[LU-20161] ofd_init0 failure leaks block device references

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.18.0
• Severity: 3
• Assignee: WC Triage
• Reporter: Oleg Drokin

Description

While playing with a inducing sysfs registration failures, I found that when ofd_init0 fails, block device reference is leaked.

This is esp. visible in our test scripts when dm-flakey unit becomes unreleasable:

[ 1528.136276] LDISKFS-fs (dm-1): mounted filesystem 2ab14afa-4991-4985-b132-417921cb718f r/w with ordered data mode. Quota mode: journalled.
[ 1528.249878] LustreError: 50539:0:(ldlm_resource.c:1027:ldlm_namespace_new()) injected sysfs registration failure for filter-lustre-OST0000_UUID
[ 1528.252224] LustreError: 50539:0:(ofd_dev.c:3139:ofd_init0()) lustre-OST0000: unable to create server namespace: rc = -17
[ 1528.252512] LustreError: 50539:0:(obd_config.c:845:class_setup()) setup lustre-OST0000 failed (-17)
[ 1528.252561] LustreError: 50539:0:(obd_config.c:2102:class_config_llog_handler()) MGC192.168.206.101@tcp: cfg command failed: rc = -17
[ 1528.252572] Lustre:    cmd=cf003 0:lustre-OST0000  1:dev  2:0  3:f  
[ 1528.252572] 
[ 1528.252744] LustreError: MGC192.168.206.101@tcp: Confguration from log lustre-OST0000 failed from MGS -17. Communication error between node & MGS, a bad configuration, or other errors. See syslog for more info
[ 1528.252756] LustreError: 50517:0:(tgt_mount.c:1837:server_start_targets()) failed to start server lustre-OST0000: -17
[ 1528.253250] LustreError: 50517:0:(tgt_mount.c:2564:server_fill_super()) Unable to start targets: -17
[ 1528.253376] LustreError: 50517:0:(tgt_mount.c:2082:server_put_super()) no obd lustre-OST0000
[ 1528.262133] workqueue: free_work [libcfs] hogged CPU for >10000us 4 times, consider switching to WQ_UNBOUND
[ 1528.270574] Lustre: server umount lustre-OST0000 complete
[ 1528.270586] LustreError: 50517:0:(super25.c:183:lustre_fill_super()) llite: Unable to mount /dev/mapper/ost1_flakey: rc = -17

attempt to unmount the fs is met with:

# bash llmountcleanup.sh 
Stopping clients: oleg601-server.virtnet /mnt/lustre (opts:-f)
Stopping clients: oleg601-server.virtnet /mnt/lustre2 (opts:-f)
Stopping /mnt/lustre-mds1 (opts:-f) on oleg601-server.virtnet
device-mapper: remove ioctl on ost1_flakey  failed: Device or resource busy
Command failed.

 

Issue Links

• is related to: LU-20160 crash in device_free() when sysfs registration fail