Small update here. I broke out of the make run and manually invoked mklibs-copy in verbose mode. It turns out that libkmod.so.2 does it fact get copied out of the /lib directory on the system being used to construct the initrd (I guess I must have had that package installed after all). See the highlighted line in the mklibs-copy output here:
https://gist.github.com/mikepurvis/a7182157f63ea01649cfca01a57f1068#file-mklibs-copy-L275 The same thing happens with libslang.so.2, a few lines further down. This behaviour seems very odd and undesirable to me— wouldn't it break the ability to generate an installer for a different architecture? These libraries should be coming from libkmod2-udeb and libslang2-udeb, respectively, right? Should there be a bug filed for this? Mike On 15 March 2018 at 16:17, Mike Purvis <mpur...@clearpath.ai> wrote: > Hi d-i developers, > > I'm working on an bootloader/installer system for some industrial > equipment. Until now, my main interaction with d-i has been grabbing the > pre-cooked kernel/initrds from Debian or Ubuntu, and tacking on a few extra > files to the tail of the initrd, like preseed, scripts, and an extra udeb > or two. > > However, I've come to a place where I'd like to have more control over > some things (incl exact kernel version) and so I want to generate my own > initrd with just the particular udebs I require. I'm basing most of my work > on the Makefile here, which details downloading a dependency tree of > packages, and creating a bare bones dpkg setup into which they are > installed: > > https://anonscm.debian.org/git/d-i/debian-installer.git/ > tree/build/Makefile#n308 > > My system actually mostly works, but one key file I'm missing is > /lib/libkmod.so.2. This file isn't contained within any of the udebs > (libkmod2-udeb is just a multi-call binary with no libs), and in fact a > little tracing reveals that it springs into existence as a result of the > mklibs-copy call which happens here: > > https://anonscm.debian.org/git/d-i/debian-installer.git/ > tree/build/Makefile#n526 > > Without this lib, systemd-udevd fails to link at runtime (called by the > init script), and the kernel panics shortly after boot. > > So my main question is, what is going on with the mklibs-copy call? Where > are the contents of libkmod.so.2 coming from? They're byte-for-byte > identical with the copy in the libkmod2 package, but in a different > location, and I don't even have that package installed. > > I'm sure there's some silly explanation here that I'm just not seeing, but > a I think I need a poke in the right direction to get there. Thanks, > > Mike >
