How to use targetcli to disable iSCSI write cache

I have exported a single disk as an iSCSI target using targetcli on a Linux host. On a remote machine I have connected to the device, and placed a ZFS volume on it (as a replication target with zrepl).

However any time I reboot the iSCSI target machine, the initiator machine running ZFS fails with write errors. After doing some research, the type of failures it has only occur when ZFS sends a write to a device, it's told the write completed successfully, but then the write is lost.

This means the iSCSI target must be caching writes and returning success before they are actually written to the disk, and when the machine is rebooted those cached writes don't always get flushed.

Is there a way to use targetcli to disable this write caching, so that writes are not returned as successful until they have actually been written to the physical disk?

I tried exporting the physical disk as the pscsi backstore instead, but this just makes the iSCSI target kernel crash as soon as the disk is accessed. (There are no VMs in use, just two native Linux servers, so no unusual SCSI commands should be used.)

/> ls
o- / .............................................................................................. [...]
  o- backstores ................................................................................... [...]
  | o- block ....................................................................... [Storage Objects: 1]
  | | o- ar0 .......... [/dev/disk/by-id/ata-ST22000NT001-3LS101_ZX20XTYG (20.0TiB) write-thru activated]
  | |   o- alua ........................................................................ [ALUA Groups: 1]
  | |     o- default_tg_pt_gp ............................................ [ALUA state: Active/optimized]
  | o- fileio ...................................................................... [Storage Objects: 0]
  | o- pscsi ....................................................................... [Storage Objects: 0]
  | o- ramdisk ..................................................................... [Storage Objects: 0]
  o- iscsi ................................................................................. [Targets: 1]
  | o- iqn.2023-07.com.example.host:sn.zx20xtyg ............................................... [TPGs: 1]
  |   o- tpg1 .................................................................... [no-gen-acls, no-auth]
  |     o- acls ............................................................................... [ACLs: 1]
  |     | o- iqn.2023-07.com.example.client:783af16f240 ................................ [Mapped LUNs: 1]
  |     |   o- mapped_lun0 ........................................................ [lun0 block/ar0 (rw)]
  |     o- luns ............................................................................... [LUNs: 1]
  |     | o- lun0 ..... [block/ar0 (/dev/disk/by-id/ata-ST22000NT001-3LS101_ZX20XTYG) (default_tg_pt_gp)]
  |     o- portals ......................................................................... [Portals: 1]
  |       o- 0.0.0.0:3260 .......................................................................... [OK]
  o- loopback .............................................................................. [Targets: 0]
  o- vhost ................................................................................. [Targets: 0]
  o- xen-pvscsi ............................................................................ [Targets: 0]

  pool: iscsi
 state: ONLINE
status: One or more devices are faulted in response to IO failures.
action: Make sure the affected devices are connected, then run 'zpool clear'.
   see: https://openzfs.github.io/openzfs-docs/msg/ZFS-8000-JQ
config: 
        
        NAME                                      STATE     READ WRITE CKSUM
        iscsi                                     ONLINE       0     0     0
          wwn-0x60014050b85f5b1d12d4875bdb68f41a  ONLINE       3   153     0
errors: List of errors unavailable: pool I/O is currently suspended

$ zpool clear iscsu
$ zpool status -v
  pool: iscsi
 state: ONLINE
status: One or more devices has experienced an error resulting in data
        corruption.  Applications may be affected.
action: Restore the file in question if possible.  Otherwise restore the
        entire pool from backup.
   see: https://openzfs.github.io/openzfs-docs/msg/ZFS-8000-8A
  scan: resilvered 360K in 00:00:02 with 0 errors on Fri Jul 28 12:58:10 2023
config:

        NAME                                      STATE     READ WRITE CKSUM
        iscsi                                     ONLINE       0     0     0
          wwn-0x60014050b85f5b1d12d4875bdb68f41a  ONLINE       0     0     0

errors: Permanent errors have been detected in the following files:

        <metadata>:<0x0>
        <metadata>:<0x101>
        <metadata>:<0x104>
        <metadata>:<0x105>
        <metadata>:<0x33>
        <metadata>:<0x34>
        <metadata>:<0x3d>
        <metadata>:<0xe8d>
        <metadata>:<0x1e1f9>
        <metadata>:<0x1e1fb>

You need to use iSCSI target sending writes with FUA (Force Unit Attention) flag set or issuing “Synchronize Cache” after writes submitted before sending write confirmation to the caller. That if you “map” the whole physical disk raw.