FYI..... Nate, if the 3ware guys want to work with us, happy to do so.
Hans Nate Amsden wrote: >hi hans. > >im sure your a very busy guy so i will try to keep it brief. I have found >what appears to me to be a critical bug in the interaction between >3ware raid10 IDE arrays and reiserfs 3.5.32. I have already contacted >3ware and their engineers are looking into it(no word yet). I hope >that you may have some idea on how i can debug this further >(I'm not a developer so i am already near my limit of my abilities) > >Systems Tested & Confirmed affected >===================================== >Linux 2.2.19+reiserfs 3.5.32+reiserfs nfsd patch + 3Ware Raid10 arrays >(I have 2 such systems) > >Systems Tested & Confirmed not affected >======================================== >2.2.19+ reiserfs 3.5.32 + reiserfs 3.5.32 nfsd patch + 3Ware Raid5 arrays >(i have 1 system) >2.2.19+ reiserfs 3.5.32 + reiserfs 3.5.32 nfsd patch + 3Ware Raid1 arrays >(i have 1 system) >2.2.19+ reiserfs 3.5.32 + reiserfs 3.5.32 nfsd patch + 3Ware raid10 array >running on an ext2 filesystem rather then a reiserfs filesystem >2.2.19+ reiserfs 3.5.32 + reiserfs 3.5.32 nfsd patch running on a non >3ware raid array(have tested 5 different systems sofar) with the reiserfs >filesystem >2.2.19 + no reiserfs patches running on ext2 filesystem > >all systems are running the same drivers & kernel patches. > >Issue >=========================================== >A symbolic link that points to itself causes either a immediate hard >lock up, or an immediate system reboot. No errors are logged, system >just dies instantly. > >How to reproduce >============================================ >(running bash, all tests were run as user 'root') >cd /path/to/reiserfs/filesystem >ln -s link1 link1 >ls -l lin<HIT TAB FOR FILENAME COMPLETION> >(system lockup or reboot at this point) > >Or, use scp to attempt to copy the symlink(or a directory tree that >contains it), system locks up or reboots when it attempts to access it. > >Exceptions >============================================= >Using 'tar' to archive the looping symlink does not cause a crash >Using 'rsync' to copy the looping symlink does not cause a crash > >Full list of kernel patches: >============================= >Reiserfs 3.5.32 >Reiserfs 3.5.32 nfsd >Openwall v4 (www.openwall.com/linux) >Intel EEPRO 100+ 1.17B Update from www.scyld.com >Latest 3ware raid driver from www.3ware.com >IDE patch from www.linux-ide.org > >Hardware configuration: >(1) Intel P3-800Mhz >768MB Memory >Intel L440GX+ Motherboard >3Ware 6800 Series 8-port IDE Raid card >6 x 80GB Maxtor drives connected to 3Ware card in raid 10 >1 x 30GB Maxtor drive connected to Intel IDE Controller >1 x IDE CDROM connected to Intel IDE Controller >1 x Floppy >2 x Quantum DLT8000 Tape drives(external) hooked to different SCSI >channels on the L440GX+ >Big 4U Rack case with half dozen high RPM fans >Climate controlled enviornment(68 degrees F 24/7), Hooked to >a APC SmartUPS 1400XLNET+Battery pack(2 hours battery backup) > >Software Configuration: >Debian GNU/Linux 2.2r4 >30GB Maxtor drive is partitioned up for / /usr /var /home /boot all running EXT2 >3Ware raid array has a 223GB reiserfs filesystem mounted on /raid >3Ware raid array has a 1GB swap partition > >How i found it >============================ >We use rsync a lot here to back up data to these big 3ware systems mostly >for off-site backups. there is a remote server at a colo that is being backed >up via rsync it has about a dozen of these looping symlinks. the >remote system is running 100% EXT2. I don't know why it has them or who >made them, but they are there. About a week ago i wanted to replicate the >data to another server and used scp from one of the 3ware systems to copy the >data over. the 3ware system rebooted. i ran more tests locally and caused >another 6-7 crashes over a period of a few hours, 100% reproducable. I >eventually >narrowed it down to the 2 systems that run raid10. The systems are relativly >solid otherwise, but doing the above causes an instant crash. I narrowed >it down to reiserfs when i attempted to duplicate the bug on a non >raid array running EXT2. i saw the system spit out "Too many levels of >symbolic links" when it tried to access the file. once i saw that >i was able to narrow it down to reiserfs & looping symlinks on the raid >array. > >all of my linux systems run 2.2 kernels, from early tests i decided >to push back any initial deployments of 2.4 until sometime in late >2002. > >I tried to include as much useful info as i could, but if you need more >info i'd be happy to provide it. > >happy holidays > >nate > > >
