Skip to contentSkip to conversation

local root exploit confirmed in 2.6.10: Linux 2.6 Kernel Capability LSM Module Local Privilege Elevation

DISCUSSION Lee Revell 3 participants 5 messages Dec 28, 2004 – Dec 29, 2004 Security (LSM)

lore
Share & tools
lore

Lee Revell

author
#0 lore raw

Frank Barknecht pointed this out on linux-audio-dev, it's a horrible bug, I confirmed it in 2.6.10, and have not seen it mentioned on the list.

Executive summary:

run "vim" as normal user. Do ":r /etc/shadow". Permission denied.

do "modprobe capability" as root in another terminal

Do ":r /etc/shadow" again in the same vim. You will be able to read and write /etc/shadow as normal user.

http://www.derkeiler.com/Mailing-Lists/securityfocus/bugtraq/2004-12/0390.html

Lee

#2 lore raw
In reply to Lee Revell
Quoted 3 lines · Lee Revell

Frank Barknecht pointed this out on linux-audio-dev, it's a horrible bug, I confirmed it in 2.6.10, and have not seen it mentioned on the list.

Although this sucks, it should be pointed out that it only grants root to users able to force the loading of a certain module, aka 'root'.

Modules have always been free to give root capabilities to all users. We don't usually ship these though :-)

Signature · 2 lines
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://lartc.org           Linux Advanced Routing & Traffic Control HOWTO
author
#3 lore raw
Quoted 7 lines · bert hubert
Quoted 3 lines · Lee Revell

Frank Barknecht pointed this out on linux-audio-dev, it's a horrible bug, I confirmed it in 2.6.10, and have not seen it mentioned on the list.

Although this sucks, it should be pointed out that it only grants root to users able to force the loading of a certain module, aka 'root'.

Not force the loading of a certain module, but predict when it will be loaded. Still, not easy to exploit.

Lee

#4 lore raw
Quoted 6 lines · Lee Revell
Quoted 5 lines · bert hubert
Quoted 3 lines · Lee Revell

Frank Barknecht pointed this out on linux-audio-dev, it's a horrible bug, I confirmed it in 2.6.10, and have not seen it mentioned on the list.

Quoted 7 lines
Quoted 2 lines

Although this sucks, it should be pointed out that it only grants root to users able to force the loading of a certain module, aka 'root'.

Not force the loading of a certain module, but predict when it will be loaded. Still, not easy to exploit.

Lee

Right, this means it is unsafe to have capabilities compiled as a module, or at least loaded after any untrusted processes start.

The attached patch, which is a simple port of a fix by Chris Wright (sent out a year ago), fixes this problem by having the dummy module track capabilities.

-serge

security/dummy.c +10 -5
Index: linux-2.6.9/security/dummy.c
===================================================================
@@ -74,12 +74,10 @@ static int dummy_acct (struct file *file
 
 static int dummy_capable (struct task_struct *tsk, int cap)
 {
-	if (cap_is_fs_cap (cap) ? tsk->fsuid == 0 : tsk->euid == 0)
-		/* capability granted */
+	if (cap_raised (tsk->cap_effective, cap))
 		return 0;
-
-	/* capability denied */
-	return -EPERM;
+	else
+		return -EPERM;
 }
 
 static int dummy_sysctl (ctl_table * table, int op)
@@ -199,6 +197,10 @@ static void dummy_bprm_apply_creds (stru
 
 	current->suid = current->euid = current->fsuid = bprm->e_uid;
 	current->sgid = current->egid = current->fsgid = bprm->e_gid;
+
+	dummy_capget(current, &current->cap_effective,
+					&current->cap_inheritable,
+					&current->cap_permitted);
 }
 
 static int dummy_bprm_set_security (struct linux_binprm *bprm)
@@ -563,6 +565,9 @@ static int dummy_task_setuid (uid_t id0,
 
 static int dummy_task_post_setuid (uid_t id0, uid_t id1, uid_t id2, int flags)
 {
+	dummy_capget(current, &current->cap_effective,
+					&current->cap_inheritable,
+					&current->cap_permitted);
 	return 0;
 }