On Thu, 2014-11-06 at 16:33 +0200, Boaz Harrosh wrote: > On 11/06/2014 12:30 PM, James Bottomley wrote: > > On Wed, 2014-11-05 at 11:30 -0800, Christoph Hellwig wrote: > >> On Wed, Nov 05, 2014 at 11:34:11AM -0500, Alan Stern wrote: > >>>> Sorry, meant to. In principle I'm OK with this as the lever for the > >>>> hack (largely because it means we don't need to do anything) but will > >>>> the distributions support it? > >>> > >>> While I can't speak for the distributions, it's reasonable to assume > >>> that if something becomes part of the upstream kernel then they will > >>> include it. > >> > >> Btw, we do have precedence for looking at partition tables from SCSI > >> code with scsi_partsize(), so the layering violation of looking at > >> EFI labels for disks sizes wouldn't be something entirely new even > >> if we did it in kernel space. > > > > We really don't want to make the decision within the kernel of whether > > we believe the partition size or the disk capacity. For these disk > > problems we need it to be the former, but if we choose that always, > > we'll get weird results on mispartitioned devices. > > > > The usual rule is no policy in the kernel and which to choose is policy, > > so just export the knob (as Alan's patch does) and then let userspace > > decide. > > > > I do not think it is a matter of policy. > > From what I understand the situation is that read_capacity_10() which is > 32bit, Max 2T byte disks. fails with 0xffffffff and read_capacity_16() (64bit) > is not supported because of wrong scsi-version or because it is blacklisted. > (or device returns not-supported) Actually, no, RC(10) returns the lowest 32 bits of the actual result. Which is out of spec, but hey, this is a USB bridge ... > Now If sd_read_capacity() succeed then off-course we choose it. Um, the problem is we can't tell ... sd_read_capacity() returns a number, it just may be wrong by a factor of 2TB. > What I'm saying > is if read-capacity fails, then should we attempt a new read_capacity_part()? We can't tell we have a failure. We can tell if the partition capacity and the read capacity differ that one of them must be wrong ... > And sure a user-mode knob if he wants to make policy, like you said, is fine > by me. > > But just the simple case of read-capacity failure should we then? We don't have a failure. This is the problem. Determining that a problem exists 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
