fix a race that early reply right after realy reply.
b=15178
Christopher Morrone, LLNL (original author)
r=nathan
r=ericm

Do not skip at_check_timed in AT_OFF mode, early replies might still be
sent out until affected RPCs will drain eventually.
b=15341
r=nathan
r=rread

individual file sync to eliminate dangerous inode locking
with I_LOCK that can lead to a deadlock.

b=15188
i=green
i=shadow

b=14661
i=green
i=shadow

-- remove BL_AST lock from lru once we mark the lock as a such;
-- do not place BL_AST lock into lru even if the lock is matched and put later;
   => no BL_AST locks in lru at all;
-- CANCELING locks are not so numerous, but to avoid its traversal by another
   thread another time, drop them from the unused list once come them across.

b=15483
i=green
i=johann

b=13397
i=johann
i=shadow

Split the export journal-api part as a separate patch.

i=Yury.Umanets
force test_66b to use FSNAME

i=Oleg.Drokin
load crc16 module if it exists

b=15152

Fixed previous lost part.

b=14751
i=green
i=bzzz

reset the time reference between calls to fsfilt_check_slow to prevent
useless slow brw_start/direct_io/... messages.

Add a new item(lq_lock) in struct lustre_qunit which protects every single qunit
and qunit_hash_lock only protects qunit hash. It will improve parellelism.
i=johann
i=andrew

b=15288
i=wangdi, shadow

fix stride_len assertion.

when current ra window is consumed, and 2 consequent read requires to enlarge
the read ahead window, and the enlargement action met window_start 0,
window_len 80, because the latest read consumes the whole window content,
stride_offset is 80, so the left of current ra window is 0, and the assertion
of stride_len should be ">= 0" rather than "> 0"

b=14257

Do not include LdLM_FLOCK locks into ldlm_pool sanity calculations.

b=14134
i=shadow, nathan

change non-fatal error messages about contacting MGS.

b=13375
i=shadow, panda

Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
Details    : If an OST is down the MDS will hang indefinitely in
             obd_statfs_rqset() waiting for the statfs data. While for
             MDS QOS usage of statfs, it should not stuck in waiting.

b=14855
i=shadow
i=adilger

On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
be defined. Add a hack in mkfs_lustre.c to work around this compile
issue.

i=Adilger
load crc16

Change #if to #ifdef to make gcc 3.x happy.

b=15327
i=oleg
i=walter

Remove specialized patch for titech, as Scjody suggested.

removal in bug 13499.

b=13499

r=shadow,nikita

Implement readv/writev (aio_read/aio_write) to greatly help NFS writes speed
on 2.6 kernels.

Increase version number to 1.6.4.53 for new tag.

Disable Adaptive Timeouts by default. Can be re-enabled by setting the ptlrpc
module parameter "at_max=600".

b=3055
i=adilger
i=nathan

b=15283
r=walter

use "get_param version" instead of "lustre_build_version"

b=13499
i=adilger
i=shadow

b=15027
i=johann
i=fanyong

update to latest e2fsprogs version

b=14763

fix for t-f

b=14872
r=nathan,ericm

Only extend(rather than reset) the timer if the time_out value from client is
bigger than current max recovery timer.

ping_evictor don't evict client too eagerly just because of it's idle.
b=15192
r=nathan
r=adilger

don't LBUG if incoming early reply size is not expected.
b=14881
r=nathan
r=adilger

b=14072
r=bobijam, johann

make error info less scary.

b=15251
i=adilger
i=bowen.zhou

Fix configure check when zlib is installed but not zlib-devel.

b=14896
i=oleg.drokin
i=tianzy

additional handling of quota cleanup in case when no init was done

Split backup and restore synopsis into separate lines so that it is clear
which option is relvant only to backup or restore.  Move the major operators
to the beginning of the option list.
b=14117

b=15033
i=adilger
i=johann

fix compile issue on Cray systems and an autoconf problem.