Automatic rollback to files copied with the backup argument

Question

I'm making an update protocol for an embedded linux system, but I want to give users the option to roll back the update if it messes with their system, especially for alpha/beta releases. So I discovered the cp command option to make automatic backups of overwritten files.

This works great! But there has to be an easy way to roll back these changes with the automatic backup files right? I don't want to force users to go and find the backups of all the changed files themself.

Thanks in advance for the help!

Answer 1

If you have space to have two parallel filesystem trees, you could possibly have something like /bin and /bin.old. Repeat for all top-level directories.

Installation would be a case of creating /bin.new, copying from the live system, overlaying the new files, and then switching /bin to /bin.old and /bin.new to /bin`, etc. Rollback would be a switch back.

# Prepare a new filesystem tree
#
rm -rf /*.new /*.old
for item in /*
do
    cp -al "$item" "$item.new"     # Links avoid using too much disk space
done

# Overlay. Because we linked in the previous step, we must remove
# (or rename) each file that we're going to change. Do not change
# or replace any file in situ
#
# If you have space for two full filesystem trees you could just copy
# instead of linking, which could simplify this update code section.
#
echo installation code goes here

# Switch over
#
OPATH="$PATH" PATH="/bin.old:/bin:/bin.new:$PATH"

for item in /*.new
do
    live="${item%.new}"

    mv -f "$live" "$live.old"
    mv -f "$item" "$live"
done

# Post-installation steps (rebuild kernel links, etc.)
#
echo post-installation code goes here

You might want to have vital system binaries in a place that could not get updated, so that the rollback didn't get burned by a poor shell. However, that's a bit of a chicken and egg situation, where you then by definition cannot update this code. Tricky.