Re: [Qemu-block] [PATCH v7 12/24] virtio-blk: Functions for op blocker management

[Kevin Wolf]

Am 26.11.2015 um 08:48 hat Stefan Hajnoczi geschrieben:
> On Wed, Nov 25, 2015 at 05:26:02PM +0100, Max Reitz wrote:
> > On 25.11.2015 17:18, Kevin Wolf wrote:
> > > Am 25.11.2015 um 17:03 hat Max Reitz geschrieben:
> > >> On 25.11.2015 16:57, Kevin Wolf wrote:
> > >>> Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
> > >>> This makes me wonder: What do we even block here any more? If I didn't
> > >>> miss anything, it's only BLOCK_OP_TYPE_BACKUP_TARGET, and I'm not sure
> > >>> why this needs to be blocked, or if we simply forgot to enable it.
> > >>
> > >> Well, even though in practice this wall of code doesn't make much sense,
> > >> of course it will be safe for potential additions of new op blockers.
> > >>
> > >> And of course we actually don't want these blockers at all anymore...
> > > 
> > > Yes, but dataplane shouldn't really be special enough any more that we
> > > want to disable features for it initially. By now it sounds more like an
> > > easy way to forget unblocking a new feature even though it would work.
> > > 
> > > So perhaps we should really just remove the blockers from dataplane.
> > > Then we don't have to answer the question above...
> > 
> > Well, maybe. I guess this is up to Stefan.
> 
> At this point blockdev.c and block jobs acquire/release AioContext,
> hence all these op blockers are being unblocked.  I think we can switch
> from whitelisting (unblocking) nearly everything to blacklisting
> (blocking) only things that aren't supported yet.

As I said, there is only one operation that isn't whitelisted today,
BLOCK_OP_TYPE_BACKUP_TARGET, and I would be surprised if it weren't just
a bug that it's not whitelisted yet.

Kevin


pgpNh__iIJ8o5.pgp
Description: PGP signature

Re: [Qemu-block] [PATCH v7 12/24] virtio-blk: Functions for op blocker management

[Max Reitz]

On 25.11.2015 16:57, Kevin Wolf wrote:
> Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
>> Put the code for setting up and removing op blockers into an own
>> function, respectively. Then, we can invoke those functions whenever a
>> BDS is removed from an virtio-blk BB or inserted into it.
>>
>> Signed-off-by: Max Reitz 
> 
> Do you know of a case where this is observable?

Actually, no.

> I don't think you can
> change the medium of a virtio-blk device, and blk_set_bs() isn't
> converted to a wrapper around blk_remove/insert_bs() yet. If this patch
> is necessary to fix something observable, the latter is probably a bug.

So I guess that implies "Otherwise, this patch should be dropped"?

>> diff --git a/hw/block/dataplane/virtio-blk.c 
>> b/hw/block/dataplane/virtio-blk.c
>> index c42ddeb..4c95d5b 100644
>> --- a/hw/block/dataplane/virtio-blk.c
>> +++ b/hw/block/dataplane/virtio-blk.c
>> @@ -39,6 +39,8 @@ struct VirtIOBlockDataPlane {
>>  EventNotifier *guest_notifier;  /* irq */
>>  QEMUBH *bh; /* bh for guest notification */
>>  
>> +Notifier insert_notifier, remove_notifier;
>> +
>>  /* Note that these EventNotifiers are assigned by value.  This is
>>   * fine as long as you do not call event_notifier_cleanup on them
>>   * (because you don't own the file descriptor or handle; you just
>> @@ -137,6 +139,54 @@ static void handle_notify(EventNotifier *e)
>>  blk_io_unplug(s->conf->conf.blk);
>>  }
>>  
>> +static void data_plane_set_up_op_blockers(VirtIOBlockDataPlane *s)
>> +{
>> +assert(!s->blocker);
>> +error_setg(>blocker, "block device is in use by data plane");
>> +blk_op_block_all(s->conf->conf.blk, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_RESIZE, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_DRIVE_DEL, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_BACKUP_SOURCE, 
>> s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_CHANGE, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_SOURCE, 
>> s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_TARGET, 
>> s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EJECT, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
>> +   s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT,
>> +   s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, 
>> BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE,
>> +   s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_MIRROR, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_STREAM, s->blocker);
>> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_REPLACE, s->blocker);
>> +}
> 
> This makes me wonder: What do we even block here any more? If I didn't
> miss anything, it's only BLOCK_OP_TYPE_BACKUP_TARGET, and I'm not sure
> why this needs to be blocked, or if we simply forgot to enable it.

Well, even though in practice this wall of code doesn't make much sense,
of course it will be safe for potential additions of new op blockers.

And of course we actually don't want these blockers at all anymore...

Max



signature.asc
Description: OpenPGP digital signature

Re: [Qemu-block] [PATCH v7 12/24] virtio-blk: Functions for op blocker management

[Kevin Wolf]

Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
> Put the code for setting up and removing op blockers into an own
> function, respectively. Then, we can invoke those functions whenever a
> BDS is removed from an virtio-blk BB or inserted into it.
> 
> Signed-off-by: Max Reitz 

Do you know of a case where this is observable? I don't think you can
change the medium of a virtio-blk device, and blk_set_bs() isn't
converted to a wrapper around blk_remove/insert_bs() yet. If this patch
is necessary to fix something observable, the latter is probably a bug.

> diff --git a/hw/block/dataplane/virtio-blk.c b/hw/block/dataplane/virtio-blk.c
> index c42ddeb..4c95d5b 100644
> --- a/hw/block/dataplane/virtio-blk.c
> +++ b/hw/block/dataplane/virtio-blk.c
> @@ -39,6 +39,8 @@ struct VirtIOBlockDataPlane {
>  EventNotifier *guest_notifier;  /* irq */
>  QEMUBH *bh; /* bh for guest notification */
>  
> +Notifier insert_notifier, remove_notifier;
> +
>  /* Note that these EventNotifiers are assigned by value.  This is
>   * fine as long as you do not call event_notifier_cleanup on them
>   * (because you don't own the file descriptor or handle; you just
> @@ -137,6 +139,54 @@ static void handle_notify(EventNotifier *e)
>  blk_io_unplug(s->conf->conf.blk);
>  }
>  
> +static void data_plane_set_up_op_blockers(VirtIOBlockDataPlane *s)
> +{
> +assert(!s->blocker);
> +error_setg(>blocker, "block device is in use by data plane");
> +blk_op_block_all(s->conf->conf.blk, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_RESIZE, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_DRIVE_DEL, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_BACKUP_SOURCE, 
> s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_CHANGE, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_SOURCE, 
> s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_TARGET, 
> s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EJECT, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
> +   s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT,
> +   s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE,
> +   s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_MIRROR, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_STREAM, s->blocker);
> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_REPLACE, s->blocker);
> +}

This makes me wonder: What do we even block here any more? If I didn't
miss anything, it's only BLOCK_OP_TYPE_BACKUP_TARGET, and I'm not sure
why this needs to be blocked, or if we simply forgot to enable it.

Kevin

Re: [Qemu-block] [PATCH v7 12/24] virtio-blk: Functions for op blocker management

[Kevin Wolf]

Am 25.11.2015 um 17:03 hat Max Reitz geschrieben:
> On 25.11.2015 16:57, Kevin Wolf wrote:
> > Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
> >> Put the code for setting up and removing op blockers into an own
> >> function, respectively. Then, we can invoke those functions whenever a
> >> BDS is removed from an virtio-blk BB or inserted into it.
> >>
> >> Signed-off-by: Max Reitz 
> > 
> > Do you know of a case where this is observable?
> 
> Actually, no.
> 
> > I don't think you can
> > change the medium of a virtio-blk device, and blk_set_bs() isn't
> > converted to a wrapper around blk_remove/insert_bs() yet. If this patch
> > is necessary to fix something observable, the latter is probably a bug.
> 
> So I guess that implies "Otherwise, this patch should be dropped"?

I'm not sure. I guess you had a reason to include these patches other
than putting the notifiers to use?

With blk_set_bs() changed, I think it would have an effect: The op
blockers would move from the old BDS to the new top-level one. This
sounds desirable to me.

> >> diff --git a/hw/block/dataplane/virtio-blk.c 
> >> b/hw/block/dataplane/virtio-blk.c
> >> index c42ddeb..4c95d5b 100644
> >> --- a/hw/block/dataplane/virtio-blk.c
> >> +++ b/hw/block/dataplane/virtio-blk.c
> >> @@ -39,6 +39,8 @@ struct VirtIOBlockDataPlane {
> >>  EventNotifier *guest_notifier;  /* irq */
> >>  QEMUBH *bh; /* bh for guest notification */
> >>  
> >> +Notifier insert_notifier, remove_notifier;
> >> +
> >>  /* Note that these EventNotifiers are assigned by value.  This is
> >>   * fine as long as you do not call event_notifier_cleanup on them
> >>   * (because you don't own the file descriptor or handle; you just
> >> @@ -137,6 +139,54 @@ static void handle_notify(EventNotifier *e)
> >>  blk_io_unplug(s->conf->conf.blk);
> >>  }
> >>  
> >> +static void data_plane_set_up_op_blockers(VirtIOBlockDataPlane *s)
> >> +{
> >> +assert(!s->blocker);
> >> +error_setg(>blocker, "block device is in use by data plane");
> >> +blk_op_block_all(s->conf->conf.blk, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_RESIZE, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_DRIVE_DEL, 
> >> s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_BACKUP_SOURCE, 
> >> s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_CHANGE, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_SOURCE, 
> >> s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_COMMIT_TARGET, 
> >> s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EJECT, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
> >> +   s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT,
> >> +   s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, 
> >> BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE,
> >> +   s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_MIRROR, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_STREAM, s->blocker);
> >> +blk_op_unblock(s->conf->conf.blk, BLOCK_OP_TYPE_REPLACE, s->blocker);
> >> +}
> > 
> > This makes me wonder: What do we even block here any more? If I didn't
> > miss anything, it's only BLOCK_OP_TYPE_BACKUP_TARGET, and I'm not sure
> > why this needs to be blocked, or if we simply forgot to enable it.
> 
> Well, even though in practice this wall of code doesn't make much sense,
> of course it will be safe for potential additions of new op blockers.
> 
> And of course we actually don't want these blockers at all anymore...

Yes, but dataplane shouldn't really be special enough any more that we
want to disable features for it initially. By now it sounds more like an
easy way to forget unblocking a new feature even though it would work.

So perhaps we should really just remove the blockers from dataplane.
Then we don't have to answer the question above...

Kevin


pgp6l8ch6aTn6.pgp
Description: PGP signature

Re: [Qemu-block] [PATCH v7 12/24] virtio-blk: Functions for op blocker management

[Stefan Hajnoczi]

On Wed, Nov 25, 2015 at 05:26:02PM +0100, Max Reitz wrote:
> On 25.11.2015 17:18, Kevin Wolf wrote:
> > Am 25.11.2015 um 17:03 hat Max Reitz geschrieben:
> >> On 25.11.2015 16:57, Kevin Wolf wrote:
> >>> Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
> >>> This makes me wonder: What do we even block here any more? If I didn't
> >>> miss anything, it's only BLOCK_OP_TYPE_BACKUP_TARGET, and I'm not sure
> >>> why this needs to be blocked, or if we simply forgot to enable it.
> >>
> >> Well, even though in practice this wall of code doesn't make much sense,
> >> of course it will be safe for potential additions of new op blockers.
> >>
> >> And of course we actually don't want these blockers at all anymore...
> > 
> > Yes, but dataplane shouldn't really be special enough any more that we
> > want to disable features for it initially. By now it sounds more like an
> > easy way to forget unblocking a new feature even though it would work.
> > 
> > So perhaps we should really just remove the blockers from dataplane.
> > Then we don't have to answer the question above...
> 
> Well, maybe. I guess this is up to Stefan.

At this point blockdev.c and block jobs acquire/release AioContext,
hence all these op blockers are being unblocked.  I think we can switch
from whitelisting (unblocking) nearly everything to blacklisting
(blocking) only things that aren't supported yet.

Stefan


signature.asc
Description: PGP signature