bpf: Warn with __bpf_trap() kfunc maybe due to uninitialized variable

Marc Suñé (Isovalent, part of Cisco) reported an issue where an
uninitialized variable caused generating bpf prog binary code not
working as expected. The reproducer is in [1] where the flags
“-Wall -Werror” are enabled, but there is no warning as the compiler
takes advantage of uninitialized variable to do aggressive optimization.
The optimized code looks like below:

      ; {
           0:       bf 16 00 00 00 00 00 00 r6 = r1
      ;       bpf_printk("Start");
           1:       18 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 r1 = 0x0 ll
                    0000000000000008:  R_BPF_64_64  .rodata
           3:       b4 02 00 00 06 00 00 00 w2 = 0x6
           4:       85 00 00 00 06 00 00 00 call 0x6
      ; DEFINE_FUNC_CTX_POINTER(data)
           5:       61 61 4c 00 00 00 00 00 w1 = *(u32 *)(r6 + 0x4c)
      ;       bpf_printk("pre ipv6_hdrlen_offset");
           6:       18 01 00 00 06 00 00 00 00 00 00 00 00 00 00 00 r1 = 0x6 ll
                    0000000000000030:  R_BPF_64_64  .rodata
           8:       b4 02 00 00 17 00 00 00 w2 = 0x17
           9:       85 00 00 00 06 00 00 00 call 0x6
      <END>

The verifier will report the following failure:
  9: (85) call bpf_trace_printk#6
  last insn is not an exit or jmp

The above verifier log does not give a clear hint about how to fix
the problem and user may take quite some time to figure out that
the issue is due to compiler taking advantage of uninitialized variable.

In llvm internals, uninitialized variable usage may generate
'unreachable' IR insn and these 'unreachable' IR insns may indicate
uninitialized variable impact on code optimization. So far, llvm
BPF backend ignores 'unreachable' IR hence the above code is generated.
With clang21 patch [2], those 'unreachable' IR insn are converted
to func __bpf_trap(). In order to maintain proper control flow
graph for bpf progs, [2] also adds an 'exit' insn after bpf_trap()
if __bpf_trap() is the last insn in the function. The new code looks like:

      ; {
           0:       bf 16 00 00 00 00 00 00 r6 = r1
      ;       bpf_printk("Start");
           1:       18 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 r1 = 0x0 ll
                    0000000000000008:  R_BPF_64_64  .rodata
           3:       b4 02 00 00 06 00 00 00 w2 = 0x6
           4:       85 00 00 00 06 00 00 00 call 0x6
      ; DEFINE_FUNC_CTX_POINTER(data)
           5:       61 61 4c 00 00 00 00 00 w1 = *(u32 *)(r6 + 0x4c)
      ;       bpf_printk("pre ipv6_hdrlen_offset");
           6:       18 01 00 00 06 00 00 00 00 00 00 00 00 00 00 00 r1 = 0x6 ll
                    0000000000000030:  R_BPF_64_64  .rodata
           8:       b4 02 00 00 17 00 00 00 w2 = 0x17
           9:       85 00 00 00 06 00 00 00 call 0x6
          10:       85 10 00 00 ff ff ff ff call -0x1
                    0000000000000050:  R_BPF_64_32  __bpf_trap
          11:       95 00 00 00 00 00 00 00 exit
      <END>

In kernel, a new kfunc __bpf_trap() is added. During insn
verification, any hit with __bpf_trap() will result in
verification failure. The kernel is able to provide better
log message for debugging.

With llvm patch [2] and without this patch (no __bpf_trap()
kfunc for existing kernel), e.g., for old kernels, the verifier
outputs
  10: <invalid kfunc call>
  kfunc '__bpf_trap' is referenced but wasn't resolved
Basically, kernel does not support __bpf_trap() kfunc.
This still didn't give clear signals about possible reason.

With llvm patch [2] and with this patch, the verifier outputs
  10: (85) call __bpf_trap#74479
  unexpected __bpf_trap() due to uninitialized variable?
It gives much better hints for verification failure.

  [1] https://github.com/msune/clang_bpf/blob/main/Makefile#L3
  [2] llvm/llvm-project#131731

Signed-off-by: Yonghong Song <[email protected]>

Author: Yonghong Song

kernel/bpf/helpers.c

@@ -3273,6 +3273,10 @@ __bpf_kfunc void bpf_local_irq_restore(unsigned long *flags__irq_flag)
 	local_irq_restore(*flags__irq_flag);
 }
 
+__bpf_kfunc void __bpf_trap(void)
+{
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(generic_btf_ids)
@@ -3386,6 +3390,7 @@ BTF_ID_FLAGS(func, bpf_copy_from_user_dynptr, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_copy_from_user_str_dynptr, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_copy_from_user_task_dynptr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
 BTF_ID_FLAGS(func, bpf_copy_from_user_task_str_dynptr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, __bpf_trap)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {

kernel/bpf/verifier.c

@@ -12105,6 +12105,7 @@ enum special_kfunc_type {
 	KF_bpf_res_spin_unlock,
 	KF_bpf_res_spin_lock_irqsave,
 	KF_bpf_res_spin_unlock_irqrestore,
+	KF___bpf_trap,
 };
 
 BTF_ID_LIST(special_kfunc_list)
@@ -12170,6 +12171,7 @@ BTF_ID(func, bpf_res_spin_lock)
 BTF_ID(func, bpf_res_spin_unlock)
 BTF_ID(func, bpf_res_spin_lock_irqsave)
 BTF_ID(func, bpf_res_spin_unlock_irqrestore)
+BTF_ID(func, __bpf_trap)
 
 static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta)
 {
@@ -13641,6 +13643,9 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 			return err;
 		}
 		__mark_btf_func_reg_size(env, regs, BPF_REG_0, sizeof(u32));
+	} else if (!insn->off && insn->imm == special_kfunc_list[KF___bpf_trap]) {
+		verbose(env, "unexpected __bpf_trap() due to uninitialized variable?\n");
+		return -EFAULT;
 	}
 
 	if (is_kfunc_destructive(&meta) && !capable(CAP_SYS_BOOT)) {