Fixing bad blocks on HDDs using


You hard drive or SMART tool reports errors when reading specific blocks similar to this message:

[3142.686141] end_request: I/O error, dev sda, sector 31415926

No matter how often you read the block, the hard drive still returns an error and does not reallocate the block.


Hard drives are programmed not to reallocate the block until someone writes to said block. This means that for normal users the program reading the bad block probably won’t fix the error by itself as most programs exhibit read-before-write usage patterns often resulting in a crash before any block is written. By using, the script presented in this post, you can force your Linux-based OS to rewrite the blocks, effectively fixing the block errors if the HDD has reallocation space left. Usage of said script is only recommended for professional IT personnel.


You can use this script, in order to automatically write the blocks yielding errors. While the data stored in those blocks will be lost forever, you won’t any read error after writing to it.

Syslog monitoring mode operates in one of several modes including automatic sequential scan. The most straightforward mode, however, is to continously scan the system log for error message like that outlined above. The tool automatically extracts the LBA (logical block address) from the system log and writes it using hdparm (use sudo apt-get install hdparm or equivalent if not already installed).

In order to use this mode, run

sudo --loop /dev/sda

in the background. In another shell, run the program yielding the error message repeatedly until the file can be read without an error. Every five seconds, will re-scan the syslog and attempt to rewrite all damaged blocks. When finished, stop fixhdd.pyusing Ctrl+C.

Sequential block scan mode

After these errors are resolved, I recommend using smartctl -t [short|long] to run a SMART test on the hard drive (even a short two-minute tests will often yield a LBA for the first bad block). After the selftest has finished, use smartctl -a to find the first LBA of first error.

For this example, we will assume the LBA of first error is 1234567. To get the offset for (i.e. the first LBA that will be scanned), substract a safety margin of about 100-1000 from the LBA of first error so the script will recognize errors occuring before the given LBA. The script will now try to read all LBAs starting from the offset, rewriting any bad blocks in the process. You can also start at offset 0 and wait several hours to days for the whole HDD to be scanned.

sudo -a -o 1234000 /dev/sda is EXTREMELY dangerous and might destroy all your data in just a few seconds. I recommend using it only if you understand the source code and know how the script and hdparm work. Even then it is your own responsibility if any of your data gets lots (in theory, there is a remote possibility of hdparm damaging your hardware, however I believe this is next to impossible). While I used several times to fix broken computers in the past, it might have critical bugs on other systems. Even hardware damage is possible, considering the power of hdparm. currently doesn’t include a simulation mode and silently bypasses the hdparm --yes-i-know-what-i-am-doing flag.

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
This script automatically re-writes sectors where
ATA read errors occur. By re-writing the sectors
(using hdparm), the HDD/SSD will be used to re-allocate
the sectors.

This script will NOT ask before overwriting data
and might DESTROY all your data. Use it under your
own responsibility and only if you know EXACTLY
what you're doing (or if you don't care).
Expect to contain critical bugs.

Runs on linux only. hdparm must be installed. must be run as root. It will only write to sectors
if reading them using hdparm yields an error.

Use --loop to watch the syslog for read errors
and rewrite all sectors where errors occur. The script will
check the log every five seconds and won't exit.

Use -a -o <offset> to scan for bad blocks starting at
LBA <offset>. Use this mode if a SMART selftest indicates an error
at a specific LBA and select an offset smaller than the given LBA.
Scanning a large number of LBAs takes a significant amount of time,
especially if many LBAs yield errors.

Use fixhdd -s <sector> to rewrite a specific LBA, but only
if reading it . Use this for correcting errors indicated by SMART
if you don't see the need for actively scanning a significant number
of blocks.

Use Ctrl+C to stop

    Revision 1.1: Fix --loop causing unary function to be called without arguments
    Revision 1.2: Fix hardcoded /dev/sda, various small improvements & fixes ; fix active scan
    Revision 1.3: Python3 ready
import subprocess
import time
import os
import stat
import sys

__author__ = "Uli Köhler"
__copyright__ = "Copyright 2015-2016 Uli Koehler"
__license__ = "Apache License v2.0"
__version__ = "1.3"
__maintainer__ = "Uli Köhler"
__email__ = ""
__status__ = "Development"

#Get list of recent bad sectors via dmesg
def getBadSectors(device):
    "Parse a list of recently read bad sectors from the syslog"
    #TODO this gets ALL bad sectors from ALL devices, not only the selected device
        out = subprocess.check_output('grep "end_request: I/O error" /var/log/syslog', shell=True)
        for line in out.split("\n"):
            line = line.strip()
            if not line: continue
            sector = int(line.rpartition(" ")[2])
            yield sector
    except subprocess.CalledProcessError:
        #usually this indicates grep has not found anything

def isSectorBad(device, sector):
        subprocess.check_output('hdparm --read-sector %d %s' % (sector, device), shell=True, stderr=None)
        return False
        return True

def resetSectorHDParm(device, sector):
    """Write to a sector using hdparm only if reading it yields a HDD error"""
    #Will throw exception on non-zero exit code
    if isSectorBad(device, sector):
        print(("Sector %d is damaged, rewriting..." % sector))
        #Maaan, this is VERY DANGEROUS!
        #Really, no kidding. Might even make things worse.
        #It could work, but it probably doesn't. Ever.
        #Don't use if your data is worth a single dime to you.
        out = subprocess.check_output('hdparm --write-sector  %d --yes-i-know-what-i-am-doing %s' % (sector, device), shell=True)
        print (out)
        print(("Sector %d is OK, ignoring" % sector))
def fixBadSectors(device, badSectors):
    "One-shot fixing of bad sectors"
    print(("Checking/Fixing %d sectors" % len(badSectors)))
    [resetSectorHDParm(device, sector) for sector in badSectors]
def checkDmesgBadSectors(device, knownGoodSectors):
    #Grab sector list from dmesg
    dmesgBadSectors = set(getBadSectors(device))
    if len(dmesgBadSectors) == 0:
        print ("No new sector errors found in syslog :-)")
        #Update set of sectors which are known to be good
        fixBadSectors(device, dmesgBadSectors)

def loopCheckForBadSectors(device):
    knownGoodSectors = set()
    while True:
        print("Waiting 5 seconds (hit Ctrl+C to interrupt)...")
        #Try again after timeout
        checkDmesgBadSectors(device, knownGoodSectors)

def isBlockDevice(filename):
    "Return if the given filename represents a valid block device"
    return stat.S_ISBLK(os.stat(filename).st_mode)

def getNumberOfSectors(device):
    "Get the physical number of LBAs for the given device"
    #Line like: 255 heads, 63 sectors/track, 60801 cylinders, total 976773168 sectors
    sectorsLine = subprocess.check_output("LANG=C fdisk -l {0} 2>/dev/null | grep ^Disk | grep sectors".format(device), shell=True)
    return int(sectorsLine.strip().split(" ")[-2])

def performActiveSectorScan(device, offset=0, n=1000):
    "Check all sectors on the hard drive for errors and fix them."
    print(("Performing active sector scan of {0} starting at {1}").format(device, offset))
    for i in range(offset, min(getNumberOfSectors(device), offset + n)):
        #Reset sector (only if it is damaged)
        resetSectorHDParm(device, i)

if __name__ == "__main__":
    # Parse arguments
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument("-s", "--sector", nargs="*", default=[], type=int, help="A list of sectors to scan (beyond those listed in ")
    parser.add_argument("--loop", action="store_true", help="Loop and scan for bad sectors every few seconds")
    parser.add_argument("-a", "--active-scan", action="store_true", help="Actively scan all blocks for errors. Use --offset to start at a specific block.")
    parser.add_argument("-o", "--offset", default=0, type=int, help="For active scan, the block to start at")
    parser.add_argument("-n", default=1000, type=int, help="For active scan, the number of blocks to scan")
    parser.add_argument("device", default="/dev/sda", help="The device to use")
    args = parser.parse_args()
    #Check if the given device is a block device after all
    if not isBlockDevice(args.device):
        print("Error: device argument must be a block device")
    print(("Trying to fix bad sectors on %s" % args.device))
    # Always perform one-shot test
    checkDmesgBadSectors(args.device, set())
    # Fix manually added bad sector list
    fixBadSectors(args.device, args.sector)
    # Active sector scan
    if args.active_scan:
        performActiveSectorScan(args.device, args.offset, args.n)
    # If enabled, loop-check
    if args.loop: loopCheckForBadSectors(args.device)

Update 2015-07-06: Fix loopCheckForBadSectors() (many thanks to Andreas Beier for reporting this bug!)