X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=69fd40fce91fd295d2e868d569d60e6e804ecf4a;hp=6c727aff4a20a62dc800e5806079742cd32d7ec1;hb=52c4843176ce73484dd405261d13048bcd1b85e5;hpb=3d10865d159072553933fb72394862c884e60079 diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h index 6c727aff..69fd40fc 100644 --- a/ccan/tdb2/private.h +++ b/ccan/tdb2/private.h @@ -1,6 +1,6 @@ #ifndef TDB_PRIVATE_H #define TDB_PRIVATE_H - /* + /* Trivial Database 2: private types and prototypes Copyright (C) Rusty Russell 2010 @@ -36,7 +36,8 @@ #include "config.h" #include #include -#ifdef HAVE_BYTESWAP_H +#include +#if HAVE_BYTESWAP_H #include #endif @@ -63,11 +64,14 @@ typedef uint64_t tdb_off_t; #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_USED_MAGIC ((uint64_t)0x1999) +#define TDB_HTABLE_MAGIC ((uint64_t)0x1888) +#define TDB_CHAIN_MAGIC ((uint64_t)0x1777) +#define TDB_FTABLE_MAGIC ((uint64_t)0x1666) +#define TDB_FREE_MAGIC ((uint64_t)0xFE) #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) @@ -90,33 +94,35 @@ typedef uint64_t tdb_off_t; #define TDB_SUBLEVEL_HASH_BITS 6 /* And 8 entries in each group, ie 8 groups per sublevel. */ #define TDB_HASH_GROUP_BITS 3 +/* This is currently 10: beyond this we chain. */ +#define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS) -/* 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 100 times larger than needed. */ +#define TDB_EXTENSION_FACTOR 100 /* We steal bits from the offsets to store hash info. */ #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1) /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */ #define TDB_OFF_UPPER_STEAL 8 #define TDB_OFF_UPPER_STEAL_EXTRA 7 -#define TDB_OFF_UPPER_STEAL_TRUNCBIT 1 -/* If this is set, hash is truncated (only 1 bit is valid). */ -#define TDB_OFF_HASH_TRUNCATED_BIT 56 -/* The bit number where we store next level of hash. */ +/* The bit number where we store extra hash bits. */ #define TDB_OFF_HASH_EXTRA_BIT 57 +#define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56 + +/* The bit number where we store the extra hash bits. */ /* Convenience mask to get actual offset. */ #define TDB_OFF_MASK \ (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK) +/* 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)) -/* 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) +/* Indicates this entry is not on an flist (can happen during coalescing) */ +#define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1) #if !HAVE_BSWAP_64 static inline uint64_t bswap_64(uint64_t x) @@ -137,8 +143,7 @@ struct tdb_used_record { 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. */ @@ -165,14 +170,9 @@ static inline uint64_t rec_extra_padding(const struct tdb_used_record *r) 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) @@ -181,30 +181,46 @@ 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 data_len; /* Not counting these two fields. */ - /* This is why the minimum record size is 16 bytes. */ - uint64_t next, prev; + uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */ + uint64_t ftable_and_len; /* Len not counting these two fields. */ + /* This is why the minimum record size is 8 bytes. */ + uint64_t next; }; -static inline unsigned int frec_zone_bits(const struct tdb_free_record *f) +static inline uint64_t frec_prev(const struct tdb_free_record *f) { - return f->magic_and_meta & ((1 << 6) - 1); + return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1); } static inline uint64_t frec_magic(const struct tdb_free_record *f) { - return f->magic_and_meta & ~((1ULL << 6) - 1); + return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL); +} + +static inline uint64_t frec_len(const struct tdb_free_record *f) +{ + return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1); +} + +static inline unsigned frec_ftable(const struct tdb_free_record *f) +{ + return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL); } -/* 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; +}; + +/* If we bottom out of the subhashes, we chain. */ +struct tdb_chain { + tdb_off_t rec[1 << TDB_HASH_GROUP_BITS]; + tdb_off_t next; }; /* this is stored at the front of every database */ @@ -214,13 +230,21 @@ struct tdb_header { 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_table; /* (First) free table. */ + 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_freetable { + 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. */ @@ -246,7 +270,7 @@ struct traverse_info { /* We ignore groups here, and treat it as a big array. */ unsigned entry; unsigned int total_buckets; - } levels[64 / TDB_SUBLEVEL_HASH_BITS]; + } levels[TDB_MAX_LEVELS + 1]; unsigned int num_levels; unsigned int toplevel_group; /* This makes delete-everything-inside-traverse work as expected. */ @@ -259,6 +283,8 @@ enum tdb_lock_flags { 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 { @@ -267,6 +293,15 @@ struct tdb_lock_type { uint32_t ltype; }; +/* This is only needed for tdb_access_commit, but used everywhere to + * simplify. */ +struct tdb_access_hdr { + struct tdb_access_hdr *next; + tdb_off_t off; + tdb_len_t len; + bool convert; +}; + struct tdb_context { /* Filename of the database. */ const char *name; @@ -290,47 +325,55 @@ struct tdb_context { int mmap_flags; /* Error code for last tdb error. */ - enum TDB_ERROR ecode; + enum TDB_ERROR ecode; /* the flags passed to tdb_open, for tdb_reopen. */ uint32_t flags; /* Logging function */ - tdb_logfn_t log; - void *log_priv; + void (*logfn)(struct tdb_context *tdb, + enum tdb_log_level level, + void *log_private, + const char *message); + void *log_private; /* Hash function. */ - tdb_hashfn_t khash; + uint64_t (*khash)(const void *key, size_t len, uint64_t seed, void *); void *hash_priv; uint64_t hash_seed; /* 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 free table are we using? */ + tdb_off_t ftable_off; + unsigned int ftable; /* IO methods: changes for transactions. */ const struct tdb_methods *methods; /* Lock information */ struct tdb_lock_type allrecord_lock; - uint64_t num_lockrecs; + size_t num_lockrecs; struct tdb_lock_type *lockrecs; + struct tdb_attribute_stats *stats; + + /* Direct access information */ + struct tdb_access_hdr *access; + /* Single list of all TDBs, to avoid multiple opens. */ struct tdb_context *next; - dev_t device; + dev_t device; ino_t inode; }; struct tdb_methods { - int (*read)(struct tdb_context *, tdb_off_t, void *, tdb_len_t); - int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t); + int (*tread)(struct tdb_context *, tdb_off_t, void *, tdb_len_t); + int (*twrite)(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, bool); }; /* @@ -366,26 +409,32 @@ int delete_from_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_ftable_init(struct tdb_context *tdb); + +/* check.c needs these to iterate through free lists. */ +tdb_off_t first_ftable(struct tdb_context *tdb); +tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable); -/* If this fails, try tdb_expand. */ +/* This returns space or TDB_OFF_ERR. */ tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen, - uint64_t hash, bool growing); + uint64_t hash, unsigned magic, 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. */ +/* Set up header for a used/ftable/htable/chain 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); + unsigned magic, uint64_t keylen, uint64_t datalen, + 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 ftable_off, unsigned bucket); + +/* Used by tdb_summary */ +size_t dead_space(struct tdb_context *tdb, tdb_off_t off); /* io.c: */ /* Initialize tdb->methods. */ @@ -398,10 +447,6 @@ void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size); void tdb_munmap(struct tdb_context *tdb); void tdb_mmap(struct tdb_context *tdb); -/* Either make a copy into pad and return that, or return ptr into mmap. - * Converts endian (ie. will use pad in that case). */ -void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len); - /* Either alloc a copy, or give direct access. Release frees or noop. */ const void *tdb_access_read(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, bool convert); @@ -432,11 +477,6 @@ tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb, tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off, uint64_t num); -/* Even on files, we can get partial writes due to signals. */ -bool tdb_pwrite_all(int fd, const void *buf, size_t len, tdb_off_t off); -bool tdb_pread_all(int fd, void *buf, size_t len, tdb_off_t off); -bool tdb_read_all(int fd, void *buf, size_t len); - /* Allocate and make a copy of some offset. */ void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len); @@ -448,6 +488,13 @@ int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off, int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off, void *rec, size_t len); +/* Adds a stat, if it's in range. */ +void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val); +#define add_stat(tdb, statname, val) \ + do { \ + if (unlikely((tdb)->stats)) \ + add_stat_((tdb), &(tdb)->stats->statname, (val)); \ + } while (0) /* lock.c: */ void tdb_lock_init(struct tdb_context *tdb); @@ -465,23 +512,31 @@ int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_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, @@ -491,59 +546,15 @@ int next_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 + +/* tdb.c: */ +void COLD tdb_logerr(struct tdb_context *tdb, + enum TDB_ERROR ecode, + enum tdb_log_level level, + const char *fmt, ...); #ifdef TDB_TRACE void tdb_trace(struct tdb_context *tdb, const char *op);