On Fri, Jan 23, 2026 at 1:21 AM Menglong Dong [off-list ref] wrote:

On 2026/1/23 16:29 Menglong Dong [off-list ref] write:

quoted

On 2026/1/23 16:07 bot+bpf-ci@kernel.org write:

quoted

quoted

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 41bbed6418b5..b3f912ad18ba 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c

[ ... ]

quoted

@@ -23779,12 +23781,13 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
                      eatype == BPF_MODIFY_RETURN) {
                          /* Load nr_args from ctx - 8 */
                          insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
-                         insn_buf[1] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
+                         insn_buf[1] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 0xFF);
+                         insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);

[ ... ]

quoted

-                         cnt = 6;
+                         cnt = 7;
                  } else {
                          insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, -EOPNOTSUPP);
                          cnt = 1;
                  }

                  new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);

[ ... ]

quoted

                  delta    += cnt - 1;

Note that the bpf_get_func_ret block above correctly updates delta after
patching.

quoted

@@ -23808,12 +23811,15 @@ static int do_misc_fixups(struct bpf_verifier_env *env)

                          /* skip 'void *__data' in btf_trace_##name() and save to reg0 */
                          insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, nr_args - 1);
+                         cnt = 1;
                  } else {
                          /* Load nr_args from ctx - 8 */
                          insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
+                         insn_buf[1] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 0xFF);
+                         cnt = 2;
                  }

-                 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1);
+                 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
                  if (!new_prog)
                          return -ENOMEM;

                  env->prog = prog = new_prog;
                  insn      = new_prog->insnsi + i + delta;
                  goto next_insn;
          }

Should there be a "delta += cnt - 1;" before assigning env->prog in this
bpf_get_func_arg_cnt block? When cnt is 2 (non-RAW_TP path), the program
grows by one instruction, but delta is not updated. This could cause
subsequent bpf_patch_insn_data() calls to use incorrect offsets.

The bpf_get_func_arg block at the beginning of this patch has:

    delta    += cnt - 1;
    env->prog = prog = new_prog;

And the bpf_get_func_ret block similarly has delta updated. But this
bpf_get_func_arg_cnt block goes directly to env->prog assignment without
updating delta.

Ah, good point, I think this is a valid problem. The selftests didn't cover
this case, and I think I'd better to use bpf_get_func_arg() and bpf_get_func_ret()
in the exit path of fsession to cover it.

Oh, the problem doesn't have much impact. The only impact
is that the verifier will check the "r0 &= 0xFF" instruction redundantly.

I'll see if there is more comment before I send next version.

Let's still fix the problem. Just go ahead and resubmit (with the fix
and improved test).

pw-bot: cr

Thanks!
Menglong Dong

quoted

Will fix it in the next version.

Thanks!
Menglong Dong

quoted

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21278745581