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