qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: "Benoît Canet" <benoit.canet@irqsave.net>
To: Fam Zheng <famz@redhat.com>
Cc: "Benoît Canet" <benoit.canet@irqsave.net>,
	kwolf@redhat.com, qemu-devel@nongnu.org, stefanha@redhat.com,
	mreitz@redhat.com
Subject: Re: [Qemu-devel] n ways block filters
Date: Mon, 17 Mar 2014 14:52:25 +0100	[thread overview]
Message-ID: <20140317135225.GA3047@irqsave.net> (raw)
In-Reply-To: <20140317031231.GA28582@T430.nay.redhat.com>

The Monday 17 Mar 2014 à 11:12:31 (+0800), Fam Zheng wrote :
> On Fri, 03/14 16:57, Benoît Canet wrote:
> > 
> > Hello list,
> > 
> > I plan to convert throttling to a block filter and write n way throttling
> > support.
> > 
> > I discussed a bit with Stefan on the list and we came to the conclusion that the
> > block filter API need group support.
> > 
> > filter group:
> > -------------
> > 
> > My current plan to implement this is to add the following fields to the BlockDriver
> > structure.
> > 
> > int bdrv_add_filter_group(const char *name, QDict options);
> > int bdrv_reconfigure_filter_group(const char *name, QDict options);
> > int bdrv_destroy_filter_group(const char *name);
> > 
> > These three extra method would allow to create, reconfigure or destroy a block
> > filter group. A block filter group contain the shared or non shared state of the
> > blockfilter. For throttling it would contains the ThrottleState structure.
> > 
> > Each block filter driver would contains a linked list of linked list where the
> > BDS are registered grouped by filter groups state.
> 
> Sorry I don't fully understand this. Does a filter group contain multiple block
> filters, and every block filter has effect on multiple BDSes? Could you give an
> example?

The driver module says block/throttle.c would contains a linked list of ThrottleGroup
in a static variable.

Each ThrottleGroup would have a name, a ThrottleState (as in utils/throttle.c)
and a linked list of members BDS.

The s variable of the bdrv_* method of throttle.c would have a pointer to the
ThrottleGroup this BDS is a member of hence the driver can access the
ThrottleState

So each BDS filter would have only one BDS parent and one BDS ->file child but
would also be member of a ThrottleGroup.

This would allow to do throttling on a group of devices.

Throttling only one device would create a ThrottleGroup of name bds->device_name
containing only the throttled BDS.

> 
> > 
> > The regular bdrv_open callback would be used to instantiate a block filter and
> > add it to a filter group. This method would also take a new-node-name for the new
> > filter. This node-name would become the name of the new filter.
> > bdrv_close would cleanup and deregister from a filter group.
> > 
> > An extra filter-group field in the option dict would allow the bdrv_open method
> > to register the newly opened block filter in it's filter group.
> > The BDS structure would have a direct pointer to it's filter group state.
> > 
> > Utility methods to do the bdrv_add_filter_group bdrv_open then bdrv_swap to
> > install a new filter can be provided by block.c. The same can be done for filter
> > close and desinstallation.
> 
> So you are defining block filter as a new kind of block driver. Is a filter
> always on top above everything else by definition?

Yes,

A filter is instanciated on top of another BDS.
Filters can be stacked on top of another to form a chain.
 
> But I am afraid BlockDriverState is already taking too many responsibilities
> here (backend, protocol driver, format driver, filter...). I was wondering if
> it is clearer to rather introduce bs->filter_list to point to a list of
> BlockFilter (a new sturcture, tailored for a block filter), and don't bother
> with bdrv_swap.

Hmm blkverify and quorum are already block filters reusing the BlockDriver
interface. The good this with this is that it would allow to move some code from
block.c to block/filter_name*.c: block throttling for example.
Crypto would be another use case.

Best regards

Benoît

> 
> Thanks,
> Fam
> 
> > 
> > Legacy throttling QMP API
> > -------------------------
> > 
> > The legacy throttling API would create throttling filters groups containing only
> > one BDS.
> > 
> > By default for every 1 way block filter block.c would create a filter group
> > using the BDS id or node-name as group name. This allow for easy filer removal
> > with the bds reference.
> > 
> > Group throttling API
> > --------------------
> > 
> > Commands would be added to create throttling filter groups reconfigure and remove
> > them.
> > 
> > Two additional commands would be added to create and insert a block filter in a
> > given group or close and remove it.
> > 
> > Before I start implementing something what are your thougths on this ?
> > 
> > Best regards
> > 
> > Benoît
> > 
> > 
> > 
> 

  reply	other threads:[~2014-03-17 13:52 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-03-14 15:57 [Qemu-devel] n ways block filters Benoît Canet
2014-03-17  3:12 ` Fam Zheng
2014-03-17 13:52   ` Benoît Canet [this message]
2014-03-17 16:02   ` Stefan Hajnoczi
2014-03-18 13:27     ` Kevin Wolf
2014-03-20 14:05       ` Benoît Canet
2014-03-20 15:12         ` Kevin Wolf
2014-03-20 15:47           ` Benoît Canet
2014-03-20 16:06           ` Benoît Canet
2014-03-24 14:53             ` Benoît Canet
2014-03-24 15:58               ` Kevin Wolf
2014-03-27 15:49                 ` Benoît Canet

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20140317135225.GA3047@irqsave.net \
    --to=benoit.canet@irqsave.net \
    --cc=famz@redhat.com \
    --cc=kwolf@redhat.com \
    --cc=mreitz@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=stefanha@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).