Re: Stable backports for qla2xxx target mode

[Greg KH <gregkh@linuxfoundation.org>]

On Thu, Aug 13, 2015 at 05:55:48PM -0700, Nicholas A. Bellinger wrote:
> On Thu, 2015-08-13 at 17:49 -0700, Greg KH wrote:
> > On Thu, Aug 13, 2015 at 02:00:45PM -0700, Nicholas A. Bellinger wrote:
> > > Hi Greg-KH,
> > > 
> > > This is a follow up on a handful of 'WTF' emails for some of the recent
> > > qla2xxx target patches from v4.2-rc5 CC'ed for v3.18.y stable.
> > > 
> > >   a6ca8878 qla2xxx: delay plogi/prli ack until existing sessions are deleted
> > >   daddf5cf qla2xxx: Abort stale cmds on qla_tgt_wq when plogi arrive
> > >   df673274 qla2xxx: added sess generations to detect RSCN update races
> > >   d20ed91b qla2xxx: disable scsi_transport_fc registration in target mode
> > >   7359df25 qla2xxx: terminate exchange when command is aborted by LIO
> > >   8b2f5ff3 qla2xxx: cleanup cmd in qla workqueue before processing TMR
> > >   e52a8b45 qla2xxx: drop cmds/tmrs arrived while session is being deleted
> > > 
> > > These are all bug-fix patches that address real-world correctness issues
> > > reported by a large customer of the qla2xxx target code, and have been
> > > reviewed + tested + signed-off-by the HW LLD maintainer.
> > 
> > And they are all fricking huge rewrites and additions to the driver,
> > none of which actually look like they should be added to a stable tree
> > at all.
> 
> Each of which addresses bugs that the largest consumer of the code (Pure
> Storage) has encountered in production.
> 
> > 
> > > Granted these patches are larger in size that I'd normally be
> > > comfortable with CC'ing for stable, but they do address real-world
> > > correctness issues seen in large scale production with qla2xxx target
> > > code.
> > > 
> > > Please consider adding these to your stable queue, to go along with the
> > > qla2xxx target patches from the same series that have already made it
> > > into v4.1.5.
> > 
> > What exactly are they fixing?  They look like they add a ton of new
> > functions to the driver, and other features.  What is so broken in the
> > driver today that warrants this type of exception to the rules?
> > 
> 
> The commit logs go into details about the exact issues, but my
> understanding is that it's a combination of OOPsen, potential
> data-corruption due to duplicate sessions, and couple different
> resource leaks.
> 
> Pure Storage and Qlogic folks, would you be so kind as to comment more
> on the specifics for Greg-KH to consider..?

I'd like the scsi maintainer to also approve such an exception, I notice
they didn't even sign off on these patches :(

greg k-h