sanitize task->comm to avoid leaking escape codes
in [Kernel]
|
From: Kees Cook on 28 Jun 2010 14:10 On Mon, Jun 28, 2010 at 07:48:38PM +0200, Stefani Seibold wrote: > Am Freitag, den 25.06.2010, 08:56 +0900 schrieb KOSAKI Motohiro: > > > Through get_task_comm() and many direct uses of task->comm in the kernel, > > > it is possible for escape codes and other non-printables to leak into > > > dmesg, syslog, etc. In the worst case, these strings could be used to > > > attack administrators using vulnerable terminal emulators, and at least > > > cause confusion through the injection of \r characters. > > > > > > This patch sanitizes task->comm to only contain printable characters > > > when it is set. Additionally, it redefines get_task_comm so that it is > > > more obvious when misused by callers (presently nothing was incorrectly > > > calling get_task_comm's unsafe use of strncpy). > > > > > > Signed-off-by: Kees Cook <kees.cook(a)canonical.com> > > > > I've reviewed this patch briefly, Here is my personal concern... > > > > On Linux, non-printable leaking is fundamental, only fixing task->comm > > doesn't solve syslog exploit issue. Probably all /proc/kmsg user should > > have escaping non-pritables code. > > > > However, task->comm is one of most easy injection data of kernel, because > > we have prctl(PR_SET_NAME), attacker don't need root privilege. So, > > conservative assumption seems guard from crappy fault. Plus, this patch > > is very small and our small TASK_COMM_LEN lead that we don't need > > big performance concern. > > > > So, I don't find demerit in this proposal. but I'm not security specialist, > > it's only personal thinking. > > > Agree. I think a escaped printk should be a more generic solution. I think sanitizing inputs is more effective than sanitizing outputs. If ->comm is safe internally, then we don't have to filter it going out on printk, audit, /proc output, etc. There is a limited number of places where a process has control over an arbitrary string in kernel structures, so the places where they are set should be fixed instead of fixing every possible usage of it on output. I wouldn't mind sanitizing printk also, but it's tangential to sanitizing task->comm when it is set. -Kees -- Kees Cook Ubuntu Security Team -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Andrew Morton on 28 Jun 2010 16:10 On Wed, 23 Jun 2010 23:23:35 +0300 Alexey Dobriyan <adobriyan(a)gmail.com> wrote: > On Wed, Jun 23, 2010 at 09:41:45PM +0200, Oleg Nesterov wrote: > > On 06/23, Kees Cook wrote: > > > > -extern char *get_task_comm(char *to, struct task_struct *tsk); > > > +#define get_task_comm(buf, task) get_task_comm_size(buf, sizeof(buf), task) > > > +extern char *get_task_comm_size(char *to, size_t len, struct task_struct *tsk); > > > > Oh, but this means that get_task_comm(ptr, task) doesn't work? > > The number of users is so small, and everyone uses TASK_COMM_LEN, > so maybe nothing should be done or "char buf[TASK_COMM_LEN]"? Yup, it would take an act of wilful daftness to pass anything other than a TASK_COMM_LEN array into get_task_comm(). I can't say I like the patch much - going in and altering a task's comm[] behind its back seems fraught with all sorts of problems which I can't immediately think of. Do we corrupt (err, "fix") /proc/pid/cmdline as well? Surely it would be better to fix the tools which display this info rather than making the kernel tell fibs. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Kees Cook on 28 Jun 2010 17:10 Hi, On Mon, Jun 28, 2010 at 01:00:28PM -0700, Andrew Morton wrote: > Surely it would be better to fix the tools which display this info > rather than making the kernel tell fibs. The strncpy in get_task_comm() is totally wrong -- it's testing the length of task->comm. Why should get_task_comm not take a destination buffer length argument? At least consider v2 of the patch -- it just fixes the get_task_comm definition and callers. But, if not, then patches to the kernel that include printk(..., get_task_comm(...) ...) shouldn't be considered flawed[1]. -Kees [1] http://lkml.org/lkml/2010/6/23/132 -- Kees Cook Ubuntu Security Team -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KOSAKI Motohiro on 28 Jun 2010 23:10 > Am Freitag, den 25.06.2010, 08:56 +0900 schrieb KOSAKI Motohiro: > > > Through get_task_comm() and many direct uses of task->comm in the kernel, > > > it is possible for escape codes and other non-printables to leak into > > > dmesg, syslog, etc. In the worst case, these strings could be used to > > > attack administrators using vulnerable terminal emulators, and at least > > > cause confusion through the injection of \r characters. > > > > > > This patch sanitizes task->comm to only contain printable characters > > > when it is set. Additionally, it redefines get_task_comm so that it is > > > more obvious when misused by callers (presently nothing was incorrectly > > > calling get_task_comm's unsafe use of strncpy). > > > > > > Signed-off-by: Kees Cook <kees.cook(a)canonical.com> > > > > I've reviewed this patch briefly, Here is my personal concern... > > > > On Linux, non-printable leaking is fundamental, only fixing task->comm > > doesn't solve syslog exploit issue. Probably all /proc/kmsg user should > > have escaping non-pritables code. > > > > However, task->comm is one of most easy injection data of kernel, because > > we have prctl(PR_SET_NAME), attacker don't need root privilege. So, > > conservative assumption seems guard from crappy fault. Plus, this patch > > is very small and our small TASK_COMM_LEN lead that we don't need > > big performance concern. > > > > So, I don't find demerit in this proposal. but I'm not security specialist, > > it's only personal thinking. > > > Agree. I think a escaped printk should be a more generic solution. Is this possible? printk() doesn't know userland locale. how do it escape correctly? When we only concern task->comm, assuming ascii-only string is enough practical. but printk generic logic should allow non-ascii, I think. I think userland reader process only know correct escaping way. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Artem Bityutskiy on 29 Jun 2010 01:00
On Wed, 2010-06-23 at 21:41 +0200, Oleg Nesterov wrote: > On 06/23, Kees Cook wrote: > > > > @@ -956,7 +957,15 @@ void set_task_comm(struct task_struct *tsk, char *buf) > > */ > > memset(tsk->comm, 0, TASK_COMM_LEN); > > wmb(); > > Off-topic. I'd wish I could understand this barrier. Since the lockless > reader doesn't do rmb() I don't see how this can help. This wmb() looks wrong to me as well. To achieve what the comment in this function says, it should be smp_wmb() and we should have smp_rmb() in the reading side, AFAIU. > OTOH, I don't > understand why it is needed, we never change ->comm[TASK_COMM_LEN-1] == '0'. I think the idea was that readers can see incomplete names, but not messed up names, consisting of old and new ones. -- Best Regards, Artem Bityutskiy (Артём Битюцкий) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |