Re: [RFC] kernel/sysctl: support setting sysctl parameters from kernel command line

[Vlastimil Babka]

On 3/17/20 10:29 PM, Kees Cook wrote:
> On Tue, Mar 17, 2020 at 02:21:05PM +0100, Vlastimil Babka wrote:
>> A recently proposed patch to add vm_swappiness command line parameter in
>> addition to existing sysctl [1] made me wonder why we don't have a general
>> support for passing sysctl parameters via command line. Googling found only
>> somebody else wondering the same [2], but I haven't found any prior discussion
>> with reasons why not to do this.
> 
> I'd like to see stuff like this (as you say, you've found some
> redundancies here which could be cleaned up a bit). I think the reason
> it hasn't happened before is that the answers have mostly revolved
> around "just set it in your initramfs". :P
> 
>> [...]
>> Hence, this patch adds a new parse_args() pass that looks for parameters
>> prefixed by 'sysctl.' and searches for them in the sysctl ctl_tables. When
>> found, the respective proc handler is invoked. The search is just a naive
>> linear one, to avoid using the whole procfs layer. It should be acceptable,
>> as the cost depends on number of sysctl. parameters passed.
> 
> I think this needs reconsidering: this RFC only searches 1 level deep,
> but sysctls are a tree. For example:

Yes, that was an oversight, easily fixed.

> kernel.yama.ptrace_scope
> mm.transparent_hugepage.enabled

Hm, transparent_hugepage is in sysfs (/sys/kernel/mm), but not sysctl, at least
in my case the sysctl tool doesn't list it. Yours does? Yay for consistency.

> net.ipv4.conf.default.rp_filter
> ...etc
> 
> If this goes in, it'll need to do full traversal.

Right.

>> The main limitation of avoiding the procfs layer is however that sysctls
>> dynamically registered by register_sysctl_table() or register_sysctl_paths()
>> cannot be set by this method.
> 
> Correct. And I like what you've done in the code: announce any unhandled
> sysctls.
> 
>> The processing is hooked right before the init process is loaded, as some
>> handlers might be more complicated than simple setters and might need some
>> subsystems to be initialized. At the moment the init process can be started and
>> eventually execute a process writing to /proc/sys/ then it should be also fine
>> to do that from the kernel.
> 
> I agree about placement.
> 
>> 
>> [1] https://lore.kernel.org/linux-doc/BL0PR02MB560167492CA4094C91589930E9FC0@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>> [2] https://unix.stackexchange.com/questions/558802/how-to-set-sysctl-using-kernel-command-line-parameter
>> 
>> Signed-off-by: Vlastimil Babka <vbabka@xxxxxxx>
>> ---
>> Hi,
>> 
>> this is an early RFC so I can get feedback whether to pursue this idea further,
>> before trying the more complicated stuff with dynamically registered sysctls.
>> For those I have some unanswered questions:
>> - Support them at all?
> 
> Maybe? It seems excessive for the initial version.

OK

>> - Do so by an internal procfs mount again, that was removed by 61a47c1ad3a4 ?
>>   Or try to keep it simple.
> 
> I think you can walk the registered sysctl structures themselves, yes?

I should be able to, yeah.

>> - If sysctls are dynamically registered at module load, process the command
>>   line sysctl arguments again? - this would be rather complicated I guess.
> 
> If it does get supported, perhaps saving them somewhere for
> register_sysctl_table() to walk when it gets called?
> 
> I like the idea if just for having to build less boiler plate for
> supporting things that I've had to plumb to both boot_params and sysctl.
> :)

Thanks, I will pursue the idea further then :)

Vlastimil