FreeBSD UFS2 Snapshots Management Environment

by Ralf S. Engelschall

The problem

The FreeBSD 5 UFS2 filesystem provides the possibility to create snapshots of live filesystems. It is already best known (and can be easily used) for allowing fsck(8) to run in the background (see rc.conf variable background_fsck) and to create consistent filesystem dumps (see dump(8) option -L).

Additionally, with the help of md(4) and mdconfig(8) you can mount snapshots as read-only filesystems, too. For illustration purposes, look at this sample session where a snapshot of /var is created:

# echo "before" >/var/tmp/foo.txt
# ls -l /var/tmp/foo.txt
-rw-r--r--  1 root  wheel  7 Sep  3 16:19 /var/tmp/foo.txt

# test -d /var/.snap || mkdir /var/.snap
# mount -u -o snapshot /var/.snap/test /var
# ls -l /var/.snap/test
-r--------  1 root  operator  2147483784 Sep  3 16:19 /var/.snap/test

# echo "after" >/var/tmp/foo.txt
# ls -l /var/tmp/foo.txt
-rw-r--r--  1 root  wheel  6 Sep  3 16:20 /var/tmp/foo.txt

# mdconfig -a -t vnode -f /var/.snap/test -u 1
# mount -o ro /dev/md1 /mnt
# ls -l /mnt/tmp/foo.txt /var/tmp/foo.txt
-rw-r--r--  1 root  wheel  7 Sep  3 16:19 /mnt/tmp/foo.txt
-rw-r--r--  1 root  wheel  6 Sep  3 16:20 /var/tmp/foo.txt
# cat /mnt/tmp/foo.txt /var/tmp/foo.txt
before
after

# umount /mnt
# mdconfig -d -u 1
# rm -f /var/.snap/test

As you can see, one can easily create a filesystem snapshot and use it to recover old states of files. Unfortunately, this is all our FreeBSD base system currently provided out-of-the-box. There is neither an easy way to regularily create new and expire old snapshots nor an easy -- or even unpriviledged -- way to access the snapshot data. Additionally, although the involved steps and commands are not really complicated, they are not convenient enough for daily work.

Keep in mind that regular filesystem backup snapshots are a great solution for easily recovering old file states within a small time frame, but only if the interface for the end-user follows the KISS principle. And at least this is IMHO not the case under our FreeBSD until now.

On the other hand, at work my home directory is placed on an NFS-mounted WAFL® filesystem provided by an ONTAP® based Filer from Network Appliance. For instance, after you have configured scheduled snapshots with the ONTAP® command snap sched 0 0 4, you can easily and at any time recover old states (as a regular user) in hourly steps for the last 4 hours:

$ echo "before" >foo.txt
$ sleep 3601
$ echo "middle" >foo.txt
$ sleep 3601
$ echo "after" >foo.txt

$ ls -la | fgrep .snapshot
$ cat .snapshot/hourly.1/foo.txt .snapshot/hourly.0/foo.txt foo.txt
before
middle
after

The solution

1. Snapshot Management Frontend

# snapshot list /tmp
# snapshot make -g3 /tmp:test
# snapshot list /tmp
/tmp    test.0
# snapshot make -g3 /tmp:test
# snapshot make -g3 /tmp:test
# snapshot list /tmp
/tmp    test.0
/tmp    test.1
/tmp    test.2
# snapshot make -g3 /tmp:test
# snapshot make -g3 /tmp:test
# snapshot list /tmp
/tmp    test.0
/tmp    test.1
/tmp    test.2
# snapshot mount /tmp:test.0 /mnt
# snapshot umount /mnt
# snapshot make -g0 /tmp:test
# snapshot list /tmp

2. Periodic and Flexible Backup Snapshot Creation

$ grep ^snapshot_ /etc/periodic.conf
snapshot_enable="YES"
snapshot_schedule="/,/usr:0:2:0 /var:0:2:4 /home:2:7:8@8,12,16,20"

3. Easy Access to Backup Snapshot Data

$ df | egrep '(/snap|/.am)'; ls -l /snap
pid1562@en2:/snap           0       0        0   100%    /snap

$ cd /home/rse
$ echo "before" >foo.txt
$ sleep 3601
$ echo "middle" >foo.txt
$ sleep 3601
$ echo "after" >foo.txt
$ cat /snap/home:hourly.1/rse/foo.txt /snap/home:hourly.0/rse/foo.txt foo.txt
before
middle
after

$ df | egrep '(/snap|/.am)'; ls -l /snap
pid1562@en2:/snap           0       0        0   100%    /snap
/dev/md1              2026030    2464  1861484     0%    /.am/en2/snap/var:hourly.0
lrwxrwxrwx  1 root  wheel  26 Sep  3 16:38 var:hourly.0 -> /.am/en2/snap/var:hourly.0
$ sleep 3600
$ df | egrep '(/snap|/.am)'; ls -l /snap
pid1562@en2:/snap           0       0        0   100%    /snap

As the surrounding df(1) and ls(1) calls show, amd(8) on-demand mounts the snapshot files. After about five minutes of no access it automatically unmounts them again.

With those three solutions, FreeBSD UFS2 snapshots are available to our unprivileged users as a short-time backup solution, allowing them to easily recover their older file states.

Implementation

The implementation is straight-forward and based on mount(8), mdconfig(8), amd(8) and cron(8).

1. Snapshot Management Frontend

First, I've implemented a little utility snapshot(8) (implementation peek: /usr/sbin/snapshot and /usr/share/man/man8/snapshot.8) which provides the convenience frontend for creating, expiring, mounting and unmounting snapshots.

2. Periodic and Flexible Backup Snapshot Creation

Second, I've implemented a periodic scheduler periodic-snapshot(8) (implementation peek: periodic-snapshot and /usr/share/man/man8/periodic-snapshot.8) which is called from /etc/crontab (periodic(8) cannot be used here because we need hourly runs and at exact times to not confuse users) and runs the snapshot make commands according to the configuration provided in /etc/periodic.conf.

3. Easy Access to Backup Snapshot Data

Third, I've set up an amd(8) map /etc/amd.map.snap in /etc/rc.conf which performs the snapshot mount and snapshot umount commands when a user accesses directories under snap.

In order to easily integrate this implementation into a FreeBSD 5-STABLE installation, download the latest tarball and install it:

# tar zxf freebsd-snapshot-*.tar.gz
# cd freebsd-snapshot-*
# make install
# /etc/rc.d/amd start

(If you later want to remove it again, just run make uninstall.)

References

See the following URLs for further details:

• mount(8)
• mksnap_ffs(8)
• mdconfig(8)
• amd(8)
• /usr/src/sys/ufs/ffs/README.snapshot
• http://www.mckusick.com/softdep/