fix calculation of sleeping time.
b=13659
r=nathan

b=17379

Print value of symlink depth before checking it.

b=17855
i=h.huang
i=tappro

mdo_{create,unlink} should use parent object's operations

b=15691
i=h.huang
i=johann

If there are too many locks on client-side, then some locks taken by
statahead maybe dropped automatically before the real "revalidate"
using them. Under such case the statahead should fail.

i=Brian
run mpi tests as MPI_USER

b=17589
i=tappro
i=yury.umanets
i=adilger

wait a bit before client umount if vfsmnt still has references.

i=johann
i=nathan
add pool refcount
replace rwlock with semaphore

b=17584
i=adilger
i=alex.zhuravlev

b=17490
i=bzzz
i=kalpak

correctly handle device paths using a subdirectory in /dev when
creating the per-device procfs directory under /proc/fs/ldiskfs.

r=shadow,johann

Do NOT call server_deregister_mount() here. This leads to
      inability cleanup cleanly and free lsi and other stuff when
      mgs calls server_put_mount() in error handling case.

i=andreas
i=nikita
i=nathan
i=manoj

Changelogs, a lightweight mechanism to track filesystem metadata and
namespace changes.

b=17850
i=h.hunag
i=johann

change the name of md stats proc entry to "md_stats".

i=Nathan
run_one: print PASS/FAIL depending on new TEST_FAILED var

b=17876
i=umka
i=pravin

b=17770
i=Elena Gryaznova

avoid using quota_usr/quota_2usr groups

b=17589

revert the previous patch, rejected by alexey.lyashkov.

i=Adilger
check_config fix for NETTYPE=ptl

b=17748
i=grev
i=adilger

the sanityN test issue is fixed

i=Adilger
conf-sanity test_32* fix

b=17371
i=Andreas Dilger

move the cleanup/setup test to the end of the series

fix mgs_setparam, which will return -EINVAL when param
related to llite(PARAM_LLITE) is set by MDT or OST

i=johann
i=nathan.rutman

i=Scjody
test_99a fix: use $TMP as working dir

b=17461
i=adilger
i=alex.zhuravlev

Ptlrpc optimizations to minimize COS overhead:
. reply state object use own locking instead of ptlrpc service one
. per-cpu dedicated threads for reply handling
. resource hash table size increase

fix resend requests for ldlm imports.

Branch HEAD
b=17695
i=umka
i=tappro

b=17589
i=yong.fan
i=yury.umanets

wait a bit while umount client if there is still ref count on mnt.

b=17867
i=yury.umanets

Compile fix: move the fld_cache_entry_delete function body
above all the references to it.

b=17742
i=h.huang
i=alexey.lyashkov

Add necessary spin-lock protection to dump_exports().

i=Adilger
fsx fix: limit bound file size by 1/2 RAM

r=shadow
- take into account the limit on the host because the higher limit the longer it takes to kill some locks.

i=Nikita
i=Rahul

	fix for lockdep warning. adds separate lockdep subclass for mdd_orphan
dir locking.

during refreshing locks waiting its I/O to complete,
take current service time into account, not only using
the timeout gotten by ldlm_get_enqueue_timeout

i=Andreas
i=Nathan.Rutman

b=16074
r=oleg,jay

small fixes for liblustre.

b=17844
i=rahul.deshmukh
i=pravin.shelar
This warning message does not need to be printed on the console during each startup

i=Adilger
i=Tappro
test_7-b fix: use do_nodes instead of loop to run rundbench

Fix double comment termination.
i=deen

i=Adilger
skip liblustre for different client/mds versions

fix an error in the test_18b of sanity-quota.sh
b=17832
i=tianzy

i=rread
i=isaac
Landing the patch making ksocklnd more os-neutral (normalizing names of
lock primitives, moving linux-spcific code to socklnd_lib-linux.c)

This is marked as 1.4.0 compat, see if it can be removed.