Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Luis Chamberlain <mcgrof@kernel.org>]

On Wed, Mar 10, 2021 at 01:11:15PM +0000, Luis Chamberlain wrote:
> I can try to modify it to include second patch first, as that is
> required. There are two separate bugs here.

I tried this, applying the syfs required changes first and then
applying your idea as a secondary patch ends up like this:

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index 4b57c84ba9d4..bb45c1e0f3e0 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -1702,6 +1702,7 @@ static void zram_reset_device(struct zram *zram)
 
 	set_capacity_and_notify(zram->disk, 0);
 	part_stat_set_all(zram->disk->part0, 0);
+	module_put(THIS_MODULE);
 
 	up_write(&zram->init_lock);
 	/* I/O operation under all of CPU are done so let's free */
@@ -1747,6 +1748,7 @@ static ssize_t disksize_store(struct device *dev,
 		goto out_free_meta;
 	}
 
+	BUG_ON(!try_module_get(THIS_MODULE));
 	zram->comp = comp;
 	zram->disksize = disksize;
 	set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);

The BUG_ON() is doable as we *know* we already have a reference to the
module due to the beginning of the other try_module_get() which would
be placed on the first patch at the top of disksize_store().

This however doesn't fix the issue. We end up in a situation where we
cannot unload the zram driver.

  Luis