Broken Corrupted Raspberry Pi SD Card
Published: 09-01-2015 | Author: Remy van Elst | Text only version of this article
Table of Contents
One of my Raspberry Pi's would not boot up after a reboot. The SD card wascorrupted, sadly beyond repair. This article walks you through the steps I tookto try to fix the SD card, including fsck, badblocks and other filesystemutilities. It also has tips to reduce the writing on the Raspberry Pi, this tosave SD cards from some amount of wear and thus possible corruption.
The machine was running as an FM transmitter so that my regular soundsystem could play podcasts (Please note that it is probably illegal to transmitwithout having your HAM license). I't would sync up my feeds and, when turnedon, started playing them.
However, after the last reboot it would not start up again. I rigged up the UARTto my machine and found out why, there were files that the system could notread, therefore init would not boot up.
The file system consistency check utility most of the time is able to find andrecover damaged filesystems. My systems do a filesystem check at every boot, andthey get a reboot regularly. The SD card has an ext4 filesystem and is 16 GB insize.
Since most of the time you can fix broken filesystems with an fsck I hooked upthe card to my machine and started the fsck:
# fsck /dev/mmcblk0p2
And it failed right away:
fsck from util-linux 2.25.2e2fsck 1.42.12 (29-Aug-2014)fsck.ext4: Filesystem revision too high while trying to open /dev/mmcblk0p2The filesystem revision is apparently too high for this version of e2fsck.(Or the filesystem superblock is corrupt)The superblock could not be read or does not describe a valid ext2/ext3/ext4filesystem. If the device is valid and it really contains an ext2/ext3/ext4filesystem (and not swap or ufs or something else), then the superblockis corrupt, and you might try running e2fsck with an alternate superblock: e2fsck -b 8193 <device> or e2fsck -b 32768 <device>
The superblock contains information about the file system such as the filesystem type, size, status, information about other metadata structures, blockcounts, inode counts, supported features, maintenance information, and more andso on (source). It is an important piece of information for the filesystem.
So the card is so broken that the first superblock is not readable. An ext4filesystem has multiple copies of it's superblock, so lets find them and usethem to try the fsck again. To find the other superblocks we can use thefollowing command:
mke2fs -n /dev/mmcblk0p2
mke2fs 1.42.12 (29-Aug-2014)/dev/mmcblk0p2 contains a ext4 file systemProceed anyway? (y,n) yCreating filesystem with 3794688 4k blocks and 950272 inodesFilesystem UUID: a56c8a06-9907-41a2-86a0-dd601212880bSuperblock backups stored on blocks: 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208
We also need to find the block size, the above command shows it as 4k, which is4096. With this information we can retry the fsck with the alternatesuperblock:
fsck -b 163840 -B 4096 /dev/mmcblk0p2
-b is the alternative superblock and
-B is the blocksize.
It didn't help much, after a lot of questions I decided to automatically fixeverything it found:
fsck from util-linux 2.25.2e2fsck 1.42.12 (29-Aug-2014)Superblock has an invalid journal (inode 8).Clear<y>? yes*** ext3 journal has been deleted - filesystem is now ext2 only ***Resize inode not valid. Recreate<y>? yesPass 1: Checking inodes, blocks, and sizesyInode 1 has EXTENTS_FL flag set on filesystem without extents support.Clear<y>? yesRoot inode has dtime set (probably due to old mke2fs). Fix<y>? yesQuota inode is not in use, but contains data. Clear<y>? yesQuota inode is not in use, but contains data. Clear<y>? yesInode 5, i_size is 2305843009213693952, should be 0. Fix<y>? yesInode 5, i_blocks is 131072, should be 0. Fix<y>? yesReserved inode 6 (<The undelete directory inode>) has invalid mode. Clear<y>? yesInode 6 has a bad extended attribute block 2064. Clear<y>? yesInode 6, i_size is 666532745924706320, should be 0. Fix<y>? yesJournal inode is not in use, but contains data. Clear<y>? yesReserved inode 9 (<Reserved inode 9>) has invalid mode. Clear<y>? yesReserved inode 10 (<Reserved inode 10>) has invalid mode. Clear<y>?Recreate journal<y>? cancelled!/dev/mmcblk0p2: e2fsck canceled./dev/mmcblk0p2: ***** FILE SYSTEM WAS MODIFIED *****
-y option to the command will automatically answer yes toeverything:
fsck -y -b 163840 -B 4096 /dev/mmcblk0p2
This goes well for a while, but fails after some time with the following error:
Error storing directory block information (inode=5542, block=0, num=33754683): Memory allocation failed/dev/mmcblk0p2: ***** FILE SYSTEM WAS MODIFIED *****Recreate journal? yesCreating journal (32768 blocks): Done.*** journal has been re-created - filesystem is now ext3 again ***e2fsck: aborted
Searching around on the web results in multiple topics suggesting to add thefollowing:
[scratch_files]directory = /var/cache/e2fsck
To the file
/etc/e2fsck.conf. The man page describes it like so:
[scratch_files] This stanza controls when e2fsck will attempt to use scratch files to reduce the need for memory.
Ted Tso explains what this option does on this mailinglist topic:
This will cause e2fsck to store certain data structures which grow large withbackup servers that have a vast number of hard-linked files in /var/cache/e2fsckinstead of in memory. This will slow down e2fsck by approximately 25%, but forlarge filesystems where you couldn't otherwise get e2fsck to complete becauseyou're exhausting the 2GB VM per-process limitation for 32-bit systems, itshould allow you to run through to completion.
You have to create the
/var/cache/e2fsck folder if it does not exist yetbefore running the fsck:
mkdir -p /var/cache/e2fsck
scratch_files stanza has more options, two of which might be important,
set dirinfo and
set dirinfo to false if your filesystem had an large number ofindividual files and not that many directories. You configure
set icount tofalse if the reverse were true. Otherwise, you configure them both on true.
There was also a good explanation of where the Out of Memory error comes from onStack Exchange, it might not be out of memory.
scratch_files setting did not improve the check. It still gave the sameerror, out of memory.
Even after playing with the dirinfo settings, or adding a 200 GB swapfile thefsck would still fail.
dmesg was also full with these kinds of logs:
[12343.5678] end_request: I/O error, dev mmcblk0, sector 50944
My last guess would be to try Spinrite. Spinrite is a hard drive recoveryand maintenande utility written by Steve Gibson from Gibson ResearchCorporation. I configured a KVM VM with the card reader device as a secondblock device and tried to run Spinrite on Level 2. However, even spinritefailed:
Booting Spinrite on a spare laptop with a cardreader with this SD card gave thesame error. It sees the device, but fails to run on it.
Spinrite has helped me multiple times recover drives, but this one might be tomuch for it.
Badblocks is a *NIX utility to search for bad blocks on a device. While it isnot usefull for creating a list of bad blocks on the SD card because SD cards donot report actual physical addresses (because of wear levelling) it does tell usif the card is broken or not.
The following badblocks command will scan and report bad blocks for the device.It is a destructive write operation, you will lose your data.
badblocks -o ./badblocks.list -w -s -v -b 4096 -c 16 /dev/mmcblk0
-o to output the badblocks list to the file
-w for thewrite operation,
-s to show progress,
-v to be verbose,
-b 4096 for theblocksize of 4K and
-c 16 to test 16 blocks at once (default is 64).
It showed me that a lot of writes were failing:
Checking for bad blocks in read-write modeFrom block 0 to 3799039Testing with pattern 0xaa: 0.01% done, 1:57 elapsed. (0/292/0 errors)^CInterrupted at block 294
The error output format means the following:
number of read errors/number of write errors/number of corruption errors
So most of the writes (292 of 294) failed, meaning the SD card is broken, beyondrepair I guess.
This SD card is a lost case. To prevent your Raspberry Pi's from writing a lotof data, and thus, wearing the SD card, you can do a couple of things.
The first one is to mount a few folders in RAM as
tmpfs. The folders are thefolders where temp files and logging is written to. This means that you won'thave syslog available, but most of the time that is not a problem.
/etc/fstab and add the following:
none /var/run tmpfs size=1M,noatime 00none /var/log tmpfs size=1M,noatime 00none /var/tmp tmpfs size=1M,noatime 00none /tmp tmpfs size=1M,noatime 00
This will mount the above folders in RAM, with a max size of 1 megabyte. The
noatime option means that the access time of a file is not updated, saving alot of writes as well. You should also add the
noatime option to your otherpartitions, for example on a standard Raspbian:
proc /proc proc defaults 0 0/dev/mmcblk0p1 /boot vfat ro,noatime 0 2/dev/mmcblk0p2 / ext4 defaults,noatime 0 1
/boot partition is also mounted read only (
noatimeoption is added.
mount -a command or reboot the machine to make this active.
Linux divides its physical RAM (random access memory) into chucks of memorycalled pages. Swapping is the process whereby a page of memory is copied to thepreconfigured space on the hard disk, called swap space, to free up that page ofmemory. The combined sizes of the physical memory and the swap space is theamount of virtual memory available.
Swappig causes a lot of writes to the SD card. You would want to turn it off tosave writes. The downside of this is that when there is not enough RAM availablethe linux OOM killer will randomly kill processes to save RAM.
Raspbian by default has a swap file, dynamically managed by the
dphsys-swapfile utility. You can turn off this utility by issueing the followingcommands:
dphys-swapfile swapoffdphys-swapfile uninstallupdate-rc.d dphys-swapfile remove
After a reboot the swap will be gone, which you can check with the
total used free shared buffers cachedMem: 484 243 241 0 42 162-/+ buffers/cache: 38 446Swap: 0 0 0
fsck at every boot
My Raspberry Pi's have a cronjob which reboots them once every seven days. Thisto apply kernel updates and just a general good procedure to see if all stillworks after a reboot. By default, fsck checks a filesystem every 30 boots(counted individually for each partition). I decided to change this to everyboot, so problems will be found and possibly fixed earlier.
To set up an fsck at every boot, execute the following command:
tune2fs -c 1 /dev/sda1
/dev/sda1 is the device or partition.
Remove the GUI
I don't use the Raspbian GUI so I decided to remove everything GUI related. Thebest way to do this is to remove
libx11 and all it's dependencies, that beingall the GUI applications:
apt-get purge --auto-remove 'libx11-.*'
This lowers the amount of disk space used, the amount of packages updated andthe amount of services running on the machine, saving both RAM and disk space.
This SD card was so corrupt I threw it away. The Raspberry Pi was reinstalledand the
pifm software was set up again. The music came from another machine,so that did not have to be copied again.
With the tips above you might be able to save your SD card if it ever getscorrupted. These tips also work on regular disks and SSD's.Tags: badblocks, blog, ext4, filesystem, fsck, mkfs, raspberry-pi, sd, swap