X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fshared%2Fmacro.h;h=2bb72f0d167626413aa5d91c1182050cc41da57d;hb=a858b64dddf79177e12ed30f5e8c47a1471c8bfe;hp=9bf81dc3cc541229bebe333d8d3ee31074ffe68b;hpb=49aa47c7fb6c6cf85f2780080e89181974efdc3b;p=elogind.git

diff --git a/src/shared/macro.h b/src/shared/macro.h
index 9bf81dc3c..2bb72f0d1 100644
--- a/src/shared/macro.h
+++ b/src/shared/macro.h
@@ -264,6 +264,24 @@ do {                                                                    \
         }                                                               \
 } while(false)
 
+/* Remove this macro, when the kernel has f_type as unsigned int or long
+ * for every architecure. Currently some 64bit architecures (like s390x)
+ * have int in the kernel, but long in userspace for f_type, so glibc
+ * extends the int to long and carries over the sign. Negative numbers are
+ * caused by the 32bit magic constants in linux/magic.h stuffed into the
+ * signed int in the kernel and these negative numbers are extended to
+ * long, which cannot be simply compared to the magic constants anymore.
+ */
+#define F_TYPE_CMP(f_type, magic)                                       \
+        __extension__ ({                                                \
+                        __SWORD_TYPE _f = (f_type);                     \
+                        const __SWORD_TYPE _c = (magic);                \
+                        const int _c32 = 1 ? (magic)                    \
+                                    : sizeof((int[magic]){0});          \
+                        (_f == _c || _f == _c32 );                      \
+                })
+
+
 /* Returns the number of chars needed to format variables of the
  * specified type as a decimal string. Adds in extra space for a
  * negative '-' prefix. */