[LU-17311]  Regression after "LU-16713 llite: writeback/commit pages under memory pressure" patch Created: 23/Nov/23  Updated: 24/Nov/23

Status: Open
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Bug Priority: Major
Reporter: Artem Blagodarenko Assignee: Qian Yingjin
Resolution: Unresolved Votes: 0
Labels: None
Environment:

Ubuntu 20.04.5
5.4.0-48-generic


Severity: 3
Rank (Obsolete): 9223372036854775807

 Description   

On Ubuntu 20.04.5 (5.4.0-48-generic kernel) with master

$sudo lustre/tests/llmount.sh
$sudo cp -R ldiskfs /mnt/lustre

Leads to the fault:

 [ 2727.782017] [0696cb192203d55d] address between user and kernel address ranges
[ 2727.828021] pc : kmem_cache_alloc+0x90/0x260
[ 2727.829114] lr : kmem_cache_alloc+0x64/0x260
[ 2727.847049]  kmem_cache_alloc+0x90/0x260
[ 2727.847163]  security_file_alloc+0x34/0x98
[ 2727.847280]  __alloc_file+0x60/0x100
[ 2727.847382]  alloc_empty_file+0x68/0x100
[ 2727.847496]  path_openat+0x58/0x13a0
[ 2727.847603]  do_filp_open+0x88/0x110
[ 2727.847873]  do_sys_open+0x188/0x2b8
[ 2727.848090]  __arm64_sys_openat+0x30/0x40
[ 2727.848246]  el0_svc_common.constprop.0+0xf4/0x200
[ 2727.849324]  el0_svc_handler+0x38/0xa8
[ 2727.849441]  el0_svc+0x10/0x180

 
Bisect shows:

git bisect good
79950e5d4c356fd4452c61f1097d53d30e779e7b is the first bad commit
commit 79950e5d4c356fd4452c61f1097d53d30e779e7b
Author: Qian Yingjin <qian@ddn.com>
Date:   Tue Jun 6 15:11:30 2023 +0700

    LU-16713 llite: writeback/commit pages under memory pressure

Revering this patch, solves the problem



 Comments   
Comment by Gerrit Updater [ 23/Nov/23 ]

"Artem Blagodarenko <ablagodarenko@ddn.com>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53220
Subject: LU-17311 llite: revert LU-16713 patch and fixes
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 76fe616dac94f39f59ca3dec78c043cc5e37cc25

Comment by Peter Jones [ 23/Nov/23 ]

Does this problem only affect older unsupported kernels like Ubuntu 20.04?

Comment by Artem Blagodarenko [ 24/Nov/23 ]

>Does this problem only affect older unsupported kernels like Ubuntu 20.04?

pjones This happened on 20.04.5 (5.4.0-48-generic kernel). I haven't checked on any other versions. I have pushed https://review.whamcloud.com/c/fs/lustre-release/+/53220 to estimate whether something changed on other builds. BTW, what Ubuntu version should I use to test supported clients for master?

Comment by Peter Jones [ 24/Nov/23 ]

Lustre 2.16 is targeting to support Ubuntu 22.04 clients. We really want to have cgroup support for Lustre and this is more complicated to do on older kernels but users wanting to use older kernels can use 2.15.x releases so this should not be an impediment to landing this code to master.

Generated at Sat Feb 10 03:34:23 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.