Re: [syzbot] [kernfs?] possible deadlock in kernfs_fop_llseek

[Christian Brauner]

On Fri, Apr 05, 2024 at 08:51:35AM +0200, Christoph Hellwig wrote:
> On Thu, Apr 04, 2024 at 12:33:40PM +0300, Amir Goldstein wrote:
> > I don't follow what you are saying.
> > Which code is in non-starter violation?
> > kernfs for calling lookup_bdev() with internal of->mutex held?
> 
> That is a huge problem, and has been causing endless annoying lockdep
> chains in the block layer for us.  If we have some way to kill this
> the whole block layer would benefit.

Why not just try and add a better resume api that forces resume to not
use a path argument neither for resume_file nor for writes to
/sys/power/resume. IOW, extend the protocol what can get written to
/sys/power/resume and then phase out the path argument. It'll take a
while but it's a possibly clean solution.