Em ter., 27 de abr. de 2021 às 13:18, Daniel Borkmann
[off-list ref] escreveu:

On 4/25/21 12:16 AM, Pedro Tammela wrote:

quoted

The 'bpf()' syscall is leaking the ENOTSUPP errno that is internal to the kernel[1].
More recent code is already using the correct EOPNOTSUPP, but changing
older return codes is not possible due to dependency concerns, so handle ENOTSUPP
in libbpf_strerror().

[1] https://lore.kernel.org/netdev/20200511165319.2251678-1-kuba@kernel.org/ (local)

Signed-off-by: Pedro Tammela <redacted>
---
  tools/lib/bpf/libbpf_errno.c | 9 +++++++++
  1 file changed, 9 insertions(+)

diff --git a/tools/lib/bpf/libbpf_errno.c b/tools/lib/bpf/libbpf_errno.c
index 0afb51f7a919..7de8bbc34a37 100644
--- a/tools/lib/bpf/libbpf_errno.c
+++ b/tools/lib/bpf/libbpf_errno.c

@@ -13,6 +13,9 @@

  #include "libbpf.h"

+/* This errno is internal to the kernel but leaks in the bpf() syscall. */
+#define ENOTSUPP 524
+
  /* make sure libbpf doesn't use kernel-only integer typedefs */
  #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64

@@ -43,6 +46,12 @@ int libbpf_strerror(int err, char *buf, size_t size)

      err = err > 0 ? err : -err;

+     if (err == ENOTSUPP) {
+             snprintf(buf, size, "Operation not supported");
+             buf[size - 1] = '\0';
+             return 0;
+     }
+
      if (err < __LIBBPF_ERRNO__START) {
              int ret;

Could you fold this into the __LIBBPF_ERRNO__START test body to denote that it
belongs outside the libbpf error range? For example, could be simplified like this:

         if (err < __LIBBPF_ERRNO__START) {
                 int ret;

                 /* Handle ENOTSUPP separate here given it's kernel internal,
                  * but for sake of error string it has the same meaning as
                  * the EOPNOTSUPP error.
                  */
                 if (err == ENOTSUPP)
                         err = EOPNOTSUPP;
                 ret = strerror_r(err, buf, size);
                 buf[size - 1] = '\0';
                 return ret;
         }

Thanks,
Daniel

Sure, looks simpler indeed.

Pedro