I've seen the "server went back in time" message while testing HEAD,
so this will fix that.  Patch already on 1.4.

bits from 3599 landed to help debugging

- ext3-inode-reuse is part of vanilla-2.4.24 series
- MDS doesn't take lock to prevent inode recreation
  (this improves performance a bit)

Corrected. OBD_MD_MDS should only be set when the directory was
created on a different node.

r=braam
update lctl.lyx

r=braam
update lmc.lyx and lconf.lyx

revert while i figure out what i did wrong.

Also reconstruct what we can in the body, which is hopefully enough.

2)add clientfsoption in lmc and lconf
3)some fix in cobd for cache miss handler

oscc_internal_create() increase pre-creation window smoothly

r=bzzz

Reconstruct create reply for CROSS_REF directories from the dentry.

use the correct bug#

more diagnostics.

more debugging

renmae 2nd ptlrcd "ptlrpcd-recov"

A small races exists between receiving a reply and timeout. This diff
prevents ptlrpc_expire_one_request from timing out a req that has
already received the reply.

THe hack on b1_2 branch to call set_info("mds_conn") on a specific OSC
instead of all of them was removed on HEAD, and this broke
mds_lov_synchronize.  Instead of introducing a new hack to do this,
for now this changes m_l_s to send the set_info() to all the connected
OSTs. This should be harmless.

- with CMD each MDS has own group on OST, so group != 0 always. this is
  why filter_should_precreate() did return 1 always, so precreation path
  didn't take a place. I've comment out check for group out for a while

- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only
- lmv_disconnect() should drop connected flag upon real disconnecting only

- race in dynlock_unlock() fixed: several threads may find refcount=0 and
  try to free lock member at the same time. thanks to Andreas!

  is called. the last one didn't drop lmv->connected to 0 and this prevent
  subsequent lmv_connect() to work properly

b=3869,1742

These are the remaining fixes from 1742 that are needed to allow a
client to reconnect during recover.  The request that triggered 3869
on the last run was actually a RESENT request that was sent before
recovery had finished, which this patch will fix.

2 some fix for regular file cow

- more debug info

b=1451

Don't allow client to reconnect if an RPC is already in progress for
that client. Added an extra counter for this, because the export
refcount is also held by locks and who knows what else.