         }
 456         return buf;
 457 }
 458
 459 /* Give a piece of tdb data to a parser */
 460 enum TDB_ERROR tdb1_parse_data(struct tdb_context *tdb, TDB_DATA key,
 461                                tdb1_off_t offset, tdb1_len_t len,
 462                                enum TDB_ERROR (*parser)(TDB_DATA key,
 463                                                         TDB_DATA data,
 464                                                         void *private_data),
 465                                void *private_data)
 466 {
 467         TDB_DATA data;
 468         enum TDB_ERROR result;
 469
 470         data.dsize = len;
 471
 472         if ((tdb->tdb1.transaction == NULL) && (tdb->file->map_ptr != NULL)) {
 473                 /*
 474                  * Optimize by avoiding the malloc/memcpy/free, point the
 475                  * parser directly at the mmap area.
 476                  */
 477                 if (tdb->tdb1.io->tdb1_oob(tdb, offset, len, 0) != 0) {
 478                         return tdb->last_error;
 479                 }
 480                 data.dptr = offset + (unsigned char *)tdb->file->map_ptr;
 481                 return parser(key, data, private_data);
 482         }
 483
 484         if (!(data.dptr = tdb1_alloc_read(tdb, offset, len))) {
 485                 return tdb->last_error;
 486         }
 487
 488         result = parser(key, data, private_data);
 489         free(data.dptr);
 490         return result;
 491 }
 492
 493 /* read/write a record */
 494 int tdb1_rec_read(struct tdb_context *tdb, tdb1_off_t offset, struct tdb1_record *rec)
 495 {
 496         if (tdb->tdb1.io->tdb1_read(tdb, offset, rec, sizeof(*rec),TDB1_DOCONV()) == -1)
 497                 return -1;
 498         if (TDB1_BAD_MAGIC(rec)) {
 499                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 500                                         "tdb1_rec_read bad magic 0x%x at offset=%d",
 501                                         rec->magic, offset);
 502                 return -1;
 503         }
 504         return tdb->tdb1.io->tdb1_oob(tdb, rec->next, sizeof(*rec), 0);
 505 }
 506
 507 int tdb1_rec_write(struct tdb_context *tdb, tdb1_off_t offset, struct tdb1_record *rec)
 508 {
 509         struct tdb1_record r = *rec;
 510         return tdb->tdb1.io->tdb1_write(tdb, offset, TDB1_CONV(r), sizeof(r));
 511 }
 512
 513 static const struct tdb1_methods io1_methods = {
 514         tdb1_read,
 515         tdb1_write,
 516         tdb1_next_hash_chain,
 517         tdb1_oob,
 518         tdb1_expand_file,
 519 };
 520
 521 /*
 522   initialise the default methods table
 523 */
 524 void tdb1_io_init(struct tdb_context *tdb)
 525 {
 526         tdb->tdb1.io = &io1_methods;
 527 }
 528
 529 enum TDB_ERROR tdb1_probe_length(struct tdb_context *tdb)
 530 {
 531         tdb->last_error = TDB_SUCCESS;
 532         tdb->tdb1.io->tdb1_oob(tdb, tdb->file->map_size, 1, true);
 533         return tdb->last_error;
 534 }