[LU-12244] lfs check subcommand no longer works as non-root user

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Affects Version/s: Lustre 2.10.6
• Labels: llnl
• Environment: Client/Server: lustre-2.10.6_2.chaos-1.ch6.x86_64 ; kernel: 3.10.0-957.5.1.3chaos.ch6.x86_64
• Severity: 3
• Assignee: Peter Jones
• Reporter: Cameron Harr

Description

With the Lustre 2.10 client, non-root users appear to no longer be able to run "lfs check <servers|osts|mds>". Strace of the command shows a "Permission denied" error accessing /sys/kernel/debug/lustre/devices, resulting in the following error at runtime: error: check: mds status failed

Our Operations staff w/o root access needs "lfs check ..." functionality to monitor and fix file system issues, so fixing this issue would helpful for us.

statfs("/sys/kernel/debug/", {f_type=DEBUGFS_MAGIC, f_bsize=4096, f_blocks=0, f_bfree=0, f_bavail=0, f_files=0, f_ffree=0, f_fsid={0, 0}, f_namelen=255, f_frsize=4096, f_flags=ST_VALID|ST_RELATIME}) = 0
stat("/sys/fs/lnet/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
stat("/sys/fs/lustre/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
stat("/sys/kernel/debug/lnet/devices", 0x7fffffff8740) = -1 EACCES (Permission denied)
stat("/sys/kernel/debug/lustre/devices", 0x7fffffff8740) = -1 EACCES (Permission denied)
stat("/proc/fs/lnet/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
stat("/proc/fs/lustre/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
stat("/proc/sys/lnet/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
stat("/proc/sys/lustre/devices", 0x7fffffff8760) = -1 ENOENT (No such file or directory)
write(2, "error: check: mds status failed\n", 32error: check: mds status failed
) = 32
exit_group(2)                           = ?
+++ exited with 2 +++

Issue Links

• is related to: LU-11850 Relocating /proc/fs/lustre/ost to /sys/kernel/debug/lustre/ost prevents non-root access
• is related to: LU-8066 Move lustre procfs handling to sysfs and debugfs.

Comments

[Olaf Faaland]

For our own recordkeeping, our local ticket: TOSS-4503

[Olaf Faaland]

Yep, that patch worked without any other patches or modification.

[Olaf Faaland]

Andreas, we haven't tried it, but we can do so today and post back.

[Andreas Dilger]

Cameron, did you try out the patch for b2_10? Did it solve your problem? It only affects the userspace tools on the client, so you wouldn't need to upgrade all of the kernel modules or take an outage to install it.

[Cameron Harr]

... And I had searched only on 2.10. Thanks.

[James A Simmons]

LU-11850 only impacts 2.12 LTS users. 

[Cameron Harr]

Thank you both. My search for related tickets missed LU-11850, which is very similar.

Sparing Redhat from some blame, we only recently started rolling out 2.10 clients (from 2.8) so this is only affecting us now.

[Andreas Dilger]

I've cherry-picked the patch to b2_10:

https://review.whamcloud.com/34782

[Andreas Dilger]

The move of the /proc/fs/lustre/devices file from procfs to debugfs was done as part of patch https://review.whamcloud.com/23428 "LU-8066 obdclass: move lustre sysctl to sysfs" landed for 2.9.56, so it has been in all 2.10 releases.

I suspect the reason it is a problem now is that RedHat has backported a change from newer kernels to their kernel that makes debugfs root-only.

[James A Simmons]

We have fixes for that which landed to newer lustre versions. You need patch:

https://review.whamcloud.com/33799