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 #ifdef HAVE_BYTESWAP_H
47 /* #define TDB_TRACE 1 */
50 #define __STRING(x) #x
53 #ifndef __STRINGSTRING
54 #define __STRINGSTRING(x) __STRING(x)
58 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
61 typedef uint64_t tdb_len_t;
62 typedef uint64_t tdb_off_t;
64 #define TDB_MAGIC_FOOD "TDB file\n"
65 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
66 #define TDB_MAGIC ((uint64_t)0x1999)
67 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
68 #define TDB_COALESCING_MAGIC ((uint64_t)0xFD)
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 #define TDB_OFF_UPPER_STEAL_TRUNCBIT 1
104 /* If this is set, hash is truncated (only 1 bit is valid). */
105 #define TDB_OFF_HASH_TRUNCATED_BIT 56
106 /* The bit number where we store next level of hash. */
107 #define TDB_OFF_HASH_EXTRA_BIT 57
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))
120 static inline uint64_t bswap_64(uint64_t x)
122 return (((x&0x000000FFULL)<<56)
123 | ((x&0x0000FF00ULL)<<48)
124 | ((x&0x00FF0000ULL)<<40)
125 | ((x&0xFF000000ULL)<<32)
126 | ((x>>8)&0xFF000000ULL)
127 | ((x>>16)&0x00FF0000ULL)
128 | ((x>>24)&0x0000FF00ULL)
129 | ((x>>32)&0x000000FFULL));
133 struct tdb_used_record {
134 /* For on-disk compatibility, we avoid bitfields:
140 uint64_t magic_and_meta;
141 /* The bottom key_len_bits*2 are key length, rest is data length. */
142 uint64_t key_and_data_len;
145 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
147 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
150 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
152 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
155 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
157 return r->key_and_data_len >> rec_key_bits(r);
160 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
162 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
165 static inline uint32_t rec_hash(const struct tdb_used_record *r)
167 return r->magic_and_meta & ((1 << 11) - 1);
170 static inline uint16_t rec_magic(const struct tdb_used_record *r)
172 return (r->magic_and_meta >> 48);
175 struct tdb_free_record {
176 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
177 uint64_t flist_and_len; /* Len not counting these two fields. */
178 /* This is why the minimum record size is 8 bytes. */
182 static inline uint64_t frec_prev(const struct tdb_free_record *f)
184 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
187 static inline uint64_t frec_magic(const struct tdb_free_record *f)
189 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
192 static inline uint64_t frec_len(const struct tdb_free_record *f)
194 return f->flist_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
197 static inline unsigned frec_flist(const struct tdb_free_record *f)
199 return f->flist_and_len >> (64 - TDB_OFF_UPPER_STEAL);
202 struct tdb_recovery_record {
204 /* Length of record. */
208 /* Old length of file before transaction. */
212 /* this is stored at the front of every database */
214 char magic_food[64]; /* for /etc/magic */
215 /* FIXME: Make me 32 bit? */
216 uint64_t version; /* version of the code */
217 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
218 uint64_t hash_seed; /* "random" seed written at creation time. */
219 tdb_off_t free_list; /* (First) free list. */
220 tdb_off_t recovery; /* Transaction recovery area. */
222 tdb_off_t reserved[26];
224 /* Top level hash table. */
225 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
228 struct tdb_freelist {
229 struct tdb_used_record hdr;
231 tdb_off_t buckets[TDB_FREE_BUCKETS];
234 /* Information about a particular (locked) hash entry. */
236 /* Full hash value of entry. */
238 /* Start and length of lock acquired. */
239 tdb_off_t hlock_start;
240 tdb_len_t hlock_range;
241 /* Start of hash group. */
242 tdb_off_t group_start;
243 /* Bucket we belong in. */
244 unsigned int home_bucket;
245 /* Bucket we (or an empty space) were found in. */
246 unsigned int found_bucket;
247 /* How many bits of the hash are already used. */
248 unsigned int hash_used;
249 /* Current working group. */
250 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
253 struct traverse_info {
254 struct traverse_level {
256 /* We ignore groups here, and treat it as a big array. */
258 unsigned int total_buckets;
259 } levels[64 / TDB_SUBLEVEL_HASH_BITS];
260 unsigned int num_levels;
261 unsigned int toplevel_group;
262 /* This makes delete-everything-inside-traverse work as expected. */
266 enum tdb_lock_flags {
267 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
270 /* If set, don't log an error on failure. */
272 /* If set, don't check for recovery (used by recovery code). */
273 TDB_LOCK_NOCHECK = 4,
276 struct tdb_lock_type {
283 /* Filename of the database. */
286 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
289 /* Are we accessing directly? (debugging check). */
292 /* Open file descriptor (undefined for TDB_INTERNAL). */
295 /* How much space has been mapped (<= current file size) */
298 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
301 /* mmap read only? */
304 /* Error code for last tdb error. */
305 enum TDB_ERROR ecode;
307 /* the flags passed to tdb_open, for tdb_reopen. */
310 /* Logging function */
319 /* Set if we are in a transaction. */
320 struct tdb_transaction *transaction;
322 /* What freelist are we using? */
326 /* IO methods: changes for transactions. */
327 const struct tdb_methods *methods;
329 /* Lock information */
330 struct tdb_lock_type allrecord_lock;
331 uint64_t num_lockrecs;
332 struct tdb_lock_type *lockrecs;
334 struct tdb_attribute_stats *stats;
336 /* Single list of all TDBs, to avoid multiple opens. */
337 struct tdb_context *next;
343 int (*read)(struct tdb_context *, tdb_off_t, void *, tdb_len_t);
344 int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
345 int (*oob)(struct tdb_context *, tdb_off_t, bool);
346 int (*expand_file)(struct tdb_context *, tdb_len_t);
347 void *(*direct)(struct tdb_context *, tdb_off_t, size_t);
354 void tdb_hash_init(struct tdb_context *tdb);
356 /* Hash random memory. */
357 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
360 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
362 /* Find and lock a hash entry (or where it would be). */
363 tdb_off_t find_and_lock(struct tdb_context *tdb,
367 struct tdb_used_record *rec,
368 struct traverse_info *tinfo);
370 int replace_in_hash(struct tdb_context *tdb,
374 int add_to_hash(struct tdb_context *tdb, struct hash_info *h,
377 int delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
380 bool is_subhash(tdb_off_t val);
383 int tdb_flist_init(struct tdb_context *tdb);
385 /* check.c needs these to iterate through free lists. */
386 tdb_off_t first_flist(struct tdb_context *tdb);
387 tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist);
389 /* If this fails, try tdb_expand. */
390 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
391 uint64_t hash, bool growing);
393 /* Put this record in a free list. */
394 int add_free_record(struct tdb_context *tdb,
395 tdb_off_t off, tdb_len_t len_with_header);
397 /* Set up header for a used record. */
398 int set_used_header(struct tdb_context *tdb,
399 struct tdb_used_record *rec,
400 uint64_t keylen, uint64_t datalen,
401 uint64_t actuallen, unsigned hashlow);
403 /* Used by tdb_check to verify. */
404 unsigned int size_to_bucket(tdb_len_t data_len);
405 tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket);
407 /* Used by tdb_summary */
408 size_t dead_space(struct tdb_context *tdb, tdb_off_t off);
411 /* Initialize tdb->methods. */
412 void tdb_io_init(struct tdb_context *tdb);
414 /* Convert endian of the buffer if required. */
415 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
417 /* Unmap and try to map the tdb. */
418 void tdb_munmap(struct tdb_context *tdb);
419 void tdb_mmap(struct tdb_context *tdb);
421 /* Either make a copy into pad and return that, or return ptr into mmap.
422 * Converts endian (ie. will use pad in that case). */
423 void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len);
425 /* Either alloc a copy, or give direct access. Release frees or noop. */
426 const void *tdb_access_read(struct tdb_context *tdb,
427 tdb_off_t off, tdb_len_t len, bool convert);
428 void *tdb_access_write(struct tdb_context *tdb,
429 tdb_off_t off, tdb_len_t len, bool convert);
431 /* Release result of tdb_access_read/write. */
432 void tdb_access_release(struct tdb_context *tdb, const void *p);
433 /* Commit result of tdb_acces_write. */
434 int tdb_access_commit(struct tdb_context *tdb, void *p);
436 /* Convenience routine to get an offset. */
437 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
439 /* Write an offset at an offset. */
440 int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val);
442 /* Clear an ondisk area. */
443 int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
445 /* Return a non-zero offset between >= start < end in this array (or end). */
446 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
451 /* Return a zero offset in this array, or num. */
452 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
455 /* Even on files, we can get partial writes due to signals. */
456 bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off);
457 bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off);
458 bool tdb_read_all(int fd, void *buf, size_t len);
460 /* Allocate and make a copy of some offset. */
461 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
463 /* Writes a converted copy of a record. */
464 int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
465 const void *rec, size_t len);
467 /* Reads record and converts it */
468 int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
469 void *rec, size_t len);
471 /* Adds a stat, if it's in range. */
472 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
473 #define add_stat(tdb, statname, val) \
475 if (unlikely((tdb)->stats)) \
476 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
480 void tdb_lock_init(struct tdb_context *tdb);
482 /* Lock/unlock a range of hashes. */
483 int tdb_lock_hashes(struct tdb_context *tdb,
484 tdb_off_t hash_lock, tdb_len_t hash_range,
485 int ltype, enum tdb_lock_flags waitflag);
486 int tdb_unlock_hashes(struct tdb_context *tdb,
488 tdb_len_t hash_range, int ltype);
490 /* Lock/unlock a particular free bucket. */
491 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
492 enum tdb_lock_flags waitflag);
493 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
495 /* Serialize transaction start. */
496 int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
497 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
499 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
500 bool tdb_has_hash_locks(struct tdb_context *tdb);
502 /* Lock entire database. */
503 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
504 enum tdb_lock_flags flags, bool upgradable);
505 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
506 int tdb_allrecord_upgrade(struct tdb_context *tdb);
508 /* Serialize db open. */
509 int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
510 void tdb_unlock_open(struct tdb_context *tdb);
511 bool tdb_has_open_lock(struct tdb_context *tdb);
513 /* Serialize db expand. */
514 int tdb_lock_expand(struct tdb_context *tdb, int ltype);
515 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
516 bool tdb_has_expansion_lock(struct tdb_context *tdb);
518 /* If it needs recovery, grab all the locks and do it. */
519 int tdb_lock_and_recover(struct tdb_context *tdb);
522 int first_in_hash(struct tdb_context *tdb, int ltype,
523 struct traverse_info *tinfo,
524 TDB_DATA *kbuf, size_t *dlen);
525 int next_in_hash(struct tdb_context *tdb, int ltype,
526 struct traverse_info *tinfo,
527 TDB_DATA *kbuf, size_t *dlen);
530 int tdb_transaction_recover(struct tdb_context *tdb);
531 bool tdb_needs_recovery(struct tdb_context *tdb);
534 void tdb_trace(struct tdb_context *tdb, const char *op);
535 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
536 void tdb_trace_open(struct tdb_context *tdb, const char *op,
537 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
538 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
539 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
540 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
542 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
543 TDB_DATA rec, int ret);
544 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
545 TDB_DATA rec, TDB_DATA ret);
546 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
547 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
549 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
550 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
552 #define tdb_trace(tdb, op)
553 #define tdb_trace_seqnum(tdb, seqnum, op)
554 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
555 #define tdb_trace_ret(tdb, op, ret)
556 #define tdb_trace_retrec(tdb, op, ret)
557 #define tdb_trace_1rec(tdb, op, rec)
558 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
559 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
560 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
561 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
562 #endif /* !TDB_TRACE */