Re: [PATCH security-next v3 18/29] LSM: Introduce lsm.enable= and lsm.disable=

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Oct 1, 2018 at 3:48 PM, John Johansen
<john.johansen@xxxxxxxxxxxxx> wrote:
> On 10/01/2018 03:27 PM, Kees Cook wrote:
>> On Mon, Oct 1, 2018 at 2:46 PM, John Johansen
>> <john.johansen@xxxxxxxxxxxxx> wrote:
>>> On 09/24/2018 05:18 PM, Kees Cook wrote:
>>>> This introduces the "lsm.enable=..." and "lsm.disable=..." boot parameters
>>>> which each can contain a comma-separated list of LSMs to enable or
>>>> disable, respectively. The string "all" matches all LSMs.
>>>>
>>>> This has very similar functionality to the existing per-LSM enable
>>>> handling ("apparmor.enabled=...", etc), but provides a centralized
>>>> place to perform the changes. These parameters take precedent over any
>>>> LSM-specific boot parameters.
>>>>
>>>> Disabling an LSM means it will not be considered when performing
>>>> initializations. Enabling an LSM means either undoing a previous
>>>> LSM-specific boot parameter disabling or a undoing a default-disabled
>>>> CONFIG setting.
>>>>
>>>> For example: "lsm.disable=apparmor apparmor.enabled=1" will result in
>>>> AppArmor being disabled. "selinux.enabled=0 lsm.enable=selinux" will
>>>> result in SELinux being enabled.
>>>>
>>>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>>>
>>> I don't like this. It brings about conflicting kernel params that are
>>> bound to confuse users. Its pretty easy for a user to understand that
>>> when they specify a parameter manually at boot, that  it overrides the
>>> build time default. But conflicting kernel parameters are a lot harder
>>> to deal with.
>>>
>>> I prefer a plain enabled= list being an override of the default build
>>> time value. Where conflicts with LSM-specific configs always result in
>>> the LSM being disabled with a complaint about the conflict.
>>>
>>> Though I have yet to be convinced its worth the cost, I do recognize
>>> it is sometimes convenient to disable a single LSM, instead of typing
>>> in a whole list of what to enable. If we have to have conflicting
>>> kernel parameters I would prefer that the conflict throw up a warning
>>> and leaving the LSM with the conflicting config disabled.
>>
>> Alright, let's drill down a bit more. I thought I had all the
>> requirements sorted out here. :)
>>
>> AppArmor and SELinux are "special" here in that they have both:
>>
>> - CONFIG for enable-ness
>> - boot param for enable-ness
>>
>> Now, the way this worked in the past was that combined with
>> CONFIG_DEFAULT_SECURITY and the link-time ordering, this resulted in a
>> way to get the LSM enabled, skipped, etc. But it was highly CONFIG
>> dependent.
>>
>> SELinux does:
>>
>> #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
>> int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
>>
>> static int __init selinux_enabled_setup(char *str)
>> {
>>         unsigned long enabled;
>>         if (!kstrtoul(str, 0, &enabled))
>>                 selinux_enabled = enabled ? 1 : 0;
>>         return 1;
>> }
>> __setup("selinux=", selinux_enabled_setup);
>> #else
>> int selinux_enabled = 1;
>> #endif
>> ...
>>         if (!security_module_enable("selinux")) {
>>                 selinux_enabled = 0;
>>                 return 0;
>>         }
>>
>>         if (!selinux_enabled) {
>>                 pr_info("SELinux:  Disabled at boot.\n");
>>                 return 0;
>>         }
>>
>>
>> AppArmor does:
>>
>> /* Boot time disable flag */
>> static bool apparmor_enabled = CONFIG_SECURITY_APPARMOR_BOOTPARAM_VALUE;
>> module_param_named(enabled, apparmor_enabled, bool, S_IRUGO);
>>
>> static int __init apparmor_enabled_setup(char *str)
>> {
>>         unsigned long enabled;
>>         int error = kstrtoul(str, 0, &enabled);
>>         if (!error)
>>                 apparmor_enabled = enabled ? 1 : 0;
>>         return 1;
>> }
>>
>> __setup("apparmor=", apparmor_enabled_setup);
>> ...
>>         if (!apparmor_enabled || !security_module_enable("apparmor")) {
>>                 aa_info_message("AppArmor disabled by boot time parameter");
>>                 apparmor_enabled = false;
>>                 return 0;
>>         }
>>
>>
>> Smack and TOMOYO each do:
>>
>>         if (!security_module_enable("smack"))
>>                 return 0;
>>
>>         if (!security_module_enable("tomoyo"))
>>                 return 0;
>>
>>
>> Capability, Integrity, Yama, and LoadPin always run init. (This series
>> fixes LoadPin to separate enable vs enforce, so we can ignore its
>> "enable" setting, which isn't an "am I active?" boolean -- its init
>> was always run.) With the enable logic is lifted out of the LSMs, we
>> want to have "implicit enable" for 6 of 8 of the LSMs. (Which is why I
>> had originally suggested CONFIG_LSM_DISABLE, since the normal state is
>> enabled.) But given your feedback, I made this "implicit disable" and
>> added CONFIG_LSM_ENABLE instead. (For which "CONFIG_LSM_ENABLE=all"
>> gets the same results.)
>>
>>
>> I think, then, the first question (mainly for you and Paul) is:
>>
>> Should we remove CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE and
>> CONFIG_SECURITY_APPARMOR_BOOTPARAM_VALUE in favor of only
>> CONFIG_LSM_ENABLE?
>>
>
> We can remove the Kconfig for the apparmor bootparam value. In fact I
> will attach that patch below. I can't get rid of the parameter as it
> is part of the userspace api. There are tools and applications
> checking /sys/module/apparmor/parameters/enabled
>
> but we can certainly default it to enabled and make it work only as a
> runtime kernel parameter to disable apparmor which is how it has been
> traditionally been used.
>
>> The answer will affect the next question: what should be done with the
>> boot parameters? AppArmor has two ways to change enablement:
>> apparmor=0/1 and apparmor.enabled=0/1. SELinux just has selinux=0/1.
>> Should those be removed in favor of "lsm.enable=..."? (And if they're
>> not removed, how do people imagine they should interact?)
>
> I am not against removing the apparmor one, it does mean retraining
> users but it is seldmon used so it may be worth dropping. If we keep
> it, it should be a disable only flag that where the use of apparmor=0
> or apparmor.enable=0 (same thing) means apparmor is disabled.

If we keep it, "apparmor=0 lsm_enable=apparmor" would mean it's
enabled. Is that okay?

> ---
>
> commit 367b8a47105c68fa170bdd14b0204555eb930476
> Author: John Johansen <john.johansen@xxxxxxxxxxxxx>
> Date:   Mon Oct 1 15:46:02 2018 -0700
>
>     apparmor: remove apparmor boot param config
>
>     The boot param value is only ever used as a means to disable apparmor.
>     Get rid of the Kconfig and a default the parameter to true.
>
>     Signed-off-by: John Johansen <john.johansen@xxxxxxxxxxxxx>
>
> diff --git a/security/apparmor/Kconfig b/security/apparmor/Kconfig
> index b6b68a7750ce..3de21f46c82a 100644
> --- a/security/apparmor/Kconfig
> +++ b/security/apparmor/Kconfig
> @@ -14,22 +14,6 @@ config SECURITY_APPARMOR
>
>           If you are unsure how to answer this question, answer N.
>
> -config SECURITY_APPARMOR_BOOTPARAM_VALUE
> -       int "AppArmor boot parameter default value"
> -       depends on SECURITY_APPARMOR
> -       range 0 1
> -       default 1
> -       help
> -         This option sets the default value for the kernel parameter
> -         'apparmor', which allows AppArmor to be enabled or disabled
> -          at boot.  If this option is set to 0 (zero), the AppArmor
> -         kernel parameter will default to 0, disabling AppArmor at
> -         boot.  If this option is set to 1 (one), the AppArmor
> -         kernel parameter will default to 1, enabling AppArmor at
> -         boot.
> -
> -         If you are unsure how to answer this question, answer 1.
> -
>  config SECURITY_APPARMOR_HASH
>         bool "Enable introspection of sha1 hashes for loaded profiles"
>         depends on SECURITY_APPARMOR
> diff --git a/security/apparmor/lsm.c b/security/apparmor/lsm.c
> index f09fea0b4db7..8e83ee52a0a3 100644
> --- a/security/apparmor/lsm.c
> +++ b/security/apparmor/lsm.c
> @@ -1303,7 +1303,7 @@ bool aa_g_paranoid_load = true;
>  module_param_named(paranoid_load, aa_g_paranoid_load, aabool, S_IRUGO);
>
>  /* Boot time disable flag */
> -static bool apparmor_enabled = CONFIG_SECURITY_APPARMOR_BOOTPARAM_VALUE;
> +static bool apparmor_enabled = true;

In the new world, this wouldn't be "= true" since its state would be
controlled by CONFIG_LSM_ENABLE (and lsm.enable=...). Is that okay?

>  module_param_named(enabled, apparmor_enabled, bool, S_IRUGO);
>
>  static int __init apparmor_enabled_setup(char *str)

I'll add this to the series, thanks!

-Kees

-- 
Kees Cook
Pixel Security



[Index of Archives]     [Linux Kernel]     [Kernel Newbies]     [x86 Platform Driver]     [Netdev]     [Linux Wireless]     [Netfilter]     [Bugtraq]     [Linux Filesystems]     [Yosemite Discussion]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Device Mapper]

  Powered by Linux