Re: [Patch 2 of 2]: scsi-dh + dm-mpath: propagate SCSI device deletion to multipath

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Mike,

I am sorry about the submission problems, there's a first time for everything...

I know this sounds superstitious but I did run quite thorough tests on this issue and I would really prefer that the proposed patch 
would go through the same code path as the one I have tested.

My suggestion is to modify the scsi_dh code as follows:

if (!scsi_dh || !get_device(&sdev->sdev_gendev) ||
    sdev->sdev_state == SDEV_CANCEL ||
    sdev->sdev_state == SDEV_DEL)
       err = SCSI_DH_NOSYS;   

if (sdev->sdev_state == SDEV_OFFLINE)
       err = SCSI_DH_DEV_OFFLINED;

if (err) {
    if (fn)
        fn(fn_data, err);

    return err;
} 

There is logic and benefits here:
1) SDEV_CANCEL/SDEV_DEV are just a stage before the DH is deleted so we can respond with the same SCSI_DH_NOSYS.
   I think it's best here to handle these together, since they are in fact the same thing. 
   In addition, when we delete a device, I don't think it goes through an offline state.
2) SDEV_OFFLINE is easily reverted by echoing to the state of the device - it's not like detaching.
   Returning SCSI_DH_DEV_OFFLINED will just call the default case and perform fail_path - no need to worry about any handler issues here.
3) It will let me sleep better at night...


Best Regards,
Menny

-----Original Message-----
From: Mike Snitzer [mailto:snitzer@xxxxxxxxxx] 
Sent: 15 December, 2010 18:10
To: Hamburger, Menny
Cc: dm-devel@xxxxxxxxxx
Subject: Re: [Patch 2 of 2]: scsi-dh + dm-mpath: propagate SCSI device deletion to multipath

On Wed, Dec 15 2010 at  3:31am -0500,
Menny_Hamburger@xxxxxxxx <Menny_Hamburger@xxxxxxxx> wrote:

> The problem:
> When a SCSI device attached to a device handler is deleted, userland processes currently performing I/O on the device will I/O hang forever.
> 
> Attached is the multipath layer part of the patch.

Please do not use attachments when submitting patches.  Please inline
the patch in the body of the mail.  Also please trim extraneous info
when you reply to mails and please do not top-post.

Your patch submissions were not well received by the dm-devel patchwork
(which automatically collects submitted dm-devel patches so they don't
get lost in the shuffle).

Also, please use unique and descriptive subjects for the patches in a
multipatch series.

As for the proposed mpath change: Babu already pointed out that you
don't need this mpath change as the default case already performs
fail_path().

So all you need is that scsi_dh patch to get upstream.  But to do that
you'll need to:
1) create a patch against the upstream scsi-misc tree (not RHEL5.5):
   git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6.git

2) send the patch to the linux-scsi@xxxxxxxxxxxxxxx mailing list (feel
   free to cc dm-devel too).

Once this change is in the upstream scsi tree it'll propagate to RHEL
releases.

Thanks,
Mike

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel


[Index of Archives]     [DM Crypt]     [Fedora Desktop]     [ATA RAID]     [Fedora Marketing]     [Fedora Packaging]     [Fedora SELinux]     [Yosemite Discussion]     [KDE Users]     [Fedora Docs]

  Powered by Linux