r=alex

 - limit number of in-flight async destroy rpcs MDS issues
   to destroy OST objects

change debug level.

missing update lvb_mtime

 - always close transaction before calling lov/osc

 - additional debug check

minor fix:
* for partial write only search capability once.

a minor reorganization of capability code.
some cleanup.

- all stats gatherd in OSC debugging time (cache loading, etc.) moved to proc. Added new proc file ./cache_stats showing them.
- some RPCs related new stats added to ./rpc_stats

fixes:
* previous implementation doesn't support multi-open/truncate by different
  users on the same inode.
* update DLD.
* cleanup.

- enabled filter_tally_write() to have brw_stats on OST

b7375
when decrypt page, should decrypt the whole page array, not only the first
page

- increased number in flight requests by factor 2. Thus twise is increased cl_dirty_max (in fact cache size)
- disabled check for in flight requests in osc_check_rpcs() - number of in flight RPCs is limited anyway by ->cl_dirty_max

- added missed fix to osc_exit_cache() from 1_4. It updates lost_grant mor carefuly what is important for cases when blocksize is not equal to page size.
- fixed warning about init of var in llog test stuff.

- added more debug in osc. Currently two ideas are working for checks why osc does not give desired band:

(1) cache size is choosen kind of cumbersome way and not enough, what makes writing thread waiting for cache in osc_enter_cache(). Cache size curremtly does not depend on how good wire is able to send RPCs to remote node what seems to be an artificial limit in some cases. If this is so, the fix would be to adjust cache size depending on much it is needed. This however does not explain why second subtest suffers from slowdown whereas first one does not.

(2) due to some reasons (limited cache or limited grant from OST) many async RPCs are falled back to synchronous IO what makes substential slowdown.

Added debug on real cluster should show what is going on there. Idea that issue on server has a little chance to be true, as server uses direct IO and there is almost not ability to be slowed down due to some cache issues, etc.

- some names cleanup in gns and capa threads related stuff to make all llite
  threads (gns, capa, close) coherent in names to each other.

add support for ost-authorization-revocation and fid verification.

 - check lop body as well

 - couple LASSERTs to catch use-after-free

land ost add/del to HEAD

- added a patch for debugging write gaps (av. gap time and sum gaps time)

- small cleanups, more detailed error messages, fied error codes in ost_rec_create_pack(),
  fixed mixed declations and non declation stuff in osc_get_info()

- fix the getting own nid - use PtlGetId now

- store node id in audit records on OSS
- move audit hook from filter_commitrw() to fsfilt_smfs_map_inode_pages()
- fix invalid nid in osc_get_info()
- rework stat/open audit hooks to store file lustre id, not parent one
- audit errors while stat/open

- fixes for making liblustre at least buildable (including fixes to compilation bug found by Brian recently)

some fixes about crypto api
1)some fixes about init inode key
2)some fixes about compute new mac

  ../../lustre/include/linux/lustre_sec.h:541: error: field `c_hash' has incomplete type

 - dedicated threads for serving OST_DESTROY and MDS_CLOSE