Re: [PATCH] libkmod-module: Remove directory existence check for KMOD_MODULE_BUILTIN

[Lucas De Marchi]

On Thu, Feb 19, 2015 at 12:25 AM, greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, Feb 18, 2015 at 11:19:14PM -0200, Lucas De Marchi wrote:
>> On Wed, Feb 18, 2015 at 8:40 PM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote:
>> > Lucas De Marchi <lucas.de.marchi@xxxxxxxxx> writes:
>> >> On Wed, Feb 18, 2015 at 2:07 AM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote:
>> >> Yeah, I just thought (an wanted that) the attributes were being
>> >> created first and then hooked up in the sysfs tree under
>> >> /sys/module/<modulename>. I.e. if the directory exists and there's no
>> >> initstate this is because it's a builtin module. I don't want to
>> >> wait/sleep on the file to appear because users of
>> >> kmod_module_get_initstate() may not tolerate this behavior.
>> >>
>> >> Looking up at the old module-init-tools, it used an ugly loop with
>> >> usleep() before trying to read the file again :-/
>> >>
>> >> Can we change kernel side guaranteeing the initstate file appears
>> >> together with the directory?
>> >
>> > Greg?  The core problem is that kmod looks for
>> > /sys/module/<name>/initstate; if it's not there, it assumes a builtin
>> > module.
>>
>> Just to make it clear:
>>
>> We try to open /sys/module/<name>/initstate. If it fails we stat
>> /sys/module/<name> checking if it exists and is a directory. If it
>> does then we assume the module is builtin.
>>
>> > However, this is racy when a module is being inserted.  Is there a way
>> > to create this sysfs file and dir atomically?
>>
>> Greg, the question is still valid since it'd be nice to have this
>> guarantee and be able to correctly reply the state with whatever is in
>> initstate file, but...
>
> You should always wait until you get the uevent that the object was
> added before poking around in sysfs.  The kernel will guarantee all of
> the needed files / directories will be created before that event is sent
> out.  That's why we added the uevent message.

for kmod I think I prefer the alternative of not needing it at all...
for daemons and other tools it makes sense indeed.

> So by just busy-spinning on the directory and ignoring the uevent, you
> are just blindly guessing as to when things are finished, which is racy
> as you see.

note there's no busy-spinning in kmod.... this was in
module-init-tools and it's what I'm saying I don't want to do. Not
something you want in a library.

-- 
Lucas De Marchi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/