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_MAGIC ((uint64_t)0x1999)
68 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
69 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
70 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
71 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
73 #define TDB_OFF_ERR ((tdb_off_t)-1)
75 /* Prevent others from opening the file. */
76 #define TDB_OPEN_LOCK 0
77 /* Doing a transaction. */
78 #define TDB_TRANSACTION_LOCK 1
80 #define TDB_EXPANSION_LOCK 2
81 /* Hash chain locks. */
82 #define TDB_HASH_LOCK_START 3
84 /* Range for hash locks. */
85 #define TDB_HASH_LOCK_RANGE_BITS 30
86 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
88 /* We have 1024 entries in the top level. */
89 #define TDB_TOPLEVEL_HASH_BITS 10
90 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
91 #define TDB_SUBLEVEL_HASH_BITS 6
92 /* And 8 entries in each group, ie 8 groups per sublevel. */
93 #define TDB_HASH_GROUP_BITS 3
95 /* Extend file by least 100 times larger than needed. */
96 #define TDB_EXTENSION_FACTOR 100
98 /* We steal bits from the offsets to store hash info. */
99 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
100 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
101 #define TDB_OFF_UPPER_STEAL 8
102 #define TDB_OFF_UPPER_STEAL_EXTRA 7
103 /* The bit number where we store extra hash bits. */
104 #define TDB_OFF_HASH_EXTRA_BIT 57
105 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
107 /* The bit number where we store the extra hash bits. */
108 /* Convenience mask to get actual offset. */
109 #define TDB_OFF_MASK \
110 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
112 /* How many buckets in a free list: see size_to_bucket(). */
113 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
115 /* We have to be able to fit a free record here. */
116 #define TDB_MIN_DATA_LEN \
117 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
119 /* Indicates this entry is not on an flist (can happen during coalescing) */
120 #define TDB_FLIST_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
123 static inline uint64_t bswap_64(uint64_t x)
125 return (((x&0x000000FFULL)<<56)
126 | ((x&0x0000FF00ULL)<<48)
127 | ((x&0x00FF0000ULL)<<40)
128 | ((x&0xFF000000ULL)<<32)
129 | ((x>>8)&0xFF000000ULL)
130 | ((x>>16)&0x00FF0000ULL)
131 | ((x>>24)&0x0000FF00ULL)
132 | ((x>>32)&0x000000FFULL));
136 struct tdb_used_record {
137 /* For on-disk compatibility, we avoid bitfields:
143 uint64_t magic_and_meta;
144 /* The bottom key_len_bits*2 are key length, rest is data length. */
145 uint64_t key_and_data_len;
148 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
150 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
153 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
155 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
158 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
160 return r->key_and_data_len >> rec_key_bits(r);
163 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
165 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
168 static inline uint32_t rec_hash(const struct tdb_used_record *r)
170 return r->magic_and_meta & ((1 << 11) - 1);
173 static inline uint16_t rec_magic(const struct tdb_used_record *r)
175 return (r->magic_and_meta >> 48);
178 struct tdb_free_record {
179 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
180 uint64_t flist_and_len; /* Len not counting these two fields. */
181 /* This is why the minimum record size is 8 bytes. */
185 static inline uint64_t frec_prev(const struct tdb_free_record *f)
187 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
190 static inline uint64_t frec_magic(const struct tdb_free_record *f)
192 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
195 static inline uint64_t frec_len(const struct tdb_free_record *f)
197 return f->flist_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
200 static inline unsigned frec_flist(const struct tdb_free_record *f)
202 return f->flist_and_len >> (64 - TDB_OFF_UPPER_STEAL);
205 struct tdb_recovery_record {
207 /* Length of record. */
211 /* Old length of file before transaction. */
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_list; /* (First) free list. */
223 tdb_off_t recovery; /* Transaction recovery area. */
225 tdb_off_t reserved[26];
227 /* Top level hash table. */
228 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
231 struct tdb_freelist {
232 struct tdb_used_record hdr;
234 tdb_off_t buckets[TDB_FREE_BUCKETS];
237 /* Information about a particular (locked) hash entry. */
239 /* Full hash value of entry. */
241 /* Start and length of lock acquired. */
242 tdb_off_t hlock_start;
243 tdb_len_t hlock_range;
244 /* Start of hash group. */
245 tdb_off_t group_start;
246 /* Bucket we belong in. */
247 unsigned int home_bucket;
248 /* Bucket we (or an empty space) were found in. */
249 unsigned int found_bucket;
250 /* How many bits of the hash are already used. */
251 unsigned int hash_used;
252 /* Current working group. */
253 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
256 struct traverse_info {
257 struct traverse_level {
259 /* We ignore groups here, and treat it as a big array. */
261 unsigned int total_buckets;
262 } levels[64 / TDB_SUBLEVEL_HASH_BITS];
263 unsigned int num_levels;
264 unsigned int toplevel_group;
265 /* This makes delete-everything-inside-traverse work as expected. */
269 enum tdb_lock_flags {
270 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
273 /* If set, don't log an error on failure. */
275 /* If set, don't check for recovery (used by recovery code). */
276 TDB_LOCK_NOCHECK = 4,
279 struct tdb_lock_type {
286 /* Filename of the database. */
289 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
292 /* Are we accessing directly? (debugging check). */
295 /* Open file descriptor (undefined for TDB_INTERNAL). */
298 /* How much space has been mapped (<= current file size) */
301 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
304 /* mmap read only? */
307 /* Error code for last tdb error. */
308 enum TDB_ERROR ecode;
310 /* the flags passed to tdb_open, for tdb_reopen. */
313 /* Logging function */
322 /* Set if we are in a transaction. */
323 struct tdb_transaction *transaction;
325 /* What freelist are we using? */
329 /* IO methods: changes for transactions. */
330 const struct tdb_methods *methods;
332 /* Lock information */
333 struct tdb_lock_type allrecord_lock;
335 struct tdb_lock_type *lockrecs;
337 struct tdb_attribute_stats *stats;
339 /* Single list of all TDBs, to avoid multiple opens. */
340 struct tdb_context *next;
346 int (*read)(struct tdb_context *, tdb_off_t, void *, tdb_len_t);
347 int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
348 int (*oob)(struct tdb_context *, tdb_off_t, bool);
349 int (*expand_file)(struct tdb_context *, tdb_len_t);
350 void *(*direct)(struct tdb_context *, tdb_off_t, size_t);
357 void tdb_hash_init(struct tdb_context *tdb);
359 /* Hash random memory. */
360 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
363 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
365 /* Find and lock a hash entry (or where it would be). */
366 tdb_off_t find_and_lock(struct tdb_context *tdb,
370 struct tdb_used_record *rec,
371 struct traverse_info *tinfo);
373 int replace_in_hash(struct tdb_context *tdb,
377 int add_to_hash(struct tdb_context *tdb, struct hash_info *h,
380 int delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
383 bool is_subhash(tdb_off_t val);
386 int tdb_flist_init(struct tdb_context *tdb);
388 /* check.c needs these to iterate through free lists. */
389 tdb_off_t first_flist(struct tdb_context *tdb);
390 tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist);
392 /* If this fails, try tdb_expand. */
393 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
394 uint64_t hash, bool growing);
396 /* Put this record in a free list. */
397 int add_free_record(struct tdb_context *tdb,
398 tdb_off_t off, tdb_len_t len_with_header);
400 /* Set up header for a used record. */
401 int set_used_header(struct tdb_context *tdb,
402 struct tdb_used_record *rec,
403 uint64_t keylen, uint64_t datalen,
404 uint64_t actuallen, unsigned hashlow);
406 /* Used by tdb_check to verify. */
407 unsigned int size_to_bucket(tdb_len_t data_len);
408 tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket);
410 /* Used by tdb_summary */
411 size_t dead_space(struct tdb_context *tdb, tdb_off_t off);
414 /* Initialize tdb->methods. */
415 void tdb_io_init(struct tdb_context *tdb);
417 /* Convert endian of the buffer if required. */
418 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
420 /* Unmap and try to map the tdb. */
421 void tdb_munmap(struct tdb_context *tdb);
422 void tdb_mmap(struct tdb_context *tdb);
424 /* Either alloc a copy, or give direct access. Release frees or noop. */
425 const void *tdb_access_read(struct tdb_context *tdb,
426 tdb_off_t off, tdb_len_t len, bool convert);
427 void *tdb_access_write(struct tdb_context *tdb,
428 tdb_off_t off, tdb_len_t len, bool convert);
430 /* Is this pointer direct? (Otherwise it's malloced) */
431 bool is_direct(const struct tdb_context *tdb, const void *p);
433 /* Release result of tdb_access_read/write. */
434 void tdb_access_release(struct tdb_context *tdb, const void *p);
435 /* Commit result of tdb_acces_write. */
436 int tdb_access_commit(struct tdb_context *tdb, void *p);
438 /* Convenience routine to get an offset. */
439 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
441 /* Write an offset at an offset. */
442 int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val);
444 /* Clear an ondisk area. */
445 int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
447 /* Return a non-zero offset between >= start < end in this array (or end). */
448 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
453 /* Return a zero offset in this array, or num. */
454 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
457 /* Even on files, we can get partial writes due to signals. */
458 bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off);
459 bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off);
460 bool tdb_read_all(int fd, void *buf, size_t len);
462 /* Allocate and make a copy of some offset. */
463 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
465 /* Writes a converted copy of a record. */
466 int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
467 const void *rec, size_t len);
469 /* Reads record and converts it */
470 int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
471 void *rec, size_t len);
473 /* Adds a stat, if it's in range. */
474 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
475 #define add_stat(tdb, statname, val) \
477 if (unlikely((tdb)->stats)) \
478 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
482 void tdb_lock_init(struct tdb_context *tdb);
484 /* Lock/unlock a range of hashes. */
485 int tdb_lock_hashes(struct tdb_context *tdb,
486 tdb_off_t hash_lock, tdb_len_t hash_range,
487 int ltype, enum tdb_lock_flags waitflag);
488 int tdb_unlock_hashes(struct tdb_context *tdb,
490 tdb_len_t hash_range, int ltype);
492 /* Lock/unlock a particular free bucket. */
493 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
494 enum tdb_lock_flags waitflag);
495 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
497 /* Serialize transaction start. */
498 int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
499 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
501 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
502 bool tdb_has_hash_locks(struct tdb_context *tdb);
504 /* Lock entire database. */
505 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
506 enum tdb_lock_flags flags, bool upgradable);
507 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
508 int tdb_allrecord_upgrade(struct tdb_context *tdb);
510 /* Serialize db open. */
511 int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
512 void tdb_unlock_open(struct tdb_context *tdb);
513 bool tdb_has_open_lock(struct tdb_context *tdb);
515 /* Serialize db expand. */
516 int tdb_lock_expand(struct tdb_context *tdb, int ltype);
517 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
518 bool tdb_has_expansion_lock(struct tdb_context *tdb);
520 /* If it needs recovery, grab all the locks and do it. */
521 int tdb_lock_and_recover(struct tdb_context *tdb);
524 int first_in_hash(struct tdb_context *tdb, int ltype,
525 struct traverse_info *tinfo,
526 TDB_DATA *kbuf, size_t *dlen);
527 int next_in_hash(struct tdb_context *tdb, int ltype,
528 struct traverse_info *tinfo,
529 TDB_DATA *kbuf, size_t *dlen);
532 int tdb_transaction_recover(struct tdb_context *tdb);
533 bool tdb_needs_recovery(struct tdb_context *tdb);
536 void COLD tdb_logerr(struct tdb_context *tdb,
537 enum TDB_ERROR ecode,
538 enum tdb_debug_level level,
539 const char *fmt, ...);
542 void tdb_trace(struct tdb_context *tdb, const char *op);
543 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
544 void tdb_trace_open(struct tdb_context *tdb, const char *op,
545 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
546 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
547 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
548 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
550 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
551 TDB_DATA rec, int ret);
552 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
553 TDB_DATA rec, TDB_DATA ret);
554 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
555 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
557 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
558 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
560 #define tdb_trace(tdb, op)
561 #define tdb_trace_seqnum(tdb, seqnum, op)
562 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
563 #define tdb_trace_ret(tdb, op, ret)
564 #define tdb_trace_retrec(tdb, op, ret)
565 #define tdb_trace_1rec(tdb, op, rec)
566 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
567 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
568 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
569 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
570 #endif /* !TDB_TRACE */