[PATCH] module support: record in vermagic ability to unload a module

An UML user reported (against 2.6.13.3/UML) he got kernel Oopses when
trying to rmmod (on a kernel with module unloading enabled) a module
compiled with module unloading disabled.  As crashing is a very correct
thing to do in that case, a solution is altering the vermagic string to
include this too.

Possibly, however, the code should not crash in this case, even if the
module didn't support unloading - it should simply abort the module
removal.  In this case, fixing that bug would be a better solution.  I've
not investigated though.

(akpm: a bit marginal - root screwed up and shot himself in the foot).

Cc: Hayim Shaul <[email protected]>
Cc: Rusty Russell <[email protected]>
Signed-off-by: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

include/linux/vermagic.h

@@ -12,12 +12,17 @@
 #else
 #define MODULE_VERMAGIC_PREEMPT ""
 #endif
+#ifdef CONFIG_MODULE_UNLOAD
+#define MODULE_VERMAGIC_MODULE_UNLOAD "mod_unload "
+#else
+#define MODULE_VERMAGIC_MODULE_UNLOAD ""
+#endif
 #ifndef MODULE_ARCH_VERMAGIC
 #define MODULE_ARCH_VERMAGIC ""
 #endif
 
 #define VERMAGIC_STRING 						\
 	UTS_RELEASE " "							\
 	MODULE_VERMAGIC_SMP MODULE_VERMAGIC_PREEMPT 			\
-	MODULE_ARCH_VERMAGIC 						\
+	MODULE_VERMAGIC_MODULE_UNLOAD MODULE_ARCH_VERMAGIC 		\
 	"gcc-" __stringify(__GNUC__) "." __stringify(__GNUC_MINOR__)