[linux-lvm] filesystem corruption...

[Anders Widman]

> On Thu, Feb 28, 2002 at 10:07:03AM +0100, Anders Widman wrote:
>> 
>> > On Thu, Feb 28, 2002 at 09:35:53AM +0100, Anders Widman wrote:
>> >> 
>> >> >> On Wednesday, February 27, 2002 04:42:28 PM -0700 Andreas Dilger <adilger at clusterfs.com> wrote:
>> >> 
>> >> >>> On Feb 27, 2002  23:36 +0100, Anders Widman wrote:
>> >> >>>> Unfortunatley, something went seriously wrong. I can't mount the disk
>> >> >>>> any more, or use reiserfsck. They (mount/reiserfsck) say there isn't
>> >> >>>> a valid filesystem on the device. Vgscan does however find all devices
>> >> >>>> and can activate the volume group, but reiserfsck doesn't work.
>> >> >>>> 
>> >> >>>> What should I do to be able to save a s much data as possible?
>> >> >>> 
>> >> >>> dd if=/dev/vg/lv of=/new/disk conv=sync,noerror
>> >> 
>> >> >> Then we should be able to use debugreiserfs and reiserfsck to 
>> >> >> find a copy of the super in the log, or just recreate it.
>> >> 
>> >> >> -chris
>> >> 
>> >> > So. How much space would this require on the 'new' disk? The same
>> >> > amount as the entire LV? If that is, it would be impossible (for me,
>> >> > economically) to do this.
>> >> 
>> >> > The disk with bad blocks is 80GB, and I have a "spare" 80GB disk, but
>> >> > not more.
>> >> 
>> >> > //Anders
>> >> 
>> >> Oh, I forgot to ask how I can recover the filsystem
>> >> header/suprtblock.. What exactly is "dd if=/dev/vg/lv of=/new/disk
>> >> conv=sync,noerror" doing to the data, except copying it to the new
>> >> device?
>> 
>> > It copies all it can read to another sane device so that youe don't loose
>> > more data if your drive turns worse.
>> 
>> > After that, you want to use "reiserfsck --rebuild-sb /dev/vg/lv".
>> 
>> Ok. Unfortunatley, I can't get enough storage to copy all data. Would
>> it be possible to just copy the data from the broken disk to a new,

> So /dev/vg/lv is bigger than that disk?

Yes, much bigger...


>> identical disk (I have a "spare" disk of the same model), and then
>> rebuild the filsystem?

> Well, then pvcreate the spare disk, add it to your VG and
> "pvmove -i /dev/BrokenDisk /dev/SaneDisk" data over.

Great. I remember reading something about this '-i' option on this
list before, just didn't remember. So this is as safe as use 'dd' to
move all data from the bad disk to the new disk?


> You need to have recent LVM 1.0.3 tools in order to use the -i option of
> pvmove, which ignores read errors and you need to patch liblvm.h with the
> following one to make pvmove work correctly.

> After that repair the filesystem in /dev/vg/lv.

This patch below is to add the support to ignore read errors?

> diff -u -B -r1.43 -r1.44
> --- LVM/tools/lib/liblvm.h      18 Feb 2002 16:37:18 -0000      1.43
> +++ LVM/tools/lib/liblvm.h      20 Feb 2002 10:49:14 -0000      1.44
> @@ -93,10 +93,10 @@
>  #include <time.h>
>  #include <limits.h>
>  #ifdef _G_LSEEK64
> -int lseek64 ( unsigned int, unsigned long long, unsigned int);
> +loff_t lseek64 ( int, loff_t, int);
>  #define llseek lseek64
>  #else
> -int llseek ( unsigned int, unsigned long long, unsigned int);
> +loff_t llseek ( int, loff_t, int);
>  #endif

>  #include <sys/ioctl.h>
> @@ -130,7 +130,7 @@
>  #define        LVMTAB                  "/etc/lvmtab"   /* LVM table of VGs */
>  #define        LVMTAB_DIR              "/etc/lvmtab.d" /* storage dir VG data */
>  #define        LVMTAB_MINSIZE   ( sizeof ( vg_t) + sizeof ( lv_t) + sizeof ( pv_t))
> -#define        LVM_DEV                 "/dev/lvm"
> +#define        LVM_DEV                 LVM_DIR_PREFIX "lvm"
>  #define        VG_BACKUP_DIR           "/etc/lvmconf"
>  #define        DISK_NAME_LEN           8
>  #define        LV_MIN_NAME_LEN         5
 
And thanks again for your help!

//Anders