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

From: John Fastabend
Date: Fri May 06 2022 - 16:52:33 EST


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@xxxxxxxxxx>
> ---
> 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.

Thanks,
John