#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
+#include <stddef.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <unistd.h>
typedef uint64_t tdb_len_t;
typedef uint64_t tdb_off_t;
-#ifndef offsetof
-#define offsetof(t,f) ((unsigned int)&((t *)0)->f)
-#endif
-
#define TDB_MAGIC_FOOD "TDB file\n"
#define TDB_VERSION ((uint64_t)(0x26011967 + 7))
#define TDB_MAGIC ((uint64_t)0x1999)
-#define TDB_FREE_MAGIC ((~(uint64_t)TDB_MAGIC) << 6)
+#define TDB_FREE_MAGIC ((uint64_t)0xFE)
+#define TDB_COALESCING_MAGIC ((uint64_t)0xFD)
#define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
-#define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
-#define TDB_RECOVERY_INVALID_MAGIC (0x0)
+#define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
+#define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
#define TDB_OFF_ERR ((tdb_off_t)-1)
/* And 8 entries in each group, ie 8 groups per sublevel. */
#define TDB_HASH_GROUP_BITS 3
-/* We start with a 64k-sized zone. */
-#define INITIAL_ZONE_BITS 16
-/* Try to create zones at least 32 times larger than allocations. */
-#define TDB_COMFORT_FACTOR_BITS 5
+/* Extend file by least 32 times larger than needed. */
+#define TDB_EXTENSION_FACTOR 32
/* We steal bits from the offsets to store hash info. */
#define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
#define TDB_OFF_MASK \
(((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
-/* We ensure buckets up to size 1 << (zone_bits - TDB_COMFORT_FACTOR_BITS). */
-/* FIXME: test this matches size_to_bucket! */
-#define BUCKETS_FOR_ZONE(zone_bits) ((zone_bits) + 2 - TDB_COMFORT_FACTOR_BITS)
+/* How many buckets in a free list: see size_to_bucket(). */
+#define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
+
+/* We have to be able to fit a free record here. */
+#define TDB_MIN_DATA_LEN \
+ (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
#if !HAVE_BSWAP_64
static inline uint64_t bswap_64(uint64_t x)
magic: 16, (highest)
key_len_bits: 5,
extra_padding: 32
- hash_bits: 5,
- zone_bits: 6 (lowest)
+ hash_bits: 11
*/
uint64_t magic_and_meta;
/* The bottom key_len_bits*2 are key length, rest is data length. */
return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
}
-static inline unsigned int rec_zone_bits(const struct tdb_used_record *r)
-{
- return r->magic_and_meta & ((1 << 6) - 1);
-}
-
static inline uint32_t rec_hash(const struct tdb_used_record *r)
{
- return (r->magic_and_meta >> 6) & ((1 << 5) - 1);
+ return r->magic_and_meta & ((1 << 11) - 1);
}
static inline uint16_t rec_magic(const struct tdb_used_record *r)
}
struct tdb_free_record {
- uint64_t magic_and_meta; /* Bottom 6 bits are zone bits. */
+ uint64_t magic_and_meta; /* TDB_OFF_UPPER_STEAL bits of magic */
uint64_t data_len; /* Not counting these two fields. */
/* This is why the minimum record size is 16 bytes. */
uint64_t next, prev;
};
-static inline unsigned int frec_zone_bits(const struct tdb_free_record *f)
+static inline uint64_t frec_magic(const struct tdb_free_record *f)
{
- return f->magic_and_meta & ((1 << 6) - 1);
+ return f->magic_and_meta >> (64 - TDB_OFF_UPPER_STEAL);
}
-static inline uint64_t frec_magic(const struct tdb_free_record *f)
+static inline uint64_t frec_flist(const struct tdb_free_record *f)
{
- return f->magic_and_meta & ~((1ULL << 6) - 1);
+ return f->magic_and_meta & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
}
-/* Each zone has its set of free lists at the head.
- *
- * For each zone we have a series of per-size buckets, and a final bucket for
- * "too big". */
-struct free_zone_header {
- /* How much does this zone cover? */
- uint64_t zone_bits;
- /* tdb_off_t buckets[free_buckets + 1] */
+struct tdb_recovery_record {
+ uint64_t magic;
+ /* Length of record. */
+ uint64_t max_len;
+ /* Length used. */
+ uint64_t len;
+ /* Old length of file before transaction. */
+ uint64_t eof;
};
/* this is stored at the front of every database */
uint64_t version; /* version of the code */
uint64_t hash_test; /* result of hashing HASH_MAGIC. */
uint64_t hash_seed; /* "random" seed written at creation time. */
+ tdb_off_t free_list; /* (First) free list. */
+ tdb_off_t recovery; /* Transaction recovery area. */
- tdb_off_t reserved[28];
+ tdb_off_t reserved[26];
/* Top level hash table. */
tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
};
+struct tdb_freelist {
+ struct tdb_used_record hdr;
+ tdb_off_t next;
+ tdb_off_t buckets[TDB_FREE_BUCKETS];
+};
+
/* Information about a particular (locked) hash entry. */
struct hash_info {
/* Full hash value of entry. */
struct traverse_info {
struct traverse_level {
tdb_off_t hashtable;
- const tdb_off_t *entries;
/* We ignore groups here, and treat it as a big array. */
unsigned entry;
unsigned int total_buckets;
TDB_LOCK_WAIT = 1,
/* If set, don't log an error on failure. */
TDB_LOCK_PROBE = 2,
+ /* If set, don't check for recovery (used by recovery code). */
+ TDB_LOCK_NOCHECK = 4,
};
struct tdb_lock_type {
/* How much space has been mapped (<= current file size) */
tdb_len_t map_size;
- /* Opened read-only? */
+ /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
bool read_only;
+ /* mmap read only? */
+ int mmap_flags;
+
/* Error code for last tdb error. */
enum TDB_ERROR ecode;
/* Set if we are in a transaction. */
struct tdb_transaction *transaction;
- /* What zone of the tdb to use, for spreading load. */
- uint64_t zone_off;
- /* Cached copy of zone header. */
- struct free_zone_header zhdr;
+ /* What freelist are we using? */
+ uint64_t flist_off;
/* IO methods: changes for transactions. */
const struct tdb_methods *methods;
int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
int (*oob)(struct tdb_context *, tdb_off_t, bool);
int (*expand_file)(struct tdb_context *, tdb_len_t);
+ void *(*direct)(struct tdb_context *, tdb_off_t, size_t);
};
/*
struct tdb_data key,
int ltype,
struct hash_info *h,
- struct tdb_used_record *rec);
+ struct tdb_used_record *rec,
+ struct traverse_info *tinfo);
int replace_in_hash(struct tdb_context *tdb,
struct hash_info *h,
bool is_subhash(tdb_off_t val);
/* free.c: */
-int tdb_zone_init(struct tdb_context *tdb);
+int tdb_flist_init(struct tdb_context *tdb);
+
+/* check.c needs these to iterate through free lists. */
+tdb_off_t first_flist(struct tdb_context *tdb);
+tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist);
/* If this fails, try tdb_expand. */
tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
uint64_t hash, bool growing);
/* Put this record in a free list. */
-int add_free_record(struct tdb_context *tdb, unsigned int zone_bits,
+int add_free_record(struct tdb_context *tdb,
tdb_off_t off, tdb_len_t len_with_header);
/* Set up header for a used record. */
int set_header(struct tdb_context *tdb,
struct tdb_used_record *rec,
uint64_t keylen, uint64_t datalen,
- uint64_t actuallen, uint64_t hash,
- unsigned int zone_bits);
+ uint64_t actuallen, unsigned hashlow);
/* Used by tdb_check to verify. */
-unsigned int size_to_bucket(unsigned int free_buckets, tdb_len_t data_len);
-tdb_off_t bucket_off(tdb_off_t zone_off, tdb_off_t bucket);
+unsigned int size_to_bucket(tdb_len_t data_len);
+tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket);
/* io.c: */
/* Initialize tdb->methods. */
/* Clear an ondisk area. */
int zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
-/* Return a non-zero offset in this array, or num. */
-tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb, tdb_off_t off,
- uint64_t num);
+/* Return a non-zero offset between >= start < end in this array (or end). */
+tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
+ tdb_off_t base,
+ uint64_t start,
+ uint64_t end);
/* Return a zero offset in this array, or num. */
tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
enum tdb_lock_flags waitflag);
void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
-/* Do we have any locks? */
-bool tdb_has_locks(struct tdb_context *tdb);
+/* Serialize transaction start. */
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
+int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
+
+/* Do we have any hash locks (ie. via tdb_chainlock) ? */
+bool tdb_has_hash_locks(struct tdb_context *tdb);
/* Lock entire database. */
int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable);
int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
+int tdb_allrecord_upgrade(struct tdb_context *tdb);
/* Serialize db open. */
-int tdb_lock_open(struct tdb_context *tdb);
+int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
void tdb_unlock_open(struct tdb_context *tdb);
+bool tdb_has_open_lock(struct tdb_context *tdb);
/* Serialize db expand. */
int tdb_lock_expand(struct tdb_context *tdb, int ltype);
void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
bool tdb_has_expansion_lock(struct tdb_context *tdb);
+/* If it needs recovery, grab all the locks and do it. */
+int tdb_lock_and_recover(struct tdb_context *tdb);
/* traverse.c: */
int first_in_hash(struct tdb_context *tdb, int ltype,
struct traverse_info *tinfo,
TDB_DATA *kbuf, size_t *dlen);
-
-#if 0
-/* Low-level locking primitives. */
-int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
- enum tdb_lock_flags flags);
-int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t offset, int ltype);
-
-int tdb_munmap(struct tdb_context *tdb);
-void tdb_mmap(struct tdb_context *tdb);
-int tdb_lock(struct tdb_context *tdb, int list, int ltype);
-int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
-bool tdb_have_locks(struct tdb_context *tdb);
-int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
-int tdb_brlock(struct tdb_context *tdb,
- int rw_type, tdb_off_t offset, size_t len,
- enum tdb_lock_flags flags);
-int tdb_brunlock(struct tdb_context *tdb,
- int rw_type, tdb_off_t offset, size_t len);
-bool tdb_have_extra_locks(struct tdb_context *tdb);
-void tdb_release_extra_locks(struct tdb_context *tdb);
-int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
-int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
-int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
- enum tdb_lock_flags flags, bool upgradable);
-int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
-int tdb_allrecord_upgrade(struct tdb_context *tdb);
-int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
-int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
-int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
-int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
-int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
-tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec);
-int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
-int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
-int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
-int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
+/* transaction.c: */
+int tdb_transaction_recover(struct tdb_context *tdb);
bool tdb_needs_recovery(struct tdb_context *tdb);
-int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
-int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
-int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct tdb_record *rec);
-unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
-int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
- tdb_off_t offset, tdb_len_t len,
- int (*parser)(TDB_DATA key, TDB_DATA data,
- void *private_data),
- void *private_data);
-tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
- struct tdb_record *rec);
-void tdb_io_init(struct tdb_context *tdb);
-int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
-int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off,
- struct tdb_record *rec);
-#endif
#ifdef TDB_TRACE
void tdb_trace(struct tdb_context *tdb, const char *op);