On Thu, Sep 12, 2013 at 09:36:43AM +0400, Michael Tokarev wrote: > 12.09.2013 07:02, Fam Zheng wrote. > >On Wed, 09/11 11:46, Richard Henderson wrote: > >>On 09/11/2013 08:48 AM, Daniel P. Berrange wrote: > >>>We know the precise list of valid modules when building QEMU, > >>>so IMHO, this should just explicitly load each known module > >>>name, and *not* readdir. Also it should do something along the > >>>lines suggested their of poisoning exported symbols with a > >>>build hash to guarantee the modules loaded match the original > >>>binary and that the symbols change on every rebuild. > >> > >>We need not mangle the symbols, which could be complicated to > >>actually implement, and irritating to work around within gdb. > >> > >Agree with this, some id or hash check should be enough. > > A solution which I proposed at the very beginning -- to export > a "hashed" init function from modules, and call it from the > main executable. Like, instead of, say, qemu_module_init(), > call qemu_module_init_0xdeadbeaf(), where 0xdeadbeaf is a > hash of some build-dependent value. This should be enough > to keep it going. > > Ofcourse, if a module lacks this function, it should not be > loaded.
Yep, that would be a reasonable way todo this. THe current patches use attribute(constructor) so QEMU doesn't actually call any explicit init function after dlopen()ing. That could easily be changed though. Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|
