openslp-devel

[Openslp-devel] [PATCH] semi-automatically refresh listening interfaces

[Gavin L. <ga...@co...>]

I had a case recently where a particular WinXP PC would almost always start
up the OpenSLPd service prior to bringing up the network interface
(presumably due to a slow DHCP server or something), which of course meant
that SLP queries on that network were not answered, since SLPD currently
only checks the interface list on startup.

So I fixed that.

The attached patch (against the OpenSLP 2.0 beta 2 release tarball) makes
the following changes to SLPD:
  * on the Linux side, sending SIGHUP to SLPD will result in it closing and
reopening the listening sockets (thereby adding or removing supported
interfaces as required), in addition to prior behaviour (reinitialising the
settings files and DA list).  Current outgoing connections and dynamic
registrations are unaffected, but there will be a brief period when the
server is unresponsive to new incoming queries from the network.
  * on the Windows side, it will interpret the SERVICE_CONTROL_PARAMCHANGE
command as an equivalent SIGHUP request.  (This provides a way to update the
config files semi-live, as on Linux.)
  * on the Windows side, it will also automatically ask Windows to notify it
of a change in the IP address table, and also treat that as a SIGHUP.
  * minor code tidy-ups and status reporting fixes.

Caveats:
  * this patch has only been tested on Windows (XP and 7); the changes
affecting the Linux side should be harmless (and beneficial) but I have not
verified that.
  * automatic detection of IPv6 address changes only works on Vista+, but
that should be ok as IPv6 generally doesn't work on XP anyway.
  * IPv6 detection has not been tested (all the networks I have access to
use only IPv4), but since the code is the same as for IPv4 it should work
fine.
  * the SIGHUP code may get triggered several times by one logical change
(presumably due to multiple internal changes); for example when a disabled
network interface is re-enabled it tends to run twice on XP and 3-4 times on
Win7 (note that different notification APIs are used on each).  While
inefficient this does not appear harmful, so I'm leaving it as is.  The
reinits are throttled by the main loop timeout, currently 1s.
  * the Linux side will not automatically detect IP address changes; code or
hook scripts to do the autodetection on Linux are left as an exercise for
some other intrepid soul.
  * this is my first foray into the OpenSLP codebase, so I may have missed
updating something necessary, or done something inefficiently -- but it
seems to be working ok so far.  I'm sure things could be improved though
(such as reinitialising only the listening sockets, not accepted sockets).

[Openslp-devel] [PATCHv2] semi-automatically refresh listening interfaces

[Gavin L. <ga...@co...>]

I've attached a variation of the previous patch (still against the 2.0 beta
2 tarball) which handles the reinit slightly differently from the original.

This one will start listening on any newly discovered (or configured)
interfaces, without closing or stopping any previously opened incoming
sockets.

The advantage over the previous patch is that it will cause less disruption
to ongoing communications, particularly if only a secondary interface is
changing.  The disadvantage is that if the service is left running for a
long time on a network that keeps assigning it different addresses
(presumably via a DHCP server with a bad memory) then it will gradually eat
up more and more resources, and probably eventually fail.  (I'm not sure
which tradeoff is better overall.)

A suggested improvement (which again I'm not quite familiar enough with the
code to do at this point) would be to track which sockets are associated
with a particular interface and close them if their interface is not still
present during the reinit.

Note that for code simplicity it's currently assuming that if the TCP listen
socket exists then all the other related sockets are still ok too.

Re: [Openslp-devel] [PATCHv2] semi-automatically refresh listening interfaces

[John C. <joh...@gm...>]

Hi Gavin,

Your dynamic ifc reinit patch has been incorporated into the repository.
Thanks for your efforts.

Regards,
John

> -----Original Message-----
> From: Gavin Lambert [mailto:ga...@co...]
> Sent: Thursday, October 25, 2012 9:32 PM
> To: ope...@li...
> Subject: [Openslp-devel] [PATCHv2] semi-automatically refresh listening
> interfaces
> 
> I've attached a variation of the previous patch (still against the 2.0
beta
> 2 tarball) which handles the reinit slightly differently from the
original.
> 
> This one will start listening on any newly discovered (or configured)
> interfaces, without closing or stopping any previously opened incoming
> sockets.
> 
> The advantage over the previous patch is that it will cause less
disruption to
> ongoing communications, particularly if only a secondary interface is
> changing.  The disadvantage is that if the service is left running for a
long time
> on a network that keeps assigning it different addresses (presumably via a
> DHCP server with a bad memory) then it will gradually eat up more and more
> resources, and probably eventually fail.  (I'm not sure which tradeoff is
better
> overall.)
> 
> A suggested improvement (which again I'm not quite familiar enough with
> the code to do at this point) would be to track which sockets are
associated
> with a particular interface and close them if their interface is not still
present
> during the reinit.
> 
> Note that for code simplicity it's currently assuming that if the TCP
listen
> socket exists then all the other related sockets are still ok too.

Re: [Openslp-devel] [PATCHv2] semi-automatically refresh listening interfaces

[Gavin L. <ga...@co...>]

Hi John,

Just noticed a minor bug (sorry, it was in my original patch):

In slpd/slp_win32.c line 135 (InterfaceMonitorInit):

-      if (!(self->pNotifyIpInterfaceChange &&
self->pNotifyIpInterfaceChange))
+      if (!(self->pNotifyIpInterfaceChange &&
self->pCancelMibChangeNotify2))

It won't actually cause any problems in practice, as one of these won't
exist without the other.  But this is more correct.

Also the comment at line 146 (regarding casting) can be removed; this was
from an older version of the code, which was solved differently in the final
patch (see line 77).

Regards,
Gavin

> -----Original Message-----
> From: John Calcote [mailto:joh...@gm...]
> Sent: Thursday, 29 November 2012 08:50
> To: 'Gavin Lambert'; ope...@li...
> Subject: RE: [Openslp-devel] [PATCHv2] semi-automatically refresh
> listening interfaces
> 
> Hi Gavin,
> 
> Your dynamic ifc reinit patch has been incorporated into the repository.
> Thanks for your efforts.
> 
> Regards,
> John
> 
> > -----Original Message-----
> > From: Gavin Lambert [mailto:ga...@co...]
> > Sent: Thursday, October 25, 2012 9:32 PM
> > To: ope...@li...
> > Subject: [Openslp-devel] [PATCHv2] semi-automatically refresh
> > listening interfaces
> >
> > I've attached a variation of the previous patch (still against the 2.0
> beta
> > 2 tarball) which handles the reinit slightly differently from the
> original.
> >
> > This one will start listening on any newly discovered (or configured)
> > interfaces, without closing or stopping any previously opened incoming
> > sockets.
> >
> > The advantage over the previous patch is that it will cause less
> disruption to
> > ongoing communications, particularly if only a secondary interface is
> > changing.  The disadvantage is that if the service is left running for
> > a
> long time
> > on a network that keeps assigning it different addresses (presumably
> > via a DHCP server with a bad memory) then it will gradually eat up
> > more and more resources, and probably eventually fail.  (I'm not sure
> > which tradeoff is
> better
> > overall.)
> >
> > A suggested improvement (which again I'm not quite familiar enough
> > with the code to do at this point) would be to track which sockets are
> associated
> > with a particular interface and close them if their interface is not
> > still
> present
> > during the reinit.
> >
> > Note that for code simplicity it's currently assuming that if the TCP
> listen
> > socket exists then all the other related sockets are still ok too.

Re: [Openslp-devel] [PATCHv2] semi-automatically refresh listening interfaces

[John C. <joh...@gm...>]

Thanks Gavin.

changeset:   1766:cdaeb8be5165
tag:         tip
user:        John Calcote <joh...@gm...>
date:        Thu Feb 28 18:59:10 2013 -0700
summary:     glambert: Fix to win32 interface change notify code.

John

> -----Original Message-----
> From: Gavin Lambert [mailto:ga...@co...]
> Sent: Thursday, February 28, 2013 6:40 PM
> To: 'John Calcote'
> Cc: ope...@li...
> Subject: RE: [Openslp-devel] [PATCHv2] semi-automatically refresh
listening
> interfaces
> 
> Hi John,
> 
> Just noticed a minor bug (sorry, it was in my original patch):
> 
> In slpd/slp_win32.c line 135 (InterfaceMonitorInit):
> 
> -      if (!(self->pNotifyIpInterfaceChange &&
> self->pNotifyIpInterfaceChange))
> +      if (!(self->pNotifyIpInterfaceChange &&
> self->pCancelMibChangeNotify2))
> 
> It won't actually cause any problems in practice, as one of these won't
exist
> without the other.  But this is more correct.
> 
> Also the comment at line 146 (regarding casting) can be removed; this was
> from an older version of the code, which was solved differently in the
final
> patch (see line 77).
> 
> Regards,
> Gavin
> 
> > -----Original Message-----
> > From: John Calcote [mailto:joh...@gm...]
> > Sent: Thursday, 29 November 2012 08:50
> > To: 'Gavin Lambert'; ope...@li...
> > Subject: RE: [Openslp-devel] [PATCHv2] semi-automatically refresh
> > listening interfaces
> >
> > Hi Gavin,
> >
> > Your dynamic ifc reinit patch has been incorporated into the repository.
> > Thanks for your efforts.
> >
> > Regards,
> > John
> >
> > > -----Original Message-----
> > > From: Gavin Lambert [mailto:ga...@co...]
> > > Sent: Thursday, October 25, 2012 9:32 PM
> > > To: ope...@li...
> > > Subject: [Openslp-devel] [PATCHv2] semi-automatically refresh
> > > listening interfaces
> > >
> > > I've attached a variation of the previous patch (still against the
> > > 2.0
> > beta
> > > 2 tarball) which handles the reinit slightly differently from the
> > original.
> > >
> > > This one will start listening on any newly discovered (or
> > > configured) interfaces, without closing or stopping any previously
> > > opened incoming sockets.
> > >
> > > The advantage over the previous patch is that it will cause less
> > disruption to
> > > ongoing communications, particularly if only a secondary interface
> > > is changing.  The disadvantage is that if the service is left
> > > running for a
> > long time
> > > on a network that keeps assigning it different addresses (presumably
> > > via a DHCP server with a bad memory) then it will gradually eat up
> > > more and more resources, and probably eventually fail.  (I'm not
> > > sure which tradeoff is
> > better
> > > overall.)
> > >
> > > A suggested improvement (which again I'm not quite familiar enough
> > > with the code to do at this point) would be to track which sockets
> > > are
> > associated
> > > with a particular interface and close them if their interface is not
> > > still
> > present
> > > during the reinit.
> > >
> > > Note that for code simplicity it's currently assuming that if the
> > > TCP
> > listen
> > > socket exists then all the other related sockets are still ok too.
>