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 #define _XOPEN_SOURCE 500
22 #define _FILE_OFFSET_BITS 64
37 #include <ccan/tdb2/tdb2.h>
38 #include <ccan/likely/likely.h>
39 #include <ccan/compiler/compiler.h>
48 /* #define TDB_TRACE 1 */
51 #define __STRING(x) #x
54 #ifndef __STRINGSTRING
55 #define __STRINGSTRING(x) __STRING(x)
59 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
62 typedef uint64_t tdb_len_t;
63 typedef uint64_t tdb_off_t;
65 #define TDB_MAGIC_FOOD "TDB file\n"
66 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
67 #define TDB_USED_MAGIC ((uint64_t)0x1999)
68 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
69 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
70 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
71 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
72 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
73 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
74 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
76 #define TDB_OFF_ERR ((tdb_off_t)-1)
77 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
79 /* Packing errors into pointers and v.v. */
80 #define TDB_PTR_IS_ERR(ptr) \
81 unlikely((void *)(ptr) >= (void *)(long)TDB_ERR_LAST)
82 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
83 #define TDB_ERR_PTR(err) ((void *)(long)(err))
85 /* Common case of returning true, false or -ve error. */
86 typedef int tdb_bool_err;
88 /* Prevent others from opening the file. */
89 #define TDB_OPEN_LOCK 0
90 /* Doing a transaction. */
91 #define TDB_TRANSACTION_LOCK 1
93 #define TDB_EXPANSION_LOCK 2
94 /* Hash chain locks. */
95 #define TDB_HASH_LOCK_START 3
97 /* Range for hash locks. */
98 #define TDB_HASH_LOCK_RANGE_BITS 30
99 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
101 /* We have 1024 entries in the top level. */
102 #define TDB_TOPLEVEL_HASH_BITS 10
103 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
104 #define TDB_SUBLEVEL_HASH_BITS 6
105 /* And 8 entries in each group, ie 8 groups per sublevel. */
106 #define TDB_HASH_GROUP_BITS 3
107 /* This is currently 10: beyond this we chain. */
108 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
110 /* Extend file by least 100 times larger than needed. */
111 #define TDB_EXTENSION_FACTOR 100
113 /* We steal bits from the offsets to store hash info. */
114 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
115 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
116 #define TDB_OFF_UPPER_STEAL 8
117 #define TDB_OFF_UPPER_STEAL_EXTRA 7
118 /* The bit number where we store extra hash bits. */
119 #define TDB_OFF_HASH_EXTRA_BIT 57
120 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
122 /* The bit number where we store the extra hash bits. */
123 /* Convenience mask to get actual offset. */
124 #define TDB_OFF_MASK \
125 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
127 /* How many buckets in a free list: see size_to_bucket(). */
128 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
130 /* We have to be able to fit a free record here. */
131 #define TDB_MIN_DATA_LEN \
132 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
134 /* Indicates this entry is not on an flist (can happen during coalescing) */
135 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
138 static inline uint64_t bswap_64(uint64_t x)
140 return (((x&0x000000FFULL)<<56)
141 | ((x&0x0000FF00ULL)<<48)
142 | ((x&0x00FF0000ULL)<<40)
143 | ((x&0xFF000000ULL)<<32)
144 | ((x>>8)&0xFF000000ULL)
145 | ((x>>16)&0x00FF0000ULL)
146 | ((x>>24)&0x0000FF00ULL)
147 | ((x>>32)&0x000000FFULL));
151 struct tdb_used_record {
152 /* For on-disk compatibility, we avoid bitfields:
158 uint64_t magic_and_meta;
159 /* The bottom key_len_bits*2 are key length, rest is data length. */
160 uint64_t key_and_data_len;
163 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
165 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
168 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
170 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
173 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
175 return r->key_and_data_len >> rec_key_bits(r);
178 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
180 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
183 static inline uint32_t rec_hash(const struct tdb_used_record *r)
185 return r->magic_and_meta & ((1 << 11) - 1);
188 static inline uint16_t rec_magic(const struct tdb_used_record *r)
190 return (r->magic_and_meta >> 48);
193 struct tdb_free_record {
194 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
195 uint64_t ftable_and_len; /* Len not counting these two fields. */
196 /* This is why the minimum record size is 8 bytes. */
200 static inline uint64_t frec_prev(const struct tdb_free_record *f)
202 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
205 static inline uint64_t frec_magic(const struct tdb_free_record *f)
207 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
210 static inline uint64_t frec_len(const struct tdb_free_record *f)
212 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
215 static inline unsigned frec_ftable(const struct tdb_free_record *f)
217 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
220 struct tdb_recovery_record {
222 /* Length of record. */
226 /* Old length of file before transaction. */
230 /* If we bottom out of the subhashes, we chain. */
232 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
236 /* this is stored at the front of every database */
238 char magic_food[64]; /* for /etc/magic */
239 /* FIXME: Make me 32 bit? */
240 uint64_t version; /* version of the code */
241 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
242 uint64_t hash_seed; /* "random" seed written at creation time. */
243 tdb_off_t free_table; /* (First) free table. */
244 tdb_off_t recovery; /* Transaction recovery area. */
246 tdb_off_t reserved[26];
248 /* Top level hash table. */
249 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
252 struct tdb_freetable {
253 struct tdb_used_record hdr;
255 tdb_off_t buckets[TDB_FREE_BUCKETS];
258 /* Information about a particular (locked) hash entry. */
260 /* Full hash value of entry. */
262 /* Start and length of lock acquired. */
263 tdb_off_t hlock_start;
264 tdb_len_t hlock_range;
265 /* Start of hash group. */
266 tdb_off_t group_start;
267 /* Bucket we belong in. */
268 unsigned int home_bucket;
269 /* Bucket we (or an empty space) were found in. */
270 unsigned int found_bucket;
271 /* How many bits of the hash are already used. */
272 unsigned int hash_used;
273 /* Current working group. */
274 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
277 struct traverse_info {
278 struct traverse_level {
280 /* We ignore groups here, and treat it as a big array. */
282 unsigned int total_buckets;
283 } levels[TDB_MAX_LEVELS + 1];
284 unsigned int num_levels;
285 unsigned int toplevel_group;
286 /* This makes delete-everything-inside-traverse work as expected. */
290 enum tdb_lock_flags {
291 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
294 /* If set, don't log an error on failure. */
296 /* If set, don't check for recovery (used by recovery code). */
297 TDB_LOCK_NOCHECK = 4,
300 struct tdb_lock_type {
306 /* This is only needed for tdb_access_commit, but used everywhere to
308 struct tdb_access_hdr {
309 struct tdb_access_hdr *next;
316 /* Filename of the database. */
319 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
322 /* Are we accessing directly? (debugging check). */
325 /* Open file descriptor (undefined for TDB_INTERNAL). */
328 /* How much space has been mapped (<= current file size) */
331 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
334 /* mmap read only? */
337 /* Error code for last tdb error. */
338 enum TDB_ERROR ecode;
340 /* the flags passed to tdb_open, for tdb_reopen. */
343 /* Logging function */
344 void (*logfn)(struct tdb_context *tdb,
345 enum tdb_log_level level,
347 const char *message);
351 uint64_t (*khash)(const void *key, size_t len, uint64_t seed, void *);
355 /* Set if we are in a transaction. */
356 struct tdb_transaction *transaction;
358 /* What free table are we using? */
359 tdb_off_t ftable_off;
362 /* IO methods: changes for transactions. */
363 const struct tdb_methods *methods;
365 /* Lock information */
366 struct tdb_lock_type allrecord_lock;
368 struct tdb_lock_type *lockrecs;
370 struct tdb_attribute_stats *stats;
372 /* Direct access information */
373 struct tdb_access_hdr *access;
375 /* Single list of all TDBs, to avoid multiple opens. */
376 struct tdb_context *next;
382 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
384 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
386 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
387 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
388 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
395 void tdb_hash_init(struct tdb_context *tdb);
397 int first_in_hash(struct tdb_context *tdb,
398 struct traverse_info *tinfo,
399 TDB_DATA *kbuf, size_t *dlen);
401 int next_in_hash(struct tdb_context *tdb,
402 struct traverse_info *tinfo,
403 TDB_DATA *kbuf, size_t *dlen);
405 /* Hash random memory. */
406 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
409 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
411 /* Find and lock a hash entry (or where it would be). */
412 tdb_off_t find_and_lock(struct tdb_context *tdb,
416 struct tdb_used_record *rec,
417 struct traverse_info *tinfo);
419 int replace_in_hash(struct tdb_context *tdb,
423 int add_to_hash(struct tdb_context *tdb, struct hash_info *h,
426 int delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
429 bool is_subhash(tdb_off_t val);
432 int tdb_ftable_init(struct tdb_context *tdb);
434 /* check.c needs these to iterate through free lists. */
435 tdb_off_t first_ftable(struct tdb_context *tdb);
436 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
438 /* This returns space or TDB_OFF_ERR. */
439 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
440 uint64_t hash, unsigned magic, bool growing);
442 /* Put this record in a free list. */
443 int add_free_record(struct tdb_context *tdb,
444 tdb_off_t off, tdb_len_t len_with_header);
446 /* Set up header for a used/ftable/htable/chain record. */
447 int set_header(struct tdb_context *tdb,
448 struct tdb_used_record *rec,
449 unsigned magic, uint64_t keylen, uint64_t datalen,
450 uint64_t actuallen, unsigned hashlow);
452 /* Used by tdb_check to verify. */
453 unsigned int size_to_bucket(tdb_len_t data_len);
454 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
456 /* Used by tdb_summary */
457 size_t dead_space(struct tdb_context *tdb, tdb_off_t off);
460 /* Initialize tdb->methods. */
461 void tdb_io_init(struct tdb_context *tdb);
463 /* Convert endian of the buffer if required. */
464 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
466 /* Unmap and try to map the tdb. */
467 void tdb_munmap(struct tdb_context *tdb);
468 void tdb_mmap(struct tdb_context *tdb);
470 /* Either alloc a copy, or give direct access. Release frees or noop. */
471 const void *tdb_access_read(struct tdb_context *tdb,
472 tdb_off_t off, tdb_len_t len, bool convert);
473 void *tdb_access_write(struct tdb_context *tdb,
474 tdb_off_t off, tdb_len_t len, bool convert);
476 /* Release result of tdb_access_read/write. */
477 void tdb_access_release(struct tdb_context *tdb, const void *p);
478 /* Commit result of tdb_acces_write. */
479 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
481 /* Convenience routine to get an offset. */
482 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
484 /* Write an offset at an offset. */
485 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
488 /* Clear an ondisk area. */
489 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
491 /* Return a non-zero offset between >= start < end in this array (or end). */
492 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
497 /* Return a zero offset in this array, or num. */
498 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
501 /* Allocate and make a copy of some offset. */
502 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
504 /* Writes a converted copy of a record. */
505 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
506 const void *rec, size_t len);
508 /* Reads record and converts it */
509 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
510 void *rec, size_t len);
512 /* Adds a stat, if it's in range. */
513 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
514 #define add_stat(tdb, statname, val) \
516 if (unlikely((tdb)->stats)) \
517 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
521 void tdb_lock_init(struct tdb_context *tdb);
523 /* Lock/unlock a range of hashes. */
524 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
525 tdb_off_t hash_lock, tdb_len_t hash_range,
526 int ltype, enum tdb_lock_flags waitflag);
527 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
529 tdb_len_t hash_range, int ltype);
531 /* Lock/unlock a particular free bucket. */
532 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
533 enum tdb_lock_flags waitflag);
534 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
536 /* Serialize transaction start. */
537 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
538 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
540 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
541 bool tdb_has_hash_locks(struct tdb_context *tdb);
543 /* Lock entire database. */
544 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
545 enum tdb_lock_flags flags, bool upgradable);
546 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
547 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
549 /* Serialize db open. */
550 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
551 enum tdb_lock_flags flags);
552 void tdb_unlock_open(struct tdb_context *tdb);
553 bool tdb_has_open_lock(struct tdb_context *tdb);
555 /* Serialize db expand. */
556 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
557 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
558 bool tdb_has_expansion_lock(struct tdb_context *tdb);
560 /* If it needs recovery, grab all the locks and do it. */
561 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
564 int tdb_transaction_recover(struct tdb_context *tdb);
565 bool tdb_needs_recovery(struct tdb_context *tdb);
568 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
569 enum TDB_ERROR ecode,
570 enum tdb_log_level level,
571 const char *fmt, ...);
574 void tdb_trace(struct tdb_context *tdb, const char *op);
575 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
576 void tdb_trace_open(struct tdb_context *tdb, const char *op,
577 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
578 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
579 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
580 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
582 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
583 TDB_DATA rec, int ret);
584 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
585 TDB_DATA rec, TDB_DATA ret);
586 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
587 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
589 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
590 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
592 #define tdb_trace(tdb, op)
593 #define tdb_trace_seqnum(tdb, seqnum, op)
594 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
595 #define tdb_trace_ret(tdb, op, ret)
596 #define tdb_trace_retrec(tdb, op, ret)
597 #define tdb_trace_1rec(tdb, op, rec)
598 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
599 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
600 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
601 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
602 #endif /* !TDB_TRACE */