Ny way to prevent a bad drive from disappearing from /dev

badblocksblock-devicedevicespartitionudev

I'm trying to recover a partition on a bad drive with ddrescue. I run:

$ sudo ddrescue -r -1 -v /dev/sdd3 OUT.img dd_rescue_logfile

and it seems to do great for a while, but after about an hour, the "current rate" drops to zero because the drive disappeared from /dev. To bring the drive back, the only thing I can think of is to reboot the system and run the ddrescue command to resume from where I left off. This makes it very difficult to run the program, as I can't just leave it and forget it for a few days – I have to constantly monitor it to make sure the disk didn't disappear. I have seen this behavior on both Arch linux, and Fedora 22.

I assume that at some point, the kernel has trouble accessing the drive and removes it from /dev. Is there any way to avoid this? To tell the kernel to keep the device there even if it looks like it's broken or non-existent?

Best Answer

You might be able to re-detect it without reboot, by unloading / re-loading the correct module (or just un-binding and re-binding the driver).

For example:

[  978.527221] sd 11:0:0:1: [sdk] Attached SCSI removable disk
#~> echo 11:0:0:1 > /sys/bus/scsi/drivers/sd/unbind
#~> echo 11:0:0:1 > /sys/bus/scsi/drivers/sd/bind
[ 5572.027119] sd 11:0:0:1: [sdk] Attached SCSI removable disk

Or if that does not work and you have no other devices hooked to the same controller, you could unbind and bind the entire controller for example via /sys/bus/pci/drivers/ahci/ if it's AHCI.

I don't have any defective HDD to test if it would actually work, but I used this method before to force re-detection of MicroSD/MMC cards in slots that were not hot-pluggable by default.

As for ddrescue dropping speed to zero, you might want to see if it supports the -a, --min-read-rate=<bytes> option so it would hopefully? consider slow regions as defective and skip them. Worst case you'll have to monitor the disk from the outside and forcibly restart ddrescue.

Related Solutions

Linux – Detecting that a device is mounted to a particular folder

You may run your backup script like that in your cron

fgrep -q /mnt/backup /proc/mounts && backup.sh

But better add the test in your script, to log the failed attempts

Uniform way to discover information about devices

There is no single standard or tool to query hardware devices on Linux systems in general. Depending on your host's architecture, and which of its components you must query, and how much detail you need about it, you may need one or more tools specific to that component. However some commands/tools are in wider use and have greater mindshare than others. Following are some--that may or may not be available for your particular host--but are nonetheless Generally Regarded As Useful and widely available from major package managers (though I only link to Debian below):

all-purpose query tools:

hwinfo. hwinfo --short - gives a useful overview of The Whole Enchilada, and info hwinfo shows many other options for querying specific subsystems in detail.
inxi is part of a larger "collection of system administration tools" with similarly general capabilities.

tools for specific subsystems:

dmidecode - processor, memory and motherboard details from BIOS
lscpu - processor details from /proc/cpuinfo
lspci - PCI devices, typically graphics cards, audio cards, network cards
lsusb - USB devices in general
ls -l /dev/disk/by-{id,label}/ - block devices and their block device files
ls -l /sys/class/net/ - network devices and their network interfaces
udevinfo - block devices, if using udev

I encourage people to expand this list if some essential tool is missing.

Best Answer

Related Solutions

Linux – Detecting that a device is mounted to a particular folder

Uniform way to discover information about devices

Related Question