Hi,
This is v3 of the series to add the kernel command line as a source
of randomness. The main change from v2 is to correctly place the
command line randomness _before_ the stack canary initialization
so the canary can take advantage of that.
Daniel Micay (1):
extract early boot entropy from the passed cmdline
Laura Abbott (1):
init: Move stack canary initialization after setup_arch
init/main.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)
--
2.13.0
From: Laura Abbott <[email protected]>
Stack canary intialization involves getting a random number.
Getting this random number may involve accessing caches or other
architectural specific features which are not available until
after the architecture is setup. Move the stack canary initialization
later to accomodate this.
Acked-by: Kees Cook <[email protected]>
Signed-off-by: Laura Abbott <[email protected]>
Signed-off-by: Laura Abbott <[email protected]>
---
v3: Ack from Kees
---
init/main.c | 11 +++++------
1 file changed, 5 insertions(+), 6 deletions(-)
diff --git a/init/main.c b/init/main.c
index 052481fbe363..21d599eaad06 100644
--- a/init/main.c
+++ b/init/main.c
@@ -515,12 +515,6 @@ asmlinkage __visible void __init start_kernel(void)
smp_setup_processor_id();
debug_objects_early_init();
- /*
- * Set up the initial canary ASAP:
- */
- add_latent_entropy();
- boot_init_stack_canary();
-
cgroup_init_early();
local_irq_disable();
@@ -534,6 +528,11 @@ asmlinkage __visible void __init start_kernel(void)
page_address_init();
pr_notice("%s", linux_banner);
setup_arch(&command_line);
+ /*
+ * Set up the the initial canary and entropy after arch
+ */
+ add_latent_entropy();
+ boot_init_stack_canary();
mm_init_cpumask(&init_mm);
setup_command_line(command_line);
setup_nr_cpu_ids();
--
2.13.0
From: Daniel Micay <[email protected]>
Existing Android bootloaders usually pass data useful as early entropy
on the kernel command-line. It may also be the case on other embedded
systems. Sample command-line from a Google Pixel running CopperheadOS:
console=ttyHSL0,115200,n8 androidboot.console=ttyHSL0
androidboot.hardware=sailfish user_debug=31 ehci-hcd.park=3
lpm_levels.sleep_disabled=1 cma=32M@0-0xffffffff buildvariant=user
veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab
androidboot.bootdevice=624000.ufshc androidboot.verifiedbootstate=yellow
androidboot.veritymode=enforcing androidboot.keymaster=1
androidboot.serialno=FA6CE0305299 androidboot.baseband=msm
mdss_mdp.panel=1:dsi:0:qcom,mdss_dsi_samsung_ea8064tg_1080p_cmd:1:none:cfg:single_dsi
androidboot.slot_suffix=_b fpsimd.fpsimd_settings=0
app_setting.use_app_setting=0 kernelflag=0x00000000 debugflag=0x00000000
androidboot.hardware.revision=PVT radioflag=0x00000000
radioflagex1=0x00000000 radioflagex2=0x00000000 cpumask=0x00000000
androidboot.hardware.ddr=4096MB,Hynix,LPDDR4 androidboot.ddrinfo=00000006
androidboot.ddrsize=4GB androidboot.hardware.color=GRA00
androidboot.hardware.ufs=32GB,Samsung androidboot.msm.hw_ver_id=268824801
androidboot.qf.st=2 androidboot.cid=11111111 androidboot.mid=G-2PW4100
androidboot.bootloader=8996-012001-1704121145
androidboot.oem_unlock_support=1 androidboot.fp_src=1
androidboot.htc.hrdump=detected androidboot.ramdump.opt=mem@2g:2g,mem@4g:2g
androidboot.bootreason=reboot androidboot.ramdump_enable=0 ro
root=/dev/dm-0 dm="system none ro,0 1 android-verity /dev/sda34"
rootwait skip_initramfs init=/init androidboot.wificountrycode=US
androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136
Among other things, it contains a value unique to the device
(androidboot.serialno=FA6CE0305299), unique to the OS builds for the
device variant (veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab)
and timings from the bootloader stages in milliseconds
(androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136).
Signed-off-by: Daniel Micay <[email protected]>
[labbott: Line-wrapped command line]
Signed-off-by: Laura Abbott <[email protected]>
---
v3: add_device_randomness comes before canary initialization, clarified comment.
---
init/main.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/init/main.c b/init/main.c
index 21d599eaad06..ba2b3a8a2382 100644
--- a/init/main.c
+++ b/init/main.c
@@ -530,8 +530,10 @@ asmlinkage __visible void __init start_kernel(void)
setup_arch(&command_line);
/*
* Set up the the initial canary and entropy after arch
+ * and after adding latent and command line entropy.
*/
add_latent_entropy();
+ add_device_randomness(command_line, strlen(command_line));
boot_init_stack_canary();
mm_init_cpumask(&init_mm);
setup_command_line(command_line);
--
2.13.0
On Wed, Aug 16, 2017 at 4:14 PM, Laura Abbott <[email protected]> wrote:
> From: Daniel Micay <[email protected]>
>
>
> Existing Android bootloaders usually pass data useful as early entropy
> on the kernel command-line. It may also be the case on other embedded
> systems. Sample command-line from a Google Pixel running CopperheadOS:
>
> console=ttyHSL0,115200,n8 androidboot.console=ttyHSL0
> androidboot.hardware=sailfish user_debug=31 ehci-hcd.park=3
> lpm_levels.sleep_disabled=1 cma=32M@0-0xffffffff buildvariant=user
> veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab
> androidboot.bootdevice=624000.ufshc androidboot.verifiedbootstate=yellow
> androidboot.veritymode=enforcing androidboot.keymaster=1
> androidboot.serialno=FA6CE0305299 androidboot.baseband=msm
> mdss_mdp.panel=1:dsi:0:qcom,mdss_dsi_samsung_ea8064tg_1080p_cmd:1:none:cfg:single_dsi
> androidboot.slot_suffix=_b fpsimd.fpsimd_settings=0
> app_setting.use_app_setting=0 kernelflag=0x00000000 debugflag=0x00000000
> androidboot.hardware.revision=PVT radioflag=0x00000000
> radioflagex1=0x00000000 radioflagex2=0x00000000 cpumask=0x00000000
> androidboot.hardware.ddr=4096MB,Hynix,LPDDR4 androidboot.ddrinfo=00000006
> androidboot.ddrsize=4GB androidboot.hardware.color=GRA00
> androidboot.hardware.ufs=32GB,Samsung androidboot.msm.hw_ver_id=268824801
> androidboot.qf.st=2 androidboot.cid=11111111 androidboot.mid=G-2PW4100
> androidboot.bootloader=8996-012001-1704121145
> androidboot.oem_unlock_support=1 androidboot.fp_src=1
> androidboot.htc.hrdump=detected androidboot.ramdump.opt=mem@2g:2g,mem@4g:2g
> androidboot.bootreason=reboot androidboot.ramdump_enable=0 ro
> root=/dev/dm-0 dm="system none ro,0 1 android-verity /dev/sda34"
> rootwait skip_initramfs init=/init androidboot.wificountrycode=US
> androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136
>
> Among other things, it contains a value unique to the device
> (androidboot.serialno=FA6CE0305299), unique to the OS builds for the
> device variant (veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab)
> and timings from the bootloader stages in milliseconds
> (androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136).
>
> Signed-off-by: Daniel Micay <[email protected]>
> [labbott: Line-wrapped command line]
> Signed-off-by: Laura Abbott <[email protected]>
Acked-by: Kees Cook <[email protected]>
Thanks!
-Kees
> ---
> v3: add_device_randomness comes before canary initialization, clarified comment.
> ---
> init/main.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/init/main.c b/init/main.c
> index 21d599eaad06..ba2b3a8a2382 100644
> --- a/init/main.c
> +++ b/init/main.c
> @@ -530,8 +530,10 @@ asmlinkage __visible void __init start_kernel(void)
> setup_arch(&command_line);
> /*
> * Set up the the initial canary and entropy after arch
> + * and after adding latent and command line entropy.
> */
> add_latent_entropy();
> + add_device_randomness(command_line, strlen(command_line));
> boot_init_stack_canary();
> mm_init_cpumask(&init_mm);
> setup_command_line(command_line);
> --
> 2.13.0
>
--
Kees Cook
Pixel Security
On Wed, Aug 16, 2017 at 04:14:58PM -0700, Laura Abbott wrote:
> From: Daniel Micay <[email protected]>
>
> Existing Android bootloaders usually pass data useful as early entropy
> on the kernel command-line. It may also be the case on other embedded
> systems.....
May I suggest a slight adjustment to the beginning commit description?
Feed the boot command-line as to the /dev/random entropy pool
Existing Android bootloaders usually pass data which may not be
known by an external attacker on the kernel command-line. It may
also be the case on other embedded systems. Sample command-line
from a Google Pixel running CopperheadOS....
The idea here is to if anything, err on the side of under-promising
the amount of security we can guarantee that this technique will
provide. For example, how hard is it really for an attacker who has
an APK installed locally to get the device serial number? Or the OS
version? And how much variability is there in the bootloader stages
in milliseconds?
I think we should definitely do this. So this is more of a request to
be very careful what we promise in the commit description, not an
objection to the change itself.
Cheers,
- Ted
On Wed, 2017-08-16 at 23:31 -0400, Theodore Ts'o wrote:
> On Wed, Aug 16, 2017 at 04:14:58PM -0700, Laura Abbott wrote:
> > From: Daniel Micay <[email protected]>
> >
> > Existing Android bootloaders usually pass data useful as early
> > entropy
> > on the kernel command-line. It may also be the case on other
> > embedded
> > systems.....
>
> May I suggest a slight adjustment to the beginning commit description?
>
> Feed the boot command-line as to the /dev/random entropy pool
>
> Existing Android bootloaders usually pass data which may not be
> known by an external attacker on the kernel command-line. It may
> also be the case on other embedded systems. Sample command-line
> from a Google Pixel running CopperheadOS....
>
> The idea here is to if anything, err on the side of under-promising
> the amount of security we can guarantee that this technique will
> provide. For example, how hard is it really for an attacker who has
> an APK installed locally to get the device serial number? Or the OS
> version? And how much variability is there in the bootloader stages
> in milliseconds?
The serial number is currently accessible to local apps up until Android
7.x so it doesn't have value if the adversary has local access. Access
to it without the READ_PHONE_STATE permission is being removed for apps
targeting Android 8.0 and will presumably be restructed for all apps at
some point in the future:
https://android-developers.googleblog.com/2017/04/changes-to-device-identifiers-in.html
Some bootloader stages vary a bit in time each boot. There's not much
variance or measurement precision so there's only a small amount of
entropy from this. The ones that consistently vary in timing do so
independently from each other so that helps a bit. Also worth noting
that before Android 8.0+, local apps can access the boot times since
it's written to a system property. After Android 8.0+, all that stuff is
inaccessible to them (no permission to get them) since there's a
whitelisting model for system property access.
> I think we should definitely do this. So this is more of a request to
> be very careful what we promise in the commit description, not an
> objection to the change itself.
I did say 'external attacker' but it could be made clearer. It's
primarily aimed at getting a tiny bit of extra entropy for the kernel
stack canary and other probabilistic exploit mitigations set up in early
boot. On non-x86 archs, i.e. 99.9% of Android devices, the kernel stack
canary remains the same after it's set up in that early boot code.
Android devices almost all have a hardware RNG and Android init blocks
until a fair bit of data is read from it along with restoring entropy
that's regularly saved while running, but unfortunately that's not
available at this point in the boot process.
The kernel could save / restore entropy using pstore (which at least
Nexus / Pixel devices have - not sure about others). I don't know how
early that could feasibly be done. Ideally it would do that combined
with early usage of the hwrng.
> I did say 'external attacker' but it could be made clearer.
Er, s/say/mean to imply/
I do think it will have some local value after Android 8 which should
start shipping in a few days though.
I'll look into having the kernel stash some entropy in pstore soon since
that seems like it could be a great improvement. I'm not sure how often
/ where it should hook into for regularly refreshing it though. Doing it
only on powering down isn't ideal.
On Thu, Aug 17, 2017 at 04:57:07PM -0400, Daniel Micay wrote:
> > I did say 'external attacker' but it could be made clearer.
>
> Er, s/say/mean to imply/
Right, that's why I had suggested modifying the first few lines of the
commit description to read something like this:
Feed the boot command-line as to the /dev/random entropy pool
Existing Android bootloaders usually pass data which may not be known
by an external attacker on the kernel command-line. It may also be
the case on other embedded systems. Sample command-line from a Google
Pixel running CopperheadOS:
(Or something like that.)
> I'll look into having the kernel stash some entropy in pstore soon since
> that seems like it could be a great improvement. I'm not sure how often
> / where it should hook into for regularly refreshing it though. Doing it
> only on powering down isn't ideal.
One thing we could do is to agree on a standard place where the
entropy would be stashed, and then have the kernel remove it from
being visible in /proc/cmdline. That's not a perfect answer, since
the user might be able to look at the command line via other
mechanisms. (For example, on x86, by looking at GRUB while the system
is booting.)
However, an attacker who is merely running code on the local system is
not likely to be gain access to that value --- so it's definitely an
improvement.
Refreshing the entry immediately after boot, and before a clean
shutdown would be ideal from a security perspective. I don't know if
there are write endurance issues with updating the pstore that
frequently, though.
- Ted
On Wed 2017-08-16 23:31:48, Theodore Ts'o wrote:
> On Wed, Aug 16, 2017 at 04:14:58PM -0700, Laura Abbott wrote:
> > From: Daniel Micay <[email protected]>
> >
> > Existing Android bootloaders usually pass data useful as early entropy
> > on the kernel command-line. It may also be the case on other embedded
> > systems.....
>
> May I suggest a slight adjustment to the beginning commit description?
>
> Feed the boot command-line as to the /dev/random entropy pool
>
> Existing Android bootloaders usually pass data which may not be
> known by an external attacker on the kernel command-line. It may
> also be the case on other embedded systems. Sample command-line
> from a Google Pixel running CopperheadOS....
>
> The idea here is to if anything, err on the side of under-promising
> the amount of security we can guarantee that this technique will
> provide. For example, how hard is it really for an attacker who has
> an APK installed locally to get the device serial number? Or the OS
> version? And how much variability is there in the bootloader stages
> in milliseconds?
>
> I think we should definitely do this. So this is more of a request to
> be very careful what we promise in the commit description, not an
> objection to the change itself.
The command line is visible to unpriviledged userspace (/proc/cmdline,
dmesg). Is that a problem?
U-boot already does some crypto stuff, so it may have some
randomness. Should we create parameter random=xxxxxxxxxxx that is
"censored" during kernel boot?
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
On Wed, Aug 30, 2017 at 2:57 AM, Pavel Machek <[email protected]> wrote:
> The command line is visible to unpriviledged userspace (/proc/cmdline,
> dmesg). Is that a problem?
These files are not exposed to untrusted processes on Android.
--
Nick Kralevich | Android Security | [email protected] | 650.214.4037