[linux-lvm] Failure recipe: 2.4.29, lvm2 lvcreate, dd reading from file on RO xfs

[linux-lvm] Failure recipe: 2.4.29, lvm2 lvcreate, dd reading from file on RO xfs

[Tom Dickson]

[-- Attachment #1: Type: text/plain, Size: 2403 bytes --]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

We have a reproduceable failure of device-mapper when running on the 2.4.29 kernel
with XFS. With the XFS logical volume mounted read only, we run a script to
continually read a one gigabyte file from the filesystem.

We run a second script that creates 3 snapshots (10 seconds between them), and
then waits 30 seconds, and throws them away.

After 7-10 iterations, the lvcreate command fails with:

trying to execute lvcreate -sns3 -L5G /dev/H99/v
~  device-mapper ioctl cmd 9 failed: Cannot allocate memory
~  Couldn't load device 'H99-s3'.
~  Problem reactivating origin v

and dmesg says:

~  device-mapper: Unable to allocate hash table space
~  device-mapper: error adding target to table
~  __alloc_pages: 0-order allocation failed (gfp=0x20/0)

At this point, anything reading from /mnt/H99 (the read only mount point) is
deadlocked.

If we run dmsetup resume H99-v, it begins to work again. However, when we tried to
unmount it (at least one time), it caused a kernel BUG() in dm-snapshot.c:1064.
The output of that kernel oops is in dmesg-dm-0204c.out.

The system has 1GB of RAM, highmem enabled, and uses the .config file attached.

We're running further tests to try and track down exactly what is happening, doing
the dd directly from /dev/H99/v seems to work. It looks like the snapshot
allocation is failing, and then failing to resume the source volume.

The failure occurs on our system while running
~  repeat-mk-NO-s3.sh and while-dd-from-abcd
(see test139scripts-0204c.tgz)

In an environment with minimal services (excluding cron to avoid any of
our custom cron actions)and with the xfs fs mounted read-only!
(same problem occurs if rw)

our kernel 2429-db0 is built with the
following patches
~  xfs-2.4.24-pre1-split-acl.bz2
~  linux-2.4.28-pre4-devmapper-ioctl.patch
~  linux-2.4.22-VFS-lock.patch

The dmsetup on the test box reports
~  dmsetup --version
~  Library version:   1.00.21-ioctl (2005-01-07)
~  Driver version:    4.1.1
and lvm reports:
~  lvm version
~  LVM version:     2.00.33 (2005-01-07)
~  Library version: 1.00.21-ioctl (2005-01-07)
~  Driver version:  4.1.1
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFCBBZx2dxAfYNwANIRAsknAJ9Vm7nhxa25u258uc1KA7NokhuSvQCdEeit
B1PZ0pViwTSwOIKaYPkgUQU=
=qaZG
-----END PGP SIGNATURE-----

[-- Attachment #2: lvm2.tgz --]
[-- Type: application/x-compressed, Size: 8684 bytes --]