[PATCH bpf] selftests/bpf: Use -4095 as the bad address for bits iterator

[PATCH bpf] selftests/bpf: Use -4095 as the bad address for bits iterator

[Hou Tao]

From: Hou Tao <houtao1@huawei.com>

As reported by Byeonguk, the bad_words test in verifier_bits_iter.c
occasionally fails on s390 host. Quoting Ilya's explanation:

  s390 kernel runs in a completely separate address space, there is no
  user/kernel split at TASK_SIZE. The same address may be valid in both
  the kernel and the user address spaces, there is no way to tell by
  looking at it. The config option related to this property is
  ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE.

  Also, unfortunately, 0 is a valid address in the s390 kernel address
  space.

Fix the issue by using -4096 as the bad address for bits iterator, as
suggested by Ilya. Verify that bpf_iter_bits_new() returns -EINVAL for
NULL address and -EFAULT for bad address.

Fixes: ebafc1e535db ("selftests/bpf: Add three test cases for bits_iter")
Reported-by: Byeonguk Jeong <jungbu2855@gmail.com>
Closes: https://lore.kernel.org/bpf/ZycSXwjH4UTvx-Cn@ub22/
Signed-off-by: Hou Tao <houtao1@huawei.com>
---
 .../selftests/bpf/progs/verifier_bits_iter.c  | 32 ++++++++++++++++---
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/bpf/progs/verifier_bits_iter.c b/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
index 156cc278e2fc..7c881bca9af5 100644
--- a/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
+++ b/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
@@ -57,9 +57,15 @@ __description("null pointer")
 __success __retval(0)
 int null_pointer(void)
 {
-	int nr = 0;
+	struct bpf_iter_bits iter;
+	int err, nr = 0;
 	int *bit;
 
+	err = bpf_iter_bits_new(&iter, NULL, 1);
+	bpf_iter_bits_destroy(&iter);
+	if (err != -EINVAL)
+		return 1;
+
 	bpf_for_each(bits, bit, NULL, 1)
 		nr++;
 	return nr;
@@ -194,15 +200,33 @@ __description("bad words")
 __success __retval(0)
 int bad_words(void)
 {
-	void *bad_addr = (void *)(3UL << 30);
-	int nr = 0;
+	void *bad_addr = (void *)-4095;
+	struct bpf_iter_bits iter;
+	volatile int nr;
 	int *bit;
+	int err;
+
+	err = bpf_iter_bits_new(&iter, bad_addr, 1);
+	bpf_iter_bits_destroy(&iter);
+	if (err != -EFAULT)
+		return 1;
 
+	nr = 0;
 	bpf_for_each(bits, bit, bad_addr, 1)
 		nr++;
+	if (nr != 0)
+		return 2;
 
+	err = bpf_iter_bits_new(&iter, bad_addr, 4);
+	bpf_iter_bits_destroy(&iter);
+	if (err != -EFAULT)
+		return 3;
+
+	nr = 0;
 	bpf_for_each(bits, bit, bad_addr, 4)
 		nr++;
+	if (nr != 0)
+		return 4;
 
-	return nr;
+	return 0;
 }
-- 
2.29.2

Re: [PATCH bpf] selftests/bpf: Use -4095 as the bad address for bits iterator

[Ilya Leoshkevich]

On Tue, 2024-11-05 at 12:30 +0800, Hou Tao wrote:
> From: Hou Tao <houtao1@huawei.com>
> 
> As reported by Byeonguk, the bad_words test in verifier_bits_iter.c
> occasionally fails on s390 host. Quoting Ilya's explanation:
> 
>   s390 kernel runs in a completely separate address space, there is
> no
>   user/kernel split at TASK_SIZE. The same address may be valid in
> both
>   the kernel and the user address spaces, there is no way to tell by
>   looking at it. The config option related to this property is
>   ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE.
> 
>   Also, unfortunately, 0 is a valid address in the s390 kernel
> address
>   space.
> 
> Fix the issue by using -4096 as the bad address for bits iterator, as
> suggested by Ilya. Verify that bpf_iter_bits_new() returns -EINVAL
> for
> NULL address and -EFAULT for bad address.

The code uses -4095, which I think is better, since it's the current
value of MAX_ERRNO, therefore, IS_ERR_VALUE() sees it as an error. It's
also not aligned, which may be an additional reason it may not be
dereferenceable on some CPUs.

Other than this discrepancy in the commit message:

Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>

> Fixes: ebafc1e535db ("selftests/bpf: Add three test cases for
> bits_iter")
> Reported-by: Byeonguk Jeong <jungbu2855@gmail.com>
> Closes: https://lore.kernel.org/bpf/ZycSXwjH4UTvx-Cn@ub22/
> Signed-off-by: Hou Tao <houtao1@huawei.com>
> ---
>  .../selftests/bpf/progs/verifier_bits_iter.c  | 32 ++++++++++++++++-
> --
>  1 file changed, 28 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
> b/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
> index 156cc278e2fc..7c881bca9af5 100644
> --- a/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
> +++ b/tools/testing/selftests/bpf/progs/verifier_bits_iter.c
> @@ -57,9 +57,15 @@ __description("null pointer")
>  __success __retval(0)
>  int null_pointer(void)
>  {
> -	int nr = 0;
> +	struct bpf_iter_bits iter;
> +	int err, nr = 0;
>  	int *bit;
>  
> +	err = bpf_iter_bits_new(&iter, NULL, 1);
> +	bpf_iter_bits_destroy(&iter);
> +	if (err != -EINVAL)
> +		return 1;
> +
>  	bpf_for_each(bits, bit, NULL, 1)
>  		nr++;
>  	return nr;
> @@ -194,15 +200,33 @@ __description("bad words")
>  __success __retval(0)
>  int bad_words(void)
>  {
> -	void *bad_addr = (void *)(3UL << 30);
> -	int nr = 0;
> +	void *bad_addr = (void *)-4095;
> +	struct bpf_iter_bits iter;
> +	volatile int nr;
>  	int *bit;
> +	int err;
> +
> +	err = bpf_iter_bits_new(&iter, bad_addr, 1);
> +	bpf_iter_bits_destroy(&iter);
> +	if (err != -EFAULT)
> +		return 1;
>  
> +	nr = 0;
>  	bpf_for_each(bits, bit, bad_addr, 1)
>  		nr++;
> +	if (nr != 0)
> +		return 2;
>  
> +	err = bpf_iter_bits_new(&iter, bad_addr, 4);
> +	bpf_iter_bits_destroy(&iter);
> +	if (err != -EFAULT)
> +		return 3;
> +
> +	nr = 0;
>  	bpf_for_each(bits, bit, bad_addr, 4)
>  		nr++;
> +	if (nr != 0)
> +		return 4;
>  
> -	return nr;
> +	return 0;
>  }

Re: [PATCH bpf] selftests/bpf: Use -4095 as the bad address for bits iterator

[Alexei Starovoitov]

On Tue, Nov 5, 2024 at 12:18 PM Ilya Leoshkevich <iii@linux.ibm.com> wrote:
>
> On Tue, 2024-11-05 at 12:30 +0800, Hou Tao wrote:
> > From: Hou Tao <houtao1@huawei.com>
> >
> > As reported by Byeonguk, the bad_words test in verifier_bits_iter.c
> > occasionally fails on s390 host. Quoting Ilya's explanation:
> >
> >   s390 kernel runs in a completely separate address space, there is
> > no
> >   user/kernel split at TASK_SIZE. The same address may be valid in
> > both
> >   the kernel and the user address spaces, there is no way to tell by
> >   looking at it. The config option related to this property is
> >   ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE.
> >
> >   Also, unfortunately, 0 is a valid address in the s390 kernel
> > address
> >   space.
> >
> > Fix the issue by using -4096 as the bad address for bits iterator, as
> > suggested by Ilya. Verify that bpf_iter_bits_new() returns -EINVAL
> > for
> > NULL address and -EFAULT for bad address.
>
> The code uses -4095, which I think is better, since it's the current
> value of MAX_ERRNO, therefore, IS_ERR_VALUE() sees it as an error. It's
> also not aligned, which may be an additional reason it may not be
> dereferenceable on some CPUs.
>
> Other than this discrepancy in the commit message:
>
> Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>

Fixed it up while applying.

Re: [PATCH bpf] selftests/bpf: Use -4095 as the bad address for bits iterator

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Tue,  5 Nov 2024 12:30:57 +0800 you wrote:
> From: Hou Tao <houtao1@huawei.com>
> 
> As reported by Byeonguk, the bad_words test in verifier_bits_iter.c
> occasionally fails on s390 host. Quoting Ilya's explanation:
> 
>   s390 kernel runs in a completely separate address space, there is no
>   user/kernel split at TASK_SIZE. The same address may be valid in both
>   the kernel and the user address spaces, there is no way to tell by
>   looking at it. The config option related to this property is
>   ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE.
> 
> [...]

Here is the summary with links:
  - [bpf] selftests/bpf: Use -4095 as the bad address for bits iterator
    https://git.kernel.org/bpf/bpf/c/6801cf7890f2

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html