Re: [PATCH] bpf: Avoid function casting when calculating immediate

From: Daniel Borkmann
Date: Mon Feb 17 2020 - 10:51:52 EST


On 1/29/20 10:36 PM, Kees Cook wrote:
In an effort to enable -Wcast-function-type in the top-level Makefile
to support Control Flow Integrity builds, rework the BPF instruction
immediate calculation macros to avoid mismatched function pointers. Since
these calculations are only ever between function address (these are
not function calls, just address calculations), they can be cast to u64
instead, where the result will be assigned to the s32 insn->imm.

Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
---
include/linux/filter.h | 6 +++---
kernel/bpf/hashtab.c | 6 +++---
kernel/bpf/verifier.c | 21 +++++++--------------
3 files changed, 13 insertions(+), 20 deletions(-)

diff --git a/include/linux/filter.h b/include/linux/filter.h
index f349e2c0884c..b5beee7bf2ea 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -340,8 +340,8 @@ static inline bool insn_is_zext(const struct bpf_insn *insn)
/* Function call */
-#define BPF_CAST_CALL(x) \
- ((u64 (*)(u64, u64, u64, u64, u64))(x))
+#define BPF_FUNC_IMM(FUNC) \
+ ((u64)(FUNC) - (u64)__bpf_call_base)

Looks good to me in general. My only concern is compilation on 32bit archs: I think
the cast needs to be of '(u64)(unsigned long)' to avoid introducing new warnings a la
'cast from pointer to integer of different size'.

Thanks,
Daniel