Re: [REGRESSION] fuse: copy_file_range() fails with EIO

[Miklos Szeredi <miklos@szeredi.hu>]

On Sat, 10 Aug 2024 at 17:12, Matthew Wilcox <willy@infradead.org> wrote:
>
> On Sat, Aug 10, 2024 at 07:56:21AM +0200, Jürg Billeter wrote:
> > Thanks for the quick response.
>
> Thanks for the quick test!
>
> > On Fri, 2024-08-09 at 17:03 +0100, Matthew Wilcox wrote:
> > > Do you have CONFIG_DEBUG_VM enabled?  There are some debugging asserts
> > > which that will enable that might indicate a problem.
> >
> > With CONFIG_DEBUG_VM enabled, I get:
> >
> > page: refcount:2 mapcount:0 mapping:00000000b2c30835 index:0x0 pfn:0x12a113
> > memcg:ffff9d8e3a660800
> > aops:0xffffffff8a056820 ino:21 dentry name:"bash"
> > flags: 0x24000000000022d(locked|referenced|uptodate|lru|workingset|node=0|zone=2)
> > raw: 024000000000022d ffffd9ce04a827c8 ffffd9ce04a84508 ffff9d8e0bbc99f0
> > raw: 0000000000000000 0000000000000000 00000002ffffffff ffff9d8e3a660800
> > page dumped because: VM_BUG_ON_FOLIO(folio_test_uptodate(folio))
>
> That's what I suspected was going wrong -- we're trying to end a read on
> a folio that is already uptodate.  Miklos, what the hell is FUSE doing
> here?

Ah, this is the fancy page cache replacement done in fuse_try_move_page().

I understand how this triggers VM_BUG_ON_FOLIO() in folio_end_read().

What I don't understand is how this results in the -EIO that Jürg reported.

Thanks,
Miklos