When adding partitions to the disk, the reference count of the disk
object is increased. then alloc partition device and called
device_add(), if the device_add() return error, the reference
count of the disk object will be reduced twice, at put_device(pdev)
and put_disk(disk). this leads to the end of the object's life cycle
prematurely, and trigger following calltrace.
__init_work+0x2d/0x50 kernel/workqueue.c:519
synchronize_rcu_expedited+0x3af/0x650 kernel/rcu/tree_exp.h:847
bdi_remove_from_list mm/backing-dev.c:938 [inline]
bdi_unregister+0x17f/0x5c0 mm/backing-dev.c:946
release_bdi+0xa1/0xc0 mm/backing-dev.c:968
kref_put include/linux/kref.h:65 [inline]
bdi_put+0x72/0xa0 mm/backing-dev.c:976
bdev_free_inode+0x11e/0x220 block/bdev.c:408
i_callback+0x3f/0x70 fs/inode.c:226
rcu_do_batch kernel/rcu/tree.c:2508 [inline]
rcu_core+0x76d/0x16c0 kernel/rcu/tree.c:2743
__do_softirq+0x1d7/0x93b kernel/softirq.c:558
invoke_softirq kernel/softirq.c:432 [inline]
__irq_exit_rcu kernel/softirq.c:636 [inline]
irq_exit_rcu+0xf2/0x130 kernel/softirq.c:648
sysvec_apic_timer_interrupt+0x93/0xc0
making disk is NULL when calling put_disk().
Reported-by: Hao Sun <[email protected]>
Signed-off-by: Zqiang <[email protected]>
---
block/partitions/core.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/block/partitions/core.c b/block/partitions/core.c
index 3a4898433c43..4cb6803e7021 100644
--- a/block/partitions/core.c
+++ b/block/partitions/core.c
@@ -424,6 +424,7 @@ static struct block_device *add_partition(struct gendisk *disk, int partno,
device_del(pdev);
out_put:
put_device(pdev);
+ disk = NULL;
out_put_disk:
put_disk(disk);
return ERR_PTR(err);
--
2.17.1
On Mon, Oct 18, 2021 at 05:11:16PM +0800, Zqiang wrote:
> --- a/block/partitions/core.c
> +++ b/block/partitions/core.c
> @@ -424,6 +424,7 @@ static struct block_device *add_partition(struct gendisk *disk, int partno,
> device_del(pdev);
> out_put:
> put_device(pdev);
> + disk = NULL;
> out_put_disk:
> put_disk(disk);
> return ERR_PTR(err);
Stylistic nitpick: I think directly returning where this currently
clears disk would be a little easier to follow.