lib/tdb2/private.h

   1 #ifndef TDB_PRIVATE_H
   2 #define TDB_PRIVATE_H
   3  /*
   4    Trivial Database 2: private types and prototypes
   5    Copyright (C) Rusty Russell 2010
   6
   7    This library is free software; you can redistribute it and/or
   8    modify it under the terms of the GNU Lesser General Public
   9    License as published by the Free Software Foundation; either
  10    version 3 of the License, or (at your option) any later version.
  11
  12    This library 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 GNU
  15    Lesser General Public License for more details.
  16
  17    You should have received a copy of the GNU Lesser General Public
  18    License along with this library; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "config.h"
  22 #ifndef HAVE_CCAN
  23 #error You need ccan to build tdb2!
  24 #endif
  25 #include "tdb2.h"
  26 #include <ccan/compiler/compiler.h>
  27 #include <ccan/likely/likely.h>
  28 #include <ccan/endian/endian.h>
  29
  30 #ifdef HAVE_LIBREPLACE
  31 #include "replace.h"
  32 #include "system/filesys.h"
  33 #include "system/time.h"
  34 #include "system/shmem.h"
  35 #include "system/select.h"
  36 #include "system/wait.h"
  37 #else
  38 #include <stdint.h>
  39 #include <stdbool.h>
  40 #include <stdlib.h>
  41 #include <stddef.h>
  42 #include <sys/time.h>
  43 #include <sys/mman.h>
  44 #include <unistd.h>
  45 #include <fcntl.h>
  46 #include <errno.h>
  47 #include <stdio.h>
  48 #include <utime.h>
  49 #include <unistd.h>
  50 #endif
  51
  52 #ifndef TEST_IT
  53 #define TEST_IT(cond)
  54 #endif
  55
  56 /* #define TDB_TRACE 1 */
  57
  58 #ifndef __STRING
  59 #define __STRING(x)    #x
  60 #endif
  61
  62 #ifndef __STRINGSTRING
  63 #define __STRINGSTRING(x) __STRING(x)
  64 #endif
  65
  66 #ifndef __location__
  67 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
  68 #endif
  69
  70 typedef uint64_t tdb_len_t;
  71 typedef uint64_t tdb_off_t;
  72
  73 #define TDB_MAGIC_FOOD "TDB file\n"
  74 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
  75 #define TDB1_VERSION (0x26011967 + 6)
  76 #define TDB_USED_MAGIC ((uint64_t)0x1999)
  77 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
  78 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
  79 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
  80 #define TDB_CAP_MAGIC ((uint64_t)0x1555)
  81 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
  82 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
  83 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
  84 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
  85
  86 /* Capability bits. */
  87 #define TDB_CAP_TYPE_MASK       0x1FFFFFFFFFFFFFFFULL
  88 #define TDB_CAP_NOCHECK         0x8000000000000000ULL
  89 #define TDB_CAP_NOWRITE         0x4000000000000000ULL
  90 #define TDB_CAP_NOOPEN          0x2000000000000000ULL
  91
  92 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)(long)TDB_ERR_LAST)
  93 #define TDB_OFF_TO_ERR(off) ((enum TDB_ERROR)(long)(off))
  94 #define TDB_ERR_TO_OFF(ecode) ((tdb_off_t)(long)(ecode))
  95
  96 /* Packing errors into pointers and v.v. */
  97 #define TDB_PTR_IS_ERR(ptr) \
  98         unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST)
  99 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
 100 #define TDB_ERR_PTR(err) ((void *)(long)(err))
 101
 102 /* Common case of returning true, false or -ve error. */
 103 typedef int tdb_bool_err;
 104
 105 /* Prevent others from opening the file. */
 106 #define TDB_OPEN_LOCK 0
 107 /* Expanding file. */
 108 #define TDB_EXPANSION_LOCK 2
 109 /* Doing a transaction. */
 110 #define TDB_TRANSACTION_LOCK 8
 111 /* Hash chain locks. */
 112 #define TDB_HASH_LOCK_START 64
 113
 114 /* Range for hash locks. */
 115 #define TDB_HASH_LOCK_RANGE_BITS 30
 116 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
 117
 118 /* We have 1024 entries in the top level. */
 119 #define TDB_TOPLEVEL_HASH_BITS 10
 120 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
 121 #define TDB_SUBLEVEL_HASH_BITS 6
 122 /* And 8 entries in each group, ie 8 groups per sublevel. */
 123 #define TDB_HASH_GROUP_BITS 3
 124 /* This is currently 10: beyond this we chain. */
 125 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
 126
 127 /* Extend file by least 100 times larger than needed. */
 128 #define TDB_EXTENSION_FACTOR 100
 129
 130 /* We steal bits from the offsets to store hash info. */
 131 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
 132 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
 133 #define TDB_OFF_UPPER_STEAL 8
 134 #define   TDB_OFF_UPPER_STEAL_EXTRA 7
 135 /* The bit number where we store extra hash bits. */
 136 #define TDB_OFF_HASH_EXTRA_BIT 57
 137 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
 138
 139 /* Additional features we understand.  Currently: none. */
 140 #define TDB_FEATURE_MASK ((uint64_t)0)
 141
 142 /* The bit number where we store the extra hash bits. */
 143 /* Convenience mask to get actual offset. */
 144 #define TDB_OFF_MASK \
 145         (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
 146
 147 /* How many buckets in a free list: see size_to_bucket(). */
 148 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
 149
 150 /* We have to be able to fit a free record here. */
 151 #define TDB_MIN_DATA_LEN        \
 152         (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
 153
 154 /* Indicates this entry is not on an flist (can happen during coalescing) */
 155 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
 156
 157 struct tdb_used_record {
 158         /* For on-disk compatibility, we avoid bitfields:
 159            magic: 16,        (highest)
 160            key_len_bits: 5,
 161            extra_padding: 32
 162            hash_bits: 11
 163         */
 164         uint64_t magic_and_meta;
 165         /* The bottom key_len_bits*2 are key length, rest is data length. */
 166         uint64_t key_and_data_len;
 167 };
 168
 169 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
 170 {
 171         return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
 172 }
 173
 174 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
 175 {
 176         return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
 177 }
 178
 179 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
 180 {
 181         return r->key_and_data_len >> rec_key_bits(r);
 182 }
 183
 184 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
 185 {
 186         return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
 187 }
 188
 189 static inline uint32_t rec_hash(const struct tdb_used_record *r)
 190 {
 191         return r->magic_and_meta & ((1 << 11) - 1);
 192 }
 193
 194 static inline uint16_t rec_magic(const struct tdb_used_record *r)
 195 {
 196         return (r->magic_and_meta >> 48);
 197 }
 198
 199 struct tdb_free_record {
 200         uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
 201         uint64_t ftable_and_len; /* Len not counting these two fields. */
 202         /* This is why the minimum record size is 8 bytes.  */
 203         uint64_t next;
 204 };
 205
 206 static inline uint64_t frec_prev(const struct tdb_free_record *f)
 207 {
 208         return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
 209 }
 210
 211 static inline uint64_t frec_magic(const struct tdb_free_record *f)
 212 {
 213         return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
 214 }
 215
 216 static inline uint64_t frec_len(const struct tdb_free_record *f)
 217 {
 218         return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
 219 }
 220
 221 static inline unsigned frec_ftable(const struct tdb_free_record *f)
 222 {
 223         return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
 224 }
 225
 226 struct tdb_recovery_record {
 227         uint64_t magic;
 228         /* Length of record (add this header to get total length). */
 229         uint64_t max_len;
 230         /* Length used. */
 231         uint64_t len;
 232         /* Old length of file before transaction. */
 233         uint64_t eof;
 234 };
 235
 236 /* If we bottom out of the subhashes, we chain. */
 237 struct tdb_chain {
 238         tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
 239         tdb_off_t next;
 240 };
 241
 242 /* this is stored at the front of every database */
 243 struct tdb_header {
 244         char magic_food[64]; /* for /etc/magic */
 245         /* FIXME: Make me 32 bit? */
 246         uint64_t version; /* version of the code */
 247         uint64_t hash_test; /* result of hashing HASH_MAGIC. */
 248         uint64_t hash_seed; /* "random" seed written at creation time. */
 249         tdb_off_t free_table; /* (First) free table. */
 250         tdb_off_t recovery; /* Transaction recovery area. */
 251
 252         uint64_t features_used; /* Features all writers understand */
 253         uint64_t features_offered; /* Features offered */
 254
 255         uint64_t seqnum; /* Sequence number for TDB_SEQNUM */
 256
 257         tdb_off_t capabilities; /* Optional linked list of capabilities. */
 258         tdb_off_t reserved[22];
 259
 260         /* Top level hash table. */
 261         tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
 262 };
 263
 264 struct tdb_freetable {
 265         struct tdb_used_record hdr;
 266         tdb_off_t next;
 267         tdb_off_t buckets[TDB_FREE_BUCKETS];
 268 };
 269
 270 struct tdb_capability {
 271         struct tdb_used_record hdr;
 272         tdb_off_t type;
 273         tdb_off_t next;
 274         /* ... */
 275 };
 276
 277 /* Information about a particular (locked) hash entry. */
 278 struct hash_info {
 279         /* Full hash value of entry. */
 280         uint64_t h;
 281         /* Start and length of lock acquired. */
 282         tdb_off_t hlock_start;
 283         tdb_len_t hlock_range;
 284         /* Start of hash group. */
 285         tdb_off_t group_start;
 286         /* Bucket we belong in. */
 287         unsigned int home_bucket;
 288         /* Bucket we (or an empty space) were found in. */
 289         unsigned int found_bucket;
 290         /* How many bits of the hash are already used. */
 291         unsigned int hash_used;
 292         /* Current working group. */
 293         tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
 294 };
 295
 296 struct traverse_info {
 297         struct traverse_level {
 298                 tdb_off_t hashtable;
 299                 /* We ignore groups here, and treat it as a big array. */
 300                 unsigned entry;
 301                 unsigned int total_buckets;
 302         } levels[TDB_MAX_LEVELS + 1];
 303         unsigned int num_levels;
 304         unsigned int toplevel_group;
 305         /* This makes delete-everything-inside-traverse work as expected. */
 306         tdb_off_t prev;
 307 };
 308
 309 typedef uint32_t tdb1_len_t;
 310 typedef uint32_t tdb1_off_t;
 311
 312 enum tdb_lock_flags {
 313         /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
 314         TDB_LOCK_NOWAIT = 0,
 315         TDB_LOCK_WAIT = 1,
 316         /* If set, don't log an error on failure. */
 317         TDB_LOCK_PROBE = 2,
 318         /* If set, don't check for recovery (used by recovery code). */
 319         TDB_LOCK_NOCHECK = 4,
 320 };
 321
 322 struct tdb_lock {
 323         struct tdb_context *owner;
 324         off_t off;
 325         uint32_t count;
 326         uint32_t ltype;
 327 };
 328
 329 /* This is only needed for tdb_access_commit, but used everywhere to
 330  * simplify. */
 331 struct tdb_access_hdr {
 332         struct tdb_access_hdr *next;
 333         tdb_off_t off;
 334         tdb_len_t len;
 335         bool convert;
 336 };
 337
 338 struct tdb_file {
 339         /* How many are sharing us? */
 340         unsigned int refcnt;
 341
 342         /* Mmap (if any), or malloc (for TDB_INTERNAL). */
 343         void *map_ptr;
 344
 345         /* How much space has been mapped (<= current file size) */
 346         tdb_len_t map_size;
 347
 348         /* The file descriptor (-1 for TDB_INTERNAL). */
 349         int fd;
 350
 351         /* Lock information */
 352         pid_t locker;
 353         struct tdb_lock allrecord_lock;
 354         size_t num_lockrecs;
 355         struct tdb_lock *lockrecs;
 356
 357         /* Identity of this file. */
 358         dev_t device;
 359         ino_t inode;
 360 };
 361
 362 struct tdb_methods {
 363         enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
 364                                 tdb_len_t);
 365         enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
 366                                  tdb_len_t);
 367         enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, tdb_len_t, bool);
 368         enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
 369         void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
 370 };
 371
 372 /*
 373   internal prototypes
 374 */
 375 /* hash.c: */
 376 uint64_t tdb_jenkins_hash(const void *key, size_t length, uint64_t seed,
 377                           void *unused);
 378
 379 enum TDB_ERROR first_in_hash(struct tdb_context *tdb,
 380                              struct traverse_info *tinfo,
 381                              TDB_DATA *kbuf, size_t *dlen);
 382
 383 enum TDB_ERROR next_in_hash(struct tdb_context *tdb,
 384                             struct traverse_info *tinfo,
 385                             TDB_DATA *kbuf, size_t *dlen);
 386
 387 /* Hash random memory. */
 388 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
 389
 390 /* Hash on disk. */
 391 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
 392
 393 /* Find and lock a hash entry (or where it would be). */
 394 tdb_off_t find_and_lock(struct tdb_context *tdb,
 395                         struct tdb_data key,
 396                         int ltype,
 397                         struct hash_info *h,
 398                         struct tdb_used_record *rec,
 399                         struct traverse_info *tinfo);
 400
 401 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
 402                                struct hash_info *h,
 403                                tdb_off_t new_off);
 404
 405 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
 406                            tdb_off_t new_off);
 407
 408 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
 409
 410 /* For tdb_check */
 411 bool is_subhash(tdb_off_t val);
 412 enum TDB_ERROR unknown_capability(struct tdb_context *tdb, const char *caller,
 413                                   tdb_off_t type);
 414
 415 /* free.c: */
 416 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
 417
 418 /* check.c needs these to iterate through free lists. */
 419 tdb_off_t first_ftable(struct tdb_context *tdb);
 420 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
 421
 422 /* This returns space or -ve error number. */
 423 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
 424                 uint64_t hash, unsigned magic, bool growing);
 425
 426 /* Put this record in a free list. */
 427 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
 428                                tdb_off_t off, tdb_len_t len_with_header,
 429                                enum tdb_lock_flags waitflag,
 430                                bool coalesce_ok);
 431
 432 /* Set up header for a used/ftable/htable/chain/capability record. */
 433 enum TDB_ERROR set_header(struct tdb_context *tdb,
 434                           struct tdb_used_record *rec,
 435                           unsigned magic, uint64_t keylen, uint64_t datalen,
 436                           uint64_t actuallen, unsigned hashlow);
 437
 438 /* Used by tdb_check to verify. */
 439 unsigned int size_to_bucket(tdb_len_t data_len);
 440 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
 441
 442 /* Used by tdb_summary */
 443 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
 444
 445 /* Adjust expansion, used by create_recovery_area */
 446 tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size);
 447
 448 /* io.c: */
 449 /* Initialize tdb->methods. */
 450 void tdb_io_init(struct tdb_context *tdb);
 451
 452 /* Convert endian of the buffer if required. */
 453 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
 454
 455 /* Unmap and try to map the tdb. */
 456 void tdb_munmap(struct tdb_file *file);
 457 enum TDB_ERROR tdb_mmap(struct tdb_context *tdb);
 458
 459 /* Either alloc a copy, or give direct access.  Release frees or noop. */
 460 const void *tdb_access_read(struct tdb_context *tdb,
 461                             tdb_off_t off, tdb_len_t len, bool convert);
 462 void *tdb_access_write(struct tdb_context *tdb,
 463                        tdb_off_t off, tdb_len_t len, bool convert);
 464
 465 /* Release result of tdb_access_read/write. */
 466 void tdb_access_release(struct tdb_context *tdb, const void *p);
 467 /* Commit result of tdb_acces_write. */
 468 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
 469
 470 /* Convenience routine to get an offset. */
 471 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
 472
 473 /* Write an offset at an offset. */
 474 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
 475                              tdb_off_t val);
 476
 477 /* Clear an ondisk area. */
 478 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
 479
 480 /* Return a non-zero offset between >= start < end in this array (or end). */
 481 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
 482                                tdb_off_t base,
 483                                uint64_t start,
 484                                uint64_t end);
 485
 486 /* Return a zero offset in this array, or num. */
 487 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
 488                             uint64_t num);
 489
 490 /* Allocate and make a copy of some offset. */
 491 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
 492
 493 /* Writes a converted copy of a record. */
 494 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
 495                                  const void *rec, size_t len);
 496
 497 /* Reads record and converts it */
 498 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
 499                                 void *rec, size_t len);
 500
 501 /* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */
 502 void tdb_inc_seqnum(struct tdb_context *tdb);
 503
 504 /* lock.c: */
 505 /* Print message because another tdb owns a lock we want. */
 506 enum TDB_ERROR owner_conflict(struct tdb_context *tdb, const char *call);
 507
 508 /* If we fork, we no longer really own locks. */
 509 bool check_lock_pid(struct tdb_context *tdb, const char *call, bool log);
 510
 511 /* Lock/unlock a range of hashes. */
 512 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
 513                                tdb_off_t hash_lock, tdb_len_t hash_range,
 514                                int ltype, enum tdb_lock_flags waitflag);
 515 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
 516                                  tdb_off_t hash_lock,
 517                                  tdb_len_t hash_range, int ltype);
 518
 519 /* For closing the file. */
 520 void tdb_lock_cleanup(struct tdb_context *tdb);
 521
 522 /* Lock/unlock a particular free bucket. */
 523 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
 524                                     enum tdb_lock_flags waitflag);
 525 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
 526
 527 /* Serialize transaction start. */
 528 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
 529 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
 530
 531 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
 532 bool tdb_has_hash_locks(struct tdb_context *tdb);
 533
 534 /* Lock entire database. */
 535 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 536                                   enum tdb_lock_flags flags, bool upgradable);
 537 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
 538 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb, off_t start);
 539
 540 /* Serialize db open. */
 541 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
 542                              int ltype, enum tdb_lock_flags flags);
 543 void tdb_unlock_open(struct tdb_context *tdb, int ltype);
 544 bool tdb_has_open_lock(struct tdb_context *tdb);
 545
 546 /* Serialize db expand. */
 547 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
 548 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
 549 bool tdb_has_expansion_lock(struct tdb_context *tdb);
 550
 551 /* If it needs recovery, grab all the locks and do it. */
 552 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
 553
 554 /* Byte-range lock wrappers for TDB1 to access. */
 555 enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
 556                           int rw_type, tdb_off_t offset, tdb_off_t len,
 557                           enum tdb_lock_flags flags);
 558
 559 enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
 560                             int rw_type, tdb_off_t offset, size_t len);
 561
 562 enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 563                              tdb_off_t offset, int ltype,
 564                              enum tdb_lock_flags flags);
 565
 566 enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
 567                                tdb_off_t off, int ltype);
 568
 569 enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
 570                                 int ltype, enum tdb_lock_flags flags,
 571                                 tdb_off_t off, tdb_off_t len);
 572
 573 /* Default lock and unlock functions. */
 574 int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *);
 575 int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *);
 576
 577 /* transaction.c: */
 578 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
 579 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
 580
 581 /* this is stored at the front of every database */
 582 struct tdb1_header {
 583         char magic_food[32]; /* for /etc/magic */
 584         uint32_t version; /* version of the code */
 585         uint32_t hash_size; /* number of hash entries */
 586         tdb1_off_t rwlocks; /* obsolete - kept to detect old formats */
 587         tdb1_off_t recovery_start; /* offset of transaction recovery region */
 588         tdb1_off_t sequence_number; /* used when TDB1_SEQNUM is set */
 589         uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */
 590         uint32_t magic2_hash; /* hash of TDB1_MAGIC. */
 591         tdb1_off_t reserved[27];
 592 };
 593
 594 struct tdb1_traverse_lock {
 595         struct tdb1_traverse_lock *next;
 596         uint32_t off;
 597         uint32_t hash;
 598         int lock_rw;
 599 };
 600
 601 struct tdb_context {
 602         /* Single list of all TDBs, to detect multiple opens. */
 603         struct tdb_context *next;
 604
 605         /* Filename of the database. */
 606         const char *name;
 607
 608         /* Logging function */
 609         void (*log_fn)(struct tdb_context *tdb,
 610                        enum tdb_log_level level,
 611                        enum TDB_ERROR ecode,
 612                        const char *message,
 613                        void *data);
 614         void *log_data;
 615
 616         /* Open flags passed to tdb_open. */
 617         int open_flags;
 618
 619         /* low level (fnctl) lock functions. */
 620         int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *);
 621         int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *);
 622         void *lock_data;
 623
 624         /* the tdb flags passed to tdb_open. */
 625         uint32_t flags;
 626
 627         /* Our statistics. */
 628         struct tdb_attribute_stats stats;
 629
 630         /* The actual file information */
 631         struct tdb_file *file;
 632
 633         /* Hash function. */
 634         uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *);
 635         void *hash_data;
 636         uint64_t hash_seed;
 637
 638         /* Our open hook, if any. */
 639         enum TDB_ERROR (*openhook)(int fd, void *data);
 640         void *openhook_data;
 641
 642         /* Last error we returned. */
 643         enum TDB_ERROR last_error;
 644
 645         struct {
 646
 647                 /* Are we accessing directly? (debugging check). */
 648                 int direct_access;
 649
 650                 /* Set if we are in a transaction. */
 651                 struct tdb_transaction *transaction;
 652
 653                 /* What free table are we using? */
 654                 tdb_off_t ftable_off;
 655                 unsigned int ftable;
 656
 657                 /* IO methods: changes for transactions. */
 658                 const struct tdb_methods *io;
 659
 660                 /* Direct access information */
 661                 struct tdb_access_hdr *access;
 662         } tdb2;
 663
 664         struct {
 665                 int traverse_read; /* read-only traversal */
 666                 int traverse_write; /* read-write traversal */
 667
 668                 struct tdb1_header header; /* a cached copy of the header */
 669                 struct tdb1_traverse_lock travlocks; /* current traversal locks */
 670                 const struct tdb1_methods *io;
 671                 struct tdb1_transaction *transaction;
 672                 int page_size;
 673                 int max_dead_records;
 674         } tdb1;
 675 };
 676
 677 #define TDB1_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
 678
 679 /* tdb1_check.c: */
 680 int tdb1_check(struct tdb_context *tdb,
 681                enum TDB_ERROR (*check)(TDB_DATA key, TDB_DATA data, void *),
 682                void *private_data);
 683
 684
 685 /* tdb1_open.c: */
 686 enum TDB_ERROR tdb1_new_database(struct tdb_context *tdb,
 687                                  struct tdb_attribute_tdb1_hashsize *hashsize,
 688                                  struct tdb_attribute_tdb1_max_dead *max_dead);
 689 enum TDB_ERROR tdb1_open(struct tdb_context *tdb,
 690                          struct tdb_attribute_tdb1_max_dead *max_dead);
 691
 692 /* tdb1_io.c: */
 693 enum TDB_ERROR tdb1_probe_length(struct tdb_context *tdb);
 694
 695 /* tdb1_lock.c: */
 696 int tdb1_allrecord_lock(struct tdb_context *tdb, int ltype,
 697                         enum tdb_lock_flags flags, bool upgradable);
 698 int tdb1_allrecord_unlock(struct tdb_context *tdb, int ltype);
 699
 700 int tdb1_chainlock(struct tdb_context *tdb, TDB_DATA key);
 701 int tdb1_chainunlock(struct tdb_context *tdb, TDB_DATA key);
 702 int tdb1_chainlock_read(struct tdb_context *tdb, TDB_DATA key);
 703 int tdb1_chainunlock_read(struct tdb_context *tdb, TDB_DATA key);
 704
 705 /* tdb1_transaction.c: */
 706 int tdb1_transaction_recover(struct tdb_context *tdb);
 707 int tdb1_transaction_cancel(struct tdb_context *tdb);
 708
 709 /* tdb1_traverse.c: */
 710 int tdb1_traverse(struct tdb_context *tdb,
 711                   int (*)(struct tdb_context *, TDB_DATA, TDB_DATA, void *),
 712                   void *private_data);
 713
 714 /* tdb1_summary.c: */
 715 char *tdb1_summary(struct tdb_context *tdb);
 716
 717 /* tdb1_tdb.c: */
 718 int tdb1_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
 719 enum TDB_ERROR tdb1_fetch(struct tdb_context *tdb, TDB_DATA key,
 720                           TDB_DATA *data);
 721 int tdb1_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf);
 722 int tdb1_delete(struct tdb_context *tdb, TDB_DATA key);
 723 int tdb1_exists(struct tdb_context *tdb, TDB_DATA key);
 724 enum TDB_ERROR tdb1_parse_record(struct tdb_context *tdb, TDB_DATA key,
 725                                  enum TDB_ERROR (*parser)(TDB_DATA key,
 726                                                           TDB_DATA data,
 727                                                           void *private_data),
 728                                  void *private_data);
 729 void tdb1_increment_seqnum_nonblock(struct tdb_context *tdb);
 730 int tdb1_get_seqnum(struct tdb_context *tdb);
 731 int tdb1_wipe_all(struct tdb_context *tdb);
 732
 733 /* tdb1_transaction.c: */
 734 int tdb1_transaction_start(struct tdb_context *tdb);
 735 int tdb1_transaction_prepare_commit(struct tdb_context *tdb);
 736 int tdb1_transaction_commit(struct tdb_context *tdb);
 737
 738 /* tdb1_traverse.c: */
 739 TDB_DATA tdb1_firstkey(struct tdb_context *tdb);
 740 TDB_DATA tdb1_nextkey(struct tdb_context *tdb, TDB_DATA key);
 741
 742 /* tdb.c: */
 743 enum TDB_ERROR COLD PRINTF_FMT(4, 5)
 744         tdb_logerr(struct tdb_context *tdb,
 745                    enum TDB_ERROR ecode,
 746                    enum tdb_log_level level,
 747                    const char *fmt, ...);
 748
 749 #ifdef TDB_TRACE
 750 void tdb_trace(struct tdb_context *tdb, const char *op);
 751 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
 752 void tdb_trace_open(struct tdb_context *tdb, const char *op,
 753                     unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
 754 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
 755 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
 756 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
 757                     TDB_DATA rec);
 758 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
 759                         TDB_DATA rec, int ret);
 760 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
 761                            TDB_DATA rec, TDB_DATA ret);
 762 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
 763                              TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
 764                              int ret);
 765 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
 766                            TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
 767 #else
 768 #define tdb_trace(tdb, op)
 769 #define tdb_trace_seqnum(tdb, seqnum, op)
 770 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
 771 #define tdb_trace_ret(tdb, op, ret)
 772 #define tdb_trace_retrec(tdb, op, ret)
 773 #define tdb_trace_1rec(tdb, op, rec)
 774 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
 775 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
 776 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
 777 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
 778 #endif /* !TDB_TRACE */
 779
 780 #endif