Hanging ext3 or USB, linux 2.6.16-rc6-mm2
I managed to hang ext3 or usb today.
I have a small machine that boot off a compactflash card. I want to use a bigger card, so I used scp to copy everything from that machine to a new 4GB card in a usb cardreader. This cardreader have never given me trouble before, but is usually used for reading.
I decided on a ext3 fs in order to avoid long fsck runs, and a minimal 4MB journal in order to not waste space. Disk seeks are supposed to be really cheap on a device with no moving parts anyway. The root reserved percentage is 1% instead of the usual 5% - more space, and fragmentation will probably not hurt much with cheap seeks.
When scp had filled the card to 71% of capacity (according to df), it stopped in the middle of a file. I first suspected network errors, but a "ls /mnt" hung.
I now have the following processes in D-state:
[khubd] [scsi_eh_4] [usb-storage] [kjournald] scp, 3 x [pdflush], 2 x ls, lsusb, and a sync.
Could this be a ext3 problem due to the small journal or something?
Or is a usb problem more likely? "Dmesg" shows an usb disconnect sometime after I mounted that filesystem, but it seems to be usblp0 which looks like the printer to me.
Helge Hafting
Quoted 22 lines
I managed to hang ext3 or usb today.
I have a small machine that boot off a compactflash card. I want to use a bigger card, so I used scp to copy everything from that machine to a new 4GB card in a usb cardreader. This cardreader have never given me trouble before, but is usually used for reading.I decided on a ext3 fs in order to avoid long fsck runs, and a minimal 4MB journal in order to not waste space. Disk seeks are supposed to be really cheap on a device with no moving parts anyway. The root reserved percentage is 1% instead of the usual 5% - more space, and fragmentation will probably not hurt much with cheap seeks.
When scp had filled the card to 71% of capacity (according to df), it stopped in the middle of a file. I first suspected network errors, but a "ls /mnt" hung.
I now have the following processes in D-state:
[khubd] [scsi_eh_4] [usb-storage] [kjournald] scp, 3 x [pdflush], 2 x ls, lsusb, and a sync.
Quoted 5 lines
Could this be a ext3 problem due to the small journal or something?
Or is a usb problem more likely? "Dmesg" shows an usb disconnect sometime after I mounted that filesystem, but it seems to be usblp0 which looks like the printer to me.
I'd guess it is some USB/block layer problem. If just ext3 hung, then
you would not see [usb-storage] and similar hung. I would need to see where each process hung to tell more..
Honza
Signature · 2 lines
Jan Kara <[email protected]> SuSE CR Labs
Quoted 11 lines · Jan Kara
Quoted 7 lines
Could this be a ext3 problem due to the small journal or something?
Or is a usb problem more likely? "Dmesg" shows an usb disconnect sometime after I mounted that filesystem, but it seems to be usblp0 which looks like the printer to me.
I'd guess it is some USB/block layer problem. If just ext3 hung, then
you would not see [usb-storage] and similar hung. I would need to see where each process hung to tell more..
You are right, it is block layer for I have seen problems with vfat formatted compactflash too now. On my home machine, the problems are intermittent. The hang can happen after writing almost one GB, or it can happen early, in "mount"
I also have a via epia board that consistently fails to use its compactflash reader, but that one is cardbus, not usb.
Helge Hafting