[yocto] [meta-security][PATCH] libcap-ng: port CVE-2014-3215 patch

Thu Nov 27 14:03:01 PST 2014

Thanks.

Armin

On 11/27/2014 10:49 AM, Joe MacDonald wrote:
> Importing the patch from meta-selinux, which itself was a backport from
> the upstream source tree.
>
> Signed-off-by: Joe MacDonald <joe_macdonald at mentor.com>
> ---
>
> I mentioned a while back that I had at least one patch in meta-selinux that may
> apply to meta-security as well.  I don't know if you guys are interested in this
> or not since the primary tool to demonstrate the exploit is seunshare, but it is
> a problem in libcap-ng itself and it is exploitable outside of the selinux
> framework.
>
> -J.
>
>   .../libcap-ng/libcap-ng/CVE-2014-3215.patch        | 91 ++++++++++++++++++++++
>   recipes-security/libcap-ng/libcap-ng_0.7.3.bb      |  4 +-
>   2 files changed, 94 insertions(+), 1 deletion(-)
>   create mode 100644 recipes-security/libcap-ng/libcap-ng/CVE-2014-3215.patch
>
> diff --git a/recipes-security/libcap-ng/libcap-ng/CVE-2014-3215.patch b/recipes-security/libcap-ng/libcap-ng/CVE-2014-3215.patch
> new file mode 100644
> index 0000000..e9164d4
> --- /dev/null
> +++ b/recipes-security/libcap-ng/libcap-ng/CVE-2014-3215.patch
> @@ -0,0 +1,91 @@
> +libcap-ng: local privilege escalation due to capng_lock
> +
> +Following the discussion here:
> +
> +   http://openwall.com/lists/oss-security/2014/04/29/7
> +
> +This is known to impact SELinux tools, however the issue could be exploited by
> +any application using the relevant functions in libcap-ng provided it is suid
> +root.
> +
> +Upstream-Status: Backport
> +
> +Signed-off-by: Joe MacDonald <joe_macdonald at mentor.com>
> +
> +diff --git a/docs/capng_lock.3 b/docs/capng_lock.3
> +index 7683119..a070c1e 100644
> +--- a/docs/capng_lock.3
> ++++ b/docs/capng_lock.3
> +@@ -8,12 +8,13 @@ int capng_lock(void);
> +
> + .SH "DESCRIPTION"
> +
> +-capng_lock will take steps to prevent children of the current process to regain full privileges if the uid is 0. This should be called while possessing the CAP_SETPCAP capability in the kernel. This function will do the following if permitted by the kernel: Set the NOROOT option on for PR_SET_SECUREBITS, set the NOROOT_LOCKED option to on for PR_SET_SECUREBITS, set the PR_NO_SETUID_FIXUP option on for PR_SET_SECUREBITS, and set the PR_NO_SETUID_FIXUP_LOCKED option on for PR_SET_SECUREBITS.
> ++capng_lock will take steps to prevent children of the current process from gaining privileges by executing setuid programs.  This should be called while possessing the CAP_SETPCAP capability in the kernel.
> +
> ++This function will do the following if permitted by the kernel:  If the kernel supports PR_SET_NO_NEW_PRIVS, it will use it.  Otherwise it will set the NOROOT option on for PR_SET_SECUREBITS, set the NOROOT_LOCKED option to on for PR_SET_SECUREBITS, set the PR_NO_SETUID_FIXUP option on for PR_SET_SECUREBITS, and set the PR_NO_SETUID_FIXUP_LOCKED option on for PR_SET_SECUREBITS.  If both fail, it will return an error.
> +
> + .SH "RETURN VALUE"
> +
> +-This returns 0 on success and a negative number on failure. -1 means a failure setting any of the PR_SET_SECUREBITS options.
> ++This returns 0 on success and a negative number on failure. -1 means a failure to use PR_SET_NO_NEW_PRIVS and a failure setting any of the PR_SET_SECUREBITS options.
> +
> + .SH "SEE ALSO"
> +
> +diff --git a/src/cap-ng.c b/src/cap-ng.c
> +index bd105ba..422f2bc 100644
> +--- a/src/cap-ng.c
> ++++ b/src/cap-ng.c
> +@@ -45,6 +45,7 @@
> +  * 2.6.24 kernel	XATTR_NAME_CAPS
> +  * 2.6.25 kernel	PR_CAPBSET_DROP, CAPABILITY_VERSION_2
> +  * 2.6.26 kernel	PR_SET_SECUREBITS, SECURE_*_LOCKED, VERSION_3
> ++ * 3.5    kernel	PR_SET_NO_NEW_PRIVS
> +  */
> +
> + /* External syscall prototypes */
> +@@ -122,6 +123,14 @@ extern int capget(cap_user_header_t header, const cap_user_data_t data);
> + #define SECURE_NO_SETUID_FIXUP_LOCKED   3  /* make bit-2 immutable */
> + #endif
> +
> ++/* prctl values that we use */
> ++#ifndef PR_SET_SECUREBITS
> ++#define PR_SET_SECUREBITS		28
> ++#endif
> ++#ifndef PR_SET_NO_NEW_PRIVS
> ++#define PR_SET_NO_NEW_PRIVS		38
> ++#endif
> ++
> + // States: new, allocated, initted, updated, applied
> + typedef enum { CAPNG_NEW, CAPNG_ERROR, CAPNG_ALLOCATED, CAPNG_INIT,
> + 	CAPNG_UPDATED, CAPNG_APPLIED } capng_states_t;
> +@@ -663,15 +672,22 @@ int capng_change_id(int uid, int gid, capng_flags_t flag)
> +
> + int capng_lock(void)
> + {
> +-#ifdef PR_SET_SECUREBITS
> +-	int rc = prctl(PR_SET_SECUREBITS,
> +-			1 << SECURE_NOROOT |
> +-			1 << SECURE_NOROOT_LOCKED |
> +-			1 << SECURE_NO_SETUID_FIXUP |
> +-			1 << SECURE_NO_SETUID_FIXUP_LOCKED, 0, 0, 0);
> ++	int rc;
> ++
> ++	// On Linux 3.5 and up, we can directly prevent ourselves and
> ++	// our descendents from gaining privileges.
> ++	if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) == 0)
> ++		return 0;
> ++
> ++	// This kernel is too old or otherwise doesn't support
> ++	// PR_SET_NO_NEW_PRIVS.  Fall back to using securebits.
> ++	rc = prctl(PR_SET_SECUREBITS,
> ++		   1 << SECURE_NOROOT |
> ++		   1 << SECURE_NOROOT_LOCKED |
> ++		   1 << SECURE_NO_SETUID_FIXUP |
> ++		   1 << SECURE_NO_SETUID_FIXUP_LOCKED, 0, 0, 0);
> + 	if (rc)
> + 		return -1;
> +-#endif
> +
> + 	return 0;
> + }
> diff --git a/recipes-security/libcap-ng/libcap-ng_0.7.3.bb b/recipes-security/libcap-ng/libcap-ng_0.7.3.bb
> index 3f225ba..1acf554 100644
> --- a/recipes-security/libcap-ng/libcap-ng_0.7.3.bb
> +++ b/recipes-security/libcap-ng/libcap-ng_0.7.3.bb
> @@ -8,7 +8,9 @@ LIC_FILES_CHKSUM = "file://COPYING;md5=94d55d512a9ba36caa9b7df079bae19f \
>   		    file://COPYING.LIB;md5=e3eda01d9815f8d24aae2dbd89b68b06"
>
>   SRC_URI = "http://people.redhat.com/sgrubb/libcap-ng/libcap-ng-${PV}.tar.gz \
> -	   file://python.patch"
> +	   file://python.patch \
> +	   file://CVE-2014-3215.patch \
> +	   "
>
>   inherit lib_package autotools pythonnative
>
>