Arch Linux

FrozenFox

Member

From: College Station, TX

Registered: 2008-03-23

Posts: 422

Website

Bizarre ext4 issue resolved; plausible explanation?

Hello everyone,

(Note: Not sure where to put this topic, but since EXT4 is part of the kernel, it could be appropriate)

I recently copied my Music folder to a Win7 partition on a new hd from my Arch EXT4. When I tried to sync itunes with the Music folder, I noticed the number of files there vs the number on the ipod/itunes didn't match. Long story short, after much sleuthing, I tracked it down to a few folders which somehow had a 0 byte size and wouldn't show any info/security/permissions in "properties", were impossible to delete (windows said the folder doesn't exist o.O) or rename, but whose mp3s inside worked perfectly and were move-able to another folder (but not back to the original, as it would just make a duplicate empty folder and who knows wtf it did with the files). These folders appeared/worked fine from within mounted ntfs in konqueror. No weird stuff shown in ls -al.

Weird eh?

I went back and ran chkdsk /f on windows' ntfs and fsck -fa (and fsck -c for badblocks) on arch's ext4 among other things. Playing around some, I found that just copying the individual folders in question still produced the same trouble, from Konqueror/Nautilus/Cli with any user level.

I found that trying to copy via konqueror to a vfat (my ipod) filesystem, konqueror/vfat would *TELL ME* there's something wrong and say it just couldn't copy/create the folder, so I tracked down a couple others this way. All other files copied worked fine. The permissions/ownership/group on all of these Music folders on Ext4 were identical according to ls -al.

Here's the EVEN MORE BIZARRE part, which I reproduced about 4 times:

If I edit the permissions from konqueror for the folders in question on ext4, and check "Write entries" for "Group" and "Others" (all but the SUID bits set) and press okay, then open it up again and UN-DO the checks I JUST made so that it was back to how it was originally, the folders no longer give any trouble. Again according to ls -al, nothing ever changed about these folders, but doing this reproducibly worked.

Anybody have a plausible guess as to wtf happened? It's obviously not windows' fault this time, as these things wouldn't copy over into vfat *before* even getting to windows. Ntfs just took it without trouble, even though it was broken.
As I understand it, EXT4 corruption is only reported to occur on files that were opened/accessed recently during improper system shutdown -- is that correct?. Half of the folders that this problem affected have 100% absolutely not been opened prior to this for maybe a year by myself, and I do not have indexing enabled. The other half had been used semi-recently.

Thanks guys. Sorry for the tldr, but it's an interesting problem and I shortened it the best I could while still giving pertinent info. I'd like to hear comments on what may've happened mostly out of curiosity and wanting to prevent it later.

Last edited by FrozenFox (2009-11-01 20:30:45)

TomB17

Member

Registered: 2009-09-02

Posts: 102

Re: Bizarre ext4 issue resolved; plausible explanation?

I can offer no assistance, FrozenFox.  The only reason I respond is because the reason I dumped EXT4 - 6 months ago was because of the 0 byte file issue.

I understand was a known bug that has been fixed but I'm not interested in subjecting myself to the EXT4 experience again any time soon but I look forward to the day it becomes stable because the chkdsk processing times on my two RAID arrays is all but killing me.  It's over 12 hours now.  I understand EXT4 could all but make the issue go away.

thetrivialstuff

Member

Registered: 2006-05-10

Posts: 191

Re: Bizarre ext4 issue resolved; plausible explanation?

Unicode characters in filenames? E.g. maybe a hyphen was actually unicode "optional soft hyphen" or some such thing. I've had similar behaviour copying MP3's between Windows and Linux when the ones on the Windows side had unicode chars and/or the ones on the Linux side had windows no-no characters (colon, questionmark, etc.).