[PATCH] Prevent ld segementation fault

[PATCH] Prevent ld segementation fault

[Martin Koegler]

vfinfo (ldmisc.c) assumes that for %B a valid bfd handle is passed as parameter.
If the bfd handle is NULL, it will segfault.
reloc_overflow (ldmain.c) passes the content of entry->u.def.section->owner
to vfinfo to be printed with %B. If the entry is in the absolute section *ABS*,
it has no associated bfd. Therefore ld will produce an segmentation fault.

Can something like the following patch be applied to the CVS?
(All it does, is to print a default value, if abfd is NULL).
Or is there a better solution to fix this problem?

Index: ldmisc.c
===================================================================
RCS file: /cvs/src/src/ld/ldmisc.c,v
retrieving revision 1.20
diff -u -r1.20 ldmisc.c
--- ldmisc.c	13 Aug 2004 03:16:01 -0000	1.20
+++ ldmisc.c	16 Dec 2004 10:21:46 -0000
@@ -188,7 +188,9 @@
 	      /* filename from a bfd */
 	      {
 		bfd *abfd = va_arg (arg, bfd *);
-		if (abfd->my_archive)
+		if (abfd == NULL)
+		  fprintf (fp, "<none>");
+		else if (abfd->my_archive)
 		  fprintf (fp, "%s(%s)", abfd->my_archive->filename,
 			   abfd->filename);
 		else


mfg Martin Kögler
e9925248@stud4.tuwien.ac.at

Re: [PATCH] Prevent ld segementation fault

[Nick Clifton]

Hi Martin,

> Can something like the following patch be applied to the CVS?
> (All it does, is to print a default value, if abfd is NULL).

Yes it can and in fact yes it has been.

> Or is there a better solution to fix this problem?

In theory users of vinfo() should check before passing a NULL BFD, but 
that is no reason for vinfo() itself not being able to cope gracefully 
with the problem.

Cheers
   Nick