check import before put it.
b=15684
i=adilger

b=15716
i=rread
i=johann

r=shadow,nikita
- fixes issue with accessing freed @ns on behalf rpc handling when some late rpc comes in fini time;
- add new comments in functions, converts existing ones to be inline wigth doxygen comments policies.

b=14929
i=zam

b=14929
i=zam

ptlrpc connect with V2 message format.

b=13821
i=wangdi
i=umka

- redesigned remote acl support, by FanYong
  b=11842
  r=pravin.shelar, eric.mei
- new sptlrpc configure interface
  b=13642
  r=yong.fan, rahul.deshmukh
- split kerberos keytab for lustre_root
  b=13873
  r=yong.fan, h.huang
- fix setuid for gss with linux keyring
  b=13899
  r=yong.fan, h.huang

  b:13815
  i:green, shadow

  b:13816
  i:shadow,yury

b=12840
i=green
i=johann

prevent self-evicting through procfs

i=tappro
i=fanyong
block reactivating mgc import until all deactivates complete
Only an issue when failing back MDT/MGS to itself (testing)

b=12606
i=shadow, nikita

Description: don't use GFP_* in generic Lustre code.
Details    : Use cfs_alloc_* functions and CFS_* flags for code portability.

Patch from nic@cray.com: add spin locks around import/export bit flag changes.

b=11315
i=adilger
i=alex

b=11013
Back out bz 11013 due to chronic test failures

b=11013
r=adilger

implement a common hash table, so it will be used as a hash class.
in this patch, finished :
1. " UUID <--> export " hash
2. " NID  <--> export " hash
3. "NetPeer <--> connect" hash

r=green
don't allocate inside a spinlock

 - b_ldlm_newlocking landed. we need it to pass some CMD2 performance tests
   NOTE: the new locking rules introduced by the patch are still experimental!

some fixes about write back cache.
1)cleanup smfs_init_inode_info
2)miss spin_unlock in class_find_client_obd
3)remove PLG_ALL flags of OBJECTS and LOGS dir on MDS
4)some minor fix in cmobd.sh for WB cache

 - all clients report their minimal transno to be replayed
 - server waits for all clients to connect in order to collect
   minimal transno's from clients and replay all requests
   in right order

 - to avoid possible lock collision during replay, we should replay all
   request before any locks

 - workaround for cascading timeouts when failure of 1 server node causes
   failures of unrelated connections. this caused different problems. for
   example, lock cancel timeouts.

- fixes and clanups in cmobd.sh and cobd.sh
- changes in lconf about module loading in OSD and MDSDEV.
- fixes in lconf about mount options forming.
- fixes in lmc about client ref managing.
- fixes in lconf about cmobd and cobd setup.

- changed mds_alloc_inode_sid() to allocate lustre_id in memory if there is
no passed one. This makes mds stuff do not use lustre_id allocated on stack
if there is only need to allocate inode sid without using it later. This
makes stack consumption smaller by 40 bytes in each such a case.

- in mds_setup() whrn comparing strings passed from userspace use strncmp()
- root lustre_id should be set up also in cases when there is no profile.
- making do not use lustre_id allocated on stack in places where it is possible.

- moving mds_finish_transno() calling to place where dput() for dchild is
already called to make sure, that unlink op got to current transaction.

- lots of changes in lconf. MDSDEV stuff is changed substantionaly. Removing
bugs prone code, replacing it by new one about setting up MDS, LMV, etc.
Added more verbose messages about MDS, LMV configuration. Removed redundant
stuff about forming mount options. It now is moved to function and be used from
few places.

- fixes in CMOBD and COBD objects about using LMV.

- fixed double client_ref adding to LMV, what led to double client config
writing on MDS setup.

that is unsigned long too.
- expanding all ->valid fields to 64bit, as it is exhaused and there is no free bits.

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.

I haven't re-tested the AMD functionality since landing, but I did
test iozone and dbench to ensure sanity.

b_cmobd_merge should not be used for any further development or testing!