trying to restore data after Harddisk breakdown

I had a lvm System which was made of 3 PVs the Harddisks hde (30 GB),hdf (80 GB) and hdg (120 GB). The first harddisk on which I created the vg was hdf, then I created 3 LVs on it. Moved the existing data from hde the VG, added hde as a PV and expanded one of the LVs. After some months I added hdg and resized all LVs. Some time later I moved a bit of free Space between the LVs but forgot to backup my configuration. Not very much longer one of the harddisks hde had a complete breakdown, it couldn't be found by the bios anymore. Since I still had waranty on it I sent it in and got some time later a replacement disk.
In the meantime I was able to extract the data from the 2 LVs which didnt span on the defect disk by using LVM2 and using
""vgscan ; vgchange -a y -P"".

Im Using now kernel 2.4.20 and device-mapper + lvm checked out from cvs yesterday night. (Compilation was not that easy, i had to make some changes for device-mapper lib/ioctl/libdevmapper.c,v
479c479
< char *outbuf = (char *) dmi + dmi->data_offset;
---

536c536
< dmt->dmi.v3->data_offset);
---

671c671
< dmi->data_offset = sizeof(struct dm_ioctl);
---

which i hope is correct since the struct dm_ioctrl had no member data_offst and reading the comments for this struct suggested that data_start should be an offset

and in lvm2 I needed to include the kernel Include files into the INCLUDE part of the MAKEFILE)

Now I have the replacement disk which should be only one large segment in one of the LVs. So I thought that it could maybe possible to get the lvm up again by exracting the actual configuration that is used by "vgscan ; vgchange -a y -P" and making a vgcfgbackup of it with lvm2 and edit the backup since it is humanreadable and restore it to the new harddisk.

so I tried to make a vgcfgbackup which worked with the -P option and got somthing which seemed a bit irritating for me:

--------------------------------------------------------
# Generated by LVM2: Sun Apr 27 09:26:39 2003

contents = "Text Format Volume Group"
version = 1

description = "Created *after* executing './vgcfgbackup -P'"

creation_host = "fuckup" # Linux fuckup 2.4.20 #1 SMP Sun Apr 27 00:32:28 CEST 2003 i686
creation_time = 1051428399 # Sun Apr 27 09:26:39 2003

Data {
id = "PUHFUx-71YI-m24t-7aZQ-J9ZM-PSrD-F10Y3t"
seqno = 0
status = ["RESIZEABLE", "PARTIAL", "READ"]
system_id = "matin1023540844"
extent_size = 8192 # 4 Megabytes
max_lv = 255
max_pv = 255

   physical_volumes {

       pv0 {
           id = "3yHeIp-kPYG-dqYk-3Qvz-czTC-m05v-Hs44sq"
           device = "/dev/hdg1"    # Hint only

           status = ["ALLOCATABLE"]
           pe_start = 8696
           pe_count = 29310    # 114.492 Gigabytes
       }

       pv1 {
           id = "834fZw-DSuv-0gGa-2313-Ud40-uqyh-mh3zbW"
           device = "/dev/hdf1"    # Hint only

           status = ["ALLOCATABLE"]
           pe_start = 8696
           pe_count = 19078    # 74.5234 Gigabytes
       }

}

   logical_volumes {

       MP3 {
           id = "000000-0000-0000-0000-0000-0000-000000"
           status = ["READ", "WRITE", "VISIBLE"]
           allocation_policy = "next free"
           read_ahead = 120
           segment_count = 1

           segment1 {
               start_extent = 0
               extent_count = 12800    # 50 Gigabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "pv0", 0
               ]
           }
       }

       DATA {
           id = "000000-0000-0000-0000-0000-0000-000001"
           status = ["READ", "WRITE", "VISIBLE"]
           allocation_policy = "next free"
           read_ahead = 120
           segment_count = 1

           segment1 {
               start_extent = 0
               extent_count = 5120    # 20 Gigabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "pv0", 15360
               ]
           }
       }

       VIDEO {
           id = "000000-0000-0000-0000-0000-0000-000002"
           status = ["READ", "WRITE", "VISIBLE"]
           allocation_policy = "next free"
           read_ahead = 120
           segment_count = 7328

           segment1 {
               start_extent = 0
               extent_count = 8830    # 34.4922 Gigabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "pv0", 20480
               ]
           }
           segment2 {
               start_extent = 8830
               extent_count = 1    # 4 Megabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "Missing", 0
               ]
           }
....  (many more Missing Segments)
           segment7326 {
               start_extent = 16154
               extent_count = 1    # 4 Megabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "Missing", 0
               ]
           }
           segment7327 {
               start_extent = 16155
               extent_count = 19078    # 74.5234 Gigabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "pv1", 0
               ]
           }
           segment7328 {
               start_extent = 35233
               extent_count = 2560    # 10 Gigabytes

               type = "striped"
               stripe_count = 1    # linear

               stripes = [
                   "pv0", 12800
               ]
           }
       }

}
}
-----------------------------------------------

first this says that the 120GB disk is my PV0 and the 80 GB is my PV1, second it says that my 2 working LVs are entirely on the 120 GB Harddisk, but I created them before on the 80 GB Disk so I thought they would be spanning over both discs.

Is this backup wrong, then how do I get the data that the system uses when I mount the two working LVs readonly?
Or has lvm moved the LVs on one of my resizes so that this data is correct and I only need to add the new PV and replace all the Missing Segments by one on the new PV?

best regards

Kai Iskratsch

It's normally safer to use a packaged release (or checkout against a label corresponding to a release): it's dangerous to assume what's at the head of CVS at any instant is going to work properly.

Yes, that's what the metadata says.

And VIDEO was split between all three disks.

If you want to try to salvage something from VIDEO, yes.

Otherwise if you just want to remove VIDEO, use vgreduce --removemissing (but read the man page first: run with --test to check it's going to do what you want it to).

Alasdair

[email protected]

thats what i wanted to try.

so i created a new pv on the new disk, then I edited the backupfile so that it has one segment on the new harddisk in it instead of the missing parts.
and then i tried to run vgcfgrestore first with the -t option to see if the backupfile is correct , and then without (since the VG is a old 1.0x VG i used the -M 1 option) in both ways i got this error message

Cannot change metadata for partial volume group Data Restore failed

how can I restore the metadata to the new disk?

best regards

Kai

There was some info about this on the list before IIRC:
If you're editing a text backup file intending to restore from it, you need to make sure it's completely how you want it first - check the VG has status READ and WRITE and not PARTIAL.

Alasdair

[email protected]

thanks for your help, this finally worked, the vg is up again and i can try to get as many data out of it as possible

kai