r=adilger,shadow

Update offset calculations for 32 bit platforms

b=18374
i=johann (patch from Tappro)
i=shadow

Return only valid cookies, empty slot for failed ones.
This fixes a memory corruption causing random oops.

b=18374
i=shadow
i=tappro

Wait for mds_fl_synced to be set during recovery before allowing clients
to reconnect. This is done to make sure that llog is properly set up.

b=18067
i=johann
i=adilger

Change oap magic LASSERT into LASSERTF to aid debugging.

i=Johann
test_53* fix: rm dirs used at start

b=17895
i=adilger
i=alex

While using HA for Lustre servers with Linux RAID, it is possible that MMP will not detect multiple mounts. To make this work we need to unplug the device queue in RAID when the MMP block is being written. Also while reading the MMP block, we should read it from disk and not the cached one.

i=Nathan
get rid of set_obd_timeout(), improve test_26*

b=16895
i=adilger (o=johann)
i=girish

dont assert on lvfs_check_rdonly()

i=Adilger
-- get rid of Hendrix rates values
-- create NUM_FILES -le IFree
-- use mdsrate for cleanup instead of rm

Update ChangeLog for release.

Branch b_release_1_6_7
b=16417
i=green
i=johann
i=bobijam

when there is any inactive ost, refuse to do quotacheck
b=18298
i=johann
i=andrew.perepechko

b=16895
i=adilger (o=johann)
i=girish

don't LASSERT on read-only device

r=johann

- temporary fix for assert caused by long connect interpret in umount time.

Branch b_release_1_6_7
b=18224
i=johhan
i=green
i=adilger
i=wangdi

Branch b_release_1_6_7
b=18238
i=green
i=umka

Branch b_release_1_6_7
b=17972
i=shadow
i=johann

1, print the actual time to wait for
   lock cancellation in "waiting_locks_callback"

2, compare resource in "ost_rw_hpreq_lock_match"

3, replace "LDLM_ERROR" with "LDLM_WARN" if a lock
   is completed after waiting more than obd_timeout

i=vitaly.fertman
i=johann

Branch b_release_1_6_7
b=18150
i=grev

b=16839
i=adilger
i=tappro
i=green

Patch from shadow.
send ACTIVATE event only if connect finished and import have state FULL.

- synchronize test with HEAD version and rename it to test_35 (just like in HEAD);

- OBD_RACE in all execution branches in ldlm_lock_decref_internal() to balance OBD_RACE in bl_ast
and prevent hanging.

b=17379
i=johann, bobijam

Fixed incorrect vaule of symlink depth number.

- reformat after test 46a

i=H.Huang
test_57b remote mds fix

i=Brian
remote rundbench run fix

checking maximal acl size before set.

Branch b_release_1_6_7
b=17636
i=johann
i=deen

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

b=18083
i=Yong Fan
i=Andrew Perepechko

Enforce op. quota file format for MDS too

  - check for OBD_FAIL_OBD_LLOG_SETUP fail_loc regardless ->lop_setup definition. This is required because
  after changes in mds_ost llog setup, which is now done on behalf of separate thread, we need to make
  llog_setup() fail in context of mds start thread to make the test actualy work. And those llogs which
  get initialized in mds start thread do not have ->lop_setup defined.

remove redundant code in previous patch

i=tappro

i=Nathan.Rutman
i=Robert.Read
zconf_umount_client{s} fix: kill ps-s before force umount

b=16117
add test scripts for b16117.
i=adilger, johann

  r=shadow

  - handle possible hangup while waiting for dd processes;
  - write concurrently in smaller steps so probablity of bl_ast is higher.

b=18049
o=johann
i=adilger
i=zhenyu.xu (bobijam)

Descriptoin: aborting recovery hang on MDS
Details    : don't throttle destroy RPCs for the MDT.