On 12/23/19 12:05 PM, Andrii Nakryiko wrote:

On Fri, Dec 20, 2019 at 10:26 PM Martin KaFai Lau [off-list ref] wrote:

quoted

This patch allows bitfield access as a scalar.

Signed-off-by: Martin KaFai Lau <redacted>
---
  kernel/bpf/btf.c | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 6e652643849b..da73b63acfc5 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c

@@ -3744,10 +3744,6 @@ int btf_struct_access(struct bpf_verifier_log *log,
         }

         for_each_member(i, t, member) {
-               if (btf_member_bitfield_size(t, member))
-                       /* bitfields are not supported yet */
-                       continue;
-
                 /* offset of the field in bytes */
                 moff = btf_member_bit_offset(t, member) / 8;
                 if (off + size <= moff)

@@ -3757,6 +3753,12 @@ int btf_struct_access(struct bpf_verifier_log *log,
                 if (off < moff)
                         continue;

+               if (btf_member_bitfield_size(t, member)) {
+                       if (off == moff && off + size <= t->size)
+                               return SCALAR_VALUE;
+                       continue;
+               }

Shouldn't this check be done before (off < moff) above?

Imagine this situation:

struct {
   int :16;
   int x:8;
};

Oh, yes, forgot the case where the first bitfield member may have no
name, in which case, `off` will not match any `moff`.

btf_struct_access is used to check vmlinux btf types. I think in
vmlinux we may not have such scenarios. So the above code should
handle vmlinux use cases properly.

But I agree with Andrii that we probably want to handle
anonymous bitfield member (which is ignored in debuginfo and BTF) properly.

Compiler will generate 4-byte load with offset 0, and then bit shifts
to extract third byte. From kernel perspective, you'll see that off=0,
but moff=2, which will get skipped.

So there are two problems, I think:
1. if member is bitfield, special handle that before (off < moff) case.
2. off == moff is too precise, I think it should be `off <= moff`, but
also check that it covers entire bitfield, e.g.:

   (off + size) * 8 >= btf_member_bit_offset(t, member) +
btf_member_bitfield_size(t, member)

Make sense or am I missing anything?

quoted

+
                 /* type of the field */
                 mtype = btf_type_by_id(btf_vmlinux, member->type);
                 mname = __btf_name_by_offset(btf_vmlinux, member->name_off);
--
2.17.1