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 #include <ccan/tdb2/tdb2.h>
32 #include <ccan/likely/likely.h>
33 #include <ccan/endian/endian.h>
39 /* #define TDB_TRACE 1 */
42 #define __STRING(x) #x
45 #ifndef __STRINGSTRING
46 #define __STRINGSTRING(x) __STRING(x)
50 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
53 typedef uint64_t tdb_len_t;
54 typedef uint64_t tdb_off_t;
56 #define TDB_MAGIC_FOOD "TDB file\n"
57 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
58 #define TDB_USED_MAGIC ((uint64_t)0x1999)
59 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
60 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
61 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
62 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
63 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
64 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
65 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
67 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
69 /* Packing errors into pointers and v.v. */
70 #define TDB_PTR_IS_ERR(ptr) \
71 unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST)
72 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
73 #define TDB_ERR_PTR(err) ((void *)(long)(err))
75 /* Common case of returning true, false or -ve error. */
76 typedef int tdb_bool_err;
78 /* Prevent others from opening the file. */
79 #define TDB_OPEN_LOCK 0
81 #define TDB_EXPANSION_LOCK 2
82 /* Doing a transaction. */
83 #define TDB_TRANSACTION_LOCK 8
84 /* Hash chain locks. */
85 #define TDB_HASH_LOCK_START 64
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 /* Additional features we understand. Currently: none. */
113 #define TDB_FEATURE_MASK ((uint64_t)0)
115 /* The bit number where we store the extra hash bits. */
116 /* Convenience mask to get actual offset. */
117 #define TDB_OFF_MASK \
118 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
120 /* How many buckets in a free list: see size_to_bucket(). */
121 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
123 /* We have to be able to fit a free record here. */
124 #define TDB_MIN_DATA_LEN \
125 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
127 /* Indicates this entry is not on an flist (can happen during coalescing) */
128 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
130 struct tdb_used_record {
131 /* For on-disk compatibility, we avoid bitfields:
137 uint64_t magic_and_meta;
138 /* The bottom key_len_bits*2 are key length, rest is data length. */
139 uint64_t key_and_data_len;
142 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
144 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
147 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
149 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
152 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
154 return r->key_and_data_len >> rec_key_bits(r);
157 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
159 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
162 static inline uint32_t rec_hash(const struct tdb_used_record *r)
164 return r->magic_and_meta & ((1 << 11) - 1);
167 static inline uint16_t rec_magic(const struct tdb_used_record *r)
169 return (r->magic_and_meta >> 48);
172 struct tdb_free_record {
173 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
174 uint64_t ftable_and_len; /* Len not counting these two fields. */
175 /* This is why the minimum record size is 8 bytes. */
179 static inline uint64_t frec_prev(const struct tdb_free_record *f)
181 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
184 static inline uint64_t frec_magic(const struct tdb_free_record *f)
186 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
189 static inline uint64_t frec_len(const struct tdb_free_record *f)
191 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
194 static inline unsigned frec_ftable(const struct tdb_free_record *f)
196 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
199 struct tdb_recovery_record {
201 /* Length of record (add this header to get total length). */
205 /* Old length of file before transaction. */
209 /* If we bottom out of the subhashes, we chain. */
211 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
215 /* this is stored at the front of every database */
217 char magic_food[64]; /* for /etc/magic */
218 /* FIXME: Make me 32 bit? */
219 uint64_t version; /* version of the code */
220 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
221 uint64_t hash_seed; /* "random" seed written at creation time. */
222 tdb_off_t free_table; /* (First) free table. */
223 tdb_off_t recovery; /* Transaction recovery area. */
225 uint64_t features_used; /* Features all writers understand */
226 uint64_t features_offered; /* Features offered */
228 uint64_t seqnum; /* Sequence number for TDB_SEQNUM */
230 tdb_off_t reserved[23];
232 /* Top level hash table. */
233 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
236 struct tdb_freetable {
237 struct tdb_used_record hdr;
239 tdb_off_t buckets[TDB_FREE_BUCKETS];
242 /* Information about a particular (locked) hash entry. */
244 /* Full hash value of entry. */
246 /* Start and length of lock acquired. */
247 tdb_off_t hlock_start;
248 tdb_len_t hlock_range;
249 /* Start of hash group. */
250 tdb_off_t group_start;
251 /* Bucket we belong in. */
252 unsigned int home_bucket;
253 /* Bucket we (or an empty space) were found in. */
254 unsigned int found_bucket;
255 /* How many bits of the hash are already used. */
256 unsigned int hash_used;
257 /* Current working group. */
258 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
261 struct traverse_info {
262 struct traverse_level {
264 /* We ignore groups here, and treat it as a big array. */
266 unsigned int total_buckets;
267 } levels[TDB_MAX_LEVELS + 1];
268 unsigned int num_levels;
269 unsigned int toplevel_group;
270 /* This makes delete-everything-inside-traverse work as expected. */
274 enum tdb_lock_flags {
275 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
278 /* If set, don't log an error on failure. */
280 /* If set, don't check for recovery (used by recovery code). */
281 TDB_LOCK_NOCHECK = 4,
285 struct tdb_context *owner;
291 /* This is only needed for tdb_access_commit, but used everywhere to
293 struct tdb_access_hdr {
294 struct tdb_access_hdr *next;
301 /* How many are sharing us? */
304 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
307 /* How much space has been mapped (<= current file size) */
310 /* The file descriptor (-1 for TDB_INTERNAL). */
313 /* Lock information */
315 struct tdb_lock allrecord_lock;
317 struct tdb_lock *lockrecs;
319 /* Identity of this file. */
325 /* Single list of all TDBs, to detect multiple opens. */
326 struct tdb_context *next;
328 /* Filename of the database. */
331 /* Logging function */
332 void (*log_fn)(struct tdb_context *tdb,
333 enum tdb_log_level level,
334 enum TDB_ERROR ecode,
339 /* Last error we returned. */
340 enum TDB_ERROR last_error;
342 /* The actual file information */
343 struct tdb_file *file;
345 /* Open flags passed to tdb_open. */
348 /* low level (fnctl) lock functions. */
349 int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *);
350 int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *);
353 /* the flags passed to tdb_open. */
356 /* Our statistics. */
357 struct tdb_attribute_stats stats;
360 uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *);
364 /* Are we accessing directly? (debugging check). */
367 /* Set if we are in a transaction. */
368 struct tdb_transaction *transaction;
370 /* What free table are we using? */
371 tdb_off_t ftable_off;
374 /* Our open hook, if any. */
375 enum TDB_ERROR (*openhook)(int fd, void *data);
378 /* IO methods: changes for transactions. */
379 const struct tdb_methods *methods;
381 /* Direct access information */
382 struct tdb_access_hdr *access;
387 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
389 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
391 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
392 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
393 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
400 tdb_bool_err first_in_hash(struct tdb_context *tdb,
401 struct traverse_info *tinfo,
402 TDB_DATA *kbuf, size_t *dlen);
404 tdb_bool_err next_in_hash(struct tdb_context *tdb,
405 struct traverse_info *tinfo,
406 TDB_DATA *kbuf, size_t *dlen);
408 /* Hash random memory. */
409 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
412 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
414 /* Find and lock a hash entry (or where it would be). */
415 tdb_off_t find_and_lock(struct tdb_context *tdb,
419 struct tdb_used_record *rec,
420 struct traverse_info *tinfo);
422 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
426 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
429 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
432 bool is_subhash(tdb_off_t val);
435 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
437 /* check.c needs these to iterate through free lists. */
438 tdb_off_t first_ftable(struct tdb_context *tdb);
439 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
441 /* This returns space or -ve error number. */
442 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
443 uint64_t hash, unsigned magic, bool growing);
445 /* Put this record in a free list. */
446 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
447 tdb_off_t off, tdb_len_t len_with_header,
448 enum tdb_lock_flags waitflag,
451 /* Set up header for a used/ftable/htable/chain record. */
452 enum TDB_ERROR set_header(struct tdb_context *tdb,
453 struct tdb_used_record *rec,
454 unsigned magic, uint64_t keylen, uint64_t datalen,
455 uint64_t actuallen, unsigned hashlow);
457 /* Used by tdb_check to verify. */
458 unsigned int size_to_bucket(tdb_len_t data_len);
459 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
461 /* Used by tdb_summary */
462 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
465 /* Initialize tdb->methods. */
466 void tdb_io_init(struct tdb_context *tdb);
468 /* Convert endian of the buffer if required. */
469 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
471 /* Unmap and try to map the tdb. */
472 void tdb_munmap(struct tdb_file *file);
473 void tdb_mmap(struct tdb_context *tdb);
475 /* Either alloc a copy, or give direct access. Release frees or noop. */
476 const void *tdb_access_read(struct tdb_context *tdb,
477 tdb_off_t off, tdb_len_t len, bool convert);
478 void *tdb_access_write(struct tdb_context *tdb,
479 tdb_off_t off, tdb_len_t len, bool convert);
481 /* Release result of tdb_access_read/write. */
482 void tdb_access_release(struct tdb_context *tdb, const void *p);
483 /* Commit result of tdb_acces_write. */
484 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
486 /* Convenience routine to get an offset. */
487 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
489 /* Write an offset at an offset. */
490 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
493 /* Clear an ondisk area. */
494 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
496 /* Return a non-zero offset between >= start < end in this array (or end). */
497 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
502 /* Return a zero offset in this array, or num. */
503 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
506 /* Allocate and make a copy of some offset. */
507 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
509 /* Writes a converted copy of a record. */
510 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
511 const void *rec, size_t len);
513 /* Reads record and converts it */
514 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
515 void *rec, size_t len);
517 /* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */
518 void tdb_inc_seqnum(struct tdb_context *tdb);
521 /* Lock/unlock a range of hashes. */
522 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
523 tdb_off_t hash_lock, tdb_len_t hash_range,
524 int ltype, enum tdb_lock_flags waitflag);
525 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
527 tdb_len_t hash_range, int ltype);
529 /* For closing the file. */
530 void tdb_lock_cleanup(struct tdb_context *tdb);
532 /* Lock/unlock a particular free bucket. */
533 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
534 enum tdb_lock_flags waitflag);
535 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
537 /* Serialize transaction start. */
538 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
539 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
541 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
542 bool tdb_has_hash_locks(struct tdb_context *tdb);
544 /* Lock entire database. */
545 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
546 enum tdb_lock_flags flags, bool upgradable);
547 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
548 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb, off_t start);
550 /* Serialize db open. */
551 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
552 int ltype, enum tdb_lock_flags flags);
553 void tdb_unlock_open(struct tdb_context *tdb, int ltype);
554 bool tdb_has_open_lock(struct tdb_context *tdb);
556 /* Serialize db expand. */
557 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
558 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
559 bool tdb_has_expansion_lock(struct tdb_context *tdb);
561 /* If it needs recovery, grab all the locks and do it. */
562 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
564 /* Byte-range lock wrappers for TDB1 to access. */
565 enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
566 int rw_type, tdb_off_t offset, tdb_off_t len,
567 enum tdb_lock_flags flags);
569 enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
570 int rw_type, tdb_off_t offset, size_t len);
572 enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
573 tdb_off_t offset, int ltype,
574 enum tdb_lock_flags flags);
576 enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
577 tdb_off_t off, int ltype);
579 enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
580 int ltype, enum tdb_lock_flags flags,
581 tdb_off_t off, tdb_off_t len);
583 /* Default lock and unlock functions. */
584 int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *);
585 int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *);
588 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
589 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
592 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
593 enum TDB_ERROR ecode,
594 enum tdb_log_level level,
595 const char *fmt, ...);
598 void tdb_trace(struct tdb_context *tdb, const char *op);
599 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
600 void tdb_trace_open(struct tdb_context *tdb, const char *op,
601 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
602 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
603 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
604 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
606 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
607 TDB_DATA rec, int ret);
608 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
609 TDB_DATA rec, TDB_DATA ret);
610 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
611 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
613 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
614 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
616 #define tdb_trace(tdb, op)
617 #define tdb_trace_seqnum(tdb, seqnum, op)
618 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
619 #define tdb_trace_ret(tdb, op, ret)
620 #define tdb_trace_retrec(tdb, op, ret)
621 #define tdb_trace_1rec(tdb, op, rec)
622 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
623 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
624 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
625 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
626 #endif /* !TDB_TRACE */