The previous patch serialized reference counting with respect to users with
valid references, but there is a chicken-egg problem between the block layer
and dm that causes the gendisk associated with a mapping to keep a reference-
less pointer to the mapped_device.
Without major reconstruction, the pointer must be referenceless or we run
into a chicken-egg problem with freeing the disk or the mapped_device. When
the block layer is actually using the mapped_device, a reference is taken.
This patch adds a flag to indicate that the mapped_device is no longer valid.
In free_dev(), we reset the ->private_data pointer, but prior to that, the
block layer can still attempt to open the device. The flag is used to
check the validity of the pointer, and return -ENXIO if it is invalid.
Otherwise, an appropriate reference is taken.
Signed-off-by: Jeff Mahoney <[email protected]>
--
drivers/md/dm.c | 25 +++++++++++++++++++++++--
1 files changed, 23 insertions(+), 2 deletions(-)
diff -ruNpX ../dontdiff linux-2.6.16-staging1/drivers/md/dm.c linux-2.6.16-staging2/drivers/md/dm.c
--- linux-2.6.16-staging1/drivers/md/dm.c 2006-04-13 16:18:20.000000000 -0400
+++ linux-2.6.16-staging2/drivers/md/dm.c 2006-04-13 16:18:20.000000000 -0400
@@ -60,6 +60,7 @@ union map_info *dm_get_mapinfo(struct bi
#define DMF_BLOCK_IO 0
#define DMF_SUSPENDED 1
#define DMF_FROZEN 2
+#define DMF_FREEING 3
struct mapped_device {
struct rw_semaphore io_lock;
@@ -216,9 +217,16 @@ static int dm_blk_open(struct inode *ino
{
struct mapped_device *md;
+ spin_lock(&_minor_lock);
md = inode->i_bdev->bd_disk->private_data;
- dm_get(md);
- return 0;
+ if (md) {
+ if (test_bit(DMF_FREEING, &md->flags) == 0)
+ dm_get(md);
+ else
+ md = NULL;
+ }
+ spin_unlock(&_minor_lock);
+ return md ? 0 : -ENXIO;
}
static int dm_blk_close(struct inode *inode, struct file *file)
@@ -870,6 +878,11 @@ static void free_dev(struct mapped_devic
mempool_destroy(md->io_pool);
del_gendisk(md->disk);
free_minor(minor);
+
+ spin_lock(&_minor_lock);
+ md->disk->private_data = NULL;
+ spin_unlock(&_minor_lock);
+
put_disk(md->disk);
blk_put_queue(md->queue);
kfree(md);
@@ -970,6 +983,11 @@ static struct mapped_device *dm_find_md(
if (md && (md == MINOR_ALLOCED || (dm_disk(md)->first_minor != minor)))
md = NULL;
+ if (md) {
+ if (test_bit(DMF_FREEING, &md->flags))
+ md = NULL;
+ }
+
spin_unlock(&_minor_lock);
return md;
@@ -1010,8 +1028,11 @@ void dm_put(struct mapped_device *md)
{
struct dm_table *map = dm_get_table(md);
+ BUG_ON(test_bit(DMF_FREEING, &md->flags));
+
if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
idr_replace(&_minor_idr, MINOR_ALLOCED, md->disk->first_minor);
+ set_bit(DMF_FREEING, &md->flags);
spin_unlock(&_minor_lock);
if (!dm_suspended(md)) {
dm_table_presuspend_targets(map);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]