usb hd crash: causes?

Kropotkin · 11-23-2006, 04:27 AM

Hi all,

An external USB harddisk of mine crashed yesterday while I was browsing some files on it in Gnome. I unmounted it and ran fsck which found and correct a ton of errors. When it was done, I remounted the disk and to my dismay all that was left was a hundred or so 0-byte files in the lost+found directory.

(Yes, I had backups)

In my two-plus years running Linux, I have never seen anything like this.

At this point, I am wondering what caused this failure. Is the drive dying? Is there something inherently unstable about heavy HD use via USB?

The disk is a two-year old 200 GB SATA drive (Maxtor) attached to an Acer laptop running Ubuntu 6.10.

unSpawn · 11-23-2006, 05:03 AM

Would be interesting to see what syslog had to say leading up to the time of crash.

Kropotkin · 11-23-2006, 03:48 PM

Ok, looking in syslog, here is the first sign of trouble (the usb drive was mounted as /dev/sda and I've removed many duplicate lines):

Code:

[...]
Nov 23 02:20:32 zeus kernel: [17972643.764000] usb 5-3: reset high speed USB device using ehci_hcd and address 5
Nov 23 02:21:04 zeus kernel: [17972675.252000] attempt to access beyond end of device
Nov 23 02:21:04 zeus kernel: [17972675.252000] sda1: rw=0, want=18444709592,
[...]

The second line is repeated many times, with different "want" values.
Then I see:

Code:

[...]
Nov 23 02:22:21 zeus kernel: [17972752.036000] Buffer I/O error on device sda1, logical block 2101352623
Nov 23 02:22:21 zeus kernel: [17972752.036000] lost page write due to I/O error on sda1
[...]

And further on:

Code:

[...]
Nov 23 02:22:26 zeus kernel: [17972757.028000] journal_bmap: journal block not found at offset 6681 on sda1
Nov 23 02:22:26 zeus kernel: [17972757.028000] Aborting journal on device sda1.
Nov 23 02:35:17 zeus kernel: [17973527.792000] EXT3-fs warning (device sda1): dx_probe: Unrecognised inode hash code 114
[...]

Then:

Code:

[...]
Nov 23 02:36:20 zeus kernel: [17973591.596000] EXT3-fs error (device sda1): ext3_readdir: bad entry in directory #2: directory entry across blocks - offset=0, inode=808593440, rec_len=12336, name_len=41
[...]

And finally at the end I dismount the drive:

Code:

[...]
Nov 23 02:36:20 zeus kernel: [17973591.596000] ext3_abort called.
Nov 23 02:36:20 zeus kernel: [17973591.596000] EXT3-fs error (device sda1): ext3_journal_start_sb: Detected aborted journal
Nov 23 02:36:20 zeus kernel: [17973591.596000] Remounting filesystem read-only
Nov 23 02:37:18 zeus kernel: [17973648.948000] EXT3-fs error (device sda1): ext3_readdir: bad entry in directory #2: directory entry across blocks - offset=0, inode=808593440, rec_len=12336, name_len=41
Nov 23 02:37:51 zeus kernel: [17973682.208000] __journal_remove_journal_head: freeing b_committed_data
Nov 23 02:37:58 zeus kernel: [17973689.176000] usb 5-3: USB disconnect, address 5
[...]

I have since reformatted the drive, and I am letting it run for a few days with some non-critical data while I decide what to do with it.

unSpawn · 11-23-2006, 05:45 PM

Ok, looking in syslog, here is the first sign of trouble

Code:

attempt to access beyond end of device

In short this line points to a corruption of the disks meta data at that point in time, so the trouble started before this line. Looking around some problems where in lower 2.6 versions (2.6.14 IIRC), some with particular chipsets (ALI?), some with really stressing the whole subsystem (tens of concurrent read and write procs), so it's nearly impossible to outline a specific situation in which it fails. The only things I could recommend is checking out newer kernel versions if possible / available, creating smaller partitions (spread risks?) and doing regular fscks (early warning).

Kropotkin · 11-23-2006, 08:27 PM

Thanks for your reply. I am running a fairly recent kernel. 2.6.17. As for your other suggestions, I'll take them into consideration.

Kropotkin · 11-24-2006, 02:02 PM

Damn, it just happened again.

Either the drive is dying or the external USB enclosure is flaky...