Monday, 9 April 2012

debian - Managing maintainer scripts for packages with multple .debs

If you've ever installed a package from command line on Linux, you must have noticed two main prompts related to package configuration: one asking what to do with installed configuration files with local changes, and one providing feedback right after the installation/upgrade/removal/purge has completed.

Under Debian the latter was most probably the postinst script, one of the Package Maintainer Scripts which is executed after installation and configuration.

These diagrams are very useful to understand what happens to the Maintainer Scripts in different circumstances: first installation, upgrade, removal, purge.

Their name is probably self-explanatory: preinst, postinst, prerm, postrm. Each of them takes zero or more arguments depending on the scenario. It can help to understand that those are really just scripts executed by dpkg - and typically they are shell scripts, sometimes with interactive prompts.

If you're building your own packages, you surely already have one of more of those scripts in the source debian/ directory. They are automatically included in the .deb by dpkg-buildpackage (or any of its wrappers) during the build.

What you may find interesting - and this is the purpose of this article - is that if you're building multiple .deb packages from the same debian/ directory (i.e. using the same debian/control makefile), having for example a single postinst script is not sufficient. Only the first built .deb will contain it, while the others won't.
The symptom in this example is that when installing the packages other than the first built, you won't see any of the prompts, feedback or actions you are delegating to postinst.

To fix this, it's sufficient to include for example a postinst script per package:
debian/control
debian/rules
[...]
debian/package_1.postinst
debian/package_2.postinst
debian/package_3.postinst

In case you're wondering, yes, you can have a single postinst script and just create a symbolic link to it for each package.

Tuesday, 3 April 2012

CANCELing a call - Trip-wires for the SIP fans

SIP is a relatively simple, text-based, human readable protocol that is now the standard de facto for VoIP signalling.
The protocol though (in my opinion!) is a little tricky, where typically the tricks are: details.

In this first post of a series of "Trip-wires for the SIP fans", I'll talk about CANCEL.

The main concept is easy: a caller may decide to cancel a call before this is answered. To do this, it sends a CANCEL request to the called party.

What's important to know is that:
A CANCEL request relates to an INVITE request, and does not relate to the SIP dialog the request may have created (or will create).

For this reason the To header tag must be the same as the INVITE request, even if meanwhile there's been a provisional response to the INVITE creating a dialog (e.g. a 180 with a tag in the To header).

From RFC 3261, 9.1:
The following procedures are used to construct a CANCEL request.  The
   Request-URI, Call-ID, To, the numeric part of CSeq, and From header
   fields in the CANCEL request MUST be identical to those in the
   request being cancelled, including tags.  A CANCEL constructed by a
   client MUST have only a single Via header field value matching the
   top Via value in the request being cancelled.  Using the same values
   for these header fields allows the CANCEL to be matched with the
   request it cancels [...].

Although not so common I guess, consider that a CANCEL can be sent also for a re-INVITE (i.e. a request to update an ongoing session, e.g. to add video). In this case there will be a tag in the To header, as the INVITE is an in-dialog request and the CANCEL just refers to it.

And if you use sipp unfortunately no, you can't rely on the [branch] syntax to assign a branch to the Via in the CANCEL request, because sipp doesn't have perception of precedent open transactions.
If the branch in the CANCEL's Via header is different than the one in the INVITE top most Via, the UAS will not be able to match the two requests and canceling will (most likely) fail.