2022-11-15 13:56:06

by Yu Kuai

Subject: [PATCH RFC v3 05/10] dm: make sure create and remove dm device won't race with open and close table

From: Yu Kuai <[email protected]>

open_table_device() and close_table_device() is protected by
table_devices_lock, hence use it to protect add_disk() and
del_gendisk().

Prepare to track per-add_disk holder relations in dm.

Signed-off-by: Yu Kuai <[email protected]>
---
drivers/md/dm.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)

diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 2917700b1e15..3728b56b364b 100644
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.c
@@ -1952,7 +1952,14 @@ static void cleanup_mapped_device(struct mapped_device *md)
spin_unlock(&_minor_lock);
if (dm_get_md_type(md) != DM_TYPE_NONE) {
dm_sysfs_exit(md);
+
+ /*
+ * Hold lock to make sure del_gendisk() won't concurrent
+ * with open/close_table_device().
+ */
+ mutex_lock(&md->table_devices_lock);
del_gendisk(md->disk);
+ mutex_unlock(&md->table_devices_lock);
}
dm_queue_destroy_crypto_profile(md->queue);
put_disk(md->disk);
@@ -2312,15 +2319,24 @@ int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
if (r)
return r;

+ /*
+ * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
+ * with open_table_device() and close_table_device().
+ */
+ mutex_lock(&md->table_devices_lock);
r = add_disk(md->disk);
+ mutex_unlock(&md->table_devices_lock);
if (r)
return r;

r = dm_sysfs_init(md);
if (r) {
+ mutex_lock(&md->table_devices_lock);
del_gendisk(md->disk);
+ mutex_unlock(&md->table_devices_lock);
return r;
}
+
md->type = type;
return 0;
}
--
2.31.1

2022-11-16 07:20:49

by Christoph Hellwig

Subject: Re: [PATCH RFC v3 05/10] dm: make sure create and remove dm device won't race with open and close table

Looks good:

Reviewed-by: Christoph Hellwig <[email protected]>

2022-11-16 22:05:41

by Mike Snitzer

Subject: Re: [PATCH RFC v3 05/10] dm: make sure create and remove dm device won't race with open and close table

On Tue, Nov 15 2022 at 9:10P -0500,
Yu Kuai <[email protected]> wrote:

> From: Yu Kuai <[email protected]>
>
> open_table_device() and close_table_device() is protected by
> table_devices_lock, hence use it to protect add_disk() and
> del_gendisk().
>
> Prepare to track per-add_disk holder relations in dm.
>
> Signed-off-by: Yu Kuai <[email protected]>
> ---
> drivers/md/dm.c | 16 ++++++++++++++++
> 1 file changed, 16 insertions(+)
>
> diff --git a/drivers/md/dm.c b/drivers/md/dm.c
> index 2917700b1e15..3728b56b364b 100644
> --- a/drivers/md/dm.c
> +++ b/drivers/md/dm.c
> @@ -1952,7 +1952,14 @@ static void cleanup_mapped_device(struct mapped_device *md)
> spin_unlock(&_minor_lock);
> if (dm_get_md_type(md) != DM_TYPE_NONE) {
> dm_sysfs_exit(md);
> +
> + /*
> + * Hold lock to make sure del_gendisk() won't concurrent
> + * with open/close_table_device().
> + */
> + mutex_lock(&md->table_devices_lock);
> del_gendisk(md->disk);
> + mutex_unlock(&md->table_devices_lock);
> }
> dm_queue_destroy_crypto_profile(md->queue);
> put_disk(md->disk);
> @@ -2312,15 +2319,24 @@ int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
> if (r)
> return r;
>
> + /*
> + * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
> + * with open_table_device() and close_table_device().
> + */
> + mutex_lock(&md->table_devices_lock);
> r = add_disk(md->disk);
> + mutex_unlock(&md->table_devices_lock);
> if (r)
> return r;
>
> r = dm_sysfs_init(md);
> if (r) {
> + mutex_lock(&md->table_devices_lock);
> del_gendisk(md->disk);
> + mutex_unlock(&md->table_devices_lock);
> return r;
> }
> +
> md->type = type;
> return 0;
> }
> --
> 2.31.1
>

In the new comments added: s/concurrent/race/ ?

But otherwise:

Reviewed-by: Mike Snitzer <[email protected]>