Re: [PATCH bpf-next 2/4] bpf: Merge pointer types also when both are PTR_TO_MEM

Daniel Borkmann <[email protected]>
Newsgroups org.kernel.vger.bpf
Message-ID <[email protected]>
On 8/14/26 2:14 AM, Eduard Zingerman wrote:
> On Thu, 2026-08-13 at 22:40 +0200, Daniel Borkmann wrote:
>> save_aux_ptr_type() only reaches the merge when reg_type_mismatch() says
>> the two types are incompatible, and that in turn requires at least one of
>> them to have a base type reg_type_mismatch_ok() rejects. PTR_TO_MEM is
>> not among those, so for two PTR_TO_MEM based types the merge never runs
>> and the recorded type stays the one of whichever path was verified first.
>>
>> That is ok as long as all PTR_TO_MEM variants can be dereferenced with
>> a plain load, which stopped being true with commit f2362a57aeff ("bpf:
>> allow void* cast using bpf_rdonly_cast()") adding PTR_TO_MEM | MEM_RDONLY
>>> PTR_UNTRUSTED. If the other path saved e.g. a PTR_TO_MEM | MEM_RINGBUF
>> first, then bpf_convert_ctx_accesses() does not rewrite the load into a
>> BPF_PROBE_MEM one, and the untrusted path faults on a plain load.
>>
>> Fix by merging the two whenever they differ and both are of PTR_TO_MEM or
>> PTR_TO_BTF_ID base instead of keying it off reg_type_mismatch(), so that
>> merge_ptr_types() gets to normalize the result in this case as well. The
>> rejection of genuinely incompatible types is left untouched.
>>
>> Fixes: f2362a57aeff ("bpf: allow void* cast using bpf_rdonly_cast()")
>> Signed-off-by: Daniel Borkmann <[email protected]>
>> ---
> 
> Would the following work?
> 
>    diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>    index 1e537bc25ef5..a02914af0f5b 100644
>    --- a/kernel/bpf/verifier.c
>    +++ b/kernel/bpf/verifier.c
>    @@ -17022,6 +17022,8 @@ static bool reg_type_mismatch_ok(enum bpf_reg_type type)
>          case PTR_TO_BTF_ID:
>          case PTR_TO_ARENA:
>                  return false;
>    +     case PTR_TO_MEM:
>    +             return !may_fault_on_deref(type);
>          default:
>                  return true;
Yes, that works and looks cleaner. Will integrate, thanks!
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.