vgscan eating memory

Hello,

after upgrading to Slackware 10.0, I ran into a problem with LVM2.

system: Slackware Linux 10.0
kernel: 2.6.7-mm5
Official slackware packages:
device-mapper-1.00.17
lvm2-2.00.15

Problem: lvm vgscan runs, eats up more and more memory until it gets OOM-killed.

When run, it says:

Reading all physical volumes. This may take a while... Ignoring additional label on /dev/hda11 at sector 1

and keeps on munching the disk.

(/dev/hda11 is a LVM partition)

I'm not sure how to debug it. strace shows n * (read,ioctl(BLKBSZGET),_llseek) and a few brk's occasionally..... No errors get logged or displayed, except the fishy notice about additional label.

When I use the OLD lvm2 binary (2.00.08), it works just fine (was running the same kernel and same version of device-mapper before, too).

Neither of the lvm packages have any patches applied to them. Slackware's package is ./configured with --enable-static_link and --with-lvm1=internal. My old package was configured with --with-lvm1=shared.

Any hints would be welcome.

--
Tarmo Samuel

I've seen a similar bug in the past, but that one got fixed.

(1) Please reproduce with the latest LVM2 release (2.00.20)

(2) Also reproduce after building with configure --disable-o_direct

(3) Then send standard lvm2 diagnostics: turn on debug logging to file in lvm.conf, run vgscan, and send me (not the list) the
(compressed) output. [If it's big, you can truncate it from
where it starts repeating itself.]

  log {
        file="/tmp/lvm2.log"
        level=7
  }

That suggests you've somehow got more than one LVM2 label on /dev/hda11:
the debug output may give more details.
[Or send me the first 2k of the device.]

What's the history of LVM commands affecting that device? (eg created with an old LVM2 pvcreate or/and the current one, or with the --labelsector option? dd onto it? vgcreate/extend? adjusted partition table?)

Alasdair

[email protected]