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>
40 #ifdef HAVE_BYTESWAP_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)
78 /* Prevent others from opening the file. */
79 #define TDB_OPEN_LOCK 0
80 /* Doing a transaction. */
81 #define TDB_TRANSACTION_LOCK 1
83 #define TDB_EXPANSION_LOCK 2
84 /* Hash chain locks. */
85 #define TDB_HASH_LOCK_START 3
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 /* The bit number where we store the extra hash bits. */
113 /* Convenience mask to get actual offset. */
114 #define TDB_OFF_MASK \
115 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
117 /* How many buckets in a free list: see size_to_bucket(). */
118 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
120 /* We have to be able to fit a free record here. */
121 #define TDB_MIN_DATA_LEN \
122 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
124 /* Indicates this entry is not on an flist (can happen during coalescing) */
125 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
128 static inline uint64_t bswap_64(uint64_t x)
130 return (((x&0x000000FFULL)<<56)
131 | ((x&0x0000FF00ULL)<<48)
132 | ((x&0x00FF0000ULL)<<40)
133 | ((x&0xFF000000ULL)<<32)
134 | ((x>>8)&0xFF000000ULL)
135 | ((x>>16)&0x00FF0000ULL)
136 | ((x>>24)&0x0000FF00ULL)
137 | ((x>>32)&0x000000FFULL));
141 struct tdb_used_record {
142 /* For on-disk compatibility, we avoid bitfields:
148 uint64_t magic_and_meta;
149 /* The bottom key_len_bits*2 are key length, rest is data length. */
150 uint64_t key_and_data_len;
153 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
155 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
158 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
160 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
163 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
165 return r->key_and_data_len >> rec_key_bits(r);
168 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
170 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
173 static inline uint32_t rec_hash(const struct tdb_used_record *r)
175 return r->magic_and_meta & ((1 << 11) - 1);
178 static inline uint16_t rec_magic(const struct tdb_used_record *r)
180 return (r->magic_and_meta >> 48);
183 struct tdb_free_record {
184 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
185 uint64_t ftable_and_len; /* Len not counting these two fields. */
186 /* This is why the minimum record size is 8 bytes. */
190 static inline uint64_t frec_prev(const struct tdb_free_record *f)
192 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
195 static inline uint64_t frec_magic(const struct tdb_free_record *f)
197 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
200 static inline uint64_t frec_len(const struct tdb_free_record *f)
202 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
205 static inline unsigned frec_ftable(const struct tdb_free_record *f)
207 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
210 struct tdb_recovery_record {
212 /* Length of record. */
216 /* Old length of file before transaction. */
220 /* If we bottom out of the subhashes, we chain. */
222 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
226 /* this is stored at the front of every database */
228 char magic_food[64]; /* for /etc/magic */
229 /* FIXME: Make me 32 bit? */
230 uint64_t version; /* version of the code */
231 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
232 uint64_t hash_seed; /* "random" seed written at creation time. */
233 tdb_off_t free_table; /* (First) free table. */
234 tdb_off_t recovery; /* Transaction recovery area. */
236 tdb_off_t reserved[26];
238 /* Top level hash table. */
239 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
242 struct tdb_freetable {
243 struct tdb_used_record hdr;
245 tdb_off_t buckets[TDB_FREE_BUCKETS];
248 /* Information about a particular (locked) hash entry. */
250 /* Full hash value of entry. */
252 /* Start and length of lock acquired. */
253 tdb_off_t hlock_start;
254 tdb_len_t hlock_range;
255 /* Start of hash group. */
256 tdb_off_t group_start;
257 /* Bucket we belong in. */
258 unsigned int home_bucket;
259 /* Bucket we (or an empty space) were found in. */
260 unsigned int found_bucket;
261 /* How many bits of the hash are already used. */
262 unsigned int hash_used;
263 /* Current working group. */
264 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
267 struct traverse_info {
268 struct traverse_level {
270 /* We ignore groups here, and treat it as a big array. */
272 unsigned int total_buckets;
273 } levels[TDB_MAX_LEVELS + 1];
274 unsigned int num_levels;
275 unsigned int toplevel_group;
276 /* This makes delete-everything-inside-traverse work as expected. */
280 enum tdb_lock_flags {
281 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
284 /* If set, don't log an error on failure. */
286 /* If set, don't check for recovery (used by recovery code). */
287 TDB_LOCK_NOCHECK = 4,
290 struct tdb_lock_type {
296 /* This is only needed for tdb_access_commit, but used everywhere to
298 struct tdb_access_hdr {
299 struct tdb_access_hdr *next;
306 /* Filename of the database. */
309 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
312 /* Are we accessing directly? (debugging check). */
315 /* Open file descriptor (undefined for TDB_INTERNAL). */
318 /* How much space has been mapped (<= current file size) */
321 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
324 /* mmap read only? */
327 /* Error code for last tdb error. */
328 enum TDB_ERROR ecode;
330 /* the flags passed to tdb_open, for tdb_reopen. */
333 /* Logging function */
342 /* Set if we are in a transaction. */
343 struct tdb_transaction *transaction;
345 /* What free table are we using? */
346 tdb_off_t ftable_off;
349 /* IO methods: changes for transactions. */
350 const struct tdb_methods *methods;
352 /* Lock information */
353 struct tdb_lock_type allrecord_lock;
355 struct tdb_lock_type *lockrecs;
357 struct tdb_attribute_stats *stats;
359 /* Direct access information */
360 struct tdb_access_hdr *access;
362 /* Single list of all TDBs, to avoid multiple opens. */
363 struct tdb_context *next;
369 int (*read)(struct tdb_context *, tdb_off_t, void *, tdb_len_t);
370 int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
371 int (*oob)(struct tdb_context *, tdb_off_t, bool);
372 int (*expand_file)(struct tdb_context *, tdb_len_t);
373 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
380 void tdb_hash_init(struct tdb_context *tdb);
382 /* Hash random memory. */
383 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
386 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
388 /* Find and lock a hash entry (or where it would be). */
389 tdb_off_t find_and_lock(struct tdb_context *tdb,
393 struct tdb_used_record *rec,
394 struct traverse_info *tinfo);
396 int replace_in_hash(struct tdb_context *tdb,
400 int add_to_hash(struct tdb_context *tdb, struct hash_info *h,
403 int delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
406 bool is_subhash(tdb_off_t val);
409 int tdb_ftable_init(struct tdb_context *tdb);
411 /* check.c needs these to iterate through free lists. */
412 tdb_off_t first_ftable(struct tdb_context *tdb);
413 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
415 /* This returns space or TDB_OFF_ERR. */
416 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
417 uint64_t hash, unsigned magic, bool growing);
419 /* Put this record in a free list. */
420 int add_free_record(struct tdb_context *tdb,
421 tdb_off_t off, tdb_len_t len_with_header);
423 /* Set up header for a used/ftable/htable/chain record. */
424 int set_header(struct tdb_context *tdb,
425 struct tdb_used_record *rec,
426 unsigned magic, uint64_t keylen, uint64_t datalen,
427 uint64_t actuallen, unsigned hashlow);
429 /* Used by tdb_check to verify. */
430 unsigned int size_to_bucket(tdb_len_t data_len);
431 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
433 /* Used by tdb_summary */
434 size_t dead_space(struct tdb_context *tdb, tdb_off_t off);
437 /* Initialize tdb->methods. */
438 void tdb_io_init(struct tdb_context *tdb);
440 /* Convert endian of the buffer if required. */
441 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
443 /* Unmap and try to map the tdb. */
444 void tdb_munmap(struct tdb_context *tdb);
445 void tdb_mmap(struct tdb_context *tdb);
447 /* Either alloc a copy, or give direct access. Release frees or noop. */
448 const void *tdb_access_read(struct tdb_context *tdb,
449 tdb_off_t off, tdb_len_t len, bool convert);
450 void *tdb_access_write(struct tdb_context *tdb,
451 tdb_off_t off, tdb_len_t len, bool convert);
453 /* Release result of tdb_access_read/write. */
454 void tdb_access_release(struct tdb_context *tdb, const void *p);
455 /* Commit result of tdb_acces_write. */
456 int tdb_access_commit(struct tdb_context *tdb, void *p);
458 /* Convenience routine to get an offset. */
459 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
461 /* Write an offset at an offset. */
462 int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val);
464 /* Clear an ondisk area. */
465 int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
467 /* Return a non-zero offset between >= start < end in this array (or end). */
468 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
473 /* Return a zero offset in this array, or num. */
474 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
477 /* Even on files, we can get partial writes due to signals. */
478 bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off);
479 bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off);
480 bool tdb_read_all(int fd, void *buf, size_t len);
482 /* Allocate and make a copy of some offset. */
483 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
485 /* Writes a converted copy of a record. */
486 int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
487 const void *rec, size_t len);
489 /* Reads record and converts it */
490 int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
491 void *rec, size_t len);
493 /* Adds a stat, if it's in range. */
494 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
495 #define add_stat(tdb, statname, val) \
497 if (unlikely((tdb)->stats)) \
498 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
502 void tdb_lock_init(struct tdb_context *tdb);
504 /* Lock/unlock a range of hashes. */
505 int tdb_lock_hashes(struct tdb_context *tdb,
506 tdb_off_t hash_lock, tdb_len_t hash_range,
507 int ltype, enum tdb_lock_flags waitflag);
508 int tdb_unlock_hashes(struct tdb_context *tdb,
510 tdb_len_t hash_range, int ltype);
512 /* Lock/unlock a particular free bucket. */
513 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
514 enum tdb_lock_flags waitflag);
515 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
517 /* Serialize transaction start. */
518 int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
519 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
521 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
522 bool tdb_has_hash_locks(struct tdb_context *tdb);
524 /* Lock entire database. */
525 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
526 enum tdb_lock_flags flags, bool upgradable);
527 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
528 int tdb_allrecord_upgrade(struct tdb_context *tdb);
530 /* Serialize db open. */
531 int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
532 void tdb_unlock_open(struct tdb_context *tdb);
533 bool tdb_has_open_lock(struct tdb_context *tdb);
535 /* Serialize db expand. */
536 int tdb_lock_expand(struct tdb_context *tdb, int ltype);
537 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
538 bool tdb_has_expansion_lock(struct tdb_context *tdb);
540 /* If it needs recovery, grab all the locks and do it. */
541 int tdb_lock_and_recover(struct tdb_context *tdb);
544 int first_in_hash(struct tdb_context *tdb, int ltype,
545 struct traverse_info *tinfo,
546 TDB_DATA *kbuf, size_t *dlen);
547 int next_in_hash(struct tdb_context *tdb, int ltype,
548 struct traverse_info *tinfo,
549 TDB_DATA *kbuf, size_t *dlen);
552 int tdb_transaction_recover(struct tdb_context *tdb);
553 bool tdb_needs_recovery(struct tdb_context *tdb);
556 void COLD tdb_logerr(struct tdb_context *tdb,
557 enum TDB_ERROR ecode,
558 enum tdb_debug_level level,
559 const char *fmt, ...);
562 void tdb_trace(struct tdb_context *tdb, const char *op);
563 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
564 void tdb_trace_open(struct tdb_context *tdb, const char *op,
565 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
566 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
567 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
568 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
570 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
571 TDB_DATA rec, int ret);
572 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
573 TDB_DATA rec, TDB_DATA ret);
574 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
575 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
577 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
578 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
580 #define tdb_trace(tdb, op)
581 #define tdb_trace_seqnum(tdb, seqnum, op)
582 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
583 #define tdb_trace_ret(tdb, op, ret)
584 #define tdb_trace_retrec(tdb, op, ret)
585 #define tdb_trace_1rec(tdb, op, rec)
586 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
587 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
588 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
589 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
590 #endif /* !TDB_TRACE */