Re: [Evms-devel] Problem: EVMS root fs & kernel 2.6x

[<ma...@op...>]

On Mon, Apr 18, 2005 at 03:06:31PM +0200, Michel Bouissou's all...
  >BUT here we have a problem: As soon as we have pivoted root to the true
  >root fs, we are now using the /dev, and /dev/evms, that resides on the
  >true root fs.
  >
  >It can happen (and I saw it happen) that, for some reason, i.e. one disk
  >has been removed, the evms_activate that was performed from within initrd
  >didn't use the same device major/minor numbers for the
  >/dev/evms/rootfsvolume, as what is stored in /dev/evms/rootfsvolume on
  >disk.
  >
  >In this situation, as soon as pivot_root is performed, we have a mismatc=
h,
  >and good chances that things go bad when rc.sysinit tries to determine if
  >the root fs needs to be fsck'd before remounting it RW... And of course
  >evms_activate cannot be performed before the root fs has been remounted
  >RW, which shouldn't happen before the root fs has been fsck'd, which
  >cannot be done as we have a device major/minor mismatch.

Wouldnt /proc/sys/kernel/real-root-dev override this?

  >This situation not only can happen, but has good chances to happen at the
  >worst moment, i.e. after a disk died and the system was rebooted with one
  >less disk and out-of-sync RAIDs etc...
  >
  >I don't see no easy workaround for this, as the only "good" solution wou=
ld
  >be to make sure that the /dev/evms directory on the "true" root fs match=
es
  >what evms_activate did within the initrd, which means that it should be
  >rewritten by the initrd, which means the initrd would need to mount the
  >"true" root fs RW at some point, which means the initrd should take good
  >care of fscking it if needed before doing so... Uhh...
  >
  >The other possible solution would be to modify EVMS itself to make sure
  >that if we have an EVMS "true root fs", then it will ALWAYS get the same
  >device major/minor numbers, i.e. by reserving a given minor number for
  >this specific use.
  >
  >Comments ?

If this is truly happening, I dont know how EVMS works AT ALL in that
case, because all the 'work' it did to create a /dev/evms tree is now
gone - pivotted away now that the real root is in place.

Hmm. I think I know what the answer is here - Kevin has mentioned this
before - emvs_activate has to run again very early in the boot sequence,
in fact before the root fs fsck, to ensure all the /dev stuff is=20
recreated. I didnt understand why at the time, but in light of what you
bring up here, now I understand.

Which begs the question - why arent we fscking the root FS while
we're in the initrd?

/kc



  >--=20
  >Michel Bouissou <mi...@bo...> OpenPGP ID 0xDDE8AC6E
  >
  >
  >
  >
  >
  >-------------------------------------------------------
  >SF email is sponsored by - The IT Product Guide
  >Read honest & candid reviews on hundreds of IT Products from real users.
  >Discover which products truly live up to the hype. Start reading now.
  >http://ads.osdn.com/?ad_ide95&alloc_id=14396&op=3Dclick
  >_______________________________________________
  >Evms-devel mailing list
  >Evm...@li...
  >To subscribe/unsubscribe, please visit:
  >https://lists.sourceforge.net/lists/listinfo/evms-devel