Long ago I was debugging kernel/module.c and kernel/kmod.c and had implemented a sanity check to avoid allocating much of the needed struct module from what user passes on finit_module() too early, ie, before checking if the module was actually loaded.
I had disregarded this work as I had bundled it up with some aliasing work I had done, so incorrectly thought this depended on the alias work. Upon a second look it does not. The module name set on userspace on info->name is not the alias but the proper module name, as such using finished_loading() should suffice once to avoid allocating the module if its already present. I've tested this with kmod.sh and found no regressions, the few minor enhancements are documented in patch 3 in detail. I'll let 0-day hammer on it as we review the patches [0] [1] but no build issues so far. [0] https://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/linux.git/log/?h=20171207-avoid-bogus-layout_and_allocate [1] https://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/linux-next.git/log/?h=20171207-avoid-bogus-layout_and_allocate Luis R. Rodriguez (3): module: add an early early_mod_check() module: move finished_loading() module: avoid allocation if module is already present and ready kernel/module.c | 81 ++++++++++++++++++++++++++++++++++----------------------- 1 file changed, 49 insertions(+), 32 deletions(-) -- 2.15.0
