Re: [RFC PATCH] mm: truncate: flush lru cache for evicted inode

[Ryusuke Konishi <konishi.ryusuke@gmail.com>]

On Mon, Jun 17, 2024 at 4:57 PM Jan Kara wrote:
>
> On Sun 16-06-24 10:39:51, Hillf Danton wrote:
> > On Sat, 15 Jun 2024 21:44:54 +0100 Matthew Wilcox wrote:
> > >
> > > I suspect this would trigger:
> > >
> > > +++ b/fs/inode.c
> > > @@ -282,6 +282,7 @@ static struct inode *alloc_inode(struct super_block *sb)
> > >  void __destroy_inode(struct inode *inode)
> > >  {
> > >         BUG_ON(inode_has_buffers(inode));
> > > +       BUG_ON(inode->i_data.nrpages);
> > >         inode_detach_wb(inode);
> > >         security_inode_free(inode);
> > >         fsnotify_inode_delete(inode);
> > >
> > Yes, it was triggered [1]
> >
> > [1] https://lore.kernel.org/lkml/00000000000084b401061af6ab80@google.com/
> >
> > and given trigger after nrpages is checked in clear_inode(),
> >
> >       iput(inode)
> >       evict(inode)
> >       truncate_inode_pages_final(&inode->i_data);
> >       clear_inode(inode);
> >       destroy_inode(inode);
> >
> > why is folio added to exiting mapping?
> >
> > #syz test https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git  83a7eefedc9b
>
> OK, so based on syzbot results this seems to be a bug in
> nilfs_evict_inode() (likely caused by corrupted filesystem so that root
> inode's link count was 0 and hence was getting deleted on iput()). I guess
> nilfs maintainers need to address these with more consistency checks of
> metadata when loading them...
>
>                                                                         Honza
> --
> Jan Kara <jack@suse.com>
> SUSE Labs, CR

Sorry for my late response.

Also, thank you for pointing out that the problem seems to be caused
via nilfs_evict_inode() by a missing consistency check of the link
count.

I'll check it out and think about how to deal with it.

Thanks,
Ryusuke Konishi