Re: Deadlock when using crypto API for block devices

On Fri, 24 Aug 2018, Mikulas Patocka wrote:

On Fri, 24 Aug 2018, Herbert Xu wrote:

quoted

On Thu, Aug 23, 2018 at 04:39:23PM -0400, Mikulas Patocka wrote:

quoted

1. don't set CRYPTO_TFM_REQ_MAY_SLEEP in dm-crypt
=================================================
If we don't set it, dm-crypt will use GFP_ATOMIC and GFP_ATOMIC may fail. 
The function init_crypt in xts.c handles the failure gracefully - but the 
question is - does the whole crypto code handle allocation failures 
gracefully? If not and if it returns -ENOMEM somewhere, it would result in 
I/O errors and data corruption.

It is safe to use GFP_ATOMIC.  First of the allocation is really
small (less than a page) so it will only fail when the system is
almost completely out of memory.

GFP_ATOMIC is used by networking code. If the system is in a situation 
when packets arrive faster than the swapper is able to swap, it will 
happen. It does happen during netwoking surge and corrupting the 
filesystem in tris situation is not acceptable.

quoted

Even when it does fail the crypto
operation will still succeed, albeit it will process things at a
smaller granularity so the performance will degrade.

A quick search through the crypto code shows that ahash_save_req and 
seqiv_aead_encrypt return -ENOMEM.

Will you fix them?

And also ablkcipher_next_slow, ablkcipher_copy_iv, skcipher_next_slow, 
skcipher_next_copy, skcipher_copy_iv, blkcipher_next_slow, 
blkcipher_copy_iv.

So, I think that dropping CRYPTO_TFM_REQ_MAY_SLEEP is not possible.

Mikulas