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
36 #include <ccan/tdb2/tdb2.h>
37 #include <ccan/likely/likely.h>
38 #ifdef HAVE_BYTESWAP_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;
64 #define offsetof(t,f) ((unsigned int)&((t *)0)->f)
67 #define TDB_MAGIC_FOOD "TDB file\n"
68 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
69 #define TDB_MAGIC ((uint64_t)0x1999)
70 #define TDB_FREE_MAGIC ((~(uint64_t)TDB_MAGIC) << 6)
71 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
72 #define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
73 #define TDB_RECOVERY_INVALID_MAGIC (0x0)
75 #define TDB_OFF_ERR ((tdb_off_t)-1)
77 /* Prevent others from opening the file. */
78 #define TDB_OPEN_LOCK 0
79 /* Doing a transaction. */
80 #define TDB_TRANSACTION_LOCK 1
82 #define TDB_EXPANSION_LOCK 2
83 /* Hash chain locks. */
84 #define TDB_HASH_LOCK_START 3
86 /* Range for hash locks. */
87 #define TDB_HASH_LOCK_RANGE_BITS 30
88 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
90 /* We have 1024 entries in the top level. */
91 #define TDB_TOPLEVEL_HASH_BITS 10
92 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
93 #define TDB_SUBLEVEL_HASH_BITS 6
94 /* And 8 entries in each group, ie 8 groups per sublevel. */
95 #define TDB_HASH_GROUP_BITS 3
97 /* We start with a 64k-sized zone. */
98 #define INITIAL_ZONE_BITS 16
99 /* Try to create zones at least 32 times larger than allocations. */
100 #define TDB_COMFORT_FACTOR_BITS 5
102 /* We steal bits from the offsets to store hash info. */
103 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
104 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
105 #define TDB_OFF_UPPER_STEAL 8
106 #define TDB_OFF_UPPER_STEAL_EXTRA 7
107 #define TDB_OFF_UPPER_STEAL_TRUNCBIT 1
108 /* If this is set, hash is truncated (only 1 bit is valid). */
109 #define TDB_OFF_HASH_TRUNCATED_BIT 56
110 /* The bit number where we store next level of hash. */
111 #define TDB_OFF_HASH_EXTRA_BIT 57
112 /* Convenience mask to get actual offset. */
113 #define TDB_OFF_MASK \
114 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
116 /* We have to be able to fit a free record here. */
117 #define TDB_MIN_DATA_LEN \
118 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
120 /* We ensure buckets up to size 1 << (zone_bits - TDB_COMFORT_FACTOR_BITS). */
121 /* FIXME: test this matches size_to_bucket! */
122 #define BUCKETS_FOR_ZONE(zone_bits) ((zone_bits) + 2 - TDB_COMFORT_FACTOR_BITS)
125 static inline uint64_t bswap_64(uint64_t x)
127 return (((x&0x000000FFULL)<<56)
128 | ((x&0x0000FF00ULL)<<48)
129 | ((x&0x00FF0000ULL)<<40)
130 | ((x&0xFF000000ULL)<<32)
131 | ((x>>8)&0xFF000000ULL)
132 | ((x>>16)&0x00FF0000ULL)
133 | ((x>>24)&0x0000FF00ULL)
134 | ((x>>32)&0x000000FFULL));
138 struct tdb_used_record {
139 /* For on-disk compatibility, we avoid bitfields:
144 zone_bits: 6 (lowest)
146 uint64_t magic_and_meta;
147 /* The bottom key_len_bits*2 are key length, rest is data length. */
148 uint64_t key_and_data_len;
151 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
153 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
156 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
158 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
161 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
163 return r->key_and_data_len >> rec_key_bits(r);
166 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
168 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
171 static inline unsigned int rec_zone_bits(const struct tdb_used_record *r)
173 return r->magic_and_meta & ((1 << 6) - 1);
176 static inline uint32_t rec_hash(const struct tdb_used_record *r)
178 return (r->magic_and_meta >> 6) & ((1 << 5) - 1);
181 static inline uint16_t rec_magic(const struct tdb_used_record *r)
183 return (r->magic_and_meta >> 48);
186 struct tdb_free_record {
187 uint64_t magic_and_meta; /* Bottom 6 bits are zone bits. */
188 uint64_t data_len; /* Not counting these two fields. */
189 /* This is why the minimum record size is 16 bytes. */
193 static inline unsigned int frec_zone_bits(const struct tdb_free_record *f)
195 return f->magic_and_meta & ((1 << 6) - 1);
198 static inline uint64_t frec_magic(const struct tdb_free_record *f)
200 return f->magic_and_meta & ~((1ULL << 6) - 1);
203 /* Each zone has its set of free lists at the head.
205 * For each zone we have a series of per-size buckets, and a final bucket for
207 struct free_zone_header {
208 /* How much does this zone cover? */
210 /* tdb_off_t buckets[free_buckets + 1] */
213 /* this is stored at the front of every database */
215 char magic_food[64]; /* for /etc/magic */
216 /* FIXME: Make me 32 bit? */
217 uint64_t version; /* version of the code */
218 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
219 uint64_t hash_seed; /* "random" seed written at creation time. */
221 tdb_off_t reserved[28];
223 /* Top level hash table. */
224 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
227 /* Information about a particular (locked) hash entry. */
229 /* Full hash value of entry. */
231 /* Start and length of lock acquired. */
232 tdb_off_t hlock_start;
233 tdb_len_t hlock_range;
234 /* Start of hash group. */
235 tdb_off_t group_start;
236 /* Bucket we belong in. */
237 unsigned int home_bucket;
238 /* Bucket we (or an empty space) were found in. */
239 unsigned int found_bucket;
240 /* How many bits of the hash are already used. */
241 unsigned int hash_used;
242 /* Current working group. */
243 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
246 struct traverse_info {
247 struct traverse_level {
249 /* We ignore groups here, and treat it as a big array. */
251 unsigned int total_buckets;
252 } levels[64 / TDB_SUBLEVEL_HASH_BITS];
253 unsigned int num_levels;
254 unsigned int toplevel_group;
255 /* This makes delete-everything-inside-traverse work as expected. */
259 enum tdb_lock_flags {
260 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
263 /* If set, don't log an error on failure. */
267 struct tdb_lock_type {
274 /* Filename of the database. */
277 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
280 /* Are we accessing directly? (debugging check). */
283 /* Open file descriptor (undefined for TDB_INTERNAL). */
286 /* How much space has been mapped (<= current file size) */
289 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
292 /* mmap read only? */
295 /* Error code for last tdb error. */
296 enum TDB_ERROR ecode;
298 /* the flags passed to tdb_open, for tdb_reopen. */
301 /* Logging function */
310 /* Set if we are in a transaction. */
311 struct tdb_transaction *transaction;
313 /* What zone of the tdb to use, for spreading load. */
315 /* Cached copy of zone header. */
316 struct free_zone_header zhdr;
318 /* IO methods: changes for transactions. */
319 const struct tdb_methods *methods;
321 /* Lock information */
322 struct tdb_lock_type allrecord_lock;
323 uint64_t num_lockrecs;
324 struct tdb_lock_type *lockrecs;
326 /* Single list of all TDBs, to avoid multiple opens. */
327 struct tdb_context *next;
333 int (*read)(struct tdb_context *, tdb_off_t, void *, tdb_len_t);
334 int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
335 int (*oob)(struct tdb_context *, tdb_off_t, bool);
336 int (*expand_file)(struct tdb_context *, tdb_len_t);
343 void tdb_hash_init(struct tdb_context *tdb);
345 /* Hash random memory. */
346 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
349 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
351 /* Find and lock a hash entry (or where it would be). */
352 tdb_off_t find_and_lock(struct tdb_context *tdb,
356 struct tdb_used_record *rec,
357 struct traverse_info *tinfo);
359 int replace_in_hash(struct tdb_context *tdb,
363 int add_to_hash(struct tdb_context *tdb, struct hash_info *h,
366 int delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
369 bool is_subhash(tdb_off_t val);
372 int tdb_zone_init(struct tdb_context *tdb);
374 /* If this fails, try tdb_expand. */
375 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
376 uint64_t hash, bool growing);
378 /* Put this record in a free list. */
379 int add_free_record(struct tdb_context *tdb, unsigned int zone_bits,
380 tdb_off_t off, tdb_len_t len_with_header);
382 /* Set up header for a used record. */
383 int set_header(struct tdb_context *tdb,
384 struct tdb_used_record *rec,
385 uint64_t keylen, uint64_t datalen,
386 uint64_t actuallen, uint64_t hash,
387 unsigned int zone_bits);
389 /* Used by tdb_check to verify. */
390 unsigned int size_to_bucket(unsigned int free_buckets, tdb_len_t data_len);
391 tdb_off_t bucket_off(tdb_off_t zone_off, tdb_off_t bucket);
394 /* Initialize tdb->methods. */
395 void tdb_io_init(struct tdb_context *tdb);
397 /* Convert endian of the buffer if required. */
398 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
400 /* Unmap and try to map the tdb. */
401 void tdb_munmap(struct tdb_context *tdb);
402 void tdb_mmap(struct tdb_context *tdb);
404 /* Either make a copy into pad and return that, or return ptr into mmap.
405 * Converts endian (ie. will use pad in that case). */
406 void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len);
408 /* Either alloc a copy, or give direct access. Release frees or noop. */
409 const void *tdb_access_read(struct tdb_context *tdb,
410 tdb_off_t off, tdb_len_t len, bool convert);
411 void *tdb_access_write(struct tdb_context *tdb,
412 tdb_off_t off, tdb_len_t len, bool convert);
414 /* Release result of tdb_access_read/write. */
415 void tdb_access_release(struct tdb_context *tdb, const void *p);
416 /* Commit result of tdb_acces_write. */
417 int tdb_access_commit(struct tdb_context *tdb, void *p);
419 /* Convenience routine to get an offset. */
420 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
422 /* Write an offset at an offset. */
423 int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val);
425 /* Clear an ondisk area. */
426 int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
428 /* Return a non-zero offset between >= start < end in this array (or end). */
429 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
434 /* Return a zero offset in this array, or num. */
435 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
438 /* Even on files, we can get partial writes due to signals. */
439 bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off);
440 bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off);
441 bool tdb_read_all(int fd, void *buf, size_t len);
443 /* Allocate and make a copy of some offset. */
444 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
446 /* Writes a converted copy of a record. */
447 int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
448 const void *rec, size_t len);
450 /* Reads record and converts it */
451 int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
452 void *rec, size_t len);
456 void tdb_lock_init(struct tdb_context *tdb);
458 /* Lock/unlock a range of hashes. */
459 int tdb_lock_hashes(struct tdb_context *tdb,
460 tdb_off_t hash_lock, tdb_len_t hash_range,
461 int ltype, enum tdb_lock_flags waitflag);
462 int tdb_unlock_hashes(struct tdb_context *tdb,
464 tdb_len_t hash_range, int ltype);
466 /* Lock/unlock a particular free bucket. */
467 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
468 enum tdb_lock_flags waitflag);
469 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
471 /* Do we have any locks? */
472 bool tdb_has_locks(struct tdb_context *tdb);
474 /* Lock entire database. */
475 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
476 enum tdb_lock_flags flags, bool upgradable);
477 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
479 /* Serialize db open. */
480 int tdb_lock_open(struct tdb_context *tdb);
481 void tdb_unlock_open(struct tdb_context *tdb);
483 /* Serialize db expand. */
484 int tdb_lock_expand(struct tdb_context *tdb, int ltype);
485 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
486 bool tdb_has_expansion_lock(struct tdb_context *tdb);
490 int first_in_hash(struct tdb_context *tdb, int ltype,
491 struct traverse_info *tinfo,
492 TDB_DATA *kbuf, size_t *dlen);
493 int next_in_hash(struct tdb_context *tdb, int ltype,
494 struct traverse_info *tinfo,
495 TDB_DATA *kbuf, size_t *dlen);
499 /* Low-level locking primitives. */
500 int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
501 enum tdb_lock_flags flags);
502 int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t offset, int ltype);
504 int tdb_munmap(struct tdb_context *tdb);
505 void tdb_mmap(struct tdb_context *tdb);
506 int tdb_lock(struct tdb_context *tdb, int list, int ltype);
507 int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
508 bool tdb_have_locks(struct tdb_context *tdb);
509 int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
510 int tdb_brlock(struct tdb_context *tdb,
511 int rw_type, tdb_off_t offset, size_t len,
512 enum tdb_lock_flags flags);
513 int tdb_brunlock(struct tdb_context *tdb,
514 int rw_type, tdb_off_t offset, size_t len);
515 bool tdb_have_extra_locks(struct tdb_context *tdb);
516 void tdb_release_extra_locks(struct tdb_context *tdb);
517 int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
518 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
519 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
520 enum tdb_lock_flags flags, bool upgradable);
521 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
522 int tdb_allrecord_upgrade(struct tdb_context *tdb);
523 int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
524 int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
525 int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
526 int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
527 int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
528 tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec);
529 int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
530 int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
531 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
532 int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
533 bool tdb_needs_recovery(struct tdb_context *tdb);
534 int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
535 int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
536 int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct tdb_record *rec);
537 unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
538 int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
539 tdb_off_t offset, tdb_len_t len,
540 int (*parser)(TDB_DATA key, TDB_DATA data,
543 tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
544 struct tdb_record *rec);
545 void tdb_io_init(struct tdb_context *tdb);
546 int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
547 int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off,
548 struct tdb_record *rec);
552 void tdb_trace(struct tdb_context *tdb, const char *op);
553 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
554 void tdb_trace_open(struct tdb_context *tdb, const char *op,
555 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
556 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
557 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
558 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
560 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
561 TDB_DATA rec, int ret);
562 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
563 TDB_DATA rec, TDB_DATA ret);
564 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
565 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
567 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
568 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
570 #define tdb_trace(tdb, op)
571 #define tdb_trace_seqnum(tdb, seqnum, op)
572 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
573 #define tdb_trace_ret(tdb, op, ret)
574 #define tdb_trace_retrec(tdb, op, ret)
575 #define tdb_trace_1rec(tdb, op, rec)
576 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
577 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
578 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
579 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
580 #endif /* !TDB_TRACE */