Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <cd6e153b-953f-0374-364e-af649c82d8b4@rasmusvillemoes.dk>
Date:   Fri, 18 Nov 2022 09:33:53 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.2
Subject: Re: [PATCH] of: fdt: parse early params before adding bootloader
 randomness
Content-Language: en-US, da
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>,
        Rob Herring <robh@kernel.org>
Cc:     devicetree@vger.kernel.org, linux-kernel@vger.kernel.org,
        Frank Rowand <frowand.list@gmail.com>
References: <20221105014613.113503-1-Jason@zx2c4.com>
 <CAL_JsqKA1_HV5V17mHkKn8X72c_UN2Jg=TYtJkt93YM6SSDMSg@mail.gmail.com>
 <Y2lBj1ZIdFRf9HdR@zx2c4.com>
From:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
In-Reply-To: <Y2lBj1ZIdFRf9HdR@zx2c4.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 07/11/2022 18.34, Jason A. Donenfeld wrote:
> On Mon, Nov 07, 2022 at 11:28:20AM -0600, Rob Herring wrote:
>> On Fri, Nov 4, 2022 at 8:46 PM Jason A. Donenfeld <Jason@zx2c4.com> wrote:
>>>
>>> FDT is examined so early that it's before the first incidental call to
>>> parse_early_param(). This is similar to EFI, except EFI actually added
>>> an explicitly call to parse_early_param(). Let's do the same here, so
>>> that specifying `random.trust_bootloader=0` is not ignored.
>>>
>>> Fixes: d97c68d178fb ("random: treat bootloader trust toggle the same way as cpu trust toggle")
>>> Cc: Rob Herring <robh@kernel.org>
>>> Cc: Frank Rowand <frowand.list@gmail.com>
>>> Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
>>> ---
>>>  drivers/of/fdt.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
>>> index 7b571a631639..6d959117fd4f 100644
>>> --- a/drivers/of/fdt.c
>>> +++ b/drivers/of/fdt.c
>>> @@ -1195,6 +1195,9 @@ int __init early_init_dt_scan_chosen(char *cmdline)
>>>
>>>         rng_seed = of_get_flat_dt_prop(node, "rng-seed", &l);
>>>         if (rng_seed && l > 0) {
>>> +               /* Parse random.trust_bootloader if it's in command line. */
>>> +               parse_early_param();
>>
>> I don't think it's good that the timing of calling this is dependent
>> on "rng-seed" being present or not. So perhaps move it up to after the
>> cmdline is set.
>>
>> Either way, the other issue is the cmdline is not necessarily fixed at
>> this point with some architectures doing their own
>> append/prepend/override of the cmdline. We can't seem to get common
>> implementation there finished. I'm doubtful that corner case would
>> actually be hit though.
> 
> Hm, yea. I'm actually now having second thoughts about this one too for
> other reasons: FDT isn't the only arch that has this issue. It's also a
> problem on x86 and m68k. Maybe the random.trust_bootloader toggle should
> just go away, since already your bootloader can do whatever it wants to
> the kernel it executes? Not sure; I'll think on it a bit I guess...

Well, for the FDT case it least one could preserve the ability for the
bootloader to say "hey, here's a bunch of bytes, but I have reason to
believe they're not as random as you'd like" by setting a boolean
"rng-seed-is-crap" [pick a better name] property alongside the rng-seed
property. Then the kernel logic could be CONFIG_TRUST_BOOT_LOADER &&
!of_get_bool("rng-seed-is-crap"). Currently, the only thing the
bootloader could do in that case is to just elide the rng-seed
completely (or set the not-actually-working cmdline flag).

I don't think it's a loss to not have a way for the bootloader to turn
on a trust-me flag.

Rasmus