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/>.
35 #include <ccan/tdb2/tdb2.h>
36 #include <ccan/likely/likely.h>
37 #include <ccan/compiler/compiler.h>
46 /* #define TDB_TRACE 1 */
49 #define __STRING(x) #x
52 #ifndef __STRINGSTRING
53 #define __STRINGSTRING(x) __STRING(x)
57 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
60 typedef uint64_t tdb_len_t;
61 typedef uint64_t tdb_off_t;
63 #define TDB_MAGIC_FOOD "TDB file\n"
64 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
65 #define TDB_USED_MAGIC ((uint64_t)0x1999)
66 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
67 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
68 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
69 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
70 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
71 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
72 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
74 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
76 /* Packing errors into pointers and v.v. */
77 #define TDB_PTR_IS_ERR(ptr) \
78 unlikely((void *)(ptr) >= (void *)(long)TDB_ERR_LAST)
79 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
80 #define TDB_ERR_PTR(err) ((void *)(long)(err))
82 /* Common case of returning true, false or -ve error. */
83 typedef int tdb_bool_err;
85 /* Prevent others from opening the file. */
86 #define TDB_OPEN_LOCK 0
87 /* Doing a transaction. */
88 #define TDB_TRANSACTION_LOCK 1
90 #define TDB_EXPANSION_LOCK 2
91 /* Hash chain locks. */
92 #define TDB_HASH_LOCK_START 3
94 /* Range for hash locks. */
95 #define TDB_HASH_LOCK_RANGE_BITS 30
96 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
98 /* We have 1024 entries in the top level. */
99 #define TDB_TOPLEVEL_HASH_BITS 10
100 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
101 #define TDB_SUBLEVEL_HASH_BITS 6
102 /* And 8 entries in each group, ie 8 groups per sublevel. */
103 #define TDB_HASH_GROUP_BITS 3
104 /* This is currently 10: beyond this we chain. */
105 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
107 /* Extend file by least 100 times larger than needed. */
108 #define TDB_EXTENSION_FACTOR 100
110 /* We steal bits from the offsets to store hash info. */
111 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
112 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
113 #define TDB_OFF_UPPER_STEAL 8
114 #define TDB_OFF_UPPER_STEAL_EXTRA 7
115 /* The bit number where we store extra hash bits. */
116 #define TDB_OFF_HASH_EXTRA_BIT 57
117 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
119 /* The bit number where we store the extra hash bits. */
120 /* Convenience mask to get actual offset. */
121 #define TDB_OFF_MASK \
122 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
124 /* How many buckets in a free list: see size_to_bucket(). */
125 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
127 /* We have to be able to fit a free record here. */
128 #define TDB_MIN_DATA_LEN \
129 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
131 /* Indicates this entry is not on an flist (can happen during coalescing) */
132 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
135 static inline uint64_t bswap_64(uint64_t x)
137 return (((x&0x000000FFULL)<<56)
138 | ((x&0x0000FF00ULL)<<48)
139 | ((x&0x00FF0000ULL)<<40)
140 | ((x&0xFF000000ULL)<<32)
141 | ((x>>8)&0xFF000000ULL)
142 | ((x>>16)&0x00FF0000ULL)
143 | ((x>>24)&0x0000FF00ULL)
144 | ((x>>32)&0x000000FFULL));
148 struct tdb_used_record {
149 /* For on-disk compatibility, we avoid bitfields:
155 uint64_t magic_and_meta;
156 /* The bottom key_len_bits*2 are key length, rest is data length. */
157 uint64_t key_and_data_len;
160 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
162 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
165 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
167 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
170 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
172 return r->key_and_data_len >> rec_key_bits(r);
175 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
177 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
180 static inline uint32_t rec_hash(const struct tdb_used_record *r)
182 return r->magic_and_meta & ((1 << 11) - 1);
185 static inline uint16_t rec_magic(const struct tdb_used_record *r)
187 return (r->magic_and_meta >> 48);
190 struct tdb_free_record {
191 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
192 uint64_t ftable_and_len; /* Len not counting these two fields. */
193 /* This is why the minimum record size is 8 bytes. */
197 static inline uint64_t frec_prev(const struct tdb_free_record *f)
199 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
202 static inline uint64_t frec_magic(const struct tdb_free_record *f)
204 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
207 static inline uint64_t frec_len(const struct tdb_free_record *f)
209 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
212 static inline unsigned frec_ftable(const struct tdb_free_record *f)
214 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
217 struct tdb_recovery_record {
219 /* Length of record. */
223 /* Old length of file before transaction. */
227 /* If we bottom out of the subhashes, we chain. */
229 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
233 /* this is stored at the front of every database */
235 char magic_food[64]; /* for /etc/magic */
236 /* FIXME: Make me 32 bit? */
237 uint64_t version; /* version of the code */
238 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
239 uint64_t hash_seed; /* "random" seed written at creation time. */
240 tdb_off_t free_table; /* (First) free table. */
241 tdb_off_t recovery; /* Transaction recovery area. */
243 tdb_off_t reserved[26];
245 /* Top level hash table. */
246 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
249 struct tdb_freetable {
250 struct tdb_used_record hdr;
252 tdb_off_t buckets[TDB_FREE_BUCKETS];
255 /* Information about a particular (locked) hash entry. */
257 /* Full hash value of entry. */
259 /* Start and length of lock acquired. */
260 tdb_off_t hlock_start;
261 tdb_len_t hlock_range;
262 /* Start of hash group. */
263 tdb_off_t group_start;
264 /* Bucket we belong in. */
265 unsigned int home_bucket;
266 /* Bucket we (or an empty space) were found in. */
267 unsigned int found_bucket;
268 /* How many bits of the hash are already used. */
269 unsigned int hash_used;
270 /* Current working group. */
271 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
274 struct traverse_info {
275 struct traverse_level {
277 /* We ignore groups here, and treat it as a big array. */
279 unsigned int total_buckets;
280 } levels[TDB_MAX_LEVELS + 1];
281 unsigned int num_levels;
282 unsigned int toplevel_group;
283 /* This makes delete-everything-inside-traverse work as expected. */
287 enum tdb_lock_flags {
288 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
291 /* If set, don't log an error on failure. */
293 /* If set, don't check for recovery (used by recovery code). */
294 TDB_LOCK_NOCHECK = 4,
297 struct tdb_lock_type {
303 /* This is only needed for tdb_access_commit, but used everywhere to
305 struct tdb_access_hdr {
306 struct tdb_access_hdr *next;
313 /* Filename of the database. */
316 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
319 /* Are we accessing directly? (debugging check). */
322 /* Open file descriptor (undefined for TDB_INTERNAL). */
325 /* How much space has been mapped (<= current file size) */
328 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
331 /* mmap read only? */
334 /* Error code for last tdb error. */
335 enum TDB_ERROR ecode;
337 /* the flags passed to tdb_open, for tdb_reopen. */
340 /* Logging function */
341 void (*logfn)(struct tdb_context *tdb,
342 enum tdb_log_level level,
344 const char *message);
348 uint64_t (*khash)(const void *key, size_t len, uint64_t seed, void *);
352 /* Set if we are in a transaction. */
353 struct tdb_transaction *transaction;
355 /* What free table are we using? */
356 tdb_off_t ftable_off;
359 /* IO methods: changes for transactions. */
360 const struct tdb_methods *methods;
362 /* Lock information */
363 struct tdb_lock_type allrecord_lock;
365 struct tdb_lock_type *lockrecs;
367 struct tdb_attribute_stats *stats;
369 /* Direct access information */
370 struct tdb_access_hdr *access;
372 /* Single list of all TDBs, to avoid multiple opens. */
373 struct tdb_context *next;
379 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
381 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
383 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
384 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
385 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
392 void tdb_hash_init(struct tdb_context *tdb);
394 tdb_bool_err first_in_hash(struct tdb_context *tdb,
395 struct traverse_info *tinfo,
396 TDB_DATA *kbuf, size_t *dlen);
398 tdb_bool_err next_in_hash(struct tdb_context *tdb,
399 struct traverse_info *tinfo,
400 TDB_DATA *kbuf, size_t *dlen);
402 /* Hash random memory. */
403 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
406 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
408 /* Find and lock a hash entry (or where it would be). */
409 tdb_off_t find_and_lock(struct tdb_context *tdb,
413 struct tdb_used_record *rec,
414 struct traverse_info *tinfo);
416 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
420 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
423 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
426 bool is_subhash(tdb_off_t val);
429 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
431 /* check.c needs these to iterate through free lists. */
432 tdb_off_t first_ftable(struct tdb_context *tdb);
433 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
435 /* This returns space or -ve error number. */
436 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
437 uint64_t hash, unsigned magic, bool growing);
439 /* Put this record in a free list. */
440 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
441 tdb_off_t off, tdb_len_t len_with_header);
443 /* Set up header for a used/ftable/htable/chain record. */
444 enum TDB_ERROR set_header(struct tdb_context *tdb,
445 struct tdb_used_record *rec,
446 unsigned magic, uint64_t keylen, uint64_t datalen,
447 uint64_t actuallen, unsigned hashlow);
449 /* Used by tdb_check to verify. */
450 unsigned int size_to_bucket(tdb_len_t data_len);
451 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
453 /* Used by tdb_summary */
454 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
457 /* Initialize tdb->methods. */
458 void tdb_io_init(struct tdb_context *tdb);
460 /* Convert endian of the buffer if required. */
461 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
463 /* Unmap and try to map the tdb. */
464 void tdb_munmap(struct tdb_context *tdb);
465 void tdb_mmap(struct tdb_context *tdb);
467 /* Either alloc a copy, or give direct access. Release frees or noop. */
468 const void *tdb_access_read(struct tdb_context *tdb,
469 tdb_off_t off, tdb_len_t len, bool convert);
470 void *tdb_access_write(struct tdb_context *tdb,
471 tdb_off_t off, tdb_len_t len, bool convert);
473 /* Release result of tdb_access_read/write. */
474 void tdb_access_release(struct tdb_context *tdb, const void *p);
475 /* Commit result of tdb_acces_write. */
476 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
478 /* Convenience routine to get an offset. */
479 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
481 /* Write an offset at an offset. */
482 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
485 /* Clear an ondisk area. */
486 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
488 /* Return a non-zero offset between >= start < end in this array (or end). */
489 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
494 /* Return a zero offset in this array, or num. */
495 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
498 /* Allocate and make a copy of some offset. */
499 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
501 /* Writes a converted copy of a record. */
502 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
503 const void *rec, size_t len);
505 /* Reads record and converts it */
506 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
507 void *rec, size_t len);
509 /* Adds a stat, if it's in range. */
510 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
511 #define add_stat(tdb, statname, val) \
513 if (unlikely((tdb)->stats)) \
514 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
518 void tdb_lock_init(struct tdb_context *tdb);
520 /* Lock/unlock a range of hashes. */
521 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
522 tdb_off_t hash_lock, tdb_len_t hash_range,
523 int ltype, enum tdb_lock_flags waitflag);
524 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
526 tdb_len_t hash_range, int ltype);
528 /* Lock/unlock a particular free bucket. */
529 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
530 enum tdb_lock_flags waitflag);
531 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
533 /* Serialize transaction start. */
534 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
535 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
537 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
538 bool tdb_has_hash_locks(struct tdb_context *tdb);
540 /* Lock entire database. */
541 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
542 enum tdb_lock_flags flags, bool upgradable);
543 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
544 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
546 /* Serialize db open. */
547 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
548 enum tdb_lock_flags flags);
549 void tdb_unlock_open(struct tdb_context *tdb);
550 bool tdb_has_open_lock(struct tdb_context *tdb);
552 /* Serialize db expand. */
553 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
554 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
555 bool tdb_has_expansion_lock(struct tdb_context *tdb);
557 /* If it needs recovery, grab all the locks and do it. */
558 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
561 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
562 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
565 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
566 enum TDB_ERROR ecode,
567 enum tdb_log_level level,
568 const char *fmt, ...);
571 void tdb_trace(struct tdb_context *tdb, const char *op);
572 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
573 void tdb_trace_open(struct tdb_context *tdb, const char *op,
574 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
575 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
576 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
577 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
579 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
580 TDB_DATA rec, int ret);
581 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
582 TDB_DATA rec, TDB_DATA ret);
583 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
584 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
586 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
587 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
589 #define tdb_trace(tdb, op)
590 #define tdb_trace_seqnum(tdb, seqnum, op)
591 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
592 #define tdb_trace_ret(tdb, op, ret)
593 #define tdb_trace_retrec(tdb, op, ret)
594 #define tdb_trace_1rec(tdb, op, rec)
595 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
596 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
597 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
598 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
599 #endif /* !TDB_TRACE */