BorgBackup #

Pull-mode backups #

Usually, a client pushes new backups to a repository. That means the client initiates the connection and obviously needs to be able to access it at any time. Sometimes, you may wish to initiate a backup from the “server” that holds the repository, if that direction is easier. For example, you might want to create backups of a VPS and store it on a server at home, which is behind a NAT etc.

This is based on the deployment docs about “pull mode” in borgbackup/borg.

Wrap borg serve with socat #

The server holding the repository can start borg serve, which communicates over standard input and output. Normally this is started as part of an SSH command, so there is no special argument to make it listen on a socket instead. Luckily, socat can do that for us:

socat UNIX-LISTEN:/tmp/borg.sock,fork \
  EXEC:"borg serve --append-only --restrict-to-repository /path/to/repository"

The deployment docs go into a little more detail about permissions on the socket. If you have multiple repositories and concurrently running jobs, you should probably use temporary directories within /tmp with proper permissions.

The “append-only” mode is confusing in borg – it probably does not do what you think it does.

It does not prevent that a client can issue deletions or prune old archives. Instead it just keeps a transaction log on the server, so that you could undo these deletions. However, you first need to find out that and when you were compromised, which does not seem to be trivially easy at this point.

Forward the socket #

Next, forward this socket to the “client” machine which has the data you want to backup:

ssh -R /run/borg.sock:/tmp/borg.sock

You probably won’t be able to use /run/borg.sock when you’re not running as root remotely; adjust accordingly.

In order to make the SSH server clean up “dangling” sockets, which could prevent a forwarding you can add StreamLocalBindUnlink yes to the client’s /etc/ssh/sshd_config.

Connect to the socket interactively #

If you ran the above interactively, you can now tell borg how to use this forwarded socket and connect to it, either using the --rsh flag or BORG_RSH environment variable. This command is what borg executes when it usually connects to a remote server over SSH; again borg wants to communicate with standard input and output of this command and expects a borg serve on the other end. So you can either wrap socat again or use the simpler nc in this case.

The path to the repository needs to “look” like an ssh:// path for borg to use the RSH command. The hostname can obviously be anything you want but the path needs to match the path to the repository on the server (as given above for borg serve).

BORG_RSH="sh -c 'exec nc -U /run/borg.sock'" \
  borg list ssh://pullsock/path/to/repository

Automate it with environment variables #

If at this point you try the “all-in-one” command from the docs above, it won’t work because ssh will mangle the commandline and won’t properly pass the --rsh flag in one piece. You can escape that with printf '%q' or ${cmd@Q} if you like.

Instead – and if you want to pass an encryption password you should do this anyway – I propose to put everything in environment variables locally and forward them properly with SSH. This requires that the sshd server on the “client” accepts the variables, so adjust AcceptEnv accordingly:

# Allow client to pass certain environment variables
AcceptEnv LANG LC_* BORG_*

Now you can put everything in BORG_* environment variables as usual and use SSH with -o SendEnv="BORG_*" like so:

export BORG_RSH="sh -c 'exec nc -U /run/borg.sock'"
export BORG_REPO="ssh://pullsock/path/to/repository"
export BORG_PASSPHRASE="MySuperSecurePassphrase"

ssh -o SendEnv="BORG_*" -R /run/borg.sock:/tmp/borg.sock \
  borg create ::{hostname}-{now} /path/to/data

Example scripts #

These are example scripts combining all of the things above. Put borgctl on the remote machine somewhere, generate a new SSH identity locally and put its public key into ~/.ssh/authorized_keys together with a “forced” command, as shown in the script. Then use borgpull locally.

#!/usr/bin/env bash
set -eu

# this file initiates the borg backup on the "far" side
# e.g. put it in a ssh forced command like so:
#  restrict,port-forwarding,command="/usr/local/bin/borgctl" ssh-ed25519 AAAAC3NzaC1lZD..... borgpull key
echo "[$(date)] running borgctl on $(hostname)"

# borg socket should be given
if ! [[ -S $BORG_SOCKET ]]; then
  echo "err: borg socket does not exist" >&2
  exit 10
  # delete socket after we're done
  trap "rm ${BORG_SOCKET@Q}" EXIT

# repository path on remote should be given
if [[ -z ${BORG_REPOPATH+defined} ]] || ! [[ $BORG_REPOPATH =~ ^/ ]]; then
  echo "err: absolute borg repository path required" >&2
  exit 10

# configure borg with environment variables
export BORG_RSH="sh -c \"exec nc -U ${BORG_SOCKET@Q}\""
export BORG_REPO="ssh://socket${BORG_REPOPATH}"

borg create --compression zstd --stats '::{hostname}-etc-{now}' /etc ;;
#!/usr/bin/env bash
set -eu

# repository to provide for targets
export BORG_REPOPATH=/path/to/repository

# start borg server on a socket
export BORG_SOCKET=$(mktemp -u /tmp/borgpull-$(date +%s)-XXXXXXXX)
socat \
  UNIX-LISTEN:"${BORG_SOCKET}",umask=077,fork,unlink-close=1 \
  EXEC:"borg serve --append-only --restrict-to-path ${BORG_REPOPATH@Q}" &
trap "kill $! 2>/dev/null || true" EXIT

# pull the backup from server
timeout 30m ssh -T \
  -o SendEnv=BORG_\* \
  -i ~/.ssh/id_borgpull \