Debugging Debian Installer
Posted: Sun, 23 January 2011 | permalink | No comments
I've been doing a bit of work recently on the netcfg portion of Debian Installer, with the largest work so far being adding IPv6 support.
Unlike most of the rest of d-i, netcfg is written in C, and as such suffers from some difficulties in the realm of debugging. Worse, the whole d-i environment is rather bereft of the usual debugging tools (strace, lsof, etc) -- but most importantly, in this case, gdb.
Thanks to some hints from Julien Cristau, I was reminded of the existence of gdbserver, a small part of gdb that allows you to debug a program running on a remote machine, using a copy of gdb running on your local machine. It's designed for debugging embedded systems, I think, because all of the heavy lifting (symbol lookups, etc) happens on your local machine, not where gdbserver is running.
Here are the steps I took to get a useful gdbserver session inside d-i:
- Get the gdb sources (apt-get source gdb) and drop into the gdbserver tree (cd gdb-x.y.z/gdb/gdbserver)
- Configure gdbserver to build a static binary (LDFLAGS=-static ./configure). This is necessary because you don't want to have to faff around with library dependencies inside the d-i environment. Then build gdbserver (make).
- Copy gdbserver over to your d-i machine/VM (I like using netcat; nc -l -p 31310 >/bin/gdbserver on the d-i machine, and nc IPADDR 31310 <gdbserver where you built gdbserver), then make it executable (chmod /bin/gdbserver)
- Fire up gdbserver in d-i, attaching it to the process you wish to debug: gdbserver host:31337 --attach PID
- Now things get a little trickier... on your local machine, you need to have an unstripped copy of the binary you're debugging. For me, this is easy, because I'm building netcfg myself (to keep the symbols, I just run DEB_BUILD_OPTIONS=noopt dpkg-buildpackage -b -rfakeroot) but I'm not sure what you'd do if you were just tracing around for fun.
- Once you've got your binary, just run gdb BINARY on your host machine to start gdb running.
- For the "even trickier" part of the show, your local machine also needs copies of all the libraries your binary is linked against. If you don't have this, when you try to debug you'll get all sorts of "wrong library or version mismatch?" errors, and your backtraces will likely break (amongst other things). Again, for me, this is easy, because I'm only running the d-i initrd, but if you're debugging something later in the installation process you might need to unpack some udebs somewhere. All I have to do is keep the initrd build tree around (which lives in the installer tree, under build/tmp/netboot/tree). Once you've got your tree of libraries, you can continue.
- You need to tell gdb about that tree of libraries, by running the following in gdb: set solib-absolute-prefix /path/to/library/tree
- Now, tell gdb to connect to your gdbserver: target remote IPADDR:31337 -- gdb should tell you "Remote debugging using IPADDR:31337", and the gdbserver should say "Remote debugging from host LOCALIP:31337".
- Now, you should be able to interact with the program being debugged just like any other. Enjoy!
Post a comment
All comments are held for moderation; markdown formatting accepted.