4 Trivial Database 2: private types and prototypes
5 Copyright (C) Rusty Russell 2010
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.
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.
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/>.
21 #include <ccan/tdb2/tdb2.h>
32 #include <ccan/likely/likely.h>
33 #include <ccan/endian/endian.h>
39 /* #define TDB_TRACE 1 */
42 #define __STRING(x) #x
45 #ifndef __STRINGSTRING
46 #define __STRINGSTRING(x) __STRING(x)
50 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
53 typedef uint64_t tdb_len_t;
54 typedef uint64_t tdb_off_t;
56 #define TDB_MAGIC_FOOD "TDB file\n"
57 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
58 #define TDB_USED_MAGIC ((uint64_t)0x1999)
59 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
60 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
61 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
62 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
63 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
64 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
65 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
67 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
69 /* Packing errors into pointers and v.v. */
70 #define TDB_PTR_IS_ERR(ptr) \
71 unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST)
72 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
73 #define TDB_ERR_PTR(err) ((void *)(long)(err))
75 /* Common case of returning true, false or -ve error. */
76 typedef int tdb_bool_err;
78 /* Prevent others from opening the file. */
79 #define TDB_OPEN_LOCK 0
81 #define TDB_EXPANSION_LOCK 2
82 /* Doing a transaction. */
83 #define TDB_TRANSACTION_LOCK 8
84 /* Hash chain locks. */
85 #define TDB_HASH_LOCK_START 64
87 /* Range for hash locks. */
88 #define TDB_HASH_LOCK_RANGE_BITS 30
89 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
91 /* We have 1024 entries in the top level. */
92 #define TDB_TOPLEVEL_HASH_BITS 10
93 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
94 #define TDB_SUBLEVEL_HASH_BITS 6
95 /* And 8 entries in each group, ie 8 groups per sublevel. */
96 #define TDB_HASH_GROUP_BITS 3
97 /* This is currently 10: beyond this we chain. */
98 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
100 /* Extend file by least 100 times larger than needed. */
101 #define TDB_EXTENSION_FACTOR 100
103 /* We steal bits from the offsets to store hash info. */
104 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
105 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
106 #define TDB_OFF_UPPER_STEAL 8
107 #define TDB_OFF_UPPER_STEAL_EXTRA 7
108 /* The bit number where we store extra hash bits. */
109 #define TDB_OFF_HASH_EXTRA_BIT 57
110 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
112 /* Additional features we understand. Currently: none. */
113 #define TDB_FEATURE_MASK ((uint64_t)0)
115 /* The bit number where we store the extra hash bits. */
116 /* Convenience mask to get actual offset. */
117 #define TDB_OFF_MASK \
118 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
120 /* How many buckets in a free list: see size_to_bucket(). */
121 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
123 /* We have to be able to fit a free record here. */
124 #define TDB_MIN_DATA_LEN \
125 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
127 /* Indicates this entry is not on an flist (can happen during coalescing) */
128 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
130 struct tdb_used_record {
131 /* For on-disk compatibility, we avoid bitfields:
137 uint64_t magic_and_meta;
138 /* The bottom key_len_bits*2 are key length, rest is data length. */
139 uint64_t key_and_data_len;
142 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
144 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
147 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
149 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
152 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
154 return r->key_and_data_len >> rec_key_bits(r);
157 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
159 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
162 static inline uint32_t rec_hash(const struct tdb_used_record *r)
164 return r->magic_and_meta & ((1 << 11) - 1);
167 static inline uint16_t rec_magic(const struct tdb_used_record *r)
169 return (r->magic_and_meta >> 48);
172 struct tdb_free_record {
173 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
174 uint64_t ftable_and_len; /* Len not counting these two fields. */
175 /* This is why the minimum record size is 8 bytes. */
179 static inline uint64_t frec_prev(const struct tdb_free_record *f)
181 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
184 static inline uint64_t frec_magic(const struct tdb_free_record *f)
186 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
189 static inline uint64_t frec_len(const struct tdb_free_record *f)
191 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
194 static inline unsigned frec_ftable(const struct tdb_free_record *f)
196 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
199 struct tdb_recovery_record {
201 /* Length of record (add this header to get total length). */
205 /* Old length of file before transaction. */
209 /* If we bottom out of the subhashes, we chain. */
211 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
215 /* this is stored at the front of every database */
217 char magic_food[64]; /* for /etc/magic */
218 /* FIXME: Make me 32 bit? */
219 uint64_t version; /* version of the code */
220 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
221 uint64_t hash_seed; /* "random" seed written at creation time. */
222 tdb_off_t free_table; /* (First) free table. */
223 tdb_off_t recovery; /* Transaction recovery area. */
225 uint64_t features_used; /* Features all writers understand */
226 uint64_t features_offered; /* Features offered */
228 uint64_t seqnum; /* Sequence number for TDB_SEQNUM */
230 tdb_off_t reserved[23];
232 /* Top level hash table. */
233 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
236 struct tdb_freetable {
237 struct tdb_used_record hdr;
239 tdb_off_t buckets[TDB_FREE_BUCKETS];
242 /* Information about a particular (locked) hash entry. */
244 /* Full hash value of entry. */
246 /* Start and length of lock acquired. */
247 tdb_off_t hlock_start;
248 tdb_len_t hlock_range;
249 /* Start of hash group. */
250 tdb_off_t group_start;
251 /* Bucket we belong in. */
252 unsigned int home_bucket;
253 /* Bucket we (or an empty space) were found in. */
254 unsigned int found_bucket;
255 /* How many bits of the hash are already used. */
256 unsigned int hash_used;
257 /* Current working group. */
258 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
261 struct traverse_info {
262 struct traverse_level {
264 /* We ignore groups here, and treat it as a big array. */
266 unsigned int total_buckets;
267 } levels[TDB_MAX_LEVELS + 1];
268 unsigned int num_levels;
269 unsigned int toplevel_group;
270 /* This makes delete-everything-inside-traverse work as expected. */
274 typedef uint32_t tdb1_len_t;
275 typedef uint32_t tdb1_off_t;
277 enum tdb_lock_flags {
278 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
281 /* If set, don't log an error on failure. */
283 /* If set, don't check for recovery (used by recovery code). */
284 TDB_LOCK_NOCHECK = 4,
288 struct tdb_context *owner;
294 /* This is only needed for tdb_access_commit, but used everywhere to
296 struct tdb_access_hdr {
297 struct tdb_access_hdr *next;
304 /* How many are sharing us? */
307 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
310 /* How much space has been mapped (<= current file size) */
313 /* The file descriptor (-1 for TDB_INTERNAL). */
316 /* Lock information */
318 struct tdb_lock allrecord_lock;
320 struct tdb_lock *lockrecs;
322 /* Identity of this file. */
328 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
330 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
332 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
333 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
334 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
341 tdb_bool_err first_in_hash(struct tdb_context *tdb,
342 struct traverse_info *tinfo,
343 TDB_DATA *kbuf, size_t *dlen);
345 tdb_bool_err next_in_hash(struct tdb_context *tdb,
346 struct traverse_info *tinfo,
347 TDB_DATA *kbuf, size_t *dlen);
349 /* Hash random memory. */
350 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
353 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
355 /* Find and lock a hash entry (or where it would be). */
356 tdb_off_t find_and_lock(struct tdb_context *tdb,
360 struct tdb_used_record *rec,
361 struct traverse_info *tinfo);
363 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
367 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
370 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
373 bool is_subhash(tdb_off_t val);
376 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
378 /* check.c needs these to iterate through free lists. */
379 tdb_off_t first_ftable(struct tdb_context *tdb);
380 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
382 /* This returns space or -ve error number. */
383 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
384 uint64_t hash, unsigned magic, bool growing);
386 /* Put this record in a free list. */
387 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
388 tdb_off_t off, tdb_len_t len_with_header,
389 enum tdb_lock_flags waitflag,
392 /* Set up header for a used/ftable/htable/chain record. */
393 enum TDB_ERROR set_header(struct tdb_context *tdb,
394 struct tdb_used_record *rec,
395 unsigned magic, uint64_t keylen, uint64_t datalen,
396 uint64_t actuallen, unsigned hashlow);
398 /* Used by tdb_check to verify. */
399 unsigned int size_to_bucket(tdb_len_t data_len);
400 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
402 /* Used by tdb_summary */
403 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
406 /* Initialize tdb->methods. */
407 void tdb_io_init(struct tdb_context *tdb);
409 /* Convert endian of the buffer if required. */
410 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
412 /* Unmap and try to map the tdb. */
413 void tdb_munmap(struct tdb_file *file);
414 void tdb_mmap(struct tdb_context *tdb);
416 /* Either alloc a copy, or give direct access. Release frees or noop. */
417 const void *tdb_access_read(struct tdb_context *tdb,
418 tdb_off_t off, tdb_len_t len, bool convert);
419 void *tdb_access_write(struct tdb_context *tdb,
420 tdb_off_t off, tdb_len_t len, bool convert);
422 /* Release result of tdb_access_read/write. */
423 void tdb_access_release(struct tdb_context *tdb, const void *p);
424 /* Commit result of tdb_acces_write. */
425 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
427 /* Convenience routine to get an offset. */
428 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
430 /* Write an offset at an offset. */
431 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
434 /* Clear an ondisk area. */
435 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
437 /* Return a non-zero offset between >= start < end in this array (or end). */
438 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
443 /* Return a zero offset in this array, or num. */
444 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
447 /* Allocate and make a copy of some offset. */
448 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
450 /* Writes a converted copy of a record. */
451 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
452 const void *rec, size_t len);
454 /* Reads record and converts it */
455 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
456 void *rec, size_t len);
458 /* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */
459 void tdb_inc_seqnum(struct tdb_context *tdb);
462 /* Lock/unlock a range of hashes. */
463 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
464 tdb_off_t hash_lock, tdb_len_t hash_range,
465 int ltype, enum tdb_lock_flags waitflag);
466 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
468 tdb_len_t hash_range, int ltype);
470 /* For closing the file. */
471 void tdb_lock_cleanup(struct tdb_context *tdb);
473 /* Lock/unlock a particular free bucket. */
474 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
475 enum tdb_lock_flags waitflag);
476 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
478 /* Serialize transaction start. */
479 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
480 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
482 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
483 bool tdb_has_hash_locks(struct tdb_context *tdb);
485 /* Lock entire database. */
486 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
487 enum tdb_lock_flags flags, bool upgradable);
488 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
489 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb, off_t start);
491 /* Serialize db open. */
492 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
493 int ltype, enum tdb_lock_flags flags);
494 void tdb_unlock_open(struct tdb_context *tdb, int ltype);
495 bool tdb_has_open_lock(struct tdb_context *tdb);
497 /* Serialize db expand. */
498 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
499 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
500 bool tdb_has_expansion_lock(struct tdb_context *tdb);
502 /* If it needs recovery, grab all the locks and do it. */
503 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
505 /* Byte-range lock wrappers for TDB1 to access. */
506 enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
507 int rw_type, tdb_off_t offset, tdb_off_t len,
508 enum tdb_lock_flags flags);
510 enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
511 int rw_type, tdb_off_t offset, size_t len);
513 enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
514 tdb_off_t offset, int ltype,
515 enum tdb_lock_flags flags);
517 enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
518 tdb_off_t off, int ltype);
520 enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
521 int ltype, enum tdb_lock_flags flags,
522 tdb_off_t off, tdb_off_t len);
524 /* Default lock and unlock functions. */
525 int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *);
526 int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *);
529 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
530 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
532 /* this is stored at the front of every database */
534 char magic_food[32]; /* for /etc/magic */
535 uint32_t version; /* version of the code */
536 uint32_t hash_size; /* number of hash entries */
537 tdb1_off_t rwlocks; /* obsolete - kept to detect old formats */
538 tdb1_off_t recovery_start; /* offset of transaction recovery region */
539 tdb1_off_t sequence_number; /* used when TDB1_SEQNUM is set */
540 uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */
541 uint32_t magic2_hash; /* hash of TDB1_MAGIC. */
542 tdb1_off_t reserved[27];
545 struct tdb1_traverse_lock {
546 struct tdb1_traverse_lock *next;
553 /* Single list of all TDBs, to detect multiple opens. */
554 struct tdb_context *next;
556 /* Filename of the database. */
559 /* Logging function */
560 void (*log_fn)(struct tdb_context *tdb,
561 enum tdb_log_level level,
562 enum TDB_ERROR ecode,
567 /* Open flags passed to tdb_open. */
570 /* low level (fnctl) lock functions. */
571 int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *);
572 int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *);
575 /* the tdb flags passed to tdb_open. */
578 /* Our statistics. */
579 struct tdb_attribute_stats stats;
581 /* The actual file information */
582 struct tdb_file *file;
585 uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *);
589 /* Our open hook, if any. */
590 enum TDB_ERROR (*openhook)(int fd, void *data);
593 /* Last error we returned. */
594 enum TDB_ERROR last_error;
598 /* Are we accessing directly? (debugging check). */
601 /* Set if we are in a transaction. */
602 struct tdb_transaction *transaction;
604 /* What free table are we using? */
605 tdb_off_t ftable_off;
608 /* IO methods: changes for transactions. */
609 const struct tdb_methods *io;
611 /* Direct access information */
612 struct tdb_access_hdr *access;
616 int traverse_read; /* read-only traversal */
617 int traverse_write; /* read-write traversal */
619 struct tdb1_header header; /* a cached copy of the header */
620 struct tdb1_traverse_lock travlocks; /* current traversal locks */
621 const struct tdb1_methods *io;
622 struct tdb1_transaction *transaction;
624 int max_dead_records;
629 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
630 enum TDB_ERROR ecode,
631 enum tdb_log_level level,
632 const char *fmt, ...);
635 void tdb_trace(struct tdb_context *tdb, const char *op);
636 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
637 void tdb_trace_open(struct tdb_context *tdb, const char *op,
638 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
639 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
640 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
641 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
643 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
644 TDB_DATA rec, int ret);
645 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
646 TDB_DATA rec, TDB_DATA ret);
647 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
648 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
650 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
651 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
653 #define tdb_trace(tdb, op)
654 #define tdb_trace_seqnum(tdb, seqnum, op)
655 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
656 #define tdb_trace_ret(tdb, op, ret)
657 #define tdb_trace_retrec(tdb, op, ret)
658 #define tdb_trace_1rec(tdb, op, rec)
659 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
660 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
661 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
662 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
663 #endif /* !TDB_TRACE */