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/>.
22 #include <ccan/likely/likely.h>
23 #include <ccan/endian/endian.h>
27 #include "system/filesys.h"
28 #include "system/time.h"
29 #include "system/shmem.h"
30 #include "system/select.h"
31 #include "system/wait.h"
51 /* #define TDB_TRACE 1 */
54 #define __STRING(x) #x
57 #ifndef __STRINGSTRING
58 #define __STRINGSTRING(x) __STRING(x)
62 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
65 typedef uint64_t tdb_len_t;
66 typedef uint64_t tdb_off_t;
68 #define TDB_MAGIC_FOOD "TDB file\n"
69 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
70 #define TDB_USED_MAGIC ((uint64_t)0x1999)
71 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
72 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
73 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
74 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
75 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
76 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
77 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
79 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
81 /* Packing errors into pointers and v.v. */
82 #define TDB_PTR_IS_ERR(ptr) \
83 unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST)
84 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
85 #define TDB_ERR_PTR(err) ((void *)(long)(err))
87 /* Common case of returning true, false or -ve error. */
88 typedef int tdb_bool_err;
90 /* Prevent others from opening the file. */
91 #define TDB_OPEN_LOCK 0
92 /* Doing a transaction. */
93 #define TDB_TRANSACTION_LOCK 1
95 #define TDB_EXPANSION_LOCK 2
96 /* Hash chain locks. */
97 #define TDB_HASH_LOCK_START 64
99 /* Range for hash locks. */
100 #define TDB_HASH_LOCK_RANGE_BITS 30
101 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
103 /* We have 1024 entries in the top level. */
104 #define TDB_TOPLEVEL_HASH_BITS 10
105 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
106 #define TDB_SUBLEVEL_HASH_BITS 6
107 /* And 8 entries in each group, ie 8 groups per sublevel. */
108 #define TDB_HASH_GROUP_BITS 3
109 /* This is currently 10: beyond this we chain. */
110 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
112 /* Extend file by least 100 times larger than needed. */
113 #define TDB_EXTENSION_FACTOR 100
115 /* We steal bits from the offsets to store hash info. */
116 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
117 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
118 #define TDB_OFF_UPPER_STEAL 8
119 #define TDB_OFF_UPPER_STEAL_EXTRA 7
120 /* The bit number where we store extra hash bits. */
121 #define TDB_OFF_HASH_EXTRA_BIT 57
122 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
124 /* Additional features we understand. Currently: none. */
125 #define TDB_FEATURE_MASK ((uint64_t)0)
127 /* The bit number where we store the extra hash bits. */
128 /* Convenience mask to get actual offset. */
129 #define TDB_OFF_MASK \
130 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
132 /* How many buckets in a free list: see size_to_bucket(). */
133 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
135 /* We have to be able to fit a free record here. */
136 #define TDB_MIN_DATA_LEN \
137 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
139 /* Indicates this entry is not on an flist (can happen during coalescing) */
140 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
142 struct tdb_used_record {
143 /* For on-disk compatibility, we avoid bitfields:
149 uint64_t magic_and_meta;
150 /* The bottom key_len_bits*2 are key length, rest is data length. */
151 uint64_t key_and_data_len;
154 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
156 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
159 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
161 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
164 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
166 return r->key_and_data_len >> rec_key_bits(r);
169 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
171 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
174 static inline uint32_t rec_hash(const struct tdb_used_record *r)
176 return r->magic_and_meta & ((1 << 11) - 1);
179 static inline uint16_t rec_magic(const struct tdb_used_record *r)
181 return (r->magic_and_meta >> 48);
184 struct tdb_free_record {
185 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
186 uint64_t ftable_and_len; /* Len not counting these two fields. */
187 /* This is why the minimum record size is 8 bytes. */
191 static inline uint64_t frec_prev(const struct tdb_free_record *f)
193 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
196 static inline uint64_t frec_magic(const struct tdb_free_record *f)
198 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
201 static inline uint64_t frec_len(const struct tdb_free_record *f)
203 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
206 static inline unsigned frec_ftable(const struct tdb_free_record *f)
208 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
211 struct tdb_recovery_record {
213 /* Length of record (add this header to get total length). */
217 /* Old length of file before transaction. */
221 /* If we bottom out of the subhashes, we chain. */
223 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
227 /* this is stored at the front of every database */
229 char magic_food[64]; /* for /etc/magic */
230 /* FIXME: Make me 32 bit? */
231 uint64_t version; /* version of the code */
232 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
233 uint64_t hash_seed; /* "random" seed written at creation time. */
234 tdb_off_t free_table; /* (First) free table. */
235 tdb_off_t recovery; /* Transaction recovery area. */
237 uint64_t features_used; /* Features all writers understand */
238 uint64_t features_offered; /* Features offered */
240 uint64_t seqnum; /* Sequence number for TDB_SEQNUM */
242 tdb_off_t reserved[23];
244 /* Top level hash table. */
245 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
248 struct tdb_freetable {
249 struct tdb_used_record hdr;
251 tdb_off_t buckets[TDB_FREE_BUCKETS];
254 /* Information about a particular (locked) hash entry. */
256 /* Full hash value of entry. */
258 /* Start and length of lock acquired. */
259 tdb_off_t hlock_start;
260 tdb_len_t hlock_range;
261 /* Start of hash group. */
262 tdb_off_t group_start;
263 /* Bucket we belong in. */
264 unsigned int home_bucket;
265 /* Bucket we (or an empty space) were found in. */
266 unsigned int found_bucket;
267 /* How many bits of the hash are already used. */
268 unsigned int hash_used;
269 /* Current working group. */
270 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
273 struct traverse_info {
274 struct traverse_level {
276 /* We ignore groups here, and treat it as a big array. */
278 unsigned int total_buckets;
279 } levels[TDB_MAX_LEVELS + 1];
280 unsigned int num_levels;
281 unsigned int toplevel_group;
282 /* This makes delete-everything-inside-traverse work as expected. */
286 enum tdb_lock_flags {
287 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
290 /* If set, don't log an error on failure. */
292 /* If set, don't check for recovery (used by recovery code). */
293 TDB_LOCK_NOCHECK = 4,
297 struct tdb_context *owner;
303 /* This is only needed for tdb_access_commit, but used everywhere to
305 struct tdb_access_hdr {
306 struct tdb_access_hdr *next;
313 /* Single list of all TDBs, to detect multiple opens. */
314 struct tdb_file *next;
316 /* How many are sharing us? */
319 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
322 /* How much space has been mapped (<= current file size) */
325 /* The file descriptor (-1 for TDB_INTERNAL). */
328 /* Lock information */
330 struct tdb_lock allrecord_lock;
332 struct tdb_lock *lockrecs;
334 /* Identity of this file. */
340 /* Filename of the database. */
343 /* Are we accessing directly? (debugging check). */
346 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
349 /* mmap read only? */
352 /* the flags passed to tdb_open, for tdb_reopen. */
355 /* Logging function */
356 void (*log_fn)(struct tdb_context *tdb,
357 enum tdb_log_level level,
363 uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *);
367 /* low level (fnctl) lock functions. */
368 int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *);
369 int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *);
372 /* Set if we are in a transaction. */
373 struct tdb_transaction *transaction;
375 /* What free table are we using? */
376 tdb_off_t ftable_off;
379 /* IO methods: changes for transactions. */
380 const struct tdb_methods *methods;
382 /* Our statistics. */
383 struct tdb_attribute_stats stats;
385 /* Direct access information */
386 struct tdb_access_hdr *access;
388 /* Last error we returned. */
389 enum TDB_ERROR last_error;
391 /* The actual file information */
392 struct tdb_file *file;
396 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
398 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
400 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
401 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
402 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
409 tdb_bool_err first_in_hash(struct tdb_context *tdb,
410 struct traverse_info *tinfo,
411 TDB_DATA *kbuf, size_t *dlen);
413 tdb_bool_err next_in_hash(struct tdb_context *tdb,
414 struct traverse_info *tinfo,
415 TDB_DATA *kbuf, size_t *dlen);
417 /* Hash random memory. */
418 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
421 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
423 /* Find and lock a hash entry (or where it would be). */
424 tdb_off_t find_and_lock(struct tdb_context *tdb,
428 struct tdb_used_record *rec,
429 struct traverse_info *tinfo);
431 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
435 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
438 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
441 bool is_subhash(tdb_off_t val);
444 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
446 /* check.c needs these to iterate through free lists. */
447 tdb_off_t first_ftable(struct tdb_context *tdb);
448 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
450 /* This returns space or -ve error number. */
451 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
452 uint64_t hash, unsigned magic, bool growing);
454 /* Put this record in a free list. */
455 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
456 tdb_off_t off, tdb_len_t len_with_header,
457 enum tdb_lock_flags waitflag,
460 /* Set up header for a used/ftable/htable/chain record. */
461 enum TDB_ERROR set_header(struct tdb_context *tdb,
462 struct tdb_used_record *rec,
463 unsigned magic, uint64_t keylen, uint64_t datalen,
464 uint64_t actuallen, unsigned hashlow);
466 /* Used by tdb_check to verify. */
467 unsigned int size_to_bucket(tdb_len_t data_len);
468 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
470 /* Used by tdb_summary */
471 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
474 /* Initialize tdb->methods. */
475 void tdb_io_init(struct tdb_context *tdb);
477 /* Convert endian of the buffer if required. */
478 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
480 /* Unmap and try to map the tdb. */
481 void tdb_munmap(struct tdb_file *file);
482 void tdb_mmap(struct tdb_context *tdb);
484 /* Either alloc a copy, or give direct access. Release frees or noop. */
485 const void *tdb_access_read(struct tdb_context *tdb,
486 tdb_off_t off, tdb_len_t len, bool convert);
487 void *tdb_access_write(struct tdb_context *tdb,
488 tdb_off_t off, tdb_len_t len, bool convert);
490 /* Release result of tdb_access_read/write. */
491 void tdb_access_release(struct tdb_context *tdb, const void *p);
492 /* Commit result of tdb_acces_write. */
493 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
495 /* Convenience routine to get an offset. */
496 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
498 /* Write an offset at an offset. */
499 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
502 /* Clear an ondisk area. */
503 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
505 /* Return a non-zero offset between >= start < end in this array (or end). */
506 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
511 /* Return a zero offset in this array, or num. */
512 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
515 /* Allocate and make a copy of some offset. */
516 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
518 /* Writes a converted copy of a record. */
519 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
520 const void *rec, size_t len);
522 /* Reads record and converts it */
523 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
524 void *rec, size_t len);
526 /* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */
527 void tdb_inc_seqnum(struct tdb_context *tdb);
530 /* Lock/unlock a range of hashes. */
531 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
532 tdb_off_t hash_lock, tdb_len_t hash_range,
533 int ltype, enum tdb_lock_flags waitflag);
534 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
536 tdb_len_t hash_range, int ltype);
538 /* For closing the file. */
539 void tdb_lock_cleanup(struct tdb_context *tdb);
541 /* Lock/unlock a particular free bucket. */
542 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
543 enum tdb_lock_flags waitflag);
544 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
546 /* Serialize transaction start. */
547 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
548 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
550 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
551 bool tdb_has_hash_locks(struct tdb_context *tdb);
553 /* Lock entire database. */
554 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
555 enum tdb_lock_flags flags, bool upgradable);
556 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
557 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
559 /* Serialize db open. */
560 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
561 int ltype, enum tdb_lock_flags flags);
562 void tdb_unlock_open(struct tdb_context *tdb, int ltype);
563 bool tdb_has_open_lock(struct tdb_context *tdb);
565 /* Serialize db expand. */
566 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
567 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
568 bool tdb_has_expansion_lock(struct tdb_context *tdb);
570 /* If it needs recovery, grab all the locks and do it. */
571 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
573 /* Default lock and unlock functions. */
574 int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *);
575 int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *);
578 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
579 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
582 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
583 enum TDB_ERROR ecode,
584 enum tdb_log_level level,
585 const char *fmt, ...);
588 void tdb_trace(struct tdb_context *tdb, const char *op);
589 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
590 void tdb_trace_open(struct tdb_context *tdb, const char *op,
591 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
592 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
593 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
594 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
596 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
597 TDB_DATA rec, int ret);
598 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
599 TDB_DATA rec, TDB_DATA ret);
600 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
601 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
603 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
604 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
606 #define tdb_trace(tdb, op)
607 #define tdb_trace_seqnum(tdb, seqnum, op)
608 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
609 #define tdb_trace_ret(tdb, op, ret)
610 #define tdb_trace_retrec(tdb, op, ret)
611 #define tdb_trace_1rec(tdb, op, rec)
612 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
613 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
614 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
615 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
616 #endif /* !TDB_TRACE */