Re: [PATCH v2 1/4] seccomp: Add sysctl to display available actions

From: Tyler Hicks
Date: Mon Feb 13 2017 - 19:25:25 EST

On 02/07/2017 06:43 PM, Kees Cook wrote:
> On Tue, Feb 7, 2017 at 4:25 PM, Tyler Hicks <tyhicks@xxxxxxxxxxxxx> wrote:
>> On 02/07/2017 06:03 PM, Kees Cook wrote:
>>> On Thu, Feb 2, 2017 at 9:37 PM, Tyler Hicks <tyhicks@xxxxxxxxxxxxx> wrote:
>>>> This patch creates a read-only sysctl containing an ordered list of
>>>> seccomp actions that the kernel supports. The ordering, from left to
>>>> right, is the lowest action value (kill) to the highest action value
>>>> (allow). Currently, a read of the sysctl file would return "kill trap
>>>> errno trace allow". The contents of this sysctl file can be useful for
>>>> userspace code as well as the system administrator.
>>>> The path to the sysctl is:
>>>> /proc/sys/kernel/seccomp/actions_avail
>>>> libseccomp and other userspace code can easily determine which actions
>>>> the current kernel supports. The set of actions supported by the current
>>>> kernel may be different than the set of action macros found in kernel
>>>> headers that were installed where the userspace code was built.
>>> This is certainly good: having a discoverable way to detect filter
>>> capabilities. I do wonder if it'd still be easier to just expose the
>>> max_log sysctl as a numeric value, since the SECCOMP_RET_* values are
>>> all part of uapi, so we can't escape their values...
>> I was very torn on whether to use a numeric or string representation
>> here. The reason I decided on string representation is because I think
>> these sysctls are mostly aimed for admins and numeric representations
>> wouldn't be easy to use. I considered added a utility to libseccomp but,
>> since the kernel code to do a string representation was so simple, I
>> went with doing it in the kernel.
> Yeah, I think I like it just because it gives a way to discover the
> UAPI "level"... I will think more about this. For v3, let's keep the
> string stuff.
>> Another possibility is exposing the SECCOMP_RET_*_NAME macros as part of
>> the uapi.
> I like keeping the UAPI minimal. ;)
>>>> +static int __init seccomp_sysctl_init(void)
>>>> +{
>>>> + struct ctl_table_header *hdr;
>>>> +
>>>> + hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
>>>> + kmemleak_not_leak(hdr);
>>> Will kmemleak complain about this if hdr is saved to a global (or not
>>> saved at all)? Also, something should be reported in the failure
>>> case...
>> I have to admit to blindly following the example set by sysctl_init() in
>> kernel/sysctl.c. I can test what kmemleak will/won't complain about and
>> report back (tomorrow at the earliest).
> Cool, no rush. I'm backlogged on reviews anyway. :)

kmemleak doesn't complain if we save it to a global. That makes sense
because it means that we have a persistent reference to the allocated

However, kmemleak doesn't complain about this allocation as-is (meaning
that I simply removed the call to kmemleak_not_leak()). From what I can
tell, this is because a reference to the allocated ctl_table_header
struct is saved when __register_sysctl_table() calls init_header(). I
think kmemleak is seeing this reference when doing scans and
(incorrectly) thinking that there's no leak.

I think the safest/cleanest thing to do is leave the call to
kmemleak_not_leak(). Let me know if you disagree.


Attachment: signature.asc
Description: OpenPGP digital signature