X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=4ecda6a5f529729518b051a53db711f582fde6db;hp=232229f5e912d076cbbdc337ea43c4ccb032ef8a;hb=850c5cfed46cd4f38df79783791969c7b30ad9da;hpb=576802602c19ed3cfda98414ffc9b118c2675931 diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h index 232229f5..4ecda6a5 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 @@ -63,14 +64,26 @@ 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_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_COALESCING_MAGIC ((uint64_t)0xFD) #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL) #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL) #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL) #define TDB_OFF_ERR ((tdb_off_t)-1) +#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((void *)(ptr) >= (void *)(long)TDB_ERR_LAST) +#define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p)) +#define TDB_ERR_PTR(err) ((void *)(long)(err)) + +/* 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 @@ -91,6 +104,8 @@ 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 @@ -116,6 +131,9 @@ 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_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1) + #if !HAVE_BSWAP_64 static inline uint64_t bswap_64(uint64_t x) { @@ -174,7 +192,7 @@ static inline uint16_t rec_magic(const struct tdb_used_record *r) struct tdb_free_record { uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */ - uint64_t flist_and_len; /* Len not counting these two fields. */ + uint64_t ftable_and_len; /* Len not counting these two fields. */ /* This is why the minimum record size is 8 bytes. */ uint64_t next; }; @@ -191,12 +209,12 @@ static inline uint64_t frec_magic(const struct tdb_free_record *f) static inline uint64_t frec_len(const struct tdb_free_record *f) { - return f->flist_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1); + return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1); } -static inline unsigned frec_flist(const struct tdb_free_record *f) +static inline unsigned frec_ftable(const struct tdb_free_record *f) { - return f->flist_and_len >> (64 - TDB_OFF_UPPER_STEAL); + return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL); } struct tdb_recovery_record { @@ -209,6 +227,12 @@ 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 */ @@ -216,7 +240,7 @@ 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_list; /* (First) free list. */ + tdb_off_t free_table; /* (First) free table. */ tdb_off_t recovery; /* Transaction recovery area. */ tdb_off_t reserved[26]; @@ -225,7 +249,7 @@ struct tdb_header { tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS]; }; -struct tdb_freelist { +struct tdb_freetable { struct tdb_used_record hdr; tdb_off_t next; tdb_off_t buckets[TDB_FREE_BUCKETS]; @@ -256,7 +280,7 @@ 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 +303,15 @@ 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; @@ -302,49 +335,55 @@ 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 freelist are we using? */ - uint64_t flist_off; - unsigned int flist; + /* 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); + void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool); }; /* @@ -353,6 +392,14 @@ struct tdb_methods { /* hash.c: */ void tdb_hash_init(struct tdb_context *tdb); +int first_in_hash(struct tdb_context *tdb, + struct traverse_info *tinfo, + TDB_DATA *kbuf, size_t *dlen); + +int 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); @@ -380,29 +427,29 @@ int delete_from_hash(struct tdb_context *tdb, struct hash_info *h); bool is_subhash(tdb_off_t val); /* free.c: */ -int tdb_flist_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_flist(struct tdb_context *tdb); -tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist); +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, tdb_off_t off, tdb_len_t len_with_header); -/* Set up header for a used record. */ -int set_used_header(struct tdb_context *tdb, - struct tdb_used_record *rec, - uint64_t keylen, uint64_t datalen, - uint64_t actuallen, unsigned hashlow); +/* Set up header for a used/ftable/htable/chain record. */ +int 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(tdb_len_t data_len); -tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket); +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); @@ -424,9 +471,6 @@ const void *tdb_access_read(struct tdb_context *tdb, void *tdb_access_write(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, bool convert); -/* Is this pointer direct? (Otherwise it's malloced) */ -bool is_direct(const struct tdb_context *tdb, const void *p); - /* Release result of tdb_access_read/write. */ void tdb_access_release(struct tdb_context *tdb, const void *p); /* Commit result of tdb_acces_write. */ @@ -451,11 +495,6 @@ 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); @@ -479,56 +518,55 @@ void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val); void tdb_lock_init(struct tdb_context *tdb); /* 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); +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 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); /* Serialize transaction start. */ -int tdb_transaction_lock(struct tdb_context *tdb, int ltype); -int tdb_transaction_unlock(struct tdb_context *tdb, int ltype); +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); -int tdb_allrecord_upgrade(struct tdb_context *tdb); +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); /* Serialize db open. */ -int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags); +enum TDB_ERROR 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); +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. */ -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); -int next_in_hash(struct tdb_context *tdb, int ltype, - struct traverse_info *tinfo, - TDB_DATA *kbuf, size_t *dlen); +enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb); /* transaction.c: */ int tdb_transaction_recover(struct tdb_context *tdb); bool tdb_needs_recovery(struct tdb_context *tdb); +/* 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); void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);