X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=4bce11a1c539ea8e39bb09c6179d495bf669e342;hp=2074e3155a0046ed737cb92cadb80e7a00d6148e;hb=c8f6f8c2dea76042f74c02eff048847e62adcca6;hpb=84a19b9bada57df46adc76d46672546a675b00a7 diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h index 2074e315..4bce11a1 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 @@ -18,26 +18,19 @@ License along with this library; if not, see . */ -#define _XOPEN_SOURCE 500 -#define _FILE_OFFSET_BITS 64 -#include -#include +#include #include +#include #include #include #include #include -#include #include #include #include #include -#include "config.h" -#include #include -#ifdef HAVE_BYTESWAP_H -#include -#endif +#include #ifndef TEST_IT #define TEST_IT(cond) @@ -60,61 +53,86 @@ 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_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_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST) + +/* Packing errors into pointers and v.v. */ +#define TDB_PTR_IS_ERR(ptr) \ + unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST) +#define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p)) +#define TDB_ERR_PTR(err) ((void *)(long)(err)) -#define TDB_OFF_ERR ((tdb_off_t)-1) +/* Common case of returning true, false or -ve error. */ +typedef int tdb_bool_err; /* Prevent others from opening the file. */ #define TDB_OPEN_LOCK 0 -/* Doing a transaction. */ -#define TDB_TRANSACTION_LOCK 1 /* Expanding file. */ #define TDB_EXPANSION_LOCK 2 +/* Doing a transaction. */ +#define TDB_TRANSACTION_LOCK 8 /* Hash chain locks. */ -#define TDB_HASH_LOCK_START 3 - -/* We start wih 256 hash buckets, and a 64k-sized zone. */ -#define INITIAL_HASH_BITS 8 -#define INITIAL_ZONE_BITS 16 - -/* Try to create zones at least 32 times larger than allocations. */ -#define TDB_COMFORT_FACTOR_BITS 5 - -/* 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) -{ - return (((x&0x000000FFULL)<<56) - | ((x&0x0000FF00ULL)<<48) - | ((x&0x00FF0000ULL)<<40) - | ((x&0xFF000000ULL)<<32) - | ((x>>8)&0xFF000000ULL) - | ((x>>16)&0x00FF0000ULL) - | ((x>>24)&0x0000FF00ULL) - | ((x>>32)&0x000000FFULL)); -} -#endif +#define TDB_HASH_LOCK_START 64 + +/* 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 +/* This is currently 10: beyond this we chain. */ +#define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS) + +/* 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 +/* The bit number where we store extra hash bits. */ +#define TDB_OFF_HASH_EXTRA_BIT 57 +#define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56 + +/* Additional features we understand. Currently: none. */ +#define TDB_FEATURE_MASK ((uint64_t)0) + +/* 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)) + +/* Indicates this entry is not on an flist (can happen during coalescing) */ +#define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1) struct tdb_used_record { /* For on-disk compatibility, we avoid bitfields: 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. */ @@ -141,14 +159,9 @@ 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) @@ -157,168 +170,240 @@ 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); } -/* 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. */ +struct tdb_recovery_record { + uint64_t magic; + /* Length of record (add this header to get total length). */ + 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 */ 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. */ + tdb_off_t free_table; /* (First) free table. */ + tdb_off_t recovery; /* Transaction recovery area. */ - struct tdb_header_volatile v; + uint64_t features_used; /* Features all writers understand */ + uint64_t features_offered; /* Features offered */ - tdb_off_t reserved[19]; + uint64_t seqnum; /* Sequence number for TDB_SEQNUM */ + + tdb_off_t reserved[23]; + + /* 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]; }; -/* 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] */ +/* 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]; }; +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[TDB_MAX_LEVELS + 1]; + unsigned int num_levels; + unsigned int toplevel_group; + /* This makes delete-everything-inside-traverse work as expected. */ + tdb_off_t prev; +}; + +typedef uint32_t tdb1_len_t; +typedef uint32_t tdb1_off_t; + enum tdb_lock_flags { /* WAIT == F_SETLKW, NOWAIT == F_SETLK */ TDB_LOCK_NOWAIT = 0, 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 { - uint32_t off; +struct tdb_lock { + struct tdb_context *owner; + off_t off; uint32_t count; uint32_t ltype; }; -struct tdb_context { - /* Filename of the database. */ - const char *name; +/* 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_file { + /* How many are sharing us? */ + unsigned int refcnt; /* Mmap (if any), or malloc (for TDB_INTERNAL). */ void *map_ptr; - /* 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? */ - bool read_only; - - /* 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; - - /* Logging function */ - tdb_logfn_t log; - void *log_priv; - - /* Hash function. */ - tdb_hashfn_t khash; - void *hash_priv; - - /* 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; - - /* IO methods: changes for transactions. */ - const struct tdb_methods *methods; + /* The file descriptor (-1 for TDB_INTERNAL). */ + int fd; /* Lock information */ - struct tdb_lock_type allrecord_lock; - uint64_t num_lockrecs; - struct tdb_lock_type *lockrecs; + pid_t locker; + struct tdb_lock allrecord_lock; + size_t num_lockrecs; + struct tdb_lock *lockrecs; - /* Single list of all TDBs, to avoid multiple opens. */ - struct tdb_context *next; - dev_t device; + /* Identity of this file. */ + 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 (*oob)(struct tdb_context *, tdb_off_t, bool); - int (*expand_file)(struct tdb_context *, tdb_len_t); + enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *, + tdb_len_t); + enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *, + tdb_len_t); + enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool); + enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t); + void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool); }; /* internal prototypes */ -/* tdb.c: */ -/* Returns true if header changed (and updates it). */ -bool header_changed(struct tdb_context *tdb); +/* hash.c: */ +uint64_t tdb_jenkins_hash(const void *key, size_t length, uint64_t seed, + void *unused); -/* Commit header to disk. */ -int write_header(struct tdb_context *tdb); +tdb_bool_err first_in_hash(struct tdb_context *tdb, + struct traverse_info *tinfo, + TDB_DATA *kbuf, size_t *dlen); + +tdb_bool_err next_in_hash(struct tdb_context *tdb, + struct traverse_info *tinfo, + TDB_DATA *kbuf, size_t *dlen); /* 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); + +enum TDB_ERROR replace_in_hash(struct tdb_context *tdb, + struct hash_info *h, + tdb_off_t new_off); + +enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h, + tdb_off_t new_off); + +enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h); +/* For tdb_check */ +bool is_subhash(tdb_off_t val); /* free.c: */ -int tdb_zone_init(struct tdb_context *tdb); +enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb); -/* If this fails, try tdb_expand. */ +/* 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); + +/* This returns space or -ve error number. */ 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, - tdb_off_t off, tdb_len_t len_with_header); +enum TDB_ERROR add_free_record(struct tdb_context *tdb, + tdb_off_t off, tdb_len_t len_with_header, + enum tdb_lock_flags waitflag, + bool coalesce_ok); -/* 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); +/* Set up header for a used/ftable/htable/chain record. */ +enum TDB_ERROR set_header(struct tdb_context *tdb, + struct tdb_used_record *rec, + 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 */ +tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off); /* io.c: */ /* Initialize tdb->methods. */ @@ -328,136 +413,226 @@ void tdb_io_init(struct tdb_context *tdb); void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size); /* Unmap and try to map the tdb. */ -void tdb_munmap(struct tdb_context *tdb); +void tdb_munmap(struct tdb_file *file); 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); +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. */ +enum TDB_ERROR 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); /* Write an offset at an offset. */ -int tdb_write_off(struct tdb_context *tdb, tdb_off_t off, tdb_off_t val); +enum TDB_ERROR 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); +enum TDB_ERROR 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, 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); /* Writes a converted copy of a record. */ -int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off, - const void *rec, size_t len); +enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off, + const void *rec, size_t len); /* Reads record and converts it */ -int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off, - void *rec, size_t len); +enum TDB_ERROR 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); +/* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */ +void tdb_inc_seqnum(struct tdb_context *tdb); /* lock.c: */ -void tdb_lock_init(struct tdb_context *tdb); +/* Lock/unlock a range of hashes. */ +enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb, + tdb_off_t hash_lock, tdb_len_t hash_range, + int ltype, enum tdb_lock_flags waitflag); +enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb, + tdb_off_t hash_lock, + tdb_len_t hash_range, int ltype); -/* 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); +/* For closing the file. */ +void tdb_lock_cleanup(struct tdb_context *tdb); /* 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); +enum TDB_ERROR 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. */ +enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype); +void 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); +enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype, + enum tdb_lock_flags flags, bool upgradable); +void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype); +enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb, off_t start); /* Serialize db open. */ -int tdb_lock_open(struct tdb_context *tdb); -void tdb_unlock_open(struct tdb_context *tdb); +enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, + int ltype, enum tdb_lock_flags flags); +void tdb_unlock_open(struct tdb_context *tdb, int ltype); +bool tdb_has_open_lock(struct tdb_context *tdb); /* Serialize db expand. */ -int tdb_lock_expand(struct tdb_context *tdb, int ltype); +enum TDB_ERROR 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. */ +enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb); -#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); +/* Byte-range lock wrappers for TDB1 to access. */ +enum TDB_ERROR tdb_brlock(struct tdb_context *tdb, + int rw_type, tdb_off_t offset, tdb_off_t len, + enum tdb_lock_flags flags); -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); -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 +enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb, + int rw_type, tdb_off_t offset, size_t len); + +enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb, + tdb_off_t offset, int ltype, + enum tdb_lock_flags flags); + +enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb, + tdb_off_t off, int ltype); + +enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb, + int ltype, enum tdb_lock_flags flags, + tdb_off_t off, tdb_off_t len); + +/* Default lock and unlock functions. */ +int tdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *); +int tdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *); + +/* transaction.c: */ +enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb); +tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb); + +/* this is stored at the front of every database */ +struct tdb1_header { + char magic_food[32]; /* for /etc/magic */ + uint32_t version; /* version of the code */ + uint32_t hash_size; /* number of hash entries */ + tdb1_off_t rwlocks; /* obsolete - kept to detect old formats */ + tdb1_off_t recovery_start; /* offset of transaction recovery region */ + tdb1_off_t sequence_number; /* used when TDB1_SEQNUM is set */ + uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */ + uint32_t magic2_hash; /* hash of TDB1_MAGIC. */ + tdb1_off_t reserved[27]; +}; + +struct tdb1_traverse_lock { + struct tdb1_traverse_lock *next; + uint32_t off; + uint32_t hash; + int lock_rw; +}; + +struct tdb_context { + /* Single list of all TDBs, to detect multiple opens. */ + struct tdb_context *next; + + /* Filename of the database. */ + const char *name; + + /* Logging function */ + void (*log_fn)(struct tdb_context *tdb, + enum tdb_log_level level, + enum TDB_ERROR ecode, + const char *message, + void *data); + void *log_data; + + /* Open flags passed to tdb_open. */ + int open_flags; + + /* low level (fnctl) lock functions. */ + int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *); + int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *); + void *lock_data; + + /* the tdb flags passed to tdb_open. */ + uint32_t flags; + + /* Our statistics. */ + struct tdb_attribute_stats stats; + + /* The actual file information */ + struct tdb_file *file; + + /* Hash function. */ + uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *); + void *hash_data; + uint64_t hash_seed; + + /* Our open hook, if any. */ + enum TDB_ERROR (*openhook)(int fd, void *data); + void *openhook_data; + + /* Last error we returned. */ + enum TDB_ERROR last_error; + + struct { + + /* Are we accessing directly? (debugging check). */ + int direct_access; + + /* Set if we are in a transaction. */ + struct tdb_transaction *transaction; + + /* What free table are we using? */ + tdb_off_t ftable_off; + unsigned int ftable; + + /* IO methods: changes for transactions. */ + const struct tdb_methods *io; + + /* Direct access information */ + struct tdb_access_hdr *access; + } tdb2; + + struct { + int traverse_read; /* read-only traversal */ + int traverse_write; /* read-write traversal */ + + struct tdb1_header header; /* a cached copy of the header */ + struct tdb1_traverse_lock travlocks; /* current traversal locks */ + const struct tdb1_methods *io; + struct tdb1_transaction *transaction; + int page_size; + int max_dead_records; + } tdb1; +}; + +/* tdb.c: */ +enum TDB_ERROR 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);