[linux-lvm] Help! PV Size calculation etc
oliver at are-b.org
Fri Mar 4 00:37:43 UTC 2005
So just before going to bed i wanted to take a quick peek at the output
of pvdata /dev/hd[egi]
The good niews is Volume Group part appears to be Ok. Same on all three,
only thing that worried me was 'NOT available/resizable' But since that
is on my other LVM also the case (even with vgchange -an) i belive this
The list of logical volume is idenitcal on all 3 aswell, so that's a
good sign too.
However what I belived to be CRC or checksum data, appears to be PV
information. For instance. The two 'old' PV's are just Physical Volumes.
/dev/hde however is a 'NEW Physical Volume' So I belive this to be one byte.
Then the PV Size. It gives me the full size of the disk. This appears to
be odd aswell, as the other two drives have both some 32mb of NOT usable
data on it. (with LVM beeing 142kb vs 134kb respct.) So I have NO clue
what value would go here on my new PV.
The PV# is 0. This should be 1 I guess, since on the other two it is 2
The Cur LV is 0 whereas the others are 1 and 2. That sounds reasonable,
but would make more sense it would be 1, 2 and 3.
The PE Size (KByte) is 0. The others have an identical value of 32768.
So i could copy that value as it is mostlikly the same.
The total PE size is unknown, but since i have a Total PE of the VG this
number should be calculatable.
The Free PE is going to remain 0 as it is on the others.
Allocated PE is most lkley going to be the same als calculated PE.
But that, really, is for tomorrow. So if someone could give me an idea
how to calculate the PV Size I'd be done I think.
> So I went ahead and did as I said in my previous post. Everything
> looks good. I was missing some header information however. I'm
> guessing the 'HM' parts are the PV information, and the 'HI' parts are
> the VG parts. Eitherway I 'copied' the missing bytes from my other two
> disks. I did this because on my other LVM setup these first few bytes
> where identical on all three disks aswell.
> The data between 0x1a0 and 0x1cf is the last cruelpit however. There
> appears to be some CRC value or checksum calculated. I did not notice
> it at first, as pvcreate did create something there aswell which
> looked very similar. However it is not. Now I checked on my other LVM
> setup, and there also three different data sets. So I'm guessing it's
> a generated checksum or something the like.
> I'm kinda lost here now however. What to do, what to do. Since it's
> late here I'm going to sleep. However I was thinking about the
> following, A) go through the LVM 1.08 source tree and find the
> checksum/crc algorithm and find out over wich area it is performed.
> B) take a look at fix UUID. A contributated tool that would fix UUID's
> in older version. Now I don't really need that I don't think as it
> might mess up things more then it fixes, however it does most likley
> contain the oh so wanted insight on where and what to
> If somebody knows where to look, or better yet, what to do that would
> be greatly appreciated.
> oliver wrote:
>> I couldn't figure out what was written at first onto this disk. The
>> first few bytes made sense, text strings about Missing operation
>> system and such. However 0x1fe and 0x1ff it said '55 AA'. Looking at
>> my 'normal' partitioned hda it made perfect sense however. That's the
>> exact point where all the 'data' e.g. partition table? stops. So
>> unless this raid controller wrote data anywhere else (got an e-mail
>> to highpoint and waiting for answer on that) I can even say that only
>> the first 0x200 addresses (512 bytes DOH) are foobared.
>> So the re-ask and re-phrase my question. Does pvcreate write
>> 'different' data each run in the first 512 bytes. Not counting the
>> UUID and the timestamp. Does vgcreate write anything else in the
>> first 512 besides the volume name on 0xac?
>> I hope some one can answer me that. If not by tomorrow round 3ish, I
>> might get bold and try it anyhow! HA! And then, all I can ask for is
>> lots of luck.
>> (The plan then, for those interested, and future archive readers is to:
>> pvcreate on hde
>> dd the first 512 bytes, edit the UUID and the timestamp (timestamp
>> will match hdg, 'just in case it needs to be older'
>> restore the backup I still have of hde (while I sleep as it takes
>> 2 hours or so with just dd if=/mnt/hd/backupfile of=/dev/hde atleast
>> the other way around did)
>> dd the first 512 bytes from above with the new UUID back onto hde.
>> Party!, I hope.
>> oliver wrote:
>>> WAS: Re: [linux-lvm] Partition table gone? Any way to restore?
>>> After further examination the 'hostname<someserialnumber>' I've
>>> mention that worried me at first appears to be a unix timestamp. I
>>> don't think that it matters much. I've noticed that on my other LVM
>>> setup where I most likley did a pvcreate /dev/hdg /dev/hdi (to
>>> create them simultaniously) since the timestamps are 100% identical.
>>> So I think I can be pretty safe by using the timestamp on my hde
>>> that I have on my hdg. I did create them the same day, I'm sure of
>>> that. (I am using LVM to span across multipledisks after all) The
>>> 3rd disk was added later, hence the different timestamp. So all I
>>> really need to know if some kind of crc/hash is computed after
>>> 0x0001000 so that I can try this.
>>> oliver wrote:
>>>> Small update. Using a hex editor i was able to more closley examine
>>>> my disk(s). Since I have 2 LVM setups I was somewhat able to
>>>> compare what I have to what I should have. I do still need some
>>>> insite however.
>>>> From what I can tell, pvcreate only modifies the first few k's of
>>>> the disk, followed by information stored from the vgcreate.
>>>> Then the lvcreate adds some more information after this. (vg00/lv
>>>> I hope i got it right so far : )
>>>> Now, the vgcreate information is identical on the two surviving
>>>> parts except for a 'hostname<someserialnumber>' part around address
>>>> Then at 0x0001000 we see the UUID of the VG passing by followed by
>>>> some data (which is identical on all three vgs.
>>>> I don't think i have to look beyond this point in space/address.
>>>> Assuming that none of the tools (besides pvcreate) write anything
>>>> before 0x0001000, I should be able to pvcreate /dev/hde edit the
>>>> UUID of that specific PV (to the same stored in the other two PV's
>>>> so the three match up again) and have a fully workable LVM set again.
>>>> The only thing that might worry me is crc information stored
>>>> somewhere after 0x0001000 (However it appears to me that there
>>>> isn't a whole bunch of data stored before 0x0001000 and the data
>>>> created by pvcreate is identical on all 3 drives with the exception
>>>> of the UUID so the crc value should match again) and the string
>>>> 'vg00'at 0x0000ac i see on all disks (the name of the vg.
>>>> What my question is (without spending days reading the sourcecode :
>>>> ) is am I correct? Assuming that all the 'important' metadata and
>>>> such is stored 'after' 0x0001000 hex, is there a good chance of it
>>>> I'm thinking of running a pvcreate, change the UUID of the PV to
>>>> what I expect it to be, add 'vg00' at 0x000ac and be happy?
>>>> I could simply try, I know, but some feedback first would be
>>>> appreciated ; ) I really don't want to loose my data.
>>>> Thanks a lot.
>>>> linux-lvm mailing list
>>>> linux-lvm at redhat.com
>>>> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>>> linux-lvm mailing list
>>> linux-lvm at redhat.com
>>> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>> linux-lvm mailing list
>> linux-lvm at redhat.com
>> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
> linux-lvm mailing list
> linux-lvm at redhat.com
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
More information about the linux-lvm