[LU-18767] qsd_id_glimpse_ast might_sleep due to multiple layers of locking

• Type: Bug
• Resolution: Won't Do
• Priority: Major
• Affects Version/s: Lustre 2.17.0
• Labels: janitory9x
• Environment: RHEL8 debugging server service issue showing up with sanity-quota testing
• Severity: 3
• Assignee: James A Simmons
• Reporter: James A Simmons

Description

Now that we fixed on quota bug with debugging kernel another one has been exposed.

BUG: sleeping function called from invalid context at kernel/locking/rwsem.c:1646

in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 24274, name: ldlm_cb00_000 [ 1186.401920] INFO: lockdep is turned off.

Call Trace: [ 1186.413231] ? dump_stack+0xbb/0x10e

? __might_sleep+0x59/0xc0 [ 1186.424798] ? down_read_nested+0x2e/0x4b0

? lquota_disk_read+0x304/0x6a0

qsd_refresh_usage+0x105/0x3d0

qsd_id_glimpse_ast+0x480/0x980 

ldlm_callback_handler+0x4a8/0x2280

The reason for this I believe is due to qsd_id_glimpse_ast() calls  lqe_write_lock() but then qsd_refresh_usage() also takes dt_read_lock(). 

Issue Links

• is related to: LU-18612 Quota over-softlimit flag remained after large file removal
• is related to: LU-16807 Resolve newer debug kernel warnings