- fixed declation of OBD_MD.. flags. They should be declared like the following:
#define OBD_MD_FLALL (0xffffffffffffffffLL)

to show compiller explicitly that they are 64bit value and should considered as such.

- fixed md_reint_rename(). It should first check if objects are the same then check if tgt is subdir of src because for the same inodes is_subdir() returns 1 and mds_reint_rename() for renaming object into itself it returned -EINVAL;

- added asserts in LMV checking if body->valid contains valid fid before accessing it to check if object is splitted.

- in lmv_intent.c after md_intent_lock() checking not only error code, but also disposition, because there are cases when we're interested in name which does not exist, lock is not taken and policy_res2 is cleared in mds_intent_policy(). In this case we should do nothing (especially access body->id1 in LMV), but exit. This is because body->id1 is zero and (body->valid & OBD_MD_FID) == 0 because say mds_getattr_lock() failed before filling out as inode is not found.

r=adilger
Do not rehash dentries in ll_inode_revalidate_it(), it is only supposed to
update inodes.

- returning correct error code from mds_statfs() amd mds_sync(), as otherwise it may cause segfault in mds_handle().

and should be restared with correct one in mds_getattr_lock().

     -  added vib
     -  zeroconf vib, ra
     -  lconf/lmc openib,vib changes

- more fixes about fid extracting on MDS. By now correct fid is always should
be returned to client.

renamed to dt_exp and lmv_exp is renamed to md_exp. Related changes.

- removed OBD_MD_FID. And fid is now fetched on MDS on all constructive requests. It
is not fetched on unlink(), rename(), etc. This is because OBD_MD_FID flag lost
the sense as lmv layer needs it for all operations (to check if object is splitted).

- fixed problem with sid of virtual directory "FIDS". By now it is saved to store just like last_rcvd and last_fid and read from there on startup.

- a lot of changes about last_fid management. last_fid_changed flag is removed and last_fid_sem is
replaced by spin lock.

so there is no need to accomodate for 2.4 kernel config differences.

So it must only warn, not abort.

remove unnecessary error msg, cleanup

- calling mds_finish_transno() in mds_reint_unlink() in all cases, not only in
cleanup_release == 1. This may cure problem we have observed yesterday (about
duplicating fids for parent and child).

- removed redundant lockh initializing in few places.
- release locks the same way as in other places (using lockh + 1, not &lockh[1]).

parent lock is not taken, so we do not need to release it.

set o_fid in mdc_obj_create for LBUG in mdt_obj_create
b:5234

the patch contains:
  1) journal_release_buffer() doesn't cause journal overflow
  2) journal_commit_thread() uses persistent buffer instead of local wbuf
     (could cause journal overflow also)
  3) optimization in journal_forget() allows to avoid needless commits

r=jacob
r=alex

r=jacob.

*   Removed S_???NAL in favour of a single S_NAL

*   NB liblustre needs attention