[PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Lorenzo Bianconi]

Update test_xdp_update_frags adding a test for a buffer size
set to (MAX_SKB_FRAGS + 2) * PAGE_SIZE. The kernel is supposed
to return -ENOMEM.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 .../bpf/prog_tests/xdp_adjust_frags.c         | 37 ++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
index 134d0ac32f59..61d5b585eb15 100644
--- a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
+++ b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
@@ -5,11 +5,12 @@
 void test_xdp_update_frags(void)
 {
 	const char *file = "./test_xdp_update_frags.o";
+	int err, prog_fd, max_skb_frags, buf_size, num;
 	struct bpf_program *prog;
 	struct bpf_object *obj;
-	int err, prog_fd;
 	__u32 *offset;
 	__u8 *buf;
+	FILE *f;
 	LIBBPF_OPTS(bpf_test_run_opts, topts);
 
 	obj = bpf_object__open(file);
@@ -99,6 +100,40 @@ void test_xdp_update_frags(void)
 	ASSERT_EQ(buf[7621], 0xbb, "xdp_update_frag buf[7621]");
 
 	free(buf);
+
+	/* test_xdp_update_frags: unsupported buffer size */
+	f = fopen("/proc/sys/net/core/max_skb_frags", "r");
+	if (!ASSERT_OK_PTR(f, "max_skb_frag file pointer"))
+		goto out;
+
+	num = fscanf(f, "%d", &max_skb_frags);
+	fclose(f);
+
+	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
+		goto out;
+
+	/* xdp_buff linear area size is always set to 4096 in the
+	 * bpf_prog_test_run_xdp routine.
+	 */
+	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
+	buf = malloc(buf_size);
+	if (!ASSERT_OK_PTR(buf, "alloc buf"))
+		goto out;
+
+	memset(buf, 0, buf_size);
+	offset = (__u32 *)buf;
+	*offset = 16;
+	buf[*offset] = 0xaa;
+	buf[*offset + 15] = 0xaa;
+
+	topts.data_in = buf;
+	topts.data_out = buf;
+	topts.data_size_in = buf_size;
+	topts.data_size_out = buf_size;
+
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
+	free(buf);
 out:
 	bpf_object__close(obj);
 }
-- 
2.34.1

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Yonghong Song]

On 2/4/22 5:58 AM, Lorenzo Bianconi wrote:
> Update test_xdp_update_frags adding a test for a buffer size
> set to (MAX_SKB_FRAGS + 2) * PAGE_SIZE. The kernel is supposed
> to return -ENOMEM.
> 
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
>   .../bpf/prog_tests/xdp_adjust_frags.c         | 37 ++++++++++++++++++-
>   1 file changed, 36 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> index 134d0ac32f59..61d5b585eb15 100644
> --- a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> +++ b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> @@ -5,11 +5,12 @@
>   void test_xdp_update_frags(void)
>   {
>   	const char *file = "./test_xdp_update_frags.o";
> +	int err, prog_fd, max_skb_frags, buf_size, num;
>   	struct bpf_program *prog;
>   	struct bpf_object *obj;
> -	int err, prog_fd;
>   	__u32 *offset;
>   	__u8 *buf;
> +	FILE *f;
>   	LIBBPF_OPTS(bpf_test_run_opts, topts);
>   
>   	obj = bpf_object__open(file);
> @@ -99,6 +100,40 @@ void test_xdp_update_frags(void)
>   	ASSERT_EQ(buf[7621], 0xbb, "xdp_update_frag buf[7621]");
>   
>   	free(buf);
> +
> +	/* test_xdp_update_frags: unsupported buffer size */
> +	f = fopen("/proc/sys/net/core/max_skb_frags", "r");
> +	if (!ASSERT_OK_PTR(f, "max_skb_frag file pointer"))
> +		goto out;

In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
but if /proc/sys/net/core/max_skb_flags is 2 or more less
than MAX_SKB_FRAGS, the test won't fail, right?

> +
> +	num = fscanf(f, "%d", &max_skb_frags);
> +	fclose(f);
> +
> +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
> +		goto out;
> +
> +	/* xdp_buff linear area size is always set to 4096 in the
> +	 * bpf_prog_test_run_xdp routine.
> +	 */
> +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
> +	buf = malloc(buf_size);
> +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
> +		goto out;
> +
> +	memset(buf, 0, buf_size);
> +	offset = (__u32 *)buf;
> +	*offset = 16;
> +	buf[*offset] = 0xaa;
> +	buf[*offset + 15] = 0xaa;
> +
> +	topts.data_in = buf;
> +	topts.data_out = buf;
> +	topts.data_size_in = buf_size;
> +	topts.data_size_out = buf_size;
> +
> +	err = bpf_prog_test_run_opts(prog_fd, &topts);
> +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> +	free(buf);
>   out:
>   	bpf_object__close(obj);
>   }

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 2859 bytes --]

> 
> 
> On 2/4/22 5:58 AM, Lorenzo Bianconi wrote:
> > Update test_xdp_update_frags adding a test for a buffer size
> > set to (MAX_SKB_FRAGS + 2) * PAGE_SIZE. The kernel is supposed
> > to return -ENOMEM.
> > 
> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > ---
> >   .../bpf/prog_tests/xdp_adjust_frags.c         | 37 ++++++++++++++++++-
> >   1 file changed, 36 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> > index 134d0ac32f59..61d5b585eb15 100644
> > --- a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> > +++ b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
> > @@ -5,11 +5,12 @@
> >   void test_xdp_update_frags(void)
> >   {
> >   	const char *file = "./test_xdp_update_frags.o";
> > +	int err, prog_fd, max_skb_frags, buf_size, num;
> >   	struct bpf_program *prog;
> >   	struct bpf_object *obj;
> > -	int err, prog_fd;
> >   	__u32 *offset;
> >   	__u8 *buf;
> > +	FILE *f;
> >   	LIBBPF_OPTS(bpf_test_run_opts, topts);
> >   	obj = bpf_object__open(file);
> > @@ -99,6 +100,40 @@ void test_xdp_update_frags(void)
> >   	ASSERT_EQ(buf[7621], 0xbb, "xdp_update_frag buf[7621]");
> >   	free(buf);
> > +
> > +	/* test_xdp_update_frags: unsupported buffer size */
> > +	f = fopen("/proc/sys/net/core/max_skb_frags", "r");
> > +	if (!ASSERT_OK_PTR(f, "max_skb_frag file pointer"))
> > +		goto out;
> 
> In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
> but if /proc/sys/net/core/max_skb_flags is 2 or more less
> than MAX_SKB_FRAGS, the test won't fail, right?

yes, you are right. Should we use the same definition used in
include/linux/skbuff.h instead? Something like:

if (65536 / page_size + 1 < 16)
	max_skb_flags = 16;
else
	max_skb_flags = 65536/page_size + 1;

Regards,
Lorenzo

> 
> > +
> > +	num = fscanf(f, "%d", &max_skb_frags);
> > +	fclose(f);
> > +
> > +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
> > +		goto out;
> > +
> > +	/* xdp_buff linear area size is always set to 4096 in the
> > +	 * bpf_prog_test_run_xdp routine.
> > +	 */
> > +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
> > +	buf = malloc(buf_size);
> > +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
> > +		goto out;
> > +
> > +	memset(buf, 0, buf_size);
> > +	offset = (__u32 *)buf;
> > +	*offset = 16;
> > +	buf[*offset] = 0xaa;
> > +	buf[*offset + 15] = 0xaa;
> > +
> > +	topts.data_in = buf;
> > +	topts.data_out = buf;
> > +	topts.data_size_in = buf_size;
> > +	topts.data_size_out = buf_size;
> > +
> > +	err = bpf_prog_test_run_opts(prog_fd, &topts);
> > +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> > +	free(buf);
> >   out:
> >   	bpf_object__close(obj);
> >   }
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Yonghong Song]

On 2/4/22 9:52 AM, Lorenzo Bianconi wrote:
>>
>>
>> On 2/4/22 5:58 AM, Lorenzo Bianconi wrote:
>>> Update test_xdp_update_frags adding a test for a buffer size
>>> set to (MAX_SKB_FRAGS + 2) * PAGE_SIZE. The kernel is supposed
>>> to return -ENOMEM.
>>>
>>> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
>>> ---
>>>    .../bpf/prog_tests/xdp_adjust_frags.c         | 37 ++++++++++++++++++-
>>>    1 file changed, 36 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
>>> index 134d0ac32f59..61d5b585eb15 100644
>>> --- a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
>>> +++ b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_frags.c
>>> @@ -5,11 +5,12 @@
>>>    void test_xdp_update_frags(void)
>>>    {
>>>    	const char *file = "./test_xdp_update_frags.o";
>>> +	int err, prog_fd, max_skb_frags, buf_size, num;
>>>    	struct bpf_program *prog;
>>>    	struct bpf_object *obj;
>>> -	int err, prog_fd;
>>>    	__u32 *offset;
>>>    	__u8 *buf;
>>> +	FILE *f;
>>>    	LIBBPF_OPTS(bpf_test_run_opts, topts);
>>>    	obj = bpf_object__open(file);
>>> @@ -99,6 +100,40 @@ void test_xdp_update_frags(void)
>>>    	ASSERT_EQ(buf[7621], 0xbb, "xdp_update_frag buf[7621]");
>>>    	free(buf);
>>> +
>>> +	/* test_xdp_update_frags: unsupported buffer size */
>>> +	f = fopen("/proc/sys/net/core/max_skb_frags", "r");
>>> +	if (!ASSERT_OK_PTR(f, "max_skb_frag file pointer"))
>>> +		goto out;
>>
>> In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
>> but if /proc/sys/net/core/max_skb_flags is 2 or more less
>> than MAX_SKB_FRAGS, the test won't fail, right?
> 
> yes, you are right. Should we use the same definition used in
> include/linux/skbuff.h instead? Something like:
> 
> if (65536 / page_size + 1 < 16)
> 	max_skb_flags = 16;
> else
> 	max_skb_flags = 65536/page_size + 1;

The maximum packet size limit 64KB won't change anytime soon.
So the above should work. Some comments to explain why using
the above formula will be good.

> 
> Regards,
> Lorenzo
> 
>>
>>> +
>>> +	num = fscanf(f, "%d", &max_skb_frags);
>>> +	fclose(f);
>>> +
>>> +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
>>> +		goto out;
>>> +
>>> +	/* xdp_buff linear area size is always set to 4096 in the
>>> +	 * bpf_prog_test_run_xdp routine.
>>> +	 */
>>> +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
>>> +	buf = malloc(buf_size);
>>> +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
>>> +		goto out;
>>> +
>>> +	memset(buf, 0, buf_size);
>>> +	offset = (__u32 *)buf;
>>> +	*offset = 16;
>>> +	buf[*offset] = 0xaa;
>>> +	buf[*offset + 15] = 0xaa;
>>> +
>>> +	topts.data_in = buf;
>>> +	topts.data_out = buf;
>>> +	topts.data_size_in = buf_size;
>>> +	topts.data_size_out = buf_size;
>>> +
>>> +	err = bpf_prog_test_run_opts(prog_fd, &topts);
>>> +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
>>> +	free(buf);
>>>    out:
>>>    	bpf_object__close(obj);
>>>    }
>>

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 1796 bytes --]

> 

[...]

> > > 
> > > In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
> > > but if /proc/sys/net/core/max_skb_flags is 2 or more less
> > > than MAX_SKB_FRAGS, the test won't fail, right?
> > 
> > yes, you are right. Should we use the same definition used in
> > include/linux/skbuff.h instead? Something like:
> > 
> > if (65536 / page_size + 1 < 16)
> > 	max_skb_flags = 16;
> > else
> > 	max_skb_flags = 65536/page_size + 1;
> 
> The maximum packet size limit 64KB won't change anytime soon.
> So the above should work. Some comments to explain why using
> the above formula will be good.

ack, I will do in v2.

Regards,
Lorenzo

> 
> > 
> > Regards,
> > Lorenzo
> > 
> > > 
> > > > +
> > > > +	num = fscanf(f, "%d", &max_skb_frags);
> > > > +	fclose(f);
> > > > +
> > > > +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
> > > > +		goto out;
> > > > +
> > > > +	/* xdp_buff linear area size is always set to 4096 in the
> > > > +	 * bpf_prog_test_run_xdp routine.
> > > > +	 */
> > > > +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
> > > > +	buf = malloc(buf_size);
> > > > +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
> > > > +		goto out;
> > > > +
> > > > +	memset(buf, 0, buf_size);
> > > > +	offset = (__u32 *)buf;
> > > > +	*offset = 16;
> > > > +	buf[*offset] = 0xaa;
> > > > +	buf[*offset + 15] = 0xaa;
> > > > +
> > > > +	topts.data_in = buf;
> > > > +	topts.data_out = buf;
> > > > +	topts.data_size_in = buf_size;
> > > > +	topts.data_size_out = buf_size;
> > > > +
> > > > +	err = bpf_prog_test_run_opts(prog_fd, &topts);
> > > > +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> > > > +	free(buf);
> > > >    out:
> > > >    	bpf_object__close(obj);
> > > >    }
> > > 
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 2239 bytes --]

> > 
> 
> [...]
> 
> > > > 
> > > > In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
> > > > but if /proc/sys/net/core/max_skb_flags is 2 or more less
> > > > than MAX_SKB_FRAGS, the test won't fail, right?
> > > 
> > > yes, you are right. Should we use the same definition used in
> > > include/linux/skbuff.h instead? Something like:
> > > 
> > > if (65536 / page_size + 1 < 16)
> > > 	max_skb_flags = 16;
> > > else
> > > 	max_skb_flags = 65536/page_size + 1;
> > 
> > The maximum packet size limit 64KB won't change anytime soon.
> > So the above should work. Some comments to explain why using
> > the above formula will be good.
> 
> ack, I will do in v2.

I can see there is a on-going discussion here [0] about increasing
MAX_SKB_FRAGS. I guess we can put on-hold this patch and see how
MAX_SKB_FRAGS will be changed.

Regards,
Lorenzo

[0] https://lore.kernel.org/all/202202031315.B425Ipe8-lkp@intel.com/t/#ma1b2c7e71fe9bc69e24642a62dadf32fda7d5f03

> 
> Regards,
> Lorenzo
> 
> > 
> > > 
> > > Regards,
> > > Lorenzo
> > > 
> > > > 
> > > > > +
> > > > > +	num = fscanf(f, "%d", &max_skb_frags);
> > > > > +	fclose(f);
> > > > > +
> > > > > +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
> > > > > +		goto out;
> > > > > +
> > > > > +	/* xdp_buff linear area size is always set to 4096 in the
> > > > > +	 * bpf_prog_test_run_xdp routine.
> > > > > +	 */
> > > > > +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
> > > > > +	buf = malloc(buf_size);
> > > > > +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
> > > > > +		goto out;
> > > > > +
> > > > > +	memset(buf, 0, buf_size);
> > > > > +	offset = (__u32 *)buf;
> > > > > +	*offset = 16;
> > > > > +	buf[*offset] = 0xaa;
> > > > > +	buf[*offset + 15] = 0xaa;
> > > > > +
> > > > > +	topts.data_in = buf;
> > > > > +	topts.data_out = buf;
> > > > > +	topts.data_size_in = buf_size;
> > > > > +	topts.data_size_out = buf_size;
> > > > > +
> > > > > +	err = bpf_prog_test_run_opts(prog_fd, &topts);
> > > > > +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> > > > > +	free(buf);
> > > > >    out:
> > > > >    	bpf_object__close(obj);
> > > > >    }
> > > > 
> > 



[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Yonghong Song]

On 2/4/22 11:08 AM, Lorenzo Bianconi wrote:
>>>
>>
>> [...]
>>
>>>>>
>>>>> In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
>>>>> but if /proc/sys/net/core/max_skb_flags is 2 or more less
>>>>> than MAX_SKB_FRAGS, the test won't fail, right?
>>>>
>>>> yes, you are right. Should we use the same definition used in
>>>> include/linux/skbuff.h instead? Something like:
>>>>
>>>> if (65536 / page_size + 1 < 16)
>>>> 	max_skb_flags = 16;
>>>> else
>>>> 	max_skb_flags = 65536/page_size + 1;
>>>
>>> The maximum packet size limit 64KB won't change anytime soon.
>>> So the above should work. Some comments to explain why using
>>> the above formula will be good.
>>
>> ack, I will do in v2.
> 
> I can see there is a on-going discussion here [0] about increasing
> MAX_SKB_FRAGS. I guess we can put on-hold this patch and see how
> MAX_SKB_FRAGS will be changed.

Thanks for the link. The new patch is going to increase
MAX_SKB_FRAGS and it is possible that will be changed again
(maybe under some config options).

The default value for
/proc/sys/net/core/max_skb_flags is MAX_SKB_FRAGS and I suspect
anybody is bothering to change it. So your patch is okay to me.
Maybe change a little bit -ENOMEM error message. current,
   ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
to
   ASSERT_EQ(err, -ENOMEM, "unsupported buffer size, possible 
non-default /proc/sys/net/core/max_skb_flags?");

> 
> Regards,
> Lorenzo
> 
> [0] https://lore.kernel.org/all/202202031315.B425Ipe8-lkp@intel.com/t/#ma1b2c7e71fe9bc69e24642a62dadf32fda7d5f03
> 
>>
>> Regards,
>> Lorenzo
>>
>>>
>>>>
>>>> Regards,
>>>> Lorenzo
>>>>
>>>>>
>>>>>> +
>>>>>> +	num = fscanf(f, "%d", &max_skb_frags);
>>>>>> +	fclose(f);
>>>>>> +
>>>>>> +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
>>>>>> +		goto out;
>>>>>> +
>>>>>> +	/* xdp_buff linear area size is always set to 4096 in the
>>>>>> +	 * bpf_prog_test_run_xdp routine.
>>>>>> +	 */
>>>>>> +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
>>>>>> +	buf = malloc(buf_size);
>>>>>> +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
>>>>>> +		goto out;
>>>>>> +
>>>>>> +	memset(buf, 0, buf_size);
>>>>>> +	offset = (__u32 *)buf;
>>>>>> +	*offset = 16;
>>>>>> +	buf[*offset] = 0xaa;
>>>>>> +	buf[*offset + 15] = 0xaa;
>>>>>> +
>>>>>> +	topts.data_in = buf;
>>>>>> +	topts.data_out = buf;
>>>>>> +	topts.data_size_in = buf_size;
>>>>>> +	topts.data_size_out = buf_size;
>>>>>> +
>>>>>> +	err = bpf_prog_test_run_opts(prog_fd, &topts);
>>>>>> +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
>>>>>> +	free(buf);
>>>>>>     out:
>>>>>>     	bpf_object__close(obj);
>>>>>>     }
>>>>>
>>>
> 
> 

Re: [PATCH bpf-next] selftest/bpf: check invalid length in test_xdp_update_frags

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 3268 bytes --]

> 
> 
> On 2/4/22 11:08 AM, Lorenzo Bianconi wrote:
> > > > 
> > > 
> > > [...]
> > > 
> > > > > > 
> > > > > > In kernel, the nr_frags checking is against MAX_SKB_FRAGS,
> > > > > > but if /proc/sys/net/core/max_skb_flags is 2 or more less
> > > > > > than MAX_SKB_FRAGS, the test won't fail, right?
> > > > > 
> > > > > yes, you are right. Should we use the same definition used in
> > > > > include/linux/skbuff.h instead? Something like:
> > > > > 
> > > > > if (65536 / page_size + 1 < 16)
> > > > > 	max_skb_flags = 16;
> > > > > else
> > > > > 	max_skb_flags = 65536/page_size + 1;
> > > > 
> > > > The maximum packet size limit 64KB won't change anytime soon.
> > > > So the above should work. Some comments to explain why using
> > > > the above formula will be good.
> > > 
> > > ack, I will do in v2.
> > 
> > I can see there is a on-going discussion here [0] about increasing
> > MAX_SKB_FRAGS. I guess we can put on-hold this patch and see how
> > MAX_SKB_FRAGS will be changed.
> 
> Thanks for the link. The new patch is going to increase
> MAX_SKB_FRAGS and it is possible that will be changed again
> (maybe under some config options).
> 
> The default value for
> /proc/sys/net/core/max_skb_flags is MAX_SKB_FRAGS and I suspect
> anybody is bothering to change it. So your patch is okay to me.
> Maybe change a little bit -ENOMEM error message. current,
>   ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> to
>   ASSERT_EQ(err, -ENOMEM, "unsupported buffer size, possible non-default
> /proc/sys/net/core/max_skb_flags?");

ack, I am fine with it.
@Alexei, Andrii: any hints about it?

Regards,
Lorenzo

> 
> > 
> > Regards,
> > Lorenzo
> > 
> > [0] https://lore.kernel.org/all/202202031315.B425Ipe8-lkp@intel.com/t/#ma1b2c7e71fe9bc69e24642a62dadf32fda7d5f03
> > 
> > > 
> > > Regards,
> > > Lorenzo
> > > 
> > > > 
> > > > > 
> > > > > Regards,
> > > > > Lorenzo
> > > > > 
> > > > > > 
> > > > > > > +
> > > > > > > +	num = fscanf(f, "%d", &max_skb_frags);
> > > > > > > +	fclose(f);
> > > > > > > +
> > > > > > > +	if (!ASSERT_EQ(num, 1, "max_skb_frags read failed"))
> > > > > > > +		goto out;
> > > > > > > +
> > > > > > > +	/* xdp_buff linear area size is always set to 4096 in the
> > > > > > > +	 * bpf_prog_test_run_xdp routine.
> > > > > > > +	 */
> > > > > > > +	buf_size = 4096 + (max_skb_frags + 1) * sysconf(_SC_PAGE_SIZE);
> > > > > > > +	buf = malloc(buf_size);
> > > > > > > +	if (!ASSERT_OK_PTR(buf, "alloc buf"))
> > > > > > > +		goto out;
> > > > > > > +
> > > > > > > +	memset(buf, 0, buf_size);
> > > > > > > +	offset = (__u32 *)buf;
> > > > > > > +	*offset = 16;
> > > > > > > +	buf[*offset] = 0xaa;
> > > > > > > +	buf[*offset + 15] = 0xaa;
> > > > > > > +
> > > > > > > +	topts.data_in = buf;
> > > > > > > +	topts.data_out = buf;
> > > > > > > +	topts.data_size_in = buf_size;
> > > > > > > +	topts.data_size_out = buf_size;
> > > > > > > +
> > > > > > > +	err = bpf_prog_test_run_opts(prog_fd, &topts);
> > > > > > > +	ASSERT_EQ(err, -ENOMEM, "unsupported buffer size");
> > > > > > > +	free(buf);
> > > > > > >     out:
> > > > > > >     	bpf_object__close(obj);
> > > > > > >     }
> > > > > > 
> > > > 
> > 
> > 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]