AC_CHECK_FUNC(epoll_create, AC_DEFINE([HAVE_EPOLL], 1, Whether Linux epoll is available))
AC_CHECK_FUNCS(backtrace_symbols dirfd getusershell pread pwrite pselect)
AC_CHECK_FUNCS(setlinebuf strlcat strlcpy strnlen)
+AC_CHECK_FUNCS(mmap utime getpagesize) dnl needed by tbd
dnl search for necessary libraries
AC_SEARCH_LIBS(gethostbyname, nsl)
/*
Unix SMB/CIFS implementation.
- Samba database functions
- Copyright (C) Andrew Tridgell 1999
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2004
- This program is free software; you can redistribute it and/or modify
- it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2 of the License, or
- (at your option) any later version.
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
- This program is distributed in the hope that it will be useful,
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- GNU General Public License for more details.
-
- You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software
- Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#ifdef __cplusplus
extern "C" {
#endif
+#include "signal.h"
/* flags to tdb_store() */
-#define TDB_REPLACE 1
-#define TDB_INSERT 2
-#define TDB_MODIFY 3
+#define TDB_REPLACE 1 /* Unused */
+#define TDB_INSERT 2 /* Don't overwrite an existing entry */
+#define TDB_MODIFY 3 /* Don't create an existing entry */
/* flags for tdb_open() */
#define TDB_DEFAULT 0 /* just a readability place holder */
#define TDB_NOMMAP 8 /* don't use mmap */
#define TDB_CONVERT 16 /* convert endian (internal use) */
#define TDB_BIGENDIAN 32 /* header is big-endian (internal use) */
-
-#define TDB_ERRCODE(code, ret) ((tdb->ecode = (code)), ret)
+#define TDB_NOSYNC 64 /* don't use synchronous transactions */
+#define TDB_SEQNUM 128 /* maintain a sequence number */
+#define TDB_VOLATILE 256 /* Activate the per-hashchain freelist, default 5 */
+#define TDB_ALLOW_NESTING 512 /* Allow transactions to nest */
+#define TDB_DISALLOW_NESTING 1024 /* Disallow transactions to nest */
/* error codes */
enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK,
- TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOEXIST, TDB_ERR_NOLOCK, TDB_ERR_LOCK_TIMEOUT };
+ TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOLOCK, TDB_ERR_LOCK_TIMEOUT,
+ TDB_ERR_NOEXIST, TDB_ERR_EINVAL, TDB_ERR_RDONLY,
+ TDB_ERR_NESTING};
-#ifndef u32
-#define u32 unsigned
-#endif
+/* debugging uses one of the following levels */
+enum tdb_debug_level {TDB_DEBUG_FATAL = 0, TDB_DEBUG_ERROR,
+ TDB_DEBUG_WARNING, TDB_DEBUG_TRACE};
-typedef struct {
- char *dptr;
+typedef struct TDB_DATA {
+ unsigned char *dptr;
size_t dsize;
} TDB_DATA;
-typedef u32 tdb_len;
-typedef u32 tdb_off;
+#ifndef PRINTF_ATTRIBUTE
+#if (__GNUC__ >= 3)
+/** Use gcc attribute to check printf fns. a1 is the 1-based index of
+ * the parameter containing the format, and a2 the index of the first
+ * argument. Note that some gcc 2.x versions don't handle this
+ * properly **/
+#define PRINTF_ATTRIBUTE(a1, a2) __attribute__ ((format (__printf__, a1, a2)))
+#else
+#define PRINTF_ATTRIBUTE(a1, a2)
+#endif
+#endif
-/* this is stored at the front of every database */
-struct tdb_header {
- char magic_food[32]; /* for /etc/magic */
- u32 version; /* version of the code */
- u32 hash_size; /* number of hash entries */
- tdb_off rwlocks;
- tdb_off reserved[31];
-};
+/* this is the context structure that is returned from a db open */
+typedef struct tdb_context TDB_CONTEXT;
-struct tdb_lock_type {
- u32 count;
- u32 ltype;
-};
+typedef int (*tdb_traverse_func)(struct tdb_context *, TDB_DATA, TDB_DATA, void *);
+typedef void (*tdb_log_func)(struct tdb_context *, enum tdb_debug_level, const char *, ...) PRINTF_ATTRIBUTE(3, 4);
+typedef unsigned int (*tdb_hash_func)(TDB_DATA *key);
-struct tdb_traverse_lock {
- struct tdb_traverse_lock *next;
- u32 off;
- u32 hash;
+struct tdb_logging_context {
+ tdb_log_func log_fn;
+ void *log_private;
};
-/* this is the context structure that is returned from a db open */
-typedef struct tdb_context {
- char *name; /* the name of the database */
- void *map_ptr; /* where it is currently mapped */
- int fd; /* open file descriptor for the database */
- tdb_len map_size; /* how much space has been mapped */
- int read_only; /* opened read-only */
- struct tdb_lock_type *locked; /* array of chain locks */
- enum TDB_ERROR ecode; /* error code for last tdb error */
- struct tdb_header header; /* a cached copy of the header */
- u32 flags; /* the flags passed to tdb_open */
- u32 *lockedkeys; /* array of locked keys: first is #keys */
- struct tdb_traverse_lock travlocks; /* current traversal locks */
- struct tdb_context *next; /* all tdbs to avoid multiple opens */
- dev_t device; /* uniquely identifies this tdb */
- ino_t inode; /* uniquely identifies this tdb */
- void (*log_fn)(struct tdb_context *tdb, int level, const char *, ...); /* logging function */
- int open_flags; /* flags used in the open - needed by reopen */
-} TDB_CONTEXT;
-
-typedef int (*tdb_traverse_func)(TDB_CONTEXT *, TDB_DATA, TDB_DATA, void *);
-typedef void (*tdb_log_func)(TDB_CONTEXT *, int , const char *, ...);
-
-TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
+struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode);
-TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode,
- tdb_log_func log_fn);
-
-int tdb_reopen(TDB_CONTEXT *tdb);
-int tdb_reopen_all(void);
-void tdb_logging_function(TDB_CONTEXT *tdb, tdb_log_func);
-enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb);
-const char *tdb_errorstr(TDB_CONTEXT *tdb);
-TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
-int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf);
-int tdb_close(TDB_CONTEXT *tdb);
-TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb);
-TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *state);
-int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[]);
-void tdb_unlockkeys(TDB_CONTEXT *tdb);
-int tdb_lockall(TDB_CONTEXT *tdb);
-void tdb_unlockall(TDB_CONTEXT *tdb);
+ const struct tdb_logging_context *log_ctx,
+ tdb_hash_func hash_fn);
+void tdb_set_max_dead(struct tdb_context *tdb, int max_dead);
+
+int tdb_reopen(struct tdb_context *tdb);
+int tdb_reopen_all(int parent_longlived);
+void tdb_set_logging_function(struct tdb_context *tdb, const struct tdb_logging_context *log_ctx);
+enum TDB_ERROR tdb_error(struct tdb_context *tdb);
+const char *tdb_errorstr(struct tdb_context *tdb);
+TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
+int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
+ int (*parser)(TDB_DATA key, TDB_DATA data,
+ void *private_data),
+ void *private_data);
+int tdb_delete(struct tdb_context *tdb, TDB_DATA key);
+int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
+int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf);
+int tdb_close(struct tdb_context *tdb);
+TDB_DATA tdb_firstkey(struct tdb_context *tdb);
+TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA key);
+int tdb_traverse(struct tdb_context *tdb, tdb_traverse_func fn, void *);
+int tdb_traverse_read(struct tdb_context *tdb, tdb_traverse_func fn, void *);
+int tdb_exists(struct tdb_context *tdb, TDB_DATA key);
+int tdb_lockall(struct tdb_context *tdb);
+int tdb_lockall_nonblock(struct tdb_context *tdb);
+int tdb_unlockall(struct tdb_context *tdb);
+int tdb_lockall_read(struct tdb_context *tdb);
+int tdb_lockall_read_nonblock(struct tdb_context *tdb);
+int tdb_unlockall_read(struct tdb_context *tdb);
+int tdb_lockall_mark(struct tdb_context *tdb);
+int tdb_lockall_unmark(struct tdb_context *tdb);
+const char *tdb_name(struct tdb_context *tdb);
+int tdb_fd(struct tdb_context *tdb);
+tdb_log_func tdb_log_fn(struct tdb_context *tdb);
+void *tdb_get_logging_private(struct tdb_context *tdb);
+int tdb_transaction_start(struct tdb_context *tdb);
+int tdb_transaction_prepare_commit(struct tdb_context *tdb);
+int tdb_transaction_commit(struct tdb_context *tdb);
+int tdb_transaction_cancel(struct tdb_context *tdb);
+int tdb_transaction_recover(struct tdb_context *tdb);
+int tdb_get_seqnum(struct tdb_context *tdb);
+int tdb_hash_size(struct tdb_context *tdb);
+size_t tdb_map_size(struct tdb_context *tdb);
+int tdb_get_flags(struct tdb_context *tdb);
+void tdb_add_flags(struct tdb_context *tdb, unsigned flag);
+void tdb_remove_flags(struct tdb_context *tdb, unsigned flag);
+void tdb_enable_seqnum(struct tdb_context *tdb);
+void tdb_increment_seqnum_nonblock(struct tdb_context *tdb);
+int tdb_check(struct tdb_context *tdb,
+ int (*check) (TDB_DATA key, TDB_DATA data, void *private_data),
+ void *private_data);
/* Low level locking functions: use with care */
-void tdb_set_lock_alarm(sig_atomic_t *palarm);
-int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key);
-int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key);
+int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key);
+
+void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *sigptr);
+
+/* wipe and repack */
+int tdb_wipe_all(struct tdb_context *tdb);
+int tdb_repack(struct tdb_context *tdb);
/* Debug functions. Not used in production. */
-void tdb_dump_all(TDB_CONTEXT *tdb);
-int tdb_printfreelist(TDB_CONTEXT *tdb);
+void tdb_dump_all(struct tdb_context *tdb);
+int tdb_printfreelist(struct tdb_context *tdb);
+int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries);
+int tdb_freelist_size(struct tdb_context *tdb);
extern TDB_DATA tdb_null;
# Makefile.am for libatalk/tdb/
noinst_LTLIBRARIES = libtdb.la
-
-libtdb_la_SOURCES = tdb.c \
- spinlock.c \
- spinlock.h
+noinst_HEADERS = tdb_private.h
+libtdb_la_SOURCES = check.c \
+ dump.c \
+ error.c \
+ freelist.c \
+ freelistcheck.c \
+ io.c \
+ lock.c \
+ open.c \
+ tdb.c \
+ transaction.c \
+ traverse.c
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Rusty Russell 2009
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+
+/* Since we opened it, these shouldn't fail unless it's recent corruption. */
+static bool tdb_check_header(struct tdb_context *tdb, tdb_off_t *recovery)
+{
+ struct tdb_header hdr;
+
+ if (tdb->methods->tdb_read(tdb, 0, &hdr, sizeof(hdr), DOCONV()) == -1)
+ return false;
+ if (strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0)
+ goto corrupt;
+
+ CONVERT(hdr);
+ if (hdr.version != TDB_VERSION)
+ goto corrupt;
+
+ if (hdr.rwlocks != 0)
+ goto corrupt;
+
+ if (hdr.hash_size == 0)
+ goto corrupt;
+
+ if (hdr.hash_size != tdb->header.hash_size)
+ goto corrupt;
+
+ if (hdr.recovery_start != 0 &&
+ hdr.recovery_start < TDB_DATA_START(tdb->header.hash_size))
+ goto corrupt;
+
+ *recovery = hdr.recovery_start;
+ return true;
+
+corrupt:
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "Header is corrupt\n"));
+ return false;
+}
+
+/* Generic record header check. */
+static bool tdb_check_record(struct tdb_context *tdb,
+ tdb_off_t off,
+ const struct tdb_record *rec)
+{
+ tdb_off_t tailer;
+
+ /* Check rec->next: 0 or points to record offset, aligned. */
+ if (rec->next > 0 && rec->next < TDB_DATA_START(tdb->header.hash_size)){
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d too small next %d\n",
+ off, rec->next));
+ goto corrupt;
+ }
+ if (rec->next + sizeof(*rec) < rec->next) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d too large next %d\n",
+ off, rec->next));
+ goto corrupt;
+ }
+ if ((rec->next % TDB_ALIGNMENT) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d misaligned next %d\n",
+ off, rec->next));
+ goto corrupt;
+ }
+ if (tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0))
+ goto corrupt;
+
+ /* Check rec_len: similar to rec->next, implies next record. */
+ if ((rec->rec_len % TDB_ALIGNMENT) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d misaligned length %d\n",
+ off, rec->rec_len));
+ goto corrupt;
+ }
+ /* Must fit tailer. */
+ if (rec->rec_len < sizeof(tailer)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d too short length %d\n",
+ off, rec->rec_len));
+ goto corrupt;
+ }
+ /* OOB allows "right at the end" access, so this works for last rec. */
+ if (tdb->methods->tdb_oob(tdb, off+sizeof(*rec)+rec->rec_len, 0))
+ goto corrupt;
+
+ /* Check tailer. */
+ if (tdb_ofs_read(tdb, off+sizeof(*rec)+rec->rec_len-sizeof(tailer),
+ &tailer) == -1)
+ goto corrupt;
+ if (tailer != sizeof(*rec) + rec->rec_len) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d invalid tailer\n", off));
+ goto corrupt;
+ }
+
+ return true;
+
+corrupt:
+ tdb->ecode = TDB_ERR_CORRUPT;
+ return false;
+}
+
+/* Grab some bytes: may copy if can't use mmap.
+ Caller has already done bounds check. */
+static TDB_DATA get_bytes(struct tdb_context *tdb,
+ tdb_off_t off, tdb_len_t len)
+{
+ TDB_DATA d;
+
+ d.dsize = len;
+
+ if (tdb->transaction == NULL && tdb->map_ptr != NULL)
+ d.dptr = (unsigned char *)tdb->map_ptr + off;
+ else
+ d.dptr = tdb_alloc_read(tdb, off, d.dsize);
+ return d;
+}
+
+/* Frees data if we're not able to simply use mmap. */
+static void put_bytes(struct tdb_context *tdb, TDB_DATA d)
+{
+ if (tdb->transaction == NULL && tdb->map_ptr != NULL)
+ return;
+ free(d.dptr);
+}
+
+/* We use the excellent Jenkins lookup3 hash; this is based on hash_word2.
+ * See: http://burtleburtle.net/bob/c/lookup3.c
+ */
+#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
+static void hash(uint32_t key, uint32_t *pc, uint32_t *pb)
+{
+ uint32_t a,b,c;
+
+ /* Set up the internal state */
+ a = b = c = 0xdeadbeef + *pc;
+ c += *pb;
+ a += key;
+ c ^= b; c -= rot(b,14);
+ a ^= c; a -= rot(c,11);
+ b ^= a; b -= rot(a,25);
+ c ^= b; c -= rot(b,16);
+ a ^= c; a -= rot(c,4);
+ b ^= a; b -= rot(a,14);
+ c ^= b; c -= rot(b,24);
+ *pc=c; *pb=b;
+}
+
+/*
+ We want to check that all free records are in the free list
+ (only once), and all free list entries are free records. Similarly
+ for each hash chain of used records.
+
+ Doing that naively (without walking hash chains, since we want to be
+ linear) means keeping a list of records which have been seen in each
+ hash chain, and another of records pointed to (ie. next pointers
+ from records and the initial hash chain heads). These two lists
+ should be equal. This will take 8 bytes per record, and require
+ sorting at the end.
+
+ So instead, we record each offset in a bitmap such a way that
+ recording it twice will cancel out. Since each offset should appear
+ exactly twice, the bitmap should be zero at the end.
+
+ The approach was inspired by Bloom Filters (see Wikipedia). For
+ each value, we flip K bits in a bitmap of size N. The number of
+ distinct arrangements is:
+
+ N! / (K! * (N-K)!)
+
+ Of course, not all arrangements are actually distinct, but testing
+ shows this formula to be close enough.
+
+ So, if K == 8 and N == 256, the probability of two things flipping the same
+ bits is 1 in 409,663,695,276,000.
+
+ Given that ldb uses a hash size of 10000, using 32 bytes per hash chain
+ (320k) seems reasonable.
+*/
+#define NUM_HASHES 8
+#define BITMAP_BITS 256
+
+static void bit_flip(unsigned char bits[], unsigned int idx)
+{
+ bits[idx / CHAR_BIT] ^= (1 << (idx % CHAR_BIT));
+}
+
+/* We record offsets in a bitmap for the particular chain it should be in. */
+static void record_offset(unsigned char bits[], tdb_off_t off)
+{
+ uint32_t h1 = off, h2 = 0;
+ unsigned int i;
+
+ /* We get two good hash values out of jhash2, so we use both. Then
+ * we keep going to produce further hash values. */
+ for (i = 0; i < NUM_HASHES / 2; i++) {
+ hash(off, &h1, &h2);
+ bit_flip(bits, h1 % BITMAP_BITS);
+ bit_flip(bits, h2 % BITMAP_BITS);
+ h2++;
+ }
+}
+
+/* Check that an in-use record is valid. */
+static bool tdb_check_used_record(struct tdb_context *tdb,
+ tdb_off_t off,
+ const struct tdb_record *rec,
+ unsigned char **hashes,
+ int (*check)(TDB_DATA, TDB_DATA, void *),
+ void *private_data)
+{
+ TDB_DATA key, data;
+
+ if (!tdb_check_record(tdb, off, rec))
+ return false;
+
+ /* key + data + tailer must fit in record */
+ if (rec->key_len + rec->data_len + sizeof(tdb_off_t) > rec->rec_len) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d too short for contents\n", off));
+ return false;
+ }
+
+ key = get_bytes(tdb, off + sizeof(*rec), rec->key_len);
+ if (!key.dptr)
+ return false;
+
+ if (tdb->hash_fn(&key) != rec->full_hash) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Record offset %d has incorrect hash\n", off));
+ goto fail_put_key;
+ }
+
+ /* Mark this offset as a known value for this hash bucket. */
+ record_offset(hashes[BUCKET(rec->full_hash)+1], off);
+ /* And similarly if the next pointer is valid. */
+ if (rec->next)
+ record_offset(hashes[BUCKET(rec->full_hash)+1], rec->next);
+
+ /* If they supply a check function and this record isn't dead,
+ get data and feed it. */
+ if (check && rec->magic != TDB_DEAD_MAGIC) {
+ data = get_bytes(tdb, off + sizeof(*rec) + rec->key_len,
+ rec->data_len);
+ if (!data.dptr)
+ goto fail_put_key;
+
+ if (check(key, data, private_data) == -1)
+ goto fail_put_data;
+ put_bytes(tdb, data);
+ }
+
+ put_bytes(tdb, key);
+ return true;
+
+fail_put_data:
+ put_bytes(tdb, data);
+fail_put_key:
+ put_bytes(tdb, key);
+ return false;
+}
+
+/* Check that an unused record is valid. */
+static bool tdb_check_free_record(struct tdb_context *tdb,
+ tdb_off_t off,
+ const struct tdb_record *rec,
+ unsigned char **hashes)
+{
+ if (!tdb_check_record(tdb, off, rec))
+ return false;
+
+ /* Mark this offset as a known value for the free list. */
+ record_offset(hashes[0], off);
+ /* And similarly if the next pointer is valid. */
+ if (rec->next)
+ record_offset(hashes[0], rec->next);
+ return true;
+}
+
+int tdb_check(struct tdb_context *tdb,
+ int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
+ void *private_data)
+{
+ unsigned int h;
+ unsigned char **hashes;
+ tdb_off_t off, recovery_start;
+ struct tdb_record rec;
+ bool found_recovery = false;
+
+ if (tdb_lockall(tdb) == -1)
+ return -1;
+
+ /* Make sure we know true size of the underlying file. */
+ tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+
+ /* Header must be OK: also gets us the recovery ptr, if any. */
+ if (!tdb_check_header(tdb, &recovery_start))
+ goto unlock;
+
+ /* We should have the whole header, too. */
+ if (tdb->map_size < TDB_DATA_START(tdb->header.hash_size)) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "File too short for hashes\n"));
+ goto unlock;
+ }
+
+ /* One big malloc: pointers then bit arrays. */
+ hashes = (unsigned char **)calloc(
+ 1, sizeof(hashes[0]) * (1+tdb->header.hash_size)
+ + BITMAP_BITS / CHAR_BIT * (1+tdb->header.hash_size));
+ if (!hashes) {
+ tdb->ecode = TDB_ERR_OOM;
+ goto unlock;
+ }
+
+ /* Initialize pointers */
+ hashes[0] = (unsigned char *)(&hashes[1+tdb->header.hash_size]);
+ for (h = 1; h < 1+tdb->header.hash_size; h++)
+ hashes[h] = hashes[h-1] + BITMAP_BITS / CHAR_BIT;
+
+ /* Freelist and hash headers are all in a row: read them. */
+ for (h = 0; h < 1+tdb->header.hash_size; h++) {
+ if (tdb_ofs_read(tdb, FREELIST_TOP + h*sizeof(tdb_off_t),
+ &off) == -1)
+ goto free;
+ if (off)
+ record_offset(hashes[h], off);
+ }
+
+ /* For each record, read it in and check it's ok. */
+ for (off = TDB_DATA_START(tdb->header.hash_size);
+ off < tdb->map_size;
+ off += sizeof(rec) + rec.rec_len) {
+ if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+ DOCONV()) == -1)
+ goto free;
+ switch (rec.magic) {
+ case TDB_MAGIC:
+ case TDB_DEAD_MAGIC:
+ if (!tdb_check_used_record(tdb, off, &rec, hashes,
+ check, private_data))
+ goto free;
+ break;
+ case TDB_FREE_MAGIC:
+ if (!tdb_check_free_record(tdb, off, &rec, hashes))
+ goto free;
+ break;
+ case TDB_RECOVERY_MAGIC:
+ case 0: /* Used for invalid (or in-progress) recovery area. */
+ if (recovery_start != off) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Unexpected recovery record at offset %d\n",
+ off));
+ goto free;
+ }
+ found_recovery = true;
+ break;
+ default:
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Bad magic 0x%x at offset %d\n",
+ rec.magic, off));
+ goto free;
+ }
+ }
+
+ /* Now, hashes should all be empty: each record exists and is referred
+ * to by one other. */
+ for (h = 0; h < 1+tdb->header.hash_size; h++) {
+ unsigned int i;
+ for (i = 0; i < BITMAP_BITS / CHAR_BIT; i++) {
+ if (hashes[h][i] != 0) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Hashes do not match records\n"));
+ goto free;
+ }
+ }
+ }
+
+ /* We must have found recovery area if there was one. */
+ if (recovery_start != 0 && !found_recovery) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,
+ "Expected %s recovery area, got %s\n",
+ recovery_start ? "a" : "no",
+ found_recovery ? "one" : "none"));
+ goto free;
+ }
+
+ free(hashes);
+ tdb_unlockall(tdb);
+ return 0;
+
+free:
+ free(hashes);
+unlock:
+ tdb_unlockall(tdb);
+ return -1;
+}
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+static tdb_off_t tdb_dump_record(struct tdb_context *tdb, int hash,
+ tdb_off_t offset)
+{
+ struct tdb_record rec;
+ tdb_off_t tailer_ofs, tailer;
+
+ if (tdb->methods->tdb_read(tdb, offset, (char *)&rec,
+ sizeof(rec), DOCONV()) == -1) {
+ printf("ERROR: failed to read record at %u\n", offset);
+ return 0;
+ }
+
+ printf(" rec: hash=%d offset=0x%08x next=0x%08x rec_len=%d "
+ "key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
+ hash, offset, rec.next, rec.rec_len, rec.key_len, rec.data_len,
+ rec.full_hash, rec.magic);
+
+ tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off_t);
+
+ if (tdb_ofs_read(tdb, tailer_ofs, &tailer) == -1) {
+ printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
+ return rec.next;
+ }
+
+ if (tailer != rec.rec_len + sizeof(rec)) {
+ printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
+ (unsigned int)tailer, (unsigned int)(rec.rec_len + sizeof(rec)));
+ }
+ return rec.next;
+}
+
+static int tdb_dump_chain(struct tdb_context *tdb, int i)
+{
+ tdb_off_t rec_ptr, top;
+
+ top = TDB_HASH_TOP(i);
+
+ if (tdb_lock(tdb, i, F_WRLCK) != 0)
+ return -1;
+
+ if (tdb_ofs_read(tdb, top, &rec_ptr) == -1)
+ return tdb_unlock(tdb, i, F_WRLCK);
+
+ if (rec_ptr)
+ printf("hash=%d\n", i);
+
+ while (rec_ptr) {
+ rec_ptr = tdb_dump_record(tdb, i, rec_ptr);
+ }
+
+ return tdb_unlock(tdb, i, F_WRLCK);
+}
+
+void tdb_dump_all(struct tdb_context *tdb)
+{
+ int i;
+ for (i=0;i<tdb->header.hash_size;i++) {
+ tdb_dump_chain(tdb, i);
+ }
+ printf("freelist:\n");
+ tdb_dump_chain(tdb, -1);
+}
+
+int tdb_printfreelist(struct tdb_context *tdb)
+{
+ int ret;
+ long total_free = 0;
+ tdb_off_t offset, rec_ptr;
+ struct tdb_record rec;
+
+ if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
+ return ret;
+
+ offset = FREELIST_TOP;
+
+ /* read in the freelist top */
+ if (tdb_ofs_read(tdb, offset, &rec_ptr) == -1) {
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return 0;
+ }
+
+ printf("freelist top=[0x%08x]\n", rec_ptr );
+ while (rec_ptr) {
+ if (tdb->methods->tdb_read(tdb, rec_ptr, (char *)&rec,
+ sizeof(rec), DOCONV()) == -1) {
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return -1;
+ }
+
+ if (rec.magic != TDB_FREE_MAGIC) {
+ printf("bad magic 0x%08x in free list\n", rec.magic);
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return -1;
+ }
+
+ printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)] (end = 0x%08x)\n",
+ rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + rec.rec_len);
+ total_free += rec.rec_len;
+
+ /* move to the next record */
+ rec_ptr = rec.next;
+ }
+ printf("total rec_len = [0x%08x (%d)]\n", (int)total_free,
+ (int)total_free);
+
+ return tdb_unlock(tdb, -1, F_WRLCK);
+}
+
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+{
+ return tdb->ecode;
+}
+
+static struct tdb_errname {
+ enum TDB_ERROR ecode; const char *estring;
+} emap[] = { {TDB_SUCCESS, "Success"},
+ {TDB_ERR_CORRUPT, "Corrupt database"},
+ {TDB_ERR_IO, "IO Error"},
+ {TDB_ERR_LOCK, "Locking error"},
+ {TDB_ERR_OOM, "Out of memory"},
+ {TDB_ERR_EXISTS, "Record exists"},
+ {TDB_ERR_NOLOCK, "Lock exists on other keys"},
+ {TDB_ERR_EINVAL, "Invalid parameter"},
+ {TDB_ERR_NOEXIST, "Record does not exist"},
+ {TDB_ERR_RDONLY, "write not permitted"} };
+
+/* Error string for the last tdb error */
+const char *tdb_errorstr(struct tdb_context *tdb)
+{
+ uint32_t i;
+ for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
+ if (tdb->ecode == emap[i].ecode)
+ return emap[i].estring;
+ return "Invalid error code";
+}
+
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* 'right' merges can involve O(n^2) cost when combined with a
+ traverse, so they are disabled until we find a way to do them in
+ O(1) time
+*/
+#define USE_RIGHT_MERGES 0
+
+/* read a freelist record and check for simple errors */
+int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct tdb_record *rec)
+{
+ if (tdb->methods->tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
+ return -1;
+
+ if (rec->magic == TDB_MAGIC) {
+ /* this happens when a app is showdown while deleting a record - we should
+ not completely fail when this happens */
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read non-free magic 0x%x at offset=%d - fixing\n",
+ rec->magic, off));
+ rec->magic = TDB_FREE_MAGIC;
+ if (tdb->methods->tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
+ return -1;
+ }
+
+ if (rec->magic != TDB_FREE_MAGIC) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read bad magic 0x%x at offset=%d\n",
+ rec->magic, off));
+ return -1;
+ }
+ if (tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
+ return -1;
+ return 0;
+}
+
+
+#if USE_RIGHT_MERGES
+/* Remove an element from the freelist. Must have alloc lock. */
+static int remove_from_freelist(struct tdb_context *tdb, tdb_off_t off, tdb_off_t next)
+{
+ tdb_off_t last_ptr, i;
+
+ /* read in the freelist top */
+ last_ptr = FREELIST_TOP;
+ while (tdb_ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
+ if (i == off) {
+ /* We've found it! */
+ return tdb_ofs_write(tdb, last_ptr, &next);
+ }
+ /* Follow chain (next offset is at start of record) */
+ last_ptr = i;
+ }
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"remove_from_freelist: not on list at off=%d\n", off));
+ return -1;
+}
+#endif
+
+
+/* update a record tailer (must hold allocation lock) */
+static int update_tailer(struct tdb_context *tdb, tdb_off_t offset,
+ const struct tdb_record *rec)
+{
+ tdb_off_t totalsize;
+
+ /* Offset of tailer from record header */
+ totalsize = sizeof(*rec) + rec->rec_len;
+ return tdb_ofs_write(tdb, offset + totalsize - sizeof(tdb_off_t),
+ &totalsize);
+}
+
+/* Add an element into the freelist. Merge adjacent records if
+ neccessary. */
+int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+ /* Allocation and tailer lock */
+ if (tdb_lock(tdb, -1, F_WRLCK) != 0)
+ return -1;
+
+ /* set an initial tailer, so if we fail we don't leave a bogus record */
+ if (update_tailer(tdb, offset, rec) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed!\n"));
+ goto fail;
+ }
+
+#if USE_RIGHT_MERGES
+ /* Look right first (I'm an Australian, dammit) */
+ if (offset + sizeof(*rec) + rec->rec_len + sizeof(*rec) <= tdb->map_size) {
+ tdb_off_t right = offset + sizeof(*rec) + rec->rec_len;
+ struct tdb_record r;
+
+ if (tdb->methods->tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: right read failed at %u\n", right));
+ goto left;
+ }
+
+ /* If it's free, expand to include it. */
+ if (r.magic == TDB_FREE_MAGIC) {
+ if (remove_from_freelist(tdb, right, r.next) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: right free failed at %u\n", right));
+ goto left;
+ }
+ rec->rec_len += sizeof(r) + r.rec_len;
+ if (update_tailer(tdb, offset, rec) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
+ goto fail;
+ }
+ }
+ }
+left:
+#endif
+
+ /* Look left */
+ if (offset - sizeof(tdb_off_t) > TDB_DATA_START(tdb->header.hash_size)) {
+ tdb_off_t left = offset - sizeof(tdb_off_t);
+ struct tdb_record l;
+ tdb_off_t leftsize;
+
+ /* Read in tailer and jump back to header */
+ if (tdb_ofs_read(tdb, left, &leftsize) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left offset read failed at %u\n", left));
+ goto update;
+ }
+
+ /* it could be uninitialised data */
+ if (leftsize == 0 || leftsize == TDB_PAD_U32) {
+ goto update;
+ }
+
+ left = offset - leftsize;
+
+ if (leftsize > offset ||
+ left < TDB_DATA_START(tdb->header.hash_size)) {
+ goto update;
+ }
+
+ /* Now read in the left record */
+ if (tdb->methods->tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
+ goto update;
+ }
+
+ /* If it's free, expand to include it. */
+ if (l.magic == TDB_FREE_MAGIC) {
+ /* we now merge the new record into the left record, rather than the other
+ way around. This makes the operation O(1) instead of O(n). This change
+ prevents traverse from being O(n^2) after a lot of deletes */
+ l.rec_len += sizeof(*rec) + rec->rec_len;
+ if (tdb_rec_write(tdb, left, &l) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_left failed at %u\n", left));
+ goto fail;
+ }
+ if (update_tailer(tdb, left, &l) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
+ goto fail;
+ }
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return 0;
+ }
+ }
+
+update:
+
+ /* Now, prepend to free list */
+ rec->magic = TDB_FREE_MAGIC;
+
+ if (tdb_ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
+ tdb_rec_write(tdb, offset, rec) == -1 ||
+ tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free record write failed at offset=%d\n", offset));
+ goto fail;
+ }
+
+ /* And we're done. */
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return 0;
+
+ fail:
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return -1;
+}
+
+
+
+/*
+ the core of tdb_allocate - called when we have decided which
+ free list entry to use
+
+ Note that we try to allocate by grabbing data from the end of an existing record,
+ not the beginning. This is so the left merge in a free is more likely to be
+ able to free up the record without fragmentation
+ */
+static tdb_off_t tdb_allocate_ofs(struct tdb_context *tdb,
+ tdb_len_t length, tdb_off_t rec_ptr,
+ struct tdb_record *rec, tdb_off_t last_ptr)
+{
+#define MIN_REC_SIZE (sizeof(struct tdb_record) + sizeof(tdb_off_t) + 8)
+
+ if (rec->rec_len < length + MIN_REC_SIZE) {
+ /* we have to grab the whole record */
+
+ /* unlink it from the previous record */
+ if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1) {
+ return 0;
+ }
+
+ /* mark it not free */
+ rec->magic = TDB_MAGIC;
+ if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+ return 0;
+ }
+ return rec_ptr;
+ }
+
+ /* we're going to just shorten the existing record */
+ rec->rec_len -= (length + sizeof(*rec));
+ if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+ return 0;
+ }
+ if (update_tailer(tdb, rec_ptr, rec) == -1) {
+ return 0;
+ }
+
+ /* and setup the new record */
+ rec_ptr += sizeof(*rec) + rec->rec_len;
+
+ memset(rec, '\0', sizeof(*rec));
+ rec->rec_len = length;
+ rec->magic = TDB_MAGIC;
+
+ if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+ return 0;
+ }
+
+ if (update_tailer(tdb, rec_ptr, rec) == -1) {
+ return 0;
+ }
+
+ return rec_ptr;
+}
+
+/* allocate some space from the free list. The offset returned points
+ to a unconnected tdb_record within the database with room for at
+ least length bytes of total data
+
+ 0 is returned if the space could not be allocated
+ */
+tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec)
+{
+ tdb_off_t rec_ptr, last_ptr, newrec_ptr;
+ struct {
+ tdb_off_t rec_ptr, last_ptr;
+ tdb_len_t rec_len;
+ } bestfit;
+ float multiplier = 1.0;
+
+ if (tdb_lock(tdb, -1, F_WRLCK) == -1)
+ return 0;
+
+ /* over-allocate to reduce fragmentation */
+ length *= 1.25;
+
+ /* Extra bytes required for tailer */
+ length += sizeof(tdb_off_t);
+ length = TDB_ALIGN(length, TDB_ALIGNMENT);
+
+ again:
+ last_ptr = FREELIST_TOP;
+
+ /* read in the freelist top */
+ if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
+ goto fail;
+
+ bestfit.rec_ptr = 0;
+ bestfit.last_ptr = 0;
+ bestfit.rec_len = 0;
+
+ /*
+ this is a best fit allocation strategy. Originally we used
+ a first fit strategy, but it suffered from massive fragmentation
+ issues when faced with a slowly increasing record size.
+ */
+ while (rec_ptr) {
+ if (tdb_rec_free_read(tdb, rec_ptr, rec) == -1) {
+ goto fail;
+ }
+
+ if (rec->rec_len >= length) {
+ if (bestfit.rec_ptr == 0 ||
+ rec->rec_len < bestfit.rec_len) {
+ bestfit.rec_len = rec->rec_len;
+ bestfit.rec_ptr = rec_ptr;
+ bestfit.last_ptr = last_ptr;
+ }
+ }
+
+ /* move to the next record */
+ last_ptr = rec_ptr;
+ rec_ptr = rec->next;
+
+ /* if we've found a record that is big enough, then
+ stop searching if its also not too big. The
+ definition of 'too big' changes as we scan
+ through */
+ if (bestfit.rec_len > 0 &&
+ bestfit.rec_len < length * multiplier) {
+ break;
+ }
+
+ /* this multiplier means we only extremely rarely
+ search more than 50 or so records. At 50 records we
+ accept records up to 11 times larger than what we
+ want */
+ multiplier *= 1.05;
+ }
+
+ if (bestfit.rec_ptr != 0) {
+ if (tdb_rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
+ goto fail;
+ }
+
+ newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr,
+ rec, bestfit.last_ptr);
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return newrec_ptr;
+ }
+
+ /* we didn't find enough space. See if we can expand the
+ database and if we can then try again */
+ if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
+ goto again;
+ fail:
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return 0;
+}
+
+
+
+/*
+ return the size of the freelist - used to decide if we should repack
+*/
+int tdb_freelist_size(struct tdb_context *tdb)
+{
+ tdb_off_t ptr;
+ int count=0;
+
+ if (tdb_lock(tdb, -1, F_RDLCK) == -1) {
+ return -1;
+ }
+
+ ptr = FREELIST_TOP;
+ while (tdb_ofs_read(tdb, ptr, &ptr) == 0 && ptr != 0) {
+ count++;
+ }
+
+ tdb_unlock(tdb, -1, F_RDLCK);
+ return count;
+}
--- /dev/null
+/*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Jeremy Allison 2006
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* Check the freelist is good and contains no loops.
+ Very memory intensive - only do this as a consistency
+ checker. Heh heh - uses an in memory tdb as the storage
+ for the "seen" record list. For some reason this strikes
+ me as extremely clever as I don't have to write another tree
+ data structure implementation :-).
+ */
+
+static int seen_insert(struct tdb_context *mem_tdb, tdb_off_t rec_ptr)
+{
+ TDB_DATA key, data;
+
+ memset(&data, '\0', sizeof(data));
+ key.dptr = (unsigned char *)&rec_ptr;
+ key.dsize = sizeof(rec_ptr);
+ return tdb_store(mem_tdb, key, data, TDB_INSERT);
+}
+
+int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries)
+{
+ struct tdb_context *mem_tdb = NULL;
+ struct tdb_record rec;
+ tdb_off_t rec_ptr, last_ptr;
+ int ret = -1;
+
+ *pnum_entries = 0;
+
+ mem_tdb = tdb_open("flval", tdb->header.hash_size,
+ TDB_INTERNAL, O_RDWR, 0600);
+ if (!mem_tdb) {
+ return -1;
+ }
+
+ if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+ tdb_close(mem_tdb);
+ return 0;
+ }
+
+ last_ptr = FREELIST_TOP;
+
+ /* Store the FREELIST_TOP record. */
+ if (seen_insert(mem_tdb, last_ptr) == -1) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ ret = -1;
+ goto fail;
+ }
+
+ /* read in the freelist top */
+ if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1) {
+ goto fail;
+ }
+
+ while (rec_ptr) {
+
+ /* If we can't store this record (we've seen it
+ before) then the free list has a loop and must
+ be corrupt. */
+
+ if (seen_insert(mem_tdb, rec_ptr)) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ ret = -1;
+ goto fail;
+ }
+
+ if (tdb_rec_free_read(tdb, rec_ptr, &rec) == -1) {
+ goto fail;
+ }
+
+ /* move to the next record */
+ last_ptr = rec_ptr;
+ rec_ptr = rec.next;
+ *pnum_entries += 1;
+ }
+
+ ret = 0;
+
+ fail:
+
+ tdb_close(mem_tdb);
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return ret;
+}
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#include "tdb_private.h"
+
+/* check for an out of bounds access - if it is out of bounds then
+ see if the database has been expanded by someone else and expand
+ if necessary
+ note that "len" is the minimum length needed for the db
+*/
+static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
+{
+ struct stat st;
+ if (len <= tdb->map_size)
+ return 0;
+ if (tdb->flags & TDB_INTERNAL) {
+ if (!probe) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond internal malloc size %d\n",
+ (int)len, (int)tdb->map_size));
+ }
+ return -1;
+ }
+
+ if (fstat(tdb->fd, &st) == -1) {
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ if (st.st_size < (size_t)len) {
+ if (!probe) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond eof at %d\n",
+ (int)len, (int)st.st_size));
+ }
+ return -1;
+ }
+
+ /* Unmap, update size, remap */
+ if (tdb_munmap(tdb) == -1) {
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ tdb->map_size = st.st_size;
+ tdb_mmap(tdb);
+ return 0;
+}
+
+/* write a lump of data at a specified offset */
+static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
+ const void *buf, tdb_len_t len)
+{
+ if (len == 0) {
+ return 0;
+ }
+
+ if (tdb->read_only || tdb->traverse_read) {
+ tdb->ecode = TDB_ERR_RDONLY;
+ return -1;
+ }
+
+ if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0)
+ return -1;
+
+ if (tdb->map_ptr) {
+ memcpy(off + (char *)tdb->map_ptr, buf, len);
+ } else {
+ ssize_t written = pwrite(tdb->fd, buf, len, off);
+ if ((written != (ssize_t)len) && (written != -1)) {
+ /* try once more */
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
+ "%d of %d bytes at %d, trying once more\n",
+ (int)written, len, off));
+ written = pwrite(tdb->fd, (const char *)buf+written,
+ len-written,
+ off+written);
+ }
+ if (written == -1) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
+ "len=%d (%s)\n", off, len, strerror(errno)));
+ return -1;
+ } else if (written != (ssize_t)len) {
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
+ "write %d bytes at %d in two attempts\n",
+ len, off));
+ return -1;
+ }
+ }
+ return 0;
+}
+
+/* Endian conversion: we only ever deal with 4 byte quantities */
+void *tdb_convert(void *buf, uint32_t size)
+{
+ uint32_t i, *p = (uint32_t *)buf;
+ for (i = 0; i < size / 4; i++)
+ p[i] = TDB_BYTEREV(p[i]);
+ return buf;
+}
+
+
+/* read a lump of data at a specified offset, maybe convert */
+static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
+ tdb_len_t len, int cv)
+{
+ if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0) {
+ return -1;
+ }
+
+ if (tdb->map_ptr) {
+ memcpy(buf, off + (char *)tdb->map_ptr, len);
+ } else {
+ ssize_t ret = pread(tdb->fd, buf, len, off);
+ if (ret != (ssize_t)len) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_read failed at %d "
+ "len=%d ret=%d (%s) map_size=%d\n",
+ (int)off, (int)len, (int)ret, strerror(errno),
+ (int)tdb->map_size));
+ return -1;
+ }
+ }
+ if (cv) {
+ tdb_convert(buf, len);
+ }
+ return 0;
+}
+
+
+
+/*
+ do an unlocked scan of the hash table heads to find the next non-zero head. The value
+ will then be confirmed with the lock held
+*/
+static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
+{
+ uint32_t h = *chain;
+ if (tdb->map_ptr) {
+ for (;h < tdb->header.hash_size;h++) {
+ if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
+ break;
+ }
+ }
+ } else {
+ uint32_t off=0;
+ for (;h < tdb->header.hash_size;h++) {
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
+ break;
+ }
+ }
+ }
+ (*chain) = h;
+}
+
+
+int tdb_munmap(struct tdb_context *tdb)
+{
+ if (tdb->flags & TDB_INTERNAL)
+ return 0;
+
+#ifdef HAVE_MMAP
+ if (tdb->map_ptr) {
+ int ret;
+
+ ret = munmap(tdb->map_ptr, tdb->map_size);
+ if (ret != 0)
+ return ret;
+ }
+#endif
+ tdb->map_ptr = NULL;
+ return 0;
+}
+
+void tdb_mmap(struct tdb_context *tdb)
+{
+ if (tdb->flags & TDB_INTERNAL)
+ return;
+
+#ifdef HAVE_MMAP
+ if (!(tdb->flags & TDB_NOMMAP)) {
+ tdb->map_ptr = mmap(NULL, tdb->map_size,
+ PROT_READ|(tdb->read_only? 0:PROT_WRITE),
+ MAP_SHARED|MAP_FILE, tdb->fd, 0);
+
+ /*
+ * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
+ */
+
+ if (tdb->map_ptr == MAP_FAILED) {
+ tdb->map_ptr = NULL;
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_mmap failed for size %d (%s)\n",
+ tdb->map_size, strerror(errno)));
+ }
+ } else {
+ tdb->map_ptr = NULL;
+ }
+#else
+ tdb->map_ptr = NULL;
+#endif
+}
+
+/* expand a file. we prefer to use ftruncate, as that is what posix
+ says to use for mmap expansion */
+static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
+{
+ char buf[8192];
+
+ if (tdb->read_only || tdb->traverse_read) {
+ tdb->ecode = TDB_ERR_RDONLY;
+ return -1;
+ }
+
+ if (ftruncate(tdb->fd, size+addition) == -1) {
+ char b = 0;
+ ssize_t written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
+ if (written == 0) {
+ /* try once more, potentially revealing errno */
+ written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
+ }
+ if (written == 0) {
+ /* again - give up, guessing errno */
+ errno = ENOSPC;
+ }
+ if (written != 1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %d failed (%s)\n",
+ size+addition, strerror(errno)));
+ return -1;
+ }
+ }
+
+ /* now fill the file with something. This ensures that the
+ file isn't sparse, which would be very bad if we ran out of
+ disk. This must be done with write, not via mmap */
+ memset(buf, TDB_PAD_BYTE, sizeof(buf));
+ while (addition) {
+ size_t n = addition>sizeof(buf)?sizeof(buf):addition;
+ ssize_t written = pwrite(tdb->fd, buf, n, size);
+ if (written == 0) {
+ /* prevent infinite loops: try _once_ more */
+ written = pwrite(tdb->fd, buf, n, size);
+ }
+ if (written == 0) {
+ /* give up, trying to provide a useful errno */
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
+ "returned 0 twice: giving up!\n"));
+ errno = ENOSPC;
+ return -1;
+ } else if (written == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write of "
+ "%d bytes failed (%s)\n", (int)n,
+ strerror(errno)));
+ return -1;
+ } else if (written != n) {
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: wrote "
+ "only %d of %d bytes - retrying\n", (int)written,
+ (int)n));
+ }
+ addition -= written;
+ size += written;
+ }
+ return 0;
+}
+
+
+/* expand the database at least size bytes by expanding the underlying
+ file and doing the mmap again if necessary */
+int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
+{
+ struct tdb_record rec;
+ tdb_off_t offset, new_size;
+
+ if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
+ return -1;
+ }
+
+ /* must know about any previous expansions by another process */
+ tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+
+ /* always make room for at least 100 more records, and at
+ least 25% more space. Round the database up to a multiple
+ of the page size */
+ new_size = MAX(tdb->map_size + size*100, tdb->map_size * 1.25);
+ size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
+
+ if (!(tdb->flags & TDB_INTERNAL))
+ tdb_munmap(tdb);
+
+ /*
+ * We must ensure the file is unmapped before doing this
+ * to ensure consistency with systems like OpenBSD where
+ * writes and mmaps are not consistent.
+ */
+
+ /* expand the file itself */
+ if (!(tdb->flags & TDB_INTERNAL)) {
+ if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
+ goto fail;
+ }
+
+ tdb->map_size += size;
+
+ if (tdb->flags & TDB_INTERNAL) {
+ char *new_map_ptr = (char *)realloc(tdb->map_ptr,
+ tdb->map_size);
+ if (!new_map_ptr) {
+ tdb->map_size -= size;
+ goto fail;
+ }
+ tdb->map_ptr = new_map_ptr;
+ } else {
+ /*
+ * We must ensure the file is remapped before adding the space
+ * to ensure consistency with systems like OpenBSD where
+ * writes and mmaps are not consistent.
+ */
+
+ /* We're ok if the mmap fails as we'll fallback to read/write */
+ tdb_mmap(tdb);
+ }
+
+ /* form a new freelist record */
+ memset(&rec,'\0',sizeof(rec));
+ rec.rec_len = size - sizeof(rec);
+
+ /* link it into the free list */
+ offset = tdb->map_size - size;
+ if (tdb_free(tdb, offset, &rec) == -1)
+ goto fail;
+
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return 0;
+ fail:
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return -1;
+}
+
+/* read/write a tdb_off_t */
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+ return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
+}
+
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+ tdb_off_t off = *d;
+ return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
+}
+
+
+/* read a lump of data, allocating the space for it */
+unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
+{
+ unsigned char *buf;
+
+ /* some systems don't like zero length malloc */
+
+ if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_OOM;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_alloc_read malloc failed len=%d (%s)\n",
+ len, strerror(errno)));
+ return NULL;
+ }
+ if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
+ SAFE_FREE(buf);
+ return NULL;
+ }
+ return buf;
+}
+
+/* Give a piece of tdb data to a parser */
+
+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_DATA data;
+ int result;
+
+ data.dsize = len;
+
+ if ((tdb->transaction == NULL) && (tdb->map_ptr != NULL)) {
+ /*
+ * Optimize by avoiding the malloc/memcpy/free, point the
+ * parser directly at the mmap area.
+ */
+ if (tdb->methods->tdb_oob(tdb, offset+len, 0) != 0) {
+ return -1;
+ }
+ data.dptr = offset + (unsigned char *)tdb->map_ptr;
+ return parser(key, data, private_data);
+ }
+
+ if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
+ return -1;
+ }
+
+ result = parser(key, data, private_data);
+ free(data.dptr);
+ return result;
+}
+
+/* read/write a record */
+int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+ if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
+ return -1;
+ if (TDB_BAD_MAGIC(rec)) {
+ /* Ensure ecode is set for log fn. */
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
+ return -1;
+ }
+ return tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0);
+}
+
+int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+ struct tdb_record r = *rec;
+ return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
+}
+
+static const struct tdb_methods io_methods = {
+ tdb_read,
+ tdb_write,
+ tdb_next_hash_chain,
+ tdb_oob,
+ tdb_expand_file,
+ tdb_brlock
+};
+
+/*
+ initialise the default methods table
+*/
+void tdb_io_init(struct tdb_context *tdb)
+{
+ tdb->methods = &io_methods;
+}
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+#define TDB_MARK_LOCK 0x80000000
+
+void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
+{
+ tdb->interrupt_sig_ptr = ptr;
+}
+
+/* a byte range locking function - return 0 on success
+ this functions locks/unlocks 1 byte at the specified offset.
+
+ On error, errno is also set so that errors are passed back properly
+ through tdb_open().
+
+ note that a len of zero means lock to end of file
+*/
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
+ int rw_type, int lck_type, int probe, size_t len)
+{
+ struct flock fl;
+ int ret;
+
+ if (tdb->flags & TDB_NOLOCK) {
+ return 0;
+ }
+
+ if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
+ tdb->ecode = TDB_ERR_RDONLY;
+ return -1;
+ }
+
+ fl.l_type = rw_type;
+ fl.l_whence = SEEK_SET;
+ fl.l_start = offset;
+ fl.l_len = len;
+ fl.l_pid = 0;
+
+ do {
+ ret = fcntl(tdb->fd,lck_type,&fl);
+
+ /* Check for a sigalarm break. */
+ if (ret == -1 && errno == EINTR &&
+ tdb->interrupt_sig_ptr &&
+ *tdb->interrupt_sig_ptr) {
+ break;
+ }
+ } while (ret == -1 && errno == EINTR);
+
+ if (ret == -1) {
+ tdb->ecode = TDB_ERR_LOCK;
+ /* Generic lock error. errno set by fcntl.
+ * EAGAIN is an expected return from non-blocking
+ * locks. */
+ if (!probe && lck_type != F_SETLK) {
+ TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
+ tdb->fd, offset, rw_type, lck_type, (int)len));
+ }
+ return -1;
+ }
+ return 0;
+}
+
+
+/*
+ upgrade a read lock to a write lock. This needs to be handled in a
+ special way as some OSes (such as solaris) have too conservative
+ deadlock detection and claim a deadlock when progress can be
+ made. For those OSes we may loop for a while.
+*/
+int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
+{
+ int count = 1000;
+ while (count--) {
+ struct timeval tv;
+ if (tdb_brlock(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
+ return 0;
+ }
+ if (errno != EDEADLK) {
+ break;
+ }
+ /* sleep for as short a time as we can - more portable than usleep() */
+ tv.tv_sec = 0;
+ tv.tv_usec = 1;
+ select(0, NULL, NULL, NULL, &tv);
+ }
+ TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock_upgrade failed at offset %d\n", offset));
+ return -1;
+}
+
+
+/* lock a list in the database. list -1 is the alloc list */
+static int _tdb_lock(struct tdb_context *tdb, int list, int ltype, int op)
+{
+ struct tdb_lock_type *new_lck;
+ int i;
+ bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
+
+ ltype &= ~TDB_MARK_LOCK;
+
+ /* a global lock allows us to avoid per chain locks */
+ if (tdb->global_lock.count &&
+ (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (list < -1 || list >= (int)tdb->header.hash_size) {
+ tdb->ecode = TDB_ERR_LOCK;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
+ list, ltype));
+ return -1;
+ }
+ if (tdb->flags & TDB_NOLOCK)
+ return 0;
+
+ for (i=0; i<tdb->num_lockrecs; i++) {
+ if (tdb->lockrecs[i].list == list) {
+ if (tdb->lockrecs[i].count == 0) {
+ /*
+ * Can't happen, see tdb_unlock(). It should
+ * be an assert.
+ */
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
+ "lck->count == 0 for list %d", list));
+ }
+ /*
+ * Just increment the in-memory struct, posix locks
+ * don't stack.
+ */
+ tdb->lockrecs[i].count++;
+ return 0;
+ }
+ }
+
+ new_lck = (struct tdb_lock_type *)realloc(
+ tdb->lockrecs,
+ sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
+ if (new_lck == NULL) {
+ errno = ENOMEM;
+ return -1;
+ }
+ tdb->lockrecs = new_lck;
+
+ /* Since fcntl locks don't nest, we do a lock for the first one,
+ and simply bump the count for future ones */
+ if (!mark_lock &&
+ tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list, ltype, op,
+ 0, 1)) {
+ return -1;
+ }
+
+ tdb->num_locks++;
+
+ tdb->lockrecs[tdb->num_lockrecs].list = list;
+ tdb->lockrecs[tdb->num_lockrecs].count = 1;
+ tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
+ tdb->num_lockrecs += 1;
+
+ return 0;
+}
+
+/* lock a list in the database. list -1 is the alloc list */
+int tdb_lock(struct tdb_context *tdb, int list, int ltype)
+{
+ int ret;
+ ret = _tdb_lock(tdb, list, ltype, F_SETLKW);
+ if (ret) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
+ "ltype=%d (%s)\n", list, ltype, strerror(errno)));
+ }
+ return ret;
+}
+
+/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
+int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
+{
+ return _tdb_lock(tdb, list, ltype, F_SETLK);
+}
+
+
+/* unlock the database: returns void because it's too late for errors. */
+ /* changed to return int it may be interesting to know there
+ has been an error --simo */
+int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
+{
+ int ret = -1;
+ int i;
+ struct tdb_lock_type *lck = NULL;
+ bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
+
+ ltype &= ~TDB_MARK_LOCK;
+
+ /* a global lock allows us to avoid per chain locks */
+ if (tdb->global_lock.count &&
+ (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->flags & TDB_NOLOCK)
+ return 0;
+
+ /* Sanity checks */
+ if (list < -1 || list >= (int)tdb->header.hash_size) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
+ return ret;
+ }
+
+ for (i=0; i<tdb->num_lockrecs; i++) {
+ if (tdb->lockrecs[i].list == list) {
+ lck = &tdb->lockrecs[i];
+ break;
+ }
+ }
+
+ if ((lck == NULL) || (lck->count == 0)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
+ return -1;
+ }
+
+ if (lck->count > 1) {
+ lck->count--;
+ return 0;
+ }
+
+ /*
+ * This lock has count==1 left, so we need to unlock it in the
+ * kernel. We don't bother with decrementing the in-memory array
+ * element, we're about to overwrite it with the last array element
+ * anyway.
+ */
+
+ if (mark_lock) {
+ ret = 0;
+ } else {
+ ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK,
+ F_SETLKW, 0, 1);
+ }
+ tdb->num_locks--;
+
+ /*
+ * Shrink the array by overwriting the element just unlocked with the
+ * last array element.
+ */
+
+ if (tdb->num_lockrecs > 1) {
+ *lck = tdb->lockrecs[tdb->num_lockrecs-1];
+ }
+ tdb->num_lockrecs -= 1;
+
+ /*
+ * We don't bother with realloc when the array shrinks, but if we have
+ * a completely idle tdb we should get rid of the locked array.
+ */
+
+ if (tdb->num_lockrecs == 0) {
+ SAFE_FREE(tdb->lockrecs);
+ }
+
+ if (ret)
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
+ return ret;
+}
+
+/*
+ get the transaction lock
+ */
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
+{
+ if (tdb->global_lock.count) {
+ return 0;
+ }
+ if (tdb->transaction_lock_count > 0) {
+ tdb->transaction_lock_count++;
+ return 0;
+ }
+
+ if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, ltype,
+ F_SETLKW, 0, 1) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_lock: failed to get transaction lock\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+ tdb->transaction_lock_count++;
+ return 0;
+}
+
+/*
+ release the transaction lock
+ */
+int tdb_transaction_unlock(struct tdb_context *tdb)
+{
+ int ret;
+ if (tdb->global_lock.count) {
+ return 0;
+ }
+ if (tdb->transaction_lock_count > 1) {
+ tdb->transaction_lock_count--;
+ return 0;
+ }
+ ret = tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ if (ret == 0) {
+ tdb->transaction_lock_count = 0;
+ }
+ return ret;
+}
+
+
+
+
+/* lock/unlock entire database */
+static int _tdb_lockall(struct tdb_context *tdb, int ltype, int op)
+{
+ bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
+
+ ltype &= ~TDB_MARK_LOCK;
+
+ /* There are no locks on read-only dbs */
+ if (tdb->read_only || tdb->traverse_read) {
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
+ tdb->global_lock.count++;
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ /* a global lock of a different type exists */
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->num_locks != 0) {
+ /* can't combine global and chain locks */
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (!mark_lock &&
+ tdb->methods->tdb_brlock(tdb, FREELIST_TOP, ltype, op,
+ 0, 4*tdb->header.hash_size)) {
+ if (op == F_SETLKW) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
+ }
+ return -1;
+ }
+
+ tdb->global_lock.count = 1;
+ tdb->global_lock.ltype = ltype;
+
+ return 0;
+}
+
+
+
+/* unlock entire db */
+static int _tdb_unlockall(struct tdb_context *tdb, int ltype)
+{
+ bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
+
+ ltype &= ~TDB_MARK_LOCK;
+
+ /* There are no locks on read-only dbs */
+ if (tdb->read_only || tdb->traverse_read) {
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->global_lock.count > 1) {
+ tdb->global_lock.count--;
+ return 0;
+ }
+
+ if (!mark_lock &&
+ tdb->methods->tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW,
+ 0, 4*tdb->header.hash_size)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
+ return -1;
+ }
+
+ tdb->global_lock.count = 0;
+ tdb->global_lock.ltype = 0;
+
+ return 0;
+}
+
+/* lock entire database with write lock */
+int tdb_lockall(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_lockall");
+ return _tdb_lockall(tdb, F_WRLCK, F_SETLKW);
+}
+
+/* lock entire database with write lock - mark only */
+int tdb_lockall_mark(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_lockall_mark");
+ return _tdb_lockall(tdb, F_WRLCK | TDB_MARK_LOCK, F_SETLKW);
+}
+
+/* unlock entire database with write lock - unmark only */
+int tdb_lockall_unmark(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_lockall_unmark");
+ return _tdb_unlockall(tdb, F_WRLCK | TDB_MARK_LOCK);
+}
+
+/* lock entire database with write lock - nonblocking varient */
+int tdb_lockall_nonblock(struct tdb_context *tdb)
+{
+ int ret = _tdb_lockall(tdb, F_WRLCK, F_SETLK);
+ tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
+ return ret;
+}
+
+/* unlock entire database with write lock */
+int tdb_unlockall(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_unlockall");
+ return _tdb_unlockall(tdb, F_WRLCK);
+}
+
+/* lock entire database with read lock */
+int tdb_lockall_read(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_lockall_read");
+ return _tdb_lockall(tdb, F_RDLCK, F_SETLKW);
+}
+
+/* lock entire database with read lock - nonblock varient */
+int tdb_lockall_read_nonblock(struct tdb_context *tdb)
+{
+ int ret = _tdb_lockall(tdb, F_RDLCK, F_SETLK);
+ tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
+ return ret;
+}
+
+/* unlock entire database with read lock */
+int tdb_unlockall_read(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_unlockall_read");
+ return _tdb_unlockall(tdb, F_RDLCK);
+}
+
+/* lock/unlock one hash chain. This is meant to be used to reduce
+ contention - it cannot guarantee how many records will be locked */
+int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
+{
+ int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+ tdb_trace_1rec(tdb, "tdb_chainlock", key);
+ return ret;
+}
+
+/* lock/unlock one hash chain, non-blocking. This is meant to be used
+ to reduce contention - it cannot guarantee how many records will be
+ locked */
+int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
+{
+ int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+ tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
+ return ret;
+}
+
+/* mark a chain as locked without actually locking it. Warning! use with great caution! */
+int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
+{
+ int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
+ tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
+ return ret;
+}
+
+/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
+int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
+{
+ tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
+ return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
+}
+
+int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
+{
+ tdb_trace_1rec(tdb, "tdb_chainunlock", key);
+ return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+ int ret;
+ ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+ tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
+ return ret;
+}
+
+int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+ tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
+ return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+
+
+/* record lock stops delete underneath */
+int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+ if (tdb->global_lock.count) {
+ return 0;
+ }
+ return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
+}
+
+/*
+ Write locks override our own fcntl readlocks, so check it here.
+ Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+ an error to fail to get the lock here.
+*/
+int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+ struct tdb_traverse_lock *i;
+ for (i = &tdb->travlocks; i; i = i->next)
+ if (i->off == off)
+ return -1;
+ return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1, 1);
+}
+
+/*
+ Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+ an error to fail to get the lock here.
+*/
+int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+ return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0, 1);
+}
+
+/* fcntl locks don't stack: avoid unlocking someone else's */
+int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+ struct tdb_traverse_lock *i;
+ uint32_t count = 0;
+
+ if (tdb->global_lock.count) {
+ return 0;
+ }
+
+ if (off == 0)
+ return 0;
+ for (i = &tdb->travlocks; i; i = i->next)
+ if (i->off == off)
+ count++;
+ return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0, 1) : 0);
+}
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
+static struct tdb_context *tdbs = NULL;
+
+
+/* This is based on the hash algorithm from gdbm */
+static unsigned int default_tdb_hash(TDB_DATA *key)
+{
+ uint32_t value; /* Used to compute the hash value. */
+ uint32_t i; /* Used to cycle through random values. */
+
+ /* Set the initial value from the key size. */
+ for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
+ value = (value + (key->dptr[i] << (i*5 % 24)));
+
+ return (1103515243 * value + 12345);
+}
+
+
+/* initialise a new database with a specified hash size */
+static int tdb_new_database(struct tdb_context *tdb, int hash_size)
+{
+ struct tdb_header *newdb;
+ size_t size;
+ int ret = -1;
+ ssize_t written;
+
+ /* We make it up in memory, then write it out if not internal */
+ size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
+ if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
+
+ /* Fill in the header */
+ newdb->version = TDB_VERSION;
+ newdb->hash_size = hash_size;
+ if (tdb->flags & TDB_INTERNAL) {
+ tdb->map_size = size;
+ tdb->map_ptr = (char *)newdb;
+ memcpy(&tdb->header, newdb, sizeof(tdb->header));
+ /* Convert the `ondisk' version if asked. */
+ CONVERT(*newdb);
+ return 0;
+ }
+ if (lseek(tdb->fd, 0, SEEK_SET) == -1)
+ goto fail;
+
+ if (ftruncate(tdb->fd, 0) == -1)
+ goto fail;
+
+ /* This creates an endian-converted header, as if read from disk */
+ CONVERT(*newdb);
+ memcpy(&tdb->header, newdb, sizeof(tdb->header));
+ /* Don't endian-convert the magic food! */
+ memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
+ /* we still have "ret == -1" here */
+ written = write(tdb->fd, newdb, size);
+ if (written == size) {
+ ret = 0;
+ } else if (written != -1) {
+ /* call write once again, this usually should return -1 and
+ * set errno appropriately */
+ size -= written;
+ written = write(tdb->fd, newdb+written, size);
+ if (written == size) {
+ ret = 0;
+ } else if (written >= 0) {
+ /* a second incomplete write - we give up.
+ * guessing the errno... */
+ errno = ENOSPC;
+ }
+ }
+
+ fail:
+ SAFE_FREE(newdb);
+ return ret;
+}
+
+
+
+static int tdb_already_open(dev_t device,
+ ino_t ino)
+{
+ struct tdb_context *i;
+
+ for (i = tdbs; i; i = i->next) {
+ if (i->device == device && i->inode == ino) {
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+/* open the database, creating it if necessary
+
+ The open_flags and mode are passed straight to the open call on the
+ database file. A flags value of O_WRONLY is invalid. The hash size
+ is advisory, use zero for a default value.
+
+ Return is NULL on error, in which case errno is also set. Don't
+ try to call tdb_error or tdb_errname, just do strerror(errno).
+
+ @param name may be NULL for internal databases. */
+struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
+ int open_flags, mode_t mode)
+{
+ return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
+}
+
+/* a default logging function */
+static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
+static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
+{
+}
+
+
+struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+ int open_flags, mode_t mode,
+ const struct tdb_logging_context *log_ctx,
+ tdb_hash_func hash_fn)
+{
+ struct tdb_context *tdb;
+ struct stat st;
+ int rev = 0, locked = 0;
+ unsigned char *vp;
+ uint32_t vertest;
+ unsigned v;
+
+ if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
+ /* Can't log this */
+ errno = ENOMEM;
+ goto fail;
+ }
+ tdb_io_init(tdb);
+ tdb->fd = -1;
+#ifdef TDB_TRACE
+ tdb->tracefd = -1;
+#endif
+ tdb->name = NULL;
+ tdb->map_ptr = NULL;
+ tdb->flags = tdb_flags;
+ tdb->open_flags = open_flags;
+ if (log_ctx) {
+ tdb->log = *log_ctx;
+ } else {
+ tdb->log.log_fn = null_log_fn;
+ tdb->log.log_private = NULL;
+ }
+ tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
+
+ /* cache the page size */
+ tdb->page_size = getpagesize();
+ if (tdb->page_size <= 0) {
+ tdb->page_size = 0x2000;
+ }
+
+ tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
+
+ if ((open_flags & O_ACCMODE) == O_WRONLY) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
+ name));
+ errno = EINVAL;
+ goto fail;
+ }
+
+ if (hash_size == 0)
+ hash_size = DEFAULT_HASH_SIZE;
+ if ((open_flags & O_ACCMODE) == O_RDONLY) {
+ tdb->read_only = 1;
+ /* read only databases don't do locking or clear if first */
+ tdb->flags |= TDB_NOLOCK;
+ tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+ }
+
+ if ((tdb->flags & TDB_ALLOW_NESTING) &&
+ (tdb->flags & TDB_DISALLOW_NESTING)) {
+ tdb->ecode = TDB_ERR_NESTING;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+ "allow_nesting and disallow_nesting are not allowed together!"));
+ errno = EINVAL;
+ goto fail;
+ }
+
+ /*
+ * TDB_ALLOW_NESTING is the default behavior.
+ * Note: this may change in future versions!
+ */
+ if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
+ tdb->flags |= TDB_ALLOW_NESTING;
+ }
+
+ /* internal databases don't mmap or lock, and start off cleared */
+ if (tdb->flags & TDB_INTERNAL) {
+ tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
+ tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+ if (tdb_new_database(tdb, hash_size) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
+ goto fail;
+ }
+ goto internal;
+ }
+
+ if ((tdb->fd = open(name, open_flags, mode)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
+ name, strerror(errno)));
+ goto fail; /* errno set by open(2) */
+ }
+
+ /* on exec, don't inherit the fd */
+ v = fcntl(tdb->fd, F_GETFD, 0);
+ fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
+
+ /* ensure there is only one process initialising at once */
+ if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
+ name, strerror(errno)));
+ goto fail; /* errno set by tdb_brlock */
+ }
+
+ /* we need to zero database if we are the only one with it open */
+ if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
+ (!tdb->read_only) &&
+ (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
+ open_flags |= O_CREAT;
+ if (ftruncate(tdb->fd, 0) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+ "failed to truncate %s: %s\n",
+ name, strerror(errno)));
+ goto fail; /* errno set by ftruncate */
+ }
+ }
+
+ errno = 0;
+ if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
+ || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
+ if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
+ if (errno == 0) {
+ errno = EIO; /* ie bad format or something */
+ }
+ goto fail;
+ }
+ rev = (tdb->flags & TDB_CONVERT);
+ } else if (tdb->header.version != TDB_VERSION
+ && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
+ /* wrong version */
+ errno = EIO;
+ goto fail;
+ }
+ vp = (unsigned char *)&tdb->header.version;
+ vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
+ (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
+ tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
+ if (!rev)
+ tdb->flags &= ~TDB_CONVERT;
+ else {
+ tdb->flags |= TDB_CONVERT;
+ tdb_convert(&tdb->header, sizeof(tdb->header));
+ }
+ if (fstat(tdb->fd, &st) == -1)
+ goto fail;
+
+ if (tdb->header.rwlocks != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
+ goto fail;
+ }
+
+ /* Is it already in the open list? If so, fail. */
+ if (tdb_already_open(st.st_dev, st.st_ino)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+ "%s (%d,%d) is already open in this process\n",
+ name, (int)st.st_dev, (int)st.st_ino));
+ errno = EBUSY;
+ goto fail;
+ }
+
+ if (!(tdb->name = (char *)strdup(name))) {
+ errno = ENOMEM;
+ goto fail;
+ }
+
+ tdb->map_size = st.st_size;
+ tdb->device = st.st_dev;
+ tdb->inode = st.st_ino;
+ tdb_mmap(tdb);
+ if (locked) {
+ if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+ "failed to take ACTIVE_LOCK on %s: %s\n",
+ name, strerror(errno)));
+ goto fail;
+ }
+
+ }
+
+ /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
+ we didn't get the initial exclusive lock as we need to let all other
+ users know we're using it. */
+
+ if (tdb_flags & TDB_CLEAR_IF_FIRST) {
+ /* leave this lock in place to indicate it's in use */
+ if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
+ goto fail;
+ }
+
+ /* if needed, run recovery */
+ if (tdb_transaction_recover(tdb) == -1) {
+ goto fail;
+ }
+
+#ifdef TDB_TRACE
+ {
+ char tracefile[strlen(name) + 32];
+
+ snprintf(tracefile, sizeof(tracefile),
+ "%s.trace.%li", name, (long)getpid());
+ tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
+ if (tdb->tracefd >= 0) {
+ tdb_enable_seqnum(tdb);
+ tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
+ open_flags);
+ } else
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
+ }
+#endif
+
+ internal:
+ /* Internal (memory-only) databases skip all the code above to
+ * do with disk files, and resume here by releasing their
+ * global lock and hooking into the active list. */
+ if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
+ goto fail;
+ tdb->next = tdbs;
+ tdbs = tdb;
+ return tdb;
+
+ fail:
+ { int save_errno = errno;
+
+ if (!tdb)
+ return NULL;
+
+#ifdef TDB_TRACE
+ close(tdb->tracefd);
+#endif
+ if (tdb->map_ptr) {
+ if (tdb->flags & TDB_INTERNAL)
+ SAFE_FREE(tdb->map_ptr);
+ else
+ tdb_munmap(tdb);
+ }
+ SAFE_FREE(tdb->name);
+ if (tdb->fd != -1)
+ if (close(tdb->fd) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
+ SAFE_FREE(tdb);
+ errno = save_errno;
+ return NULL;
+ }
+}
+
+/*
+ * Set the maximum number of dead records per hash chain
+ */
+
+void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
+{
+ tdb->max_dead_records = max_dead;
+}
+
+/**
+ * Close a database.
+ *
+ * @returns -1 for error; 0 for success.
+ **/
+int tdb_close(struct tdb_context *tdb)
+{
+ struct tdb_context **i;
+ int ret = 0;
+
+ tdb_trace(tdb, "tdb_close");
+ if (tdb->transaction) {
+ _tdb_transaction_cancel(tdb);
+ }
+
+ if (tdb->map_ptr) {
+ if (tdb->flags & TDB_INTERNAL)
+ SAFE_FREE(tdb->map_ptr);
+ else
+ tdb_munmap(tdb);
+ }
+ SAFE_FREE(tdb->name);
+ if (tdb->fd != -1) {
+ ret = close(tdb->fd);
+ tdb->fd = -1;
+ }
+ SAFE_FREE(tdb->lockrecs);
+
+ /* Remove from contexts list */
+ for (i = &tdbs; *i; i = &(*i)->next) {
+ if (*i == tdb) {
+ *i = tdb->next;
+ break;
+ }
+ }
+
+#ifdef TDB_TRACE
+ close(tdb->tracefd);
+#endif
+ memset(tdb, 0, sizeof(*tdb));
+ SAFE_FREE(tdb);
+
+ return ret;
+}
+
+/* register a loging function */
+void tdb_set_logging_function(struct tdb_context *tdb,
+ const struct tdb_logging_context *log_ctx)
+{
+ tdb->log = *log_ctx;
+}
+
+void *tdb_get_logging_private(struct tdb_context *tdb)
+{
+ return tdb->log.log_private;
+}
+
+static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
+{
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+ !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
+ struct stat st;
+#endif
+
+ if (tdb->flags & TDB_INTERNAL) {
+ return 0; /* Nothing to do. */
+ }
+
+ if (tdb->num_locks != 0 || tdb->global_lock.count) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
+ goto fail;
+ }
+
+ if (tdb->transaction != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
+ goto fail;
+ }
+
+/* If we have real pread & pwrite, we can skip reopen. */
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+ !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
+ if (tdb_munmap(tdb) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
+ goto fail;
+ }
+ if (close(tdb->fd) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
+ tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
+ if (tdb->fd == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
+ goto fail;
+ }
+ if (fstat(tdb->fd, &st) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
+ goto fail;
+ }
+ if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
+ goto fail;
+ }
+ tdb_mmap(tdb);
+#endif /* fake pread or pwrite */
+
+ if (active_lock &&
+ (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
+ goto fail;
+ }
+
+ return 0;
+
+fail:
+ tdb_close(tdb);
+ return -1;
+}
+
+/* reopen a tdb - this can be used after a fork to ensure that we have an independent
+ seek pointer from our parent and to re-establish locks */
+int tdb_reopen(struct tdb_context *tdb)
+{
+ return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
+}
+
+/* reopen all tdb's */
+int tdb_reopen_all(int parent_longlived)
+{
+ struct tdb_context *tdb;
+
+ for (tdb=tdbs; tdb; tdb = tdb->next) {
+ bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
+
+ /*
+ * If the parent is longlived (ie. a
+ * parent daemon architecture), we know
+ * it will keep it's active lock on a
+ * tdb opened with CLEAR_IF_FIRST. Thus
+ * for child processes we don't have to
+ * add an active lock. This is essential
+ * to improve performance on systems that
+ * keep POSIX locks as a non-scalable data
+ * structure in the kernel.
+ */
+ if (parent_longlived) {
+ /* Ensure no clear-if-first. */
+ active_lock = false;
+ }
+
+ if (tdb_reopen_internal(tdb, active_lock) != 0)
+ return -1;
+ }
+
+ return 0;
+}
+++ /dev/null
-/*
- Unix SMB/CIFS implementation.
- Samba database functions
- Copyright (C) Anton Blanchard 2001
-
- This program is free software; you can redistribute it and/or modify
- it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2 of the License, or
- (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- GNU General Public License for more details.
-
- You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software
- Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
-*/
-#define STANDALONE 1
-
-#if HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#if STANDALONE
-#include <stdlib.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <string.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <sys/stat.h>
-#include <time.h>
-#include <signal.h>
-#include "spinlock.h"
-
-#define DEBUG
-#else
-#include "includes.h"
-#endif
-
-#ifdef USE_SPINLOCKS
-
-/*
- * ARCH SPECIFIC
- */
-
-#if defined(SPARC_SPINLOCKS)
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
- unsigned int result;
-
- asm volatile("ldstub [%1], %0"
- : "=r" (result)
- : "r" (lock)
- : "memory");
-
- return (result == 0) ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
- asm volatile("":::"memory");
- *lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
- *lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
- return (*lock != 0);
-}
-
-#elif defined(POWERPC_SPINLOCKS)
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
- unsigned int result;
-
- __asm__ __volatile__(
-"1: lwarx %0,0,%1\n\
- cmpwi 0,%0,0\n\
- li %0,0\n\
- bne- 2f\n\
- li %0,1\n\
- stwcx. %0,0,%1\n\
- bne- 1b\n\
- isync\n\
-2:" : "=&r"(result)
- : "r"(lock)
- : "cr0", "memory");
-
- return (result == 1) ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
- asm volatile("eieio":::"memory");
- *lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
- *lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
- return (*lock != 0);
-}
-
-#elif defined(INTEL_SPINLOCKS)
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
- int oldval;
-
- asm volatile("xchgl %0,%1"
- : "=r" (oldval), "=m" (*lock)
- : "0" (0)
- : "memory");
-
- return oldval > 0 ? 0 : EBUSY;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
- asm volatile("":::"memory");
- *lock = 1;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
- *lock = 1;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
- return (*lock != 1);
-}
-
-#elif defined(MIPS_SPINLOCKS)
-
-static inline unsigned int load_linked(unsigned long addr)
-{
- unsigned int res;
-
- __asm__ __volatile__("ll\t%0,(%1)"
- : "=r" (res)
- : "r" (addr));
-
- return res;
-}
-
-static inline unsigned int store_conditional(unsigned long addr, unsigned int value)
-{
- unsigned int res;
-
- __asm__ __volatile__("sc\t%0,(%2)"
- : "=r" (res)
- : "0" (value), "r" (addr));
- return res;
-}
-
-static inline int __spin_trylock(spinlock_t *lock)
-{
- unsigned int mw;
-
- do {
- mw = load_linked(lock);
- if (mw)
- return EBUSY;
- } while (!store_conditional(lock, 1));
-
- asm volatile("":::"memory");
-
- return 0;
-}
-
-static inline void __spin_unlock(spinlock_t *lock)
-{
- asm volatile("":::"memory");
- *lock = 0;
-}
-
-static inline void __spin_lock_init(spinlock_t *lock)
-{
- *lock = 0;
-}
-
-static inline int __spin_is_locked(spinlock_t *lock)
-{
- return (*lock != 0);
-}
-
-#else
-#error Need to implement spinlock code in spinlock.c
-#endif
-
-/*
- * OS SPECIFIC
- */
-
-static void yield_cpu(void)
-{
- struct timespec tm;
-
-#ifdef USE_SCHED_YIELD
- sched_yield();
-#else
- /* Linux will busy loop for delays < 2ms on real time tasks */
- tm.tv_sec = 0;
- tm.tv_nsec = 2000000L + 1;
- nanosleep(&tm, NULL);
-#endif
-}
-
-static int this_is_smp(void)
-{
- return 0;
-}
-
-/*
- * GENERIC
- */
-
-static int smp_machine = 0;
-
-static inline void __spin_lock(spinlock_t *lock)
-{
- int ntries = 0;
-
- while(__spin_trylock(lock)) {
- while(__spin_is_locked(lock)) {
- if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
- continue;
- yield_cpu();
- }
- }
-}
-
-static void __read_lock(tdb_rwlock_t *rwlock)
-{
- int ntries = 0;
-
- while(1) {
- __spin_lock(&rwlock->lock);
-
- if (!(rwlock->count & RWLOCK_BIAS)) {
- rwlock->count++;
- __spin_unlock(&rwlock->lock);
- return;
- }
-
- __spin_unlock(&rwlock->lock);
-
- while(rwlock->count & RWLOCK_BIAS) {
- if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
- continue;
- yield_cpu();
- }
- }
-}
-
-static void __write_lock(tdb_rwlock_t *rwlock)
-{
- int ntries = 0;
-
- while(1) {
- __spin_lock(&rwlock->lock);
-
- if (rwlock->count == 0) {
- rwlock->count |= RWLOCK_BIAS;
- __spin_unlock(&rwlock->lock);
- return;
- }
-
- __spin_unlock(&rwlock->lock);
-
- while(rwlock->count != 0) {
- if (smp_machine && ntries++ < MAX_BUSY_LOOPS)
- continue;
- yield_cpu();
- }
- }
-}
-
-static void __write_unlock(tdb_rwlock_t *rwlock)
-{
- __spin_lock(&rwlock->lock);
-
-#ifdef DEBUG
- if (!(rwlock->count & RWLOCK_BIAS))
- fprintf(stderr, "bug: write_unlock\n");
-#endif
-
- rwlock->count &= ~RWLOCK_BIAS;
- __spin_unlock(&rwlock->lock);
-}
-
-static void __read_unlock(tdb_rwlock_t *rwlock)
-{
- __spin_lock(&rwlock->lock);
-
-#ifdef DEBUG
- if (!rwlock->count)
- fprintf(stderr, "bug: read_unlock\n");
-
- if (rwlock->count & RWLOCK_BIAS)
- fprintf(stderr, "bug: read_unlock\n");
-#endif
-
- rwlock->count--;
- __spin_unlock(&rwlock->lock);
-}
-
-/* TDB SPECIFIC */
-
-/* lock a list in the database. list -1 is the alloc list */
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type)
-{
- tdb_rwlock_t *rwlocks;
-
- if (!tdb->map_ptr) return -1;
- rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
-
- switch(rw_type) {
- case F_RDLCK:
- __read_lock(&rwlocks[list+1]);
- break;
-
- case F_WRLCK:
- __write_lock(&rwlocks[list+1]);
- break;
-
- default:
- return TDB_ERRCODE(TDB_ERR_LOCK, -1);
- }
- return 0;
-}
-
-/* unlock the database. */
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type)
-{
- tdb_rwlock_t *rwlocks;
-
- if (!tdb->map_ptr) return -1;
- rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
-
- switch(rw_type) {
- case F_RDLCK:
- __read_unlock(&rwlocks[list+1]);
- break;
-
- case F_WRLCK:
- __write_unlock(&rwlocks[list+1]);
- break;
-
- default:
- return TDB_ERRCODE(TDB_ERR_LOCK, -1);
- }
-
- return 0;
-}
-
-int tdb_create_rwlocks(int fd, unsigned int hash_size)
-{
- unsigned size, i;
- tdb_rwlock_t *rwlocks;
-
- size = (hash_size + 1) * sizeof(tdb_rwlock_t);
- rwlocks = malloc(size);
- if (!rwlocks)
- return -1;
-
- for(i = 0; i < hash_size+1; i++) {
- __spin_lock_init(&rwlocks[i].lock);
- rwlocks[i].count = 0;
- }
-
- /* Write it out (appending to end) */
- if (write(fd, rwlocks, size) != size) {
- free(rwlocks);
- return -1;
- }
- smp_machine = this_is_smp();
- free(rwlocks);
- return 0;
-}
-
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
-{
- tdb_rwlock_t *rwlocks;
- unsigned i;
-
- if (tdb->header.rwlocks == 0) return 0;
- if (!tdb->map_ptr) return -1;
-
- /* We're mmapped here */
- rwlocks = (tdb_rwlock_t *)((char *)tdb->map_ptr + tdb->header.rwlocks);
- for(i = 0; i < tdb->header.hash_size+1; i++) {
- __spin_lock_init(&rwlocks[i].lock);
- rwlocks[i].count = 0;
- }
- return 0;
-}
-#else
-int tdb_create_rwlocks(int fd, unsigned int hash_size) { return 0; }
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type) { return -1; }
-
-/* Non-spinlock version: remove spinlock pointer */
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb)
-{
- tdb_off off = (tdb_off)((char *)&tdb->header.rwlocks
- - (char *)&tdb->header);
-
- tdb->header.rwlocks = 0;
- if (lseek(tdb->fd, off, SEEK_SET) != off
- || write(tdb->fd, (void *)&tdb->header.rwlocks,
- sizeof(tdb->header.rwlocks))
- != sizeof(tdb->header.rwlocks))
- return -1;
- return 0;
-}
-#endif
+++ /dev/null
-#ifndef __SPINLOCK_H__
-#define __SPINLOCK_H__
-
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#include <atalk/tdb.h>
-
-#ifdef USE_SPINLOCKS
-
-#define RWLOCK_BIAS 0x1000UL
-
-/* OS SPECIFIC */
-#define MAX_BUSY_LOOPS 1000
-#undef USE_SCHED_YIELD
-
-/* ARCH SPECIFIC */
-/* We should make sure these are padded to a cache line */
-#if defined(SPARC_SPINLOCKS)
-typedef volatile char spinlock_t;
-#elif defined(POWERPC_SPINLOCKS)
-typedef volatile unsigned long spinlock_t;
-#elif defined(INTEL_SPINLOCKS)
-typedef volatile int spinlock_t;
-#elif defined(MIPS_SPINLOCKS)
-typedef volatile unsigned long spinlock_t;
-#else
-#error Need to implement spinlock code in spinlock.h
-#endif
-
-typedef struct {
- spinlock_t lock;
- volatile int count;
-} tdb_rwlock_t;
-
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_create_rwlocks(int fd, unsigned int hash_size);
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb);
-
-#else /* !USE_SPINLOCKS */
-#if 0
-#define tdb_create_rwlocks(fd, hash_size) 0
-#define tdb_spinlock(tdb, list, rw_type) (-1)
-#define tdb_spinunlock(tdb, list, rw_type) (-1)
-#else
-int tdb_spinlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_spinunlock(TDB_CONTEXT *tdb, int list, int rw_type);
-int tdb_create_rwlocks(int fd, unsigned int hash_size);
-#endif
-int tdb_clear_spinlocks(TDB_CONTEXT *tdb);
-#endif
-
-#endif
/*
Unix SMB/CIFS implementation.
- Samba database functions
- Copyright (C) Andrew Tridgell 1999-2000
- Copyright (C) Luke Kenneth Casson Leighton 2000
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
- This program is free software; you can redistribute it and/or modify
- it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2 of the License, or
- (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- GNU General Public License for more details.
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
- You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software
- Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
-*/
-
-#define STANDALONE 1
-
-/* NOTE: If you use tdbs under valgrind, and in particular if you run
- * tdbtorture, you may get spurious "uninitialized value" warnings. I
- * think this is because valgrind doesn't understand that the mmap'd
- * area may be written to by other processes. Memory can, from the
- * point of view of the grinded process, spontaneously become
- * initialized.
- *
- * I can think of a few solutions. [mbp 20030311]
- *
- * 1 - Write suppressions for Valgrind so that it doesn't complain
- * about this. Probably the most reasonable but people need to
- * remember to use them.
- *
- * 2 - Use IO not mmap when running under valgrind. Not so nice.
- *
- * 3 - Use the special valgrind macros to mark memory as valid at the
- * right time. Probably too hard -- the process just doesn't know.
- */
-
-#ifdef STANDALONE
-#if HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#define _XOPEN_SOURCE 600
-#include <unistd.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <string.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <signal.h>
-#include "spinlock.h"
-#else
-#include "includes.h"
-#endif
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
-#define TDB_MAGIC_FOOD "TDB file\n"
-#define TDB_VERSION (0x26011967 + 6)
-#define TDB_MAGIC (0x26011999U)
-#define TDB_FREE_MAGIC (~TDB_MAGIC)
-#define TDB_DEAD_MAGIC (0xFEE1DEAD)
-#define TDB_ALIGNMENT 4
-#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
-#define DEFAULT_HASH_SIZE 131
-#define TDB_PAGE_SIZE 0x2000
-#define FREELIST_TOP (sizeof(struct tdb_header))
-#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
-#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
-#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
-#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
-#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
-
-/* NB assumes there is a local variable called "tdb" that is the
- * current context, also takes doubly-parenthesized print-style
- * argument. */
-#define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0)
-
-/* lock offsets */
-#define GLOBAL_LOCK 0
-#define ACTIVE_LOCK 4
-
-#ifndef MAP_FILE
-#define MAP_FILE 0
-#endif
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
-#ifndef MAP_FAILED
-#define MAP_FAILED ((void *)-1)
-#endif
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
-/* free memory if the pointer is valid and zero the pointer */
-#ifndef SAFE_FREE
-#define SAFE_FREE(x) do { if ((x) != NULL) {free((x)); (x)=NULL;} } while(0)
-#endif
+#include "tdb_private.h"
-#define BUCKET(hash) ((hash) % tdb->header.hash_size)
TDB_DATA tdb_null;
-/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
-static TDB_CONTEXT *tdbs = NULL;
-
-static int tdb_munmap(TDB_CONTEXT *tdb)
-{
- if (tdb->flags & TDB_INTERNAL)
- return 0;
-
-#ifdef HAVE_MMAP
- if (tdb->map_ptr) {
- int ret = munmap(tdb->map_ptr, tdb->map_size);
- if (ret != 0)
- return ret;
- }
-#endif
- tdb->map_ptr = NULL;
- return 0;
-}
-
-static void tdb_mmap(TDB_CONTEXT *tdb)
+/*
+ non-blocking increment of the tdb sequence number if the tdb has been opened using
+ the TDB_SEQNUM flag
+*/
+void tdb_increment_seqnum_nonblock(struct tdb_context *tdb)
{
- if (tdb->flags & TDB_INTERNAL)
+ tdb_off_t seqnum=0;
+
+ if (!(tdb->flags & TDB_SEQNUM)) {
return;
-
-#ifdef HAVE_MMAP
- if (!(tdb->flags & TDB_NOMMAP)) {
- tdb->map_ptr = mmap(NULL, tdb->map_size,
- PROT_READ|(tdb->read_only? 0:PROT_WRITE),
- MAP_SHARED|MAP_FILE, tdb->fd, 0);
-
- /*
- * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
- */
-
- if (tdb->map_ptr == MAP_FAILED) {
- tdb->map_ptr = NULL;
- TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n",
- tdb->map_size, strerror(errno)));
- }
- } else {
- tdb->map_ptr = NULL;
}
-#else
- tdb->map_ptr = NULL;
-#endif
-}
-/* Endian conversion: we only ever deal with 4 byte quantities */
-static void *convert(void *buf, u32 size)
-{
- u32 i, *p = buf;
- for (i = 0; i < size / 4; i++)
- p[i] = TDB_BYTEREV(p[i]);
- return buf;
-}
-#define DOCONV() (tdb->flags & TDB_CONVERT)
-#define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
-
-/* the body of the database is made of one list_struct for the free space
- plus a separate data list for each hash value */
-struct list_struct {
- tdb_off next; /* offset of the next record in the list */
- tdb_len rec_len; /* total byte length of record */
- tdb_len key_len; /* byte length of key */
- tdb_len data_len; /* byte length of data */
- u32 full_hash; /* the full 32 bit hash of the key */
- u32 magic; /* try to catch errors */
- /* the following union is implied:
- union {
- char record[rec_len];
- struct {
- char key[key_len];
- char data[data_len];
- }
- u32 totalsize; (tailer)
- }
+ /* we ignore errors from this, as we have no sane way of
+ dealing with them.
*/
-};
-
-/***************************************************************
- Allow a caller to set a "alarm" flag that tdb can check to abort
- a blocking lock on SIGALRM.
-***************************************************************/
-
-static sig_atomic_t *palarm_fired;
-
-void tdb_set_lock_alarm(sig_atomic_t *palarm)
-{
- palarm_fired = palarm;
+ tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+ seqnum++;
+ tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
}
-/* a byte range locking function - return 0 on success
- this functions locks/unlocks 1 byte at the specified offset.
-
- On error, errno is also set so that errors are passed back properly
- through tdb_open(). */
-static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset,
- int rw_type, int lck_type, int probe)
-{
- struct flock fl;
- int ret;
-
- if (tdb->flags & TDB_NOLOCK)
- return 0;
- if ((rw_type == F_WRLCK) && (tdb->read_only)) {
- errno = EACCES;
- return -1;
- }
-
- fl.l_type = rw_type;
- fl.l_whence = SEEK_SET;
- fl.l_start = offset;
- fl.l_len = 1;
- fl.l_pid = 0;
-
- do {
- ret = fcntl(tdb->fd,lck_type,&fl);
- if (ret == -1 && errno == EINTR && palarm_fired && *palarm_fired)
- break;
- } while (ret == -1 && errno == EINTR);
-
- if (ret == -1) {
- if (!probe && lck_type != F_SETLK) {
- /* Ensure error code is set for log fun to examine. */
- if (errno == EINTR && palarm_fired && *palarm_fired)
- tdb->ecode = TDB_ERR_LOCK_TIMEOUT;
- else
- tdb->ecode = TDB_ERR_LOCK;
- TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n",
- tdb->fd, offset, rw_type, lck_type));
- }
- /* Was it an alarm timeout ? */
- if (errno == EINTR && palarm_fired && *palarm_fired)
- return TDB_ERRCODE(TDB_ERR_LOCK_TIMEOUT, -1);
- /* Otherwise - generic lock error. */
- /* errno set by fcntl */
- return TDB_ERRCODE(TDB_ERR_LOCK, -1);
- }
- return 0;
-}
-
-/* lock a list in the database. list -1 is the alloc list */
-static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
-{
- if (list < -1 || list >= (int)tdb->header.hash_size) {
- TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n",
- list, ltype));
- return -1;
- }
- if (tdb->flags & TDB_NOLOCK)
- return 0;
-
- /* Since fcntl locks don't nest, we do a lock for the first one,
- and simply bump the count for future ones */
- if (tdb->locked[list+1].count == 0) {
- if (!tdb->read_only && tdb->header.rwlocks) {
- if (tdb_spinlock(tdb, list, ltype)) {
- TDB_LOG((tdb, 0, "tdb_lock spinlock failed on list ltype=%d\n",
- list, ltype));
- return -1;
- }
- } else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
- TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n",
- list, ltype, strerror(errno)));
- return -1;
- }
- tdb->locked[list+1].ltype = ltype;
- }
- tdb->locked[list+1].count++;
- return 0;
-}
-
-/* unlock the database: returns void because it's too late for errors. */
- /* changed to return int it may be interesting to know there
- has been an error --simo */
-static int tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype)
-{
- int ret = -1;
-
- if (tdb->flags & TDB_NOLOCK)
- return 0;
-
- /* Sanity checks */
- if (list < -1 || list >= (int)tdb->header.hash_size) {
- TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
- return ret;
- }
-
- if (tdb->locked[list+1].count==0) {
- TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
- return ret;
- }
-
- if (tdb->locked[list+1].count == 1) {
- /* Down to last nested lock: unlock underneath */
- if (!tdb->read_only && tdb->header.rwlocks) {
- ret = tdb_spinunlock(tdb, list, ltype);
- } else {
- ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
- }
- } else {
- ret = 0;
- }
- tdb->locked[list+1].count--;
-
- if (ret)
- TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n"));
- return ret;
-}
-
-/* This is based on the hash algorithm from gdbm */
-static u32 tdb_hash(TDB_DATA *key)
-{
- u32 value; /* Used to compute the hash value. */
- u32 i; /* Used to cycle through random values. */
-
- /* Set the initial value from the key size. */
- for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
- value = (value + (key->dptr[i] << (i*5 % 24)));
-
- return (1103515243 * value + 12345);
-}
-
-/* check for an out of bounds access - if it is out of bounds then
- see if the database has been expanded by someone else and expand
- if necessary
- note that "len" is the minimum length needed for the db
+/*
+ increment the tdb sequence number if the tdb has been opened using
+ the TDB_SEQNUM flag
*/
-static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
-{
- struct stat st;
- if (len <= tdb->map_size)
- return 0;
- if (tdb->flags & TDB_INTERNAL) {
- if (!probe) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n",
- (int)len, (int)tdb->map_size));
- }
- return TDB_ERRCODE(TDB_ERR_IO, -1);
- }
-
- if (fstat(tdb->fd, &st) == -1)
- return TDB_ERRCODE(TDB_ERR_IO, -1);
-
- if (st.st_size < (size_t)len) {
- if (!probe) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
- (int)len, (int)st.st_size));
- }
- return TDB_ERRCODE(TDB_ERR_IO, -1);
- }
-
- /* Unmap, update size, remap */
- if (tdb_munmap(tdb) == -1)
- return TDB_ERRCODE(TDB_ERR_IO, -1);
- tdb->map_size = st.st_size;
- tdb_mmap(tdb);
- return 0;
-}
-
-/* write a lump of data at a specified offset */
-static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
-{
- if (tdb_oob(tdb, off + len, 0) != 0)
- return -1;
-
- if (tdb->map_ptr)
- memcpy(off + (char *)tdb->map_ptr, buf, len);
-#ifdef HAVE_PWRITE
- else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
-#else
- else if (lseek(tdb->fd, off, SEEK_SET) != off
- || write(tdb->fd, buf, len) != (ssize_t)len) {
-#endif
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
- off, len, strerror(errno)));
- return TDB_ERRCODE(TDB_ERR_IO, -1);
- }
- return 0;
-}
-
-/* read a lump of data at a specified offset, maybe convert */
-static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
-{
- if (tdb_oob(tdb, off + len, 0) != 0)
- return -1;
-
- if (tdb->map_ptr)
- memcpy(buf, off + (char *)tdb->map_ptr, len);
-#ifdef HAVE_PREAD
- else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) {
-#else
- else if (lseek(tdb->fd, off, SEEK_SET) != off
- || read(tdb->fd, buf, len) != (ssize_t)len) {
-#endif
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
- off, len, strerror(errno)));
- return TDB_ERRCODE(TDB_ERR_IO, -1);
- }
- if (cv)
- convert(buf, len);
- return 0;
-}
-
-/* read a lump of data, allocating the space for it */
-static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
-{
- char *buf;
-
- if (!(buf = malloc(len))) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_OOM;
- TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
- len, strerror(errno)));
- return TDB_ERRCODE(TDB_ERR_OOM, buf);
- }
- if (tdb_read(tdb, offset, buf, len, 0) == -1) {
- SAFE_FREE(buf);
- return NULL;
- }
- return buf;
-}
-
-/* read/write a tdb_off */
-static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
-{
- return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
-}
-static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
-{
- tdb_off off = *d;
- return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
-}
-
-/* read/write a record */
-static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
- if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
- return -1;
- if (TDB_BAD_MAGIC(rec)) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_CORRUPT;
- TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
- return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
- }
- return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
-}
-static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
- struct list_struct r = *rec;
- return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
-}
-
-/* read a freelist record and check for simple errors */
-static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec)
-{
- if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
- return -1;
-
- if (rec->magic == TDB_MAGIC) {
- /* this happens when a app is showdown while deleting a record - we should
- not completely fail when this happens */
- TDB_LOG((tdb, 0,"rec_free_read non-free magic at offset=%d - fixing\n",
- rec->magic, off));
- rec->magic = TDB_FREE_MAGIC;
- if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
- return -1;
- }
-
- if (rec->magic != TDB_FREE_MAGIC) {
- /* Ensure ecode is set for log fn. */
- tdb->ecode = TDB_ERR_CORRUPT;
- TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n",
- rec->magic, off));
- return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
- }
- if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
- return -1;
- return 0;
-}
-
-/* update a record tailer (must hold allocation lock) */
-static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
- const struct list_struct *rec)
-{
- tdb_off totalsize;
-
- /* Offset of tailer from record header */
- totalsize = sizeof(*rec) + rec->rec_len;
- return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
- &totalsize);
-}
-
-static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
-{
- struct list_struct rec;
- tdb_off tailer_ofs, tailer;
-
- if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
- printf("ERROR: failed to read record at %u\n", offset);
- return 0;
- }
-
- printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
- offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
-
- tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
- if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
- printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
- return rec.next;
- }
-
- if (tailer != rec.rec_len + sizeof(rec)) {
- printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
- (unsigned)tailer, (unsigned)(rec.rec_len + sizeof(rec)));
- }
- return rec.next;
-}
-
-static int tdb_dump_chain(TDB_CONTEXT *tdb, int i)
-{
- tdb_off rec_ptr, top;
-
- top = TDB_HASH_TOP(i);
-
- if (tdb_lock(tdb, i, F_WRLCK) != 0)
- return -1;
-
- if (ofs_read(tdb, top, &rec_ptr) == -1)
- return tdb_unlock(tdb, i, F_WRLCK);
-
- if (rec_ptr)
- printf("hash=%d\n", i);
-
- while (rec_ptr) {
- rec_ptr = tdb_dump_record(tdb, rec_ptr);
- }
-
- return tdb_unlock(tdb, i, F_WRLCK);
-}
-
-void tdb_dump_all(TDB_CONTEXT *tdb)
-{
- unsigned int i;
- for (i=0;i<tdb->header.hash_size;i++) {
- tdb_dump_chain(tdb, i);
- }
- printf("freelist:\n");
- tdb_dump_chain(tdb, -1);
-}
-
-int tdb_printfreelist(TDB_CONTEXT *tdb)
-{
- int ret;
- long total_free = 0;
- tdb_off offset, rec_ptr;
- struct list_struct rec;
-
- if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
- return ret;
-
- offset = FREELIST_TOP;
-
- /* read in the freelist top */
- if (ofs_read(tdb, offset, &rec_ptr) == -1) {
- tdb_unlock(tdb, -1, F_WRLCK);
- return 0;
- }
-
- printf("freelist top=[0x%08x]\n", rec_ptr );
- while (rec_ptr) {
- if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
- tdb_unlock(tdb, -1, F_WRLCK);
- return -1;
- }
-
- if (rec.magic != TDB_FREE_MAGIC) {
- printf("bad magic 0x%08x in free list\n", rec.magic);
- tdb_unlock(tdb, -1, F_WRLCK);
- return -1;
- }
-
- printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec.next, rec.rec_len, rec.rec_len );
- total_free += rec.rec_len;
-
- /* move to the next record */
- rec_ptr = rec.next;
- }
- printf("total rec_len = [0x%08x (%d)]\n", (int)total_free,
- (int)total_free);
-
- return tdb_unlock(tdb, -1, F_WRLCK);
-}
-
-/* Remove an element from the freelist. Must have alloc lock. */
-static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
-{
- tdb_off last_ptr, i;
-
- /* read in the freelist top */
- last_ptr = FREELIST_TOP;
- while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
- if (i == off) {
- /* We've found it! */
- return ofs_write(tdb, last_ptr, &next);
- }
- /* Follow chain (next offset is at start of record) */
- last_ptr = i;
- }
- TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
- return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
-}
-
-/* Add an element into the freelist. Merge adjacent records if
- neccessary. */
-static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
-{
- tdb_off right, left;
-
- /* Allocation and tailer lock */
- if (tdb_lock(tdb, -1, F_WRLCK) != 0)
- return -1;
-
- /* set an initial tailer, so if we fail we don't leave a bogus record */
- if (update_tailer(tdb, offset, rec) != 0) {
- TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
- goto fail;
- }
-
- /* Look right first (I'm an Australian, dammit) */
- right = offset + sizeof(*rec) + rec->rec_len;
- if (right + sizeof(*rec) <= tdb->map_size) {
- struct list_struct r;
-
- if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
- goto left;
- }
-
- /* If it's free, expand to include it. */
- if (r.magic == TDB_FREE_MAGIC) {
- if (remove_from_freelist(tdb, right, r.next) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
- goto left;
- }
- rec->rec_len += sizeof(r) + r.rec_len;
- }
- }
-
-left:
- /* Look left */
- left = offset - sizeof(tdb_off);
- if (left > TDB_HASH_TOP(tdb->header.hash_size-1)) {
- struct list_struct l;
- tdb_off leftsize;
-
- /* Read in tailer and jump back to header */
- if (ofs_read(tdb, left, &leftsize) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
- goto update;
- }
- left = offset - leftsize;
-
- /* Now read in record */
- if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
- goto update;
- }
-
- /* If it's free, expand to include it. */
- if (l.magic == TDB_FREE_MAGIC) {
- if (remove_from_freelist(tdb, left, l.next) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
- goto update;
- } else {
- offset = left;
- rec->rec_len += leftsize;
- }
- }
- }
-
-update:
- if (update_tailer(tdb, offset, rec) == -1) {
- TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
- goto fail;
- }
-
- /* Now, prepend to free list */
- rec->magic = TDB_FREE_MAGIC;
-
- if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
- rec_write(tdb, offset, rec) == -1 ||
- ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
- TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
- goto fail;
- }
-
- /* And we're done. */
- tdb_unlock(tdb, -1, F_WRLCK);
- return 0;
-
- fail:
- tdb_unlock(tdb, -1, F_WRLCK);
- return -1;
-}
-
-
-/* expand a file. we prefer to use ftruncate, as that is what posix
- says to use for mmap expansion */
-static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
-{
- char buf[1024];
-#if HAVE_FTRUNCATE_EXTEND
- if (ftruncate(tdb->fd, size+addition) != 0) {
- TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n",
- size+addition, strerror(errno)));
- return -1;
- }
-#else
- char b = 0;
-
-#ifdef HAVE_PWRITE
- if (pwrite(tdb->fd, &b, 1, (size+addition) - 1) != 1) {
-#else
- if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 ||
- write(tdb->fd, &b, 1) != 1) {
-#endif
- TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n",
- size+addition, strerror(errno)));
- return -1;
- }
-#endif
-
- /* now fill the file with something. This ensures that the file isn't sparse, which would be
- very bad if we ran out of disk. This must be done with write, not via mmap */
- memset(buf, 0x42, sizeof(buf));
- while (addition) {
- int n = addition>sizeof(buf)?sizeof(buf):addition;
-#ifdef HAVE_PWRITE
- int ret = pwrite(tdb->fd, buf, n, size);
-#else
- int ret;
- if (lseek(tdb->fd, size, SEEK_SET) != size)
- return -1;
- ret = write(tdb->fd, buf, n);
-#endif
- if (ret != n) {
- TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n",
- n, strerror(errno)));
- return -1;
- }
- addition -= n;
- size += n;
- }
- return 0;
-}
-
-
-/* expand the database at least size bytes by expanding the underlying
- file and doing the mmap again if necessary */
-static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
+static void tdb_increment_seqnum(struct tdb_context *tdb)
{
- struct list_struct rec;
- tdb_off offset;
-
- if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
- TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
- return -1;
- }
-
- /* must know about any previous expansions by another process */
- tdb_oob(tdb, tdb->map_size + 1, 1);
-
- /* always make room for at least 10 more records, and round
- the database up to a multiple of TDB_PAGE_SIZE */
- size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size;
-
- if (!(tdb->flags & TDB_INTERNAL))
- tdb_munmap(tdb);
-
- /*
- * We must ensure the file is unmapped before doing this
- * to ensure consistency with systems like OpenBSD where
- * writes and mmaps are not consistent.
- */
-
- /* expand the file itself */
- if (!(tdb->flags & TDB_INTERNAL)) {
- if (expand_file(tdb, tdb->map_size, size) != 0)
- goto fail;
+ if (!(tdb->flags & TDB_SEQNUM)) {
+ return;
}
- tdb->map_size += size;
-
- if (tdb->flags & TDB_INTERNAL)
- tdb->map_ptr = realloc(tdb->map_ptr, tdb->map_size);
- else {
- /*
- * We must ensure the file is remapped before adding the space
- * to ensure consistency with systems like OpenBSD where
- * writes and mmaps are not consistent.
- */
-
- /* We're ok if the mmap fails as we'll fallback to read/write */
- tdb_mmap(tdb);
+ if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1, 1) != 0) {
+ return;
}
- /* form a new freelist record */
- memset(&rec,'\0',sizeof(rec));
- rec.rec_len = size - sizeof(rec);
+ tdb_increment_seqnum_nonblock(tdb);
- /* link it into the free list */
- offset = tdb->map_size - size;
- if (tdb_free(tdb, offset, &rec) == -1)
- goto fail;
-
- tdb_unlock(tdb, -1, F_WRLCK);
- return 0;
- fail:
- tdb_unlock(tdb, -1, F_WRLCK);
- return -1;
+ tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1, 1);
}
-/* allocate some space from the free list. The offset returned points
- to a unconnected list_struct within the database with room for at
- least length bytes of total data
-
- 0 is returned if the space could not be allocated
- */
-static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
- struct list_struct *rec)
+static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
{
- tdb_off rec_ptr, last_ptr, newrec_ptr;
- struct list_struct newrec;
-
- if (tdb_lock(tdb, -1, F_WRLCK) == -1)
- return 0;
-
- /* Extra bytes required for tailer */
- length += sizeof(tdb_off);
-
- again:
- last_ptr = FREELIST_TOP;
-
- /* read in the freelist top */
- if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
- goto fail;
-
- /* keep looking until we find a freelist record big enough */
- while (rec_ptr) {
- if (rec_free_read(tdb, rec_ptr, rec) == -1)
- goto fail;
-
- if (rec->rec_len >= length) {
- /* found it - now possibly split it up */
- if (rec->rec_len > length + MIN_REC_SIZE) {
- /* Length of left piece */
- length = TDB_ALIGN(length, TDB_ALIGNMENT);
-
- /* Right piece to go on free list */
- newrec.rec_len = rec->rec_len
- - (sizeof(*rec) + length);
- newrec_ptr = rec_ptr + sizeof(*rec) + length;
-
- /* And left record is shortened */
- rec->rec_len = length;
- } else
- newrec_ptr = 0;
-
- /* Remove allocated record from the free list */
- if (ofs_write(tdb, last_ptr, &rec->next) == -1)
- goto fail;
-
- /* Update header: do this before we drop alloc
- lock, otherwise tdb_free() might try to
- merge with us, thinking we're free.
- (Thanks Jeremy Allison). */
- rec->magic = TDB_MAGIC;
- if (rec_write(tdb, rec_ptr, rec) == -1)
- goto fail;
-
- /* Did we create new block? */
- if (newrec_ptr) {
- /* Update allocated record tailer (we
- shortened it). */
- if (update_tailer(tdb, rec_ptr, rec) == -1)
- goto fail;
-
- /* Free new record */
- if (tdb_free(tdb, newrec_ptr, &newrec) == -1)
- goto fail;
- }
-
- /* all done - return the new record offset */
- tdb_unlock(tdb, -1, F_WRLCK);
- return rec_ptr;
- }
- /* move to the next record */
- last_ptr = rec_ptr;
- rec_ptr = rec->next;
- }
- /* we didn't find enough space. See if we can expand the
- database and if we can then try again */
- if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
- goto again;
- fail:
- tdb_unlock(tdb, -1, F_WRLCK);
- return 0;
-}
-
-/* initialise a new database with a specified hash size */
-static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
-{
- struct tdb_header *newdb;
- int size, ret = -1;
-
- /* We make it up in memory, then write it out if not internal */
- size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
- if (!(newdb = calloc(size, 1)))
- return TDB_ERRCODE(TDB_ERR_OOM, -1);
-
- /* Fill in the header */
- newdb->version = TDB_VERSION;
- newdb->hash_size = hash_size;
-#ifdef USE_SPINLOCKS
- newdb->rwlocks = size;
-#endif
- if (tdb->flags & TDB_INTERNAL) {
- tdb->map_size = size;
- tdb->map_ptr = (char *)newdb;
- memcpy(&tdb->header, newdb, sizeof(tdb->header));
- /* Convert the `ondisk' version if asked. */
- CONVERT(*newdb);
- return 0;
- }
- if (lseek(tdb->fd, 0, SEEK_SET) == -1)
- goto fail;
-
- if (ftruncate(tdb->fd, 0) == -1)
- goto fail;
-
- /* This creates an endian-converted header, as if read from disk */
- CONVERT(*newdb);
- memcpy(&tdb->header, newdb, sizeof(tdb->header));
- /* Don't endian-convert the magic food! */
- memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
- if (write(tdb->fd, newdb, size) != size)
- ret = -1;
- else
- ret = tdb_create_rwlocks(tdb->fd, hash_size);
-
- fail:
- SAFE_FREE(newdb);
- return ret;
+ return memcmp(data.dptr, key.dptr, data.dsize);
}
/* Returns 0 on fail. On success, return offset of record, and fills
in rec */
-static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
- struct list_struct *r)
+static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
+ struct tdb_record *r)
{
- tdb_off rec_ptr;
+ tdb_off_t rec_ptr;
/* read in the hash top */
- if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
return 0;
/* keep looking until we find the right record */
while (rec_ptr) {
- if (rec_read(tdb, rec_ptr, r) == -1)
+ if (tdb_rec_read(tdb, rec_ptr, r) == -1)
return 0;
- if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) {
- char *k;
- /* a very likely hit - read the key */
- k = tdb_alloc_read(tdb, rec_ptr + sizeof(*r),
- r->key_len);
- if (!k)
- return 0;
-
- if (memcmp(key.dptr, k, key.dsize) == 0) {
- SAFE_FREE(k);
- return rec_ptr;
- }
- SAFE_FREE(k);
+ if (!TDB_DEAD(r) && hash==r->full_hash
+ && key.dsize==r->key_len
+ && tdb_parse_data(tdb, key, rec_ptr + sizeof(*r),
+ r->key_len, tdb_key_compare,
+ NULL) == 0) {
+ return rec_ptr;
+ }
+ /* detect tight infinite loop */
+ if (rec_ptr == r->next) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_find: loop detected.\n"));
+ return 0;
}
rec_ptr = r->next;
}
- return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
-}
-
-/* If they do lockkeys, check that this hash is one they locked */
-static int tdb_keylocked(TDB_CONTEXT *tdb, u32 hash)
-{
- u32 i;
- if (!tdb->lockedkeys)
- return 1;
- for (i = 0; i < tdb->lockedkeys[0]; i++)
- if (tdb->lockedkeys[i+1] == hash)
- return 1;
- return TDB_ERRCODE(TDB_ERR_NOLOCK, 0);
+ tdb->ecode = TDB_ERR_NOEXIST;
+ return 0;
}
/* As tdb_find, but if you succeed, keep the lock */
-static tdb_off tdb_find_lock(TDB_CONTEXT *tdb, TDB_DATA key, int locktype,
- struct list_struct *rec)
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
+ struct tdb_record *rec)
{
- u32 hash, rec_ptr;
+ uint32_t rec_ptr;
- hash = tdb_hash(&key);
- if (!tdb_keylocked(tdb, hash))
- return 0;
if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
return 0;
if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
return rec_ptr;
}
-enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
-{
- return tdb->ecode;
-}
-
-static struct tdb_errname {
- enum TDB_ERROR ecode; const char *estring;
-} emap[] = { {TDB_SUCCESS, "Success"},
- {TDB_ERR_CORRUPT, "Corrupt database"},
- {TDB_ERR_IO, "IO Error"},
- {TDB_ERR_LOCK, "Locking error"},
- {TDB_ERR_OOM, "Out of memory"},
- {TDB_ERR_EXISTS, "Record exists"},
- {TDB_ERR_NOLOCK, "Lock exists on other keys"},
- {TDB_ERR_NOEXIST, "Record does not exist"} };
-
-/* Error string for the last tdb error */
-const char *tdb_errorstr(TDB_CONTEXT *tdb)
-{
- u32 i;
- for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
- if (tdb->ecode == emap[i].ecode)
- return emap[i].estring;
- return "Invalid error code";
-}
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
/* update an entry in place - this only works if the new data size
is <= the old data size and the key exists.
on failure return -1.
*/
-
-static int tdb_update(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf)
+static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, TDB_DATA dbuf)
{
- struct list_struct rec;
- tdb_off rec_ptr;
+ struct tdb_record rec;
+ tdb_off_t rec_ptr;
/* find entry */
- if (!(rec_ptr = tdb_find(tdb, key, tdb_hash(&key), &rec)))
+ if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
return -1;
+ /* it could be an exact duplicate of what is there - this is
+ * surprisingly common (eg. with a ldb re-index). */
+ if (rec.key_len == key.dsize &&
+ rec.data_len == dbuf.dsize &&
+ rec.full_hash == hash) {
+ TDB_DATA data = _tdb_fetch(tdb, key);
+ if (data.dsize == dbuf.dsize &&
+ memcmp(data.dptr, dbuf.dptr, data.dsize) == 0) {
+ if (data.dptr) {
+ free(data.dptr);
+ }
+ return 0;
+ }
+ if (data.dptr) {
+ free(data.dptr);
+ }
+ }
+
+
/* must be long enough key, data and tailer */
- if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
+ if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off_t)) {
tdb->ecode = TDB_SUCCESS; /* Not really an error */
return -1;
}
- if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+ if (tdb->methods->tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
dbuf.dptr, dbuf.dsize) == -1)
return -1;
if (dbuf.dsize != rec.data_len) {
/* update size */
rec.data_len = dbuf.dsize;
- return rec_write(tdb, rec_ptr, &rec);
+ return tdb_rec_write(tdb, rec_ptr, &rec);
}
return 0;
/* find an entry in the database given a key */
/* If an entry doesn't exist tdb_err will be set to
* TDB_ERR_NOEXIST. If a key has no data attached
- * tdb_err will not be set. Both will return a
- * zero pptr and zero dsize.
+ * then the TDB_DATA will have zero length but
+ * a non-zero pointer
*/
-
-TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
{
- tdb_off rec_ptr;
- struct list_struct rec;
+ tdb_off_t rec_ptr;
+ struct tdb_record rec;
TDB_DATA ret;
+ uint32_t hash;
/* find which hash bucket it is in */
- if (!(rec_ptr = tdb_find_lock(tdb,key,F_RDLCK,&rec)))
+ hash = tdb->hash_fn(&key);
+ if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
return tdb_null;
- if (rec.data_len)
- ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
- rec.data_len);
- else
- ret.dptr = NULL;
+ ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+ rec.data_len);
ret.dsize = rec.data_len;
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
return ret;
}
+TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
+{
+ TDB_DATA ret = _tdb_fetch(tdb, key);
+
+ tdb_trace_1rec_retrec(tdb, "tdb_fetch", key, ret);
+ return ret;
+}
+
+/*
+ * Find an entry in the database and hand the record's data to a parsing
+ * function. The parsing function is executed under the chain read lock, so it
+ * should be fast and should not block on other syscalls.
+ *
+ * DONT CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
+ *
+ * For mmapped tdb's that do not have a transaction open it points the parsing
+ * function directly at the mmap area, it avoids the malloc/memcpy in this
+ * case. If a transaction is open or no mmap is available, it has to do
+ * malloc/read/parse/free.
+ *
+ * This is interesting for all readers of potentially large data structures in
+ * the tdb records, ldb indexes being one example.
+ */
+
+int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
+ int (*parser)(TDB_DATA key, TDB_DATA data,
+ void *private_data),
+ void *private_data)
+{
+ tdb_off_t rec_ptr;
+ struct tdb_record rec;
+ int ret;
+ uint32_t hash;
+
+ /* find which hash bucket it is in */
+ hash = tdb->hash_fn(&key);
+
+ if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
+ tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, -1);
+ tdb->ecode = TDB_ERR_NOEXIST;
+ return 0;
+ }
+ tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, 0);
+
+ ret = tdb_parse_data(tdb, key, rec_ptr + sizeof(rec) + rec.key_len,
+ rec.data_len, parser, private_data);
+
+ tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+
+ return ret;
+}
+
/* check if an entry in the database exists
note that 1 is returned if the key is found and 0 is returned if not found
this doesn't match the conventions in the rest of this module, but is
compatible with gdbm
*/
-int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
+static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
- struct list_struct rec;
+ struct tdb_record rec;
- if (tdb_find_lock(tdb, key, F_RDLCK, &rec) == 0)
+ if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
return 0;
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
return 1;
}
-/* record lock stops delete underneath */
-static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
- return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
-}
-/*
- Write locks override our own fcntl readlocks, so check it here.
- Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
- an error to fail to get the lock here.
-*/
-
-static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
- struct tdb_traverse_lock *i;
- for (i = &tdb->travlocks; i; i = i->next)
- if (i->off == off)
- return -1;
- return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
-}
-
-/*
- Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
- an error to fail to get the lock here.
-*/
-
-static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
+int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
{
- return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
-}
-/* fcntl locks don't stack: avoid unlocking someone else's */
-static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
-{
- struct tdb_traverse_lock *i;
- u32 count = 0;
+ uint32_t hash = tdb->hash_fn(&key);
+ int ret;
- if (off == 0)
- return 0;
- for (i = &tdb->travlocks; i; i = i->next)
- if (i->off == off)
- count++;
- return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
+ ret = tdb_exists_hash(tdb, key, hash);
+ tdb_trace_1rec_ret(tdb, "tdb_exists", key, ret);
+ return ret;
}
/* actually delete an entry in the database given the offset */
-static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
+int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct tdb_record *rec)
{
- tdb_off last_ptr, i;
- struct list_struct lastrec;
+ tdb_off_t last_ptr, i;
+ struct tdb_record lastrec;
- if (tdb->read_only) return -1;
+ if (tdb->read_only || tdb->traverse_read) return -1;
- if (write_lock_record(tdb, rec_ptr) == -1) {
+ if (((tdb->traverse_write != 0) && (!TDB_DEAD(rec))) ||
+ tdb_write_lock_record(tdb, rec_ptr) == -1) {
/* Someone traversing here: mark it as dead */
rec->magic = TDB_DEAD_MAGIC;
- return rec_write(tdb, rec_ptr, rec);
+ return tdb_rec_write(tdb, rec_ptr, rec);
}
- if (write_unlock_record(tdb, rec_ptr) != 0)
+ if (tdb_write_unlock_record(tdb, rec_ptr) != 0)
return -1;
/* find previous record in hash chain */
- if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
return -1;
for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
- if (rec_read(tdb, i, &lastrec) == -1)
+ if (tdb_rec_read(tdb, i, &lastrec) == -1)
return -1;
/* unlink it: next ptr is at start of record. */
if (last_ptr == 0)
last_ptr = TDB_HASH_TOP(rec->full_hash);
- if (ofs_write(tdb, last_ptr, &rec->next) == -1)
+ if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1)
return -1;
/* recover the space */
return 0;
}
-/* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
-static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
- struct list_struct *rec)
+static int tdb_count_dead(struct tdb_context *tdb, uint32_t hash)
{
- int want_next = (tlock->off != 0);
-
- /* No traversal allows if you've called tdb_lockkeys() */
- if (tdb->lockedkeys)
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
-
- /* Lock each chain from the start one. */
- for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
- if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
- return -1;
-
- /* No previous record? Start at top of chain. */
- if (!tlock->off) {
- if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
- &tlock->off) == -1)
- goto fail;
- } else {
- /* Otherwise unlock the previous record. */
- if (unlock_record(tdb, tlock->off) != 0)
- goto fail;
- }
+ int res = 0;
+ tdb_off_t rec_ptr;
+ struct tdb_record rec;
+
+ /* read in the hash top */
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+ return 0;
- if (want_next) {
- /* We have offset of old record: grab next */
- if (rec_read(tdb, tlock->off, rec) == -1)
- goto fail;
- tlock->off = rec->next;
- }
+ while (rec_ptr) {
+ if (tdb_rec_read(tdb, rec_ptr, &rec) == -1)
+ return 0;
- /* Iterate through chain */
- while( tlock->off) {
- tdb_off current;
- if (rec_read(tdb, tlock->off, rec) == -1)
- goto fail;
- if (!TDB_DEAD(rec)) {
- /* Woohoo: we found one! */
- if (lock_record(tdb, tlock->off) != 0)
- goto fail;
- return tlock->off;
- }
- /* Try to clean dead ones from old traverses */
- current = tlock->off;
- tlock->off = rec->next;
- if (do_delete(tdb, current, rec) != 0)
- goto fail;
+ if (rec.magic == TDB_DEAD_MAGIC) {
+ res += 1;
}
- tdb_unlock(tdb, tlock->hash, F_WRLCK);
- want_next = 0;
+ rec_ptr = rec.next;
}
- /* We finished iteration without finding anything */
- return TDB_ERRCODE(TDB_SUCCESS, 0);
-
- fail:
- tlock->off = 0;
- if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0)
- TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
- return -1;
+ return res;
}
-/* traverse the entire database - calling fn(tdb, key, data) on each element.
- return -1 on error or the record count traversed
- if fn is NULL then it is not called
- a non-zero return value from fn() indicates that the traversal should stop
- */
-int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *state)
+/*
+ * Purge all DEAD records from a hash chain
+ */
+static int tdb_purge_dead(struct tdb_context *tdb, uint32_t hash)
{
- TDB_DATA key, dbuf;
- struct list_struct rec;
- struct tdb_traverse_lock tl = { NULL, 0, 0 };
- int ret, count = 0;
+ int res = -1;
+ struct tdb_record rec;
+ tdb_off_t rec_ptr;
- /* This was in the initializaton, above, but the IRIX compiler
- * did not like it. crh
- */
- tl.next = tdb->travlocks.next;
-
- /* fcntl locks don't stack: beware traverse inside traverse */
- tdb->travlocks.next = &tl;
-
- /* tdb_next_lock places locks on the record returned, and its chain */
- while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
- count++;
- /* now read the full record */
- key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec),
- rec.key_len + rec.data_len);
- if (!key.dptr) {
- ret = -1;
- if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0)
- goto out;
- if (unlock_record(tdb, tl.off) != 0)
- TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
- goto out;
- }
- key.dsize = rec.key_len;
- dbuf.dptr = key.dptr + rec.key_len;
- dbuf.dsize = rec.data_len;
-
- /* Drop chain lock, call out */
- if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) {
- ret = -1;
- goto out;
+ if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+ return -1;
+ }
+
+ /* read in the hash top */
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+ goto fail;
+
+ while (rec_ptr) {
+ tdb_off_t next;
+
+ if (tdb_rec_read(tdb, rec_ptr, &rec) == -1) {
+ goto fail;
}
- if (fn && fn(tdb, key, dbuf, state)) {
- /* They want us to terminate traversal */
- ret = count;
- if (unlock_record(tdb, tl.off) != 0) {
- TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));;
- ret = -1;
- }
- tdb->travlocks.next = tl.next;
- SAFE_FREE(key.dptr);
- return count;
+
+ next = rec.next;
+
+ if (rec.magic == TDB_DEAD_MAGIC
+ && tdb_do_delete(tdb, rec_ptr, &rec) == -1) {
+ goto fail;
}
- SAFE_FREE(key.dptr);
+ rec_ptr = next;
}
-out:
- tdb->travlocks.next = tl.next;
- if (ret < 0)
- return -1;
- else
- return count;
+ res = 0;
+ fail:
+ tdb_unlock(tdb, -1, F_WRLCK);
+ return res;
}
-/* find the first entry in the database and return its key */
-TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
+/* delete an entry in the database given a key */
+static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
- TDB_DATA key;
- struct list_struct rec;
+ tdb_off_t rec_ptr;
+ struct tdb_record rec;
+ int ret;
- /* release any old lock */
- if (unlock_record(tdb, tdb->travlocks.off) != 0)
- return tdb_null;
- tdb->travlocks.off = tdb->travlocks.hash = 0;
+ if (tdb->max_dead_records != 0) {
- if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
- return tdb_null;
- /* now read the key */
- key.dsize = rec.key_len;
- key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
- if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
- TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
- return key;
-}
+ /*
+ * Allow for some dead records per hash chain, mainly for
+ * tdb's with a very high create/delete rate like locking.tdb.
+ */
-/* find the next entry in the database, returning its key */
-TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
-{
- u32 oldhash;
- TDB_DATA key = tdb_null;
- struct list_struct rec;
- char *k = NULL;
-
- /* Is locked key the old key? If so, traverse will be reliable. */
- if (tdb->travlocks.off) {
- if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
- return tdb_null;
- if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
- || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
- rec.key_len))
- || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
- /* No, it wasn't: unlock it and start from scratch */
- if (unlock_record(tdb, tdb->travlocks.off) != 0)
- return tdb_null;
- if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
- return tdb_null;
- tdb->travlocks.off = 0;
+ if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+ return -1;
+
+ if (tdb_count_dead(tdb, hash) >= tdb->max_dead_records) {
+ /*
+ * Don't let the per-chain freelist grow too large,
+ * delete all existing dead records
+ */
+ tdb_purge_dead(tdb, hash);
+ }
+
+ if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
+ tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+ return -1;
}
- SAFE_FREE(k);
+ /*
+ * Just mark the record as dead.
+ */
+ rec.magic = TDB_DEAD_MAGIC;
+ ret = tdb_rec_write(tdb, rec_ptr, &rec);
}
+ else {
+ if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK,
+ &rec)))
+ return -1;
- if (!tdb->travlocks.off) {
- /* No previous element: do normal find, and lock record */
- tdb->travlocks.off = tdb_find_lock(tdb, oldkey, F_WRLCK, &rec);
- if (!tdb->travlocks.off)
- return tdb_null;
- tdb->travlocks.hash = BUCKET(rec.full_hash);
- if (lock_record(tdb, tdb->travlocks.off) != 0) {
- TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
- return tdb_null;
- }
+ ret = tdb_do_delete(tdb, rec_ptr, &rec);
}
- oldhash = tdb->travlocks.hash;
-
- /* Grab next record: locks chain and returned record,
- unlocks old record */
- if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
- key.dsize = rec.key_len;
- key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
- key.dsize);
- /* Unlock the chain of this new record */
- if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
- TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+
+ if (ret == 0) {
+ tdb_increment_seqnum(tdb);
}
- /* Unlock the chain of old record */
- if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
- TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
- return key;
+
+ if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_delete: WARNING tdb_unlock failed!\n"));
+ return ret;
}
-/* delete an entry in the database given a key */
-int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
+int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
{
- tdb_off rec_ptr;
- struct list_struct rec;
+ uint32_t hash = tdb->hash_fn(&key);
int ret;
- if (!(rec_ptr = tdb_find_lock(tdb, key, F_WRLCK, &rec)))
- return -1;
- ret = do_delete(tdb, rec_ptr, &rec);
- if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
- TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
+ ret = tdb_delete_hash(tdb, key, hash);
+ tdb_trace_1rec_ret(tdb, "tdb_delete", key, ret);
return ret;
}
-/* store an element in the database, replacing any existing element
- with the same key
+/*
+ * See if we have a dead record around with enough space
+ */
+static tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
+ struct tdb_record *r, tdb_len_t length)
+{
+ tdb_off_t rec_ptr;
+
+ /* read in the hash top */
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+ return 0;
+
+ /* keep looking until we find the right record */
+ while (rec_ptr) {
+ if (tdb_rec_read(tdb, rec_ptr, r) == -1)
+ return 0;
+
+ if (TDB_DEAD(r) && r->rec_len >= length) {
+ /*
+ * First fit for simple coding, TODO: change to best
+ * fit
+ */
+ return rec_ptr;
+ }
+ rec_ptr = r->next;
+ }
+ return 0;
+}
- return 0 on success, -1 on failure
-*/
-int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
+static int _tdb_store(struct tdb_context *tdb, TDB_DATA key,
+ TDB_DATA dbuf, int flag, uint32_t hash)
{
- struct list_struct rec;
- u32 hash;
- tdb_off rec_ptr;
+ struct tdb_record rec;
+ tdb_off_t rec_ptr;
char *p = NULL;
- int ret = 0;
-
- /* find which hash bucket it is in */
- hash = tdb_hash(&key);
- if (!tdb_keylocked(tdb, hash))
- return -1;
- if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
- return -1;
+ int ret = -1;
/* check for it existing, on insert. */
if (flag == TDB_INSERT) {
- if (tdb_exists(tdb, key)) {
+ if (tdb_exists_hash(tdb, key, hash)) {
tdb->ecode = TDB_ERR_EXISTS;
goto fail;
}
} else {
/* first try in-place update, on modify or replace. */
- if (tdb_update(tdb, key, dbuf) == 0)
- goto out;
- if (flag == TDB_MODIFY && tdb->ecode == TDB_ERR_NOEXIST)
+ if (tdb_update_hash(tdb, key, hash, dbuf) == 0) {
+ goto done;
+ }
+ if (tdb->ecode == TDB_ERR_NOEXIST &&
+ flag == TDB_MODIFY) {
+ /* if the record doesn't exist and we are in TDB_MODIFY mode then
+ we should fail the store */
goto fail;
+ }
}
/* reset the error code potentially set by the tdb_update() */
tdb->ecode = TDB_SUCCESS;
care. Doing this first reduces fragmentation, and avoids
coalescing with `allocated' block before it's updated. */
if (flag != TDB_INSERT)
- tdb_delete(tdb, key);
+ tdb_delete_hash(tdb, key, hash);
/* Copy key+value *before* allocating free space in case malloc
fails and we are left with a dead spot in the tdb. */
if (dbuf.dsize)
memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
- /* now we're into insert / modify / replace of a record which
- * we know could not be optimised by an in-place store (for
- * various reasons). */
- if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
+ if (tdb->max_dead_records != 0) {
+ /*
+ * Allow for some dead records per hash chain, look if we can
+ * find one that can hold the new record. We need enough space
+ * for key, data and tailer. If we find one, we don't have to
+ * consult the central freelist.
+ */
+ rec_ptr = tdb_find_dead(
+ tdb, hash, &rec,
+ key.dsize + dbuf.dsize + sizeof(tdb_off_t));
+
+ if (rec_ptr != 0) {
+ rec.key_len = key.dsize;
+ rec.data_len = dbuf.dsize;
+ rec.full_hash = hash;
+ rec.magic = TDB_MAGIC;
+ if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
+ || tdb->methods->tdb_write(
+ tdb, rec_ptr + sizeof(rec),
+ p, key.dsize + dbuf.dsize) == -1) {
+ goto fail;
+ }
+ goto done;
+ }
+ }
+
+ /*
+ * We have to allocate some space from the freelist, so this means we
+ * have to lock it. Use the chance to purge all the DEAD records from
+ * the hash chain under the freelist lock.
+ */
+
+ if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+ goto fail;
+ }
+
+ if ((tdb->max_dead_records != 0)
+ && (tdb_purge_dead(tdb, hash) == -1)) {
+ tdb_unlock(tdb, -1, F_WRLCK);
+ goto fail;
+ }
+
+ /* we have to allocate some space */
+ rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec);
+
+ tdb_unlock(tdb, -1, F_WRLCK);
+
+ if (rec_ptr == 0) {
goto fail;
+ }
/* Read hash top into next ptr */
- if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
goto fail;
rec.key_len = key.dsize;
rec.magic = TDB_MAGIC;
/* write out and point the top of the hash chain at it */
- if (rec_write(tdb, rec_ptr, &rec) == -1
- || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
- || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
+ if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
+ || tdb->methods->tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
+ || tdb_ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
/* Need to tdb_unallocate() here */
goto fail;
}
- out:
+
+ done:
+ ret = 0;
+ fail:
+ if (ret == 0) {
+ tdb_increment_seqnum(tdb);
+ }
+
SAFE_FREE(p);
- tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
return ret;
-fail:
- ret = -1;
- goto out;
}
-/* Attempt to append data to an entry in place - this only works if the new data size
- is <= the old data size and the key exists.
- on failure return -1. Record must be locked before calling.
+/* store an element in the database, replacing any existing element
+ with the same key
+
+ return 0 on success, -1 on failure
*/
-static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
{
- struct list_struct rec;
- tdb_off rec_ptr;
-
- /* find entry */
- if (!(rec_ptr = tdb_find(tdb, key, tdb_hash(&key), &rec)))
- return -1;
-
- /* Append of 0 is always ok. */
- if (new_dbuf.dsize == 0)
- return 0;
+ uint32_t hash;
+ int ret;
- /* must be long enough for key, old data + new data and tailer */
- if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + sizeof(tdb_off)) {
- /* No room. */
- tdb->ecode = TDB_SUCCESS; /* Not really an error */
+ if (tdb->read_only || tdb->traverse_read) {
+ tdb->ecode = TDB_ERR_RDONLY;
+ tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, -1);
return -1;
}
- if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len,
- new_dbuf.dptr, new_dbuf.dsize) == -1)
+ /* find which hash bucket it is in */
+ hash = tdb->hash_fn(&key);
+ if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
return -1;
- /* update size */
- rec.data_len += new_dbuf.dsize;
- return rec_write(tdb, rec_ptr, &rec);
+ ret = _tdb_store(tdb, key, dbuf, flag, hash);
+ tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, ret);
+ tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+ return ret;
}
/* Append to an entry. Create if not exist. */
-
-int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
{
- struct list_struct rec;
- u32 hash;
- tdb_off rec_ptr;
- char *p = NULL;
- int ret = 0;
- size_t new_data_size = 0;
+ uint32_t hash;
+ TDB_DATA dbuf;
+ int ret = -1;
/* find which hash bucket it is in */
- hash = tdb_hash(&key);
- if (!tdb_keylocked(tdb, hash))
- return -1;
+ hash = tdb->hash_fn(&key);
if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
return -1;
- /* first try in-place. */
- if (tdb_append_inplace(tdb, key, new_dbuf) == 0)
- goto out;
-
- /* reset the error code potentially set by the tdb_append_inplace() */
- tdb->ecode = TDB_SUCCESS;
-
- /* find entry */
- if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
- if (tdb->ecode != TDB_ERR_NOEXIST)
- goto fail;
+ dbuf = _tdb_fetch(tdb, key);
- /* Not found - create. */
-
- ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT);
- goto out;
+ if (dbuf.dptr == NULL) {
+ dbuf.dptr = (unsigned char *)malloc(new_dbuf.dsize);
+ } else {
+ unsigned int new_len = dbuf.dsize + new_dbuf.dsize;
+ unsigned char *new_dptr;
+
+ /* realloc '0' is special: don't do that. */
+ if (new_len == 0)
+ new_len = 1;
+ new_dptr = (unsigned char *)realloc(dbuf.dptr, new_len);
+ if (new_dptr == NULL) {
+ free(dbuf.dptr);
+ }
+ dbuf.dptr = new_dptr;
}
- new_data_size = rec.data_len + new_dbuf.dsize;
-
- /* Copy key+old_value+value *before* allocating free space in case malloc
- fails and we are left with a dead spot in the tdb. */
-
- if (!(p = (char *)malloc(key.dsize + new_data_size))) {
+ if (dbuf.dptr == NULL) {
tdb->ecode = TDB_ERR_OOM;
- goto fail;
+ goto failed;
}
- /* Copy the key in place. */
- memcpy(p, key.dptr, key.dsize);
+ memcpy(dbuf.dptr + dbuf.dsize, new_dbuf.dptr, new_dbuf.dsize);
+ dbuf.dsize += new_dbuf.dsize;
- /* Now read the old data into place. */
- if (rec.data_len &&
- tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + key.dsize, rec.data_len, 0) == -1)
- goto fail;
+ ret = _tdb_store(tdb, key, dbuf, 0, hash);
+ tdb_trace_2rec_retrec(tdb, "tdb_append", key, new_dbuf, dbuf);
+
+failed:
+ tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+ SAFE_FREE(dbuf.dptr);
+ return ret;
+}
- /* Finally append the new data. */
- if (new_dbuf.dsize)
- memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize);
- /* delete any existing record - if it doesn't exist we don't
- care. Doing this first reduces fragmentation, and avoids
- coalescing with `allocated' block before it's updated. */
+/*
+ return the name of the current tdb file
+ useful for external logging functions
+*/
+const char *tdb_name(struct tdb_context *tdb)
+{
+ return tdb->name;
+}
- tdb_delete(tdb, key);
+/*
+ return the underlying file descriptor being used by tdb, or -1
+ useful for external routines that want to check the device/inode
+ of the fd
+*/
+int tdb_fd(struct tdb_context *tdb)
+{
+ return tdb->fd;
+}
- if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec)))
- goto fail;
+/*
+ return the current logging function
+ useful for external tdb routines that wish to log tdb errors
+*/
+tdb_log_func tdb_log_fn(struct tdb_context *tdb)
+{
+ return tdb->log.log_fn;
+}
- /* Read hash top into next ptr */
- if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
- goto fail;
- rec.key_len = key.dsize;
- rec.data_len = new_data_size;
- rec.full_hash = hash;
- rec.magic = TDB_MAGIC;
+/*
+ get the tdb sequence number. Only makes sense if the writers opened
+ with TDB_SEQNUM set. Note that this sequence number will wrap quite
+ quickly, so it should only be used for a 'has something changed'
+ test, not for code that relies on the count of the number of changes
+ made. If you want a counter then use a tdb record.
+
+ The aim of this sequence number is to allow for a very lightweight
+ test of a possible tdb change.
+*/
+int tdb_get_seqnum(struct tdb_context *tdb)
+{
+ tdb_off_t seqnum=0;
- /* write out and point the top of the hash chain at it */
- if (rec_write(tdb, rec_ptr, &rec) == -1
- || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+new_data_size)==-1
- || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
- /* Need to tdb_unallocate() here */
- goto fail;
- }
+ tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+ return seqnum;
+}
- out:
- SAFE_FREE(p);
- tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
- return ret;
+int tdb_hash_size(struct tdb_context *tdb)
+{
+ return tdb->header.hash_size;
+}
-fail:
- ret = -1;
- goto out;
+size_t tdb_map_size(struct tdb_context *tdb)
+{
+ return tdb->map_size;
}
-static int tdb_already_open(dev_t device,
- ino_t ino)
+int tdb_get_flags(struct tdb_context *tdb)
{
- TDB_CONTEXT *i;
-
- for (i = tdbs; i; i = i->next) {
- if (i->device == device && i->inode == ino) {
- return 1;
- }
+ return tdb->flags;
+}
+
+void tdb_add_flags(struct tdb_context *tdb, unsigned flags)
+{
+ if ((flags & TDB_ALLOW_NESTING) &&
+ (flags & TDB_DISALLOW_NESTING)) {
+ tdb->ecode = TDB_ERR_NESTING;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_add_flags: "
+ "allow_nesting and disallow_nesting are not allowed together!"));
+ return;
}
- return 0;
+ if (flags & TDB_ALLOW_NESTING) {
+ tdb->flags &= ~TDB_DISALLOW_NESTING;
+ }
+ if (flags & TDB_DISALLOW_NESTING) {
+ tdb->flags &= ~TDB_ALLOW_NESTING;
+ }
+
+ tdb->flags |= flags;
}
-/* open the database, creating it if necessary
+void tdb_remove_flags(struct tdb_context *tdb, unsigned flags)
+{
+ if ((flags & TDB_ALLOW_NESTING) &&
+ (flags & TDB_DISALLOW_NESTING)) {
+ tdb->ecode = TDB_ERR_NESTING;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_remove_flags: "
+ "allow_nesting and disallow_nesting are not allowed together!"));
+ return;
+ }
+
+ if (flags & TDB_ALLOW_NESTING) {
+ tdb->flags |= TDB_DISALLOW_NESTING;
+ }
+ if (flags & TDB_DISALLOW_NESTING) {
+ tdb->flags |= TDB_ALLOW_NESTING;
+ }
- The open_flags and mode are passed straight to the open call on the
- database file. A flags value of O_WRONLY is invalid. The hash size
- is advisory, use zero for a default value.
+ tdb->flags &= ~flags;
+}
- Return is NULL on error, in which case errno is also set. Don't
- try to call tdb_error or tdb_errname, just do strerror(errno).
- @param name may be NULL for internal databases. */
-TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
- int open_flags, mode_t mode)
+/*
+ enable sequence number handling on an open tdb
+*/
+void tdb_enable_seqnum(struct tdb_context *tdb)
{
- return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL);
+ tdb->flags |= TDB_SEQNUM;
}
-TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
- int open_flags, mode_t mode,
- tdb_log_func log_fn)
+/*
+ add a region of the file to the freelist. Length is the size of the region in bytes,
+ which includes the free list header that needs to be added
+ */
+static int tdb_free_region(struct tdb_context *tdb, tdb_off_t offset, ssize_t length)
{
- TDB_CONTEXT *tdb;
- struct stat st;
- int rev = 0, locked;
- unsigned char *vp;
- u32 vertest;
-
- if (!(tdb = calloc(1, sizeof *tdb))) {
- /* Can't log this */
- errno = ENOMEM;
- goto fail;
+ struct tdb_record rec;
+ if (length <= sizeof(rec)) {
+ /* the region is not worth adding */
+ return 0;
}
- tdb->fd = -1;
- tdb->name = NULL;
- tdb->map_ptr = NULL;
- tdb->lockedkeys = NULL;
- tdb->flags = tdb_flags;
- tdb->open_flags = open_flags;
- tdb->log_fn = log_fn;
-
- if ((open_flags & O_ACCMODE) == O_WRONLY) {
- TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
- name));
- errno = EINVAL;
- goto fail;
+ if (length + offset > tdb->map_size) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: adding region beyond end of file\n"));
+ return -1;
}
-
- if (hash_size == 0)
- hash_size = DEFAULT_HASH_SIZE;
- if ((open_flags & O_ACCMODE) == O_RDONLY) {
- tdb->read_only = 1;
- /* read only databases don't do locking or clear if first */
- tdb->flags |= TDB_NOLOCK;
- tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+ memset(&rec,'\0',sizeof(rec));
+ rec.rec_len = length - sizeof(rec);
+ if (tdb_free(tdb, offset, &rec) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: failed to add free record\n"));
+ return -1;
}
+ return 0;
+}
- /* internal databases don't mmap or lock, and start off cleared */
- if (tdb->flags & TDB_INTERNAL) {
- tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
- tdb->flags &= ~TDB_CLEAR_IF_FIRST;
- if (tdb_new_database(tdb, hash_size) != 0) {
- TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!"));
- goto fail;
- }
- goto internal;
+/*
+ wipe the entire database, deleting all records. This can be done
+ very fast by using a global lock. The entire data portion of the
+ file becomes a single entry in the freelist.
+
+ This code carefully steps around the recovery area, leaving it alone
+ */
+int tdb_wipe_all(struct tdb_context *tdb)
+{
+ int i;
+ tdb_off_t offset = 0;
+ ssize_t data_len;
+ tdb_off_t recovery_head;
+ tdb_len_t recovery_size = 0;
+
+ if (tdb_lockall(tdb) != 0) {
+ return -1;
}
- if ((tdb->fd = open(name, open_flags, mode)) == -1) {
- TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
- name, strerror(errno)));
- goto fail; /* errno set by open(2) */
+ tdb_trace(tdb, "tdb_wipe_all");
+
+ /* see if the tdb has a recovery area, and remember its size
+ if so. We don't want to lose this as otherwise each
+ tdb_wipe_all() in a transaction will increase the size of
+ the tdb by the size of the recovery area */
+ if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery head\n"));
+ goto failed;
}
- /* ensure there is only one process initialising at once */
- if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
- name, strerror(errno)));
- goto fail; /* errno set by tdb_brlock */
+ if (recovery_head != 0) {
+ struct tdb_record rec;
+ if (tdb->methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery record\n"));
+ return -1;
+ }
+ recovery_size = rec.rec_len + sizeof(rec);
}
- /* we need to zero database if we are the only one with it open */
- if ((locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))
- && (tdb_flags & TDB_CLEAR_IF_FIRST)) {
- open_flags |= O_CREAT;
- if (ftruncate(tdb->fd, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_open_ex: "
- "failed to truncate %s: %s\n",
- name, strerror(errno)));
- goto fail; /* errno set by ftruncate */
+ /* wipe the hashes */
+ for (i=0;i<tdb->header.hash_size;i++) {
+ if (tdb_ofs_write(tdb, TDB_HASH_TOP(i), &offset) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write hash %d\n", i));
+ goto failed;
}
}
- if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
- || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
- || (tdb->header.version != TDB_VERSION
- && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
- /* its not a valid database - possibly initialise it */
- if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
- errno = EIO; /* ie bad format or something */
- goto fail;
- }
- rev = (tdb->flags & TDB_CONVERT);
+ /* wipe the freelist */
+ if (tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write freelist\n"));
+ goto failed;
}
- vp = (unsigned char *)&tdb->header.version;
- vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
- (((u32)vp[2]) << 8) | (u32)vp[3];
- tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
- if (!rev)
- tdb->flags &= ~TDB_CONVERT;
- else {
- tdb->flags |= TDB_CONVERT;
- convert(&tdb->header, sizeof(tdb->header));
+
+ /* add all the rest of the file to the freelist, possibly leaving a gap
+ for the recovery area */
+ if (recovery_size == 0) {
+ /* the simple case - the whole file can be used as a freelist */
+ data_len = (tdb->map_size - TDB_DATA_START(tdb->header.hash_size));
+ if (tdb_free_region(tdb, TDB_DATA_START(tdb->header.hash_size), data_len) != 0) {
+ goto failed;
+ }
+ } else {
+ /* we need to add two freelist entries - one on either
+ side of the recovery area
+
+ Note that we cannot shift the recovery area during
+ this operation. Only the transaction.c code may
+ move the recovery area or we risk subtle data
+ corruption
+ */
+ data_len = (recovery_head - TDB_DATA_START(tdb->header.hash_size));
+ if (tdb_free_region(tdb, TDB_DATA_START(tdb->header.hash_size), data_len) != 0) {
+ goto failed;
+ }
+ /* and the 2nd free list entry after the recovery area - if any */
+ data_len = tdb->map_size - (recovery_head+recovery_size);
+ if (tdb_free_region(tdb, recovery_head+recovery_size, data_len) != 0) {
+ goto failed;
+ }
}
- if (fstat(tdb->fd, &st) == -1)
- goto fail;
- /* Is it already in the open list? If so, fail. */
- if (tdb_already_open(st.st_dev, st.st_ino)) {
- TDB_LOG((tdb, 2, "tdb_open_ex: "
- "%s (%d,%d) is already open in this process\n",
- name, st.st_dev, st.st_ino));
- errno = EBUSY;
- goto fail;
+ if (tdb_unlockall(tdb) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to unlock\n"));
+ goto failed;
}
- if (!(tdb->name = (char *)strdup(name))) {
- errno = ENOMEM;
- goto fail;
+ return 0;
+
+failed:
+ tdb_unlockall(tdb);
+ return -1;
+}
+
+struct traverse_state {
+ bool error;
+ struct tdb_context *dest_db;
+};
+
+/*
+ traverse function for repacking
+ */
+static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private_data)
+{
+ struct traverse_state *state = (struct traverse_state *)private_data;
+ if (tdb_store(state->dest_db, key, data, TDB_INSERT) != 0) {
+ state->error = true;
+ return -1;
}
+ return 0;
+}
- tdb->map_size = st.st_size;
- tdb->device = st.st_dev;
- tdb->inode = st.st_ino;
- tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
- if (!tdb->locked) {
- TDB_LOG((tdb, 2, "tdb_open_ex: "
- "failed to allocate lock structure for %s\n",
- name));
- errno = ENOMEM;
- goto fail;
+/*
+ repack a tdb
+ */
+int tdb_repack(struct tdb_context *tdb)
+{
+ struct tdb_context *tmp_db;
+ struct traverse_state state;
+
+ tdb_trace(tdb, "tdb_repack");
+
+ if (tdb_transaction_start(tdb) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to start transaction\n"));
+ return -1;
}
- tdb_mmap(tdb);
- if (locked) {
- if (!tdb->read_only)
- if (tdb_clear_spinlocks(tdb) != 0) {
- TDB_LOG((tdb, 0, "tdb_open_ex: "
- "failed to clear spinlock\n"));
- goto fail;
- }
- if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_open_ex: "
- "failed to take ACTIVE_LOCK on %s: %s\n",
- name, strerror(errno)));
- goto fail;
- }
+
+ tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
+ if (tmp_db == NULL) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to create tmp_db\n"));
+ tdb_transaction_cancel(tdb);
+ return -1;
}
- /* leave this lock in place to indicate it's in use */
- if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
- goto fail;
- internal:
- /* Internal (memory-only) databases skip all the code above to
- * do with disk files, and resume here by releasing their
- * global lock and hooking into the active list. */
- if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
- goto fail;
- tdb->next = tdbs;
- tdbs = tdb;
- return tdb;
+ state.error = false;
+ state.dest_db = tmp_db;
- fail:
- { int save_errno = errno;
+ if (tdb_traverse_read(tdb, repack_traverse, &state) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying out\n"));
+ tdb_transaction_cancel(tdb);
+ tdb_close(tmp_db);
+ return -1;
+ }
- if (!tdb)
- return NULL;
-
- if (tdb->map_ptr) {
- if (tdb->flags & TDB_INTERNAL)
- SAFE_FREE(tdb->map_ptr);
- else
- tdb_munmap(tdb);
+ if (state.error) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during traversal\n"));
+ tdb_transaction_cancel(tdb);
+ tdb_close(tmp_db);
+ return -1;
}
- SAFE_FREE(tdb->name);
- if (tdb->fd != -1)
- if (close(tdb->fd) != 0)
- TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
- SAFE_FREE(tdb->locked);
- SAFE_FREE(tdb);
- errno = save_errno;
- return NULL;
+
+ if (tdb_wipe_all(tdb) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to wipe database\n"));
+ tdb_transaction_cancel(tdb);
+ tdb_close(tmp_db);
+ return -1;
}
-}
-/**
- * Close a database.
- *
- * @returns -1 for error; 0 for success.
- **/
-int tdb_close(TDB_CONTEXT *tdb)
-{
- TDB_CONTEXT **i;
- int ret = 0;
-
- if (tdb->map_ptr) {
- if (tdb->flags & TDB_INTERNAL)
- SAFE_FREE(tdb->map_ptr);
- else
- tdb_munmap(tdb);
+ state.error = false;
+ state.dest_db = tdb;
+
+ if (tdb_traverse_read(tmp_db, repack_traverse, &state) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying back\n"));
+ tdb_transaction_cancel(tdb);
+ tdb_close(tmp_db);
+ return -1;
}
- SAFE_FREE(tdb->name);
- if (tdb->fd != -1)
- ret = close(tdb->fd);
- SAFE_FREE(tdb->locked);
- SAFE_FREE(tdb->lockedkeys);
-
- /* Remove from contexts list */
- for (i = &tdbs; *i; i = &(*i)->next) {
- if (*i == tdb) {
- *i = tdb->next;
- break;
- }
+
+ if (state.error) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during second traversal\n"));
+ tdb_transaction_cancel(tdb);
+ tdb_close(tmp_db);
+ return -1;
}
- memset(tdb, 0, sizeof(*tdb));
- SAFE_FREE(tdb);
+ tdb_close(tmp_db);
- return ret;
+ if (tdb_transaction_commit(tdb) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to commit\n"));
+ return -1;
+ }
+
+ return 0;
}
-/* lock/unlock entire database */
-int tdb_lockall(TDB_CONTEXT *tdb)
+#ifdef TDB_TRACE
+static void tdb_trace_write(struct tdb_context *tdb, const char *str)
{
- u32 i;
-
- /* There are no locks on read-only dbs */
- if (tdb->read_only)
- return TDB_ERRCODE(TDB_ERR_LOCK, -1);
- if (tdb->lockedkeys)
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
- for (i = 0; i < tdb->header.hash_size; i++)
- if (tdb_lock(tdb, i, F_WRLCK))
- break;
-
- /* If error, release locks we have... */
- if (i < tdb->header.hash_size) {
- u32 j;
-
- for ( j = 0; j < i; j++)
- tdb_unlock(tdb, j, F_WRLCK);
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
+ if (write(tdb->tracefd, str, strlen(str)) != strlen(str)) {
+ close(tdb->tracefd);
+ tdb->tracefd = -1;
}
+}
- return 0;
+static void tdb_trace_start(struct tdb_context *tdb)
+{
+ tdb_off_t seqnum=0;
+ char msg[sizeof(tdb_off_t) * 4 + 1];
+
+ tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+ snprintf(msg, sizeof(msg), "%u ", seqnum);
+ tdb_trace_write(tdb, msg);
}
-void tdb_unlockall(TDB_CONTEXT *tdb)
+
+static void tdb_trace_end(struct tdb_context *tdb)
{
- u32 i;
- for (i=0; i < tdb->header.hash_size; i++)
- tdb_unlock(tdb, i, F_WRLCK);
+ tdb_trace_write(tdb, "\n");
}
-int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[])
+static void tdb_trace_end_ret(struct tdb_context *tdb, int ret)
{
- u32 i, j, hash;
-
- /* Can't lock more keys if already locked */
- if (tdb->lockedkeys)
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
- if (!(tdb->lockedkeys = malloc(sizeof(u32) * (number+1))))
- return TDB_ERRCODE(TDB_ERR_OOM, -1);
- /* First number in array is # keys */
- tdb->lockedkeys[0] = number;
-
- /* Insertion sort by bucket */
- for (i = 0; i < number; i++) {
- hash = tdb_hash(&keys[i]);
- for (j = 0; j < i && BUCKET(tdb->lockedkeys[j+1]) < BUCKET(hash); j++);
- memmove(&tdb->lockedkeys[j+2], &tdb->lockedkeys[j+1], sizeof(u32) * (i-j));
- tdb->lockedkeys[j+1] = hash;
- }
- /* Finally, lock in order */
- for (i = 0; i < number; i++)
- if (tdb_lock(tdb, i, F_WRLCK))
- break;
-
- /* If error, release locks we have... */
- if (i < number) {
- for ( j = 0; j < i; j++)
- tdb_unlock(tdb, j, F_WRLCK);
- SAFE_FREE(tdb->lockedkeys);
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
- }
- return 0;
+ char msg[sizeof(ret) * 4 + 4];
+ snprintf(msg, sizeof(msg), " = %i\n", ret);
+ tdb_trace_write(tdb, msg);
}
-/* Unlock the keys previously locked by tdb_lockkeys() */
-void tdb_unlockkeys(TDB_CONTEXT *tdb)
+static void tdb_trace_record(struct tdb_context *tdb, TDB_DATA rec)
{
- u32 i;
- if (!tdb->lockedkeys)
+ char msg[20 + rec.dsize*2], *p;
+ unsigned int i;
+
+ /* We differentiate zero-length records from non-existent ones. */
+ if (rec.dptr == NULL) {
+ tdb_trace_write(tdb, " NULL");
return;
- for (i = 0; i < tdb->lockedkeys[0]; i++)
- tdb_unlock(tdb, tdb->lockedkeys[i+1], F_WRLCK);
- SAFE_FREE(tdb->lockedkeys);
+ }
+
+ /* snprintf here is purely cargo-cult programming. */
+ p = msg;
+ p += snprintf(p, sizeof(msg), " %zu:", rec.dsize);
+ for (i = 0; i < rec.dsize; i++)
+ p += snprintf(p, 2, "%02x", rec.dptr[i]);
+
+ tdb_trace_write(tdb, msg);
}
-/* lock/unlock one hash chain. This is meant to be used to reduce
- contention - it cannot guarantee how many records will be locked */
-int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
+void tdb_trace(struct tdb_context *tdb, const char *op)
{
- return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_end(tdb);
}
-int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
+void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op)
{
- return tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
+ char msg[sizeof(tdb_off_t) * 4 + 1];
+
+ snprintf(msg, sizeof(msg), "%u ", seqnum);
+ tdb_trace_write(tdb, msg);
+ tdb_trace_write(tdb, op);
+ tdb_trace_end(tdb);
}
-#if 0
-static int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+void tdb_trace_open(struct tdb_context *tdb, const char *op,
+ unsigned hash_size, unsigned tdb_flags, unsigned open_flags)
{
- return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_RDLCK);
+ char msg[128];
+
+ snprintf(msg, sizeof(msg),
+ "%s %u 0x%x 0x%x", op, hash_size, tdb_flags, open_flags);
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, msg);
+ tdb_trace_end(tdb);
}
-#endif
-#if 0
-static int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret)
{
- return tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_RDLCK);
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_end_ret(tdb, ret);
}
-#endif
-/* register a loging function */
-void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))
+void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret)
{
- tdb->log_fn = fn;
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_write(tdb, " =");
+ tdb_trace_record(tdb, ret);
+ tdb_trace_end(tdb);
}
-
-/* reopen a tdb - this is used after a fork to ensure that we have an independent
- seek pointer from our parent and to re-establish locks */
-int tdb_reopen(TDB_CONTEXT *tdb)
+void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec)
{
- struct stat st;
-
- if (tdb_munmap(tdb) != 0) {
- TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
- goto fail;
- }
- if (close(tdb->fd) != 0)
- TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
- tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
- if (tdb->fd == -1) {
- TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
- goto fail;
- }
- if (fstat(tdb->fd, &st) != 0) {
- TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
- goto fail;
- }
- if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
- TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
- goto fail;
- }
- tdb_mmap(tdb);
- if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
- goto fail;
- }
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_record(tdb, rec);
+ tdb_trace_end(tdb);
+}
- return 0;
+void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec, int ret)
+{
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_record(tdb, rec);
+ tdb_trace_end_ret(tdb, ret);
+}
-fail:
- tdb_close(tdb);
- return -1;
+void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec, TDB_DATA ret)
+{
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_record(tdb, rec);
+ tdb_trace_write(tdb, " =");
+ tdb_trace_record(tdb, ret);
+ tdb_trace_end(tdb);
}
-/* reopen all tdb's */
-int tdb_reopen_all(void)
+void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
+ int ret)
{
- TDB_CONTEXT *tdb;
+ char msg[1 + sizeof(ret) * 4];
- for (tdb=tdbs; tdb; tdb = tdb->next) {
- if (tdb_reopen(tdb) != 0) return -1;
- }
+ snprintf(msg, sizeof(msg), " %#x", flag);
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_record(tdb, rec1);
+ tdb_trace_record(tdb, rec2);
+ tdb_trace_write(tdb, msg);
+ tdb_trace_end_ret(tdb, ret);
+}
- return 0;
+void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret)
+{
+ tdb_trace_start(tdb);
+ tdb_trace_write(tdb, op);
+ tdb_trace_record(tdb, rec1);
+ tdb_trace_record(tdb, rec2);
+ tdb_trace_write(tdb, " =");
+ tdb_trace_record(tdb, ret);
+ tdb_trace_end(tdb);
}
+#endif
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library - private includes
+
+ Copyright (C) Andrew Tridgell 2005
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#if 0
+#include "replace.h"
+#include "system/filesys.h"
+#include "system/time.h"
+#include "system/shmem.h"
+#include "system/select.h"
+#include "system/wait.h"
+#include "tdb.h"
+#endif
+
+#include <unistd.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <sys/select.h>
+#include <sys/mman.h>
+#include <sys/time.h>
+#include <time.h>
+#include <utime.h>
+#include <sys/wait.h>
+#include <string.h>
+#include <stdlib.h>
+#include <limits.h>
+#include <stdio.h>
+#include <errno.h>
+#include <sys/param.h>
+
+#ifndef __STRING
+#define __STRING(x) #x
+#endif
+
+#ifndef __STRINGSTRING
+#define __STRINGSTRING(x) __STRING(x)
+#endif
+
+#ifndef __LINESTR__
+#define __LINESTR__ __STRINGSTRING(__LINE__)
+#endif
+
+#ifndef __location__
+#define __location__ __FILE__ ":" __LINESTR__
+#endif
+
+#include "atalk/tdb.h"
+
+/* #define TDB_TRACE 1 */
+#ifndef HAVE_GETPAGESIZE
+#define getpagesize() 0x2000
+#endif
+
+typedef uint32_t tdb_len_t;
+typedef uint32_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 (0x26011967 + 6)
+#define TDB_MAGIC (0x26011999U)
+#define TDB_FREE_MAGIC (~TDB_MAGIC)
+#define TDB_DEAD_MAGIC (0xFEE1DEAD)
+#define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
+#define TDB_ALIGNMENT 4
+#define DEFAULT_HASH_SIZE 131
+#define FREELIST_TOP (sizeof(struct tdb_header))
+#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
+#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
+#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
+#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
+#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
+#define TDB_HASHTABLE_SIZE(tdb) ((tdb->header.hash_size+1)*sizeof(tdb_off_t))
+#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + sizeof(tdb_off_t))
+#define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
+#define TDB_SEQNUM_OFS offsetof(struct tdb_header, sequence_number)
+#define TDB_PAD_BYTE 0x42
+#define TDB_PAD_U32 0x42424242
+
+/* NB assumes there is a local variable called "tdb" that is the
+ * current context, also takes doubly-parenthesized print-style
+ * argument. */
+#define TDB_LOG(x) tdb->log.log_fn x
+
+#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);
+void tdb_trace_open(struct tdb_context *tdb, const char *op,
+ unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
+void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
+void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
+void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec);
+void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec, int ret);
+void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec, TDB_DATA ret);
+void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
+ int ret);
+void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
+ TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
+#else
+#define tdb_trace(tdb, op)
+#define tdb_trace_seqnum(tdb, seqnum, op)
+#define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
+#define tdb_trace_ret(tdb, op, ret)
+#define tdb_trace_retrec(tdb, op, ret)
+#define tdb_trace_1rec(tdb, op, rec)
+#define tdb_trace_1rec_ret(tdb, op, rec, ret)
+#define tdb_trace_1rec_retrec(tdb, op, rec, ret)
+#define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
+#define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
+#endif /* !TDB_TRACE */
+
+/* lock offsets */
+#define GLOBAL_LOCK 0
+#define ACTIVE_LOCK 4
+#define TRANSACTION_LOCK 8
+
+/* free memory if the pointer is valid and zero the pointer */
+#ifndef SAFE_FREE
+#define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
+#endif
+
+#define BUCKET(hash) ((hash) % tdb->header.hash_size)
+
+#define DOCONV() (tdb->flags & TDB_CONVERT)
+#define CONVERT(x) (DOCONV() ? tdb_convert(&x, sizeof(x)) : &x)
+
+
+/* the body of the database is made of one tdb_record for the free space
+ plus a separate data list for each hash value */
+struct tdb_record {
+ tdb_off_t next; /* offset of the next record in the list */
+ tdb_len_t rec_len; /* total byte length of record */
+ tdb_len_t key_len; /* byte length of key */
+ tdb_len_t data_len; /* byte length of data */
+ uint32_t full_hash; /* the full 32 bit hash of the key */
+ uint32_t magic; /* try to catch errors */
+ /* the following union is implied:
+ union {
+ char record[rec_len];
+ struct {
+ char key[key_len];
+ char data[data_len];
+ }
+ uint32_t totalsize; (tailer)
+ }
+ */
+};
+
+
+/* this is stored at the front of every database */
+struct tdb_header {
+ char magic_food[32]; /* for /etc/magic */
+ uint32_t version; /* version of the code */
+ uint32_t hash_size; /* number of hash entries */
+ tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
+ tdb_off_t recovery_start; /* offset of transaction recovery region */
+ tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
+ tdb_off_t reserved[29];
+};
+
+struct tdb_lock_type {
+ int list;
+ uint32_t count;
+ uint32_t ltype;
+};
+
+struct tdb_traverse_lock {
+ struct tdb_traverse_lock *next;
+ uint32_t off;
+ uint32_t hash;
+ int lock_rw;
+};
+
+
+struct tdb_methods {
+ int (*tdb_read)(struct tdb_context *, tdb_off_t , void *, tdb_len_t , int );
+ int (*tdb_write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
+ void (*next_hash_chain)(struct tdb_context *, uint32_t *);
+ int (*tdb_oob)(struct tdb_context *, tdb_off_t , int );
+ int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
+ int (*tdb_brlock)(struct tdb_context *, tdb_off_t , int, int, int, size_t);
+};
+
+struct tdb_context {
+ char *name; /* the name of the database */
+ void *map_ptr; /* where it is currently mapped */
+ int fd; /* open file descriptor for the database */
+ tdb_len_t map_size; /* how much space has been mapped */
+ int read_only; /* opened read-only */
+ int traverse_read; /* read-only traversal */
+ int traverse_write; /* read-write traversal */
+ struct tdb_lock_type global_lock;
+ int num_lockrecs;
+ struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
+ enum TDB_ERROR ecode; /* error code for last tdb error */
+ struct tdb_header header; /* a cached copy of the header */
+ uint32_t flags; /* the flags passed to tdb_open */
+ struct tdb_traverse_lock travlocks; /* current traversal locks */
+ struct tdb_context *next; /* all tdbs to avoid multiple opens */
+ dev_t device; /* uniquely identifies this tdb */
+ ino_t inode; /* uniquely identifies this tdb */
+ struct tdb_logging_context log;
+ unsigned int (*hash_fn)(TDB_DATA *key);
+ int open_flags; /* flags used in the open - needed by reopen */
+ unsigned int num_locks; /* number of chain locks held */
+ const struct tdb_methods *methods;
+ struct tdb_transaction *transaction;
+ int page_size;
+ int max_dead_records;
+ int transaction_lock_count;
+#ifdef TDB_TRACE
+ int tracefd;
+#endif
+ volatile sig_atomic_t *interrupt_sig_ptr;
+};
+
+
+/*
+ internal prototypes
+*/
+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);
+int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, int rw_type, int lck_type, int probe, size_t len);
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
+int tdb_transaction_unlock(struct tdb_context *tdb);
+int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len);
+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);
+void *tdb_convert(void *buf, uint32_t size);
+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);
+int _tdb_transaction_cancel(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);
+
+
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 2005
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/*
+ transaction design:
+
+ - only allow a single transaction at a time per database. This makes
+ using the transaction API simpler, as otherwise the caller would
+ have to cope with temporary failures in transactions that conflict
+ with other current transactions
+
+ - keep the transaction recovery information in the same file as the
+ database, using a special 'transaction recovery' record pointed at
+ by the header. This removes the need for extra journal files as
+ used by some other databases
+
+ - dynamically allocated the transaction recover record, re-using it
+ for subsequent transactions. If a larger record is needed then
+ tdb_free() the old record to place it on the normal tdb freelist
+ before allocating the new record
+
+ - during transactions, keep a linked list of writes all that have
+ been performed by intercepting all tdb_write() calls. The hooked
+ transaction versions of tdb_read() and tdb_write() check this
+ linked list and try to use the elements of the list in preference
+ to the real database.
+
+ - don't allow any locks to be held when a transaction starts,
+ otherwise we can end up with deadlock (plus lack of lock nesting
+ in posix locks would mean the lock is lost)
+
+ - if the caller gains a lock during the transaction but doesn't
+ release it then fail the commit
+
+ - allow for nested calls to tdb_transaction_start(), re-using the
+ existing transaction record. If the inner transaction is cancelled
+ then a subsequent commit will fail
+
+ - keep a mirrored copy of the tdb hash chain heads to allow for the
+ fast hash heads scan on traverse, updating the mirrored copy in
+ the transaction version of tdb_write
+
+ - allow callers to mix transaction and non-transaction use of tdb,
+ although once a transaction is started then an exclusive lock is
+ gained until the transaction is committed or cancelled
+
+ - the commit stategy involves first saving away all modified data
+ into a linearised buffer in the transaction recovery area, then
+ marking the transaction recovery area with a magic value to
+ indicate a valid recovery record. In total 4 fsync/msync calls are
+ needed per commit to prevent race conditions. It might be possible
+ to reduce this to 3 or even 2 with some more work.
+
+ - check for a valid recovery record on open of the tdb, while the
+ global lock is held. Automatically recover from the transaction
+ recovery area if needed, then continue with the open as
+ usual. This allows for smooth crash recovery with no administrator
+ intervention.
+
+ - if TDB_NOSYNC is passed to flags in tdb_open then transactions are
+ still available, but no transaction recovery area is used and no
+ fsync/msync calls are made.
+
+ - if TDB_ALLOW_NESTING is passed to flags in tdb open, or added using
+ tdb_add_flags() transaction nesting is enabled.
+ It resets the TDB_DISALLOW_NESTING flag, as both cannot be used together.
+ The default is that transaction nesting is allowed.
+ Note: this default may change in future versions of tdb.
+
+ Beware. when transactions are nested a transaction successfully
+ completed with tdb_transaction_commit() can be silently unrolled later.
+
+ - if TDB_DISALLOW_NESTING is passed to flags in tdb open, or added using
+ tdb_add_flags() transaction nesting is disabled.
+ It resets the TDB_ALLOW_NESTING flag, as both cannot be used together.
+ An attempt create a nested transaction will fail with TDB_ERR_NESTING.
+ The default is that transaction nesting is allowed.
+ Note: this default may change in future versions of tdb.
+*/
+
+
+/*
+ hold the context of any current transaction
+*/
+struct tdb_transaction {
+ /* we keep a mirrored copy of the tdb hash heads here so
+ tdb_next_hash_chain() can operate efficiently */
+ uint32_t *hash_heads;
+
+ /* the original io methods - used to do IOs to the real db */
+ const struct tdb_methods *io_methods;
+
+ /* the list of transaction blocks. When a block is first
+ written to, it gets created in this list */
+ uint8_t **blocks;
+ uint32_t num_blocks;
+ uint32_t block_size; /* bytes in each block */
+ uint32_t last_block_size; /* number of valid bytes in the last block */
+
+ /* non-zero when an internal transaction error has
+ occurred. All write operations will then fail until the
+ transaction is ended */
+ int transaction_error;
+
+ /* when inside a transaction we need to keep track of any
+ nested tdb_transaction_start() calls, as these are allowed,
+ but don't create a new transaction */
+ int nesting;
+
+ /* set when a prepare has already occurred */
+ bool prepared;
+ tdb_off_t magic_offset;
+
+ /* set when the GLOBAL_LOCK has been taken */
+ bool global_lock_taken;
+
+ /* old file size before transaction */
+ tdb_len_t old_map_size;
+
+ /* we should re-pack on commit */
+ bool need_repack;
+};
+
+
+/*
+ read while in a transaction. We need to check first if the data is in our list
+ of transaction elements, then if not do a real read
+*/
+static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
+ tdb_len_t len, int cv)
+{
+ uint32_t blk;
+
+ /* break it down into block sized ops */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_read(tdb, off, buf, len2, cv) != 0) {
+ return -1;
+ }
+ len -= len2;
+ off += len2;
+ buf = (void *)(len2 + (char *)buf);
+ }
+
+ if (len == 0) {
+ return 0;
+ }
+
+ blk = off / tdb->transaction->block_size;
+
+ /* see if we have it in the block list */
+ if (tdb->transaction->num_blocks <= blk ||
+ tdb->transaction->blocks[blk] == NULL) {
+ /* nope, do a real read */
+ if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
+ goto fail;
+ }
+ return 0;
+ }
+
+ /* it is in the block list. Now check for the last block */
+ if (blk == tdb->transaction->num_blocks-1) {
+ if (len > tdb->transaction->last_block_size) {
+ goto fail;
+ }
+ }
+
+ /* now copy it out of this block */
+ memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
+ if (cv) {
+ tdb_convert(buf, len);
+ }
+ return 0;
+
+fail:
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%d len=%d\n", off, len));
+ tdb->ecode = TDB_ERR_IO;
+ tdb->transaction->transaction_error = 1;
+ return -1;
+}
+
+
+/*
+ write while in a transaction
+*/
+static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
+ const void *buf, tdb_len_t len)
+{
+ uint32_t blk;
+
+ /* Only a commit is allowed on a prepared transaction */
+ if (tdb->transaction->prepared) {
+ tdb->ecode = TDB_ERR_EINVAL;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: transaction already prepared, write not allowed\n"));
+ tdb->transaction->transaction_error = 1;
+ return -1;
+ }
+
+ /* if the write is to a hash head, then update the transaction
+ hash heads */
+ if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
+ off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
+ uint32_t chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
+ memcpy(&tdb->transaction->hash_heads[chain], buf, len);
+ }
+
+ /* break it up into block sized chunks */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_write(tdb, off, buf, len2) != 0) {
+ return -1;
+ }
+ len -= len2;
+ off += len2;
+ if (buf != NULL) {
+ buf = (const void *)(len2 + (const char *)buf);
+ }
+ }
+
+ if (len == 0) {
+ return 0;
+ }
+
+ blk = off / tdb->transaction->block_size;
+ off = off % tdb->transaction->block_size;
+
+ if (tdb->transaction->num_blocks <= blk) {
+ uint8_t **new_blocks;
+ /* expand the blocks array */
+ if (tdb->transaction->blocks == NULL) {
+ new_blocks = (uint8_t **)malloc(
+ (blk+1)*sizeof(uint8_t *));
+ } else {
+ new_blocks = (uint8_t **)realloc(
+ tdb->transaction->blocks,
+ (blk+1)*sizeof(uint8_t *));
+ }
+ if (new_blocks == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ goto fail;
+ }
+ memset(&new_blocks[tdb->transaction->num_blocks], 0,
+ (1+(blk - tdb->transaction->num_blocks))*sizeof(uint8_t *));
+ tdb->transaction->blocks = new_blocks;
+ tdb->transaction->num_blocks = blk+1;
+ tdb->transaction->last_block_size = 0;
+ }
+
+ /* allocate and fill a block? */
+ if (tdb->transaction->blocks[blk] == NULL) {
+ tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
+ if (tdb->transaction->blocks[blk] == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ tdb->transaction->transaction_error = 1;
+ return -1;
+ }
+ if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size;
+ if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
+ len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
+ }
+ if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size,
+ tdb->transaction->blocks[blk],
+ len2, 0) != 0) {
+ SAFE_FREE(tdb->transaction->blocks[blk]);
+ tdb->ecode = TDB_ERR_IO;
+ goto fail;
+ }
+ if (blk == tdb->transaction->num_blocks-1) {
+ tdb->transaction->last_block_size = len2;
+ }
+ }
+ }
+
+ /* overwrite part of an existing block */
+ if (buf == NULL) {
+ memset(tdb->transaction->blocks[blk] + off, 0, len);
+ } else {
+ memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+ }
+ if (blk == tdb->transaction->num_blocks-1) {
+ if (len + off > tdb->transaction->last_block_size) {
+ tdb->transaction->last_block_size = len + off;
+ }
+ }
+
+ return 0;
+
+fail:
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n",
+ (blk*tdb->transaction->block_size) + off, len));
+ tdb->transaction->transaction_error = 1;
+ return -1;
+}
+
+
+/*
+ write while in a transaction - this varient never expands the transaction blocks, it only
+ updates existing blocks. This means it cannot change the recovery size
+*/
+static int transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
+ const void *buf, tdb_len_t len)
+{
+ uint32_t blk;
+
+ /* break it up into block sized chunks */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_write_existing(tdb, off, buf, len2) != 0) {
+ return -1;
+ }
+ len -= len2;
+ off += len2;
+ if (buf != NULL) {
+ buf = (const void *)(len2 + (const char *)buf);
+ }
+ }
+
+ if (len == 0) {
+ return 0;
+ }
+
+ blk = off / tdb->transaction->block_size;
+ off = off % tdb->transaction->block_size;
+
+ if (tdb->transaction->num_blocks <= blk ||
+ tdb->transaction->blocks[blk] == NULL) {
+ return 0;
+ }
+
+ if (blk == tdb->transaction->num_blocks-1 &&
+ off + len > tdb->transaction->last_block_size) {
+ if (off >= tdb->transaction->last_block_size) {
+ return 0;
+ }
+ len = tdb->transaction->last_block_size - off;
+ }
+
+ /* overwrite part of an existing block */
+ memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+
+ return 0;
+}
+
+
+/*
+ accelerated hash chain head search, using the cached hash heads
+*/
+static void transaction_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
+{
+ uint32_t h = *chain;
+ for (;h < tdb->header.hash_size;h++) {
+ /* the +1 takes account of the freelist */
+ if (0 != tdb->transaction->hash_heads[h+1]) {
+ break;
+ }
+ }
+ (*chain) = h;
+}
+
+/*
+ out of bounds check during a transaction
+*/
+static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
+{
+ if (len <= tdb->map_size) {
+ return 0;
+ }
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+}
+
+/*
+ transaction version of tdb_expand().
+*/
+static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t size,
+ tdb_off_t addition)
+{
+ /* add a write to the transaction elements, so subsequent
+ reads see the zero data */
+ if (transaction_write(tdb, size, NULL, addition) != 0) {
+ return -1;
+ }
+
+ tdb->transaction->need_repack = true;
+
+ return 0;
+}
+
+/*
+ brlock during a transaction - ignore them
+*/
+static int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset,
+ int rw_type, int lck_type, int probe, size_t len)
+{
+ return 0;
+}
+
+static const struct tdb_methods transaction_methods = {
+ transaction_read,
+ transaction_write,
+ transaction_next_hash_chain,
+ transaction_oob,
+ transaction_expand_file,
+ transaction_brlock
+};
+
+
+/*
+ start a tdb transaction. No token is returned, as only a single
+ transaction is allowed to be pending per tdb_context
+*/
+int tdb_transaction_start(struct tdb_context *tdb)
+{
+ /* some sanity checks */
+ if (tdb->read_only || (tdb->flags & TDB_INTERNAL) || tdb->traverse_read) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
+ tdb->ecode = TDB_ERR_EINVAL;
+ return -1;
+ }
+
+ /* cope with nested tdb_transaction_start() calls */
+ if (tdb->transaction != NULL) {
+ if (!(tdb->flags & TDB_ALLOW_NESTING)) {
+ tdb->ecode = TDB_ERR_NESTING;
+ return -1;
+ }
+ tdb->transaction->nesting++;
+ TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n",
+ tdb->transaction->nesting));
+ return 0;
+ }
+
+ if (tdb->num_locks != 0 || tdb->global_lock.count) {
+ /* the caller must not have any locks when starting a
+ transaction as otherwise we'll be screwed by lack
+ of nested locks in posix */
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction with locks held\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ if (tdb->travlocks.next != NULL) {
+ /* you cannot use transactions inside a traverse (although you can use
+ traverse inside a transaction) as otherwise you can end up with
+ deadlock */
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ return -1;
+ }
+
+ tdb->transaction = (struct tdb_transaction *)
+ calloc(sizeof(struct tdb_transaction), 1);
+ if (tdb->transaction == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
+
+ /* a page at a time seems like a reasonable compromise between compactness and efficiency */
+ tdb->transaction->block_size = tdb->page_size;
+
+ /* get the transaction write lock. This is a blocking lock. As
+ discussed with Volker, there are a number of ways we could
+ make this async, which we will probably do in the future */
+ if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
+ SAFE_FREE(tdb->transaction->blocks);
+ SAFE_FREE(tdb->transaction);
+ return -1;
+ }
+
+ /* get a read lock from the freelist to the end of file. This
+ is upgraded to a write lock during the commit */
+ if (tdb_brlock(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ goto fail;
+ }
+
+ /* setup a copy of the hash table heads so the hash scan in
+ traverse can be fast */
+ tdb->transaction->hash_heads = (uint32_t *)
+ calloc(tdb->header.hash_size+1, sizeof(uint32_t));
+ if (tdb->transaction->hash_heads == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ goto fail;
+ }
+ if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
+ TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to read hash heads\n"));
+ tdb->ecode = TDB_ERR_IO;
+ goto fail;
+ }
+
+ /* make sure we know about any file expansions already done by
+ anyone else */
+ tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+ tdb->transaction->old_map_size = tdb->map_size;
+
+ /* finally hook the io methods, replacing them with
+ transaction specific methods */
+ tdb->transaction->io_methods = tdb->methods;
+ tdb->methods = &transaction_methods;
+
+ /* Trace at the end, so we get sequence number correct. */
+ tdb_trace(tdb, "tdb_transaction_start");
+ return 0;
+
+fail:
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_transaction_unlock(tdb);
+ SAFE_FREE(tdb->transaction->blocks);
+ SAFE_FREE(tdb->transaction->hash_heads);
+ SAFE_FREE(tdb->transaction);
+ return -1;
+}
+
+
+/*
+ sync to disk
+*/
+static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
+{
+ if (tdb->flags & TDB_NOSYNC) {
+ return 0;
+ }
+
+ if (fsync(tdb->fd) != 0) {
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: fsync failed\n"));
+ return -1;
+ }
+#ifdef HAVE_MMAP
+ if (tdb->map_ptr) {
+ tdb_off_t moffset = offset & ~(tdb->page_size-1);
+ if (msync(moffset + (char *)tdb->map_ptr,
+ length + (offset - moffset), MS_SYNC) != 0) {
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: msync failed - %s\n",
+ strerror(errno)));
+ return -1;
+ }
+ }
+#endif
+ return 0;
+}
+
+
+int _tdb_transaction_cancel(struct tdb_context *tdb)
+{
+ int i, ret = 0;
+
+ if (tdb->transaction == NULL) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
+ return -1;
+ }
+
+ if (tdb->transaction->nesting != 0) {
+ tdb->transaction->transaction_error = 1;
+ tdb->transaction->nesting--;
+ return 0;
+ }
+
+ tdb->map_size = tdb->transaction->old_map_size;
+
+ /* free all the transaction blocks */
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ if (tdb->transaction->blocks[i] != NULL) {
+ free(tdb->transaction->blocks[i]);
+ }
+ }
+ SAFE_FREE(tdb->transaction->blocks);
+
+ if (tdb->transaction->magic_offset) {
+ const struct tdb_methods *methods = tdb->transaction->io_methods;
+ uint32_t zero = 0;
+
+ /* remove the recovery marker */
+ if (methods->tdb_write(tdb, tdb->transaction->magic_offset, &zero, 4) == -1 ||
+ transaction_sync(tdb, tdb->transaction->magic_offset, 4) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_cancel: failed to remove recovery magic\n"));
+ ret = -1;
+ }
+ }
+
+ if (tdb->transaction->global_lock_taken) {
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb->transaction->global_lock_taken = false;
+ }
+
+ /* remove any global lock created during the transaction */
+ if (tdb->global_lock.count != 0) {
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 4*tdb->header.hash_size);
+ tdb->global_lock.count = 0;
+ }
+
+ /* remove any locks created during the transaction */
+ if (tdb->num_locks != 0) {
+ for (i=0;i<tdb->num_lockrecs;i++) {
+ tdb_brlock(tdb,FREELIST_TOP+4*tdb->lockrecs[i].list,
+ F_UNLCK,F_SETLKW, 0, 1);
+ }
+ tdb->num_locks = 0;
+ tdb->num_lockrecs = 0;
+ SAFE_FREE(tdb->lockrecs);
+ }
+
+ /* restore the normal io methods */
+ tdb->methods = tdb->transaction->io_methods;
+
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_transaction_unlock(tdb);
+ SAFE_FREE(tdb->transaction->hash_heads);
+ SAFE_FREE(tdb->transaction);
+
+ return ret;
+}
+
+/*
+ cancel the current transaction
+*/
+int tdb_transaction_cancel(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_transaction_cancel");
+ return _tdb_transaction_cancel(tdb);
+}
+
+/*
+ work out how much space the linearised recovery data will consume
+*/
+static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
+{
+ tdb_len_t recovery_size = 0;
+ int i;
+
+ recovery_size = sizeof(uint32_t);
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
+ break;
+ }
+ if (tdb->transaction->blocks[i] == NULL) {
+ continue;
+ }
+ recovery_size += 2*sizeof(tdb_off_t);
+ if (i == tdb->transaction->num_blocks-1) {
+ recovery_size += tdb->transaction->last_block_size;
+ } else {
+ recovery_size += tdb->transaction->block_size;
+ }
+ }
+
+ return recovery_size;
+}
+
+/*
+ allocate the recovery area, or use an existing recovery area if it is
+ large enough
+*/
+static int tdb_recovery_allocate(struct tdb_context *tdb,
+ tdb_len_t *recovery_size,
+ tdb_off_t *recovery_offset,
+ tdb_len_t *recovery_max_size)
+{
+ struct tdb_record rec;
+ const struct tdb_methods *methods = tdb->transaction->io_methods;
+ tdb_off_t recovery_head;
+
+ if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery head\n"));
+ return -1;
+ }
+
+ rec.rec_len = 0;
+
+ if (recovery_head != 0 &&
+ methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
+ return -1;
+ }
+
+ *recovery_size = tdb_recovery_size(tdb);
+
+ if (recovery_head != 0 && *recovery_size <= rec.rec_len) {
+ /* it fits in the existing area */
+ *recovery_max_size = rec.rec_len;
+ *recovery_offset = recovery_head;
+ return 0;
+ }
+
+ /* we need to free up the old recovery area, then allocate a
+ new one at the end of the file. Note that we cannot use
+ tdb_allocate() to allocate the new one as that might return
+ us an area that is being currently used (as of the start of
+ the transaction) */
+ if (recovery_head != 0) {
+ if (tdb_free(tdb, recovery_head, &rec) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to free previous recovery area\n"));
+ return -1;
+ }
+ }
+
+ /* the tdb_free() call might have increased the recovery size */
+ *recovery_size = tdb_recovery_size(tdb);
+
+ /* round up to a multiple of page size */
+ *recovery_max_size = TDB_ALIGN(sizeof(rec) + *recovery_size, tdb->page_size) - sizeof(rec);
+ *recovery_offset = tdb->map_size;
+ recovery_head = *recovery_offset;
+
+ if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
+ (tdb->map_size - tdb->transaction->old_map_size) +
+ sizeof(rec) + *recovery_max_size) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to create recovery area\n"));
+ return -1;
+ }
+
+ /* remap the file (if using mmap) */
+ methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+
+ /* we have to reset the old map size so that we don't try to expand the file
+ again in the transaction commit, which would destroy the recovery area */
+ tdb->transaction->old_map_size = tdb->map_size;
+
+ /* write the recovery header offset and sync - we can sync without a race here
+ as the magic ptr in the recovery record has not been set */
+ CONVERT(recovery_head);
+ if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD,
+ &recovery_head, sizeof(tdb_off_t)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
+ return -1;
+ }
+ if (transaction_write_existing(tdb, TDB_RECOVERY_HEAD, &recovery_head, sizeof(tdb_off_t)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
+ return -1;
+ }
+
+ return 0;
+}
+
+
+/*
+ setup the recovery data that will be used on a crash during commit
+*/
+static int transaction_setup_recovery(struct tdb_context *tdb,
+ tdb_off_t *magic_offset)
+{
+ tdb_len_t recovery_size;
+ unsigned char *data, *p;
+ const struct tdb_methods *methods = tdb->transaction->io_methods;
+ struct tdb_record *rec;
+ tdb_off_t recovery_offset, recovery_max_size;
+ tdb_off_t old_map_size = tdb->transaction->old_map_size;
+ uint32_t magic, tailer;
+ int i;
+
+ /*
+ check that the recovery area has enough space
+ */
+ if (tdb_recovery_allocate(tdb, &recovery_size,
+ &recovery_offset, &recovery_max_size) == -1) {
+ return -1;
+ }
+
+ data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
+ if (data == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
+
+ rec = (struct tdb_record *)data;
+ memset(rec, 0, sizeof(*rec));
+
+ rec->magic = 0;
+ rec->data_len = recovery_size;
+ rec->rec_len = recovery_max_size;
+ rec->key_len = old_map_size;
+ CONVERT(rec);
+
+ /* build the recovery data into a single blob to allow us to do a single
+ large write, which should be more efficient */
+ p = data + sizeof(*rec);
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ tdb_off_t offset;
+ tdb_len_t length;
+
+ if (tdb->transaction->blocks[i] == NULL) {
+ continue;
+ }
+
+ offset = i * tdb->transaction->block_size;
+ length = tdb->transaction->block_size;
+ if (i == tdb->transaction->num_blocks-1) {
+ length = tdb->transaction->last_block_size;
+ }
+
+ if (offset >= old_map_size) {
+ continue;
+ }
+ if (offset + length > tdb->transaction->old_map_size) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
+ free(data);
+ tdb->ecode = TDB_ERR_CORRUPT;
+ return -1;
+ }
+ memcpy(p, &offset, 4);
+ memcpy(p+4, &length, 4);
+ if (DOCONV()) {
+ tdb_convert(p, 8);
+ }
+ /* the recovery area contains the old data, not the
+ new data, so we have to call the original tdb_read
+ method to get it */
+ if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
+ free(data);
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ p += 8 + length;
+ }
+
+ /* and the tailer */
+ tailer = sizeof(*rec) + recovery_max_size;
+ memcpy(p, &tailer, 4);
+ CONVERT(p);
+
+ /* write the recovery data to the recovery area */
+ if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery data\n"));
+ free(data);
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ if (transaction_write_existing(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery data\n"));
+ free(data);
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ /* as we don't have ordered writes, we have to sync the recovery
+ data before we update the magic to indicate that the recovery
+ data is present */
+ if (transaction_sync(tdb, recovery_offset, sizeof(*rec) + recovery_size) == -1) {
+ free(data);
+ return -1;
+ }
+
+ free(data);
+
+ magic = TDB_RECOVERY_MAGIC;
+ CONVERT(magic);
+
+ *magic_offset = recovery_offset + offsetof(struct tdb_record, magic);
+
+ if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ if (transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery magic\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ /* ensure the recovery magic marker is on disk */
+ if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
+ return -1;
+ }
+
+ return 0;
+}
+
+static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
+{
+ const struct tdb_methods *methods;
+
+ if (tdb->transaction == NULL) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: no transaction\n"));
+ return -1;
+ }
+
+ if (tdb->transaction->prepared) {
+ tdb->ecode = TDB_ERR_EINVAL;
+ _tdb_transaction_cancel(tdb);
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction already prepared\n"));
+ return -1;
+ }
+
+ if (tdb->transaction->transaction_error) {
+ tdb->ecode = TDB_ERR_IO;
+ _tdb_transaction_cancel(tdb);
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction error pending\n"));
+ return -1;
+ }
+
+
+ if (tdb->transaction->nesting != 0) {
+ return 0;
+ }
+
+ /* check for a null transaction */
+ if (tdb->transaction->blocks == NULL) {
+ return 0;
+ }
+
+ methods = tdb->transaction->io_methods;
+
+ /* if there are any locks pending then the caller has not
+ nested their locks properly, so fail the transaction */
+ if (tdb->num_locks || tdb->global_lock.count) {
+ tdb->ecode = TDB_ERR_LOCK;
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: locks pending on commit\n"));
+ _tdb_transaction_cancel(tdb);
+ return -1;
+ }
+
+ /* upgrade the main transaction lock region to a write lock */
+ if (tdb_brlock_upgrade(tdb, FREELIST_TOP, 0) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to upgrade hash locks\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ _tdb_transaction_cancel(tdb);
+ return -1;
+ }
+
+ /* get the global lock - this prevents new users attaching to the database
+ during the commit */
+ if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to get global lock\n"));
+ tdb->ecode = TDB_ERR_LOCK;
+ _tdb_transaction_cancel(tdb);
+ return -1;
+ }
+
+ tdb->transaction->global_lock_taken = true;
+
+ if (!(tdb->flags & TDB_NOSYNC)) {
+ /* write the recovery data to the end of the file */
+ if (transaction_setup_recovery(tdb, &tdb->transaction->magic_offset) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: failed to setup recovery data\n"));
+ _tdb_transaction_cancel(tdb);
+ return -1;
+ }
+ }
+
+ tdb->transaction->prepared = true;
+
+ /* expand the file to the new size if needed */
+ if (tdb->map_size != tdb->transaction->old_map_size) {
+ if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
+ tdb->map_size -
+ tdb->transaction->old_map_size) == -1) {
+ tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: expansion failed\n"));
+ _tdb_transaction_cancel(tdb);
+ return -1;
+ }
+ tdb->map_size = tdb->transaction->old_map_size;
+ methods->tdb_oob(tdb, tdb->map_size + 1, 1);
+ }
+
+ /* Keep the global lock until the actual commit */
+
+ return 0;
+}
+
+/*
+ prepare to commit the current transaction
+*/
+int tdb_transaction_prepare_commit(struct tdb_context *tdb)
+{
+ tdb_trace(tdb, "tdb_transaction_prepare_commit");
+ return _tdb_transaction_prepare_commit(tdb);
+}
+
+/*
+ commit the current transaction
+*/
+int tdb_transaction_commit(struct tdb_context *tdb)
+{
+ const struct tdb_methods *methods;
+ int i;
+ bool need_repack;
+
+ if (tdb->transaction == NULL) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
+ return -1;
+ }
+
+ tdb_trace(tdb, "tdb_transaction_commit");
+
+ if (tdb->transaction->transaction_error) {
+ tdb->ecode = TDB_ERR_IO;
+ _tdb_transaction_cancel(tdb);
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: transaction error pending\n"));
+ return -1;
+ }
+
+
+ if (tdb->transaction->nesting != 0) {
+ tdb->transaction->nesting--;
+ return 0;
+ }
+
+ /* check for a null transaction */
+ if (tdb->transaction->blocks == NULL) {
+ _tdb_transaction_cancel(tdb);
+ return 0;
+ }
+
+ if (!tdb->transaction->prepared) {
+ int ret = _tdb_transaction_prepare_commit(tdb);
+ if (ret)
+ return ret;
+ }
+
+ methods = tdb->transaction->io_methods;
+
+ /* perform all the writes */
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ tdb_off_t offset;
+ tdb_len_t length;
+
+ if (tdb->transaction->blocks[i] == NULL) {
+ continue;
+ }
+
+ offset = i * tdb->transaction->block_size;
+ length = tdb->transaction->block_size;
+ if (i == tdb->transaction->num_blocks-1) {
+ length = tdb->transaction->last_block_size;
+ }
+
+ if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
+
+ /* we've overwritten part of the data and
+ possibly expanded the file, so we need to
+ run the crash recovery code */
+ tdb->methods = methods;
+ tdb_transaction_recover(tdb);
+
+ _tdb_transaction_cancel(tdb);
+
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
+ return -1;
+ }
+ SAFE_FREE(tdb->transaction->blocks[i]);
+ }
+
+ SAFE_FREE(tdb->transaction->blocks);
+ tdb->transaction->num_blocks = 0;
+
+ /* ensure the new data is on disk */
+ if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+ return -1;
+ }
+
+ /*
+ TODO: maybe write to some dummy hdr field, or write to magic
+ offset without mmap, before the last sync, instead of the
+ utime() call
+ */
+
+ /* on some systems (like Linux 2.6.x) changes via mmap/msync
+ don't change the mtime of the file, this means the file may
+ not be backed up (as tdb rounding to block sizes means that
+ file size changes are quite rare too). The following forces
+ mtime changes when a transaction completes */
+#ifdef HAVE_UTIME
+ utime(tdb->name, NULL);
+#endif
+
+ need_repack = tdb->transaction->need_repack;
+
+ /* use a transaction cancel to free memory and remove the
+ transaction locks */
+ _tdb_transaction_cancel(tdb);
+
+ if (need_repack) {
+ return tdb_repack(tdb);
+ }
+
+ return 0;
+}
+
+
+/*
+ recover from an aborted transaction. Must be called with exclusive
+ database write access already established (including the global
+ lock to prevent new processes attaching)
+*/
+int tdb_transaction_recover(struct tdb_context *tdb)
+{
+ tdb_off_t recovery_head, recovery_eof;
+ unsigned char *data, *p;
+ uint32_t zero = 0;
+ struct tdb_record rec;
+
+ /* find the recovery area */
+ if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery head\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ if (recovery_head == 0) {
+ /* we have never allocated a recovery record */
+ return 0;
+ }
+
+ /* read the recovery record */
+ if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
+ sizeof(rec), DOCONV()) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery record\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ if (rec.magic != TDB_RECOVERY_MAGIC) {
+ /* there is no valid recovery data */
+ return 0;
+ }
+
+ if (tdb->read_only) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: attempt to recover read only database\n"));
+ tdb->ecode = TDB_ERR_CORRUPT;
+ return -1;
+ }
+
+ recovery_eof = rec.key_len;
+
+ data = (unsigned char *)malloc(rec.data_len);
+ if (data == NULL) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to allocate recovery data\n"));
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
+
+ /* read the full recovery data */
+ if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
+ rec.data_len, 0) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery data\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ /* recover the file data */
+ p = data;
+ while (p+8 < data + rec.data_len) {
+ uint32_t ofs, len;
+ if (DOCONV()) {
+ tdb_convert(p, 8);
+ }
+ memcpy(&ofs, p, 4);
+ memcpy(&len, p+4, 4);
+
+ if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
+ free(data);
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ p += 8 + len;
+ }
+
+ free(data);
+
+ if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync recovery\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ /* if the recovery area is after the recovered eof then remove it */
+ if (recovery_eof <= recovery_head) {
+ if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery head\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ }
+
+ /* remove the recovery magic */
+ if (tdb_ofs_write(tdb, recovery_head + offsetof(struct tdb_record, magic),
+ &zero) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ /* reduce the file size to the old size */
+ tdb_munmap(tdb);
+ if (ftruncate(tdb->fd, recovery_eof) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to reduce to recovery size\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ tdb->map_size = recovery_eof;
+ tdb_mmap(tdb);
+
+ if (transaction_sync(tdb, 0, recovery_eof) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+
+ TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_recover: recovered %d byte database\n",
+ recovery_eof));
+
+ /* all done */
+ return 0;
+}
--- /dev/null
+ /*
+ Unix SMB/CIFS implementation.
+
+ trivial database library
+
+ Copyright (C) Andrew Tridgell 1999-2005
+ Copyright (C) Paul `Rusty' Russell 2000
+ Copyright (C) Jeremy Allison 2000-2003
+
+ ** NOTE! The following LGPL license applies to the tdb
+ ** library. This does NOT imply that all of Samba is released
+ ** under the LGPL
+
+ This library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 3 of the License, or (at your option) any later version.
+
+ This library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+#define TDB_NEXT_LOCK_ERR ((tdb_off_t)-1)
+
+/* Uses traverse lock: 0 = finish, TDB_NEXT_LOCK_ERR = error,
+ other = record offset */
+static tdb_off_t tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
+ struct tdb_record *rec)
+{
+ int want_next = (tlock->off != 0);
+
+ /* Lock each chain from the start one. */
+ for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
+ if (!tlock->off && tlock->hash != 0) {
+ /* this is an optimisation for the common case where
+ the hash chain is empty, which is particularly
+ common for the use of tdb with ldb, where large
+ hashes are used. In that case we spend most of our
+ time in tdb_brlock(), locking empty hash chains.
+
+ To avoid this, we do an unlocked pre-check to see
+ if the hash chain is empty before starting to look
+ inside it. If it is empty then we can avoid that
+ hash chain. If it isn't empty then we can't believe
+ the value we get back, as we read it without a
+ lock, so instead we get the lock and re-fetch the
+ value below.
+
+ Notice that not doing this optimisation on the
+ first hash chain is critical. We must guarantee
+ that we have done at least one fcntl lock at the
+ start of a search to guarantee that memory is
+ coherent on SMP systems. If records are added by
+ others during the search then thats OK, and we
+ could possibly miss those with this trick, but we
+ could miss them anyway without this trick, so the
+ semantics don't change.
+
+ With a non-indexed ldb search this trick gains us a
+ factor of around 80 in speed on a linux 2.6.x
+ system (testing using ldbtest).
+ */
+ tdb->methods->next_hash_chain(tdb, &tlock->hash);
+ if (tlock->hash == tdb->header.hash_size) {
+ continue;
+ }
+ }
+
+ if (tdb_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
+ return TDB_NEXT_LOCK_ERR;
+
+ /* No previous record? Start at top of chain. */
+ if (!tlock->off) {
+ if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
+ &tlock->off) == -1)
+ goto fail;
+ } else {
+ /* Otherwise unlock the previous record. */
+ if (tdb_unlock_record(tdb, tlock->off) != 0)
+ goto fail;
+ }
+
+ if (want_next) {
+ /* We have offset of old record: grab next */
+ if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+ goto fail;
+ tlock->off = rec->next;
+ }
+
+ /* Iterate through chain */
+ while( tlock->off) {
+ tdb_off_t current;
+ if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+ goto fail;
+
+ /* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
+ if (tlock->off == rec->next) {
+ tdb->ecode = TDB_ERR_CORRUPT;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: loop detected.\n"));
+ goto fail;
+ }
+
+ if (!TDB_DEAD(rec)) {
+ /* Woohoo: we found one! */
+ if (tdb_lock_record(tdb, tlock->off) != 0)
+ goto fail;
+ return tlock->off;
+ }
+
+ /* Try to clean dead ones from old traverses */
+ current = tlock->off;
+ tlock->off = rec->next;
+ if (!(tdb->read_only || tdb->traverse_read) &&
+ tdb_do_delete(tdb, current, rec) != 0)
+ goto fail;
+ }
+ tdb_unlock(tdb, tlock->hash, tlock->lock_rw);
+ want_next = 0;
+ }
+ /* We finished iteration without finding anything */
+ tdb->ecode = TDB_SUCCESS;
+ return 0;
+
+ fail:
+ tlock->off = 0;
+ if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
+ return TDB_NEXT_LOCK_ERR;
+}
+
+/* traverse the entire database - calling fn(tdb, key, data) on each element.
+ return -1 on error or the record count traversed
+ if fn is NULL then it is not called
+ a non-zero return value from fn() indicates that the traversal should stop
+ */
+static int tdb_traverse_internal(struct tdb_context *tdb,
+ tdb_traverse_func fn, void *private_data,
+ struct tdb_traverse_lock *tl)
+{
+ TDB_DATA key, dbuf;
+ struct tdb_record rec;
+ int ret = 0, count = 0;
+ tdb_off_t off;
+
+ /* This was in the initializaton, above, but the IRIX compiler
+ * did not like it. crh
+ */
+ tl->next = tdb->travlocks.next;
+
+ /* fcntl locks don't stack: beware traverse inside traverse */
+ tdb->travlocks.next = tl;
+
+ /* tdb_next_lock places locks on the record returned, and its chain */
+ while ((off = tdb_next_lock(tdb, tl, &rec)) != 0) {
+ if (off == TDB_NEXT_LOCK_ERR) {
+ ret = -1;
+ goto out;
+ }
+ count++;
+ /* now read the full record */
+ key.dptr = tdb_alloc_read(tdb, tl->off + sizeof(rec),
+ rec.key_len + rec.data_len);
+ if (!key.dptr) {
+ ret = -1;
+ if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
+ goto out;
+ if (tdb_unlock_record(tdb, tl->off) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
+ goto out;
+ }
+ key.dsize = rec.key_len;
+ dbuf.dptr = key.dptr + rec.key_len;
+ dbuf.dsize = rec.data_len;
+
+ tdb_trace_1rec_retrec(tdb, "traverse", key, dbuf);
+
+ /* Drop chain lock, call out */
+ if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
+ ret = -1;
+ SAFE_FREE(key.dptr);
+ goto out;
+ }
+ if (fn && fn(tdb, key, dbuf, private_data)) {
+ /* They want us to terminate traversal */
+ tdb_trace_ret(tdb, "tdb_traverse_end", count);
+ if (tdb_unlock_record(tdb, tl->off) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
+ ret = -1;
+ }
+ SAFE_FREE(key.dptr);
+ goto out;
+ }
+ SAFE_FREE(key.dptr);
+ }
+ tdb_trace(tdb, "tdb_traverse_end");
+out:
+ tdb->travlocks.next = tl->next;
+ if (ret < 0)
+ return -1;
+ else
+ return count;
+}
+
+
+/*
+ a write style traverse - temporarily marks the db read only
+*/
+int tdb_traverse_read(struct tdb_context *tdb,
+ tdb_traverse_func fn, void *private_data)
+{
+ struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
+ int ret;
+
+ /* we need to get a read lock on the transaction lock here to
+ cope with the lock ordering semantics of solaris10 */
+ if (tdb_transaction_lock(tdb, F_RDLCK)) {
+ return -1;
+ }
+
+ tdb->traverse_read++;
+ tdb_trace(tdb, "tdb_traverse_read_start");
+ ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+ tdb->traverse_read--;
+
+ tdb_transaction_unlock(tdb);
+
+ return ret;
+}
+
+/*
+ a write style traverse - needs to get the transaction lock to
+ prevent deadlocks
+
+ WARNING: The data buffer given to the callback fn does NOT meet the
+ alignment restrictions malloc gives you.
+*/
+int tdb_traverse(struct tdb_context *tdb,
+ tdb_traverse_func fn, void *private_data)
+{
+ struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
+ int ret;
+
+ if (tdb->read_only || tdb->traverse_read) {
+ return tdb_traverse_read(tdb, fn, private_data);
+ }
+
+ if (tdb_transaction_lock(tdb, F_WRLCK)) {
+ return -1;
+ }
+
+ tdb->traverse_write++;
+ tdb_trace(tdb, "tdb_traverse_start");
+ ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+ tdb->traverse_write--;
+
+ tdb_transaction_unlock(tdb);
+
+ return ret;
+}
+
+
+/* find the first entry in the database and return its key */
+TDB_DATA tdb_firstkey(struct tdb_context *tdb)
+{
+ TDB_DATA key;
+ struct tdb_record rec;
+ tdb_off_t off;
+
+ /* release any old lock */
+ if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
+ return tdb_null;
+ tdb->travlocks.off = tdb->travlocks.hash = 0;
+ tdb->travlocks.lock_rw = F_RDLCK;
+
+ /* Grab first record: locks chain and returned record. */
+ off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
+ if (off == 0 || off == TDB_NEXT_LOCK_ERR) {
+ tdb_trace_retrec(tdb, "tdb_firstkey", tdb_null);
+ return tdb_null;
+ }
+ /* now read the key */
+ key.dsize = rec.key_len;
+ key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
+
+ tdb_trace_retrec(tdb, "tdb_firstkey", key);
+
+ /* Unlock the hash chain of the record we just read. */
+ if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
+ return key;
+}
+
+/* find the next entry in the database, returning its key */
+TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
+{
+ uint32_t oldhash;
+ TDB_DATA key = tdb_null;
+ struct tdb_record rec;
+ unsigned char *k = NULL;
+ tdb_off_t off;
+
+ /* Is locked key the old key? If so, traverse will be reliable. */
+ if (tdb->travlocks.off) {
+ if (tdb_lock(tdb,tdb->travlocks.hash,tdb->travlocks.lock_rw))
+ return tdb_null;
+ if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
+ || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
+ rec.key_len))
+ || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
+ /* No, it wasn't: unlock it and start from scratch */
+ if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
+ tdb_trace_1rec_retrec(tdb, "tdb_nextkey",
+ oldkey, tdb_null);
+ SAFE_FREE(k);
+ return tdb_null;
+ }
+ if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0) {
+ SAFE_FREE(k);
+ return tdb_null;
+ }
+ tdb->travlocks.off = 0;
+ }
+
+ SAFE_FREE(k);
+ }
+
+ if (!tdb->travlocks.off) {
+ /* No previous element: do normal find, and lock record */
+ tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
+ if (!tdb->travlocks.off) {
+ tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, tdb_null);
+ return tdb_null;
+ }
+ tdb->travlocks.hash = BUCKET(rec.full_hash);
+ if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
+ return tdb_null;
+ }
+ }
+ oldhash = tdb->travlocks.hash;
+
+ /* Grab next record: locks chain and returned record,
+ unlocks old record */
+ off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
+ if (off != TDB_NEXT_LOCK_ERR && off != 0) {
+ key.dsize = rec.key_len;
+ key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
+ key.dsize);
+ /* Unlock the chain of this new record */
+ if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+ }
+ /* Unlock the chain of old record */
+ if (tdb_unlock(tdb, BUCKET(oldhash), tdb->travlocks.lock_rw) != 0)
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+ tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, key);
+ return key;
+}
+