X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=01172732980689ba3dd62a699ded3b38580ad9a9;hp=1a81286d5e36280a0fe9329474921ab6c480c9ff;hb=f5087965ebdb24618ca59854b7a819e21c9fdf78;hpb=fe55330a60e4e14ea6cac2ff40d50eddca4cf140 diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h index 1a81286d..01172732 100644 --- a/ccan/tdb2/private.h +++ b/ccan/tdb2/private.h @@ -36,6 +36,7 @@ #include "config.h" #include #include +#include #ifdef HAVE_BYTESWAP_H #include #endif @@ -65,7 +66,6 @@ typedef uint64_t tdb_off_t; #define TDB_VERSION ((uint64_t)(0x26011967 + 7)) #define TDB_MAGIC ((uint64_t)0x1999) #define TDB_FREE_MAGIC ((uint64_t)0xFE) -#define TDB_COALESCING_MAGIC ((uint64_t)0xFD) #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL) #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL) #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL) @@ -91,6 +91,8 @@ 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) /* Extend file by least 100 times larger than needed. */ #define TDB_EXTENSION_FACTOR 100 @@ -100,11 +102,11 @@ typedef uint64_t tdb_off_t; /* 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) @@ -116,6 +118,9 @@ typedef uint64_t tdb_off_t; #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_FLIST_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1) + #if !HAVE_BSWAP_64 static inline uint64_t bswap_64(uint64_t x) { @@ -209,6 +214,12 @@ struct tdb_recovery_record { 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[64]; /* for /etc/magic */ @@ -256,7 +267,7 @@ 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. */ @@ -279,6 +290,15 @@ 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; @@ -308,8 +328,8 @@ struct tdb_context { uint32_t flags; /* Logging function */ - tdb_logfn_t log; - void *log_priv; + tdb_logfn_t logfn; + void *log_private; /* Hash function. */ tdb_hashfn_t khash; @@ -328,11 +348,14 @@ struct tdb_context { /* 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; @@ -344,7 +367,7 @@ struct tdb_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); + void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool); }; /* @@ -418,10 +441,6 @@ 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); @@ -530,6 +549,12 @@ int next_in_hash(struct tdb_context *tdb, int ltype, int tdb_transaction_recover(struct tdb_context *tdb); bool tdb_needs_recovery(struct tdb_context *tdb); +/* tdb.c: */ +void COLD tdb_logerr(struct tdb_context *tdb, + enum TDB_ERROR ecode, + enum tdb_debug_level level, + const char *fmt, ...); + #ifdef TDB_TRACE void tdb_trace(struct tdb_context *tdb, const char *op); void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);