[LU-17437] convert xattr to acl problem

• Status: Open
• Project: Lustre
• Affects Version/s: Lustre 2.15.4
• Type: Bug
• Priority: Minor
• Reporter: Götz Waschk
• Assignee: WC Triage
• Resolution: Unresolved

Environment

Alma Linux 9.3 with Lustre 2.15.4 on the client
Scientific Linux 7.9 with Lustre 2.12.8 on the server

• Severity: 3

Description

I am trying to access a file that has an ACL as root and I cannot open it:

// -rw-rw-r--+ 1 giavitto ultrasat 22838 Jan  9 13:48 /lustre/fs24/group/ultrasat/first_light/udp_packets/sn204_shooting_v1-3_15s_pinhole_natural_light_1_with_2_zero_exposures_and_hotfixes_misalignment/2024-01-09T12-43-13/frame_009_image_0.png
# getfacl '/lustre/fs24/group/ultrasat/first_light/udp_packets/sn204_shooting_v1-3_15s_pinhole_natural_light_1_with_2_zero_exposures_and_hotfixes_misalignment/2024-01-09T12-43-13/frame_009_image_0.png' 
getfacl: Removing leading '/' from absolute path names
# file: lustre/fs24/group/ultrasat/first_light/udp_packets/sn204_shooting_v1-3_15s_pinhole_natural_light_1_with_2_zero_exposures_and_hotfixes_misalignment/2024-01-09T12-43-13/frame_009_image_0.png
# owner: giavitto
# group: ultrasat
user::rw-
group::rwx                      #effective:rw-
group:ultrasat:rwx              #effective:rw-
mask::rw-
other::r--

When I try to copy the file, I get the error

cp: cannot open '/lustre/fs24/group/ultrasat/first_light/udp_packets/sn204_shooting_v1-3_15s_pinhole_natural_light_1_with_2_zero_exposures_and_hotfixes_misalignment/2024-01-09T12-43-13/frame_009_image_0.png' for reading: Operation not supported

and in the syslog this message:

LustreError: 2481150:0:(mdc_acl.c:50:mdc_unpack_acl()) convert xattr to acl: -95

Comments

[Patrick Farrell]

Are you able to reproduce this problem by creating a new file with these properties or is it unique to this existing file?

[Götz Waschk]

I haven't managed to create new files with the same problem. There are several files with this problem in the same directory, they all have the same ACL, but some are readable and some are not.

Deleting the ACL with setfacl -b makes the files readable again.

[Andreas Dilger]

You could grab the binary ACL xattr from the MDT inodes with debugfs, and then compare them? Alternately, run the test with full debug (preferably on the MDS with a local mountpoint, so that you get the full set of logs:

# lctl set_param debug=all
# lctl clear; cp ...; lctl dk

and then review the logs to see where the "-95" error is coming from.

I recall some issues with inheriting very large ACLs, but that doesn't appear to be the case here.

[Götz Waschk]

Unfortunately, I have removed the ACLs from that directory so the users could access their files and I haven't found any other inaccessible files on this file system.