Re: [PATCH 0/6] 9pfs: fix fstat() after unlink() (with a Linux guest)

[Greg Kurz]

On Mon, 25 Nov 2024 11:23:39 +0100
Christian Schoenebeck <qemu_oss@crudebyte.com> wrote:

> On Monday, November 25, 2024 9:45:54 AM CET Greg Kurz wrote:
> > On Sun, 24 Nov 2024 17:28:40 +0100
> > Christian Schoenebeck <qemu_oss@crudebyte.com> wrote:
> > 
> > > This fixes an infamous, long standing bug:
> > > https://gitlab.com/qemu-project/qemu/-/issues/103
> > > 
> > 
> > \o/
> > 
> > It is great if you manage to fix that once and far all !
> > 
> > > * Actual fix of this bug is patch 5.
> > > 
> > > * Patches 1 and 6 add a test case to verify the expected behaviour.
> > > 
> > > * The other patches (2, 3, 4) are basically just minor cleanup patches 
> > > more
> > >   or less (un)related that I simply did not bother to send separately.
> > > 
> > > Probably there are still other 9p request types that should be fixed for 
> > > this
> > > use-after-unlink idiom, but this series fixes the mentioned bug report as
> > > described by reporter, so fair enough to round this up here for now.
> > > 
> > 
> > When I last worked on that issue I had spotted some other places to fix.
> > 
> > Maybe you can find some ideas for future work at :
> > 
> > https://github.com/gkurz/qemu/tree/9p-attr-fixes
> 
> Was there a reason why you left those patches on the attic?
> 

Lack of cycles

> What I am seeing is that it was not fixing Tgetattr (i.e. fstat() on guest),
> so it wouldn't have fixed the original reporter's scenario, but they would
> have brought things forward. So just wondering ...
> 

Yeah the fix for Tgetattr was in some other series I had sent at the time but
I did not get much feeback then...

> /Christian
> 
> 

Cheers,

-- 
Greg