Red Hat 8.0: the saga continues...

Tobiasly · 10-10-2002, 09:57 AM

OK, I'm starting to get honked off at this thing. This upgrade that was supposed to take a few hours to get at least a blank but workable state is now taking days. Not that I didn't expect that to happen...

My first problem came with my RAID drives. I have 3 SCSI drives, partitioned identically. The / and /boot partitions are RAID 1 (2 + 1 spare), and most everything else is RAID 5. So when I went into 8.0 setup, I told it to keep the partitions, but reformat them and start over (I never ever simply "upgrade" an OS).

Well, at some point a while back, my /dev/sda had a brain fart and the RAID switched it off like it should have (I think the cable was loose and got bumped or something). I thought I added it back in though, but apparently not on the /boot partition -- it was still using just /dev/sdb and sdc. So apparently Disk Druid, when installing 8.0, noticed that sda wasn't being used in the corresponding "md" device, and left it that way. I told it to use all three drives, but I think it decided /dev/sda should be the spare insead of sdc.

Not that big a problem, except that it loaded the boot loader onto the /dev/sda device (the actual BIOS boot device that corresponded to the md /boot mount point), but wrote the GRUB files to the "md" device, which had /dev/sda removed! In a RAID 1 array, every drive gets the same image, but /dev/sda didn't get the image the installer thought it would, since it was removed.

So, after installing, I reboot the system and see all my old GRUB configs! It took me a good several hours to figure that one out. I mounted each of the RAID 1 member drives separately, figured out what was going on, and attempted to hot-add sda back in. But /proc/mdstat didn't show it being resynced, and the new files weren't copied over, so I simply copied them over myself. Mistake -- I guess the new GRUB files weren't on the same sector, because then the system wouldn't even boot.

So, I bit the bullet, quit being a GUI wuss, used fdisk to delete those fucking partitions and set 'em up the right way, and drove on. I don't like how Disk Druid wants to move your partitions around and rename them and shit anyway. I have those partitions in a particular physical order for a reason, thank you very much.

Now, the next prob. I have two ethernet cards -- eth0 connects to my home LAN, and eth1 connects to my cable modem. Forget that multi-homed shit. eth1 is getting configured through DHCP just fine, but I can't get to anything external -- no web, no pings, nothing. When I try to ping the DNS servers I'm assigned, it times out. The problem is with this box, 'cause it works fine when I hook the cable modem up to another box (I get the same DNS servers and everything). I can't resolve any hostnames either -- even if I specify which DNS server to use with "dig". resolv.conf looks fine.

So I read the ping man page, and saw a "-I" parameter where you can specify which device you want to ping <I>from</I>. So I did a "ping -I /dev/eth1 xxx.yyy..." and it worked. So it appears that the problem is, that the box doesn't know that it should use eth1 for external stuff and eth0 for internal. I don't know how it knew that before, but it used to work. I have a sneaking suspicion it has something to do with the fact that Red Hat replaced the previous "pump" DHCP client with "dhclient", and it's not setting some parameter correctly.

Any ideas on this one?

10-10-2002, 10:07 AM

man route

Tobiasly · 10-10-2002, 10:17 AM

That has to set the record for the highest Cellar question-to-answer word count ratio.

Thanks, I'll check it out when I get home.

juju · 10-10-2002, 10:29 AM

Yeah, it sounds like you need to set up your routing table. That tells Linux what ethernet card to use for what IP range. I don't really understand how it all works myself, or i'd give a more detailed explanation. I just trudge through the Linux Networking HOWTO whenever I need to configure it. Or, more specifically, here

Are there any experts on route out there that can give a more specific explanation?

juju · 10-10-2002, 10:31 AM

I think it's funny how linux users think that the man pages are this immense bastion of helpful knowledge. When I was a linux newbie (and i'm not saying Tobiasly is by any stretch), I could understand a man page about as much as I could understand a page written in Chinese.

10-10-2002, 11:36 AM

Different strokes for different folks. Man pages have been exceptionally useful to me throughout my UNIX career. I had the exact same problem Tobiasly is describing, and it was a route problem - that man route fixed.

Tobiasly · 10-10-2002, 01:23 PM

Man pages just take some getting used to -- if they're set up right, like most thankfully are these days, they can be pretty helpful. If you don't know exactly what you're looking for, but know about something that's somewhat related, you can usually go to the bottom and check out the "SEE ALSO" and "FILES" sections, which often is what I need.

I'm dismayed by the apparent trend of man pages being left out-of-date in favor of info pages. Info sucks, ascii-based Emacs sucks, and I can never remember the keystrokes to get back where I was.

Looking at the "route" page jogged my memory -- I believe I might have un-checked the "routed" package during install, which would explain why everything wasn't "just working" like it was. I try to maintain some semblance of security by removing all these daemons I never use or that are potentially dangerous (i.e. finger, telnet, rsh) and may have gotten a bit trigger-happy!

10-10-2002, 09:57 AM	#1
Tobiasly hot Join Date: Mar 2002 Location: Jeffersonville, IN (near Louisville) Posts: 892	Red Hat 8.0: the saga continues... OK, I'm starting to get honked off at this thing. This upgrade that was supposed to take a few hours to get at least a blank but workable state is now taking days. Not that I didn't expect that to happen... My first problem came with my RAID drives. I have 3 SCSI drives, partitioned identically. The / and /boot partitions are RAID 1 (2 + 1 spare), and most everything else is RAID 5. So when I went into 8.0 setup, I told it to keep the partitions, but reformat them and start over (I never ever simply "upgrade" an OS). Well, at some point a while back, my /dev/sda had a brain fart and the RAID switched it off like it should have (I think the cable was loose and got bumped or something). I thought I added it back in though, but apparently not on the /boot partition -- it was still using just /dev/sdb and sdc. So apparently Disk Druid, when installing 8.0, noticed that sda wasn't being used in the corresponding "md" device, and left it that way. I told it to use all three drives, but I think it decided /dev/sda should be the spare insead of sdc. Not that big a problem, except that it loaded the boot loader onto the /dev/sda device (the actual BIOS boot device that corresponded to the md /boot mount point), but wrote the GRUB files to the "md" device, which had /dev/sda removed! In a RAID 1 array, every drive gets the same image, but /dev/sda didn't get the image the installer thought it would, since it was removed. So, after installing, I reboot the system and see all my old GRUB configs! It took me a good several hours to figure that one out. I mounted each of the RAID 1 member drives separately, figured out what was going on, and attempted to hot-add sda back in. But /proc/mdstat didn't show it being resynced, and the new files weren't copied over, so I simply copied them over myself. Mistake -- I guess the new GRUB files weren't on the same sector, because then the system wouldn't even boot. So, I bit the bullet, quit being a GUI wuss, used fdisk to delete those fucking partitions and set 'em up the right way, and drove on. I don't like how Disk Druid wants to move your partitions around and rename them and shit anyway. I have those partitions in a particular physical order for a reason, thank you very much. Now, the next prob. I have two ethernet cards -- eth0 connects to my home LAN, and eth1 connects to my cable modem. Forget that multi-homed shit. eth1 is getting configured through DHCP just fine, but I can't get to anything external -- no web, no pings, nothing. When I try to ping the DNS servers I'm assigned, it times out. The problem is with this box, 'cause it works fine when I hook the cable modem up to another box (I get the same DNS servers and everything). I can't resolve any hostnames either -- even if I specify which DNS server to use with "dig". resolv.conf looks fine. So I read the ping man page, and saw a "-I" parameter where you can specify which device you want to ping <I>from</I>. So I did a "ping -I /dev/eth1 xxx.yyy..." and it worked. So it appears that the problem is, that the box doesn't know that it should use eth1 for external stuff and eth0 for internal. I don't know how it knew that before, but it used to work. I have a sneaking suspicion it has something to do with the fact that Red Hat replaced the previous "pump" DHCP client with "dhclient", and it's not setting some parameter correctly. Any ideas on this one?

10-10-2002, 10:17 AM	#3
Tobiasly hot Join Date: Mar 2002 Location: Jeffersonville, IN (near Louisville) Posts: 892	That has to set the record for the highest Cellar question-to-answer word count ratio. Thanks, I'll check it out when I get home.

10-10-2002, 10:29 AM	#4
juju no one of consequence Join Date: Jun 2001 Location: Arkansas Posts: 2,839	Yeah, it sounds like you need to set up your routing table. That tells Linux what ethernet card to use for what IP range. I don't really understand how it all works myself, or i'd give a more detailed explanation. I just trudge through the Linux Networking HOWTO whenever I need to configure it. Or, more specifically, here Are there any experts on route out there that can give a more specific explanation?

10-10-2002, 10:31 AM	#5
juju no one of consequence Join Date: Jun 2001 Location: Arkansas Posts: 2,839	I think it's funny how linux users think that the man pages are this immense bastion of helpful knowledge. When I was a linux newbie (and i'm not saying Tobiasly is by any stretch), I could understand a man page about as much as I could understand a page written in Chinese.

10-10-2002, 11:36 AM	#6
dave Guest Posts: n/a	Different strokes for different folks. Man pages have been exceptionally useful to me throughout my UNIX career. I had the exact same problem Tobiasly is describing, and it was a route problem - that man route fixed.

10-10-2002, 01:23 PM	#7
Tobiasly hot Join Date: Mar 2002 Location: Jeffersonville, IN (near Louisville) Posts: 892	Man pages just take some getting used to -- if they're set up right, like most thankfully are these days, they can be pretty helpful. If you don't know exactly what you're looking for, but know about something that's somewhat related, you can usually go to the bottom and check out the "SEE ALSO" and "FILES" sections, which often is what I need. I'm dismayed by the apparent trend of man pages being left out-of-date in favor of info pages. Info sucks, ascii-based Emacs sucks, and I can never remember the keystrokes to get back where I was. Looking at the "route" page jogged my memory -- I believe I might have un-checked the "routed" package during install, which would explain why everything wasn't "just working" like it was. I try to maintain some semblance of security by removing all these daemons I never use or that are potentially dangerous (i.e. finger, telnet, rsh) and may have gotten a bit trigger-happy!

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)