[DRBD-user] drbd uml 2.6.6

Lars Ellenberg Lars.Ellenberg at linbit.com
Wed May 12 18:06:01 CEST 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


/ 2004-05-12 16:19:08 +0100
\ Mark Olliver:
> 
> Hi,
> 
> I just want to check if anyone has managed or knows of any problems of 
> using kernel 2.6.6 with the skas uml patch applied and drbd??

I was using 2.6.5, and since about three hours I'm using 2.6.6 UML as my
main test setup. If you want, I can upload my config and full patch
somewhere on drbd.org.

basically it is 
 kernel.org 2.6.6,
plus some patches from http://www.user-mode-linux.org/~blaisorblade/ ,
namely:
 uml-patch-2.6.6-01.bz2,  # which applies cleanly
 patch-2.6.6-v2,          # some offsets, no fuzz
 and finally
 host+Uml-skas3-2.6.6-v1.patch
 
there has been exactly *one* reject in one of the last two patches,
iirc for include/asm-um/irq.h, which was easily applied by hand.

then, of course, in drbd, make kernel-patch KDIR=/...2.6.6+UML.../
and applie.

make menuconfig ARCH=um, enable DRBD, make linux ARCH=um,
and there you go.

generic problems:

  the 2.6.6 uml hostfs seems to have reintroduced a weird bug, not closing
  file desctiptors *on the host* properly, thus the UML instance runs into
  "too many open files". so don't use hostfs too much, until they sorted
  that out, or someone please explain to me what I am doing wrong...
  it seemed to have worked fine with my previous setup,
  2.6.5 + uml-patch-2.6.4-1 + some skas and whatnot patches


drbd related:

  drbd seems to work fine, as well as module, and as "monolithic" build.

  XFS seems to work on top of DRBD within uml, and survives a tiobench.
  but sometimes, if you copy many files, it get stuck waiting for some xfs
  internal semaphore to be released, which never happens.

  this only shows on top of drbd, not using the underlying ubd directly,
  but seems to be a xfs internal problem, since it even shows when DRBD
  is "virgin standalone", just passing request to the lower level device,
  and passing end_io events up to the upper levels. with a recently
  introduce atomic counter I can *prove* that drbd did the correct endio.

  only xfs seems to not up its own semaphore properly...
  as soon as this is reproducable in real systems (where currently xfs on
  top of DRBD seems to have all sorts of other weird problems, which I
  doubt are DRBDs fault... only xfs does not expect the restrictions that
  the DRBD requests queue imposes), this needs to be sorted out with the
  XFS guys... 


otherwise, doing uml is fun.
hav a look into drbd/testing, where there is a script named
uml-screen-debug, which I found very useful to start my umls.


	Lars Ellenberg



More information about the drbd-user mailing list