ext4 oops when running xfstests on cifs - kernel 5.0.0-x (Disco) only

[Guilherme G. Piccoli]

I've found an issue with kernel 5.0.0-24 while trying to verify LP#1824981 in a qemu guest - xfstests running on CIFS triggers a crash in ext4 with the following signature:

BUG: unable to handle kernel NULL pointer dereference at 000000000000012c
#PF error: [normal kernel read fault]
Oops: 0000 [#1] SMP PTI
CPU: 3 PID: 24980 Comm: 223 Not tainted 5.0.0-24-generic #25-Ubuntu
RIP: 0010:call_filldir+0xa2/0x140
[...]
Call Trace:
 ext4_readdir+0x869/0xa80
 iterate_dir+0x9a/0x1b0
 __x64_sys_getdents+0xa6/0x140
 ? __x64_sys_getdents+0x140/0x140
 do_syscall_64+0x5a/0x110

The full oops is attached below.

I've tested on the following kernels, always in the Disco guest:
*Reproduced -> 5.0.0-7 (built that one), 5.0.0-13, 5.0.0-21, 5.0.0-23, 5.0.0-24 (proposed)
*No Issue -> 4.18.0-25 (from Cosmic), 5.2.0-10 (from Eoan), and 5.0, 5.1 and 5.2 from mainline builds (https://kernel.ubuntu.com/~kernel-ppa/mainline/)

To trigger that I've ran the xfstests on a cifs mount point according to the instructions in https://wiki.samba.org/index.php/Xfstesting-cifs . The issue shows around test generic/208 usually.

I've started a "manual" bisect - isolated all commits from v5.0..5.0.0-7, excluded all related with debian folder (and efi-lockdown, due to their complexity) and I'm reverting in a binary search fashion. Reverted 105 commits and still reproduced, and in the next set I've failed to build.

I'll investigate the build fault and also I'll try to reproduce in a Bionic VM with bionic-hwe kernel version 5.0 .

[Guilherme G. Piccoli (gpiccoli)]

I've also faced this oops.

[Guilherme G. Piccoli (gpiccoli)]

Disco is EOLed by now, so if there's a report of this same issue in a newer/supported release, please comment here and we can investigate that front.
Thanks,

Guilherme