Re: [CRASH] crypto: virtio: crash when modprobing tcrypt on 5.5-rc7 / next-20200122

From: LABBE Corentin
Date: Tue Feb 04 2020 - 04:08:18 EST


On Mon, Feb 03, 2020 at 11:01:57PM +0000, Ard Biesheuvel wrote:
> On Mon, 3 Feb 2020 at 20:53, LABBE Corentin <clabbe@xxxxxxxxxxxx> wrote:
> >
> > On Thu, Jan 23, 2020 at 06:52:29AM -0500, Michael S. Tsirkin wrote:
> > > On Thu, Jan 23, 2020 at 11:10:00AM +0100, LABBE Corentin wrote:
> > > > Hello
> > > >
> > > > When modprobing tcrypt on qemu 4.1.0 I get a kernel panic on 5.5-rc7 and next-20200122
> > > > qemu is started by:
> > > > /usr/bin/qemu-system-x86_64 -cpu host -enable-kvm -nographic -net nic,model=e1000,macaddr=52:54:00:12:34:58 -net tap -m 512 -monitor none -object cryptodev-backend-builtin,id=cryptodev0 -device virtio-crypto-pci,id=crypto0,cryptodev=cryptodev0 -append 'console=ttyS0 root=/dev/ram0 ip=dhcp' -kernel /var/lib/lava/dispatcher/tmp/41332/deployimages-td18675m/kernel/bzImage -initrd /var/lib/lava/dispatcher/tmp/41332/deployimages-td18675m/ramdisk/rootfs.cpio.gz -drive format=qcow2,file=/var/lib/lava/dispatcher/tmp/41332/apply-overlay-guest-icy4k1ol/lava-guest.qcow2,media=disk,if=ide,id=lavatest
> > > >
> > > > [ 112.771925] general protection fault: 0000 [#1] SMP PTI
> > > > [ 112.772686] CPU: 0 PID: 126 Comm: virtio0-engine Not tainted 5.5.0-rc7+ #1
> > > > [ 112.773576] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20190711_202441-buildvm-armv7-10.arm.fedoraproject.org-2.fc31 04/01/2014
>
> Interesting. So your QEMU was cross-built on an ARMv7 machine?? Or
> just the guest firmware?

This qemu is built and run on x86 (gentoo).
The firmware is not built (pre-built version).

>
>
> > > > [ 112.775319] RIP: 0010:sg_next+0x0/0x20
> > > > [ 112.775821] Code: cc cc cc cc cc cc cc cc cc cc c7 47 10 00 00 00 00 89 57 0c 48 89 37 89 4f 08 c3 0f 1f 44 00 00 66 2e 0f 1f 84 00 00 00 00 00 <f6> 07 02 75 17 48 8b 57 20 48 8d 47 20 48 89 d1 48 83 e1 fc 83 e2
> > > > [ 112.778330] RSP: 0018:ffffa92440237d90 EFLAGS: 00010006
> > > > [ 112.779071] RAX: fefefefe00000000 RBX: 000000000000000a RCX: fefefefe00000000
> > > > [ 112.780081] RDX: 0000000000000001 RSI: ffff9b19da1a2180 RDI: fefefefe00000000
> > > > [ 112.781081] RBP: ffff9b19da1a2198 R08: ffff9b19dfb24ee8 R09: 0000000000000a20
> > > > [ 112.782079] R10: ffff9b19da125010 R11: 0000000000000000 R12: ffff9b19da1a21b8
> > > > [ 112.783079] R13: 0000000000000003 R14: ffff9b19da1a2180 R15: 0000000000000004
> > > > [ 112.784077] FS: 0000000000000000(0000) GS:ffff9b19de400000(0000) knlGS:0000000000000000
> > > > [ 112.785202] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > [ 112.786030] CR2: 00007f18a157b050 CR3: 000000001040a004 CR4: 0000000000060ef0
> > > > [ 112.787034] Call Trace:
> > > > [ 112.787393] virtqueue_add_sgs+0x4c/0x90
> > > > [ 112.787998] virtio_crypto_skcipher_crypt_req+0x310/0x3e0
> > > > [ 112.788817] crypto_pump_work+0x10c/0x240
> > > > [ 112.789420] ? __kthread_init_worker+0x50/0x50
> > > > [ 112.790082] kthread_worker_fn+0x89/0x180
> > > > [ 112.790690] kthread+0x10e/0x130
> > > > [ 112.791182] ? kthread_park+0x80/0x80
> > > > [ 112.791736] ret_from_fork+0x35/0x40
> > > > [ 112.792282] Modules linked in: cts lzo salsa20_generic camellia_x86_64 camellia_generic fcrypt pcbc tgr192 anubis wp512 khazad tea michael_mic arc4 cast6_generic cast5_generic cast_common deflate sha512_ssse3 sha512_generic cfb ofb serpent_sse2_x86_64 serpent_generic lrw twofish_x86_64_3way twofish_x86_64 crypto_simd cryptd glue_helper twofish_generic twofish_common blowfish_x86_64 blowfish_generic blowfish_common md4 tcrypt(+)
> > > > [ 112.797652] ---[ end trace 4a8142d4a08c2518 ]---
> > > > [ 112.798320] RIP: 0010:sg_next+0x0/0x20
> > > > [ 112.798865] Code: cc cc cc cc cc cc cc cc cc cc c7 47 10 00 00 00 00 89 57 0c 48 89 37 89 4f 08 c3 0f 1f 44 00 00 66 2e 0f 1f 84 00 00 00 00 00 <f6> 07 02 75 17 48 8b 57 20 48 8d 47 20 48 89 d1 48 83 e1 fc 83 e2
> > > > [ 112.801452] RSP: 0018:ffffa92440237d90 EFLAGS: 00010006
> > > > [ 112.802189] RAX: fefefefe00000000 RBX: 000000000000000a RCX: fefefefe00000000
> > > > [ 112.803190] RDX: 0000000000000001 RSI: ffff9b19da1a2180 RDI: fefefefe00000000
> > > > [ 112.804192] RBP: ffff9b19da1a2198 R08: ffff9b19dfb24ee8 R09: 0000000000000a20
> > > > [ 112.805201] R10: ffff9b19da125010 R11: 0000000000000000 R12: ffff9b19da1a21b8
> > > > [ 112.806195] R13: 0000000000000003 R14: ffff9b19da1a2180 R15: 0000000000000004
> > > > [ 112.807222] FS: 0000000000000000(0000) GS:ffff9b19de400000(0000) knlGS:0000000000000000
> > > > [ 112.808352] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > [ 112.809169] CR2: 00007f18a157b050 CR3: 000000001040a004 CR4: 0000000000060ef0
> > > >
> > > > I have tested also 5.4.14
> > > > and I got random freeze with:
> > > > qemu-system-x86_64: virtio: zero sized buffers are not allowed
> > > >
> > > > Regards
> > >
> > > did any of previous versions work for you?
> > > Any chance of a bisect?
> > >
> >
> > Hello
> >
> > Bisecting lead to 500e6807ce93b1fdc7d5b827c5cc167cc35630db ("crypto: virtio - implement missing support for output IVs")
> > Qemu 4.2.0 also fail like this but not 1:3.1+dfsg-8+deb10u2
> >
>
> That commit adds scatterwalk_map_and_copy() calls in two places, and
> the only way I see that could cause corruption here is when
> req->cryptlen < AES_BLOCK_SIZE.
>
> Could you please try whether returning an error early in
> __virtio_crypto_skcipher_do_req() if req->cryptlen < AES_BLOCK_SIZE
> fixes the issue?

I did the following:
--- a/drivers/crypto/virtio/virtio_crypto_algs.c
+++ b/drivers/crypto/virtio/virtio_crypto_algs.c
@@ -360,6 +360,9 @@ __virtio_crypto_skcipher_do_req(struct virtio_crypto_sym_request *vc_sym_req,
int sg_total;
uint8_t *iv;

+ if (req->cryptlen < AES_BLOCK_SIZE)
+ return -EINVAL;
+
src_nents = sg_nents_for_len(req->src, req->cryptlen);
dst_nents = sg_nents(req->dst);

Without any change in the crash.
So the problem is probably present since the introduction of the driver but hidden because it never pass the selftests.