Messages in this thread | | | Subject | Re: [PATCH bpf-next v2 1/2] bpf: Unify data extension operation of jited_ksyms and jited_linfo | From | Pu Lehui <> | Date | Sat, 7 May 2022 08:51:00 +0800 |
| |
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@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 > . >
| |