#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 <stdint.h>
-#include <stdbool.h>
+#include <ccan/tdb2/tdb2.h>
#include <stdlib.h>
#include <stddef.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
-#include <string.h>
#include <errno.h>
#include <stdio.h>
#include <utime.h>
#include <unistd.h>
-#include "config.h"
-#include <ccan/tdb2/tdb2.h>
#include <ccan/likely/likely.h>
-#include <ccan/compiler/compiler.h>
-#if HAVE_BYTESWAP_H
-#include <byteswap.h>
-#endif
+#include <ccan/endian/endian.h>
#ifndef TEST_IT
#define TEST_IT(cond)
#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
-/* 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
+#define TDB_HASH_LOCK_START 64
/* Range for hash locks. */
#define TDB_HASH_LOCK_RANGE_BITS 30
#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 \
/* 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)
-{
- 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
-
struct tdb_used_record {
/* For on-disk compatibility, we avoid bitfields:
magic: 16, (highest)
struct tdb_recovery_record {
uint64_t magic;
- /* Length of record. */
+ /* Length of record (add this header to get total length). */
uint64_t max_len;
/* Length used. */
uint64_t len;
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];
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_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;
};
bool convert;
};
-struct tdb_context {
- /* Filename of the database. */
- const char *name;
+struct tdb_file {
+ /* 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;
-
- /* Open file descriptor (undefined for TDB_INTERNAL). */
- int fd;
-
/* How much space has been mapped (<= current file size) */
tdb_len_t map_size;
- /* 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;
-
- /* the flags passed to tdb_open, for tdb_reopen. */
- uint32_t flags;
-
- /* Logging function */
- tdb_logfn_t logfn;
- void *log_private;
-
- /* 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 free table are we using? */
- tdb_off_t ftable_off;
- unsigned int ftable;
-
- /* 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;
+ pid_t locker;
+ struct tdb_lock allrecord_lock;
size_t num_lockrecs;
- struct tdb_lock_type *lockrecs;
-
- struct tdb_attribute_stats *stats;
+ struct tdb_lock *lockrecs;
- /* Direct access information */
- struct tdb_access_hdr *access;
-
- /* 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
*/
/* hash.c: */
-void tdb_hash_init(struct tdb_context *tdb);
+uint64_t tdb_jenkins_hash(const void *key, size_t length, uint64_t seed,
+ void *unused);
+
+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_ftable_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_ftable(struct tdb_context *tdb);
tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
-/* This returns space or TDB_OFF_ERR. */
+/* This returns space or -ve error number. */
tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
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,
+ enum tdb_lock_flags waitflag,
+ bool coalesce_ok);
/* 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);
+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 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 alloc a copy, or give direct access. Release frees or noop. */
/* 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,
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);
-/* 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)
+/* 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. */
-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, off_t start);
/* Serialize db open. */
-int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags);
-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. */
-int tdb_lock_and_recover(struct tdb_context *tdb);
+enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
+
+/* 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);
+
+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);
-/* 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_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: */
-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);
+
+/* 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: */
-void COLD tdb_logerr(struct tdb_context *tdb,
- enum TDB_ERROR ecode,
- enum tdb_debug_level level,
- const char *fmt, ...);
+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);