|| ||Serge Hallyn <firstname.lastname@example.org>|
|| ||Re: [patch 3/3] lsm: add bsdjail documentation|
|| ||Wed, 06 Oct 2004 15:25:58 -0500|
|| ||Chris Wright <email@example.com>, firstname.lastname@example.org,
Attached is a patch carrying the documentation for the bsdjail LSM.
Signed-off-by: Serge E. Hallyn <email@example.com>
diff -Nrup linux-2.6.9-rc3-bk6/Documentation/bsdjail.txt linux-2.6.9-rc3-bk6-jail/Documentation/bsdjail.txt
--- linux-2.6.9-rc3-bk6/Documentation/bsdjail.txt 1969-12-31 18:00:00.000000000 -0600
+++ linux-2.6.9-rc3-bk6-jail/Documentation/bsdjail.txt 2004-10-06 10:51:46.000000000 -0500
@@ -0,0 +1,99 @@
+BSD Jail Linux Security Module
+Serge E. Hallyn <firstname.lastname@example.org>
+Implements a subset of the BSD Jail functionality as a Linux LSM.
+What is currently implemented:
+ If a proces is in a jail, it:
+ 1. Is locked under a chroot (as are all children) which is not
+ vulnerable to the well-known chdir(..)(etc)chroot(.) escape.
+ 2. Cannot mount or umount
+ 3. Cannot send signals outside of jail
+ 4. Cannot ptrace processes outside of jail
+ 5. Cannot create devices
+ 6. Cannot renice processes
+ 7. Cannot load or unload modules
+ 8. Cannot change network settings
+ 9. May be assigned a specific ip address which will be used
+ for all it's socket binds.
+ 10. Cannot see contents of /proc/<pid> entries of processes not in the
+ same jail. (We hide their existence for convenience's sake, but
+ their existance can still be detected using, for instance, statfs)
+ 11. Has no CAP_SYS_RAWIO capability (no ioperm/iopl)
+ 12. May not share IPC resources with processes outside its own jail.
+ 13. May find it's valid network address (if restricted) under
+The security of this module is very much dependent on the security
+of the rest of the system. You must carefully think through your
+use of the system.
+ 1. If you leave /dev/hda1 in the jail, processes in the
+ jail can access that filesystem (i.e. /sbin/debugfs).
+ 2. If you provide root access within a jail, this can of
+ course be used to setuid binaries in the jail. Combined
+ with an unjailed regular user account, this gives jailed
+ users unjailed root access. (thanks to Brad Spender for
+ pointing this out). To protect against this, use jails
+ in private namespaces, with the jail filesystems mounted
+ ONLY within the jail namespaces. For instance:
+$ # (Make sure /dev/hdc5 is not mounted anywhere)
+$ new_namespace_shell /bin/bash
+$ mount /dev/hdc5 /opt
+$ mount -t proc proc /opt/proc
+$ echo -n "root /opt" > /proc/$$/attr/exec
+$ echo -n "ip 220.127.116.11" > /proc/$$/attr/exec
+$ exec /bin/sh
+$ apachectl start
+How to use:
+ 1. modprobe bsdjail
+ [ 1.5 /sbin/ifconfig eth0:0 18.104.22.168;
+ 1.6 /sbin/route add -host 22.214.171.124 dev eth0:0
+ (optional) ]
+ 2. Make sure the root filesystem (ie /dev/hdc5) is not mounted
+ anywhere else.
+ 3. exec_private_namespace /bin/sh
+ 4. mount /dev/hdc5 /opt
+ 5. mount -t proc proc /opt/proc
+ 6. echo -n "root /opt" > /proc/$$/attr/exec
+ echo -n "ip 126.96.36.199" > /proc/$$/attr/exec (optional)
+ 7. exec /bin/sh
+ 8. sshd
+ 9. exit
+The new shell will now run in a private jail on the filesystem on
+/dev/hdc5. If proc has been mounted under /dev/hdc5, then a "ps -auxw"
+under the jailed shell will show only entries for processes started under
+If a private IP was specified for the jail, then
+ cat /proc/$$/attr/current
+will show the address for the private network device. Other network
+devices will be visible through /sbin/ifconfig -a, but not usable.
+If the reading process is not in a jail, then
+ cat /proc/$$/attr/current
+returns information about the root and ip * for the target process,
+or "Not Jailed" if the target process is not jailed.
+Cat /proc/$$/attr/exec gives a list of the valid keywords to cat into
+/proc/$$/attr/exec when starting a jail.
+Current valid keywords for creating a jail are:
+ root: Root of jail's fs
+ ip: Ip addr for this jail
+ nrtask: Number of tasks in this jail
+ nice: The nice level for this jail. (maybe should be min/max?)
+ slice: Max timeslice per process
+ data: Max size of DATA segment per process
+ memlock: Max size of memory which can be locked per process
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/