OSDN Git Service

bpf: Support variable length array in tracing programs
authorYonghong Song <yhs@fb.com>
Sat, 9 May 2020 17:59:16 +0000 (10:59 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Sun, 10 May 2020 00:05:27 +0000 (17:05 -0700)
In /proc/net/ipv6_route, we have
  struct fib6_info {
    struct fib6_table *fib6_table;
    ...
    struct fib6_nh fib6_nh[0];
  }
  struct fib6_nh {
    struct fib_nh_common nh_common;
    struct rt6_info **rt6i_pcpu;
    struct rt6_exception_bucket *rt6i_exception_bucket;
  };
  struct fib_nh_common {
    ...
    u8 nhc_gw_family;
    ...
  }

The access:
  struct fib6_nh *fib6_nh = &rt->fib6_nh;
  ... fib6_nh->nh_common.nhc_gw_family ...

This patch ensures such an access is handled properly.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200509175916.2476853-1-yhs@fb.com
kernel/bpf/btf.c

index c490fbd..dcd2331 100644 (file)
@@ -3833,6 +3833,7 @@ int btf_struct_access(struct bpf_verifier_log *log,
        const struct btf_type *mtype, *elem_type = NULL;
        const struct btf_member *member;
        const char *tname, *mname;
+       u32 vlen;
 
 again:
        tname = __btf_name_by_offset(btf_vmlinux, t->name_off);
@@ -3841,7 +3842,43 @@ again:
                return -EINVAL;
        }
 
+       vlen = btf_type_vlen(t);
        if (off + size > t->size) {
+               /* If the last element is a variable size array, we may
+                * need to relax the rule.
+                */
+               struct btf_array *array_elem;
+
+               if (vlen == 0)
+                       goto error;
+
+               member = btf_type_member(t) + vlen - 1;
+               mtype = btf_type_skip_modifiers(btf_vmlinux, member->type,
+                                               NULL);
+               if (!btf_type_is_array(mtype))
+                       goto error;
+
+               array_elem = (struct btf_array *)(mtype + 1);
+               if (array_elem->nelems != 0)
+                       goto error;
+
+               moff = btf_member_bit_offset(t, member) / 8;
+               if (off < moff)
+                       goto error;
+
+               /* Only allow structure for now, can be relaxed for
+                * other types later.
+                */
+               elem_type = btf_type_skip_modifiers(btf_vmlinux,
+                                                   array_elem->type, NULL);
+               if (!btf_type_is_struct(elem_type))
+                       goto error;
+
+               off = (off - moff) % elem_type->size;
+               return btf_struct_access(log, elem_type, off, size, atype,
+                                        next_btf_id);
+
+error:
                bpf_log(log, "access beyond struct %s at off %u size %u\n",
                        tname, off, size);
                return -EACCES;