Re: [PATCH][next] bpf: fix swapped arguments in calls to check_buffer_access

On 7/28/20 3:43 AM, Daniel Borkmann wrote:

On 7/27/20 11:39 PM, Yonghong Song wrote:

quoted

On 7/27/20 10:54 AM, Colin King wrote:

quoted

From: Colin Ian King <redacted>

There are a couple of arguments of the boolean flag zero_size_allowed
and the char pointer buf_info when calling to function 
check_buffer_access
that are swapped by mistake. Fix these by swapping them to correct
the argument ordering.

Addresses-Coverity: ("Array compared to 0")
Fixes: afbf21dce668 ("bpf: Support readonly/readwrite buffers in 
verifier")
Signed-off-by: Colin Ian King <redacted>

Thanks for the fix!
Acked-by: Yonghong Song <redacted>

Sigh, thanks for the fix Colin, applied! Yonghong, could you follow-up with
BPF selftest test cases that exercise these paths? Thx

This will be triggered with a verifier rejection path, e.g., negative 
offset from the base. I will send a follow-up patch soon.

BTW, using llvm to build the kernel (without this change), the compiler
actually issues a warning:

-bash-4.4$ make -j100 LLVM=1 && make LLVM=1 vmlinux
   GEN     Makefile
...
   CC      kernel/bpf/verifier.o
/data/users/yhs/work/net-next/kernel/bpf/verifier.c:3481:18: warning: 
expression which evaluates to zero treate$
  as a null pointer constant of type 'const char *' 
[-Wnon-literal-null-conversion]
                                           "rdonly", false,
                                                     ^~~~~
/data/users/yhs/work/net-next/kernel/bpf/verifier.c:3487:16: warning: 
expression which evaluates to zero treate$
  as a null pointer constant of type 'const char *' 
[-Wnon-literal-null-conversion]
                                           "rdwr", false,
                                                   ^~~~~
2 warnings generated.
   AR      kernel/bpf/built-in.a

Looks like I need to use LLVM compiler more often...