[PATCH bpf v2 0/1] bpf: btf: Fix endianness of bitfields

[PATCH bpf v2 0/1] bpf: btf: Fix endianness of bitfields

[Okash Khawaja]

Hi,

Following are the changes from v1:

- use compile time check for endianness
- factor out right_shift_bits from the check as it is same in both
- change size of bit shift variables to match length of bit offset in
BTF specification

Please note that I have considered non-big endians to be little endians
for our purposes. Let me know if you think that's a problem. We need to
then decide what to do in case neither of macros are defined.

Thanks,
Okash

[PATCH bpf v2 1/1] bpf: btf: Fix bitfield extraction for big endian

[Okash Khawaja]

[-- Attachment #1: fix-btf-bitfields.patch --]
[-- Type: text/plain, Size: 2222 bytes --]

When extracting bitfield from a number, btf_int_bits_seq_show() builds
a mask and accesses least significant byte of the number in a way
specific to little-endian. This patch fixes that by checking endianness
of the machine and then shifting left and right the unneeded bits.

Thanks to Martin Lau for the help in navigating potential pitfalls when
dealing with endianess and for the final solution.

Fixes: b00b8daec828 ("bpf: btf: Add pretty print capability for data with BTF type info")
Signed-off-by: Okash Khawaja <osk@fb.com>

---
 kernel/bpf/btf.c |   30 +++++++++++++-----------------
 1 file changed, 13 insertions(+), 17 deletions(-)

--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -991,16 +991,13 @@ static void btf_int_bits_seq_show(const
 				  void *data, u8 bits_offset,
 				  struct seq_file *m)
 {
+	u16 left_shift_bits, right_shift_bits;
 	u32 int_data = btf_type_int(t);
 	u16 nr_bits = BTF_INT_BITS(int_data);
 	u16 total_bits_offset;
 	u16 nr_copy_bytes;
 	u16 nr_copy_bits;
-	u8 nr_upper_bits;
-	union {
-		u64 u64_num;
-		u8  u8_nums[8];
-	} print_num;
+	u64 print_num;
 
 	total_bits_offset = bits_offset + BTF_INT_OFFSET(int_data);
 	data += BITS_ROUNDDOWN_BYTES(total_bits_offset);
@@ -1008,21 +1005,20 @@ static void btf_int_bits_seq_show(const
 	nr_copy_bits = nr_bits + bits_offset;
 	nr_copy_bytes = BITS_ROUNDUP_BYTES(nr_copy_bits);
 
-	print_num.u64_num = 0;
-	memcpy(&print_num.u64_num, data, nr_copy_bytes);
+	print_num = 0;
+	memcpy(&print_num, data, nr_copy_bytes);
 
-	/* Ditch the higher order bits */
-	nr_upper_bits = BITS_PER_BYTE_MASKED(nr_copy_bits);
-	if (nr_upper_bits) {
-		/* We need to mask out some bits of the upper byte. */
-		u8 mask = (1 << nr_upper_bits) - 1;
+#ifdef __BIG_ENDIAN_BITFIELD
+	left_shift_bits = bits_offset;
+#else
+	left_shift_bits = BITS_PER_U64 - nr_copy_bits;
+#endif
+	right_shift_bits = BITS_PER_U64 - nr_bits;
 
-		print_num.u8_nums[nr_copy_bytes - 1] &= mask;
-	}
+	print_num <<= left_shift_bits;
+	print_num >>= right_shift_bits;
 
-	print_num.u64_num >>= bits_offset;
-
-	seq_printf(m, "0x%llx", print_num.u64_num);
+	seq_printf(m, "0x%llx", print_num);
 }
 
 static void btf_int_seq_show(const struct btf *btf, const struct btf_type *t,

Re: [PATCH bpf v2 1/1] bpf: btf: Fix bitfield extraction for big endian

[Martin KaFai Lau]

On Tue, Jul 10, 2018 at 02:33:07PM -0700, Okash Khawaja wrote:
> When extracting bitfield from a number, btf_int_bits_seq_show() builds
> a mask and accesses least significant byte of the number in a way
> specific to little-endian. This patch fixes that by checking endianness
> of the machine and then shifting left and right the unneeded bits.
> 
> Thanks to Martin Lau for the help in navigating potential pitfalls when
> dealing with endianess and for the final solution.
> 
> Fixes: b00b8daec828 ("bpf: btf: Add pretty print capability for data with BTF type info")
> Signed-off-by: Okash Khawaja <osk@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>

Re: [PATCH bpf v2 1/1] bpf: btf: Fix bitfield extraction for big endian

[Daniel Borkmann]

On 07/11/2018 01:46 AM, Martin KaFai Lau wrote:
> On Tue, Jul 10, 2018 at 02:33:07PM -0700, Okash Khawaja wrote:
>> When extracting bitfield from a number, btf_int_bits_seq_show() builds
>> a mask and accesses least significant byte of the number in a way
>> specific to little-endian. This patch fixes that by checking endianness
>> of the machine and then shifting left and right the unneeded bits.
>>
>> Thanks to Martin Lau for the help in navigating potential pitfalls when
>> dealing with endianess and for the final solution.
>>
>> Fixes: b00b8daec828 ("bpf: btf: Add pretty print capability for data with BTF type info")
>> Signed-off-by: Okash Khawaja <osk@fb.com>
> Acked-by: Martin KaFai Lau <kafai@fb.com>

Applied to bpf, thanks Okash!