Re: [PATCH bpf-next v13 1/8] bpf: Extend BPF syscall with common attributes support

["Alexei Starovoitov" <alexei.starovoitov@gmail.com>]

On Mon May 11, 2026 at 9:19 AM PDT, bot+bpf-ci wrote:
>> diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
>> index f5639d5ac331..50055ab73649 100644
>> --- a/include/linux/syscalls.h
>> +++ b/include/linux/syscalls.h
>> @@ -936,7 +936,8 @@ asmlinkage long sys_seccomp(unsigned int op, unsigned int flags,
>>  asmlinkage long sys_getrandom(char __user *buf, size_t count,
>>  			      unsigned int flags);
>>  asmlinkage long sys_memfd_create(const char __user *uname_ptr, unsigned int flags);
>> -asmlinkage long sys_bpf(int cmd, union bpf_attr __user *attr, unsigned int size);
>> +asmlinkage long sys_bpf(int cmd, union bpf_attr __user *attr, unsigned int size,
>> +			struct bpf_common_attr __user *attr_common, unsigned int size_common);
>>  asmlinkage long sys_execveat(int dfd, const char __user *filename,
>>  			const char __user *const __user *argv,
>>  			const char __user *const __user *envp, int flags);
>
> [ ... ]
>
>> diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
>> index 677be9a47347..16ff0968fc21 100644
>> --- a/tools/include/uapi/linux/bpf.h
>> +++ b/tools/include/uapi/linux/bpf.h
>> @@ -994,6 +994,7 @@ enum bpf_cmd {
>>  	BPF_PROG_STREAM_READ_BY_FD,
>>  	BPF_PROG_ASSOC_STRUCT_OPS,
>>  	__MAX_BPF_CMD,
>> +	BPF_COMMON_ATTRS = 1 << 16, /* Indicate carrying syscall common attrs. */
>>  };
>
> [ ... ]
>
>> @@ -1500,6 +1501,13 @@ struct bpf_stack_build_id {
>>  	};
>>  };
>>
>> +struct bpf_common_attr {
>> +	__u64 log_buf;
>> +	__u32 log_size;
>> +	__u32 log_level;
>> +	__u32 log_true_size;
>> +};
>> +
>
> Should log_buf use __aligned_u64 instead of plain __u64?
>
> Looking at other pointer-bearing fields in the BPF UAPI, they consistently
> use __aligned_u64. For example, the existing BPF_PROG_LOAD log_buf in union
> bpf_attr is declared as:
>
>     __aligned_u64 log_buf;  /* user supplied buffer */
>
> The __aligned_u64 attribute forces 8-byte alignment on all architectures,
> including 32-bit systems where native unsigned long long alignment is only
> 4 bytes. Without this attribute, struct bpf_common_attr will have different
> layouts between 32-bit and 64-bit ABIs:
>
> - 32-bit: 20 bytes, 4-byte aligned
> - 64-bit: 24 bytes (4 bytes of tail padding), 8-byte aligned
>
> While the current __sys_bpf() implementation with bpf_check_uarg_tail_zero()
> and min_t() capped copying handles this gracefully, deviating from the
> established BPF UAPI convention could create problems for future extensions.
> If another __u64 field is added later, the inter-field offsets would diverge
> between ABIs.
>
> Would it be better to use __aligned_u64 log_buf to match union bpf_attr and
> preserve identical struct layouts across 32/64-bit ABIs?

bot has a point.

pw-bot: cr