Re: [PATCH V2 0/4] zram: fix two races and one zram leak

On Thu, Oct 21, 2021 at 10:39:46AM -0700, Luis Chamberlain wrote:

On Wed, Oct 20, 2021 at 02:40:05PM -0700, Minchan Kim wrote:

quoted

On Wed, Oct 20, 2021 at 09:55:44AM +0800, Ming Lei wrote:

quoted

Hello,

Fixes three issues reported by Luis Chamberlain with one simpler approach:

- race between between zram_reset_device() and disksize_store() (1/4)

- zram leak during unloading module, which is one race between resetting
and removing device (2/4)

- race between zram_remove and disksize_store (3/4)

Also replace replace fsync_bdev with sync_blockdev since no one opens
it.(4/4)

V2:
	- take another approach to avoid failing of zram_remove()
	- add patch to address race between zram_reset_device() and
	  disksize_store()

Thanks for breaking the problems down, Ming.

To me, the whole patchset looks good to me since each patch solves
the problem step by step and finally fix.

Luis, do you have any concern of this patchset to solve the cpuhp
problem? (Sorry in advance if I miss some concerns if you raised
in different thread. I'm totally lost).

Running tests against this now. Will report back!

So indeed with these patches I end up in the situation where we if
if spawn two ltp zram02.sh runs and cancel then randomly and start
them again:

zram: Can't change algorithm for initialized device

And after that only if you do:

swapoff /dev/zram0

Only then can you restart the tests again.

I had note seen that with my patch fix, but But Ming noted that he
did see that, and I trust him, although I can't reproduce that issue.

And from at lest a testing perspective then:

Tested-by: Luis Chamberlain <mcgrof@kernel.org>

I'll go and do the line-by-line code review now.

  Luis