[LU-18664] distribute_txn_commit_batchid_update() is called in an atomic context

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.17.0
• Labels: janitor9x
• Environment: RHEL8 running with debug kernel.
• Severity: 3
• Assignee: James A Simmons
• Reporter: James A Simmons

Description

The function distribute_txn_commit_batchid_update() is called in an atomic context and its memory allocation is using the wrong flags so it could sleep.
https://testing.whamcloud.com/gerrit-janitor/54439/testresults/sanity3-ldiskfs-DNE-rocky8.10_x86_64-rocky8.10_x86_64/

Lustre: DEBUG MARKER: == sanity test 60g: transaction abort won't cause MDT hung === 01:29:59 (1737268199)
Lustre: *** cfs_fail_loc=19a, val=0***
------------[ cut here ]------------
do not call blocking ops when !TASK_RUNNING; state=402 set at [<00000000878da86f>] distribute_txn_commit_thread+0x98/0x1020 [ptlrpc]
WARNING: CPU: 0 PID: 11190 at kernel/sched/core.c:7471 __might_sleep+0x9d/0xc0
CPU: 0 PID: 11190 Comm: dist_txn-1 4.18.0rh8.10-debug #7
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.2-1.fc38 04/01/2014 [ 438.807736] RIP: 0010:__might_sleep+0x9d/0xc0
Call Trace:
? __might_sleep+0x9d/0xc0
? report_bug+0x113/0x140
? do_error_trap+0xb6/0x130
? do_invalid_op+0x46/0x60
? __might_sleep+0x9d/0xc0
? invalid_op+0x14/0x20
? distribute_txn_commit_batchid_update+0x68/0xa90 [ptlrpc]
? __might_sleep+0x9d/0xc0
? __might_sleep+0x95/0xc0
slab_pre_alloc_hook.constprop.59+0x13d/0x1f0
kmem_cache_alloc_trace+0x5b/0x380
distribute_txn_commit_batchid_update+0x68/0xa90 [ptlrpc] 
distribute_txn_commit_thread+0xab5/0x1020 [ptlrpc]
kthread+0x1d7/0x210

Issue Links

• is related to: LU-17884 Soft lockup after test ends
• is related to: LU-16807 Resolve newer debug kernel warnings
• is related to: LU-13986 livelock is possible in distribute_txn_commit_thread()