Data Corruption bug with Samba's vfs_iouring and Linux 5.6.7/5.7rc3

[Jeremy Allison]

On Tue, May 12, 2020 at 12:50:02PM +0530, Anoop C S wrote:
> On Mon, 2020-05-11 at 21:54 +0200, Stefan Metzmacher via samba-
> technical wrote:
> > Am 09.05.20 um 04:50 schrieb Jeremy Allison via samba-technical:
> > > On Sat, May 09, 2020 at 12:04:55AM +0200, Stefan Metzmacher via
> > > samba-technical wrote:
> > > > Am 08.05.20 um 23:50 schrieb Jeremy Allison:
> > > > > RB+ from me if you add a comment header above the function
> > > > > as well as in the commit so it explains what it's doing.
> > > > > 
> > > > > Feel free to crib my ascii art to explain in the header
> > > > > comment too :-).
> > > > > 
> > > > > Thanks for the cleanup !
> > > > 
> > > > I'll do that on Monday.
> > > 
> > > Here's an updated version with added
> > > comments inside
> > > 
> > > [PATCH 20/29] vfs_io_uring: avoid stack recursion of
> > > vfs_io_uring_queue_run()
> > > 
> > > so I could actually (hopefully) understand
> > > it if I have to go back and look at it in 5 years time :-).
> > 
> > Thanks! I've done some experiments with offset < 0 and it's not
> > allowed
> > for SMB2 read nor write.
> > 
> > I'll post the updated patchset to the merge request shortly.
> > 
> > It would be great if someone could verify it with a 5.6 kernel.
> 
> Latest patch set[1] could successfully copy from and to vfs_io_uring
> enabled ext4 share using Windows explorer against Samba server with
> kernel v5.6.10.	SHA256 checksum of copied files matches original value.
> 
> [1] https://attachments.samba.org/attachment.cgi?id=15971

Thanks for the confirmation Anoop. I've pushed Metze's latest
patchset to master, and we can back-port to 4.12.next as soon
as it lands.

I really appreciate your help in reproducing the problem on
the latest kernels !

Jeremy.