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