we have rh 4.2 5.0 and 5.1 running various versions of raid code. the 1005 was the easiest by far. sorry if this seems rudimentary, but there are lurkers here who can use all of this. Ultra-quick start Raid (too quick) Howto: 1. rm -rf /usr/src/linux cause we want clean sources. redhat rpms are not clean, as they contain patches already. 2. cd /usr/src; tar xzf linux-2.0.35----.tgz get the .35 kern src at sunsite.unc.edu etc. 3. gunzip raid0145-19981110-2.0-----.gz get latest nov. release (i think this finally compiles) 4. patch -p0 < raid0145-1998xxx 5. cd linux; make menuconfig put in the stuff you need. raid non-module for autostart. 6. make zImage or bzImage if you need :) 7. make modules; make modules_install 8. tar xzf raidtools.tgz (get this from linux.kernel.org/pub/linux/daemons/raid/alpha too) 9. configure and make raidtools. (must have patched kernel and done a make first) 10. make install of raidtools. 11. setup new kernel as label linux in your /etc/lilo.conf make old kernel "linux.old" (we do this cause redhats rc scripts use lilo to determine which modules dir to depmod -a, and when we boot from floppy we will call linux, so there must be a "linux" in the mbr with exactly the same ) 12. run lilo to write the bootloader's map, etc. 13. use fdisk to change partition type on each raid piece to fd 14. reboot with new kernel. (we have not made raid disks yet, so kernel will complain about bad superblocks, if so, we are almost there. one time complaining is good :-) 15. setup /etc/raidtab (there is example with raidtools) 16. run mkraid /dev/md0 thru mdx. (i do them individually rather than with -a so i can watch outputs.) 17. raidstart /dev/md0 thru mdx 18. mke2fs (or other fs) /dev/md0 thru mdx 19. edit /etc/fstab to reflect mount points (make sure mount points actually exist) 20. mount /raid1 or what ever. (if this not work error messages are useful) 21. reboot. 22. all raid1 and up should restart. raid 0 and linear require persistent superblocks (/etc/raidtab setting during mkraid) 23. if you want to make root raid, edit /etc/fstab to say that your root partition is /dev/mdx then copy info over to raid partition (see harddrive upgrade howto) be careful here and dont copy /raid to /raid or you will enter a loop that fills the drive. also dont copy /proc, just make the dir in /raid/ 24. mkbootdisk (this creates boot floppy with lilo this is why we need to install new kernel as "linux" in lilo, cause of redhat boot scripts, and why we edited the fstab, cause that is how mkbootdisk knows what to drev the root device to.) 25. put / back to your current root in /etc/fstab (this gives us a failover IMPORTANT!) 26. boot from floppy. (raid root should start, if not, take out floppy and reboot with linux.old from the HD) 27. you still have to boot from floppy, though there are a couple alternatives. 27.1 patch lilo to understand your root raid1 device. i have not done this but seen it. 27.2 still use floppy. i do this, not a big deal. 27.3 use a boot rom on network card to get kernel into ram from another machine. (seems silly, point of raid is fault tolerance, but we have to boot from ANOTHER computer!) 27.4 the current solution i am working on is a bootrom on a homemade card that hooks int18h (like rombasic used to do) and loads the kernel off of a secondary flashrom into ram high (like bzImages) need the secondary rom cause the largest bios-scanned space is 128k, and i cant make a 128 k kernel :-) hope this helps those of you using redhat. allan noah sysad pfeiffer univ. www.pfeiffer.edu
