Linux-Crypto Archive on lore.kernel.org
 help / color / Atom feed
* [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread
@ 2019-11-10 13:55 Maciej S. Szmigiero
  2019-11-15  6:08 ` Herbert Xu
       [not found] ` <5dcee409.1c69fb81.f5027.48ad@mx.google.com>
  0 siblings, 2 replies; 5+ messages in thread
From: Maciej S. Szmigiero @ 2019-11-10 13:55 UTC (permalink / raw)
  To: Theodore Ts'o
  Cc: Herbert Xu, Arnd Bergmann, Greg Kroah-Hartman,
	Sebastian Andrzej Siewior, Keerthy, Stephen Boyd, linux-crypto,
	linux-kernel

Since commit 59b569480dc8
("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
there is a race in add_hwgenerator_randomness() between freezing and
stopping the calling kthread.

This commit changed wait_event_interruptible() call with
kthread_freezable_should_stop() as a condition into wait_event_freezable()
with just kthread_should_stop() as a condition to fix a warning that
kthread_freezable_should_stop() might sleep inside the wait.

wait_event_freezable() ultimately calls __refrigerator() with its
check_kthr_stop argument set to false, which causes it to keep the kthread
frozen even if somebody calls kthread_stop() on it.

Calling wait_event_freezable() with kthread_should_stop() as a condition
is racy because it doesn't take into account the situation where this
condition becomes true on a kthread marked for freezing only after this
condition has already been checked.

Calling freezing() should avoid the issue that the commit 59b569480dc8 has
fixed, as it is only a checking function, it doesn't actually do the
freezing.

add_hwgenerator_randomness() has two post-boot users: in khwrng the
kthread will be frozen anyway by call to kthread_freezable_should_stop()
in its main loop, while its second user (ath9k-hwrng) is not freezable at
all.

This change allows a VM with virtio-rng loaded to write s2disk image
successfully.

Fixes: 59b569480dc8 ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
---
 drivers/char/random.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index de434feb873a..2f87910dd498 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -2500,8 +2500,8 @@ void add_hwgenerator_randomness(const char *buffer, size_t count,
 	 * We'll be woken up again once below random_write_wakeup_thresh,
 	 * or when the calling thread is about to terminate.
 	 */
-	wait_event_freezable(random_write_wait,
-			kthread_should_stop() ||
+	wait_event_interruptible(random_write_wait,
+			kthread_should_stop() || freezing(current) ||
 			ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
 	mix_pool_bytes(poolp, buffer, count);
 	credit_entropy_bits(poolp, entropy);

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread
  2019-11-10 13:55 [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread Maciej S. Szmigiero
@ 2019-11-15  6:08 ` Herbert Xu
       [not found] ` <5dcee409.1c69fb81.f5027.48ad@mx.google.com>
  1 sibling, 0 replies; 5+ messages in thread
From: Herbert Xu @ 2019-11-15  6:08 UTC (permalink / raw)
  To: Maciej S. Szmigiero
  Cc: Theodore Ts'o, Arnd Bergmann, Greg Kroah-Hartman,
	Sebastian Andrzej Siewior, Keerthy, Stephen Boyd, linux-crypto,
	linux-kernel

On Sun, Nov 10, 2019 at 02:55:42PM +0100, Maciej S. Szmigiero wrote:
> Since commit 59b569480dc8
> ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
> there is a race in add_hwgenerator_randomness() between freezing and
> stopping the calling kthread.
> 
> This commit changed wait_event_interruptible() call with
> kthread_freezable_should_stop() as a condition into wait_event_freezable()
> with just kthread_should_stop() as a condition to fix a warning that
> kthread_freezable_should_stop() might sleep inside the wait.
> 
> wait_event_freezable() ultimately calls __refrigerator() with its
> check_kthr_stop argument set to false, which causes it to keep the kthread
> frozen even if somebody calls kthread_stop() on it.
> 
> Calling wait_event_freezable() with kthread_should_stop() as a condition
> is racy because it doesn't take into account the situation where this
> condition becomes true on a kthread marked for freezing only after this
> condition has already been checked.
> 
> Calling freezing() should avoid the issue that the commit 59b569480dc8 has
> fixed, as it is only a checking function, it doesn't actually do the
> freezing.
> 
> add_hwgenerator_randomness() has two post-boot users: in khwrng the
> kthread will be frozen anyway by call to kthread_freezable_should_stop()
> in its main loop, while its second user (ath9k-hwrng) is not freezable at
> all.
> 
> This change allows a VM with virtio-rng loaded to write s2disk image
> successfully.
> 
> Fixes: 59b569480dc8 ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
> Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
> ---
>  drivers/char/random.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)

Patch applied.  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

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread
       [not found] ` <5dcee409.1c69fb81.f5027.48ad@mx.google.com>
@ 2019-11-16 23:01   ` Maciej S. Szmigiero
  2019-11-17  0:56     ` Herbert Xu
  0 siblings, 1 reply; 5+ messages in thread
From: Maciej S. Szmigiero @ 2019-11-16 23:01 UTC (permalink / raw)
  To: Stephen Boyd
  Cc: Theodore Ts'o, Herbert Xu, Arnd Bergmann, Greg Kroah-Hartman,
	Sebastian Andrzej Siewior, Keerthy, linux-crypto, linux-kernel

On 15.11.2019 18:44, Stephen Boyd wrote:
> Quoting Maciej S. Szmigiero (2019-11-10 05:55:42)
>> Since commit 59b569480dc8
>> ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
>> there is a race in add_hwgenerator_randomness() between freezing and
>> stopping the calling kthread.
>>
>> This commit changed wait_event_interruptible() call with
>> kthread_freezable_should_stop() as a condition into wait_event_freezable()
>> with just kthread_should_stop() as a condition to fix a warning that
>> kthread_freezable_should_stop() might sleep inside the wait.
>>
>> wait_event_freezable() ultimately calls __refrigerator() with its
>> check_kthr_stop argument set to false, which causes it to keep the kthread
>> frozen even if somebody calls kthread_stop() on it.
>>
>> Calling wait_event_freezable() with kthread_should_stop() as a condition
>> is racy because it doesn't take into account the situation where this
>> condition becomes true on a kthread marked for freezing only after this
>> condition has already been checked.
>>
>> Calling freezing() should avoid the issue that the commit 59b569480dc8 has
>> fixed, as it is only a checking function, it doesn't actually do the
>> freezing.
>>
>> add_hwgenerator_randomness() has two post-boot users: in khwrng the
>> kthread will be frozen anyway by call to kthread_freezable_should_stop()
>> in its main loop, while its second user (ath9k-hwrng) is not freezable at
>> all.
>>
>> This change allows a VM with virtio-rng loaded to write s2disk image
>> successfully.
>>
>> Fixes: 59b569480dc8 ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")
>> Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
>> ---
>>  drivers/char/random.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/char/random.c b/drivers/char/random.c
>> index de434feb873a..2f87910dd498 100644
>> --- a/drivers/char/random.c
>> +++ b/drivers/char/random.c
>> @@ -2500,8 +2500,8 @@ void add_hwgenerator_randomness(const char *buffer, size_t count,
>>          * We'll be woken up again once below random_write_wakeup_thresh,
>>          * or when the calling thread is about to terminate.
>>          */
>> -       wait_event_freezable(random_write_wait,
>> -                       kthread_should_stop() ||
>> +       wait_event_interruptible(random_write_wait,
>> +                       kthread_should_stop() || freezing(current) ||
>>                         ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
> 
> Is it a problem that this wakes up, sees that it should freeze but then
> calls mix_pool_bytes() and credit_entropy_bits()? It looks like
> credit_entropy_bits() will try to wakeup a reader task that is already
> frozen (see the wake_up_interruptible(&random_read_wait) call).

If a reader (user space) task is frozen then it is no longer waiting
on this waitqueue - at least if I understand correctly how the freezer
works for user space tasks, that is by interrupting waits via a fake
signal.

> At one> point I was checking to see if the task was freezing and avoided calling
> those functions so we could get back to kthread_freezable_should_stop()
> in the kthread and actually freeze.

Yes, but I think mixing some extra valid data into random buffer in this
very rare situation shouldn't hurt.
 
>>         mix_pool_bytes(poolp, buffer, count);
>>         credit_entropy_bits(poolp, entropy);
> 
> It's almost like we need a wait_event_freezable_stoppable() API that
> will freeze if freezing() and break out if the kthread is stopped and
> otherwise wait for a wakeup to test the condition. Basically the same
> sort of API that we have for wait_event_freezable() but we pass true for
> the check_kthr_stop flag. Then we can have something like this:
> 
> 	wait_event_freezable_stoppable(
> 		ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
> 	if (!kthread_should_stop()) {
> 		mix_pool_bytes(...);
> 		credit_entropy_bits(...);
> 	}
> 

This API could be added but will there be other users for it?
I mean I think it might not be worth adding a new core kernel API for
such a specific, and workaroundable, issue.

But, on the other hand, maybe this would be a cleaner solution, even
though it would be more complicated.

By the way, the same goes for something like set_freezable_only()
function for the second khwrng freezing issue that I have mentioned
in my Nov 10 UTC message.

Maciej

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread
  2019-11-16 23:01   ` Maciej S. Szmigiero
@ 2019-11-17  0:56     ` Herbert Xu
  2019-11-17  4:09       ` Theodore Y. Ts'o
  0 siblings, 1 reply; 5+ messages in thread
From: Herbert Xu @ 2019-11-17  0:56 UTC (permalink / raw)
  To: Maciej S. Szmigiero
  Cc: Stephen Boyd, Theodore Ts'o, Arnd Bergmann,
	Greg Kroah-Hartman, Sebastian Andrzej Siewior, Keerthy,
	linux-crypto, linux-kernel

On Sun, Nov 17, 2019 at 12:01:20AM +0100, Maciej S. Szmigiero wrote:
>
> If a reader (user space) task is frozen then it is no longer waiting
> on this waitqueue - at least if I understand correctly how the freezer
> works for user space tasks, that is by interrupting waits via a fake
> signal.

At this point I'm just going to revert the whole thing and we can
sort it out in the next development cycle.

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

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread
  2019-11-17  0:56     ` Herbert Xu
@ 2019-11-17  4:09       ` Theodore Y. Ts'o
  0 siblings, 0 replies; 5+ messages in thread
From: Theodore Y. Ts'o @ 2019-11-17  4:09 UTC (permalink / raw)
  To: Herbert Xu
  Cc: Maciej S. Szmigiero, Stephen Boyd, Arnd Bergmann,
	Greg Kroah-Hartman, Sebastian Andrzej Siewior, Keerthy,
	linux-crypto, linux-kernel

On Sun, Nov 17, 2019 at 08:56:41AM +0800, Herbert Xu wrote:
> On Sun, Nov 17, 2019 at 12:01:20AM +0100, Maciej S. Szmigiero wrote:
> >
> > If a reader (user space) task is frozen then it is no longer waiting
> > on this waitqueue - at least if I understand correctly how the freezer
> > works for user space tasks, that is by interrupting waits via a fake
> > signal.
> 
> At this point I'm just going to revert the whole thing and we can
> sort it out in the next development cycle.

Thanks, I hadn't planned on taking any /dev/random changes this cycle
because it had gotten too late and the ext4 tree had gotten unusually
busy.

						- Ted

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, back to index

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-11-10 13:55 [PATCH] random: Don't freeze in add_hwgenerator_randomness() if stopping kthread Maciej S. Szmigiero
2019-11-15  6:08 ` Herbert Xu
     [not found] ` <5dcee409.1c69fb81.f5027.48ad@mx.google.com>
2019-11-16 23:01   ` Maciej S. Szmigiero
2019-11-17  0:56     ` Herbert Xu
2019-11-17  4:09       ` Theodore Y. Ts'o

Linux-Crypto Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-crypto/0 linux-crypto/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-crypto linux-crypto/ https://lore.kernel.org/linux-crypto \
		linux-crypto@vger.kernel.org
	public-inbox-index linux-crypto

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-crypto


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git