#ifndef TDB_PRIVATE_H
#define TDB_PRIVATE_H
- /*
+ /*
Trivial Database 2: private types and prototypes
Copyright (C) Rusty Russell 2010
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
-#define _XOPEN_SOURCE 500
-#define _FILE_OFFSET_BITS 64
+#include "config.h"
#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <utime.h>
#include <unistd.h>
-#include "config.h"
#include <ccan/tdb2/tdb2.h>
#include <ccan/likely/likely.h>
-#ifdef HAVE_BYTESWAP_H
+#include <ccan/compiler/compiler.h>
+#if HAVE_BYTESWAP_H
#include <byteswap.h>
#endif
#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((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))
+
+/* 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
#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 32 times larger than needed. */
-#define TDB_EXTENSION_FACTOR 32
+/* 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
+
+/* 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)
#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)
{
}
struct tdb_free_record {
- uint64_t magic_and_meta; /* TDB_OFF_UPPER_STEAL bits of magic */
- 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 uint64_t frec_prev(const struct tdb_free_record *f)
+{
+ 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 >> (64 - TDB_OFF_UPPER_STEAL);
+ return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
}
-static inline uint64_t frec_flist(const struct tdb_free_record *f)
+static inline uint64_t frec_len(const struct tdb_free_record *f)
{
- return f->magic_and_meta & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
+ 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);
}
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 */
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];
+ uint64_t features_used; /* Features all writers understand */
+ uint64_t features_offered; /* Features offered */
+
+ 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_freelist {
+struct tdb_freetable {
struct tdb_used_record hdr;
tdb_off_t next;
tdb_off_t buckets[TDB_FREE_BUCKETS];
/* 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. */
TDB_LOCK_NOCHECK = 4,
};
-struct tdb_lock_type {
+struct tdb_lock {
+ struct tdb_context *owner;
uint32_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 {
+ /* Single list of all TDBs, to detect multiple opens. */
+ struct tdb_file *next;
+
+ /* How many are sharing us? */
+ unsigned int refcnt;
/* Mmap (if any), or malloc (for TDB_INTERNAL). */
void *map_ptr;
- /* Are we accessing directly? (debugging check). */
- int direct_access;
+ /* How much space has been mapped (<= current file size) */
+ tdb_len_t map_size;
- /* Open file descriptor (undefined for TDB_INTERNAL). */
+ /* The file descriptor (-1 for TDB_INTERNAL). */
int fd;
- /* How much space has been mapped (<= current file size) */
- tdb_len_t map_size;
+ /* Lock information */
+ pid_t locker;
+ struct tdb_lock allrecord_lock;
+ size_t num_lockrecs;
+ struct tdb_lock *lockrecs;
+
+ /* Identity of this file. */
+ dev_t device;
+ ino_t inode;
+};
+
+struct tdb_context {
+ /* Filename of the database. */
+ const char *name;
+
+ /* Are we accessing directly? (debugging check). */
+ int direct_access;
/* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
bool read_only;
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;
+ /* 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;
- struct tdb_lock_type *lockrecs;
+ struct tdb_attribute_stats *stats;
- /* Single list of all TDBs, to avoid multiple opens. */
- struct tdb_context *next;
- dev_t device;
- ino_t inode;
+ /* Direct access information */
+ struct tdb_access_hdr *access;
+
+ /* Last error we returned. */
+ enum TDB_ERROR last_error;
+
+ /* The actual file information */
+ struct tdb_file *file;
};
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);
- void *(*direct)(struct tdb_context *, tdb_off_t, size_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);
};
/*
/* hash.c: */
void tdb_hash_init(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);
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);
+enum TDB_ERROR 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);
+enum TDB_ERROR 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);
+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_flist_init(struct tdb_context *tdb);
+enum TDB_ERROR 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 -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,
- 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);
-/* 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, unsigned hashlow);
+/* 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(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);
+tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
/* io.c: */
/* Initialize tdb->methods. */
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);
/* 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);
+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 between >= start < end in this array (or end). */
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);
/* 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);
+/* 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);
+/* 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: */
/* 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);
+
+/* 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);
/* 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);
+enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
+tdb_bool_err 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);