[dm-devel] [RFC PATCH 0/2] multipath: change default devnode blacklist

[Martin Wilck]

On Fri, 2020-06-05 at 17:27 -0500, Benjamin Marzinski wrote:
> On Fri, Jun 05, 2020 at 07:20:26PM +0000, Martin Wilck wrote:
> > Hi Ben,
> > 
> > On Thu, 2020-06-04 at 19:30 -0500, Benjamin Marzinski wrote:
> > > I recently got a request to add the Oracle ASM filer driver
> > > devices
> > > to
> > > multipath's builtin devnode blacklist.  However, instead of
> > > having to
> > > do
> > > always this for each device type individually, I decided to make
> > > multipath blacklist all non scsi, nvme, and dasd devnodes by
> > > default.
> > > This is what the multipath udev rules already do. If people don't
> > > like
> > > this solution, the alternative is to add another line to the
> > > default
> > > devnode blacklist like "^(asm/|oracleafd/|ofsctl)".
> > 
> > Thanks, this looks ok. But I'd like to propose an alternative idea:
> > Extend the RE syntax in our config file to allow negated regular 
> > expressions. Like this:
> > 
> > blacklist {
> >     devnode "!(^(sd[a-z]|dasd[a-z]|nvme[0-9]))"
> > }
> > 
> > The "!(${RE})" construct would mean "everything that does not match
> > ${RE}".
> > This logic would only be applied to an entire regex.
> > If a user needs a RE matching with "!(" and ending with ")", she
> > can escape 
> > the exclamation mark "\!(like this)".
> > 
> > AFAICS this could be implemented quite easily (by adding a "bool
> > negate" field
> > in struct blentry and some simple parser logic), and could be
> > applied to other 
> > REs in the config file as well. We could print this with "multipath
> > -t", and 
> > we wouldn't need to document an exception.
> > 
> > It's also pretty much backwards-compatible, I don't think many
> > people use
> > regexes starting with "!(" for multipath these days.
> 
> Sure, but since we can only really support negating the whole regular
> expression, and c regular expressions don't treat '!' as a special
> character, why do we need the prentheses around the regular
> expression?
> It seems like we can just treat regular expressions starting with '!'
> as
> negated, and ones starting with "\!" as starting with a literal '!'.
> 
> Do you think that there is much chance that users have blacklist
> strings
> that start with '!'? There are no devnodes, udev properties, or
> protocols that start with that.  I don't know of a UUID format that
> has
> an exclamation point, and while it's possible that a product string
> starts with one, it seems really unlikely.
> 

Sure, just using "!" would be fine, too, and simpler. Does this mean
you agree with my proposal in general?

Martin

-- 
Dr. Martin Wilck <mwilck at suse.com>, Tel. +49 (0)911 74053 2107
SUSE  Software Solutions Germany GmbH
HRB 36809, AG Nürnberg GF: Felix
Imendörffer