Re: [PATCH] random: Initialize vsprintf's pointer hash once the random core is ready.

[Petr Mladek]

On Fri 2022-07-29 10:52:58, Sebastian Andrzej Siewior wrote:
> The printk code invokes vnsprintf in order to compute the complete
> string before adding it into its buffer. This happens in an IRQ-off
> region which leads to a warning on PREEMPT_RT in the random code if the
> format strings contains a %p for pointer printing. This happens because
> the random core acquires locks which become sleeping locks on PREEMPT_RT
> which must not be acquired with disabled interrupts and or preemption.
> By default the pointers are hashed which requires a random value on the
> first invocation (either by printk or another user which comes first.
> 
> One could argue that there is no need for printk to disable interrupts
> during the vsprintf() invocation which would fix the just mentioned
> problem. However printk itself can be invoked in a context with
> disabled interrupts which would lead to the very same problem.
> 
> This late init via printk can be avoided by explicitly initializing
> vsprintf's random value once the random-core has been initialized.
> 
> Remove the on demand init from __ptr_to_hashval() and keep the -EAGAIN if
> the init has not yet been performed. Move the actual init bits to
> vsprintf_init_hash_pointer() which are invoked from random-core once it
> has been initialized and get_random_bytes() is available.

> --- a/drivers/char/random.c
> +++ b/drivers/char/random.c
> @@ -221,10 +222,15 @@ static void crng_reseed(void)
>  		++next_gen;
>  	WRITE_ONCE(base_crng.generation, next_gen);
>  	WRITE_ONCE(base_crng.birth, jiffies);
> -	if (!static_branch_likely(&crng_is_ready))
> +	if (!static_branch_likely(&crng_is_ready)) {
>  		crng_init = CRNG_READY;
> +		init_hash_pointer = true;

I am not familiar with the crng code. I wonder if the following would work:

	if (!static_branch_likely(&crng_is_ready) && crng_init != CRNG_READY) {
		crng_init = CRNG_READY;
		init_hash_pointer = true;
	}

The point is that vsprintf_init_hash_pointer() will be called only by
the first caller. It would allow to remove the @filling spin lock.

> +	}
>  	spin_unlock_irqrestore(&base_crng.lock, flags);
>  	memzero_explicit(key, sizeof(key));
> +
> +	if (init_hash_pointer)
> +		vsprintf_init_hash_pointer();
>  }
>  
>  /*
> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
> index 3c1853a9d1c09..6fa2ebb9f9b9e 100644
> --- a/lib/vsprintf.c
> +++ b/lib/vsprintf.c
> @@ -751,36 +751,30 @@ static int __init debug_boot_weak_hash_enable(char *str)
>  early_param("debug_boot_weak_hash", debug_boot_weak_hash_enable);
>  
>  static DEFINE_STATIC_KEY_FALSE(filled_random_ptr_key);
> +static siphash_key_t ptr_key __read_mostly;
>  
> -static void enable_ptr_key_workfn(struct work_struct *work)
> +void vsprintf_init_hash_pointer(void)
>  {
> -	static_branch_enable(&filled_random_ptr_key);
> +	static DEFINE_SPINLOCK(filling);
> +	unsigned long flags;
> +	static bool filled;
> +
> +	spin_lock_irqsave(&filling, flags);
> +	if (!filled) {
> +		get_random_bytes(&ptr_key, sizeof(ptr_key));
> +		filled = true;
> +		static_branch_enable(&filled_random_ptr_key);

This can't be called in an atomic context. Is crng_reseed() always
called in a non-atomic context?

That said, the static branch is an overkill. vsprintf() is a slow
path. It should be enough to use a simple boolean. It might require
a simple memory barrier to serialize @ptr_key and the new boolean
read&write.

> +	}
> +	spin_unlock_irqrestore(&filling, flags);
>  }

Best Regards,
Petr