On 04/14/2011 11:08 AM, NeilBrown wrote:
>
> I would mark mddev_lock as __must_check__ (or however it is spelt) and fix
> the errors produced.
>
> I think the call in md_stop_writes is the only one where an uninterruptible
> wait is needed - I'd probaby just change that to an explicit "mutex_lock",
> but I wouldn't be against creating "mddev_lock_uninterruptible" for that case.
>
That is fine for me, but actually I do not see a good way to fix
rdev_size_store as well (at least for mddev_lock(my_mddev)).
Please find another try.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Alexey Khoroshilov <khoroshilov@xxxxxxxxx>
---
drivers/md/md.c | 22 ++++++++++++++++------
1 files changed, 16 insertions(+), 6 deletions(-)
diff --git a/drivers/md/md.c b/drivers/md/md.c
index b12b377..4f83f16 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -623,11 +623,16 @@ static mddev_t * mddev_find(dev_t unit)
goto retry;
}
-static inline int mddev_lock(mddev_t * mddev)
+static inline int __must_check mddev_lock(mddev_t *mddev)
{
return mutex_lock_interruptible(&mddev->reconfig_mutex);
}
+static inline void mddev_lock_uninterruptible(mddev_t *mddev)
+{
+ return mutex_lock(&mddev->reconfig_mutex);
+}
+
static inline int mddev_is_locked(mddev_t *mddev)
{
return mutex_is_locked(&mddev->reconfig_mutex);
@@ -2610,13 +2615,18 @@ rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
*/
mddev_t *mddev;
int overlap = 0;
+ int intr = 0;
struct list_head *tmp;
mddev_unlock(my_mddev);
for_each_mddev(mddev, tmp) {
mdk_rdev_t *rdev2;
- mddev_lock(mddev);
+ if (mddev_lock(mddev)) {
+ mddev_put(mddev);
+ intr = 1;
+ break;
+ }
list_for_each_entry(rdev2, &mddev->disks, same_set)
if (rdev->bdev == rdev2->bdev &&
rdev != rdev2 &&
@@ -2632,8 +2642,8 @@ rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
break;
}
}
- mddev_lock(my_mddev);
- if (overlap) {
+ mddev_lock_uninterruptible(my_mddev);
+ if (overlap || intr) {
/* Someone else could have slipped in a size
* change here, but doing so is just silly.
* We put oldsectors back because we *know* it is
@@ -2641,7 +2651,7 @@ rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
* itself
*/
rdev->sectors = oldsectors;
- return -EBUSY;
+ return overlap ? -EBUSY : -EINTR;
}
}
return len;
@@ -4748,7 +4758,7 @@ static void __md_stop_writes(mddev_t *mddev)
void md_stop_writes(mddev_t *mddev)
{
- mddev_lock(mddev);
+ mddev_lock_uninterruptible(mddev);
__md_stop_writes(mddev);
mddev_unlock(mddev);
}
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
