linux-kernel - Re: [dm-devel] [PATCH 0/9] Generate uevents for all DM events

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 5 Oct 2016 08:26:41 +0200
From:   Hannes Reinecke <hare@...e.de>
To:     Andy Grover <agrover@...hat.com>,
        Greg KH <gregkh@...uxfoundation.org>, dm-devel@...hat.com,
        linux-kernel@...r.kernel.org, snitzer@...hat.com
Subject: Re: [dm-devel] [PATCH 0/9] Generate uevents for all DM events

On 10/05/2016 02:40 AM, Alasdair G Kergon wrote:
> On Tue, Oct 04, 2016 at 04:39:28PM -0700, Andy Grover wrote:
>> devicemapper is using uevents for:
>> a. dm-verity detected corruption
>> b. dm-multipath: path failed or reinstated
>> c. dm device renamed
>> d. there's also some use in md and bcache.
>>
>> devicemapper uses DM_EVENT ioctl (yuck) for:
>> 1. dm-thin pool data/metadata filling up (hit a threshold)
>> 2. dm-cache is now clean
>> 3. dm-log flushed or log failed
>> 4. dm-raid error detected or sync complete
> 
>> there doesn't seem to be much technical differentiation between the  
>> two lists.
> 
> The distinction in dm is that events in the first category may affect
> the availability of the device: they represent major (and hopefully
> rare) changes.
> 
> Events in the second category are just notifications: no impact on /dev,
> no need to trigger udev rules, and their use will continue to be
> extended, and (rarely at the moment) could be frequent (which is no
> problem for the existing polling-based mechanism).
> 
>> Instead of using uevent for everything, we could go to a separate  
>> genetlink for 1-4 instead of making them use uevent like a-d, but we'd  
>> end up with two different userspace notification techniques.
> 
> We see these as two different categories of notifications, and prefer
> the greater flexibility a mechanism independent of uevents would
> provide.  The team has discussed several alternatives over the years but
> didn't make a decision as we've not yet reached a point where we're
> straining the existing mechanism too far.
> 
I would love to remove the dm-event usage from multipathing.
As it stands using dm-events from multipathing is a massive resource
drain (we have to allocate a waiting thread for each device), and we
more-or-less disregard that information anyway as it is an ex-post
notification made _after_ someone modified the table.
Given that in most cases it was actually _us_ doing the reconfiguration
it has very little value.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@...e.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)