Now that we have an explicit base_crng generation counter, we don't need
a separate one for batched entropy. Rather, we can just move the
generation forward every time we change crng_init state.

Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
---
 drivers/char/random.c | 28 +++++++---------------------
 1 file changed, 7 insertions(+), 21 deletions(-)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 999f1d164e72..f4d432305869 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -431,8 +431,6 @@ static DEFINE_PER_CPU(struct crng, crngs) = {
 
 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 
-static void invalidate_batched_entropy(void);
-
 /*
  * crng_fast_load() can be called by code in the interrupt service
  * path.  So we can't afford to dilly-dally. Returns the number of
@@ -455,7 +453,7 @@ static size_t crng_fast_load(const void *cp, size_t len)
 		src++; crng_init_cnt++; len--; ret++;
 	}
 	if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
-		invalidate_batched_entropy();
+		++base_crng.generation;
 		crng_init = 1;
 	}
 	spin_unlock_irqrestore(&base_crng.lock, flags);
@@ -536,7 +534,6 @@ static void crng_reseed(void)
 	spin_unlock_irqrestore(&base_crng.lock, flags);
 	memzero_explicit(key, sizeof(key));
 	if (finalize_init) {
-		invalidate_batched_entropy();
 		process_random_ready_list();
 		wake_up_interruptible(&crng_init_wait);
 		kill_fasync(&fasync, SIGIO, POLL_IN);
@@ -1278,7 +1275,7 @@ int __init rand_initialize(void)
 
 	extract_entropy(base_crng.key, sizeof(base_crng.key));
 	if (arch_init && trust_cpu && crng_init < 2) {
-		invalidate_batched_entropy();
+		++base_crng.generation;
 		crng_init = 2;
 		pr_notice("crng init done (trusting CPU's manufacturer)\n");
 	}
@@ -1628,8 +1625,6 @@ static int __init random_sysctls_init(void)
 device_initcall(random_sysctls_init);
 #endif	/* CONFIG_SYSCTL */
 
-static atomic_t batch_generation = ATOMIC_INIT(0);
-
 struct batched_entropy {
 	union {
 		/* We make this 1.5x a ChaCha block, so that we get the
@@ -1642,8 +1637,8 @@ struct batched_entropy {
 		u32 entropy_u32[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u32))];
 	};
 	local_lock_t lock;
+	unsigned long generation;
 	unsigned int position;
-	int generation;
 };
 
 /*
@@ -1662,14 +1657,14 @@ u64 get_random_u64(void)
 	unsigned long flags;
 	struct batched_entropy *batch;
 	static void *previous;
-	int next_gen;
+	unsigned long next_gen;
 
 	warn_unseeded_randomness(&previous);
 
 	local_lock_irqsave(&batched_entropy_u64.lock, flags);
 	batch = raw_cpu_ptr(&batched_entropy_u64);
 
-	next_gen = atomic_read(&batch_generation);
+	next_gen = READ_ONCE(base_crng.generation);
 	if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0 ||
 	    next_gen != batch->generation) {
 		_get_random_bytes(batch->entropy_u64, sizeof(batch->entropy_u64));
@@ -1695,14 +1690,14 @@ u32 get_random_u32(void)
 	unsigned long flags;
 	struct batched_entropy *batch;
 	static void *previous;
-	int next_gen;
+	unsigned long next_gen;
 
 	warn_unseeded_randomness(&previous);
 
 	local_lock_irqsave(&batched_entropy_u32.lock, flags);
 	batch = raw_cpu_ptr(&batched_entropy_u32);
 
-	next_gen = atomic_read(&batch_generation);
+	next_gen = READ_ONCE(base_crng.generation);
 	if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0 ||
 	    next_gen != batch->generation) {
 		_get_random_bytes(batch->entropy_u32, sizeof(batch->entropy_u32));
@@ -1718,15 +1713,6 @@ u32 get_random_u32(void)
 }
 EXPORT_SYMBOL(get_random_u32);
 
-/* It's important to invalidate all potential batched entropy that might
- * be stored before the crng is initialized, which we can do lazily by
- * bumping the generation counter.
- */
-static void invalidate_batched_entropy(void)
-{
-	atomic_inc(&batch_generation);
-}
-
 /**
  * randomize_page - Generate a random, page aligned address
  * @start:	The smallest acceptable address the caller will take.
-- 
2.35.0

Am Wed, Feb 09, 2022 at 10:54:06PM +0100 schrieb Jason A. Donenfeld:
> Now that we have an explicit base_crng generation counter, we don't need
> a separate one for batched entropy. Rather, we can just move the
> generation forward every time we change crng_init state.
> 
> Cc: Dominik Brodowski <linux@dominikbrodowski.net>
> Cc: Theodore Ts'o <tytso@mit.edu>
> Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
> ---
>  drivers/char/random.c | 28 +++++++---------------------
>  1 file changed, 7 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/char/random.c b/drivers/char/random.c
> index 999f1d164e72..f4d432305869 100644
> --- a/drivers/char/random.c
> +++ b/drivers/char/random.c
> @@ -431,8 +431,6 @@ static DEFINE_PER_CPU(struct crng, crngs) = {
>  
>  static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
>  
> -static void invalidate_batched_entropy(void);
> -
>  /*
>   * crng_fast_load() can be called by code in the interrupt service
>   * path.  So we can't afford to dilly-dally. Returns the number of
> @@ -455,7 +453,7 @@ static size_t crng_fast_load(const void *cp, size_t len)
>  		src++; crng_init_cnt++; len--; ret++;
>  	}
>  	if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
> -		invalidate_batched_entropy();
> +		++base_crng.generation;
>  		crng_init = 1;
>  	}
>  	spin_unlock_irqrestore(&base_crng.lock, flags);

This will only ever increase base_crng.generation from 0 to 1, and the
proper lock is held. The base_crng.key has changed, so it's appropriate
to state that it has reached a new generation.

> @@ -536,7 +534,6 @@ static void crng_reseed(void)
>  	spin_unlock_irqrestore(&base_crng.lock, flags);
>  	memzero_explicit(key, sizeof(key));
>  	if (finalize_init) {
> -		invalidate_batched_entropy();
>  		process_random_ready_list();
>  		wake_up_interruptible(&crng_init_wait);
>  		kill_fasync(&fasync, SIGIO, POLL_IN);

In crng_reseed(), base_crng.generation is incremented above while holding
the lock, and checked that it doesn't reach ULONG_MAX. OK.

> @@ -1278,7 +1275,7 @@ int __init rand_initialize(void)
>  
>  	extract_entropy(base_crng.key, sizeof(base_crng.key));
>  	if (arch_init && trust_cpu && crng_init < 2) {
> -		invalidate_batched_entropy();
> +		++base_crng.generation;
>  		crng_init = 2;
>  		pr_notice("crng init done (trusting CPU's manufacturer)\n");
>  	}

Here we do not need to take a lock (single-threaded operation), can only be
at generation 0 or 1, and the base_crng.key has changed. Which leads me to
ask: shouldn't we increase the generation counter always (or at least if
arch_init is true)? And just make icnrementing crng_init to 2 depending on
trust_cpu?

To sum it up:

	Reviewed-by: Dominik Brodowski <linux@dominikbrodowski.net>

Thanks,
	Dominik

Hi Dominik,

On Thu, Feb 10, 2022 at 7:04 AM Dominik Brodowski
<linux@dominikbrodowski.net> wrote:
> Here we do not need to take a lock (single-threaded operation), can only be
> at generation 0 or 1, and the base_crng.key has changed. Which leads me to
> ask: shouldn't we increase the generation counter always (or at least if
> arch_init is true)? And just make icnrementing crng_init to 2 depending on
> trust_cpu?

Interesting consideration. I think incrementing the generation counter
there unconditionally can't hurt. It should be done every time the
base_crng key changes, which there it clearly does since we're
extracting into it. I'll go ahead and do that.

Jason

Now that we have an explicit base_crng generation counter, we don't need
a separate one for batched entropy. Rather, we can just move the
generation forward every time we change crng_init state or update the
base_crng key.

Cc: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
---
v2 always increments the generation after extraction, as suggested by
Dominik.

 drivers/char/random.c | 29 ++++++++---------------------
 1 file changed, 8 insertions(+), 21 deletions(-)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 5beb421ec12b..57d36f13e3a6 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -431,8 +431,6 @@ static DEFINE_PER_CPU(struct crng, crngs) = {
 
 static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 
-static void invalidate_batched_entropy(void);
-
 /*
  * crng_fast_load() can be called by code in the interrupt service
  * path.  So we can't afford to dilly-dally. Returns the number of
@@ -455,7 +453,7 @@ static size_t crng_fast_load(const void *cp, size_t len)
 		src++; crng_init_cnt++; len--; ret++;
 	}
 	if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
-		invalidate_batched_entropy();
+		++base_crng.generation;
 		crng_init = 1;
 	}
 	spin_unlock_irqrestore(&base_crng.lock, flags);
@@ -530,7 +528,6 @@ static void crng_reseed(void)
 	WRITE_ONCE(base_crng.generation, next_gen);
 	base_crng.birth = jiffies;
 	if (crng_init < 2) {
-		invalidate_batched_entropy();
 		crng_init = 2;
 		finalize_init = true;
 	}
@@ -1277,8 +1274,9 @@ int __init rand_initialize(void)
 	mix_pool_bytes(utsname(), sizeof(*(utsname())));
 
 	extract_entropy(base_crng.key, sizeof(base_crng.key));
+	++base_crng.generation;
+
 	if (arch_init && trust_cpu && crng_init < 2) {
-		invalidate_batched_entropy();
 		crng_init = 2;
 		pr_notice("crng init done (trusting CPU's manufacturer)\n");
 	}
@@ -1628,8 +1626,6 @@ static int __init random_sysctls_init(void)
 device_initcall(random_sysctls_init);
 #endif	/* CONFIG_SYSCTL */
 
-static atomic_t batch_generation = ATOMIC_INIT(0);
-
 struct batched_entropy {
 	union {
 		/* We make this 1.5x a ChaCha block, so that we get the
@@ -1642,8 +1638,8 @@ struct batched_entropy {
 		u32 entropy_u32[CHACHA_BLOCK_SIZE * 3 / (2 * sizeof(u32))];
 	};
 	local_lock_t lock;
+	unsigned long generation;
 	unsigned int position;
-	int generation;
 };
 
 /*
@@ -1662,14 +1658,14 @@ u64 get_random_u64(void)
 	unsigned long flags;
 	struct batched_entropy *batch;
 	static void *previous;
-	int next_gen;
+	unsigned long next_gen;
 
 	warn_unseeded_randomness(&previous);
 
 	local_lock_irqsave(&batched_entropy_u64.lock, flags);
 	batch = raw_cpu_ptr(&batched_entropy_u64);
 
-	next_gen = atomic_read(&batch_generation);
+	next_gen = READ_ONCE(base_crng.generation);
 	if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0 ||
 	    next_gen != batch->generation) {
 		_get_random_bytes(batch->entropy_u64, sizeof(batch->entropy_u64));
@@ -1695,14 +1691,14 @@ u32 get_random_u32(void)
 	unsigned long flags;
 	struct batched_entropy *batch;
 	static void *previous;
-	int next_gen;
+	unsigned long next_gen;
 
 	warn_unseeded_randomness(&previous);
 
 	local_lock_irqsave(&batched_entropy_u32.lock, flags);
 	batch = raw_cpu_ptr(&batched_entropy_u32);
 
-	next_gen = atomic_read(&batch_generation);
+	next_gen = READ_ONCE(base_crng.generation);
 	if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0 ||
 	    next_gen != batch->generation) {
 		_get_random_bytes(batch->entropy_u32, sizeof(batch->entropy_u32));
@@ -1718,15 +1714,6 @@ u32 get_random_u32(void)
 }
 EXPORT_SYMBOL(get_random_u32);
 
-/* It's important to invalidate all potential batched entropy that might
- * be stored before the crng is initialized, which we can do lazily by
- * bumping the generation counter.
- */
-static void invalidate_batched_entropy(void)
-{
-	atomic_inc(&batch_generation);
-}
-
 /**
  * randomize_page - Generate a random, page aligned address
  * @start:	The smallest acceptable address the caller will take.
-- 
2.35.0