lkml.org 
[lkml]   [2022]   [Jul]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [EXT] Re: [REGRESSION] qla2xxx: tape drive not removed after unplug FC cable
Hi Thorsten,

On Mon, 4 Jul 2022, 5:06am, Thorsten Leemhuis wrote:

> On 23.06.22 01:03, Arun Easi wrote:
> > On Wed, 22 Jun 2022, 7:56am, Tony Battersby wrote:
> >
> >> On 6/21/22 18:05, Arun Easi wrote:
> >>> Thanks for the info. Just to reiterate, you've reported two issues (though
> >>> this log was showing only 1 of them).
> >>>
> >>> Issue 1 - Tape device never disappears when removed
> >>> Issue 2 - When a direct connected tape 1 was replaced with tape 2, tape 2
> >>> was not discovered.
> >>>
> >>> For Issue-2, please try the attached patch. This may not be the final fix,
> >>> but wanted to check if that would fix the issue for you.
> >>>
> >>> For Issue-1, the behavior was intentional, though that behavior needs
> >>> refinement. These tape drives support something called FC sequence level
> >>> error recovery (added in FCP-2), which can make tape I/Os survive even
> >>> across a short cable pull. This is not a simple retry of the I/O, rather a
> >>> retry done at the FC sequence level that gives the IO a better chance of
> >>> revival. In other words, the said patch that caused regression, while
> >>> introduces an incorrect reporting of the state of the device, makes backup
> >>> more resilient.
> >>>
> >>> Now, onto the behavior when device state is reported immediately. What we
> >>> have observed, at least with one tape drive from a major vendor, is that,
> >>> across a device loss and device back case with both the events reported to
> >>> upper layers, the backup operation was getting failed. This is due to a
> >>> REPORT LUNS command being issued during device reappearance reporting
> >>> (fc_remote_port_add -> SCSI scan), which the tape drive was not expecting
> >>> and caused the backup to fail.
> >>>
> >>> I know that some tape drives do not support multiple commands to it at the
> >>> same time, but not sure if that is still the norm these days.
> >>>
> >>> So, perhaps one way to make the behavior better, is to either report the
> >>> disappearing device a bit delayed or have intelligence added in SCSI scan
> >>> to detect ongoing tape IO operations and delay/avoid the REPORT LUNs.
> >>> Former is a more contained (in the LLD) fix.
> >>>
> >>> Regards,
> >>> -Arun
> >>
> >> Your patch does fix Issue-2 for me.  For Issue-1, it would be fine with
> >> me if qla2xxx reported device removal to the upper level a bit delayed,
> >> as you said.
> >>
> >
> > Thanks for testing and verifying the patch.
>
> BTW, that patch should have 'Link:' tags pointing to all reports about
> this issue, e.g. the start of this thread.

Thanks, will add, I was not aware of this.

>
> These tags are important, as they allow others to look into the
> backstory now and years from now. That is why they should be placed in
> cases like this, as Documentation/process/submitting-patches.rst and
> Documentation/process/5.Posting.rst explain in more detail.
> Additionally, my regression tracking bot ‘regzbot’ relies on these tags
> to automatically connect reports with patches that are posted or
> committed to fix the reported issue. BTW, let me tell regzbot to monitor
> this thread:
>
> > We will post the patch upstream after due testing.
> That was more than two weeks ago now and I didn't see any progress. Or
> did I miss it?

No, the fix is being prepared to be posted.

Regards,
-Arun
\
 
 \ /
  Last update: 2022-07-06 19:30    [W:0.136 / U:0.012 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site