X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=6780fdadd421303ee9aec389cae999b8b96d755c;hp=1b6cebea229e38d2422cb9acfa45f37873c587f5;hb=8afb9681b3be7149cf8913a5aefe915194fd20f7;hpb=ee6f11b307599200a574208372bc962eff81a9b1 diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h index 1b6cebea..6780fdad 100644 --- a/ccan/tdb2/private.h +++ b/ccan/tdb2/private.h @@ -23,6 +23,7 @@ #include #include #include +#include #include #include #include @@ -60,19 +61,13 @@ 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) +#define TDB_FREE_MAGIC ((~(uint64_t)TDB_MAGIC) << 6) #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL) #define TDB_RECOVERY_MAGIC (0xf53bc0e7U) #define TDB_RECOVERY_INVALID_MAGIC (0x0) -#define TDB_EXTRA_HASHBITS (11) /* We steal 11 bits to stash hash info. */ -#define TDB_EXTRA_HASHBITS_NUM (3) #define TDB_OFF_ERR ((tdb_off_t)-1) @@ -80,13 +75,48 @@ typedef uint64_t tdb_off_t; #define TDB_OPEN_LOCK 0 /* Doing a transaction. */ #define TDB_TRANSACTION_LOCK 1 +/* Expanding file. */ +#define TDB_EXPANSION_LOCK 2 /* Hash chain locks. */ -#define TDB_HASH_LOCK_START 2 - -/* We start wih 256 hash buckets, 10 free buckets. A 4k-sized zone. */ -#define INITIAL_HASH_BITS 8 -#define INITIAL_FREE_BUCKETS 10 -#define INITIAL_ZONE_BITS 12 +#define TDB_HASH_LOCK_START 3 + +/* Range for hash locks. */ +#define TDB_HASH_LOCK_RANGE_BITS 30 +#define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS) + +/* We have 1024 entries in the top level. */ +#define TDB_TOPLEVEL_HASH_BITS 10 +/* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */ +#define TDB_SUBLEVEL_HASH_BITS 6 +/* 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 + +/* 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. */ +#define TDB_OFF_HASH_EXTRA_BIT 57 +/* Convenience mask to get actual offset. */ +#define TDB_OFF_MASK \ + (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK) + +/* 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) #if !HAVE_BSWAP_64 static inline uint64_t bswap_64(uint64_t x) @@ -106,8 +136,9 @@ struct tdb_used_record { /* For on-disk compatibility, we avoid bitfields: magic: 16, (highest) key_len_bits: 5, - hash:11, - extra_padding: 32 (lowest) + extra_padding: 32 + hash_bits: 5, + zone_bits: 6 (lowest) */ uint64_t magic_and_meta; /* The bottom key_len_bits*2 are key length, rest is data length. */ @@ -131,12 +162,17 @@ static inline uint64_t rec_data_length(const struct tdb_used_record *r) static inline uint64_t rec_extra_padding(const struct tdb_used_record *r) { - return r->magic_and_meta & 0xFFFFFFFF; + return (r->magic_and_meta >> 11) & 0xFFFFFFFF; } -static inline uint64_t rec_hash(const struct tdb_used_record *r) +static inline unsigned int rec_zone_bits(const struct tdb_used_record *r) { - return ((r->magic_and_meta >> 32) & ((1ULL << 11) - 1)); + 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); } static inline uint16_t rec_magic(const struct tdb_used_record *r) @@ -145,33 +181,76 @@ static inline uint16_t rec_magic(const struct tdb_used_record *r) } struct tdb_free_record { - uint64_t magic; + 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; }; -/* These parts can change while we have db open. */ -struct tdb_header_volatile { - uint64_t generation; /* Makes sure it changes on every update. */ - uint64_t hash_bits; /* Entries in hash table. */ - uint64_t hash_off; /* Offset of hash table. */ - uint64_t num_zones; /* How many zones in the file. */ - uint64_t zone_bits; /* Size of zones. */ - uint64_t free_buckets; /* How many buckets in each zone. */ - uint64_t free_off; /* Arrays of free entries. */ +static inline unsigned int frec_zone_bits(const struct tdb_free_record *f) +{ + return f->magic_and_meta & ((1 << 6) - 1); +} + +static inline uint64_t frec_magic(const struct tdb_free_record *f) +{ + return f->magic_and_meta & ~((1ULL << 6) - 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] */ }; /* this is stored at the front of every database */ struct tdb_header { - char magic_food[32]; /* for /etc/magic */ + char magic_food[64]; /* for /etc/magic */ + /* FIXME: Make me 32 bit? */ 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. */ - struct tdb_header_volatile v; + tdb_off_t reserved[28]; + + /* Top level hash table. */ + tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS]; +}; + +/* Information about a particular (locked) hash entry. */ +struct hash_info { + /* Full hash value of entry. */ + uint64_t h; + /* Start and length of lock acquired. */ + tdb_off_t hlock_start; + tdb_len_t hlock_range; + /* Start of hash group. */ + tdb_off_t group_start; + /* Bucket we belong in. */ + unsigned int home_bucket; + /* Bucket we (or an empty space) were found in. */ + unsigned int found_bucket; + /* How many bits of the hash are already used. */ + unsigned int hash_used; + /* Current working group. */ + tdb_off_t group[1 << TDB_HASH_GROUP_BITS]; +}; - tdb_off_t reserved[19]; +struct traverse_info { + struct traverse_level { + tdb_off_t hashtable; + /* We ignore groups here, and treat it as a big array. */ + unsigned entry; + unsigned int total_buckets; + } levels[64 / TDB_SUBLEVEL_HASH_BITS]; + unsigned int num_levels; + unsigned int toplevel_group; + /* This makes delete-everything-inside-traverse work as expected. */ + tdb_off_t prev; }; enum tdb_lock_flags { @@ -195,23 +274,24 @@ struct tdb_context { /* Mmap (if any), or malloc (for TDB_INTERNAL). */ void *map_ptr; + /* Are we accessing directly? (debugging check). */ + int direct_access; + /* Open file descriptor (undefined for TDB_INTERNAL). */ int fd; /* 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; - /* A cached copy of the header */ - struct tdb_header header; - /* (for debugging). */ - bool header_uptodate; - /* the flags passed to tdb_open, for tdb_reopen. */ uint32_t flags; @@ -222,12 +302,15 @@ struct tdb_context { /* Hash function. */ tdb_hashfn_t khash; 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 last_zone; + uint64_t zone_off; + /* Cached copy of zone header. */ + struct free_zone_header zhdr; /* IO methods: changes for transactions. */ const struct tdb_methods *methods; @@ -253,40 +336,56 @@ struct tdb_methods { /* internal prototypes */ -/* tdb.c: */ -/* Returns true if header changed (and updates it). */ -bool header_changed(struct tdb_context *tdb); - -/* Commit header to disk. */ -int write_header(struct tdb_context *tdb); +/* hash.c: */ +void tdb_hash_init(struct tdb_context *tdb); /* Hash random memory. */ uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len); -/* offset of hash table entry for this list/hash value */ -tdb_off_t hash_off(struct tdb_context *tdb, uint64_t list); +/* Hash on disk. */ +uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off); + +/* Find and lock a hash entry (or where it would be). */ +tdb_off_t find_and_lock(struct tdb_context *tdb, + struct tdb_data key, + int ltype, + struct hash_info *h, + struct tdb_used_record *rec, + struct traverse_info *tinfo); + +int replace_in_hash(struct tdb_context *tdb, + struct hash_info *h, + tdb_off_t new_off); + +int add_to_hash(struct tdb_context *tdb, struct hash_info *h, + tdb_off_t new_off); +int delete_from_hash(struct tdb_context *tdb, struct hash_info *h); + +/* For tdb_check */ +bool is_subhash(tdb_off_t val); /* free.c: */ -void tdb_zone_init(struct tdb_context *tdb); +int tdb_zone_init(struct tdb_context *tdb); /* 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, +int add_free_record(struct tdb_context *tdb, unsigned int zone_bits, 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); + uint64_t actuallen, unsigned hashlow, + unsigned int zone_bits); /* Used by tdb_check to verify. */ -unsigned int size_to_bucket(struct tdb_context *tdb, tdb_len_t data_len); -tdb_off_t zone_of(struct tdb_context *tdb, tdb_off_t off); +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); /* io.c: */ /* Initialize tdb->methods. */ @@ -306,7 +405,13 @@ 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); +void *tdb_access_write(struct tdb_context *tdb, + tdb_off_t off, tdb_len_t len, bool convert); + +/* Release result of tdb_access_read/write. */ void tdb_access_release(struct tdb_context *tdb, const void *p); +/* Commit result of tdb_acces_write. */ +int tdb_access_commit(struct tdb_context *tdb, void *p); /* Convenience routine to get an offset. */ tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off); @@ -317,9 +422,11 @@ int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val); /* 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, @@ -341,21 +448,22 @@ 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); -/* Hash on disk. */ -uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off); /* lock.c: */ void tdb_lock_init(struct tdb_context *tdb); -/* Lock/unlock a particular hash list. */ -tdb_off_t tdb_lock_list(struct tdb_context *tdb, uint64_t hash, - int ltype, enum tdb_lock_flags waitflag); -int tdb_unlock_list(struct tdb_context *tdb, tdb_off_t list, int ltype); +/* Lock/unlock a range of hashes. */ +int tdb_lock_hashes(struct tdb_context *tdb, + tdb_off_t hash_lock, tdb_len_t hash_range, + int ltype, enum tdb_lock_flags waitflag); +int tdb_unlock_hashes(struct tdb_context *tdb, + tdb_off_t hash_lock, + tdb_len_t hash_range, int ltype); -/* Lock/unlock a particular free list. */ -int tdb_lock_free_list(struct tdb_context *tdb, tdb_off_t flist, - enum tdb_lock_flags waitflag); -void tdb_unlock_free_list(struct tdb_context *tdb, tdb_off_t flist); +/* Lock/unlock a particular free bucket. */ +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); @@ -368,9 +476,21 @@ int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype); /* Serialize db open. */ int tdb_lock_open(struct tdb_context *tdb); void tdb_unlock_open(struct tdb_context *tdb); -/* Expand the file. */ -int tdb_expand(struct tdb_context *tdb, tdb_len_t klen, tdb_len_t dlen, - bool growing); + +/* 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); + + +/* traverse.c: */ +int first_in_hash(struct tdb_context *tdb, int ltype, + struct traverse_info *tinfo, + TDB_DATA *kbuf, size_t *dlen); +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. */