On Wed, Apr 21, 2021 at 02:33:46PM +0200, Martin Wilck wrote:
> On Wed, 2021-04-21 at 18:40 +0800, Ming Lei wrote:
> > On Wed, Apr 21, 2021 at 12:19:00PM +0200, Martin Wilck wrote:
> > > On Wed, 2021-04-21 at 15:55 +0800, Ming Lei wrote:
> > > > Hello Guys,
> > > >
> > > > fnic uses the following way to walk scsi commands in failure
> > > > handling,
> > > > which is obvious wrong, because caller of scsi_host_find_tag has
> > > > to
> > > > guarantee that the tag is active.
> > > >
> > > > for (tag = 0; tag < fnic->fnic_max_tag_id; tag++) {
> > > > ...
> > > > sc = scsi_host_find_tag(fnic->lport->host, tag);
> > > > ...
> > > > }
> > > >
> > > > Fix the issue by using blk_mq_tagset_busy_iter() to walk
> > > > request/scsi_command.
> > >
> > > How does this relate to Hannes' previous patch?
> > > https://marc.info/?l=linux-scsi&m=161400059528859&w=2
> >
> > oops, this patch is actually same or similar with Hannes's.
> >
> > Given these patches are bug fix, can we cherry-pick them for 5.13?
>
> No objections in principle, but the differences between your patch and
> Hannes' are pretty large. I couldn't tell which one is more
> appropriate.
>
> Question: Both your patch set and Hannes' patch replace a couple of
> scsi_host_find_tag() calls in the fnic driver, while leaving some
> others in place. It's not clear to me why we can be sure that no
> corruption occurs in any of the latter. Could you explain?
Caller of scsi_host_find_tag() has to guarantee that the passed tag
is active.
The changed functions do pass invalid tag to scsi_host_find_tag(), so
we need to fix them.
Thanks,
Ming
