[linux-lvm] segfaults and "VGDATA .. inconsistent" beta7 tools

[Paul Jakma]

On Wed, 16 May 2001, Heinz J. Mauelshagen wrote:

> The lvm library sets up a device name/number cache in order to
> map between those. If pv_create_name_from_kdev_t() returns
> (char*)NULL, an entry for /dev/ida/c0d0p6 couldn't be found in
> the cache :-( Even though /dev/ida/ is a supported namespace, the
> entry couldn't have made it into the cache, because the stat() on
> the device returned an error.
>
> In order to check this, you could for eg. add a "printf ("found
> %s\n", devpath)" after the sprintf() in lvm_add_dir_cache() in
> lvm_dir_cache.c and and a "printf ("ADDED %s\n", devpath)" after
> the cache_size++ in the same function.
>
> BTW: long time ago, we were faced with problems related to messy /dev/ entries.
>      In case you don't run devfs you need to make sure that /dev/ entries are
>      all right.


hmm... devfs.... that got me thinking. i run devfs, but mounted on
/devfs instead of /dev for informational purposes. search through the
archive shows that having devfs compiled in but not mounted on /dev
is a known problem as devfs changes /proc/partitions.

compile without devfs and hey presto both machines work fine...

> No.
> See above.

i think i've answered it myself.

thing is, i still want to run devfs, but not on /dev. if i convert
all the file names in /dev to be the same as in devfs then LVM would
work, wouldn't it?

regards,
-- 
Paul Jakma	paul at clubi.ie	paul at jakma.org
PGP5 key: http://www.clubi.ie/jakma/publickey.txt
-------------------------------------------
Fortune:
Committees have become so important nowadays that subcommittees have to
be appointed to do the work.