So it's nvme_admin_abort_cmd here
[ 74.017450] run blktests nvme/012 at 2021-02-09 21:41:55
[ 74.111311] loop: module loaded
[ 74.125717] loop0: detected capacity change from 2097152 to 0
[ 74.141026] nvmet: adding nsid 1 to subsystem blktests-subsystem-1
[ 74.149395] nvmet_tcp: enabling port 0 (127.0.0.1:4420)
[ 74.158298] nvmet: creating controller 1 for subsystem
blktests-subsystem-1 for NQN
nqn.2014-08.org.nvmexpress:uuid:41131d88-02ca-4ccc-87b3-6ca3f28b13a4.
[ 74.158742] nvme nvme0: creating 48 I/O queues.
[ 74.163391] nvme nvme0: mapped 48/0/0 default/read/poll queues.
[ 74.184623] nvme nvme0: new ctrl: NQN "blktests-subsystem-1", addr
127.0.0.1:4420
[ 75.235059] nvme_tcp: rq 38 opcode 8
[ 75.238653] blk_update_request: I/O error, dev nvme0c0n1, sector
1048624 op 0x9:(WRITE_ZEROES) flags 0x2800800 phys_seg 0 prio class 0
[ 75.380179] XFS (nvme0n1): Mounting V5 Filesystem
[ 75.387457] XFS (nvme0n1): Ending clean mount
[ 75.388555] xfs filesystem being mounted at /mnt/blktests supports
timestamps until 2038 (0x7fffffff)
[ 91.035659] XFS (nvme0n1): Unmounting Filesystem
[ 91.043334] nvme nvme0: Removing ctrl: NQN "blktests-subsystem-1"
But write-zeores is also data less command and should not fail.
And it has a bio, which means that nvme-tcp tries to init an iter
for it when it shouldn't. So the actual offending commit is:
cb9b870fba3e, which cleaned up how the iter is initialized but
introduced this issue.