This article is being updated. Please be aware the content herein, not limited to version numbers and slight syntax changes, may not match the output from the most recent versions of Bright. This notation will be removed when the content has been updated.
Unix-like systems such as Solaris and FreeBSD provide kernel facilities that limit the information that a user can obtain about the processes run by other users. Linux does this too, as of kernel 3.2.20 [1]. With this, a system administrator is able to restrict users to view only their own processes. You can also permit specific groups access to view all processes, which can be handy for system administrators. The restriction feature has already been pushed back into the kernel for Red Hat Enterprise Linux 6.3 [2], and from there to CentOS 6.3 and Scientific Linux 6.3. Recently, this feature was also backported to RHEL 5.9 [3].
If you are using Bright Cluster Manager version 6.0 and above, and you run it on one of these distributions, then all you have to do to make this security feature available is to:
- change the mount options of the /proc filesystem
- remount proc on all the nodes.
The settings take effect right away on the node or node category.
Example: Assume that the head node is called darkstar-head. Assume we have a category default. This category has the processing nodes, ie the nodes on which you want to restrict the ability users to view others’ processes
[root@darkstar-head ~]# cmsh[darkstar-head]% category [darkstar-head->category]% use default [darkstar-head->category[default]]% fsmounts [darkstar-head->category[default]->fsmounts]% use /proc [darkstar-head->category[default]->fsmounts[/proc]]% set mountoptions defaults,nosuid,hidepid=2[darkstar-head->category*[default*]->fsmounts*[/proc*]]% commit[darkstar-head->category[default]->fsmounts[/proc]]% quit[root@darkstar-head ~]#
The hidepid option accepts three values:
- 0 (zero) sets the /proc filesystem to the default configuration. This allows a user to view all processes on the system.
- 1 (one) allows a user to access only the /proc/[PID] directories that the user owns. PID directories under /proc that do not belong to the user are restricted.
- 2 restricts the user from viewing all PIDs, including all under /proc.
After you enter the commit command, the changes go into the node fstab files. The changes will take effect during the next reboot. However, if you need the changes to take effect immediately, you will have to ro remount the /proc filesystem on all the nodes. You can do this by running the pexec command on the head node.
pexec "mount -o rw,hidepid=2,remount /proc"Finding nodes that are alive...Running pexec.------------------------------------node001: Working directory: /root------------------------------------node002: Working directory: /root------------------------------------
Finally, if you have multiple system administrators on the system and you want to permit them to see all processes on the system for troubleshooting, you can add the gid option, to specify a group ID that is permitted to view all process on the system regardless of the hidepid option.
...[darkstar-head->category[default]->fsmounts[/proc]]% set mountoptions defaults,nosuid,hidepid=2,gid=100...