tdb2: move file operations into separate structure

This moves the fd and locking information into a new 'struct tdb_file',
opening the way for it to be shared by multiple tdb_open calls on the
same file.

ccan/tdb2/free.c

@@ -661,7 +661,7 @@ static enum TDB_ERROR tdb_expand(struct tdb_context *tdb, tdb_len_t size)
 
 	/* Need to hold a hash lock to expand DB: transactions rely on it. */
 	if (!(tdb->flags & TDB_NOLOCK)
-	    && !tdb->allrecord_lock.count && !tdb_has_hash_locks(tdb)) {
+	    && !tdb->file->allrecord_lock.count && !tdb_has_hash_locks(tdb)) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_expand: must hold lock during expand");
 	}

ccan/tdb2/io.c

@@ -49,7 +49,7 @@ void tdb_mmap(struct tdb_context *tdb)
 		return;
 
 	tdb->map_ptr = mmap(NULL, tdb->map_size, tdb->mmap_flags,
-			    MAP_SHARED, tdb->fd, 0);
+			    MAP_SHARED, tdb->file->fd, 0);
 
 	/*
 	 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
@@ -96,7 +96,7 @@ static enum TDB_ERROR tdb_oob(struct tdb_context *tdb, tdb_off_t len,
 		return ecode;
 	}
 
-	if (fstat(tdb->fd, &st) != 0) {
+	if (fstat(tdb->file->fd, &st) != 0) {
 		tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 			   "Failed to fstat file: %s", strerror(errno));
 		tdb_unlock_expand(tdb, F_RDLCK);
@@ -245,7 +245,7 @@ static enum TDB_ERROR tdb_write(struct tdb_context *tdb, tdb_off_t off,
 		memcpy(off + (char *)tdb->map_ptr, buf, len);
 	} else {
 		ssize_t ret;
-		ret = pwrite(tdb->fd, buf, len, off);
+		ret = pwrite(tdb->file->fd, buf, len, off);
 		if (ret != len) {
 			/* This shouldn't happen: we avoid sparse files. */
 			if (ret >= 0)
@@ -274,7 +274,7 @@ static enum TDB_ERROR tdb_read(struct tdb_context *tdb, tdb_off_t off,
 	if (tdb->map_ptr) {
 		memcpy(buf, off + (char *)tdb->map_ptr, len);
 	} else {
-		ssize_t r = pread(tdb->fd, buf, len, off);
+		ssize_t r = pread(tdb->file->fd, buf, len, off);
 		if (r != len) {
 			return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 					  "tdb_read failed with %zi at %zu "
@@ -374,7 +374,7 @@ static enum TDB_ERROR fill(struct tdb_context *tdb,
 {
 	while (len) {
 		size_t n = len > size ? size : len;
-		ssize_t ret = pwrite(tdb->fd, buf, n, off);
+		ssize_t ret = pwrite(tdb->file->fd, buf, n, off);
 		if (ret != n) {
 			if (ret >= 0)
 				errno = ENOSPC;
@@ -418,7 +418,7 @@ static enum TDB_ERROR tdb_expand_file(struct tdb_context *tdb,
 		tdb_munmap(tdb);
 
 		/* If this fails, we try to fill anyway. */
-		if (ftruncate(tdb->fd, tdb->map_size + addition))
+		if (ftruncate(tdb->file->fd, tdb->map_size + addition))
 			;
 
 		/* now fill the file with something. This ensures that the

ccan/tdb2/lock.c

@@ -42,10 +42,10 @@ static int fcntl_lock(struct tdb_context *tdb,
 
 	add_stat(tdb, lock_lowlevel, 1);
 	if (waitflag)
-		return fcntl(tdb->fd, F_SETLKW, &fl);
+		return fcntl(tdb->file->fd, F_SETLKW, &fl);
 	else {
 		add_stat(tdb, lock_nonblock, 1);
-		return fcntl(tdb->fd, F_SETLK, &fl);
+		return fcntl(tdb->file->fd, F_SETLK, &fl);
 	}
 }
 
@@ -116,7 +116,7 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 	fl.l_len = len;
 	fl.l_pid = 0;
 
-	return fcntl(tdb->fd, F_SETLKW, &fl);
+	return fcntl(tdb->file->fd, F_SETLKW, &fl);
 }
 
 /* a byte range locking function - return 0 on success
@@ -161,7 +161,7 @@ static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
 				   "tdb_brlock failed (fd=%d) at"
 				   " offset %zu rw_type=%d flags=%d len=%zu:"
 				   " %s",
-				   tdb->fd, (size_t)offset, rw_type,
+				   tdb->file->fd, (size_t)offset, rw_type,
 				   flags, (size_t)len, strerror(errno));
 		}
 		return TDB_ERR_LOCK;
@@ -186,7 +186,7 @@ static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_brunlock failed (fd=%d) at offset %zu"
 				  " rw_type=%d len=%zu",
-				  tdb->fd, (size_t)offset, rw_type,
+				  tdb->file->fd, (size_t)offset, rw_type,
 				  (size_t)len);
 	}
 	return TDB_SUCCESS;
@@ -202,14 +202,14 @@ enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
 {
 	int count = 1000;
 
-	if (tdb->allrecord_lock.count != 1) {
+	if (tdb->file->allrecord_lock.count != 1) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_allrecord_upgrade failed:"
 				  " count %u too high",
-				  tdb->allrecord_lock.count);
+				  tdb->file->allrecord_lock.count);
 	}
 
-	if (tdb->allrecord_lock.off != 1) {
+	if (tdb->file->allrecord_lock.off != 1) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_allrecord_upgrade failed:"
 				  " already upgraded?");
@@ -220,8 +220,8 @@ enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
 		if (tdb_brlock(tdb, F_WRLCK,
 			       TDB_HASH_LOCK_START, 0,
 			       TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
-			tdb->allrecord_lock.ltype = F_WRLCK;
-			tdb->allrecord_lock.off = 0;
+			tdb->file->allrecord_lock.ltype = F_WRLCK;
+			tdb->file->allrecord_lock.off = 0;
 			return TDB_SUCCESS;
 		}
 		if (errno != EDEADLK) {
@@ -241,9 +241,9 @@ static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
 {
 	unsigned int i;
 
-	for (i=0; i<tdb->num_lockrecs; i++) {
-		if (tdb->lockrecs[i].off == offset) {
-			return &tdb->lockrecs[i];
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off == offset) {
+			return &tdb->file->lockrecs[i];
 		}
 	}
 	return NULL;
@@ -303,7 +303,7 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 		return TDB_SUCCESS;
 	}
 
-	if (tdb->num_lockrecs
+	if (tdb->file->num_lockrecs
 	    && offset >= TDB_HASH_LOCK_START
 	    && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
@@ -311,15 +311,15 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 	}
 
 	new_lck = (struct tdb_lock_type *)realloc(
-		tdb->lockrecs,
-		sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
+		tdb->file->lockrecs,
+		sizeof(*tdb->file->lockrecs) * (tdb->file->num_lockrecs+1));
 	if (new_lck == NULL) {
 		return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 				  "tdb_nest_lock:"
 				  " unable to allocate %zu lock struct",
-				  tdb->num_lockrecs + 1);
+				  tdb->file->num_lockrecs + 1);
 	}
-	tdb->lockrecs = new_lck;
+	tdb->file->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 */
@@ -330,7 +330,7 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 
 	/* First time we grab a lock, perhaps someone died in commit? */
 	if (!(flags & TDB_LOCK_NOCHECK)
-	    && tdb->num_lockrecs == 0) {
+	    && tdb->file->num_lockrecs == 0) {
 		tdb_bool_err berr = tdb_needs_recovery(tdb);
 		if (berr != false) {
 			tdb_brunlock(tdb, ltype, offset, 1);
@@ -348,10 +348,10 @@ static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
 		}
 	}
 
-	tdb->lockrecs[tdb->num_lockrecs].off = offset;
-	tdb->lockrecs[tdb->num_lockrecs].count = 1;
-	tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
-	tdb->num_lockrecs++;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].off = offset;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].count = 1;
+	tdb->file->lockrecs[tdb->file->num_lockrecs].ltype = ltype;
+	tdb->file->num_lockrecs++;
 
 	return TDB_SUCCESS;
 }
@@ -389,7 +389,7 @@ static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
 	 * Shrink the array by overwriting the element just unlocked with the
 	 * last array element.
 	 */
-	*lck = tdb->lockrecs[--tdb->num_lockrecs];
+	*lck = tdb->file->lockrecs[--tdb->file->num_lockrecs];
 
 	return ecode;
 }
@@ -452,17 +452,18 @@ enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
 	enum TDB_ERROR ecode;
 	tdb_bool_err berr;
 
-	if (tdb->allrecord_lock.count
-	    && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
-		tdb->allrecord_lock.count++;
+	if (tdb->file->allrecord_lock.count
+	    && (ltype == F_RDLCK
+		|| tdb->file->allrecord_lock.ltype == F_WRLCK)) {
+		tdb->file->allrecord_lock.count++;
 		return TDB_SUCCESS;
 	}
 
-	if (tdb->allrecord_lock.count) {
+	if (tdb->file->allrecord_lock.count) {
 		/* a global lock of a different type exists */
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
 				  "tdb_allrecord_lock: already have %s lock",
-				  tdb->allrecord_lock.ltype == F_RDLCK
+				  tdb->file->allrecord_lock.ltype == F_RDLCK
 				  ? "read" : "write");
 	}
 
@@ -507,11 +508,11 @@ again:
 		return ecode;
 	}
 
-	tdb->allrecord_lock.count = 1;
+	tdb->file->allrecord_lock.count = 1;
 	/* If it's upgradable, it's actually exclusive so we can treat
 	 * it as a write lock. */
-	tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
-	tdb->allrecord_lock.off = upgradable;
+	tdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
+	tdb->file->allrecord_lock.off = upgradable;
 
 	/* Now check for needing recovery. */
 	if (flags & TDB_LOCK_NOCHECK)
@@ -543,7 +544,8 @@ void tdb_unlock_open(struct tdb_context *tdb)
 
 bool tdb_has_open_lock(struct tdb_context *tdb)
 {
-	return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
+	return !(tdb->flags & TDB_NOLOCK)
+		&& find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
 }
 
 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype)
@@ -561,15 +563,15 @@ void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
 /* unlock entire db */
 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
 {
-	if (tdb->allrecord_lock.count == 0) {
+	if (tdb->file->allrecord_lock.count == 0) {
 		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
 			   "tdb_allrecord_unlock: not locked!");
 		return;
 	}
 
 	/* Upgradable locks are marked as write locks. */
-	if (tdb->allrecord_lock.ltype != ltype
-	    && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
+	if (tdb->file->allrecord_lock.ltype != ltype
+	    && (!tdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
 		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 			   "tdb_allrecord_unlock: have %s lock",
 			   tdb->allrecord_lock.ltype == F_RDLCK
@@ -577,13 +579,13 @@ void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
 		return;
 	}
 
-	if (tdb->allrecord_lock.count > 1) {
-		tdb->allrecord_lock.count--;
+	if (tdb->file->allrecord_lock.count > 1) {
+		tdb->file->allrecord_lock.count--;
 		return;
 	}
 
-	tdb->allrecord_lock.count = 0;
-	tdb->allrecord_lock.ltype = 0;
+	tdb->file->allrecord_lock.count = 0;
+	tdb->file->allrecord_lock.ltype = 0;
 
 	tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
 }
@@ -597,10 +599,10 @@ bool tdb_has_hash_locks(struct tdb_context *tdb)
 {
 	unsigned int i;
 
-	for (i=0; i<tdb->num_lockrecs; i++) {
-		if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
-		    && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
-					       + TDB_HASH_LOCK_RANGE))
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off >= TDB_HASH_LOCK_START
+		    && tdb->file->lockrecs[i].off < (TDB_HASH_LOCK_START
+						     + TDB_HASH_LOCK_RANGE))
 			return true;
 	}
 	return false;
@@ -610,8 +612,11 @@ static bool tdb_has_free_lock(struct tdb_context *tdb)
 {
 	unsigned int i;
 
-	for (i=0; i<tdb->num_lockrecs; i++) {
-		if (tdb->lockrecs[i].off
+	if (tdb->flags & TDB_NOLOCK)
+		return false;
+
+	for (i=0; i<tdb->file->num_lockrecs; i++) {
+		if (tdb->file->lockrecs[i].off
 		    > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
 			return true;
 	}
@@ -628,16 +633,16 @@ enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
 		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
 
 	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count &&
-	    (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
+	if (tdb->file->allrecord_lock.count &&
+	    (ltype == tdb->file->allrecord_lock.ltype || ltype == F_RDLCK)) {
 		return TDB_SUCCESS;
 	}
 
-	if (tdb->allrecord_lock.count) {
+	if (tdb->file->allrecord_lock.count) {
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
 				  "tdb_lock_hashes:"
 				  " already have %s allrecordlock",
-				  tdb->allrecord_lock.ltype == F_RDLCK
+				  tdb->file->allrecord_lock.ltype == F_RDLCK
 				  ? "read" : "write");
 	}
 
@@ -662,9 +667,12 @@ enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
 	unsigned lock = TDB_HASH_LOCK_START
 		+ (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
 
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
 	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count) {
-		if (tdb->allrecord_lock.ltype == F_RDLCK
+	if (tdb->file->allrecord_lock.count) {
+		if (tdb->file->allrecord_lock.ltype == F_RDLCK
 		    && ltype == F_WRLCK) {
 			return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 					  "tdb_unlock_hashes RO allrecord!");
@@ -691,9 +699,12 @@ enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
 {
 	assert(b_off >= sizeof(struct tdb_header));
 
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
 	/* a allrecord lock allows us to avoid per chain locks */
-	if (tdb->allrecord_lock.count) {
-		if (tdb->allrecord_lock.ltype == F_WRLCK)
+	if (tdb->file->allrecord_lock.count) {
+		if (tdb->file->allrecord_lock.ltype == F_WRLCK)
 			return 0;
 		return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 				  "tdb_lock_free_bucket with"
@@ -713,15 +724,8 @@ enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
 
 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
 {
-	if (tdb->allrecord_lock.count)
+	if (tdb->file->allrecord_lock.count)
 		return;
 
 	tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
 }
-
-void tdb_lock_init(struct tdb_context *tdb)
-{
-	tdb->num_lockrecs = 0;
-	tdb->lockrecs = NULL;
-	tdb->allrecord_lock.count = 0;
-}

ccan/tdb2/open.c

 #include "private.h"
 
-/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
-static struct tdb_context *tdbs = NULL;
+/* all lock info, to detect double-opens (fcntl file don't nest!) */
+static struct tdb_file *files = NULL;
 
-static bool tdb_already_open(dev_t device, ino_t ino)
+static struct tdb_file *find_file(dev_t device, ino_t ino)
 {
-	struct tdb_context *i;
+	struct tdb_file *i;
 
-	for (i = tdbs; i; i = i->next) {
+	for (i = files; i; i = i->next) {
 		if (i->device == device && i->inode == ino) {
-			return true;
+			break;
 		}
 	}
-
-	return false;
+	return i;
 }
 
 static bool read_all(int fd, void *buf, size_t len)
@@ -143,19 +142,19 @@ static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
 		memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 		return TDB_SUCCESS;
 	}
-	if (lseek(tdb->fd, 0, SEEK_SET) == -1) {
+	if (lseek(tdb->file->fd, 0, SEEK_SET) == -1) {
 		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 				  "tdb_new_database:"
 				  " failed to seek: %s", strerror(errno));
 	}
 
-	if (ftruncate(tdb->fd, 0) == -1) {
+	if (ftruncate(tdb->file->fd, 0) == -1) {
 		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 				  "tdb_new_database:"
 				  " failed to truncate: %s", strerror(errno));
 	}
 
-	rlen = write(tdb->fd, &newdb, sizeof(newdb));
+	rlen = write(tdb->file->fd, &newdb, sizeof(newdb));
 	if (rlen != sizeof(newdb)) {
 		if (rlen >= 0)
 			errno = ENOSPC;
@@ -190,16 +189,15 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 	tdb->name = NULL;
 	tdb->map_ptr = NULL;
 	tdb->direct_access = 0;
-	tdb->fd = -1;
 	tdb->map_size = sizeof(struct tdb_header);
 	tdb->flags = tdb_flags;
 	tdb->logfn = NULL;
 	tdb->transaction = NULL;
 	tdb->stats = NULL;
 	tdb->access = NULL;
+	tdb->file = NULL;
 	tdb_hash_init(tdb);
 	tdb_io_init(tdb);
-	tdb_lock_init(tdb);
 
 	while (attr) {
 		switch (attr->base.attr) {
@@ -266,18 +264,56 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		return tdb;
 	}
 
-	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
-		/* errno set by open(2) */
-		saved_errno = errno;
-		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+	if (stat(name, &st) != -1)
+		tdb->file = find_file(st.st_dev, st.st_ino);
+
+	if (!tdb->file) {
+		int fd;
+
+		if ((fd = open(name, open_flags, mode)) == -1) {
+			/* errno set by open(2) */
+			saved_errno = errno;
+			tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 				   "tdb_open: could not open file %s: %s",
 				   name, strerror(errno));
-		goto fail;
-	}
+			goto fail;
+		}
+
+		/* on exec, don't inherit the fd */
+		v = fcntl(fd, F_GETFD, 0);
+		fcntl(fd, F_SETFD, v | FD_CLOEXEC);
+
+		if (fstat(fd, &st) == -1) {
+			saved_errno = errno;
+			tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				   "tdb_open: could not stat open %s: %s",
+				   name, strerror(errno));
+			goto fail;
+		}
+
+		tdb->file = malloc(sizeof(*tdb->file));
+		if (!tdb->file) {
+			saved_errno = ENOMEM;
+			tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+				   "tdb_open: could alloc file");
+			goto fail;
+		}
 
-	/* on exec, don't inherit the fd */
-	v = fcntl(tdb->fd, F_GETFD, 0);
-        fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
+		tdb->file->next = files;
+		tdb->file->num_lockrecs = 0;
+		tdb->file->lockrecs = NULL;
+		tdb->file->allrecord_lock.count = 0;
+		tdb->file->fd = fd;
+		tdb->file->device = st.st_dev;
+		tdb->file->inode = st.st_ino;
+	} else {
+		/* FIXME */
+		ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+				   "tdb_open: %s (%d,%d) is already open in"
+				   " this process",
+				   name, (int)st.st_dev, (int)st.st_ino);
+ 		goto fail;
+ 	}
 
 	/* ensure there is only one process initialising at once */
 	ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
@@ -286,7 +322,7 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 	}
 
 	/* If they used O_TRUNC, read will return 0. */
-	rlen = read(tdb->fd, &hdr, sizeof(hdr));
+	rlen = read(tdb->file->fd, &hdr, sizeof(hdr));
 	if (rlen == 0 && (open_flags & O_CREAT)) {
 		ecode = tdb_new_database(tdb, seed, &hdr);
 		if (ecode != TDB_SUCCESS) {
@@ -330,24 +366,6 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		goto fail;
 	}
 
-	if (fstat(tdb->fd, &st) == -1) {
-		saved_errno = errno;
-		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
-				   "tdb_open: could not stat open %s: %s",
-				   name, strerror(errno));
-		goto fail;
-	}
-
-	/* Is it already in the open list?  If so, fail. */
-	if (tdb_already_open(st.st_dev, st.st_ino)) {
-		/* FIXME */
-		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
-				   "tdb_open: %s (%d,%d) is already open"
-				   " in this process",
-				   name, (int)st.st_dev, (int)st.st_ino);
-		goto fail;
-	}
-
 	tdb->name = strdup(name);
 	if (!tdb->name) {
 		ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
@@ -365,8 +383,6 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 			goto fail;
 	}
 
-	tdb->device = st.st_dev;
-	tdb->inode = st.st_ino;
 	tdb_unlock_open(tdb);
 
 	/* This make sure we have current map_size and mmap. */
@@ -390,8 +406,8 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		goto fail;
 	}
 
-	tdb->next = tdbs;
-	tdbs = tdb;
+	/* Add to linked list. */
+	files = tdb->file;
 	return tdb;
 
  fail:
@@ -426,13 +442,16 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		} else
 			tdb_munmap(tdb);
 	}
-	free(tdb->lockrecs);
 	free((char *)tdb->name);
-	if (tdb->fd != -1)
-		if (close(tdb->fd) != 0)
+	if (tdb->file) {
+		if (close(tdb->file->fd) != 0)
 			tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
-				   "tdb_open: failed to close tdb->fd"
+				   "tdb_open: failed to close tdb fd"
 				   " on error: %s", strerror(errno));
+		free(tdb->file->lockrecs);
+		free(tdb->file);
+	}
+
 	free(tdb);
 	errno = saved_errno;
 	return NULL;
@@ -440,7 +459,6 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 
 int tdb_close(struct tdb_context *tdb)
 {
-	struct tdb_context **i;
 	int ret = 0;
 
 	tdb_trace(tdb, "tdb_close");
@@ -456,18 +474,19 @@ int tdb_close(struct tdb_context *tdb)
 			tdb_munmap(tdb);
 	}
 	free((char *)tdb->name);
-	if (tdb->fd != -1) {
-		ret = close(tdb->fd);
-		tdb->fd = -1;
-	}
-	free(tdb->lockrecs);
-
-	/* Remove from contexts list */
-	for (i = &tdbs; *i; i = &(*i)->next) {
-		if (*i == tdb) {
-			*i = tdb->next;
-			break;
+	if (tdb->file) {
+		struct tdb_file **i;
+		ret = close(tdb->file->fd);
+
+		/* Remove from files list */
+		for (i = &files; *i; i = &(*i)->next) {
+			if (*i == tdb->file) {
+				*i = tdb->file->next;
+				break;
+			}
 		}
+		free(tdb->file->lockrecs);
+		free(tdb->file);
 	}
 
 #ifdef TDB_TRACE

ccan/tdb2/private.h

@@ -315,6 +315,23 @@ struct tdb_access_hdr {
 	bool convert;
 };
 
+struct tdb_file {
+	/* Single list of all TDBs, to detect multiple opens. */
+	struct tdb_file *next;
+
+	/* The file descriptor. */
+	int fd;
+
+	/* Lock information */
+	struct tdb_lock_type allrecord_lock;
+	size_t num_lockrecs;
+	struct tdb_lock_type *lockrecs;
+
+	/* Identity of this file. */
+	dev_t device;
+	ino_t inode;
+};
+
 struct tdb_context {
 	/* Filename of the database. */
 	const char *name;
@@ -325,9 +342,6 @@ struct tdb_context {
 	/* Are we accessing directly? (debugging check). */
 	int direct_access;
 
-	 /* Open file descriptor (undefined for TDB_INTERNAL). */
-	int fd;
-
 	/* How much space has been mapped (<= current file size) */
 	tdb_len_t map_size;
 
@@ -365,20 +379,13 @@ struct tdb_context {
 	/* IO methods: changes for transactions. */
 	const struct tdb_methods *methods;
 
-	/* Lock information */
-	struct tdb_lock_type allrecord_lock;
-	size_t num_lockrecs;
-	struct tdb_lock_type *lockrecs;
-
 	struct tdb_attribute_stats *stats;
 
 	/* Direct access information */
 	struct tdb_access_hdr *access;
 
-	/* Single list of all TDBs, to avoid multiple opens. */
-	struct tdb_context *next;
-	dev_t device;
-	ino_t inode;
+	/* The actual file information */
+	struct tdb_file *file;
 };
 
 struct tdb_methods {
@@ -521,8 +528,6 @@ void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
 	} while (0)
 
 /* lock.c: */
-void tdb_lock_init(struct tdb_context *tdb);
-
 /* Lock/unlock a range of hashes. */
 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
 			       tdb_off_t hash_lock, tdb_len_t hash_range,

ccan/tdb2/test/failtest_helper.h

@@ -4,7 +4,7 @@
 #include <stdbool.h>
 
 /* FIXME: Check these! */
-#define INITIAL_TDB_MALLOC	"open.c", 184, FAILTEST_MALLOC
+#define INITIAL_TDB_MALLOC	"open.c", 183, FAILTEST_MALLOC
 #define URANDOM_OPEN		"open.c", 43, FAILTEST_OPEN
 #define URANDOM_READ		"open.c", 23, FAILTEST_READ
 

ccan/tdb2/test/run-03-coalesce.c

@@ -94,7 +94,7 @@ int main(int argc, char *argv[])
 	/* Lock and coalesce. */
 	ok1(tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == 0);
 	ok1(coalesce(tdb, layout->elem[1].base.off, b_off, 1024) == 1);
-	ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+	ok1(tdb->file->allrecord_lock.count == 0 && tdb->file->num_lockrecs == 0);
 	ok1(free_record_length(tdb, layout->elem[1].base.off)
 	    == 1024 + sizeof(struct tdb_used_record) + 2048);
 	ok1(tdb_check(tdb, NULL, NULL) == 0);
@@ -117,7 +117,7 @@ int main(int argc, char *argv[])
 	/* Lock and coalesce. */
 	ok1(tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == 0);
 	ok1(coalesce(tdb, layout->elem[1].base.off, b_off, 1024) == 1);
-	ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+	ok1(tdb->file->allrecord_lock.count == 0 && tdb->file->num_lockrecs == 0);
 	ok1(free_record_length(tdb, layout->elem[1].base.off)
 	    == 1024 + sizeof(struct tdb_used_record) + 512);
 	ok1(tdb_check(tdb, NULL, NULL) == 0);
@@ -141,7 +141,8 @@ int main(int argc, char *argv[])
 	/* Lock and coalesce. */
 	ok1(tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == 0);
 	ok1(coalesce(tdb, layout->elem[1].base.off, b_off, 1024) == 1);
-	ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+	ok1(tdb->file->allrecord_lock.count == 0
+	    && tdb->file->num_lockrecs == 0);
 	ok1(free_record_length(tdb, layout->elem[1].base.off)
 	    == 1024 + sizeof(struct tdb_used_record) + 512
 	    + sizeof(struct tdb_used_record) + 256);

ccan/tdb2/test/run-04-basichash.c

@@ -60,9 +60,9 @@ int main(int argc, char *argv[])
 		ok1(h.hlock_start == 0);
 		ok1(h.hlock_range == 
 		    1ULL << (64-(TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS)));
-		ok1((tdb->flags & TDB_NOLOCK) || tdb->num_lockrecs == 1);
+		ok1((tdb->flags & TDB_NOLOCK) || tdb->file->num_lockrecs == 1);
 		ok1((tdb->flags & TDB_NOLOCK)
-		    || tdb->lockrecs[0].off == TDB_HASH_LOCK_START);
+		    || tdb->file->lockrecs[0].off == TDB_HASH_LOCK_START);
 		/* FIXME: Check lock length */
 
 		/* Allocate a new record. */
@@ -101,9 +101,9 @@ int main(int argc, char *argv[])
 		ok1(h.hlock_start == 0);
 		ok1(h.hlock_range == 
 		    1ULL << (64-(TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS)));
-		ok1((tdb->flags & TDB_NOLOCK) || tdb->num_lockrecs == 1);
+		ok1((tdb->flags & TDB_NOLOCK) || tdb->file->num_lockrecs == 1);
 		ok1((tdb->flags & TDB_NOLOCK)
-		    || tdb->lockrecs[0].off == TDB_HASH_LOCK_START);
+		    || tdb->file->lockrecs[0].off == TDB_HASH_LOCK_START);
 		/* FIXME: Check lock length */
 
 		ok1(tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
@@ -127,9 +127,9 @@ int main(int argc, char *argv[])
 		ok1(h.hlock_start == 0);
 		ok1(h.hlock_range == 
 		    1ULL << (64-(TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS)));
-		ok1((tdb->flags & TDB_NOLOCK) || tdb->num_lockrecs == 1);
+		ok1((tdb->flags & TDB_NOLOCK) || tdb->file->num_lockrecs == 1);
 		ok1((tdb->flags & TDB_NOLOCK)
-		    || tdb->lockrecs[0].off == TDB_HASH_LOCK_START);
+		    || tdb->file->lockrecs[0].off == TDB_HASH_LOCK_START);
 		/* FIXME: Check lock length */
 
 		/* Make it expand 0'th bucket. */
@@ -165,9 +165,9 @@ int main(int argc, char *argv[])
 		ok1(h.hlock_start == 0);
 		ok1(h.hlock_range == 
 		    1ULL << (64-(TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS)));
-		ok1((tdb->flags & TDB_NOLOCK) || tdb->num_lockrecs == 1);
+		ok1((tdb->flags & TDB_NOLOCK) || tdb->file->num_lockrecs == 1);
 		ok1((tdb->flags & TDB_NOLOCK)
-		    || tdb->lockrecs[0].off == TDB_HASH_LOCK_START);
+		    || tdb->file->lockrecs[0].off == TDB_HASH_LOCK_START);
 		/* FIXME: Check lock length */
 
 		/* Simple delete should work. */
@@ -196,9 +196,9 @@ int main(int argc, char *argv[])
 		ok1(h.hlock_start == 0);
 		ok1(h.hlock_range == 
 		    1ULL << (64-(TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS)));
-		ok1((tdb->flags & TDB_NOLOCK) || tdb->num_lockrecs == 1);
+		ok1((tdb->flags & TDB_NOLOCK) || tdb->file->num_lockrecs == 1);
 		ok1((tdb->flags & TDB_NOLOCK)
-		    || tdb->lockrecs[0].off == TDB_HASH_LOCK_START);
+		    || tdb->file->lockrecs[0].off == TDB_HASH_LOCK_START);
 		/* FIXME: Check lock length */
 
 		ok1(expand_group(tdb, &h) == 0);

ccan/tdb2/test/run-13-delete.c

@@ -180,7 +180,8 @@ int main(int argc, char *argv[])
 		/* Check mixed bitpattern. */
 		test_val(tdb, 0x123456789ABCDEF0ULL);
 
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
+				   && tdb->file->num_lockrecs == 0));
 		tdb_close(tdb);
 
 		/* Deleting these entries in the db gave problems. */

ccan/tdb2/test/run-15-append.c

@@ -69,7 +69,8 @@ int main(int argc, char *argv[])
 				moves++;
 			oldoff = newoff;
 		}
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
+				   && tdb->file->num_lockrecs == 0));
 		/* We should increase by 50% each time... */
 		ok(moves <= ilog64(j / SIZE_STEP)*2, "Moved %u times", moves);
 		tdb_close(tdb);
@@ -100,7 +101,8 @@ int main(int argc, char *argv[])
 				moves++;
 			oldoff = newoff;
 		}
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
+				   && tdb->file->num_lockrecs == 0));
 		/* We should increase by 50% each time... */
 		ok(moves <= ilog64(j / SIZE_STEP)*2, "Moved %u times", moves);
 		tdb_close(tdb);
@@ -122,7 +124,8 @@ int main(int argc, char *argv[])
 		ok1(data.dsize == MAX_SIZE);
 		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
 		free(data.dptr);
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
+				   && tdb->file->num_lockrecs == 0));
 		tdb_close(tdb);
 	}
 

ccan/tdb2/test/run-55-transaction.c

@@ -43,7 +43,8 @@ int main(int argc, char *argv[])
 
 		/* Cancelling a transaction means no store */
 		tdb_transaction_cancel(tdb);
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(tdb->file->allrecord_lock.count == 0
+		    && tdb->file->num_lockrecs == 0);
 		ok1(tdb_check(tdb, NULL, NULL) == 0);
 		ok1(tdb_fetch(tdb, key, &data) == TDB_ERR_NOEXIST);
 
@@ -57,7 +58,8 @@ int main(int argc, char *argv[])
 		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
 		free(data.dptr);
 		ok1(tdb_transaction_commit(tdb) == 0);
-		ok1(tdb->allrecord_lock.count == 0 && tdb->num_lockrecs == 0);
+		ok1(tdb->file->allrecord_lock.count == 0
+		    && tdb->file->num_lockrecs == 0);
 		ok1(tdb_check(tdb, NULL, NULL) == 0);
 		ok1(tdb_fetch(tdb, key, &data) == TDB_SUCCESS);
 		ok1(data.dsize == 1000);

ccan/tdb2/test/run-57-die-during-transaction.c

@@ -153,7 +153,7 @@ reset:
 
 	if (setjmp(jmpbuf) != 0) {
 		/* We're partway through.  Simulate our death. */
-		close(tdb->fd);
+		close(tdb->file->fd);
 		forget_locking();
 		in_transaction = false;
 

ccan/tdb2/test/run-remap-in-read_traverse.c

@@ -51,7 +51,7 @@ int main(int argc, char *argv[])
 		       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
 
 	ok1(external_agent_operation(agent, OPEN, filename) == SUCCESS);
-	i = add_records_to_grow(agent, tdb->fd, tdb->map_size);
+	i = add_records_to_grow(agent, tdb->file->fd, tdb->map_size);
 
 	/* Do a traverse. */
 	ok1(tdb_traverse(tdb, NULL, NULL) == i);

ccan/tdb2/transaction.c

@@ -430,7 +430,7 @@ static enum TDB_ERROR transaction_sync(struct tdb_context *tdb,
 		return TDB_SUCCESS;
 	}
 
-	if (fsync(tdb->fd) != 0) {
+	if (fsync(tdb->file->fd) != 0) {
 		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 				  "tdb_transaction: fsync failed: %s",
 				  strerror(errno));
@@ -495,8 +495,8 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 		}
 	}
 
-	if (tdb->allrecord_lock.count)
-		tdb_allrecord_unlock(tdb, tdb->allrecord_lock.ltype);
+	if (tdb->file->allrecord_lock.count)
+		tdb_allrecord_unlock(tdb, tdb->file->allrecord_lock.ltype);
 
 	/* restore the normal io methods */
 	tdb->methods = tdb->transaction->io_methods;