Re: [PATCH bpf-next] bpf: fix segmentation fault of test_progs

From: Andrii Nakryiko <andrii.nakryiko@gmail.com>
To: Jianlin Lv <Jianlin.Lv@arm.com>
Cc: bpf <bpf@vger.kernel.org>,
	"David S. Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>,
	Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Yonghong Song <yhs@fb.com>,
	open list <linux-kernel@vger.kernel.org>,
	Networking <netdev@vger.kernel.org>
Subject: Re: [PATCH bpf-next] bpf: fix segmentation fault of test_progs
Date: Fri, 7 Aug 2020 13:13:25 -0700	[thread overview]
Message-ID: <CAEf4Bza2vJYRbzo2Qbx_XRBPsS-n3dxhOaK+vzjqrhb2wUaCnQ@mail.gmail.com> (raw)
In-Reply-To: <20200807172016.150952-1-Jianlin.Lv@arm.com>

On Fri, Aug 7, 2020 at 10:21 AM Jianlin Lv <Jianlin.Lv@arm.com> wrote:
>
> test_progs reports the segmentation fault as below
>
> $ sudo ./test_progs -t mmap --verbose
> test_mmap:PASS:skel_open_and_load 0 nsec
> ......
> test_mmap:PASS:adv_mmap1 0 nsec
> test_mmap:PASS:adv_mmap2 0 nsec
> test_mmap:PASS:adv_mmap3 0 nsec
> test_mmap:PASS:adv_mmap4 0 nsec
> Segmentation fault
>
> This issue was triggered because mmap() and munmap() used inconsistent
> length parameters; mmap() creates a new mapping of 3*page_size, but the
> length parameter set in the subsequent re-map and munmap() functions is
> 4*page_size; this leads to the destruction of the process space.
>
> Another issue is that when unmap the second page fails, the length
> parameter to delete tmp1 mappings should be 3*page_size.
>
> Signed-off-by: Jianlin Lv <Jianlin.Lv@arm.com>
> ---
>  tools/testing/selftests/bpf/prog_tests/mmap.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/tools/testing/selftests/bpf/prog_tests/mmap.c b/tools/testing/selftests/bpf/prog_tests/mmap.c
> index 43d0b5578f46..2070cfe19cac 100644
> --- a/tools/testing/selftests/bpf/prog_tests/mmap.c
> +++ b/tools/testing/selftests/bpf/prog_tests/mmap.c
> @@ -192,7 +192,7 @@ void test_mmap(void)
>         /* unmap second page: pages 1, 3 mapped */
>         err = munmap(tmp1 + page_size, page_size);
>         if (CHECK(err, "adv_mmap2", "errno %d\n", errno)) {
> -               munmap(tmp1, map_sz);
> +               munmap(tmp1, 3 * page_size);

this is a good catch, thank you!

>                 goto cleanup;
>         }
>
> @@ -207,8 +207,8 @@ void test_mmap(void)
>         CHECK(tmp1 + page_size != tmp2, "adv_mmap4",
>               "tmp1: %p, tmp2: %p\n", tmp1, tmp2);
>
> -       /* re-map all 4 pages */
> -       tmp2 = mmap(tmp1, 4 * page_size, PROT_READ, MAP_SHARED | MAP_FIXED,
> +       /* re-map all 3 pages */
> +       tmp2 = mmap(tmp1, 3 * page_size, PROT_READ, MAP_SHARED | MAP_FIXED,
>                     data_map_fd, 0);

"all 3 pages" is a lie, there are 4. I'd still want to work with all 4
pages. How about we mmap() 4 pages of anonymous memory first, then do
all the mmap() with MAP_FIXED, re-using that memory range. That will
ensure that we are not stepping on any other allocated memory, right?

>         if (CHECK(tmp2 == MAP_FAILED, "adv_mmap5", "errno %d\n", errno)) {
>                 munmap(tmp1, 3 * page_size); /* unmap page 1 */
> @@ -226,7 +226,7 @@ void test_mmap(void)
>         CHECK_FAIL(map_data->val[2] != 321);
>         CHECK_FAIL(map_data->val[far] != 3 * 321);
>
> -       munmap(tmp2, 4 * page_size);
> +       munmap(tmp2, 3 * page_size);
>
>         /* map all 4 pages, but with pg_off=1 page, should fail */
>         tmp1 = mmap(NULL, 4 * page_size, PROT_READ, MAP_SHARED | MAP_FIXED,
> --
> 2.17.1
>