Re: [PATCH v10 1/2] crypto: skcipher AF_ALG - overhaul memory management

Am Freitag, 23. Juni 2017, 11:06:30 CEST schrieb Stephan Müller:

Hi Herbert,

Am Freitag, 23. Juni 2017, 08:10:48 CEST schrieb Herbert Xu:

Hi Herbert,

quoted

On Wed, Jun 21, 2017 at 10:03:02PM +0200, Stephan Müller wrote:

quoted

+	/* convert iovecs of output buffers into RX SGL */
+	while (len < ctx->used && msg_data_left(msg)) {

How are we supposed to reach the wait path when ctx->used == 0?

Right.

May I ask whether that wait is correct to begin with? The recvmsg is
protected by a lock_sock. Thus, if the code is waiting, the lock is still
held. So, how can data be inserted into the socket by sendmsg/sendpage
while recvmsg is waiting? Don't we have a deadlock here?

Please ignore that comment. I have implemented a new test set where the 
multithreaded operation of symmetric ciphers are tested. See [1]. Without the 
len < ctx->used, the multithreaded behavior works as intended.

I will send an update shortly.

Thanks

Ciao
Stephan

[1] https://github.com/smuellerDD/libkcapi/commit/
18fe09d7a39ecab8669f599c06a7964c18bb1a1f