[PATCH bpf-next v2 1/2] bpf: Unify data extension operation of jited_ksyms and jited_linfo

[Pu Lehui]

On 2022/5/7 4:52, John Fastabend wrote:
> Pu Lehui wrote:
>> We found that 32-bit environment can not print bpf line info due
>> to data inconsistency between jited_ksyms[0] and jited_linfo[0].
>>
>> For example:
>> jited_kyms[0] = 0xb800067c, jited_linfo[0] = 0xffffffffb800067c
>>
>> We know that both of them store bpf func address, but due to the
>> different data extension operations when extended to u64, they may
>> not be the same. We need to unify the data extension operations of
>> them.
>>
>> Signed-off-by: Pu Lehui <pulehui at huawei.com>
>> ---
>>   kernel/bpf/syscall.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
>> index e9e3e49c0eb7..18137ea5190d 100644
>> --- a/kernel/bpf/syscall.c
>> +++ b/kernel/bpf/syscall.c
>> @@ -3871,13 +3871,16 @@ static int bpf_prog_get_info_by_fd(struct file *file,
>>   		info.nr_jited_line_info = 0;
>>   	if (info.nr_jited_line_info && ulen) {
>>   		if (bpf_dump_raw_ok(file->f_cred)) {
>> +			unsigned long jited_linfo_addr;
>>   			__u64 __user *user_linfo;
>>   			u32 i;
>>   
>>   			user_linfo = u64_to_user_ptr(info.jited_line_info);
>>   			ulen = min_t(u32, info.nr_jited_line_info, ulen);
>>   			for (i = 0; i < ulen; i++) {
>> -				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
>> +				jited_linfo_addr = (unsigned long)
>> +					prog->aux->jited_linfo[i];
>> +				if (put_user((__u64) jited_linfo_addr,
>>   					     &user_linfo[i]))
> 
> the logic is fine but i'm going to nitpick a bit this 4 lines is ugly
> just make it slightly longer than 80chars or use a shoarter name? For
> example,
> 
> 			for (i = 0; i < ulen; i++) {
> 				unsigned long l;
> 
> 				l = (unsigned long) prog->aux->jited_linfo[i];
> 				if (put_user((__u64) l, &user_linfo[i]))
> 
> is much nicer -- no reason to smash single assignment across multiple
> lines. My $.02.
> 

Okay, It sounds good. I will make change in next version. Thanks.

> Thanks,
> John
> .
>