kernel panic when kill user process about cryptodev-linux HOT 11 OPEN

Yes, you have gotten the point, the key point is "different destination buffer", and no crash happened with in-place operations.
So sorry to have bother you, wish you have a happy vacation.
Thanks!

from cryptodev-linux.

What AES (from the panic) test did you run and what kernel algortithm(s) did you enabled? What are the versions of kernel and cryptodev?

from cryptodev-linux.

aes_test was a testing program of user space, and it was coded after the file examples/aes.c. I put the three functions: aes_ctx_init， aes_encrypt，aes_ctx_deinit into a while loop for a longtime test, and killed the aes_test while it was running. Kernel algortithm was xts(aes), and I just added several lines to crypto_create_session to support this algortithm. The version of cryptodev is 1.9, and kernel is 4.1.46. The
kernel algortithm(s) I enabled were: stdrng, lzo, crct10dif, crc32c, ecb(arc4), arc4, aes, des3_ede, des, sha224, sha256, sha1, md5, md4.

from cryptodev-linux.

Did you get the crash also with vanilla examples/aes.c or with some of the tests? You could modify them to run a little longer for an easy time to CTRL-C.
The kernel algorithms question was to see if you run a software or a hardware AES driver - it is irrelevant though since the crash is on a different code path.

You can try this branch, we did not see issues with it at the time. If you do, let me know your results

https://github.com/cristian-stoica/cryptodev-linux/tree/umaster

from cryptodev-linux.

I just used the branch you gave to test again, and modified examples/aes.c, made the source address and destination address different in the aes_encrypt function. I tested kill -9 or -15 while running for several times, the crash always happened. Now I'm not sure the problem belongs to cryptodev or the kernel, I will change the kernel version to test it again, and maybe tomorrow I could give you the result. Thank you very much!

from cryptodev-linux.

If I understand you correctly you always get the crash with different destination buffer but not with in-place operations? This sounds like something worth investigating.
I'm going in vacation so we may continue the discussion in a couple of weeks. Otherwise feel free to submit a pull request if you stumble upon a patch.

from cryptodev-linux.

This week , I did the same test with other kernel version——4.10.0-42-generic，and it also reported calltrace information as following:
[10420.249240] [] release_user_pages+0x90/0x100 [cryptodev]
[10420.249246] [] get_userbuf+0x228/0x310 [cryptodev]
[10420.249252] [] crypto_run+0x1c8/0x4a8 [cryptodev]
[10420.249257] [] cryptodev_ioctl+0x520/0x7d8 [cryptodev]
[10420.249260] [] do_vfs_ioctl+0xc4/0x8e8
[10420.249262] [] SyS_ioctl+0x94/0xa8
[10420.249264] [] el0_svc_naked+0x24/0x28
[10420.249266] Code: 52800080 d37df035 f8617841 8b150042 (f9400023)
[10420.249268] ---[ end trace a5d9b3b46941a846 ]---

So I think this problem has nothing with kernel version. I will be glad to research it with you when you return to work.

from cryptodev-linux.

@JaybeamChing Am I wrong to ask if you've got a patch for this?

from cryptodev-linux.

Not yet, I am still not sure how to solve this problem.

from cryptodev-linux.

Do you have any idea how to fix this bug now?

from cryptodev-linux.

@JaybeamChing I'm sorry I don't have a solution for you. I am currently involved in other projects and not actively debugging this issue.

from cryptodev-linux.