Re: [PATCH] firmware_loader: Fix use-after-free Read in firmware_loading_store

[Anirudh Rayabharam]

On Thu, Jul 15, 2021 at 04:21:05PM -0700, Luis Chamberlain wrote:
> On Thu, Jul 15, 2021 at 04:46:24PM -0600, Shuah Khan wrote:
> > On 7/15/21 4:28 PM, Luis Chamberlain wrote:
> > > On Fri, Jul 09, 2021 at 10:38:12AM -0600, Shuah Khan wrote:
> > > > However I am seeing the following over and over again in the
> > > > log - hence I think it is safer to check the aborted status
> > > > in __fw_load_abort().
> > > > 
> > > > ? __list_del_entry_valid+0xe0/0xf0
> > > > [  348.604808][T12994]  __list_del_entry_valid+0xe0/0xf0
> > > > [  348.610020][T12994]  firmware_loading_store+0x141/0x650
> > > > [  348.615761][T12994]  ? firmware_data_write+0x4e0/0x4e0
> > > > [  348.621064][T12994]  ? sysfs_file_ops+0x1c0/0x1c0
> > > > [  348.625921][T12994]  dev_attr_store+0x50/0x80
> > > > 
> > > > Also the fallback logic takes actions based on errors as in
> > > > fw_load_sysfs_fallback() that returns -EAGAIN which would
> > > > trigger request_firmware() again.
> > > > 
> > > > Based on all of this I think this fix is needed, if only I can
> > > > test for sure.
> > > 
> > > Shuah, curious if you had read this patch from Anirudh Rayabharam
> > > and my response to that v4 patch iteration?
> > > 
> > > https://lkml.kernel.org/r/20210518155921.4181-1-mail@xxxxxxxxxxxxx
> > > 
> > 
> > Yes. I realized I am trying to fix the same problem we have been
> > discussing. :) Sorry for the noise.
> 
> No worries, and thanks again for you help!
> 
> > Ignore my patch. I will follow the thread.
> 
> OK ! I think all we need is just Anirudh to split his patch to
> remove the -EAGAIN return value in a separate patch as a first step,
> documenting in the commmit log that:
> 
> The only motivation on her part with using -EAGAIN on commit
> 0542ad88fbdd81bb ("firmware loader: Fix _request_firmware_load()
> return val for fw load abort") was to distinguish the error from
> -ENOMEM, and so there is no real reason in keeping it. Keeping

Are you sure about this? As per Shuah's explanation [1], it sounds to
me like certain media drivers explicitly check for -EAGAIN to retry
the firmware load. Shuah, is my understanding correct?

[1]: https://lore.kernel.org/lkml/04b5bb2f-edf7-5b43-585a-3267d83bd8c3@xxxxxxxxxxxxxxxxxxx/

> -ETIMEDOUT is much telling of what the reason for a failure is,
> so just use that.
> 
> Then his second patch would be simplified without the -EAGAIN
> condition.
> 
> All I asked was to confirm that the -ETIMEDOUT was indeed propagated.

Yes, -ETIMEDOUT is indeed propagated by fw_sysfs_wait_timeout.

> 
> Anirudh, sorry for the trouble, but can I ask you for a v5 with two
> patches as described above?

Sure, I will do that. 

	- Anirudh.

> 
>   Luis