Re: [RFC PATCH v11.1 2/2] mm, pmem, xfs: Introduce MF_MEM_REMOVE for unbind

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Apr 25, 2023 at 03:23:15PM +0200, Jan Kara wrote:
> On Tue 25-04-23 20:47:35, Shiyang Ruan wrote:
> > 
> > 
> > 在 2023/4/20 20:09, Jan Kara 写道:
> > > On Thu 20-04-23 10:07:39, Shiyang Ruan wrote:
> > > > 在 2023/4/12 18:52, Shiyang Ruan 写道:
> > > > > This is a RFC HOTFIX.
> > > > > 
> > > > > This hotfix adds a exclusive forzen state to make sure any others won't
> > > > > thaw the fs during xfs_dax_notify_failure():
> > > > > 
> > > > >     #define SB_FREEZE_EXCLUSIVE	(SB_FREEZE_COMPLETE + 2)
> > > > > Using +2 here is because Darrick's patch[0] is using +1.  So, should we
> > > > > make these definitions global?
> > > > > 
> > > > > Another thing I can't make up my mind is: when another freezer has freeze
> > > > > the fs, should we wait unitl it finish, or print a warning in dmesg and
> > > > > return -EBUSY?
> > > > > 
> > > > > Since there are at least 2 places needs exclusive forzen state, I think
> > > > > we can refactor helper functions of freeze/thaw for them.  e.g.
> > > > >     int freeze_super_exclusive(struct super_block *sb, int frozen);
> > > > >     int thaw_super_exclusive(struct super_block *sb, int frozen);
> > > > > 
> > > > > [0] https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/commit/?h=repair-fscounters&id=c3a0d1de4d54ffb565dbc7092dfe1fb851940669
> > > 
> > > I'm OK with the idea of new freeze state that does not allow userspace to
> > > thaw the filesystem. But I don't really like the guts of filesystem
> > > freezing being replicated inside XFS. It is bad enough that they are
> > > replicated in [0], replicating them *once more* in another XFS file shows
> > > we are definitely doing something wrong. And Luis will need yet another
> > > incantation of the exlusive freeze for suspend-to-disk. So please guys get
> > > together and reorganize the generic freezing code so that it supports
> > > exclusive freeze (for in-kernel users) and works for your usecases instead
> > > of replicating it inside XFS...
> > 
> > I agree that too much replicating code is not good.  It's necessary to
> > create a generic exclusive freeze/thaw for all users.  But for me, I don't
> > have the confidence to do it well, because it requires good design and code
> > changes will involve other filesystems.  It's diffcult.
> > 
> > However, I hope to be able to make progress on this unbind feature. Thus, I
> > tend to refactor a common helper function for xfs first, and update the code
> > later when the generic freeze is done.
> 
> I think Darrick was thinking about working on a proper generic interface.
> So please coordinate with him.

I'll post a vfs generic kernelfreeze series later today.

One thing I haven't figured out yet is what's supposed to happen when
PREREMOVE is called on a frozen filesystem.  We don't want userspace to
be able to thaw the fs while PREREMOVE is running, so I /guess/ that
means we need some method for the kernel to take over a userspace
freeze and then put it back when we're done?

--D

> 								Honza
> 
> -- 
> Jan Kara <jack@xxxxxxxx>
> SUSE Labs, CR



[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [NTFS 3]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [NTFS 3]     [Samba]     [Device Mapper]     [CEPH Development]

  Powered by Linux