btrfs-rescrub-errors

btrfs-rescrub-errors

#!/bin/bash

# Scan dmesg for btrfs error messages, and re-scrub the respective
# offsets.  Needs btrfs scrub with -s / -e switches:
# https://patchwork.kernel.org/patch/11268681/

set -eEuo pipefail
shopt -s lastpipe

blocksize=$((1*1024*1024*1024))

declare -A log2phys_cache

function log2phys() {
	local dev=$1
	local logaddr=$2
	log2phys_reply=${log2phys_cache[$logaddr $dev]-}
	if [[ -z "$log2phys_reply" ]]
	then
		log2phys_reply=$(btrfs-map-logical "$dev" -l "$logaddr" 2>/dev/null)
		log2phys_cache[$logaddr $dev]=$log2phys_reply
	fi
}

declare -A bad

printf 'Parsing dmesg...\n' 1>&2

dmesg |
	while read -r line
	do
		if [[ "$line" =~ \ error\ at\ logical\ [^\ ]*\ on\ dev\ (.*),\ physical\ ([^ ]*), ]]
		then
			dev=${BASH_REMATCH[1]}
			physaddr=${BASH_REMATCH[2]}
			printf '%s %s\n' "$physaddr" "$dev"
		elif [[ "$line" =~ \ error\ at\ logical\ ([^ ]*)\ on\ dev\ ([^ ]*)$ ]]
		then
			logaddr=${BASH_REMATCH[1]}
			dev=${BASH_REMATCH[2]}
			log2phys "$dev" "$logaddr"
			printf %s "$log2phys_reply" | \
				while read -r line
				do
					if [[ "$line" =~ (^mirror [^ ]* logical [^ ]* physical ([^ ]*) device (.*)$) ]]
					then
						physaddr=${BASH_REMATCH[2]}
						dev=${BASH_REMATCH[3]}
					else
						printf 'Failed to parse btrfs-map-logical output: %q\n' "$line" 1>&2
						continue
					fi
				done
			printf '%s %s\n' "$physaddr" "$dev"
		fi
	done |
	while read -r addr dev
	do
		block=$((addr/blocksize))
		bad[$((block-1)) $dev]=1
		bad[$((block  )) $dev]=1
		bad[$((block+1)) $dev]=1
	done

printf '%s\n' "${!bad[@]}" |
	sort -n |
	while read -r block dev
	do
		if [[ ! -v "bad[$block $dev]" ]]
		then
			continue
		fi

		unset 'bad[$block $dev]'
		block2=$block
		while [[ -v "bad[$((block2+1)) $dev]" ]]
		do
			block2=$((block2+1))
			unset 'bad[$block2 $dev]'
		done

		addr1=$((block *blocksize))
		addr2=$((block2*blocksize))
		printf 'Rescanning %s - %s on %s:\n' "$addr1" "$addr2" "$dev" 1>&2
		btrfs scrub start -B -s "$addr1" -e "$addr2" "$dev"
	done