Re: 6.17rc5: btrfs scrub, Freezing user space processes failed

[Qu Wenruo <wqu@suse.com>]

在 2025/9/16 08:46, Chris Murphy 写道:
> The storage stack may be relevant: USB flash drive -> dm-crypt -> Btrfs

And for your original report of no response, the problem is since btrfs 
is blocking the suspension, there should be no extra reason why the 
system hangs.

Unless there are other corner cases like the USB device is powered off.

If you can reproduce the bug, please catch the dying message using 
something like netconsole.

> 
> Darrick Wong notes that in https://elixir.bootlin.com/linux/v6.15/source/fs/btrfs/ioctl.c#L3159
> 
> btrfs_ioctl_scrub calls mnt_want_write_file for the duration of the scrub, and mnt_want_write_file takes SB_FREEZE_WRITE and holds that all the way to the end,  which means you can't fsfreeze the filesystem

Yes, that's already a known problem and both David and I were working on 
this in the past:

https://lore.kernel.org/linux-btrfs/9606fae20bff6c1fbe14dc7b067f3b333c2a955b.1751847905.git.wqu@suse.com/

https://lore.kernel.org/linux-btrfs/20250708132540.28285-1-dsterba@suse.com/


My solution is to cancel scrub which is the simplest solution.
David's solution is pause scrub/balance using extra callbacks and a more 
complex mechanism.

Thanks,
Qu

> 
> So how did this ever work? Folks do use btrfsmaintenance with scrub and trim timers, and a laptop can sleep at any time. We can't inhibit this indefinitely.
> 
> Perhaps scrub and balance can be paused if pm suspend/hibernate is requested? Just make it a non-factor.
> 
> Chris Murphy
>