ZFS Tools

Various tools which assist with day-to-day ZFS business. Most of these were written around 2007, and some of what they do (like recursive snapshot sending) is now built into the real ZFS tooling.

zr

Provides a nice way of recovering or reverting individual files from snapshots. Use this from the live filesystem.

zp

Complementary to zr, promotes files out of a snapshot. Use this from inside a snapshot directory.

zfs_scrub_telemetry

Wraps a zfs scrub, sending telemetry about it to Wavefront.

zfs_snapshot

I run this from cron, and it takes automatic snapshots of my systems. Solaris has the auto-snapshot SMF service for this now, but zfs_snapshot predates that by several years and, IMO, is simpler to use.

zfs_snapshot [-pr] [-t day|month|date|time|now] [-o ptn,ptn] zfs
zfs_snapshot [-pr] [-t day|month|date|time|now] [-o ptn,ptn] -d dir

Options are as follows:

• -d dir: snapshots the dataset containing the directory dir.
• -t type: the type of snapshot to make. This applies an automatic naming convention. For example:
  • -t day @wednesday
  • -t month @january
  • -t date @2008-30-01
  • -t time @08:45
  • -t now @2008-30-01_08:45:00
• -o: lets you supply a comma-separated list of filesystems which you do no want to snapshot. Wildcards are allowed, but must be quoted. Useful for omitting, say, all log directories.
• -p: "dry-run" mode. Instead of removing the snaps, the script will print the zfs commands it would use, and exit.
• -r: recurse down the dataset heirachy. (zfs can do this itself now, with its own -r option.)

Existing snapshots with the same name are overwritten. With no arguments, the script snapshots every dataset it finds, though you can omit things with -o.

Examples

From my own crontab.

# Snapshot ZFS filesystems every day
#
45 6 * * * /usr/local/bin/zfs_snapshot -t day -o "*build,*logs"

# Snapshot home dirs twice an hour
#
0,30 * * * * /usr/local/bin/zfs_snapshot -t time space/export/home

# Snapshot monthly
#
45 5 1 * * /usr/local/bin/zfs_snapshot -t month -o "*patches,*build,*logs"

zfs_remove_snap

This script batch-removes ZFS snapshots. It takes the following options:

• -d dir: removes all snapshots in the dataset which contains the given directory.
• -D name: removes all snapshots belonging to datasets which contain name, in any pool on the machine. For instance, -D logs removes all snapshots on all */logs datasets.
• -n name: removes snapshots with the given name. That is, the part following the @. So, say you snapshot with the weekday name every day and you want to remove all Monday's snaps, use -n monday.
• -r: tells the script to recurse down datasets. So, if you have a pool called tank, -r tank would remove every snapshot it holds.
• -p: "dry-run" mode. Instead of removing the snaps, the script will print the zfs commands it would use, and exit.
• -h: print usage and exit.

zfs_real_usage

The way ZFS reports space can be a little confusing. This script tells you what datasets and snapshots are taking up real-estate on disk. It sorts from the least to the most, and takes no options. If you want to filter, grep is your friend.

I've found this script useful when I need to clear some space, and some deeply buried snapshot somewhere is hogging a stack of room.

zfs_scrub

This is a wrapper to zfs scrub, which I used to run monthly from cron. It scrubs every pool on the box sequentially, and when the scrub completes, it mails you the result. If you want to be contacted whether the scrub passes or fails, uncomment the MAILSUCCESS variable. With that commented out, the script only mails when the scrub finds errors.

By default the script writes its actions to the syslog, via the LOCAL7 facility.

zfs_send_stream

This was written in the days before Sun gave us recursive snapshotting and sending. I wrote it to do full transfers of pools, including all their snapshots and attributes, between machines.

It has two modes, which we'll call 'send' and 'restore'. Send mode is invoked like this:

zfs_send_stream -s <remote_host> -p <remote_pool> <dataset>

• -s user@host: this is fed straight to ssh -C. It's your responsibility to set up the users at each end, and do any necessary key exchange.
• -p pool: Doesn't necessarily have to be a pool (though I think it did in the original version.) It's the top-level target dataset on the receiving host.
• n: by default, the script will create a file which records all the properties of the datasets being sent (more of which in a moment.) If you don't want these, use this flag.

Restore mode is run on the receiving host like this:

zfs_send_stream -r <file>

It reads in the properties file I told you about in the -n option, and applies them to the received datasets.

find_rogue_snapshots

I have a snapshot naming scheme. This script finds snapshots which do not fit that scheme. Won't be useful to anyone else.

fix_zpool_caches

A quick fix for a dodgy computer I used to have. Won't be useful to anyone.