[PATCH] base/class.c: prevent ooops due to insert/remove race

From: Mark Lord
Date: Thu Nov 29 2007 - 12:45:56 EST


(reposting for linux-usb-devel list)

Greg KH wrote:
On Wed, Nov 28, 2007 at 11:00:36PM -0500, Mark Lord wrote:
While doing insert/remove (quickly) tests on USB, I managed to trigger
an Oops on 2.6.23.8 on the call to strlen() in make_class_name().

This patch prevents this oops.

There is still the larger problem of the overall race
that caused this in the first place, but much of the rest
of the code in class.c appears to also do NULL checks to
avoid Oops'ing, so this continues the tradition.

Signed-off-by: Mark Lord <mlord@xxxxxxxxx>

As this is a bandage over the real problem, I'd prefer to not apply this
one right now until we find the root cause.
...

Ahh... okay.. but since nobody is actually actively trying
to "find the root cause", there's a problem with that approach.

Rather than have systems continue to crash (reboot required to regain
use of USB after the Oops), here is an improved version of the patch
that addresses Greg's concern.

This version still prevents the Oops, but also flags the BUG so that
it will not go unnoticed.

This should make everyone happier.

* * * *

While doing insert/remove (quickly) tests on USB, I managed to trigger
an Oops on 2.6.23.1 on the call to strlen() in make_class_name().

This patch prevents the oops, but still keeps the bug visible.

There is still the larger problem of the overall race
that caused this in the first place, but much of the rest
of the code in class.c appears to also do NULL checks to
avoid Oops'ing, so this continues the tradition.

Signed-off-by: Mark Lord <mlord@xxxxxxxxx> ---

Patch applies to both 2.6.24 and 2.6.23.


While doing insert/remove (quickly) tests on USB, I managed to trigger
an Oops on 2.6.23.1 on the call to strlen() in make_class_name().

This patch prevents the oops, but still keeps the bug visible.

There is still the larger problem of the overall race
that caused this in the first place, but much of the rest
of the code in class.c appears to also do NULL checks to
avoid Oops'ing, so this continues the tradition.

Signed-off-by: Mark Lord <mlord@xxxxxxxxx> ---

Patch applies to both 2.6.24 and 2.6.23.

--- linux/drivers/base/class.c.orig 2007-11-29 10:51:43.000000000 -0500
+++ linux/drivers/base/class.c 2007-11-29 10:54:42.000000000 -0500
@@ -354,6 +354,10 @@
char *class_name;
int size;

+ if (!name) {
+ BUG_ON(1);
+ return NULL;
+ }
size = strlen(name) + strlen(kobject_name(kobj)) + 2;

class_name = kmalloc(size, GFP_KERNEL);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/