Re: [PATCH 3/4] pipe: mark blocking pipe read and FIFO open sleeps as freezable

[daijunbing]

Hi,

Please drop this patch from vfs-7.2.misc.

After further review, I believe this change is not sufficiently justified.
In particular, I do not have a solid enough analysis to show that making
the anon_pipe_read() wait freezable is safe with respect to freeze-time
lock/resource constraints across all relevant call paths.

While this change has been running on our devices for about a year without
any reported issues, I am not confident that this is sufficient to 
justify it for upstream use in the general case.

This concern only affects this patch; the rest of the series should be
unaffected.

Thanks,


在 2026/5/30 18:10, Jan Kara 写道:
> Making sure we don't hold any lock is rather non-trivial in some cases. For
> example in your case here, anon_pipe_read() is used in .read_iter method so
> you should make sure all places calling .read_iter don't hold some lock. If
> nothing else, I'm missing this analysis in the changelog and I actually
> strongly suspect it is not true in some cor