Hi Rusty,
Today's linux-next merge of the modules tree got a conflict in
kernel/module.c between commit caabe240574a ("MODSIGN: Move the magic
string to the end of a module and eliminate the search") from Linus' tree
and commit 0250abdeec54 ("module: add syscall to load module from fd")
from the modules tree.
I fixed it up (I think - see below) and can carry the fix as necessary
(no action is required).
I do wonder why the above change in Linus' tree seems to have bypassed
the modules maintainer.
--
Cheers,
Stephen Rothwell [email protected]
diff --cc kernel/module.c
index 6085f5e,261bf82..0000000
--- a/kernel/module.c
+++ b/kernel/module.c
@@@ -2420,18 -2422,27 +2422,18 @@@ static inline void kmemleak_load_module
#endif
#ifdef CONFIG_MODULE_SIG
- static int module_sig_check(struct load_info *info,
- const void *mod, unsigned long *_len)
+ static int module_sig_check(struct load_info *info)
{
int err = -ENOKEY;
- unsigned long markerlen = sizeof(MODULE_SIG_STRING) - 1;
- unsigned long len = *_len;
+ const unsigned long markerlen = sizeof(MODULE_SIG_STRING) - 1;
+ const void *mod = info->hdr;
- const void *p = mod, *end = mod + info->len;
++ unsigned long len = info->len;
- /* Poor man's memmem. */
- while ((p = memchr(p, MODULE_SIG_STRING[0], end - p))) {
- if (p + markerlen > end)
- break;
-
- if (memcmp(p, MODULE_SIG_STRING, markerlen) == 0) {
- const void *sig = p + markerlen;
- /* Truncate module up to signature. */
- info->len = p - mod;
- err = mod_verify_sig(mod, info->len,
- sig, end - sig);
- break;
- }
- p++;
+ if (len > markerlen &&
+ memcmp(mod + len - markerlen, MODULE_SIG_STRING, markerlen) == 0) {
+ /* We truncate the module to discard the signature */
- *_len -= markerlen;
- err = mod_verify_sig(mod, _len);
++ info->len -= markerlen;
++ err = mod_verify_sig(mod, &info->len);
}
if (!err) {
On Sun, Oct 21, 2012 at 6:35 PM, Stephen Rothwell <[email protected]> wrote:
> Hi Rusty,
>
> Today's linux-next merge of the modules tree got a conflict in
> kernel/module.c between commit caabe240574a ("MODSIGN: Move the magic
> string to the end of a module and eliminate the search") from Linus' tree
> and commit 0250abdeec54 ("module: add syscall to load module from fd")
> from the modules tree.
>
> I fixed it up (I think - see below) and can carry the fix as necessary
> (no action is required).
>
> I do wonder why the above change in Linus' tree seems to have bypassed
> the modules maintainer.
>
> --
> Cheers,
> Stephen Rothwell [email protected]
>
> diff --cc kernel/module.c
> index 6085f5e,261bf82..0000000
> --- a/kernel/module.c
> +++ b/kernel/module.c
> @@@ -2420,18 -2422,27 +2422,18 @@@ static inline void kmemleak_load_module
> #endif
>
> #ifdef CONFIG_MODULE_SIG
> - static int module_sig_check(struct load_info *info,
> - const void *mod, unsigned long *_len)
> + static int module_sig_check(struct load_info *info)
> {
> int err = -ENOKEY;
> - unsigned long markerlen = sizeof(MODULE_SIG_STRING) - 1;
> - unsigned long len = *_len;
> + const unsigned long markerlen = sizeof(MODULE_SIG_STRING) - 1;
> + const void *mod = info->hdr;
I got rid of "mod" when I did my merge of these changes against
Linus's tree, but I had to cast it in the memcmp, so probably best to
just keep "mod".
> - const void *p = mod, *end = mod + info->len;
> ++ unsigned long len = info->len;
>
> - /* Poor man's memmem. */
> - while ((p = memchr(p, MODULE_SIG_STRING[0], end - p))) {
> - if (p + markerlen > end)
> - break;
> -
> - if (memcmp(p, MODULE_SIG_STRING, markerlen) == 0) {
> - const void *sig = p + markerlen;
> - /* Truncate module up to signature. */
> - info->len = p - mod;
> - err = mod_verify_sig(mod, info->len,
> - sig, end - sig);
> - break;
> - }
> - p++;
> + if (len > markerlen &&
> + memcmp(mod + len - markerlen, MODULE_SIG_STRING, markerlen) == 0) {
> + /* We truncate the module to discard the signature */
> - *_len -= markerlen;
> - err = mod_verify_sig(mod, _len);
> ++ info->len -= markerlen;
> ++ err = mod_verify_sig(mod, &info->len);
> }
>
> if (!err) {
-Kees
--
Kees Cook
Chrome OS Security
Stephen Rothwell <[email protected]> writes:
> Hi Rusty,
>
> Today's linux-next merge of the modules tree got a conflict in
> kernel/module.c between commit caabe240574a ("MODSIGN: Move the magic
> string to the end of a module and eliminate the search") from Linus' tree
> and commit 0250abdeec54 ("module: add syscall to load module from fd")
> from the modules tree.
>
> I fixed it up (I think - see below) and can carry the fix as necessary
> (no action is required).
>
> I do wonder why the above change in Linus' tree seems to have bypassed
> the modules maintainer.
Yes. David tacked it on to a real bugfix which went straight to Linus;
having fixed my mess I couldn't complain too much (though I had NAKed
similar patches in the past).
I will rebase my -next tree today.
Cheers,
Rusty.