Re: [PATCH v2] module: Harden STRICT_MODULE_RWX

From: Jessica Yu <jeyu@kernel.org>
To: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>,
	linux-kernel@vger.kernel.org,
	Thomas Gleixner <tglx@linutronix.de>,
	keescook@chromium.org, Josh Poimboeuf <jpoimboe@redhat.com>,
	Miroslav Benes <mbenes@suse.cz>
Subject: Re: [PATCH v2] module: Harden STRICT_MODULE_RWX
Date: Mon, 10 Aug 2020 17:06:50 +0200	[thread overview]
Message-ID: <20200810150647.GB8612@linux-8ccs> (raw)
In-Reply-To: <20200810092523.GA8612@linux-8ccs>

+++ Jessica Yu [10/08/20 11:25 +0200]:
>+++ Mauro Carvalho Chehab [08/08/20 10:12 +0200]:
>[snip]
>>Right now, what happens is:
>>
>>	# modprobe wlcore
>>	modprobe: ERROR: could not insert 'wlcore': Exec format error
>>
>>This seems to be failing for all modules, as doesn't show anything
>>probed.
>>
>>Btw, IMO, it would be useful to have some pr_debug() infra in order to
>>explain why insmod is failing, or to have more error codes used there,
>>as nothing was printed at dmesg. That makes harder to debug issues
>>there. I ended losing a lot of time yesterday rebuilding the Kernel
>>and checking the FS, before deciding to add some printks inside the
>>Kernel ;-)
>>
>>In order for modprobe to start working again, I had to apply this
>>dirty hack:
>>
>><hack>
>>diff --git a/kernel/module.c b/kernel/module.c
>>index 910a57640818..10d590dc48ad 100644
>>--- a/kernel/module.c
>>+++ b/kernel/module.c
>>@@ -2051,11 +2051,12 @@ static int module_enforce_rwx_sections(Elf_Ehdr *hdr, Elf_Shdr *sechdrs,
>>	const unsigned long shf_wx = SHF_WRITE|SHF_EXECINSTR;
>>	int i;
>>
>>+#if 0
>>	for (i = 0; i < hdr->e_shnum; i++) {
>>		if ((sechdrs[i].sh_flags & shf_wx) == shf_wx)
>>			return -ENOEXEC;
>>	}
>>-
>>+#endif
>>	return 0;
>>}
>></hack>

[ I somehow munged the To: header in the last mail. Sorry about that,
it's fixed now. ]

>All this hunk does it reject loading modules that have any sections
>that have both the writable and executable flags. You're saying it's
>happening for all modules on your setup - I am curious as to which
>sections have both these flags - what does readelf -S tell you?

Hmm, I was not able to reproduce this with a cross-compiled kernel
using the attached config (gcc 9.3.0 with vanilla v5.8 kernel). I am
curious if the failing sections are also SHF_ALLOC - in that case, the
code is doing what it is intended to do, which is rejecting loading
any modules with writable and executable sections. If the problematic
sections are *not* SHF_ALLOC, then we might be able to work around
that by ignoring non-SHF_ALLOC sections as they are not copied to the
final module location anyway.

Jessica