On 2021-01-08 22:41, Casey Schaufler wrote:
On 1/8/2021 1:49 AM, Preeti Nagar wrote:
The changes introduce a new security feature, RunTime Integrity Check
(RTIC), designed to protect Linux Kernel at runtime. The motivation
behind these changes is:
1. The system protection offered by SE for Android relies on the
assumption of kernel integrity. If the kernel itself is compromised
(by
a perhaps as yet unknown future vulnerability), SE for Android
security
mechanisms could potentially be disabled and rendered ineffective.
2. Qualcomm Snapdragon devices use Secure Boot, which adds
cryptographic
checks to each stage of the boot-up process, to assert the
authenticity
of all secure software images that the device executes. However, due
to
various vulnerabilities in SW modules, the integrity of the system can
be
compromised at any time after device boot-up, leading to un-authorized
SW executing.
It would be helpful if you characterized the "various vulnerabilities"
rather than simply asserting their existence. This would allow the
reviewer
to determine if the proposed patch addresses the issue.
There might not currently be vulnerabilities, but the system is meant
more
specifically to harden valuable assets against future compromises. The
key
value add is a third party independent entity keeping a watch on crucial
kernel assets.
The feature's idea is to move some sensitive kernel structures to a
separate page and monitor further any unauthorized changes to these,
from higher Exception Levels using stage 2 MMU. Moving these to a
different page will help avoid getting page faults from un-related
data.
I've always been a little slow when it comes to understanding the
details of advanced memory management facilities. That's part of
why I work in access control. Could you expand this a bit, so that
someone who doesn't already know how your stage 2 MMU works might
be able to evaluate what you're doing here.
Sure, will include more details. The mechanism we have been working on
removes the write permissions for HLOS in the stage 2 page tables for
the regions to be monitored, such that any modification attempts to
these
will lead to faults being generated and handled by handlers. If the
protected assets are moved to a separate page, faults will be generated
corresponding to change attempts to these assets only. If not moved to a
separate page, write attempts to un-related data which is present on the
monitored pages will also be generated.
Using this mechanism, some sensitive variables of the kernel which are
initialized after init or are updated rarely can also be protected
from
simple overwrites and attacks trying to modify these.
How would this interact with or complement __read_mostly?
Currently, the mechanism we are working on developing is
independent of __read_mostly. This is something we can look more into
while working further on the mechanism.
Currently, the change moves selinux_state structure to a separate
page. In
future we plan to move more security-related kernel assets to this
page to
enhance protection.
What's special about selinux_state? What about the SELinux policy?
How would I, as maintainer of the Smack security module, know if
some Smack data should be treated the same way?
We are investigating more of the SELinux related and other kernel assets
which can be included in the protection. The basis of selinux_state is
because disabling of SELinux is one of the common attack vectors in
Android. We understand any kernel assets, unauthorized changes to which
can give way to security or any other type of attack can be considered
to
be a potential asset to be added to the protection.
We want to seek your suggestions and comments on the idea and the
changes
in the patch.
Signed-off-by: Preeti Nagar <pnagar@xxxxxxxxxxxxxx>
---
include/asm-generic/vmlinux.lds.h | 10 ++++++++++
include/linux/init.h | 4 ++++
security/Kconfig | 10 ++++++++++
security/selinux/hooks.c | 4 ++++
4 files changed, 28 insertions(+)
diff --git a/include/asm-generic/vmlinux.lds.h
b/include/asm-generic/vmlinux.lds.h
index b2b3d81..158dbc2 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -770,6 +770,15 @@
*(.scommon) \
}
+#ifdef CONFIG_SECURITY_RTIC
+#define RTIC_BSS \
+ . = ALIGN(PAGE_SIZE); \
+ KEEP(*(.bss.rtic)) \
+ . = ALIGN(PAGE_SIZE);
+#else
+#define RTIC_BSS
+#endif
+
/*
* Allow archectures to redefine BSS_FIRST_SECTIONS to add extra
* sections to the front of bss.
@@ -782,6 +791,7 @@
. = ALIGN(bss_align); \
.bss : AT(ADDR(.bss) - LOAD_OFFSET) { \
BSS_FIRST_SECTIONS \
+ RTIC_BSS \
. = ALIGN(PAGE_SIZE); \
*(.bss..page_aligned) \
. = ALIGN(PAGE_SIZE); \
diff --git a/include/linux/init.h b/include/linux/init.h
index 7b53cb3..617adcf 100644
--- a/include/linux/init.h
+++ b/include/linux/init.h
@@ -300,6 +300,10 @@ void __init parse_early_options(char *cmdline);
/* Data marked not to be saved by software suspend */
#define __nosavedata __section(".data..nosave")
+#ifdef CONFIG_SECURITY_RTIC
+#define __rticdata __section(".bss.rtic")
+#endif
+
#ifdef MODULE
#define __exit_p(x) x
#else
diff --git a/security/Kconfig b/security/Kconfig
index 7561f6f..66b61b9 100644
--- a/security/Kconfig
+++ b/security/Kconfig
@@ -291,5 +291,15 @@ config LSM
source "security/Kconfig.hardening"
+config SECURITY_RTIC
+ bool "RunTime Integrity Check feature"
Shouldn't this depend on the architecture(s) supporting the
feature?
+ help
+ RTIC(RunTime Integrity Check) feature is to protect Linux kernel
+ at runtime. This relocates some of the security sensitive kernel
+ structures to a separate page aligned special section.
+
+ This is to enable monitoring and protection of these kernel assets
+ from a higher exception level(EL) against any unauthorized
changes.
"if you are unsure ..."
We just thought keeping it generic might be a better idea, thus, moved
the
changes to generic files from arch-specific files and thus, kept config
also
independent of the arch. Can surely make this config arch dependent if
that is
a better approach?
+
endmenu
diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index 6b1826f..7add17c 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c
@@ -104,7 +104,11 @@
#include "audit.h"
#include "avc_ss.h"
+#ifdef CONFIG_SECURITY_RTIC
+struct selinux_state selinux_state __rticdata;
+#else
struct selinux_state selinux_state;
+#endif
Shouldn't the __rticdata tag be applied always, and its
definition take care of the cases where it doesn't do anything?
Will update this change in the next version of the patch. Thank you.
/* SECMARK reference count */
static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
