[PATCH] Using possibly corrupted structure in atm/he.c

[PATCH] Using possibly corrupted structure in atm/he.c

[Felipe W Damasio]

[-- Attachment #1: Type: text/plain, Size: 582 bytes --]

	Hi Chas,

	Patch against 2.6-test5.

	If copy_from_user returns != 0, it means the the regs structure wasn't 
filled correctly, and since its fields are used to determine which ioctl 
the user is requesting the kernel could oops.

	And as long as we're covering the subject, the patch also audits 
copy_to_user on the same function to check a possible failure to copy 
the result back to userspace.

	These bugs were found by smatch.

	Please consider applying.

	Cheers,

Felipe
-- 
It's most certainly GNU/Linux, not Linux. Read more at
http://www.gnu.org/gnu/why-gnu-linux.html

[-- Attachment #2: atm_he-copy.patch --]
[-- Type: text/plain, Size: 889 bytes --]

--- linux-2.6.0-test5/drivers/atm/he.c	Mon Sep  8 16:49:54 2003
+++ linux-2.6.0-test5-fwd/drivers/atm/he.c	Sun Sep 21 01:13:39 2003
@@ -2860,8 +2860,10 @@
 			if (!capable(CAP_NET_ADMIN))
 				return -EPERM;
 
-			copy_from_user(&reg, (struct he_ioctl_reg *) arg,
-						sizeof(struct he_ioctl_reg));
+			if (copy_from_user(&reg, (struct he_ioctl_reg *) arg,
+						sizeof(struct he_ioctl_reg)))
+				return -EFAULT;
+			
 			spin_lock_irqsave(&he_dev->global_lock, flags);
 			switch (reg.type) {
 				case HE_REGTYPE_PCI:
@@ -2885,8 +2887,9 @@
 			}
 			spin_unlock_irqrestore(&he_dev->global_lock, flags);
 			if (err == 0)
-				copy_to_user((struct he_ioctl_reg *) arg, &reg,
-							sizeof(struct he_ioctl_reg));
+				if (copy_to_user((struct he_ioctl_reg *) arg, &reg,
+							sizeof(struct he_ioctl_reg)))
+					return -EFAULT;
 			break;
 		default:
 #ifdef CONFIG_ATM_HE_USE_SUNI

Re: [PATCH] Using possibly corrupted structure in atm/he.c

[chas williams]

In message <3F6D2832.8040609@terra.com.br>,Felipe W Damasio writes:
>	If copy_from_user returns != 0, it means the the regs structure wasn't 
>filled correctly, and since its fields are used to determine which ioctl 
>the user is requesting the kernel could oops.
>
>	And as long as we're covering the subject, the patch also audits 
>copy_to_user on the same function to check a possible failure to copy 
>the result back to userspace.

indeed these are broken as written.  i will get this sent up the chain.