Refresh TPGs after deleting a backstore

[iammattcoleman]

Resolves Red Hat Bugzilla #1571728

[iammattcoleman]

@pkalever When using the daemon, 'This LUN does not exist in configFS' appears much more frequently if multiple targets are being modified with individual targetcli commands (not using batch mode) by several simultaneous users or scripts. Can you review/test this change?

[pkalever]

@iammattcoleman certainly yes!

I remember we take blocking locks on each request.

BTW how did you reproduce this issue?
Can you add the steps that you have used to reproduce?

Thanks!

[iammattcoleman]

@pkalever The simplest way to reproduce it is to create a file (e.g. /tmp/lun-does-not-exist-in-configfs) with the following contents...

/backstores/fileio create test1 /tmp/test1.img 10M
/iscsi create iqn.2020-04.com.datto:test
/iscsi/iqn.2020-04.com.datto:test/tpg1/luns create /backstores/fileio/test1
/backstores/fileio delete test1
ls
/iscsi delete iqn.2020-04.com.datto:test

...and then run...

targetcli </tmp/lun-does-not-exist-in-configfs
rm /tmp/test1.img

[maurizio-lombardi]

Hello Matt,

I see that the code you added to ui_command_delete() contains references to iscsi, but what happens if we try to reproduce the bug with a different type of target?

[iammattcoleman]

@maurizio-lombardi That's my main concern with this pull request. I only have experience with iSCSI.

[gonzoleeman]

I wasn't aware of any other protocol that has target portal groups.

[iammattcoleman]

@gonzoleeman I wasn't either, but apparently they all do. targetcli is clear about it for vhost targets, but doesn't say anything about loopback targets:

/> /loopback create
Created target naa.50014055630997a6.
/> /vhost create
Created target naa.5001405ef6e5af1a.
Created TPG 1.

Interestingly, they both show target portal groups in ConfigFS:

# find /sys/kernel/config/target -type d -name 'tpgt_*'
/sys/kernel/config/target/loopback/naa.50014055630997a6/tpgt_1
/sys/kernel/config/target/vhost/naa.5001405ef6e5af1a/tpgt_1

@maurizio-lombardi I'm closing this pull request for now, since it definitely needs fixing. I'll reopen it with a proper fix tomorrow.

[iammattcoleman]

@gonzoleeman I opened an updated version of this pull request that handles all fabric types, including ones that don't have target portal groups: #171