I need to transfer an entire linux root filesystem off of a failing hard drive onto another computer with an open, available partition. I've pretty sure this involves tar and ssh, but I can't remember exactly how to do this.

I'm imaging probably using a live cd on the new/target host to run something like this:

ssh user@failingharddrivehost "some tar command | piped into something else"

migrated from stackoverflow.com May 11 '13 at 5:09

This question came from our site for professional and enthusiast programmers.

  • offtopic. not a programming question. but try ssh user@failingsys "tar cfz - /" > oldsys.tar.gz – Marc B May 11 '13 at 4:46
  • You want to take care that you do not tar /dev/ (e.g. /dev/random, /dev/sdX, ... ). Ditto /proc/ – Hennes May 11 '13 at 10:00

Use rsync. From the new host, you can use

rsync -avP --numeric-ids --exclude='/dev' --exclude='/proc' --exclude='/sys' root@failedharddrivehost:/ /path/to/destination/

I wouldn't try involving something like tar because it probably won't work when there are broken files.

  • Ended up using rsync, as describe above by etagenklo and @Hennes. Migration went well. Just need to fix grub, but that shouldn't be too bad. – CHK May 13 '13 at 14:03
  • 7
    I'd add -AHX as flags to rsync as well in order to preserve acls, xattrs and hardlinks resulting in an even more exact copy of the original fs. – Alexander Remesch Aug 5 '16 at 8:53
  • 4
    I'd add -x to ignore items not on the root filesystem, then you can skip the --exclude arguments for various paths. – Alex Jan 15 '18 at 1:54

If both computers are on the same (safe) LAN, I recommend a different approach using netcat. This is usually much faster as it doesn't encrypt the data.

root@good_host$ cd good_partition; netcat -l -p 1234 | tar xvpmf -
root@bad_host$ tar -cv -f- --exclude=/proc --exclude=/sys / | netcat good_host.ip 1234

which opens a listening port 1234 on the good machine netcat -l -p 1234 and pipes the incoming data to tar to extract (preserving mtime and permissions). The bad host sends the data to this port, also using tar and netcat. I included some --exclude parameters, as /proc and /sys are virtual filesystems and hence useless on the new host. (especially the file representing your RAM in (/proc/kcore) will add an unnecessary amount of data).

However, you should (also) consider to make a dd dump of the failing drive's partitions:

user@good_host$ cd good_partition; netcat -l -p 1234 > dump_of_bad_partition_1.dd
root@bad_host$ dd if=/dev/sda1 | netcat good_host.ip 1234

where you had to adopt /dev/sda1 to the right device. Do that with other partitions on the failing drive, too.

With that dump you are sure, that you did not miss any important metadata (like ACLs) which tar won't capture.


Why are you combine with directory excluding? Isn't better idea mount the same device into another directory? modern kernels allows this way. for example, you have mounted

/dev/sda1 as / then do: mkdir /CLEANROOT mount /dev/sda1 /CLEANROOT

after this you have: /dev/sda1 as / /dev/sda1 as /CLEANROOT

This is the same filesystem visible with two places, but /CLEANROOT haven't got additive mounts. Then you can tar or rsync /CLEANROOT without any exclusions instead copying / with exclusions.

Of course you must copy another data partitions when you've got some.

Copying partition is first step for server recovery. another is regenerate boot sectors, otherwise system wont boot from copied disk. Usefull is rescue mode when you boot from install/rescue CD or pendrive.


Do you have physical access to the failing host?

If you do then boot from a live CD. Then use:

  • dump (dump/restores whole filesystems including its permissions).
  • Tar with /dev excluded. You can combine this with outputting to std_out and piping that though netcat
    The exclude syntax is: tar --exclude='/dev'.
  • or rsync with the same excludes. E.g.
    rsync -zvr --exclude /dev/ / destination_computer_name_or_ip
  • or use dd like this:
    nc -l 4242 | gunzip | cat > my_full_disk_backup_of_PC_named_foo
    dd if=/dev/sda of=- bs=1M | gzip | nc -p 4242 name_of_the_destination

If you can not boot from a live CD then some most of the above solutions will stay the same, but:

  1. Some files may be in use/locked.
  2. Make sure to exclude not just /dev/ but also /proc/.
    E.g. tar --exclude='/dev' --exclude='/proc'
  • Yes I have physical (and root) access to both hosts. Going to try with rsync. the netcat crashed for the failing box for unknown reasons. the target filesystem doesn't want to be mounted. it fails with: #mount /dev/sda1 /mnt/fedora mount: unknown filesystem type 'LVM2_member' – CHK May 11 '13 at 22:59

Here is a description of how to copy files using tar and ssh. Basically, you would run one of the following, depending on whether you want to copy local -> remote or remote -> local:

tar cf - files... | ssh remotehost -c 'cd /destination && tar xvf -'

ssh remotehost -c 'cd /destination && tar cf - files' | tar xvf -

you should consider using rsync

the follwoing command assumes 2 things:

  1. you are on the system with failing hard drive
  2. new partition has a minimal linux installation with ssh enabled.

rsync / new_partition:/wherever/you/want/

Note: the trailing / is important, otherwise your files will end up in one directory level above

  • The OP probably want to use the -a parameter (and perhaps the -A), as it preserves times, ownership, symlinks (and perhaps ACLs) etc. and the -e ssh parameter, as the data should be transferred to another computer. So rsync -aAv -e ssh root@failingharddrivehost:/ /good_computer/newpartition_mountpoint – mpy May 11 '13 at 7:42

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.