On Wed, 2015-10-07 at 22:23 +0200, Hannes Reinecke wrote:
> On 10/06/2015 07:17 PM, James Bottomley wrote:
> > On Tue, 2015-10-06 at 10:14 -0700, Lee Duncan wrote:
> >> On 10/06/2015 02:40 AM, Christoph Hellwig wrote:
> >>>> struct Scsi_Host *scsi_host_lookup(unsigned short hostnum)
> >>>> {
> >>>> - struct device *cdev;
> >>>> - struct Scsi_Host *shost = NULL;
> >>>> -
> >>>> - cdev = class_find_device(&shost_class, NULL, &hostnum,
> >>>> - __scsi_host_match);
> >>>> - if (cdev) {
> >>>> - shost = scsi_host_get(class_to_shost(cdev));
> >>>> - put_device(cdev);
> >>>> - }
> >>>> + struct Scsi_Host *shost;
> >>>> +
> >>>> + spin_lock(&host_index_lock);
> >>>> + shost = idr_find(&host_index_idr, hostnum);
> >>>> + spin_unlock(&host_index_lock);
> >>>> +
> >>>> return shost;
> >>>
> >>> How does this actually grab a reference to the host?
> >>
> >> Good catch -- I should have noticed that.
> >>
> >> I will resubmit the patch.
> >
> > I'll wait to see what you produce, but I don't think, using a separate
> > idr array, you can close the race window between lookup and get. One of
> > the nice things about using the cdev iterator is that the get is part of
> > the lookup process.
> >
> Hmm. Should be possible if you free up the IDR in scsi_remove_host(),
> just after setting the state to SHOST_DEL.
Then all lookups fail between _del and _release which is different
behaviour from current. In theory that's a very short window because we
should have removed all the devices synchronously in _del, so I don't
think there's any adverse consequences but it's something that would
need investigating.
To be honest, with the host number patch, I'm starting to come to the
conclusion that if it isn't broken, don't fix it because of the risks.
What are we trying to fix, anyway? The host number wrapping at 32 bits?
James
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
