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

[Anoop C S]

On Thu, 2020-05-07 at 23:27 -0700, Jeremy Allison via samba-technical
wrote:
> On Fri, May 08, 2020 at 10:39:38AM +0530, Anoop C S wrote:
> > Your patch fixes mismatch in SHA256 checksum of 100 bin files
> > copied
> > using Windows explorer in my environment.
> > 
> > > If so I'll fix it up to be production-ready,
> > > (for example I think I can get rid of the
> > > immediate event useage) fix the pwrite case
> > > and then cut it onto bite-sized reviewable chunks.
> 
> OK, here's the "production ready" version.
> Gets rid of the crappy immediate event.
> 
> Still doesn't fix the io_uring pwrite issue

Just to mention...

...at least copying good set of files into vfs_io_uring enabled share
using Windows explorer didn't cause any mismatch on final SHA256
checksum.

> but if can confirm that this version also fixes
> the problem (and it's a much cleaner patchset :-)

This patch set also fixes corruption with copying of files from
vfs_io_uring enabled share.

> then I'll add the pwrite fixes next and I'll
> try running in CI.

Cool.

> Then we should have a fix for io_uring in 4.12 !
> 
> Cheers,
> 
> Jeremy.