Re: [PATCH bpf-next v4 2/2] selftests/bpf: check whether s32 is sufficient for kfunc offset

From: Hou Tao <houtao1@huawei.com>
To: Yonghong Song <yhs@fb.com>, Hou Tao <hotforest@gmail.com>,
	Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Martin KaFai Lau <kafai@fb.com>,
	Andrii Nakryiko <andrii@kernel.org>,
	"David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>,
	John Fastabend <john.fastabend@gmail.com>,
	<netdev@vger.kernel.org>, <bpf@vger.kernel.org>
Subject: Re: [PATCH bpf-next v4 2/2] selftests/bpf: check whether s32 is sufficient for kfunc offset
Date: Tue, 8 Feb 2022 10:47:28 +0800	[thread overview]
Message-ID: <c8ae6afe-d383-12b9-10df-57dc1da9b566@huawei.com> (raw)
In-Reply-To: <585256c2-318f-3895-8175-e7d7cbe1cd4f@fb.com>

Hi,

On 2/8/2022 2:33 AM, Yonghong Song wrote:
>
>
> On 2/5/22 8:31 PM, Hou Tao wrote:
>> In add_kfunc_call(), bpf_kfunc_desc->imm with type s32 is used to
>> represent the offset of called kfunc from __bpf_call_base, so
>> add a test to ensure that the offset will not be overflowed.
>>
>> Signed-off-by: Hou Tao <houtao1@huawei.com>
>> ---
>>   .../selftests/bpf/prog_tests/ksyms_module.c   | 42 +++++++++++++++++++
>>   1 file changed, 42 insertions(+)
>>
[...]
>> +    /* Ensure kfunc call is supported */
>> +    skel = test_ksyms_module__open_and_load();
>> +    if (!ASSERT_OK_PTR(skel, "test_ksyms_module__open"))
>> +        return;
>> +
>> +    err = kallsyms_find("bpf_testmod_test_mod_kfunc", &kfunc_addr);
>> +    if (!ASSERT_OK(err, "find kfunc addr"))
>> +        goto cleanup;
>> +
>> +    err = kallsyms_find("__bpf_call_base", &base_addr);
>> +    if (!ASSERT_OK(err, "find base addr"))
>> +        goto cleanup;
>> +
>> +    used_offset = kfunc_addr - base_addr;
>> +    actual_offset = kfunc_addr - base_addr;
>> +    ASSERT_EQ((long long)used_offset, actual_offset, "kfunc offset
>> overflowed");
>
> I am a little bit confused about motivation here. Maybe I missed something. If
> we indeed have kfunc offset overflow,
> should kernel verifier just reject the program? Specially,
> we should make the above test_ksyms_module__open_and_load()
> fail?
In add_kfunc_call(), the calculation of imm doesn't consider the overflow
of s32. So test_ksyms_module__open_and_load() will succeed. I think the
better solution is to put the overflow check in add_kfunc_call(), so will
drop this patch and add the overflow check in add_kfunc_call() instead.

Regards,
Tao

[...]