Applicable to:
- Plesk for Linux
Symptoms
- The server runs CloudLinux OS with CageFS enabled.
- After WP Toolkit is updated, the
/usr/local/bin/wpwrapper cannot load WP-CLI for users with CageFS enabled. -
The
/usr/local/psa/admin/plib/modules/wp-toolkit/vendor/wp-cli/wpt-wp-cli.phpfile exists on the server but is not visible inside the user's cage:# cagefs_enter_user USER /bin/sh -c '[ -f /usr/local/psa/admin/plib/modules/wp-toolkit/vendor/wp-cli/wpt-wp-cli.php ] && echo VISIBLE || echo MISSING'
> MISSING -
The
/var/log/plesk/panel.logfile may contain a failed removal of the old WP Toolkit directory during the update:CONFIG_TEXT: removing wp-toolkit-old-2026-10-15-03-13-39/vendor/wp-cli failed: Device or resource busy
Cause
This is a WP Toolkit bug with ID PPPM-15580.
When WP Toolkit is updated, its module directory is replaced and the old one is deleted, but the CageFS skeleton bind mount of the vendor/wp-cli directory stays attached to the deleted directory.
Resolution
The bug will be fixed in one of the future Plesk updates and the bug number will be published to changelogs once it is released.
As a workaround, remount CageFS after the WP Toolkit update has finished:
- Connect to the server via SSH.
-
Check whether the CageFS mount is stale:
# MP=/usr/share/cagefs-skeleton/usr/local/psa/admin/plib/modules/wp-toolkit/vendor/wp-cli
# findmnt -T "$MP" -o TARGET,SOURCE
Note: The mount is stale if the SOURCE column contains
wp-toolkit-old-or ends with//deleted. Continue only in this case. -
Check that the file is missing inside a cage:
Note: Replace USER with any user that has CageFS enabled.
# cagefs_enter_user USER /bin/sh -c '[ -f /usr/local/psa/admin/plib/modules/wp-toolkit/vendor/wp-cli/wpt-wp-cli.php ] && echo VISIBLE || echo MISSING'
-
If Steps 2 & 3 shows issues then remount CageFS:
Warning: The following command stops all processes running inside existing cages, including shell sessions and long-running scripts of CageFS users, and resets LVE limits to their configured values. Run it in a low-traffic period.
Note: The
cagefsctl --updatecommand does not repair this state. Editing/etc/cagefs/cagefs.mpor runningcagefsctl --force-updateis not required.# cagefsctl --do-not-ask --remount-all
-
Verify that the mount follows the current directory and the file is visible again:
# findmnt -T "$MP" -o TARGET,SOURCE
# cagefs_enter_user USER /bin/sh -c '[ -f /usr/local/psa/admin/plib/modules/wp-toolkit/vendor/wp-cli/wpt-wp-cli.php ] && echo VISIBLE || echo MISSING'
The SOURCE column must show
wp-toolkit/vendor/wp-cliwithoutwp-toolkit-old-or//deleted, and the second command must printVISIBLE.
- Source: Zendesk ticket 96096063, escalated in PPS-19432. Bug PPPM-15580 is in status New, with no fix version or ETA.
- Reproduced in a lab on CloudLinux 9.8, Plesk 18.0.81, cagefs 7.6.49, upgrading WP Toolkit 6.11.4 to 6.11.5. A manual swap of the directory (copy, rename, move, delete) gives the same result.
- After the upgrade the skeleton mount points to
wp-toolkit-old-<timestamp>/vendor/wp-cli//deletedand the skeleton inode differs from the host inode.cagefsctl --update --do-not-ask --silentleaves it stale,cagefsctl --remount-allrepairs it even with a cage process running. WP Toolkit maintenance and--force-updateare not needed. - The
!line in/etc/cagefs/cagefs.mpis added by WP Toolkit maintenance (a read-only bind mount ofvendor/wp-cliinto the skeleton), not an exclusion. It is not the cause of this issue. - The "Device or resource busy" error on removal of the old directory was seen on the customer servers but not reproduced in the lab.
- The remount was verified only in the lab, not on customer servers.
Comments
Please sign in to leave a comment.