Prune corrupted Docker layers in overlayfs2 storage

prune-corrupted-layer-metadata.sh

#!/bin/sh
# (c) 2023 Stefan Agner
# This scripts attempts to clear corrupted Docker overlay2 storage metadata
# which can be left over after a power failure.
# See also: https://github.com/moby/moby/issues/42964
# 
# Typically the error message when attempting to pull the image for which
# corrupted layers are in the storage looks like:
# failed to register layer: error creating overlay mount to /mnt/data/docker/overlay2/6ee02298ee75a4f96e77f90551673cb700f29867f9ad1c4e20b8c816bfcf0735/merged: too many levels of symbolic links

# Docker storage directory, intended for Home Assistant OS
docker_storage_dir="/mnt/data/docker"

# Warning message
echo "WARNING: This script attempts to clear corrupted Docker overlay2 storage metadata."
echo "Use with caution, as it may result in data loss. Ensure you have backups before running this script."

# Initial sanity check for the overlay2 directory
overlay2_dir="$docker_storage_dir/overlay2"
layerdb_dir="$docker_storage_dir/image/overlay2/layerdb/sha256"

if [ ! -d "$overlay2_dir" ]; then
    echo "Error: Directory $overlay2_dir does not exist."
    exit 1
fi

# Find unique cache IDs
cache_ids=$(find "$overlay2_dir" -type f -size 0 \( -name lower -o -name link \) -exec sh -c 'basename $(dirname {})' \; | sort -u)

# Check if cache IDs are found
if [ -z "$cache_ids" ]; then
    echo "No cache IDs with size 0 found. Exiting."
    exit 0
fi

# Count the number of cache IDs found
num_cache_ids=$(echo "$cache_ids" | wc -w)

# Prompt the user for confirmation
echo "Found $num_cache_ids cache IDs with size 0. Do you want to continue and delete them? (yes/no)"
read user_confirmation

# Check user's response
if [ "$user_confirmation" != "yes" ]; then
    echo "Operation canceled. No directories were deleted."
    exit 0
fi

# Initialize counters for summary
deleted_cache_dirs=0
deleted_layerdb_dirs=0

# Iterate through cache IDs
for cache_id in $cache_ids; do
    # Find corresponding directory in layerdb
    layerdb_dir_path=$(dirname $(grep -r -e "$cache_id" "$layerdb_dir"/*/cache-id))
    cache_dir="$overlay2_dir/$cache_id"
    
    # Remove cache directory
    if [ -d "$cache_dir" ]; then
        rm -rf "$cache_dir"
        deleted_cache_dirs=$((deleted_cache_dirs + 1))
    fi
    
    # Remove LayerDB directory
    if [ -d "$layerdb_dir_path" ]; then
        rm -rf "$layerdb_dir_path"
        deleted_layerdb_dirs=$((deleted_layerdb_dirs + 1))
    fi
done

# Print summary
echo "Summary: Deleted $deleted_cache_dirs cache directories and $deleted_layerdb_dirs LayerDB directories."