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]))
return -EFAULT;
}
--
2.25.1
