Merge tag '20201024-v4-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wtarreau/prandom

Pull random32 updates from Willy Tarreau:
 "Make prandom_u32() less predictable.

  This is the cleanup of the latest series of prandom_u32
  experimentations consisting in using SipHash instead of Tausworthe to
  produce the randoms used by the network stack.

  The changes to the files were kept minimal, and the controversial
  commit that used to take noise from the fast_pool (f227e3ec3b5c) was
  reverted. Instead, a dedicated "net_rand_noise" per_cpu variable is
  fed from various sources of activities (networking, scheduling) to
  perturb the SipHash state using fast, non-trivially predictable data,
  instead of keeping it fully deterministic. The goal is essentially to
  make any occasional memory leakage or brute-force attempt useless.

  The resulting code was verified to be very slightly faster on x86_64
  than what is was with the controversial commit above, though this
  remains barely above measurement noise. It was also tested on i386 and
  arm, and build- tested only on arm64"

Link: https://lore.kernel.org/netdev/20200808152628.GA27941@SDF.ORG/

* tag '20201024-v4-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wtarreau/prandom:
  random32: add a selftest for the prandom32 code
  random32: add noise from network and scheduling activity
  random32: make prandom_u32() output unpredictable

1 files changed, 345 insertions, 180 deletions

diff --git a/lib/random32.c b/lib/random32.c
index dfb9981ab798..4d0e05e471d7 100644
--- a/lib/random32.c
+++ b/lib/random32.c
@@ -38,19 +38,10 @@
 #include <linux/jiffies.h>
 #include <linux/random.h>
 #include <linux/sched.h>
+#include <linux/bitops.h>
 #include <asm/unaligned.h>
 #include <trace/events/random.h>
 
-#ifdef CONFIG_RANDOM32_SELFTEST
-static void __init prandom_state_selftest(void);
-#else
-static inline void prandom_state_selftest(void)
-{
-}
-#endif
-
-DEFINE_PER_CPU(struct rnd_state, net_rand_state)  __latent_entropy;
-
 /**
  *	prandom_u32_state - seeded pseudo-random number generator.
  *	@state: pointer to state structure holding seeded state.
@@ -71,26 +62,6 @@ u32 prandom_u32_state(struct rnd_state *state)
 EXPORT_SYMBOL(prandom_u32_state);
 
 /**
- *	prandom_u32 - pseudo random number generator
- *
- *	A 32 bit pseudo-random number is generated using a fast
- *	algorithm suitable for simulation. This algorithm is NOT
- *	considered safe for cryptographic use.
- */
-u32 prandom_u32(void)
-{
-	struct rnd_state *state = &get_cpu_var(net_rand_state);
-	u32 res;
-
-	res = prandom_u32_state(state);
-	trace_prandom_u32(res);
-	put_cpu_var(net_rand_state);
-
-	return res;
-}
-EXPORT_SYMBOL(prandom_u32);
-
-/**
  *	prandom_bytes_state - get the requested number of pseudo-random bytes
  *
  *	@state: pointer to state structure holding seeded state.
@@ -121,20 +92,6 @@ void prandom_bytes_state(struct rnd_state *state, void *buf, size_t bytes)
 }
 EXPORT_SYMBOL(prandom_bytes_state);
 
-/**
- *	prandom_bytes - get the requested number of pseudo-random bytes
- *	@buf: where to copy the pseudo-random bytes to
- *	@bytes: the requested number of bytes
- */
-void prandom_bytes(void *buf, size_t bytes)
-{
-	struct rnd_state *state = &get_cpu_var(net_rand_state);
-
-	prandom_bytes_state(state, buf, bytes);
-	put_cpu_var(net_rand_state);
-}
-EXPORT_SYMBOL(prandom_bytes);
-
 static void prandom_warmup(struct rnd_state *state)
 {
 	/* Calling RNG ten times to satisfy recurrence condition */
@@ -150,96 +107,6 @@ static void prandom_warmup(struct rnd_state *state)
 	prandom_u32_state(state);
 }
 
-static u32 __extract_hwseed(void)
-{
-	unsigned int val = 0;
-
-	(void)(arch_get_random_seed_int(&val) ||
-	       arch_get_random_int(&val));
-
-	return val;
-}
-
-static void prandom_seed_early(struct rnd_state *state, u32 seed,
-			       bool mix_with_hwseed)
-{
-#define LCG(x)	 ((x) * 69069U)	/* super-duper LCG */
-#define HWSEED() (mix_with_hwseed ? __extract_hwseed() : 0)
-	state->s1 = __seed(HWSEED() ^ LCG(seed),        2U);
-	state->s2 = __seed(HWSEED() ^ LCG(state->s1),   8U);
-	state->s3 = __seed(HWSEED() ^ LCG(state->s2),  16U);
-	state->s4 = __seed(HWSEED() ^ LCG(state->s3), 128U);
-}
-
-/**
- *	prandom_seed - add entropy to pseudo random number generator
- *	@entropy: entropy value
- *
- *	Add some additional entropy to the prandom pool.
- */
-void prandom_seed(u32 entropy)
-{
-	int i;
-	/*
-	 * No locking on the CPUs, but then somewhat random results are, well,
-	 * expected.
-	 */
-	for_each_possible_cpu(i) {
-		struct rnd_state *state = &per_cpu(net_rand_state, i);
-
-		state->s1 = __seed(state->s1 ^ entropy, 2U);
-		prandom_warmup(state);
-	}
-}
-EXPORT_SYMBOL(prandom_seed);
-
-/*
- *	Generate some initially weak seeding values to allow
- *	to start the prandom_u32() engine.
- */
-static int __init prandom_init(void)
-{
-	int i;
-
-	prandom_state_selftest();
-
-	for_each_possible_cpu(i) {
-		struct rnd_state *state = &per_cpu(net_rand_state, i);
-		u32 weak_seed = (i + jiffies) ^ random_get_entropy();
-
-		prandom_seed_early(state, weak_seed, true);
-		prandom_warmup(state);
-	}
-
-	return 0;
-}
-core_initcall(prandom_init);
-
-static void __prandom_timer(struct timer_list *unused);
-
-static DEFINE_TIMER(seed_timer, __prandom_timer);
-
-static void __prandom_timer(struct timer_list *unused)
-{
-	u32 entropy;
-	unsigned long expires;
-
-	get_random_bytes(&entropy, sizeof(entropy));
-	prandom_seed(entropy);
-
-	/* reseed every ~60 seconds, in [40 .. 80) interval with slack */
-	expires = 40 + prandom_u32_max(40);
-	seed_timer.expires = jiffies + msecs_to_jiffies(expires * MSEC_PER_SEC);
-
-	add_timer(&seed_timer);
-}
-
-static void __init __prandom_start_seed_timer(void)
-{
-	seed_timer.expires = jiffies + msecs_to_jiffies(40 * MSEC_PER_SEC);
-	add_timer(&seed_timer);
-}
-
 void prandom_seed_full_state(struct rnd_state __percpu *pcpu_state)
 {
 	int i;
@@ -259,51 +126,6 @@ void prandom_seed_full_state(struct rnd_state __percpu *pcpu_state)
 }
 EXPORT_SYMBOL(prandom_seed_full_state);
 
-/*
- *	Generate better values after random number generator
- *	is fully initialized.
- */
-static void __prandom_reseed(bool late)
-{
-	unsigned long flags;
-	static bool latch = false;
-	static DEFINE_SPINLOCK(lock);
-
-	/* Asking for random bytes might result in bytes getting
-	 * moved into the nonblocking pool and thus marking it
-	 * as initialized. In this case we would double back into
-	 * this function and attempt to do a late reseed.
-	 * Ignore the pointless attempt to reseed again if we're
-	 * already waiting for bytes when the nonblocking pool
-	 * got initialized.
-	 */
-
-	/* only allow initial seeding (late == false) once */
-	if (!spin_trylock_irqsave(&lock, flags))
-		return;
-
-	if (latch && !late)
-		goto out;
-
-	latch = true;
-	prandom_seed_full_state(&net_rand_state);
-out:
-	spin_unlock_irqrestore(&lock, flags);
-}
-
-void prandom_reseed_late(void)
-{
-	__prandom_reseed(true);
-}
-
-static int __init prandom_reseed(void)
-{
-	__prandom_reseed(false);
-	__prandom_start_seed_timer();
-	return 0;
-}
-late_initcall(prandom_reseed);
-
 #ifdef CONFIG_RANDOM32_SELFTEST
 static struct prandom_test1 {
 	u32 seed;
@@ -423,7 +245,28 @@ static struct prandom_test2 {
 	{  407983964U, 921U,  728767059U },
 };
 
-static void __init prandom_state_selftest(void)
+static u32 __extract_hwseed(void)
+{
+	unsigned int val = 0;
+
+	(void)(arch_get_random_seed_int(&val) ||
+	       arch_get_random_int(&val));
+
+	return val;
+}
+
+static void prandom_seed_early(struct rnd_state *state, u32 seed,
+			       bool mix_with_hwseed)
+{
+#define LCG(x)	 ((x) * 69069U)	/* super-duper LCG */
+#define HWSEED() (mix_with_hwseed ? __extract_hwseed() : 0)
+	state->s1 = __seed(HWSEED() ^ LCG(seed),        2U);
+	state->s2 = __seed(HWSEED() ^ LCG(state->s1),   8U);
+	state->s3 = __seed(HWSEED() ^ LCG(state->s2),  16U);
+	state->s4 = __seed(HWSEED() ^ LCG(state->s3), 128U);
+}
+
+static int __init prandom_state_selftest(void)
 {
 	int i, j, errors = 0, runs = 0;
 	bool error = false;
@@ -463,5 +306,327 @@ static void __init prandom_state_selftest(void)
 		pr_warn("prandom: %d/%d self tests failed\n", errors, runs);
 	else
 		pr_info("prandom: %d self tests passed\n", runs);
+	return 0;
 }
+core_initcall(prandom_state_selftest);
 #endif
+
+/*
+ * The prandom_u32() implementation is now completely separate from the
+ * prandom_state() functions, which are retained (for now) for compatibility.
+ *
+ * Because of (ab)use in the networking code for choosing random TCP/UDP port
+ * numbers, which open DoS possibilities if guessable, we want something
+ * stronger than a standard PRNG.  But the performance requirements of
+ * the network code do not allow robust crypto for this application.
+ *
+ * So this is a homebrew Junior Spaceman implementation, based on the
+ * lowest-latency trustworthy crypto primitive available, SipHash.
+ * (The authors of SipHash have not been consulted about this abuse of
+ * their work.)
+ *
+ * Standard SipHash-2-4 uses 2n+4 rounds to hash n words of input to
+ * one word of output.  This abbreviated version uses 2 rounds per word
+ * of output.
+ */
+
+struct siprand_state {
+	unsigned long v0;
+	unsigned long v1;
+	unsigned long v2;
+	unsigned long v3;
+};
+
+static DEFINE_PER_CPU(struct siprand_state, net_rand_state) __latent_entropy;
+DEFINE_PER_CPU(unsigned long, net_rand_noise);
+EXPORT_PER_CPU_SYMBOL(net_rand_noise);
+
+/*
+ * This is the core CPRNG function.  As "pseudorandom", this is not used
+ * for truly valuable things, just intended to be a PITA to guess.
+ * For maximum speed, we do just two SipHash rounds per word.  This is
+ * the same rate as 4 rounds per 64 bits that SipHash normally uses,
+ * so hopefully it's reasonably secure.
+ *
+ * There are two changes from the official SipHash finalization:
+ * - We omit some constants XORed with v2 in the SipHash spec as irrelevant;
+ *   they are there only to make the output rounds distinct from the input
+ *   rounds, and this application has no input rounds.
+ * - Rather than returning v0^v1^v2^v3, return v1+v3.
+ *   If you look at the SipHash round, the last operation on v3 is
+ *   "v3 ^= v0", so "v0 ^ v3" just undoes that, a waste of time.
+ *   Likewise "v1 ^= v2".  (The rotate of v2 makes a difference, but
+ *   it still cancels out half of the bits in v2 for no benefit.)
+ *   Second, since the last combining operation was xor, continue the
+ *   pattern of alternating xor/add for a tiny bit of extra non-linearity.
+ */
+static inline u32 siprand_u32(struct siprand_state *s)
+{
+	unsigned long v0 = s->v0, v1 = s->v1, v2 = s->v2, v3 = s->v3;
+	unsigned long n = raw_cpu_read(net_rand_noise);
+
+	v3 ^= n;
+	PRND_SIPROUND(v0, v1, v2, v3);
+	PRND_SIPROUND(v0, v1, v2, v3);
+	v0 ^= n;
+	s->v0 = v0;  s->v1 = v1;  s->v2 = v2;  s->v3 = v3;
+	return v1 + v3;
+}
+
+
+/**
+ *	prandom_u32 - pseudo random number generator
+ *
+ *	A 32 bit pseudo-random number is generated using a fast
+ *	algorithm suitable for simulation. This algorithm is NOT
+ *	considered safe for cryptographic use.
+ */
+u32 prandom_u32(void)
+{
+	struct siprand_state *state = get_cpu_ptr(&net_rand_state);
+	u32 res = siprand_u32(state);
+
+	trace_prandom_u32(res);
+	put_cpu_ptr(&net_rand_state);
+	return res;
+}
+EXPORT_SYMBOL(prandom_u32);
+
+/**
+ *	prandom_bytes - get the requested number of pseudo-random bytes
+ *	@buf: where to copy the pseudo-random bytes to
+ *	@bytes: the requested number of bytes
+ */
+void prandom_bytes(void *buf, size_t bytes)
+{
+	struct siprand_state *state = get_cpu_ptr(&net_rand_state);
+	u8 *ptr = buf;
+
+	while (bytes >= sizeof(u32)) {
+		put_unaligned(siprand_u32(state), (u32 *)ptr);
+		ptr += sizeof(u32);
+		bytes -= sizeof(u32);
+	}
+
+	if (bytes > 0) {
+		u32 rem = siprand_u32(state);
+
+		do {
+			*ptr++ = (u8)rem;
+			rem >>= BITS_PER_BYTE;
+		} while (--bytes > 0);
+	}
+	put_cpu_ptr(&net_rand_state);
+}
+EXPORT_SYMBOL(prandom_bytes);
+
+/**
+ *	prandom_seed - add entropy to pseudo random number generator
+ *	@entropy: entropy value
+ *
+ *	Add some additional seed material to the prandom pool.
+ *	The "entropy" is actually our IP address (the only caller is
+ *	the network code), not for unpredictability, but to ensure that
+ *	different machines are initialized differently.
+ */
+void prandom_seed(u32 entropy)
+{
+	int i;
+
+	add_device_randomness(&entropy, sizeof(entropy));
+
+	for_each_possible_cpu(i) {
+		struct siprand_state *state = per_cpu_ptr(&net_rand_state, i);
+		unsigned long v0 = state->v0, v1 = state->v1;
+		unsigned long v2 = state->v2, v3 = state->v3;
+
+		do {
+			v3 ^= entropy;
+			PRND_SIPROUND(v0, v1, v2, v3);
+			PRND_SIPROUND(v0, v1, v2, v3);
+			v0 ^= entropy;
+		} while (unlikely(!v0 || !v1 || !v2 || !v3));
+
+		WRITE_ONCE(state->v0, v0);
+		WRITE_ONCE(state->v1, v1);
+		WRITE_ONCE(state->v2, v2);
+		WRITE_ONCE(state->v3, v3);
+	}
+}
+EXPORT_SYMBOL(prandom_seed);
+
+/*
+ *	Generate some initially weak seeding values to allow
+ *	the prandom_u32() engine to be started.
+ */
+static int __init prandom_init_early(void)
+{
+	int i;
+	unsigned long v0, v1, v2, v3;
+
+	if (!arch_get_random_long(&v0))
+		v0 = jiffies;
+	if (!arch_get_random_long(&v1))
+		v1 = random_get_entropy();
+	v2 = v0 ^ PRND_K0;
+	v3 = v1 ^ PRND_K1;
+
+	for_each_possible_cpu(i) {
+		struct siprand_state *state;
+
+		v3 ^= i;
+		PRND_SIPROUND(v0, v1, v2, v3);
+		PRND_SIPROUND(v0, v1, v2, v3);
+		v0 ^= i;
+
+		state = per_cpu_ptr(&net_rand_state, i);
+		state->v0 = v0;  state->v1 = v1;
+		state->v2 = v2;  state->v3 = v3;
+	}
+
+	return 0;
+}
+core_initcall(prandom_init_early);
+
+
+/* Stronger reseeding when available, and periodically thereafter. */
+static void prandom_reseed(struct timer_list *unused);
+
+static DEFINE_TIMER(seed_timer, prandom_reseed);
+
+static void prandom_reseed(struct timer_list *unused)
+{
+	unsigned long expires;
+	int i;
+
+	/*
+	 * Reinitialize each CPU's PRNG with 128 bits of key.
+	 * No locking on the CPUs, but then somewhat random results are,
+	 * well, expected.
+	 */
+	for_each_possible_cpu(i) {
+		struct siprand_state *state;
+		unsigned long v0 = get_random_long(), v2 = v0 ^ PRND_K0;
+		unsigned long v1 = get_random_long(), v3 = v1 ^ PRND_K1;
+#if BITS_PER_LONG == 32
+		int j;
+
+		/*
+		 * On 32-bit machines, hash in two extra words to
+		 * approximate 128-bit key length.  Not that the hash
+		 * has that much security, but this prevents a trivial
+		 * 64-bit brute force.
+		 */
+		for (j = 0; j < 2; j++) {
+			unsigned long m = get_random_long();
+
+			v3 ^= m;
+			PRND_SIPROUND(v0, v1, v2, v3);
+			PRND_SIPROUND(v0, v1, v2, v3);
+			v0 ^= m;
+		}
+#endif
+		/*
+		 * Probably impossible in practice, but there is a
+		 * theoretical risk that a race between this reseeding
+		 * and the target CPU writing its state back could
+		 * create the all-zero SipHash fixed point.
+		 *
+		 * To ensure that never happens, ensure the state
+		 * we write contains no zero words.
+		 */
+		state = per_cpu_ptr(&net_rand_state, i);
+		WRITE_ONCE(state->v0, v0 ? v0 : -1ul);
+		WRITE_ONCE(state->v1, v1 ? v1 : -1ul);
+		WRITE_ONCE(state->v2, v2 ? v2 : -1ul);
+		WRITE_ONCE(state->v3, v3 ? v3 : -1ul);
+	}
+
+	/* reseed every ~60 seconds, in [40 .. 80) interval with slack */
+	expires = round_jiffies(jiffies + 40 * HZ + prandom_u32_max(40 * HZ));
+	mod_timer(&seed_timer, expires);
+}
+
+/*
+ * The random ready callback can be called from almost any interrupt.
+ * To avoid worrying about whether it's safe to delay that interrupt
+ * long enough to seed all CPUs, just schedule an immediate timer event.
+ */
+static void prandom_timer_start(struct random_ready_callback *unused)
+{
+	mod_timer(&seed_timer, jiffies);
+}
+
+#ifdef CONFIG_RANDOM32_SELFTEST
+/* Principle: True 32-bit random numbers will all have 16 differing bits on
+ * average. For each 32-bit number, there are 601M numbers differing by 16
+ * bits, and 89% of the numbers differ by at least 12 bits. Note that more
+ * than 16 differing bits also implies a correlation with inverted bits. Thus
+ * we take 1024 random numbers and compare each of them to the other ones,
+ * counting the deviation of correlated bits to 16. Constants report 32,
+ * counters 32-log2(TEST_SIZE), and pure randoms, around 6 or lower. With the
+ * u32 total, TEST_SIZE may be as large as 4096 samples.
+ */
+#define TEST_SIZE 1024
+static int __init prandom32_state_selftest(void)
+{
+	unsigned int x, y, bits, samples;
+	u32 xor, flip;
+	u32 total;
+	u32 *data;
+
+	data = kmalloc(sizeof(*data) * TEST_SIZE, GFP_KERNEL);
+	if (!data)
+		return 0;
+
+	for (samples = 0; samples < TEST_SIZE; samples++)
+		data[samples] = prandom_u32();
+
+	flip = total = 0;
+	for (x = 0; x < samples; x++) {
+		for (y = 0; y < samples; y++) {
+			if (x == y)
+				continue;
+			xor = data[x] ^ data[y];
+			flip |= xor;
+			bits = hweight32(xor);
+			total += (bits - 16) * (bits - 16);
+		}
+	}
+
+	/* We'll return the average deviation as 2*sqrt(corr/samples), which
+	 * is also sqrt(4*corr/samples) which provides a better resolution.
+	 */
+	bits = int_sqrt(total / (samples * (samples - 1)) * 4);
+	if (bits > 6)
+		pr_warn("prandom32: self test failed (at least %u bits"
+			" correlated, fixed_mask=%#x fixed_value=%#x\n",
+			bits, ~flip, data[0] & ~flip);
+	else
+		pr_info("prandom32: self test passed (less than %u bits"
+			" correlated)\n",
+			bits+1);
+	kfree(data);
+	return 0;
+}
+core_initcall(prandom32_state_selftest);
+#endif /*  CONFIG_RANDOM32_SELFTEST */
+
+/*
+ * Start periodic full reseeding as soon as strong
+ * random numbers are available.
+ */
+static int __init prandom_init_late(void)
+{
+	static struct random_ready_callback random_ready = {
+		.func = prandom_timer_start
+	};
+	int ret = add_random_ready_callback(&random_ready);
+
+	if (ret == -EALREADY) {
+		prandom_timer_start(&random_ready);
+		ret = 0;
+	}
+	return ret;
+}
+late_initcall(prandom_init_late);