[PATCH v6 0/5] Seeding DRBG with more entropy

[PATCH v6 0/5] Seeding DRBG with more entropy

[Stephan Mueller]

Hi,

as of now, the DRBG is only seeded from get_random_bytes. In various
circumstances, the nonblocking_pool behind get_random_bytes may not be fully
seeded from hardware events at the time the DRBG requires to be seeded.
Based on the discussion in [1], the DRBG seeding is updated such that it
does not completely rely on get_random_bytes any more.

The seeding approach can be characterized as follows:

1. pull buffer of size entropy + nonce from get_random_bytes

2. pull another buffer of size entropy + nonce from my Jitter RNG

3. concatenate both buffers

4. seed the DRBG with the concatenated buffer

5. trigger the async invocation of the blocking API for accessing
   the nonblocking pool with a buffer of size entropy

6. return the DRBG instance to the caller without waiting for the completion
   of step 5

7. at some point in time, the blocking API returns with a full buffer
   which is then used to re-seed the DRBG

This way, we will get entropy during the first initialization without
blocking.

The patch set adds a blocking API to access the nonblocking pool to wait
until the nonblocking pool is initialized.

Note: the DRBG and Jitter RNG patches are against the current cryptodev-2.6
tree.

The new Jitter RNG is an RNG that has large set of tests and was presented on
LKML some time back. After speaking with mathematicians at NIST, that Jitter
RNG approach would be acceptable from their side as a noise source. Note, I
personally think that the Jitter RNG has sufficient entropy in almost all
circumstances (see the massive testing I conducted on all more widely used
CPUs as shown in [2]).

Changes v6:
* patch 01: simplify patch by just adding a blocking API call to random.c as
  suggested by Herbert Xu.
* patch 03: move the async operation into this patch: the DRBG is in control
  of the async work.

Changes v5:
* drop patch 01 and therefore drop the creation of a kernel pool
* change patch 02 to use the nonblocking pool and block until the nonblocking
  pool is initialized or until the cancel operation is triggered.

Changes v4:
* Patch 02: Change get_blocking_random_bytes_cb to allow callers to call it
  multiple times without re-initializing the work data structure. Furthermore,
  only change the pointers to the output buffer and callback if work is not
  pending to avoid race conditions.
* Patch 04: No canceling of seeding during drbg_seed as the invocation of
  get_blocking_random_bytes_cb can now be done repeatedly without
  re-initializing the work data structure.

Changes v3:
* Patch 01: Correct calculation of entropy count as pointed out by Herbert Xu
* Patch 06: Correct a trivial coding issue in jent_entropy_init for
  checking JENT_EMINVARVAR reported by cppcheck

Changes v2:
* Use Dual BSD/GPL license in MODULE_LICENSE as suggested by
  Paul Bolle <pebolle@tiscali.nl>
* Patch 05, drbg_dealloc_state: only deallocate Jitter RNG if one was
  instantiated in the first place. There are two main reasons why the Jitter RNG
  may not be allocated: either it is not available as kernel module/in vmlinuz
  or during init time of the Jitter RNG, the performed testing shows that the
  underlying hardware is not suitable for the Jitter RNG (e.g. has a too coarse
  timer).


[1] http://www.mail-archive.com/linux-crypto@vger.kernel.org/msg13891.html

[2] http://www.chronox.de/jent.html


Stephan Mueller (5):
  random: Blocking API for accessing nonblocking_pool
  crypto: drbg - prepare for async seeding
  crypto: drbg - add async seeding operation
  crypto: drbg - use Jitter RNG to obtain seed
  crypto: add jitterentropy RNG

 crypto/Kconfig         |  10 +
 crypto/Makefile        |   2 +
 crypto/drbg.c          | 140 ++++++--
 crypto/jitterentropy.c | 909 +++++++++++++++++++++++++++++++++++++++++++++++++
 crypto/testmgr.c       |   4 +
 drivers/char/random.c  |  14 +
 include/crypto/drbg.h  |   5 +
 include/linux/random.h |   1 +
 8 files changed, 1058 insertions(+), 27 deletions(-)
 create mode 100644 crypto/jitterentropy.c

-- 
2.1.0

[PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

The added API calls provide a synchronous function call
get_blocking_random_bytes where the caller is blocked until
the nonblocking_pool is initialized.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 drivers/char/random.c  | 14 ++++++++++++++
 include/linux/random.h |  1 +
 2 files changed, 15 insertions(+)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 9cd6968..6f71354 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -1245,6 +1245,20 @@ void get_random_bytes(void *buf, int nbytes)
 EXPORT_SYMBOL(get_random_bytes);
 
 /*
+ * Equivalent function to get_random_bytes with the difference that this
+ * function blocks the request until the nonblocking_pool is initialized.
+ */
+void get_blocking_random_bytes(void *buf, int nbytes)
+{
+	if (unlikely(nonblocking_pool.initialized == 0))
+		wait_event_interruptible(urandom_init_wait,
+					 nonblocking_pool.initialized);
+	extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
+}
+EXPORT_SYMBOL(get_blocking_random_bytes);
+
+
+/*
  * This function will use the architecture-specific hardware random
  * number generator if it is available.  The arch-specific hw RNG will
  * almost certainly be faster than what we can do in software, but it
diff --git a/include/linux/random.h b/include/linux/random.h
index b05856e..0926d78 100644
--- a/include/linux/random.h
+++ b/include/linux/random.h
@@ -14,6 +14,7 @@ extern void add_input_randomness(unsigned int type, unsigned int code,
 extern void add_interrupt_randomness(int irq, int irq_flags);
 
 extern void get_random_bytes(void *buf, int nbytes);
+extern void get_blocking_random_bytes(void *buf, int nbytes);
 extern void get_random_bytes_arch(void *buf, int nbytes);
 void generate_random_uuid(unsigned char uuid_out[16]);
 extern int random_int_secret_init(void);
-- 
2.1.0

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Wed, May 13, 2015 at 09:54:41PM +0200, Stephan Mueller wrote:
>  /*
> + * Equivalent function to get_random_bytes with the difference that this
> + * function blocks the request until the nonblocking_pool is initialized.
> + */
> +void get_blocking_random_bytes(void *buf, int nbytes)
> +{
> +	if (unlikely(nonblocking_pool.initialized == 0))
> +		wait_event_interruptible(urandom_init_wait,
> +					 nonblocking_pool.initialized);
> +	extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);

So what if the wait was interrupted? You are going to extract
entropy from an empty pool.

Anyway, you still haven't addressed my primary concern with this
model which is the potential for dead-lock.  Sleeping for an open
period of time like this in a work queue is bad form.  It may also
lead to dead-locks if whatever you're waiting for happened to use
the same work thread.

That's why I think you should simply provide a function and data
pointer which random.c can then stash onto a list to call when
the pool is ready.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Freitag, 15. Mai 2015, 14:46:26 schrieb Herbert Xu:

Hi Herbert,

> On Wed, May 13, 2015 at 09:54:41PM +0200, Stephan Mueller wrote:
> >  /*
> > 
> > + * Equivalent function to get_random_bytes with the difference that this
> > + * function blocks the request until the nonblocking_pool is initialized.
> > + */
> > +void get_blocking_random_bytes(void *buf, int nbytes)
> > +{
> > +	if (unlikely(nonblocking_pool.initialized == 0))
> > +		wait_event_interruptible(urandom_init_wait,
> > +					 nonblocking_pool.initialized);
> > +	extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
> 
> So what if the wait was interrupted? You are going to extract
> entropy from an empty pool.
> 
> Anyway, you still haven't addressed my primary concern with this
> model which is the potential for dead-lock.  Sleeping for an open
> period of time like this in a work queue is bad form.  It may also
> lead to dead-locks if whatever you're waiting for happened to use
> the same work thread.
> 
> That's why I think you should simply provide a function and data
> pointer which random.c can then stash onto a list to call when
> the pool is ready.

Thanks for the hint to the list. Before handing in another formal patch, may i ask for checking the following approach? I would think that this one should cover your concerns.

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 9cd6968..9bc2a57 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -409,6 +409,19 @@ static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 static DECLARE_WAIT_QUEUE_HEAD(urandom_init_wait);
 static struct fasync_struct *fasync;
 
+static LIST_HEAD(random_wait_list);
+static DEFINE_MUTEX(random_wait_list_mutex);
+struct random_work {
+	struct list_head	list;
+	struct work_struct	rw_work;
+	void			*rw_buf;
+	int			rw_len;
+	void			*rw_private;
+	void			(*rw_cb)(void *buf, int buflen,
+					 void *private);
+};
+static void process_random_waiters(void);
+
 /**********************************************************************
  *
  * OS independent entropy store.   Here are the functions which handle
@@ -660,6 +673,7 @@ retry:
 		r->entropy_total = 0;
 		if (r == &nonblocking_pool) {
 			prandom_reseed_late();
+			process_random_waiters();
 			wake_up_interruptible(&urandom_init_wait);
 			pr_notice("random: %s pool is initialized\n", r->name);
 		}
@@ -1778,3 +1792,64 @@ void add_hwgenerator_randomness(const char *buffer, size_t count,
 	credit_entropy_bits(poolp, entropy);
 }
 EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
+
+static void process_random_waiters(void)
+{
+	struct random_work *rw = NULL;
+
+	mutex_lock(&random_wait_list_mutex);
+	while (!list_empty(&random_wait_list)) {
+		rw = list_first_entry(&random_wait_list, struct random_work,
+				      list);
+		list_del(&rw->list);
+		schedule_work(&rw->rw_work);
+	}
+	mutex_unlock(&random_wait_list_mutex);
+}
+
+static void get_blocking_random_bytes_work(struct work_struct *work)
+{
+	struct random_work *rw = container_of(work, struct random_work,
+					      rw_work);
+
+	get_random_bytes(rw->rw_buf, rw->rw_len);
+	rw->rw_cb(rw->rw_buf, rw->rw_len, rw->rw_private);
+	kfree(rw);
+}
+
+/*
+ * Equivalent function to get_random_bytes with the difference that this
+ * function blocks the request until the nonblocking_pool is initialized.
+ */
+int get_blocking_random_bytes_cb(void *buf, int nbytes, void *private,
+				 void (*cb)(void *buf, int buflen,
+					    void *private))
+{
+	struct random_work *rw = NULL;
+	int ret = 0;
+
+	mutex_lock(&random_wait_list_mutex);
+	list_for_each_entry(rw, &random_wait_list, list)
+		if (buf == rw->rw_buf)
+			goto out;
+
+	rw = kmalloc(sizeof(struct random_work), GFP_KERNEL);
+	if (!rw) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	INIT_WORK(&rw->rw_work, get_blocking_random_bytes_work);
+	rw->rw_buf = buf;
+	rw->rw_len = nbytes;
+	rw->rw_private = private;
+	rw->rw_cb = cb;
+	list_add_tail(&rw->list, &random_wait_list);
+
+out:
+	mutex_unlock(&random_wait_list_mutex);
+	if (nonblocking_pool.initialized)
+		process_random_waiters();
+
+	return ret;
+}
+EXPORT_SYMBOL(get_blocking_random_bytes_cb);
diff --git a/include/linux/random.h b/include/linux/random.h
index b05856e..b57525f 100644
--- a/include/linux/random.h
+++ b/include/linux/random.h
@@ -15,6 +15,9 @@ extern void add_interrupt_randomness(int irq, int irq_flags);
 
 extern void get_random_bytes(void *buf, int nbytes);
 extern void get_random_bytes_arch(void *buf, int nbytes);
+extern int get_blocking_random_bytes_cb(void *buf, int nbytes, void *private,
+					void (*cb)(void *buf, int buflen,
+						   void *private));
 void generate_random_uuid(unsigned char uuid_out[16]);
 extern int random_int_secret_init(void);
 
-- 
2.1.0



-- 
Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Mon, May 18, 2015 at 07:32:01AM +0200, Stephan Mueller wrote:
> 
> Thanks for the hint to the list. Before handing in another formal patch, may i ask for checking the following approach? I would think that this one should cover your concerns.

Yes this is definitely going in the right direction.

> +/*
> + * Equivalent function to get_random_bytes with the difference that this
> + * function blocks the request until the nonblocking_pool is initialized.
> + */
> +int get_blocking_random_bytes_cb(void *buf, int nbytes, void *private,
> +				 void (*cb)(void *buf, int buflen,
> +					    void *private))

You can simplify this further and get rid of buf/nbytes.  All
we need to know is whether the pool is ready.  Everything else
can come from private.

> +	struct random_work *rw = NULL;
> +	int ret = 0;

I think this function should return 0 if the pool is ready now,
-EINPROGRESS if it's not (indicating that the callback will be
called when it is ready) and otherwise an error.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Montag, 18. Mai 2015, 17:21:31 schrieb Herbert Xu:

Hi Herbert,

>> +/*
>> + * Equivalent function to get_random_bytes with the difference that this
>> + * function blocks the request until the nonblocking_pool is initialized.
>> + */
>> +int get_blocking_random_bytes_cb(void *buf, int nbytes, void *private,
>> +				 void (*cb)(void *buf, int buflen,
>> +					    void *private))
>
>You can simplify this further and get rid of buf/nbytes.  All
>we need to know is whether the pool is ready.  Everything else
>can come from private.
>
So, the async function is now just a notification of the caller. Sounds good 
with me.

>> +	struct random_work *rw = NULL;
>> +	int ret = 0;
>
>I think this function should return 0 if the pool is ready now,
>-EINPROGRESS if it's not (indicating that the callback will be
>called when it is ready) and otherwise an error.

Ok, will come in the next patch.

Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Montag, 18. Mai 2015, 15:07:10 schrieb Stephan Mueller:

Hi Herbert,

>>
>>You can simplify this further and get rid of buf/nbytes.  All
>>we need to know is whether the pool is ready.  Everything else
>>can come from private.
>
>So, the async function is now just a notification of the caller. Sounds good
>with me.
>

I am just running into an interesting problem with a missing cancel operation: 
a caller instantiates a DRBG handle and invokes the seeding operation. The 
nonblocking_pool is not initialized. Therefore, the callback is put onto the 
list for being processed later.

Now, the caller releases the DRBG handle *before* the callback is triggered.

The callback is triggered with a pointer that is invalid, but the pointer is 
non-NULL. Therefore, I am not sure how to validate the pointer in the callback 
function.

I would think that there is no other way than to add a cancel API call that 
allows removing a list entry from the wait list.


Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Theodore Ts'o]

On Mon, May 18, 2015 at 03:26:13PM +0200, Stephan Mueller wrote:
> 
> I am just running into an interesting problem with a missing cancel operation: 
> a caller instantiates a DRBG handle and invokes the seeding operation. The 
> nonblocking_pool is not initialized. Therefore, the callback is put onto the 
> list for being processed later.
> 
> Now, the caller releases the DRBG handle *before* the callback is triggered.
> 
> The callback is triggered with a pointer that is invalid, but the pointer is 
> non-NULL. Therefore, I am not sure how to validate the pointer in the callback 
> function.

The simplest thing to do is to put a refcount on inside the DRBG
handle structure.  The caller instantiates the DRBG handle, and
invokes the the DRBG.  The DRBG, since it is kicking off an
asynchronous operation, increments the refcount.

Both the caller and the callback function, before they exit, drop the
refcount, and if they see the refcount is zero, they free the DRBG
handle and the memory where the random seed is to be (or has been)
deposited.

This is the same pattern that the block I/O layer uses with a bio
struct.  In the BIO case, it's important since the callback function
could have been called and returned before the caller gets control
back from the bio_submit() call.  Or the struct bio may contain an
EOPNOTSUPP error, in which case there will be no callback function
dispatched.  So long as everyone handles the refcount rules, it all
works out.

Regards,

					- Ted

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Montag, 18. Mai 2015, 11:02:34 schrieb Theodore Ts'o:

Hi Theodore, Herbert,
> 
> The simplest thing to do is to put a refcount on inside the DRBG
> handle structure.  The caller instantiates the DRBG handle, and
> invokes the the DRBG.  The DRBG, since it is kicking off an
> asynchronous operation, increments the refcount.

That is a good idea. After experimenting with the refcount, I see that kernel 
crypto API release function of crypto_destroy_tfm unconditionally destroys the 
crypto handle by freeing it.

So, if a caller releases the DRBG handle, the DRBG code cannot prevent the 
destruction of its context with a refcount.

Herbert, do you have any ideas?

-- 
Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 07:58:25AM +0200, Stephan Mueller wrote:
> 
> Herbert, do you have any ideas?

On the /dev/random side,

1) Add a struct module argument in addition to func/data.
2) Grab module ref count when func/data is registered.
3) Drop module ref count after func returns.

On the drbg side,

1) Allocate data pointer before func/data registration, it should
contain a flag indicating whether drbg is still alive.
2) In cra_exit, zap the flag in allocated data.
3) In func immediately return if flag indicates drbg is dead.
4) Free allocated data pointer when func is done.

Obviously you need to add some locking so that func doesn't race
against cra_exit or any other drbg paths that it intersects.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Dienstag, 19. Mai 2015, 15:22:27 schrieb Herbert Xu:

Hi Herbert,

> On Tue, May 19, 2015 at 07:58:25AM +0200, Stephan Mueller wrote:
> > Herbert, do you have any ideas?
> 
> On the /dev/random side,
> 
> 1) Add a struct module argument in addition to func/data.
> 2) Grab module ref count when func/data is registered.
> 3) Drop module ref count after func returns.
> 
> On the drbg side,
> 
> 1) Allocate data pointer before func/data registration, it should
> contain a flag indicating whether drbg is still alive.
> 2) In cra_exit, zap the flag in allocated data.
> 3) In func immediately return if flag indicates drbg is dead.
> 4) Free allocated data pointer when func is done.
> 
> Obviously you need to add some locking so that func doesn't race
> against cra_exit or any other drbg paths that it intersects.

Thank you for the hints. I will follow your guidance.

Just for my edification: why is this (rather complex sounding) approach 
preferred over a simple cancel API? Other async APIs (e.g. the AIO syscalls 
with io_cancel) have such cancel operations.

Such cancel function would be as simple as:

void get_blocking_random_bytes_cancel(void *private)
{
        struct random_work *rw, *tmp;

        mutex_lock(&random_wait_list_mutex);
        list_for_each_entry_safe(rw, tmp, &random_wait_list, list) {
                if (private == rw->rw_private) {
                        list_del(&rw->list);
                        kfree(rw);
                        break;
                }
        }
        mutex_unlock(&random_wait_list_mutex);
}

-- 
Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 09:35:15AM +0200, Stephan Mueller wrote:
> 
> Thank you for the hints. I will follow your guidance.
> 
> Just for my edification: why is this (rather complex sounding) approach 
> preferred over a simple cancel API? Other async APIs (e.g. the AIO syscalls 
> with io_cancel) have such cancel operations.
> 
> Such cancel function would be as simple as:

You're right.  The cancel function is indeed simpler.  I can
certainly live with that.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Dienstag, 19. Mai 2015, 15:51:55 schrieb Herbert Xu:

Hi Herbert,

>You're right.  The cancel function is indeed simpler.  I can
>certainly live with that.

Thank you. I will test my patch a bit more and then release it with the 
discussed changes.

Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Theodore Ts'o]

On Tue, May 19, 2015 at 03:51:55PM +0800, Herbert Xu wrote:
> On Tue, May 19, 2015 at 09:35:15AM +0200, Stephan Mueller wrote:
> > 
> > Thank you for the hints. I will follow your guidance.
> > 
> > Just for my edification: why is this (rather complex sounding) approach 
> > preferred over a simple cancel API? Other async APIs (e.g. the AIO syscalls 
> > with io_cancel) have such cancel operations.
> > 
> > Such cancel function would be as simple as:
> 
> You're right.  The cancel function is indeed simpler.  I can
> certainly live with that.

I'm not at all convinced it's simpler.  I see it add a huge amount of
hair to drivers/char/random.c that is only used by a single caller.

I'm also not sure I understand herbert's suggstion of adding a struct
module to the /dev/random code.  Why is this helpful, and what is the
race that you are trying to protect against?

Finally, this is only going to block *once*, when the system is
initially botting up.  Why is it so important that we get the
asynchronous nature of this right, and why can't we solve it simply by
just simply doing the work in a workqueue, with a completion barrier
getting triggered once /dev/random initializes itself, and just simply
blocking the module unload until /dev/random is initialized?

	     	    	   	 	     - Ted

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 09:50:28AM -0400, Theodore Ts'o wrote:
>
> Finally, this is only going to block *once*, when the system is
> initially botting up.  Why is it so important that we get the
> asynchronous nature of this right, and why can't we solve it simply by
> just simply doing the work in a workqueue, with a completion barrier
> getting triggered once /dev/random initializes itself, and just simply
> blocking the module unload until /dev/random is initialized?

I guess I'm still thinking of the old work queue code before
Tejun's cmwq work.  Yes blocking in a work queue should be fine
as there is usually just one DRBG instance.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Dienstag, 19. Mai 2015, 22:18:05 schrieb Herbert Xu:

Hi Herbert,

>On Tue, May 19, 2015 at 09:50:28AM -0400, Theodore Ts'o wrote:
>> Finally, this is only going to block *once*, when the system is
>> initially botting up.  Why is it so important that we get the
>> asynchronous nature of this right, and why can't we solve it simply by
>> just simply doing the work in a workqueue, with a completion barrier
>> getting triggered once /dev/random initializes itself, and just simply
>> blocking the module unload until /dev/random is initialized?
>
>I guess I'm still thinking of the old work queue code before
>Tejun's cmwq work.  Yes blocking in a work queue should be fine
>as there is usually just one DRBG instance.

The current modification with patch 1 to random.c is the smallest change to 
date. Is that then appropriate?

Herbert, based on your comment now, would the currently discussed patch with 
waiting in the work queue in patch 3 appropriate? Or what would you like to 
see changed?

Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 04:27:54PM +0200, Stephan Mueller wrote:
>
> The current modification with patch 1 to random.c is the smallest change to 
> date. Is that then appropriate?
> 
> Herbert, based on your comment now, would the currently discussed patch with 
> waiting in the work queue in patch 3 appropriate? Or what would you like to 
> see changed?

Sorry but I have lost track as to which patch is which.  Please post
the changes to random.c and we can all take a look.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Dienstag, 19. Mai 2015, 22:30:22 schrieb Herbert Xu:

Hi Herbert,

>On Tue, May 19, 2015 at 04:27:54PM +0200, Stephan Mueller wrote:
>> The current modification with patch 1 to random.c is the smallest change to
>> date. Is that then appropriate?
>> 
>> Herbert, based on your comment now, would the currently discussed patch
>> with
>> waiting in the work queue in patch 3 appropriate? Or what would you like to
>> see changed?
>
>Sorry but I have lost track as to which patch is which.  Please post
>the changes to random.c and we can all take a look.

It would be the patch from the start of the thread as follows.

The added API calls provide a synchronous function call
get_blocking_random_bytes where the caller is blocked until
the nonblocking_pool is initialized.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 drivers/char/random.c  | 14 ++++++++++++++
 include/linux/random.h |  1 +
 2 files changed, 15 insertions(+)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 9cd6968..6f71354 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -1245,6 +1245,20 @@ void get_random_bytes(void *buf, int nbytes)
 EXPORT_SYMBOL(get_random_bytes);
 
 /*
+ * Equivalent function to get_random_bytes with the difference that this
+ * function blocks the request until the nonblocking_pool is initialized.
+ */
+void get_blocking_random_bytes(void *buf, int nbytes)
+{
+       if (unlikely(nonblocking_pool.initialized == 0))
+               wait_event_interruptible(urandom_init_wait,
+                                        nonblocking_pool.initialized);
+       extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
+}
+EXPORT_SYMBOL(get_blocking_random_bytes);
+
+
+/*
  * This function will use the architecture-specific hardware random
  * number generator if it is available.  The arch-specific hw RNG will
  * almost certainly be faster than what we can do in software, but it
diff --git a/include/linux/random.h b/include/linux/random.h
index b05856e..0926d78 100644
--- a/include/linux/random.h
+++ b/include/linux/random.h
@@ -14,6 +14,7 @@ extern void add_input_randomness(unsigned int type, unsigned 
int code,
 extern void add_interrupt_randomness(int irq, int irq_flags);
 
 extern void get_random_bytes(void *buf, int nbytes);
+extern void get_blocking_random_bytes(void *buf, int nbytes);
 extern void get_random_bytes_arch(void *buf, int nbytes);
 void generate_random_uuid(unsigned char uuid_out[16]);
 extern int random_int_secret_init(void);

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 04:36:58PM +0200, Stephan Mueller wrote:
> diff --git a/drivers/char/random.c b/drivers/char/random.c
> index 9cd6968..6f71354 100644
> --- a/drivers/char/random.c
> +++ b/drivers/char/random.c
> @@ -1245,6 +1245,20 @@ void get_random_bytes(void *buf, int nbytes)
>  EXPORT_SYMBOL(get_random_bytes);
>  
>  /*
> + * Equivalent function to get_random_bytes with the difference that this
> + * function blocks the request until the nonblocking_pool is initialized.
> + */
> +void get_blocking_random_bytes(void *buf, int nbytes)
> +{
> +       if (unlikely(nonblocking_pool.initialized == 0))
> +               wait_event_interruptible(urandom_init_wait,
> +                                        nonblocking_pool.initialized);
> +       extract_entropy(&nonblocking_pool, buf, nbytes, 0, 0);
> +}
> +EXPORT_SYMBOL(get_blocking_random_bytes);

You still need to handle the case where wait_event_interruptible
returns an error.  Otherwise this looks fine.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Mittwoch, 20. Mai 2015, 06:55:33 schrieb Herbert Xu:

Hi Herbert,
>
>You still need to handle the case where wait_event_interruptible
>returns an error.  Otherwise this looks fine.

Thank you. I would suggest to add a while loop around the call that sets up 
the wait if it terminated with ERESTARTSYS. This way we can maintain a void 
function that is almost identical to the get_random_bytes function.


Ciao
Stephan

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Herbert Xu]

On Tue, May 19, 2015 at 10:18:05PM +0800, Herbert Xu wrote:
> On Tue, May 19, 2015 at 09:50:28AM -0400, Theodore Ts'o wrote:
> >
> > Finally, this is only going to block *once*, when the system is
> > initially botting up.  Why is it so important that we get the
> > asynchronous nature of this right, and why can't we solve it simply by
> > just simply doing the work in a workqueue, with a completion barrier
> > getting triggered once /dev/random initializes itself, and just simply
> > blocking the module unload until /dev/random is initialized?
> 
> I guess I'm still thinking of the old work queue code before
> Tejun's cmwq work.  Yes blocking in a work queue should be fine
> as there is usually just one DRBG instance.

It looks like waiting for it in a workqueue isn't good enough
after all.  I just got this in a KVM machine:

INFO: task kworker/0:1:121 blocked for more than 120 seconds.
      Tainted: G           O    4.1.0-rc1+ #34
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kworker/0:1     D ffff88001eb47d18     0   121      2 0x00000000
Workqueue: events drbg_async_seed [drbg]
 ffff88001eb47d18 ffff88001e1bcec0 ffff88001eb84010 0000000000000246
 ffff88001eb48000 0000000000000020 ffff88001d54ea50 0000000000000000
 ffff88001f613080 ffff88001eb47d38 ffffffff813fe692 0000000000000020
Call Trace:
 [<ffffffff813fe692>] schedule+0x32/0x80
 [<ffffffff812c1415>] get_blocking_random_bytes+0x65/0xa0
 [<ffffffff810825e0>] ? add_wait_queue+0x60/0x60
 [<ffffffffa03dd14c>] drbg_async_seed+0x2c/0xc0 [drbg]
 [<ffffffff81063369>] process_one_work+0x129/0x310
 [<ffffffff810636a9>] worker_thread+0x119/0x430
 [<ffffffff813fe51b>] ? __schedule+0x7fb/0x85e
 [<ffffffff81063590>] ? process_scheduled_works+0x40/0x40
 [<ffffffff81068ed4>] kthread+0xc4/0xe0
 [<ffffffff81060000>] ? proc_cap_handler+0x180/0x1b0
 [<ffffffff81068e10>] ? kthread_freezable_should_stop+0x60/0x60
 [<ffffffff81401ee2>] ret_from_fork+0x42/0x70
 [<ffffffff81068e10>] ? kthread_freezable_should_stop+0x60/0x60

Steffen, I think we need to revisit the idea of having a list
of callbacks.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH v6 1/5] random: Blocking API for accessing nonblocking_pool

[Stephan Mueller]

Am Freitag, 5. Juni 2015, 13:28:06 schrieb Herbert Xu:

Hi Herbert,
> 
> Steffen, I think we need to revisit the idea of having a list
> of callbacks.

Ok, I will reactivate my patch with the list.
> 
> Cheers,


-- 
Ciao
Stephan

[PATCH v6 2/5] crypto: drbg - prepare for async seeding

[Stephan Mueller]

In order to prepare for the addition of the asynchronous seeding call,
the invocation of seeding the DRBG is moved out into a helper function.

In addition, a block of memory is allocated during initialization time
that will be used as a scratchpad for obtaining entropy. That scratchpad
is used for the initial seeding operation as well as by the
asynchronous seeding call. The memory must be zeroized every time the
DRBG seeding call succeeds to avoid entropy data lingering in memory.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 crypto/drbg.c         | 81 ++++++++++++++++++++++++++++++++++-----------------
 include/crypto/drbg.h |  2 ++
 2 files changed, 56 insertions(+), 27 deletions(-)

diff --git a/crypto/drbg.c b/crypto/drbg.c
index 23d444e..36dfece 100644
--- a/crypto/drbg.c
+++ b/crypto/drbg.c
@@ -1041,6 +1041,21 @@ static struct drbg_state_ops drbg_hash_ops = {
  * Functions common for DRBG implementations
  ******************************************************************/
 
+static inline int __drbg_seed(struct drbg_state *drbg, struct list_head *seed,
+			      int reseed)
+{
+	int ret = drbg->d_ops->update(drbg, seed, reseed);
+
+	if (ret)
+		return ret;
+
+	drbg->seeded = true;
+	/* 10.1.1.2 / 10.1.1.3 step 5 */
+	drbg->reseed_ctr = 1;
+
+	return ret;
+}
+
 /*
  * Seeding or reseeding of the DRBG
  *
@@ -1056,8 +1071,6 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 		     bool reseed)
 {
 	int ret = 0;
-	unsigned char *entropy = NULL;
-	size_t entropylen = 0;
 	struct drbg_string data1;
 	LIST_HEAD(seedlist);
 
@@ -1073,26 +1086,10 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 				 drbg->test_data.len);
 		pr_devel("DRBG: using test entropy\n");
 	} else {
-		/*
-		 * Gather entropy equal to the security strength of the DRBG.
-		 * With a derivation function, a nonce is required in addition
-		 * to the entropy. A nonce must be at least 1/2 of the security
-		 * strength of the DRBG in size. Thus, entropy * nonce is 3/2
-		 * of the strength. The consideration of a nonce is only
-		 * applicable during initial seeding.
-		 */
-		entropylen = drbg_sec_strength(drbg->core->flags);
-		if (!entropylen)
-			return -EFAULT;
-		if (!reseed)
-			entropylen = ((entropylen + 1) / 2) * 3;
 		pr_devel("DRBG: (re)seeding with %zu bytes of entropy\n",
-			 entropylen);
-		entropy = kzalloc(entropylen, GFP_KERNEL);
-		if (!entropy)
-			return -ENOMEM;
-		get_random_bytes(entropy, entropylen);
-		drbg_string_fill(&data1, entropy, entropylen);
+			 drbg->seed_buf_len);
+		get_random_bytes(drbg->seed_buf, drbg->seed_buf_len);
+		drbg_string_fill(&data1, drbg->seed_buf, drbg->seed_buf_len);
 	}
 	list_add_tail(&data1.list, &seedlist);
 
@@ -1111,16 +1108,24 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 		memset(drbg->C, 0, drbg_statelen(drbg));
 	}
 
-	ret = drbg->d_ops->update(drbg, &seedlist, reseed);
+	ret = __drbg_seed(drbg, &seedlist, reseed);
+
+	/*
+	 * Clear the initial entropy buffer as the async call may not overwrite
+	 * that buffer for quite some time.
+	 */
+	memzero_explicit(drbg->seed_buf, drbg->seed_buf_len);
 	if (ret)
 		goto out;
-
-	drbg->seeded = true;
-	/* 10.1.1.2 / 10.1.1.3 step 5 */
-	drbg->reseed_ctr = 1;
+	/*
+	 * For all subsequent seeding calls, we only need the seed buffer
+	 * equal to the security strength of the DRBG. We undo the calculation
+	 * in drbg_alloc_state.
+	 */
+	if (!reseed)
+		drbg->seed_buf_len = drbg->seed_buf_len / 3 * 2;
 
 out:
-	kzfree(entropy);
 	return ret;
 }
 
@@ -1143,6 +1148,8 @@ static inline void drbg_dealloc_state(struct drbg_state *drbg)
 	drbg->prev = NULL;
 	drbg->fips_primed = false;
 #endif
+	kzfree(drbg->seed_buf);
+	drbg->seed_buf = NULL;
 }
 
 /*
@@ -1204,6 +1211,26 @@ static inline int drbg_alloc_state(struct drbg_state *drbg)
 		if (!drbg->scratchpad)
 			goto err;
 	}
+
+	/*
+	 * Gather entropy equal to the security strength of the DRBG.
+	 * With a derivation function, a nonce is required in addition
+	 * to the entropy. A nonce must be at least 1/2 of the security
+	 * strength of the DRBG in size. Thus, entropy * nonce is 3/2
+	 * of the strength. The consideration of a nonce is only
+	 * applicable during initial seeding.
+	 */
+	drbg->seed_buf_len = drbg_sec_strength(drbg->core->flags);
+	if (!drbg->seed_buf_len) {
+		ret = -EFAULT;
+		goto err;
+	}
+	/* ensure we have sufficient buffer space for initial seed */
+	drbg->seed_buf_len = ((drbg->seed_buf_len + 1) / 2) * 3;
+	drbg->seed_buf = kzalloc(drbg->seed_buf_len, GFP_KERNEL);
+	if (!drbg->seed_buf)
+		goto err;
+
 	return 0;
 
 err:
diff --git a/include/crypto/drbg.h b/include/crypto/drbg.h
index 480d7a0..b052698 100644
--- a/include/crypto/drbg.h
+++ b/include/crypto/drbg.h
@@ -119,6 +119,8 @@ struct drbg_state {
 	bool fips_primed;	/* Continuous test primed? */
 	unsigned char *prev;	/* FIPS 140-2 continuous test value */
 #endif
+	u8 *seed_buf;			/* buffer holding the seed */
+	size_t seed_buf_len;
 	const struct drbg_state_ops *d_ops;
 	const struct drbg_core *core;
 	struct drbg_string test_data;
-- 
2.1.0

[PATCH v6 3/5] crypto: drbg - add async seeding operation

[Stephan Mueller]

The async seeding operation is triggered during initalization right
after the first non-blocking seeding is completed. As required by the
asynchronous operation of random.c, a callback function is provided that
is triggered by random.c once entropy is available. That callback
function performs the actual seeding of the DRBG.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 crypto/drbg.c         | 25 +++++++++++++++++++++++++
 include/crypto/drbg.h |  2 ++
 2 files changed, 27 insertions(+)

diff --git a/crypto/drbg.c b/crypto/drbg.c
index 36dfece..2b2738e 100644
--- a/crypto/drbg.c
+++ b/crypto/drbg.c
@@ -1056,6 +1056,24 @@ static inline int __drbg_seed(struct drbg_state *drbg, struct list_head *seed,
 	return ret;
 }
 
+static void drbg_async_seed(struct work_struct *work)
+{
+	struct drbg_string data;
+	LIST_HEAD(seedlist);
+	struct drbg_state *drbg = container_of(work, struct drbg_state,
+					       seed_work);
+	int ret = 0;
+
+	get_blocking_random_bytes(drbg->seed_buf, drbg->seed_buf_len);
+
+	drbg_string_fill(&data, drbg->seed_buf, drbg->seed_buf_len);
+	list_add_tail(&data.list, &seedlist);
+	mutex_lock(&drbg->drbg_mutex);
+	ret = __drbg_seed(drbg, &seedlist, true);
+	memzero_explicit(drbg->seed_buf, drbg->seed_buf_len);
+	mutex_unlock(&drbg->drbg_mutex);
+}
+
 /*
  * Seeding or reseeding of the DRBG
  *
@@ -1125,6 +1143,10 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 	if (!reseed)
 		drbg->seed_buf_len = drbg->seed_buf_len / 3 * 2;
 
+	/* Invoke asynchronous seeding unless DRBG is in test mode. */
+	if (!list_empty(&drbg->test_data.list))
+		schedule_work(&drbg->seed_work);
+
 out:
 	return ret;
 }
@@ -1134,6 +1156,7 @@ static inline void drbg_dealloc_state(struct drbg_state *drbg)
 {
 	if (!drbg)
 		return;
+	cancel_work_sync(&drbg->seed_work);
 	kzfree(drbg->V);
 	drbg->V = NULL;
 	kzfree(drbg->C);
@@ -1231,6 +1254,8 @@ static inline int drbg_alloc_state(struct drbg_state *drbg)
 	if (!drbg->seed_buf)
 		goto err;
 
+	INIT_WORK(&drbg->seed_work, drbg_async_seed);
+
 	return 0;
 
 err:
diff --git a/include/crypto/drbg.h b/include/crypto/drbg.h
index b052698..46994b2 100644
--- a/include/crypto/drbg.h
+++ b/include/crypto/drbg.h
@@ -51,6 +51,7 @@
 #include <linux/fips.h>
 #include <linux/mutex.h>
 #include <linux/list.h>
+#include <linux/workqueue.h>
 
 /*
  * Concatenation Helper and string operation helper
@@ -119,6 +120,7 @@ struct drbg_state {
 	bool fips_primed;	/* Continuous test primed? */
 	unsigned char *prev;	/* FIPS 140-2 continuous test value */
 #endif
+	struct work_struct seed_work;	/* asynchronous seeding support */
 	u8 *seed_buf;			/* buffer holding the seed */
 	size_t seed_buf_len;
 	const struct drbg_state_ops *d_ops;
-- 
2.1.0

[PATCH v6 4/5] crypto: drbg - use Jitter RNG to obtain seed

[Stephan Mueller]

During initialization, the DRBG now tries to allocate a handle of the
Jitter RNG. If such a Jitter RNG is available during seeding, the DRBG
pulls the required entropy/nonce string from get_random_bytes and
concatenates it with a string of equal size from the Jitter RNG. That
combined string is now the seed for the DRBG.

Written differently, the initial seed of the DRBG is now:

get_random_bytes(entropy/nonce) || jitterentropy (entropy/nonce)

If the Jitter RNG is not available, the DRBG only seeds from
get_random_bytes.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 crypto/drbg.c         | 46 ++++++++++++++++++++++++++++++++++++++++------
 include/crypto/drbg.h |  1 +
 2 files changed, 41 insertions(+), 6 deletions(-)

diff --git a/crypto/drbg.c b/crypto/drbg.c
index 2b2738e..1224c0c 100644
--- a/crypto/drbg.c
+++ b/crypto/drbg.c
@@ -1104,10 +1104,25 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 				 drbg->test_data.len);
 		pr_devel("DRBG: using test entropy\n");
 	} else {
-		pr_devel("DRBG: (re)seeding with %zu bytes of entropy\n",
-			 drbg->seed_buf_len);
+		/* Get seed from in-kernel /dev/urandom */
 		get_random_bytes(drbg->seed_buf, drbg->seed_buf_len);
-		drbg_string_fill(&data1, drbg->seed_buf, drbg->seed_buf_len);
+
+		/* Get seed from Jitter RNG */
+		if (!drbg->jent ||
+		    crypto_rng_get_bytes(drbg->jent,
+					 drbg->seed_buf + drbg->seed_buf_len,
+					 drbg->seed_buf_len)) {
+			pr_info("DRBG: could not obtain random data from Jitter RNG\n");
+			drbg_string_fill(&data1, drbg->seed_buf,
+					 drbg->seed_buf_len);
+			pr_devel("DRBG: (re)seeding with %zu bytes of entropy\n",
+				 drbg->seed_buf_len);
+		} else {
+			drbg_string_fill(&data1, drbg->seed_buf,
+					 drbg->seed_buf_len * 2);
+			pr_devel("DRBG: (re)seeding with %zu bytes of entropy\n",
+				 drbg->seed_buf_len * 2);
+		}
 	}
 	list_add_tail(&data1.list, &seedlist);
 
@@ -1132,7 +1147,7 @@ static int drbg_seed(struct drbg_state *drbg, struct drbg_string *pers,
 	 * Clear the initial entropy buffer as the async call may not overwrite
 	 * that buffer for quite some time.
 	 */
-	memzero_explicit(drbg->seed_buf, drbg->seed_buf_len);
+	memzero_explicit(drbg->seed_buf, drbg->seed_buf_len * 2);
 	if (ret)
 		goto out;
 	/*
@@ -1173,6 +1188,10 @@ static inline void drbg_dealloc_state(struct drbg_state *drbg)
 #endif
 	kzfree(drbg->seed_buf);
 	drbg->seed_buf = NULL;
+	if (drbg->jent) {
+		crypto_free_rng(drbg->jent);
+		drbg->jent = NULL;
+	}
 }
 
 /*
@@ -1248,14 +1267,29 @@ static inline int drbg_alloc_state(struct drbg_state *drbg)
 		ret = -EFAULT;
 		goto err;
 	}
-	/* ensure we have sufficient buffer space for initial seed */
+	/*
+	 * Ensure we have sufficient buffer space for initial seed which
+	 * consists of the seed from get_random_bytes and the Jitter RNG.
+	 */
 	drbg->seed_buf_len = ((drbg->seed_buf_len + 1) / 2) * 3;
-	drbg->seed_buf = kzalloc(drbg->seed_buf_len, GFP_KERNEL);
+	drbg->seed_buf = kzalloc(drbg->seed_buf_len * 2, GFP_KERNEL);
 	if (!drbg->seed_buf)
 		goto err;
 
 	INIT_WORK(&drbg->seed_work, drbg_async_seed);
 
+	drbg->jent = crypto_alloc_rng("jitterentropy_rng", 0, 0);
+	if(IS_ERR(drbg->jent))
+	{
+		pr_info("DRBG: could not allocate Jitter RNG handle for seeding\n");
+		/*
+		 * As the Jitter RNG is a module that may not be present, we
+		 * continue with the operation and do not fully tie the DRBG
+		 * to the Jitter RNG.
+		 */
+		drbg->jent = NULL;
+	}
+
 	return 0;
 
 err:
diff --git a/include/crypto/drbg.h b/include/crypto/drbg.h
index 46994b2..c3f208d 100644
--- a/include/crypto/drbg.h
+++ b/include/crypto/drbg.h
@@ -123,6 +123,7 @@ struct drbg_state {
 	struct work_struct seed_work;	/* asynchronous seeding support */
 	u8 *seed_buf;			/* buffer holding the seed */
 	size_t seed_buf_len;
+	struct crypto_rng *jent;
 	const struct drbg_state_ops *d_ops;
 	const struct drbg_core *core;
 	struct drbg_string test_data;
-- 
2.1.0

[PATCH v6 5/5] crypto: add jitterentropy RNG

[Stephan Mueller]

The CPU Jitter RNG provides a source of good entropy by
collecting CPU executing time jitter. The entropy in the CPU
execution time jitter is magnified by the CPU Jitter Random
Number Generator. The CPU Jitter Random Number Generator uses
the CPU execution timing jitter to generate a bit stream
which complies with different statistical measurements that
determine the bit stream is random.

The CPU Jitter Random Number Generator delivers entropy which
follows information theoretical requirements. Based on these
studies and the implementation, the caller can assume that
one bit of data extracted from the CPU Jitter Random Number
Generator holds one bit of entropy.

The CPU Jitter Random Number Generator provides a decentralized
source of entropy, i.e. every caller can operate on a private
state of the entropy pool.

The RNG does not have any dependencies on any other service
in the kernel. The RNG only needs a high-resolution time
stamp.

Further design details, the cryptographic assessment and
large array of test results are documented at
http://www.chronox.de/jent.html.

CC: Andreas Steffen <andreas.steffen@strongswan.org>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Sandy Harris <sandyinchina@gmail.com>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
---
 crypto/Kconfig         |  10 +
 crypto/Makefile        |   2 +
 crypto/jitterentropy.c | 909 +++++++++++++++++++++++++++++++++++++++++++++++++
 crypto/testmgr.c       |   4 +
 4 files changed, 925 insertions(+)
 create mode 100644 crypto/jitterentropy.c

diff --git a/crypto/Kconfig b/crypto/Kconfig
index eba55b4..717f2d7 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -1478,9 +1478,19 @@ config CRYPTO_DRBG
 	tristate
 	default CRYPTO_DRBG_MENU if (CRYPTO_DRBG_HMAC || CRYPTO_DRBG_HASH || CRYPTO_DRBG_CTR)
 	select CRYPTO_RNG
+	select CRYPTO_JITTERENTROPY
 
 endif	# if CRYPTO_DRBG_MENU
 
+config CRYPTO_JITTERENTROPY
+	tristate "Jitterentropy Non-Deterministic Random Number Generator"
+	help
+	  The Jitterentropy RNG is a noise that is intended
+	  to provide seed to another RNG. The RNG does not
+	  perform any cryptographic whitening of the generated
+	  random numbers. This Jitterentropy RNG registers with
+	  the kernel crypto API and can be used by any caller.
+
 config CRYPTO_USER_API
 	tristate
 
diff --git a/crypto/Makefile b/crypto/Makefile
index 97b7d3a..2f450ef 100644
--- a/crypto/Makefile
+++ b/crypto/Makefile
@@ -94,6 +94,8 @@ obj-$(CONFIG_CRYPTO_RNG2) += rng.o
 obj-$(CONFIG_CRYPTO_RNG2) += krng.o
 obj-$(CONFIG_CRYPTO_ANSI_CPRNG) += ansi_cprng.o
 obj-$(CONFIG_CRYPTO_DRBG) += drbg.o
+CFLAGS_jitterentropy.o = -O0
+obj-$(CONFIG_CRYPTO_JITTERENTROPY) += jitterentropy.o
 obj-$(CONFIG_CRYPTO_TEST) += tcrypt.o
 obj-$(CONFIG_CRYPTO_GHASH) += ghash-generic.o
 obj-$(CONFIG_CRYPTO_USER_API) += af_alg.o
diff --git a/crypto/jitterentropy.c b/crypto/jitterentropy.c
new file mode 100644
index 0000000..1ebe58a
--- /dev/null
+++ b/crypto/jitterentropy.c
@@ -0,0 +1,909 @@
+/*
+ * Non-physical true random number generator based on timing jitter.
+ *
+ * Copyright Stephan Mueller <smueller@chronox.de>, 2014
+ *
+ * Design
+ * ======
+ *
+ * See http://www.chronox.de/jent.html
+ *
+ * License
+ * =======
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, and the entire permission notice in its entirety,
+ *    including the disclaimer of warranties.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the author may not be used to endorse or promote
+ *    products derived from this software without specific prior
+ *    written permission.
+ *
+ * ALTERNATIVELY, this product may be distributed under the terms of
+ * the GNU General Public License, in which case the provisions of the GPL2 are
+ * required INSTEAD OF the above restrictions.  (This clause is
+ * necessary due to a potential bad interaction between the GPL and
+ * the restrictions contained in a BSD-style copyright.)
+ *
+ * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
+ * WHICH ARE HEREBY DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
+ * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
+ * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+/*
+ * This Jitterentropy RNG is based on the jitterentropy library
+ * version 1.1.0 provided at http://www.chronox.de/jent.html
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/fips.h>
+#include <linux/time.h>
+#include <linux/crypto.h>
+#include <crypto/internal/rng.h>
+
+#ifdef __OPTIMIZE__
+ #error "The CPU Jitter random number generator must not be compiled with optimizations. See documentation. Use the compiler switch -O0 for compiling jitterentropy.c."
+#endif
+
+/* The entropy pool */
+struct rand_data {
+	/* all data values that are vital to maintain the security
+	 * of the RNG are marked as SENSITIVE. A user must not
+	 * access that information while the RNG executes its loops to
+	 * calculate the next random value. */
+	__u64 data;		/* SENSITIVE Actual random number */
+	__u64 old_data;		/* SENSITIVE Previous random number */
+	__u64 prev_time;	/* SENSITIVE Previous time stamp */
+#define DATA_SIZE_BITS ((sizeof(__u64)) * 8)
+	__u64 last_delta;	/* SENSITIVE stuck test */
+	__s64 last_delta2;	/* SENSITIVE stuck test */
+	unsigned int stuck:1;	/* Time measurement stuck */
+	unsigned int osr;	/* Oversample rate */
+	unsigned int stir:1;		/* Post-processing stirring */
+	unsigned int disable_unbias:1;	/* Deactivate Von-Neuman unbias */
+#define JENT_MEMORY_BLOCKS 64
+#define JENT_MEMORY_BLOCKSIZE 32
+#define JENT_MEMORY_ACCESSLOOPS 128
+#define JENT_MEMORY_SIZE (JENT_MEMORY_BLOCKS*JENT_MEMORY_BLOCKSIZE)
+	unsigned char *mem;	/* Memory access location with size of
+				 * memblocks * memblocksize */
+	unsigned int memlocation; /* Pointer to byte in *mem */
+	unsigned int memblocks;	/* Number of memory blocks in *mem */
+	unsigned int memblocksize; /* Size of one memory block in bytes */
+	unsigned int memaccessloops; /* Number of memory accesses per random
+				      * bit generation */
+};
+
+/* Flags that can be used to initialize the RNG */
+#define JENT_DISABLE_STIR (1<<0) /* Disable stirring the entropy pool */
+#define JENT_DISABLE_UNBIAS (1<<1) /* Disable the Von-Neuman Unbiaser */
+#define JENT_DISABLE_MEMORY_ACCESS (1<<2) /* Disable memory access for more
+					   * entropy, saves MEMORY_SIZE RAM for
+					   * entropy collector */
+
+#define DRIVER_NAME     "jitterentropy"
+
+/* -- error codes for init function -- */
+#define JENT_ENOTIME		1 /* Timer service not available */
+#define JENT_ECOARSETIME	2 /* Timer too coarse for RNG */
+#define JENT_ENOMONOTONIC	3 /* Timer is not monotonic increasing */
+#define JENT_EMINVARIATION	4 /* Timer variations too small for RNG */
+#define JENT_EVARVAR		5 /* Timer does not produce variations of
+				   * variations (2nd derivation of time is
+				   * zero). */
+#define JENT_EMINVARVAR		6 /* Timer variations of variations is tooi
+				   * small. */
+
+/***************************************************************************
+ * Helper functions
+ ***************************************************************************/
+
+static inline void jent_get_nstime(__u64 *out)
+{
+	struct timespec ts;
+	__u64 tmp = 0;
+
+	tmp = random_get_entropy();
+
+	/*
+	 * If random_get_entropy does not return a value (which is possible on,
+	 * for example, MIPS), invoke __getnstimeofday
+	 * hoping that there are timers we can work with.
+	 *
+	 * The list of available timers can be obtained from
+	 * /sys/devices/system/clocksource/clocksource0/available_clocksource
+	 * and are registered with clocksource_register()
+	 */
+	if ((0 == tmp) &&
+#ifndef MODULE
+	   (0 == timekeeping_valid_for_hres()) &&
+#endif
+	   (0 == __getnstimeofday(&ts))) {
+		tmp = ts.tv_sec;
+		tmp = tmp << 32;
+		tmp = tmp | ts.tv_nsec;
+	}
+
+	*out = tmp;
+}
+
+
+/**
+ * Update of the loop count used for the next round of
+ * an entropy collection.
+ *
+ * Input:
+ * @ec entropy collector struct -- may be NULL
+ * @bits is the number of low bits of the timer to consider
+ * @min is the number of bits we shift the timer value to the right at
+ *	the end to make sure we have a guaranteed minimum value
+ *
+ * @return Newly calculated loop counter
+ */
+static __u64 jent_loop_shuffle(struct rand_data *ec,
+			       unsigned int bits, unsigned int min)
+{
+	__u64 time = 0;
+	__u64 shuffle = 0;
+	unsigned int i = 0;
+	unsigned int mask = (1<<bits) - 1;
+
+	jent_get_nstime(&time);
+	/*
+	 * mix the current state of the random number into the shuffle
+	 * calculation to balance that shuffle a bit more
+	 */
+	if (ec)
+		time ^= ec->data;
+	/*
+	 * we fold the time value as much as possible to ensure that as many
+	 * bits of the time stamp are included as possible
+	 */
+	for (i = 0; (DATA_SIZE_BITS / bits) > i; i++) {
+		shuffle ^= time & mask;
+		time = time >> bits;
+	}
+
+	/*
+	 * We add a lower boundary value to ensure we have a minimum
+	 * RNG loop count.
+	 */
+	return (shuffle + (1<<min));
+}
+
+/***************************************************************************
+ * Noise sources
+ ***************************************************************************/
+
+/**
+ * CPU Jitter noise source -- this is the noise source based on the CPU
+ *			      execution time jitter
+ *
+ * This function folds the time into one bit units by iterating
+ * through the DATA_SIZE_BITS bit time value as follows: assume our time value
+ * is 0xabcd
+ * 1st loop, 1st shift generates 0xd000
+ * 1st loop, 2nd shift generates 0x000d
+ * 2nd loop, 1st shift generates 0xcd00
+ * 2nd loop, 2nd shift generates 0x000c
+ * 3rd loop, 1st shift generates 0xbcd0
+ * 3rd loop, 2nd shift generates 0x000b
+ * 4th loop, 1st shift generates 0xabcd
+ * 4th loop, 2nd shift generates 0x000a
+ * Now, the values at the end of the 2nd shifts are XORed together.
+ *
+ * The code is deliberately inefficient and shall stay that way. This function
+ * is the root cause why the code shall be compiled without optimization. This
+ * function not only acts as folding operation, but this function's execution
+ * is used to measure the CPU execution time jitter. Any change to the loop in
+ * this function implies that careful retesting must be done.
+ *
+ * Input:
+ * @ec entropy collector struct -- may be NULL
+ * @time time stamp to be folded
+ * @loop_cnt if a value not equal to 0 is set, use the given value as number of
+ *	     loops to perform the folding
+ *
+ * Output:
+ * @folded result of folding operation
+ *
+ * @return Number of loops the folding operation is performed
+ */
+static __u64 jent_fold_time(struct rand_data *ec, __u64 time,
+			    __u64 *folded, __u64 loop_cnt)
+{
+	unsigned int i;
+	__u64 j = 0;
+	__u64 new = 0;
+#define MAX_FOLD_LOOP_BIT 4
+#define MIN_FOLD_LOOP_BIT 0
+	__u64 fold_loop_cnt =
+		jent_loop_shuffle(ec, MAX_FOLD_LOOP_BIT, MIN_FOLD_LOOP_BIT);
+
+	/*
+	 * testing purposes -- allow test app to set the counter, not
+	 * needed during runtime
+	 */
+	if (loop_cnt)
+		fold_loop_cnt = loop_cnt;
+	for (j = 0; j < fold_loop_cnt; j++) {
+		new = 0;
+		for (i = 1; (DATA_SIZE_BITS) >= i; i++) {
+			__u64 tmp = time << (DATA_SIZE_BITS - i);
+
+			tmp = tmp >> (DATA_SIZE_BITS - 1);
+			new ^= tmp;
+		}
+	}
+	*folded = new;
+	return fold_loop_cnt;
+}
+
+/**
+ * Memory Access noise source -- this is a noise source based on variations in
+ *				 memory access times
+ *
+ * This function performs memory accesses which will add to the timing
+ * variations due to an unknown amount of CPU wait states that need to be
+ * added when accessing memory. The memory size should be larger than the L1
+ * caches as outlined in the documentation and the associated testing.
+ *
+ * The L1 cache has a very high bandwidth, albeit its access rate is  usually
+ * slower than accessing CPU registers. Therefore, L1 accesses only add minimal
+ * variations as the CPU has hardly to wait. Starting with L2, significant
+ * variations are added because L2 typically does not belong to the CPU any more
+ * and therefore a wider range of CPU wait states is necessary for accesses.
+ * L3 and real memory accesses have even a wider range of wait states. However,
+ * to reliably access either L3 or memory, the ec->mem memory must be quite
+ * large which is usually not desirable.
+ *
+ * Input:
+ * @ec Reference to the entropy collector with the memory access data -- if
+ *     the reference to the memory block to be accessed is NULL, this noise
+ *     source is disabled
+ * @loop_cnt if a value not equal to 0 is set, use the given value as number of
+ *	     loops to perform the folding
+ *
+ * @return Number of memory access operations
+ */
+static unsigned int jent_memaccess(struct rand_data *ec, __u64 loop_cnt)
+{
+	unsigned char *tmpval = NULL;
+	unsigned int wrap = 0;
+	__u64 i = 0;
+#define MAX_ACC_LOOP_BIT 7
+#define MIN_ACC_LOOP_BIT 0
+	__u64 acc_loop_cnt =
+		jent_loop_shuffle(ec, MAX_ACC_LOOP_BIT, MIN_ACC_LOOP_BIT);
+
+	if (NULL == ec || NULL == ec->mem)
+		return 0;
+	wrap = ec->memblocksize * ec->memblocks;
+
+	/*
+	 * testing purposes -- allow test app to set the counter, not
+	 * needed during runtime
+	 */
+	if (loop_cnt)
+		acc_loop_cnt = loop_cnt;
+
+	for (i = 0; i < (ec->memaccessloops + acc_loop_cnt); i++) {
+		tmpval = ec->mem + ec->memlocation;
+		/*
+		 * memory access: just add 1 to one byte,
+		 * wrap at 255 -- memory access implies read
+		 * from and write to memory location
+		 */
+		*tmpval = (*tmpval + 1) & 0xff;
+		/*
+		 * Addition of memblocksize - 1 to pointer
+		 * with wrap around logic to ensure that every
+		 * memory location is hit evenly
+		 */
+		ec->memlocation = ec->memlocation + ec->memblocksize - 1;
+		ec->memlocation = ec->memlocation % wrap;
+	}
+	return i;
+}
+
+/***************************************************************************
+ * Start of entropy processing logic
+ ***************************************************************************/
+
+/**
+ * Stuck test by checking the:
+ *	1st derivation of the jitter measurement (time delta)
+ *	2nd derivation of the jitter measurement (delta of time deltas)
+ *	3rd derivation of the jitter measurement (delta of delta of time deltas)
+ *
+ * All values must always be non-zero.
+ *
+ * Input:
+ * @ec Reference to entropy collector
+ * @current_delta Jitter time delta
+ *
+ * @return
+ *	0 jitter measurement not stuck (good bit)
+ *	1 jitter measurement stuck (reject bit)
+ */
+static void jent_stuck(struct rand_data *ec, __u64 current_delta)
+{
+	__s64 delta2 = ec->last_delta - current_delta;
+	__s64 delta3 = delta2 - ec->last_delta2;
+
+	ec->last_delta = current_delta;
+	ec->last_delta2 = delta2;
+
+	if (!current_delta || !delta2 || !delta3)
+		ec->stuck = 1;
+}
+
+/**
+ * This is the heart of the entropy generation: calculate time deltas and
+ * use the CPU jitter in the time deltas. The jitter is folded into one
+ * bit. You can call this function the "random bit generator" as it
+ * produces one random bit per invocation.
+ *
+ * WARNING: ensure that ->prev_time is primed before using the output
+ *	    of this function! This can be done by calling this function
+ *	    and not using its result.
+ *
+ * Input:
+ * @entropy_collector Reference to entropy collector
+ *
+ * @return One random bit
+ */
+static __u64 jent_measure_jitter(struct rand_data *ec)
+{
+	__u64 time = 0;
+	__u64 data = 0;
+	__u64 current_delta = 0;
+
+	/* Invoke one noise source before time measurement to add variations */
+	jent_memaccess(ec, 0);
+
+	/*
+	 * Get time stamp and calculate time delta to previous
+	 * invocation to measure the timing variations
+	 */
+	jent_get_nstime(&time);
+	current_delta = time - ec->prev_time;
+	ec->prev_time = time;
+
+	/* Now call the next noise sources which also folds the data */
+	jent_fold_time(ec, current_delta, &data, 0);
+
+	/*
+	 * Check whether we have a stuck measurement. The enforcement
+	 * is performed after the stuck value has been mixed into the
+	 * entropy pool.
+	 */
+	jent_stuck(ec, current_delta);
+
+	return data;
+}
+
+/**
+ * Von Neuman unbias as explained in RFC 4086 section 4.2. As shown in the
+ * documentation of that RNG, the bits from jent_measure_jitter are considered
+ * independent which implies that the Von Neuman unbias operation is applicable.
+ * A proof of the Von-Neumann unbias operation to remove skews is given in the
+ * document "A proposal for: Functionality classes for random number
+ * generators", version 2.0 by Werner Schindler, section 5.4.1.
+ *
+ * Input:
+ * @entropy_collector Reference to entropy collector
+ *
+ * @return One random bit
+ */
+static __u64 jent_unbiased_bit(struct rand_data *entropy_collector)
+{
+	do {
+		__u64 a = jent_measure_jitter(entropy_collector);
+		__u64 b = jent_measure_jitter(entropy_collector);
+
+		if (a == b)
+			continue;
+		if (1 == a)
+			return 1;
+		else
+			return 0;
+	} while (1);
+}
+
+/**
+ * Shuffle the pool a bit by mixing some value with a bijective function (XOR)
+ * into the pool.
+ *
+ * The function generates a mixer value that depends on the bits set and the
+ * location of the set bits in the random number generated by the entropy
+ * source. Therefore, based on the generated random number, this mixer value
+ * can have 2**64 different values. That mixer value is initialized with the
+ * first two SHA-1 constants. After obtaining the mixer value, it is XORed into
+ * the random number.
+ *
+ * The mixer value is not assumed to contain any entropy. But due to the XOR
+ * operation, it can also not destroy any entropy present in the entropy pool.
+ *
+ * Input:
+ * @entropy_collector Reference to entropy collector
+ */
+static void jent_stir_pool(struct rand_data *entropy_collector)
+{
+	/*
+	 * to shut up GCC on 32 bit, we have to initialize the 64 variable
+	 * with two 32 bit variables
+	 */
+	union c {
+		__u64 u64;
+		__u32 u32[2];
+	};
+	/*
+	 * This constant is derived from the first two 32 bit initialization
+	 * vectors of SHA-1 as defined in FIPS 180-4 section 5.3.1
+	 */
+	union c constant;
+	/*
+	 * The start value of the mixer variable is derived from the third
+	 * and fourth 32 bit initialization vector of SHA-1 as defined in
+	 * FIPS 180-4 section 5.3.1
+	 */
+	union c mixer;
+	unsigned int i = 0;
+
+	/*
+	 * Store the SHA-1 constants in reverse order to make up the 64 bit
+	 * value -- this applies to a little endian system, on a big endian
+	 * system, it reverses as expected. But this really does not matter
+	 * as we do not rely on the specific numbers. We just pick the SHA-1
+	 * constants as they have a good mix of bit set and unset.
+	 */
+	constant.u32[1] = 0x67452301;
+	constant.u32[0] = 0xefcdab89;
+	mixer.u32[1] = 0x98badcfe;
+	mixer.u32[0] = 0x10325476;
+
+	for (i = 0; i < DATA_SIZE_BITS; i++) {
+		/*
+		 * get the i-th bit of the input random number and only XOR
+		 * the constant into the mixer value when that bit is set
+		 */
+		if ((entropy_collector->data >> i) & 1)
+			mixer.u64 ^= constant.u64;
+		mixer.u64 = rol64(mixer.u64, 1);
+	}
+	entropy_collector->data ^= mixer.u64;
+}
+
+/**
+ * Generator of one 64 bit random number
+ * Function fills rand_data->data
+ *
+ * Input:
+ * @ec Reference to entropy collector
+ */
+static void jent_gen_entropy(struct rand_data *ec)
+{
+	unsigned int k = 0;
+
+	/* priming of the ->prev_time value */
+	jent_measure_jitter(ec);
+
+	while (1) {
+		__u64 data = 0;
+
+		if (ec->disable_unbias == 1)
+			data = jent_measure_jitter(ec);
+		else
+			data = jent_unbiased_bit(ec);
+
+		/* enforcement of the jent_stuck test */
+		if (ec->stuck) {
+			/*
+			 * We only mix in the bit considered not appropriate
+			 * without the LSFR. The reason is that if we apply
+			 * the LSFR and we do not rotate, the 2nd bit with LSFR
+			 * will cancel out the first LSFR application on the
+			 * bad bit.
+			 *
+			 * And we do not rotate as we apply the next bit to the
+			 * current bit location again.
+			 */
+			ec->data ^= data;
+			ec->stuck = 0;
+			continue;
+		}
+
+		/*
+		 * Fibonacci LSFR with polynom of
+		 *  x^64 + x^61 + x^56 + x^31 + x^28 + x^23 + 1 which is
+		 *  primitive according to
+		 *   http://poincare.matf.bg.ac.rs/~ezivkovm/publications/primpol1.pdf
+		 * (the shift values are the polynom values minus one
+		 * due to counting bits from 0 to 63). As the current
+		 * position is always the LSB, the polynom only needs
+		 * to shift data in from the left without wrap.
+		 */
+		ec->data ^= data;
+		ec->data ^= ((ec->data >> 63) & 1);
+		ec->data ^= ((ec->data >> 60) & 1);
+		ec->data ^= ((ec->data >> 55) & 1);
+		ec->data ^= ((ec->data >> 30) & 1);
+		ec->data ^= ((ec->data >> 27) & 1);
+		ec->data ^= ((ec->data >> 22) & 1);
+		ec->data = rol64(ec->data, 1);
+
+		/*
+		 * We multiply the loop value with ->osr to obtain the
+		 * oversampling rate requested by the caller
+		 */
+		if (++k >= (DATA_SIZE_BITS * ec->osr))
+			break;
+	}
+	if (ec->stir)
+		jent_stir_pool(ec);
+}
+
+/**
+ * The continuous test required by FIPS 140-2 -- the function automatically
+ * primes the test if needed.
+ *
+ * Return:
+ * 0 if FIPS test passed
+ * < 0 if FIPS test failed
+ */
+static void jent_fips_test(struct rand_data *ec)
+{
+	if (!fips_enabled)
+		return;
+
+	/* prime the FIPS test */
+	if (!ec->old_data) {
+		ec->old_data = ec->data;
+		jent_gen_entropy(ec);
+	}
+
+	if (ec->data == ec->old_data)
+		panic(DRIVER_NAME ": Duplicate output detected\n");
+
+	ec->old_data = ec->data;
+}
+
+
+/**
+ * Entry function: Obtain entropy for the caller.
+ *
+ * This function invokes the entropy gathering logic as often to generate
+ * as many bytes as requested by the caller. The entropy gathering logic
+ * creates 64 bit per invocation.
+ *
+ * This function truncates the last 64 bit entropy value output to the exact
+ * size specified by the caller.
+ *
+ * Input:
+ * @ec Reference to entropy collector
+ * @data pointer to buffer for storing random data -- buffer must already
+ *	 exist
+ * @len size of the buffer, specifying also the requested number of random
+ *	in bytes
+ *
+ * @return 0 when request is fulfilled or an error
+ *
+ * The following error codes can occur:
+ *	-1	entropy_collector is NULL
+ */
+static ssize_t jent_read_entropy(struct rand_data *ec, u8 *data, size_t len)
+{
+	u8 *p = data;
+
+	if (!ec)
+		return -EINVAL;
+
+	while (0 < len) {
+		size_t tocopy;
+
+		jent_gen_entropy(ec);
+		jent_fips_test(ec);
+		if ((DATA_SIZE_BITS / 8) < len)
+			tocopy = (DATA_SIZE_BITS / 8);
+		else
+			tocopy = len;
+		memcpy(p, &ec->data, tocopy);
+
+		len -= tocopy;
+		p += tocopy;
+	}
+
+	return 0;
+}
+
+/***************************************************************************
+ * Initialization logic
+ ***************************************************************************/
+
+static struct rand_data *jent_entropy_collector_alloc(unsigned int osr,
+						      unsigned int flags)
+{
+	struct rand_data *entropy_collector;
+
+	entropy_collector = kzalloc(sizeof(struct rand_data), GFP_KERNEL);
+	if (!entropy_collector)
+		return NULL;
+
+	if (!(flags & JENT_DISABLE_MEMORY_ACCESS)) {
+		/* Allocate memory for adding variations based on memory
+		 * access
+		 */
+		entropy_collector->mem = kzalloc(JENT_MEMORY_SIZE, GFP_KERNEL);
+		if (!entropy_collector->mem) {
+			kfree(entropy_collector);
+			return NULL;
+		}
+		entropy_collector->memblocksize = JENT_MEMORY_BLOCKSIZE;
+		entropy_collector->memblocks = JENT_MEMORY_BLOCKS;
+		entropy_collector->memaccessloops = JENT_MEMORY_ACCESSLOOPS;
+	}
+
+	/* verify and set the oversampling rate */
+	if (0 == osr)
+		osr = 1; /* minimum sampling rate is 1 */
+	entropy_collector->osr = osr;
+
+	entropy_collector->stir = 1;
+	if (flags & JENT_DISABLE_STIR)
+		entropy_collector->stir = 0;
+	if (flags & JENT_DISABLE_UNBIAS)
+		entropy_collector->disable_unbias = 1;
+
+	/* fill the data pad with non-zero values */
+	jent_gen_entropy(entropy_collector);
+
+	return entropy_collector;
+}
+
+static void jent_entropy_collector_free(struct rand_data *entropy_collector)
+{
+	if (entropy_collector->mem)
+		kzfree(entropy_collector->mem);
+	entropy_collector->mem = NULL;
+	if (entropy_collector)
+		kzfree(entropy_collector);
+	entropy_collector = NULL;
+}
+
+static int jent_entropy_init(void)
+{
+	int i;
+	__u64 delta_sum = 0;
+	__u64 old_delta = 0;
+	int time_backwards = 0;
+	int count_var = 0;
+	int count_mod = 0;
+
+	/* We could perform statistical tests here, but the problem is
+	 * that we only have a few loop counts to do testing. These
+	 * loop counts may show some slight skew and we produce
+	 * false positives.
+	 *
+	 * Moreover, only old systems show potentially problematic
+	 * jitter entropy that could potentially be caught here. But
+	 * the RNG is intended for hardware that is available or widely
+	 * used, but not old systems that are long out of favor. Thus,
+	 * no statistical tests.
+	 */
+
+	/*
+	 * We could add a check for system capabilities such as clock_getres or
+	 * check for CONFIG_X86_TSC, but it does not make much sense as the
+	 * following sanity checks verify that we have a high-resolution
+	 * timer.
+	 */
+	/*
+	 * TESTLOOPCOUNT needs some loops to identify edge systems. 100 is
+	 * definitely too little.
+	 */
+#define TESTLOOPCOUNT 300
+#define CLEARCACHE 100
+	for (i = 0; (TESTLOOPCOUNT + CLEARCACHE) > i; i++) {
+		__u64 time = 0;
+		__u64 time2 = 0;
+		__u64 folded = 0;
+		__u64 delta = 0;
+		unsigned int lowdelta = 0;
+
+		jent_get_nstime(&time);
+		jent_fold_time(NULL, time, &folded, 1<<MIN_FOLD_LOOP_BIT);
+		jent_get_nstime(&time2);
+
+		/* test whether timer works */
+		if (!time || !time2)
+			return JENT_ENOTIME;
+		delta = time2 - time;
+		/*
+		 * test whether timer is fine grained enough to provide
+		 * delta even when called shortly after each other -- this
+		 * implies that we also have a high resolution timer
+		 */
+		if (!delta)
+			return JENT_ECOARSETIME;
+
+		/*
+		 * up to here we did not modify any variable that will be
+		 * evaluated later, but we already performed some work. Thus we
+		 * already have had an impact on the caches, branch prediction,
+		 * etc. with the goal to clear it to get the worst case
+		 * measurements.
+		 */
+		if (CLEARCACHE > i)
+			continue;
+
+		/* test whether we have an increasing timer */
+		if (!(time2 > time))
+			time_backwards++;
+
+		/*
+		 * Avoid modulo of 64 bit integer to allow code to compile
+		 * on 32 bit architectures.
+		 */
+		lowdelta = time2 - time;
+		if (!(lowdelta % 100))
+			count_mod++;
+
+		/*
+		 * ensure that we have a varying delta timer which is necessary
+		 * for the calculation of entropy -- perform this check
+		 * only after the first loop is executed as we need to prime
+		 * the old_data value
+		 */
+		if (i) {
+			if (delta != old_delta)
+				count_var++;
+			if (delta > old_delta)
+				delta_sum += (delta - old_delta);
+			else
+				delta_sum += (old_delta - delta);
+		}
+		old_delta = delta;
+	}
+
+	/*
+	 * we allow up to three times the time running backwards.
+	 * CLOCK_REALTIME is affected by adjtime and NTP operations. Thus,
+	 * if such an operation just happens to interfere with our test, it
+	 * should not fail. The value of 3 should cover the NTP case being
+	 * performed during our test run.
+	 */
+	if (3 < time_backwards)
+		return JENT_ENOMONOTONIC;
+	/* Error if the time variances are always identical */
+	if (!delta_sum)
+		return JENT_EVARVAR;
+
+	/*
+	 * Variations of deltas of time must on average be larger
+	 * than 1 to ensure the entropy estimation
+	 * implied with 1 is preserved
+	 */
+	if (delta_sum <= 1)
+		return JENT_EMINVARVAR;
+
+	/*
+	 * Ensure that we have variations in the time stamp below 10 for at
+	 * least 10% of all checks -- on some platforms, the counter
+	 * increments in multiples of 100, but not always
+	 */
+	if ((TESTLOOPCOUNT/10 * 9) < count_mod)
+		return JENT_ECOARSETIME;
+
+	return 0;
+}
+
+/***************************************************************************
+ * Kernel crypto API interface
+ ***************************************************************************/
+
+struct jitterentropy {
+	spinlock_t jent_lock;
+	struct rand_data *entropy_collector;
+};
+
+static int jent_kcapi_init(struct crypto_tfm *tfm)
+{
+	struct jitterentropy *rng = crypto_tfm_ctx(tfm);
+	int ret = 0;
+
+	rng->entropy_collector = jent_entropy_collector_alloc(1, 0);
+	if (!rng->entropy_collector)
+		ret = -ENOMEM;
+
+	spin_lock_init(&rng->jent_lock);
+	return ret;
+}
+
+static void jent_kcapi_cleanup(struct crypto_tfm *tfm)
+{
+	struct jitterentropy *rng = crypto_tfm_ctx(tfm);
+
+	spin_lock(&rng->jent_lock);
+	if (rng->entropy_collector)
+		jent_entropy_collector_free(rng->entropy_collector);
+	rng->entropy_collector = NULL;
+	spin_unlock(&rng->jent_lock);
+}
+
+static int jent_kcapi_random(struct crypto_rng *tfm,
+			     const u8 *src, unsigned int slen,
+			     u8 *rdata, unsigned int dlen)
+{
+	struct jitterentropy *rng = crypto_rng_ctx(tfm);
+	int ret = 0;
+
+	spin_lock(&rng->jent_lock);
+	ret = jent_read_entropy(rng->entropy_collector, rdata, dlen);
+	spin_unlock(&rng->jent_lock);
+
+	return ret;
+}
+
+static int jent_kcapi_reset(struct crypto_rng *tfm,
+			    const u8 *seed, unsigned int slen)
+{
+	return 0;
+}
+
+static struct rng_alg jent_alg = {
+	.generate		= jent_kcapi_random,
+	.seed			= jent_kcapi_reset,
+	.seedsize		= 0,
+	.base			= {
+		.cra_name               = "jitterentropy_rng",
+		.cra_driver_name        = "jitterentropy_rng",
+		.cra_priority           = 100,
+		.cra_ctxsize            = sizeof(struct jitterentropy),
+		.cra_module             = THIS_MODULE,
+		.cra_init               = jent_kcapi_init,
+		.cra_exit               = jent_kcapi_cleanup,
+
+	}
+};
+
+static int __init jent_mod_init(void)
+{
+	int ret = 0;
+
+	ret = jent_entropy_init();
+	if (ret) {
+		pr_info(DRIVER_NAME ": Initialization failed with host not compliant with requirements: %d\n", ret);
+		return -EFAULT;
+	}
+	return crypto_register_rng(&jent_alg);
+}
+
+static void __exit jent_mod_exit(void)
+{
+	crypto_unregister_rng(&jent_alg);
+}
+
+module_init(jent_mod_init);
+module_exit(jent_mod_exit);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Stephan Mueller <smueller@chronox.de>");
+MODULE_DESCRIPTION("Non-physical True Random Number Generator based on CPU Jitter");
+MODULE_ALIAS_CRYPTO("jitterentropy_rng");
diff --git a/crypto/testmgr.c b/crypto/testmgr.c
index 1817252..277b3ac 100644
--- a/crypto/testmgr.c
+++ b/crypto/testmgr.c
@@ -3106,6 +3106,10 @@ static const struct alg_test_desc alg_test_descs[] = {
 			}
 		}
 	}, {
+		.alg = "jitterentropy_rng",
+		.fips_allowed = 1,
+		.test = alg_test_null,
+	}, {
 		.alg = "lrw(aes)",
 		.test = alg_test_skcipher,
 		.suite = {
-- 
2.1.0