03:14.45 | jsmith | stickster: You awake? |
03:15.00 | plarsen | hi there |
03:15.09 | jsmith | Hey plarsen |
03:15.35 | plarsen | I see nobody posted a topic for tomorrow? |
03:15.40 | plarsen | I won't be coming .... |
03:15.55 | plarsen | so whatever happens, send me a blurb and I'll put it on the site. |
03:16.35 | jsmith | I'm not sure I'll make it either |
03:16.41 | jsmith | Too much other stuff going on :-( |
03:17.13 | stickster | Busy schedule tomorrow, straight from LUG to rehearsal in Herndon |
03:18.40 | plarsen | I'm sure you'll find something to do ;) |
03:19.42 | jsmith | stickster: Quick question if you have a second |
03:19.55 | jsmith | stickster: A friend of mine has two drives in a RAID 1 array |
03:20.06 | jsmith | stickster: But apparently only one of the disks has a master boot record |
03:20.17 | jsmith | stickster: What's the easiest way to see which drive's MBR is correct? |
03:20.23 | plarsen | hardware or software raid |
03:20.24 | plarsen | ? |
03:20.32 | plarsen | I'm allowed to "interfear" ;) |
03:20.33 | jsmith | software |
03:20.48 | jsmith | Yes, you're allow to cause me to fear |
03:21.18 | plarsen | jsmith: Whatever disk he's got active is the rigth one ... if you do an mdstatus you should see the raid is "off" ?? |
03:21.31 | stickster | You can 'dd if=/dev/sdX of=/tmp/mbrcheck count=1' and look at the result... the MBR will have the AA55 word at offset $1FE |
03:21.54 | stickster | It appears on Intel as 55AA in their ordering |
03:22.34 | plarsen | jsmith: Didn't he mirror the whole disk, or just the partition? |
03:22.52 | stickster | If he's booting off it in SW RAID it can't be the whole disk |
03:22.58 | stickster | Well, on most distros it can't. |
03:23.14 | stickster | I hear in 2.6.21 it may be possible to put /boot on a RAID partition |
03:23.52 | plarsen | stickster: hehe - with RAID1 you can easily boot ;) I've run md on whole disks before ..... it's been a while though. |
03:24.01 | plarsen | @ work I do everything hardware raid |
03:24.08 | plarsen | So I don't mess too much with md these days |
03:24.26 | stickster | I've only got some HW RAID and SAN at work |
03:25.04 | stickster | But I was pretty sure my installer didn't allow creation of a mirror for /boot |
03:25.06 | plarsen | I've got an md (software) raid on my home box here |
03:25.12 | plarsen | But it just hums along ;) I don't touch it |
03:25.26 | stickster | You ended up making a /boot on one disk, and even if you mirror the disk the MBR and /boot is only used on one |
03:25.36 | plarsen | and yes, that's a partition based mirror. |
03:25.42 | stickster | Yup |
03:25.48 | plarsen | stickster: right - but it will work ;) |
03:26.01 | plarsen | stickster: so if you loose your primary, you just need to "dd" it back ;) |
03:27.17 | plarsen | jsmith: he would only be booting of one of the drives if the mirror is done on partition level. I don't doubt when Stickster says it can't be done on the full disk. |
03:27.34 | plarsen | jsmith: So why does it matter if the MBR on the secondary drive is bad? |
03:28.07 | stickster | I don't think he said it was bad, just that he wanted to find out which was the MBR being used... how 'bout it jsmith? |
03:28.31 | jsmith | stickster: Yeah, it's not bad... just one of the MBRs works, and the other doesn't. |
03:28.34 | plarsen | stickster: bios disk 0? |
03:28.43 | jsmith | stickster: He just booted the disks individually until he found which MBR worked |
03:28.51 | jsmith | stickster: And then used dd to copy it to the other disk |
03:34.43 | plarsen | hmmm stickster; I still don't see anything in the md man pages that should prevent the full disk (non-partition) approach? I could have SWORN I did that back in the 2.4 days. |
03:34.53 | plarsen | The installer may not allow it ... but it can be done?? |
03:44.31 | stickster | plarsen: I'm not sure how that would work. RAID arrays require a special superblock; how would that resolve with the required MBR for the BIOS? |
03:45.00 | jsmith | G'night... |
03:45.12 | stickster | gnight jsmith |
03:45.23 | plarsen | stickster: same way that there's room for both mbr and partition tables? Different addresses? |
03:45.28 | plarsen | night jsmith |
03:45.44 | stickster | plarsen: Uh, no, the first partition table is in the MBR at sector 0 on the disk |
03:46.15 | plarsen | stickster: nope .. you can wipe mbr without the partition table ... ? |
03:46.25 | plarsen | mbr is just "jump" instructions |
03:46.46 | stickster | plarsen: Right, wiping from $000 -> $1BD |
03:46.53 | stickster | inside sector 0 |
03:47.40 | stickster | $1BE -> $1FD is the first partition table |
03:48.32 | plarsen | That wasn't really my point. MD "super block" can easily be placed elsewhere? |
03:49.44 | plarsen | super-block is at the end of the device. There shouldn't be a conflict. |
03:50.10 | stickster | Yes, I just read that myself |
03:50.38 | stickster | But again, how would the user be prevented from writing an ext3 partition space over that area? |
03:50.57 | plarsen | Of course, if your RAID0/3/5 your disk, your "content" is kinda screwed. Hence the need for a seperate /boot |
03:51.28 | plarsen | prevented? The same way he's prevented from overwriting other parts of the disk? |
03:51.31 | plarsen | Not sure I understand |
03:52.09 | plarsen | What's the difference between preventing a user access from a partition that participates in an md device, and a disk device? |
03:52.30 | stickster | The partition is a type 0xFD, not ext3 |
03:52.51 | plarsen | Right |
03:52.52 | stickster | That way the md driver using the partition can control access to the end of the device, keeping the file system from overwriting it |
03:53.13 | plarsen | ehhhhh - partition types are for "fun" only ... ??? |
03:53.27 | plarsen | Remember, there's gonna be NO partition table on a non RAID1 device. |
03:54.01 | stickster | Not correct, you can assign three or more partitions in RAID 5, for instance. |
03:54.19 | stickster | But let's not cloud the issue |
03:54.34 | stickster | The point right now is, imagine a box with two hard disks. |
03:54.36 | plarsen | riiight - but on the single disk partition, there's no partition table. It's only in the logical assembled RAID5 md device. |
03:54.45 | plarsen | ok |
03:55.30 | stickster | Imagine your scheme where you have the whole disks in a RAID1 mirror. |
03:55.45 | stickster | You are booting off these devices. |
03:55.58 | stickster | At what level does the system know it has a RAID? The md driver |
03:56.48 | stickster | i.e. kernel or later, not in BIOS |
03:57.00 | plarsen | correct - you need a non-raid boot device/partition. I totally agree. But once that's done, the md config etc. is all located in initrd |
03:57.35 | stickster | Well, I'm not concerned about how you boot exactly, but using the whole drive as a RAID is not safe when you're using any kind of programmatic installer. |
03:58.13 | stickster | Because if the superblock is at the end of the disk, and there's no external marker for its existence, the installer would allow users to create partitions whose filesystems overwrote the md superblock. |
03:58.50 | stickster | Thus, the point of RAID partitions as the building blocks |
03:59.07 | stickster | Now, that's not to say you can't make RAIDs from disks, or even boot from disks using RAID |
03:59.56 | plarsen | hmmm - why can't you just check for a valid super-block on device level ?? |
04:00.42 | plarsen | But I see your point. You were looking at this from an installation perspective |
04:01.38 | stickster | that's how most users are going to encounter it. Exceedingly few will purposely go through the pain of doing all the installation steps manually to achieve a completely mirrored system |
04:02.15 | stickster | But I'll have to look at anaconda code to see how they handle this |
04:02.18 | plarsen | well, that's excatly my problem with md .... to 'recover' you actually have to save your partition table seperately |
04:02.33 | plarsen | Because if that goes "poof" your system is down, regardless that you mirrored your system |
04:02.41 | stickster | Yes, but that's a good idea whether RAID or not |
04:03.28 | stickster | You remember the old boot diskettes we used to get :-) |
04:03.54 | plarsen | :) yes, if you're not using hardware raid that is a good idea ;) And the difference is, with hardware raid that physical aspect is mirrored too. |
04:04.17 | plarsen | heheh yeah, still have my old collection here. I think I still have a DOS 3.3 ;) |
04:06.21 | plarsen | so that's my issue for doing a full device RAID. But you're right, it's a chicken and egg issue. If there's no partition table/mbr, the bios won't boot |
04:06.28 | stickster | Yeah, checking the clone option in anaconda -- it clones the partition layout where the partitions are 0xFD |
04:10.30 | plarsen | yeah, it's about bedtime for me too. |
04:10.56 | plarsen | If you guys get me stuff from the meeting tomorrow evening, I should have time to put it up before I go out of town for 5 days. |
04:11.01 | plarsen | Night |
04:23.40 | stickster_afk | I'll see what I can do, but after the meeting I hit the road and may not be back online until nighttime |
13:05.31 | stickster_afk | plarsen_away: Looks like you marked the meeting at 10:00am instead of 9:00 on the website |
22:00.58 | *** join/#fredlug jsmith (n=jsmith@000-170-517.area3.spcsdns.net) |
22:16.45 | *** join/#fredlug IrishW0lf_afk (n=william@70-41-145-17.cust.wildblue.net) |