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_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
78 /* Packing errors into pointers and v.v. */
79 #define TDB_PTR_IS_ERR(ptr) \
80 unlikely((void *)(ptr) >= (void *)(long)TDB_ERR_LAST)
81 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
82 #define TDB_ERR_PTR(err) ((void *)(long)(err))
84 /* Common case of returning true, false or -ve error. */
85 typedef int tdb_bool_err;
87 /* Prevent others from opening the file. */
88 #define TDB_OPEN_LOCK 0
89 /* Doing a transaction. */
90 #define TDB_TRANSACTION_LOCK 1
92 #define TDB_EXPANSION_LOCK 2
93 /* Hash chain locks. */
94 #define TDB_HASH_LOCK_START 3
96 /* Range for hash locks. */
97 #define TDB_HASH_LOCK_RANGE_BITS 30
98 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
100 /* We have 1024 entries in the top level. */
101 #define TDB_TOPLEVEL_HASH_BITS 10
102 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
103 #define TDB_SUBLEVEL_HASH_BITS 6
104 /* And 8 entries in each group, ie 8 groups per sublevel. */
105 #define TDB_HASH_GROUP_BITS 3
106 /* This is currently 10: beyond this we chain. */
107 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
109 /* Extend file by least 100 times larger than needed. */
110 #define TDB_EXTENSION_FACTOR 100
112 /* We steal bits from the offsets to store hash info. */
113 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
114 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
115 #define TDB_OFF_UPPER_STEAL 8
116 #define TDB_OFF_UPPER_STEAL_EXTRA 7
117 /* The bit number where we store extra hash bits. */
118 #define TDB_OFF_HASH_EXTRA_BIT 57
119 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
121 /* The bit number where we store the extra hash bits. */
122 /* Convenience mask to get actual offset. */
123 #define TDB_OFF_MASK \
124 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
126 /* How many buckets in a free list: see size_to_bucket(). */
127 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
129 /* We have to be able to fit a free record here. */
130 #define TDB_MIN_DATA_LEN \
131 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
133 /* Indicates this entry is not on an flist (can happen during coalescing) */
134 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
137 static inline uint64_t bswap_64(uint64_t x)
139 return (((x&0x000000FFULL)<<56)
140 | ((x&0x0000FF00ULL)<<48)
141 | ((x&0x00FF0000ULL)<<40)
142 | ((x&0xFF000000ULL)<<32)
143 | ((x>>8)&0xFF000000ULL)
144 | ((x>>16)&0x00FF0000ULL)
145 | ((x>>24)&0x0000FF00ULL)
146 | ((x>>32)&0x000000FFULL));
150 struct tdb_used_record {
151 /* For on-disk compatibility, we avoid bitfields:
157 uint64_t magic_and_meta;
158 /* The bottom key_len_bits*2 are key length, rest is data length. */
159 uint64_t key_and_data_len;
162 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
164 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
167 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
169 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
172 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
174 return r->key_and_data_len >> rec_key_bits(r);
177 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
179 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
182 static inline uint32_t rec_hash(const struct tdb_used_record *r)
184 return r->magic_and_meta & ((1 << 11) - 1);
187 static inline uint16_t rec_magic(const struct tdb_used_record *r)
189 return (r->magic_and_meta >> 48);
192 struct tdb_free_record {
193 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
194 uint64_t ftable_and_len; /* Len not counting these two fields. */
195 /* This is why the minimum record size is 8 bytes. */
199 static inline uint64_t frec_prev(const struct tdb_free_record *f)
201 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
204 static inline uint64_t frec_magic(const struct tdb_free_record *f)
206 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
209 static inline uint64_t frec_len(const struct tdb_free_record *f)
211 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
214 static inline unsigned frec_ftable(const struct tdb_free_record *f)
216 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
219 struct tdb_recovery_record {
221 /* Length of record. */
225 /* Old length of file before transaction. */
229 /* If we bottom out of the subhashes, we chain. */
231 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
235 /* this is stored at the front of every database */
237 char magic_food[64]; /* for /etc/magic */
238 /* FIXME: Make me 32 bit? */
239 uint64_t version; /* version of the code */
240 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
241 uint64_t hash_seed; /* "random" seed written at creation time. */
242 tdb_off_t free_table; /* (First) free table. */
243 tdb_off_t recovery; /* Transaction recovery area. */
245 tdb_off_t reserved[26];
247 /* Top level hash table. */
248 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
251 struct tdb_freetable {
252 struct tdb_used_record hdr;
254 tdb_off_t buckets[TDB_FREE_BUCKETS];
257 /* Information about a particular (locked) hash entry. */
259 /* Full hash value of entry. */
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];
276 struct traverse_info {
277 struct traverse_level {
279 /* We ignore groups here, and treat it as a big array. */
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. */
289 enum tdb_lock_flags {
290 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
293 /* If set, don't log an error on failure. */
295 /* If set, don't check for recovery (used by recovery code). */
296 TDB_LOCK_NOCHECK = 4,
299 struct tdb_lock_type {
305 /* This is only needed for tdb_access_commit, but used everywhere to
307 struct tdb_access_hdr {
308 struct tdb_access_hdr *next;
315 /* Filename of the database. */
318 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
321 /* Are we accessing directly? (debugging check). */
324 /* Open file descriptor (undefined for TDB_INTERNAL). */
327 /* How much space has been mapped (<= current file size) */
330 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
333 /* mmap read only? */
336 /* Error code for last tdb error. */
337 enum TDB_ERROR ecode;
339 /* the flags passed to tdb_open, for tdb_reopen. */
342 /* Logging function */
343 void (*logfn)(struct tdb_context *tdb,
344 enum tdb_log_level level,
346 const char *message);
350 uint64_t (*khash)(const void *key, size_t len, uint64_t seed, void *);
354 /* Set if we are in a transaction. */
355 struct tdb_transaction *transaction;
357 /* What free table are we using? */
358 tdb_off_t ftable_off;
361 /* IO methods: changes for transactions. */
362 const struct tdb_methods *methods;
364 /* Lock information */
365 struct tdb_lock_type allrecord_lock;
367 struct tdb_lock_type *lockrecs;
369 struct tdb_attribute_stats *stats;
371 /* Direct access information */
372 struct tdb_access_hdr *access;
374 /* Single list of all TDBs, to avoid multiple opens. */
375 struct tdb_context *next;
381 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
383 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
385 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
386 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
387 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
394 void tdb_hash_init(struct tdb_context *tdb);
396 tdb_bool_err first_in_hash(struct tdb_context *tdb,
397 struct traverse_info *tinfo,
398 TDB_DATA *kbuf, size_t *dlen);
400 tdb_bool_err next_in_hash(struct tdb_context *tdb,
401 struct traverse_info *tinfo,
402 TDB_DATA *kbuf, size_t *dlen);
404 /* Hash random memory. */
405 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
408 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
410 /* Find and lock a hash entry (or where it would be). */
411 tdb_off_t find_and_lock(struct tdb_context *tdb,
415 struct tdb_used_record *rec,
416 struct traverse_info *tinfo);
418 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
422 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
425 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
428 bool is_subhash(tdb_off_t val);
431 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
433 /* check.c needs these to iterate through free lists. */
434 tdb_off_t first_ftable(struct tdb_context *tdb);
435 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
437 /* This returns space or -ve error number. */
438 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
439 uint64_t hash, unsigned magic, bool growing);
441 /* Put this record in a free list. */
442 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
443 tdb_off_t off, tdb_len_t len_with_header);
445 /* Set up header for a used/ftable/htable/chain record. */
446 enum TDB_ERROR set_header(struct tdb_context *tdb,
447 struct tdb_used_record *rec,
448 unsigned magic, uint64_t keylen, uint64_t datalen,
449 uint64_t actuallen, unsigned hashlow);
451 /* Used by tdb_check to verify. */
452 unsigned int size_to_bucket(tdb_len_t data_len);
453 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
455 /* Used by tdb_summary */
456 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
459 /* Initialize tdb->methods. */
460 void tdb_io_init(struct tdb_context *tdb);
462 /* Convert endian of the buffer if required. */
463 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
465 /* Unmap and try to map the tdb. */
466 void tdb_munmap(struct tdb_context *tdb);
467 void tdb_mmap(struct tdb_context *tdb);
469 /* Either alloc a copy, or give direct access. Release frees or noop. */
470 const void *tdb_access_read(struct tdb_context *tdb,
471 tdb_off_t off, tdb_len_t len, bool convert);
472 void *tdb_access_write(struct tdb_context *tdb,
473 tdb_off_t off, tdb_len_t len, bool convert);
475 /* Release result of tdb_access_read/write. */
476 void tdb_access_release(struct tdb_context *tdb, const void *p);
477 /* Commit result of tdb_acces_write. */
478 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
480 /* Convenience routine to get an offset. */
481 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
483 /* Write an offset at an offset. */
484 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
487 /* Clear an ondisk area. */
488 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
490 /* Return a non-zero offset between >= start < end in this array (or end). */
491 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
496 /* Return a zero offset in this array, or num. */
497 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
500 /* Allocate and make a copy of some offset. */
501 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
503 /* Writes a converted copy of a record. */
504 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
505 const void *rec, size_t len);
507 /* Reads record and converts it */
508 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
509 void *rec, size_t len);
511 /* Adds a stat, if it's in range. */
512 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
513 #define add_stat(tdb, statname, val) \
515 if (unlikely((tdb)->stats)) \
516 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
520 void tdb_lock_init(struct tdb_context *tdb);
522 /* Lock/unlock a range of hashes. */
523 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
524 tdb_off_t hash_lock, tdb_len_t hash_range,
525 int ltype, enum tdb_lock_flags waitflag);
526 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
528 tdb_len_t hash_range, int ltype);
530 /* Lock/unlock a particular free bucket. */
531 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
532 enum tdb_lock_flags waitflag);
533 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
535 /* Serialize transaction start. */
536 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
537 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
539 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
540 bool tdb_has_hash_locks(struct tdb_context *tdb);
542 /* Lock entire database. */
543 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
544 enum tdb_lock_flags flags, bool upgradable);
545 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
546 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
548 /* Serialize db open. */
549 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
550 enum tdb_lock_flags flags);
551 void tdb_unlock_open(struct tdb_context *tdb);
552 bool tdb_has_open_lock(struct tdb_context *tdb);
554 /* Serialize db expand. */
555 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
556 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
557 bool tdb_has_expansion_lock(struct tdb_context *tdb);
559 /* If it needs recovery, grab all the locks and do it. */
560 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
563 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
564 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
567 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
568 enum TDB_ERROR ecode,
569 enum tdb_log_level level,
570 const char *fmt, ...);
573 void tdb_trace(struct tdb_context *tdb, const char *op);
574 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
575 void tdb_trace_open(struct tdb_context *tdb, const char *op,
576 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
577 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
578 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
579 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
581 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
582 TDB_DATA rec, int ret);
583 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
584 TDB_DATA rec, TDB_DATA ret);
585 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
586 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
588 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
589 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
591 #define tdb_trace(tdb, op)
592 #define tdb_trace_seqnum(tdb, seqnum, op)
593 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
594 #define tdb_trace_ret(tdb, op, ret)
595 #define tdb_trace_retrec(tdb, op, ret)
596 #define tdb_trace_1rec(tdb, op, rec)
597 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
598 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
599 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
600 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
601 #endif /* !TDB_TRACE */