I'm trying to set up a space-efficient rotating backup scheme with rsnapshot / rsync from a server to a Hetzner storagebox. I'm having a hard time understanding how hard links on the destination are affecting the disk usage being reported. In short: even though hard links seem in place on the backup destination, they don't seem to be taken into account in disk usage, but instead counted as full files.

Since the destination folder of rsnapshot must be on the local file system, I have set up a workflow consisting of 2 parts:

  1. create a local snapshot with rsnapshot, in a local folder on the source server
  2. rsync that local snapshot over SSH with rsync to the destination

That seems to work well and fast, but I have one concern: disk usage reported on the destination (with du -sh) seems to cumulate the size of all snapshots, even though they seem to have been copied properly using hard links. Note: since Hetzner storageboxes don't allow interactive SSH login, I'm inspecting this backup destination as a mounted volume with CIFS.

For example, after 3 rounds of this rsnaphsot + rsync combo, the destination folder contains daily.0, daily.1, and daily.2 folders. When checking random files in those snapshot folders for hard links, I do get expected results:

  1. find /mnt/user.your-storagebox.de/rsync-backup/ -name "output.file" -print0 | xargs -0 ls -li:

    351317 -rw-rw---- 3 root root 8650 Dec 15 11:25 /mnt/user.your-storagebox.de/rsync-backup/daily.0/home/user/output.file
    351317 -rw-rw---- 3 root root 8650 Dec 15 11:25 /mnt/user.your-storagebox.de/rsync-backup/daily.1/home/user/output.file
    351317 -rw-rw---- 3 root root 8650 Dec 15 11:25 /mnt/user.your-storagebox.de/rsync-backup/daily.2/home/user/output.file

    returns 3 files with identical inodes, and a link count of 3 (as expected)

  2. find /mnt/user.your-storagebox.de/rsync-backup/ -samefile /mnt/user.your-storagebox.de/rsync-backup/daily.0/home/user/output.file


    returns 3 files (as expected)

I guess this indicates that these snapshots have been properly copied to the destination as hard links.

Yet... when checking their disk usage on the destination location: du -sh /mnt/user.your-storagebox.de/rsync-backup, a value of 12G is returned. This is unexpected, since the original source folder only is about 4G. Apparently, disk usage is calculated cumulatively, despite the hard links?

OTOH, when inspecting the destination folder via rsnapshot du, I'm getting output that does seem to take hard links into account:

4.3G    /mnt/user.your-storagebox.de/rsync-backup/daily.0/
41K     /mnt/user.your-storagebox.de/rsync-backup/daily.1/
41K     /mnt/user.your-storagebox.de/rsync-backup/daily.2/
4.3G    total

This is confusing: either the snapshots are being copied with hard links, and should take up minimal space (which seems to be the case when inspecting the inodes), or they are not and are taking up much more space than expected (as suggested by the du -sh output).

My main concern is: is the disk usage reported on this mounted volume correct or nog? Are there any caveats w.r.t. the use of du -sh on mounted volumes I should be aware of?


My version of du (Debian, du (GNU coreutils) 8.30) handles files with hardlinks and counts the multiple instances only once. It would appear that yours does not. You can verify this fairly easily, though

Prepare the scenario

mkdir zzz                 # Scenario workspace
tar cf zzz/etc.1 /etc     # Ignore "tar: Removing leading `/' from member names"

Trial #1. Two files copied but not hardlinked

cp zzz/etc.1 zzz/etc.2    # Create copy

du -s zzz/etc.1           # 2580 KB, in my instance
du -s zzz/etc.2           # As you would expect, the same value
du -s zzz                 # 5164 KB, because the files are "different"

Trial #2. Two files hardlinked together

rm zzz/etc.2
ln zzz/etc.1 zzz/etc.2    # Create hardlink

du -s zzz/etc.1           # Unchanged from above, of course, 2850 KB
du -s zzz/etc.2           # As you would expect, still the same value
du -s zzz                 # For me, this is still the same value, 2580 KB

If your instance of du cannot handle multiple instances of the same hardlinked file, your trial #2 will return the total of both etc.1 and etc.2 just like it did for trial #1.

Using this information you can determine whether your version of du is being misleading, or that the files really are using up more disk space that you would expect. (Given your other metrics I'm fairly sure it's the former.)

  • Many thanks, @roaima. Following your exact steps, I'm getting the results you listed in trial #2: hard links are counted only once. This is what I'd expect from my rsync'ed rsnapshot backup, but even though the files in those respective snapshot folders do seem to have the same inode and are listed as hard links (with ls -li), du seems to count them double, resulting in 12G total size. OTOH, the rsnapshot du command does treat these snapshot folders as hard links, resulting in about 4G total size. Also, dh seems to count only 4G, so the behaviour of du is still puzzling. – rvdb Dec 21 '20 at 13:42
  • Ah, but when passing the separate snapshot folders to du, this shows different results! For example: du -sh /mnt/user.your-storagebox.de/rsync-backup/* does produce results similar to the rsnapshot du command: 4.3G /mnt/user.your-storagebox.de/rsync-backup/daily.0/ 41K /mnt/user.your-storagebox.de/rsync-backup/daily.1/ 41K /mnt/user.your-storagebox.de/rsync-backup/daily.2/ This at least suggests that hardlinks are being preserved. – rvdb Dec 21 '20 at 15:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.