On Mon 12-06-23 11:36:57, Darrick J. Wong wrote: > On Mon, Jun 12, 2023 at 01:35:26PM +0200, Jan Kara wrote: > > What we could be doing to limit unnecessary waiting is that we'd update > > freeze_holders already when we enter freeze_super() and lock s_umount > > (bailing if our holder type is already set). That way we'd have at most one > > process for each holder type freezing the fs / waiting for freezing to > > complete. > > <shrug> I don't know how often we even really have threads contending > for s_umount and elevated freeze state. How about we go with the > simpler wait_for_partially_frozen and see if complaints crop up? Yeah, I'm for the simpler approach as well. This was more a suggestion if you think that is not viable. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
