linux-kernel - Re: [PATCH 1/4] module: add syscall to load module from fd

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAMOw1v4KOcH+OckoFzWEr68aUweKD-6gH83BHmBm-1c7PqhCpQ@mail.gmail.com>
Date:	Tue, 23 Oct 2012 14:25:17 -0200
From:	Lucas De Marchi <lucas.demarchi@...fusion.mobi>
To:	Kees Cook <keescook@...omium.org>
Cc:	Rusty Russell <rusty@...tcorp.com.au>, mtk.manpages@...il.com,
	"H. Peter Anvin" <hpa@...or.com>, linux-kernel@...r.kernel.org,
	jonathon@...masters.org
Subject: Re: [PATCH 1/4] module: add syscall to load module from fd

On Tue, Oct 23, 2012 at 1:42 PM, Kees Cook <keescook@...omium.org> wrote:
> On Mon, Oct 22, 2012 at 9:08 PM, Lucas De Marchi
> <lucas.demarchi@...fusion.mobi> wrote:
>> On Tue, Oct 23, 2012 at 1:40 AM, Kees Cook <keescook@...omium.org> wrote:
>>> On Mon, Oct 22, 2012 at 7:37 PM, Lucas De Marchi
>>> <lucas.demarchi@...fusion.mobi> wrote:
>>>> On Mon, Oct 22, 2012 at 5:39 AM, Rusty Russell <rusty@...tcorp.com.au> wrote:
>>>>> "Michael Kerrisk (man-pages)" <mtk.manpages@...il.com> writes:
>>>>>>> FIX: add flags arg to sys_finit_module()
>>>>>>>
>>>>>>> Thanks to Michael Kerrisk for keeping us honest.
>>>>>>
>>>>>> w00t! Thanks, Rusty ;-).
>>>>>>
>>>>>> Acked-by: Michael Kerrisk <mtk.manpages@...il.com>
>>>>>
>>>>> Here's the version I ended up with when I added two flags.
>>>>>
>>>>> Lucas, is this useful to you?
>>>>>
>>>>> BTW Michael: why aren't the syscall man pages in the kernel source?
>>>>>
>>>>> Thanks,
>>>>> Rusty.
>>>>>
>>>>> module: add flags arg to sys_finit_module()
>>>>>
>>>>> Thanks to Michael Kerrisk for keeping us honest.  These flags are actually
>>>>> useful for eliminating the only case where kmod has to mangle a module's
>>>>> internals: for overriding module versioning.
>>>>>
>>>>> Signed-off-by: Rusty Russell <rusty@...tcorp.com.au>
>>>
>>> Acked-by: Kees Cook <keescook@...omium.org>
>>>
>>>> I wonder if we shouldn't get a new init_module2() as well, adding the
>>>> flags parameter. Of course this would be in another patch.
>>>>
>>>> My worries are that for compressed modules we still need to use
>>>> init_module() and then --force won't work with signed modules.
>>>
>>> For those cases, I think it should remain up to userspace to do the
>>> decompress and use init_module(). The code I'd written for patching
>>> module-init-tools basically just kept the fd around if it didn't need
>>> to mangle the module, and it would use finit_module (written before
>>> the flags argument was added):
>>>
>>>         /* request kernel linkage */
>>> -       ret = init_module(module->data, module->len, opts);
>>> +       if (fd < 0)
>>> +               ret = init_module(module->data, module->len, opts);
>>> +       else {
>>> +           ret = finit_module(fd, opts);
>>> +           if (ret != 0 && errno == ENOSYS)
>>> +                   ret = init_module(module->data, module->len, opts);
>>> +       }
>>>         if (ret != 0) {
>>>
>>> (And yes, I realize kmod is what'll actually be getting this logic.
>>> This was for my testing in Chrome OS, which is still using
>>> module-init-tools.)
>>
>> sure... but do you realize this will fail in case kernel is checking
>> module signature and we passed --force to modprobe (therefore mangled
>> the decompressed memory area)?
>
> Hm, yeah, userspace mangling of a module plus signing would fail.
> Seems like mangling and signing aren't compatible. Doing it in
> kernel-space (as now written for finit_module) solves that, but it
> means that now compression isn't possible if you need both signing and
> mangling.
>
> I'm not a user of signing, compression, or mangling, so I'm probably a
> bit unimaginative here. It seems like the case for needing all three
> is pretty uncommon. (e.g. if you're doing compression, you're probably
> building embedded images, which means you're unlikely to need
> --force.)


Some desktop distros ship compressed modules by default. I received
feedback from distros some months ago that this is basically because
of the disk space, not performance.  However some measurements I did
in a regular laptop with spinning disk showed a small advantage
performance-wise, too  (with SSD is another story and uncompressed
wins by a large margin)

Since this only affects users of --force option, I think it only
affects module developers, who could uncompress the module, call
depmod again, and modprobe it. ( Mixing compressed and uncompressed
modules used not to work in module-init-tools and earlier versions of
kmod wrt dependencies, but now it should be a seamless operation )


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