On Wed, Jul 16, 2025 at 09:05:05AM -0700, Yonghong Song wrote:
On 7/16/25 3:13 AM, Shung-Hsi Yu wrote:
Hi Andrii and Yonghong,
On Fri, May 23, 2025 at 09:13:40PM -0700, Yonghong Song wrote:
Add two tests:
- one test has 'rX <op> r10' where rX is not r10, and
- another test has 'rX <op> rY' where rX and rY are not r10 but there is an early insn 'rX = r10'.
Without previous verifier change, both tests will fail.
Signed-off-by: Yonghong Song yonghong.song@linux.dev
.../selftests/bpf/progs/verifier_precision.c | 53 +++++++++++++++++++ 1 file changed, 53 insertions(+)
I was looking this commit (5ffb537e416e) since it was a BPF selftest test for CVE-2025-38279, but upon looking I found that the commit differs from the patch, there is an extra hunk that changed kernel/bpf/verifier.c that wasn't found the Yonghong's original patch.
I suppose it was meant to be squashed into the previous commit e2d2115e56c4 "bpf: Do not include stack ptr register in precision backtracking bookkeeping"?
Andrii made some change to my original patch for easy understanding. See https://lore.kernel.org/bpf/20250524041335.4046126-1-yonghong.song@linux.dev Quoted below: " I've moved it inside the preceding if/else (twice), so it's more obvious that BPF_X deal with both src_reg and dst_reg, and BPF_K case deals only with BPF_K. The end result is the same, but I found this way a bit easier to follow. Applied to bpf-next, thanks.
Argh, indeed I missed the sibling thread. Thanks for point that out.
Shung-Hsi
...