On Thu, 23 Aug 2001, Keith Owens wrote:
> The definition of struct archdata in kernel and modutils can be
> different, a new kernel layout with an old modutils is legal but fatal
> unless you code for it. The correct test for archdata is
>
> if (!mod_member_present(mp, archdata_start) ||
> (mp->archdata_end - mp->archdata_start) <=
> offsetof(struct archdata, dbe_table_end))
> continue;
>
> Do not use archdata unless it is at least large enough to contain
> dbe_table_end. That test also takes care of NULL pointers, end - start
> == 0 for NULL.
Hmm, your suggested code checks if the passed struct is long enough for
dbe_table_start only -- what about dbe_table_end? The following code:
ap = (struct archdata *)(mod->archdata_start);
if (!mod_member_present(mp, archdata_start) ||
(mp->archdata_end - mp->archdata_start) <
offsetof(struct archdata, dbe_table_end) + sizeof(ap->dbe_table_end))
continue;
should be stricter. While modutils as released won't ever pass a smaller
struct, someone may modify them or use another program to invoke
init_module(), so we need to protect the kernel against bogus data.
> The rest of the code looks OK, except it needs a global change of
> arch_init_module: to module_arch_init: to match the macro name.
OK, I'll do it. It should have been done for ia64 in the first place.
Or should it be changed into "<arch>_init_module" to match functions' real
names?
> Do you have the corresponding modutils patch or shall I do it?
I've send it to you separately just after the kernel patch. Should I
resend it?
Maciej
--
+ Maciej W. Rozycki, Technical University of Gdansk, Poland +
+--------------------------------------------------------------+
+ e-mail: macro@ds2.pg.gda.pl, PGP key available +
