[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [RFC PATCH] block: Fix eject -f for locked devices
From: |
Markus Armbruster |
Subject: |
Re: [Qemu-devel] [RFC PATCH] block: Fix eject -f for locked devices |
Date: |
Wed, 23 Feb 2011 15:11:22 +0100 |
User-agent: |
Gnus/5.13 (Gnus v5.13) Emacs/23.1 (gnu/linux) |
Sorry for my slow reply.
Blue Swirl <address@hidden> writes:
> On Fri, Feb 18, 2011 at 5:16 PM, Markus Armbruster <address@hidden> wrote:
>> From 8cd4978c9be6ff2bcc414bb1c1b258b96b9a74c1 Mon Sep 17 00:00:00 2001
>> From: Markus Armbruster <address@hidden>
>> Date: Fri, 18 Feb 2011 15:54:02 +0100
>>
>> After forcefully ejecting media locked by the guest, you can't ever
>> again insert new media.
>>
>> Example:
>>
>> (qemu) info block
>> hda: type=hd removable=0 file=test.img ro=0 drv=raw encrypted=0
>> cd: type=cdrom removable=1 locked=1 file=x.iso ro=0 drv=raw encrypted=0
>> (qemu) eject cd
>> Device 'cd' is locked
>> (qemu) eject -f cd
>> (qemu) info block
>> hda: type=hd removable=0 file=test.img ro=0 drv=raw encrypted=0
>> cd: type=cdrom removable=1 locked=1 [not inserted]
>> (qemu) change cd x.iso
>> Device 'cd' is locked
>>
>> Signed-off-by: Markus Armbruster <address@hidden>
>> ---
>> I'm not entirely sure this is the appropriate fix, and that's why
>> there's RFC in the subject.
>>
>> Both IDE and SCSI devices expose their drive's BlockDriverState member
>> locked to the guest via mode sense.
>>
>> What does real hardware do when I force-eject media (typically by
>> rummaging in that little hole with a paperclip)? Does it actively
>> notify the OS? Does mode sense change?
>
> No idea, but IIRC the drive is still usable after that,
That's what I'd expect. If the OS recovers from losing the media, the
drive should be fine.
> so locking the
> drive does not look correct.
I'm not sure I get you here.
>> A possible alternative fix is to make do_change_block() ignore
>> bdrv_is_locked() when inserting media into an empty drive.
>
> Then the meaning of locked would change, maybe eject_disabled would
> then describe the state better.
I like the current meaning of BlockDriverState member locked just fine,
it nicely mirrors the SCSI / ATAPI mode sense page bit.
I'm worried about unwanted side effects of my change. For instance,
what about this code in do_snapshot_blkdev():
flags = bs->open_flags;
bdrv_close(bs);
ret = bdrv_open(bs, filename, flags, drv);
I'm afraid my change would make it screw up bs->locked. Jes?