Description of problem:
On a 2x2 distributed replicate volume. Mounted a fuse client and kept sanity tests to run. Immedietly stopped the volume and restarted it. The sanity test (executing lftest of ltp test suite) hung.
Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.
  
Actual results:


Expected results:


Additional info:
Statedump information ===>

[global.callpool.stack.2]
uid=0
gid=0
pid=1666
unique=114637
op=WRITE
type=1
cnt=6

[global.callpool.stack.2.frame.1]
ref_count=1
translator=mirror-write-behind
complete=0

[global.callpool.stack.2.frame.2]
ref_count=0
translator=mirror-client-2
complete=1
parent=mirror-replicate-1
wind_from=afr_getxattr
wind_to=children[call_child]->fops->getxattr
unwind_from=client3_1_getxattr_cbk
unwind_to=afr_getxattr_cbk

[global.callpool.stack.2.frame.3]
ref_count=0
translator=mirror-replicate-1
complete=1
parent=mirror-dht
wind_from=dht_getxattr
wind_to=subvol->fops->getxattr
unwind_from=afr_getxattr_cbk
unwind_to=dht_getxattr_cbk

[global.callpool.stack.2.frame.4]
ref_count=0
translator=mirror-replicate-0
complete=0
parent=mirror-dht
wind_from=dht_getxattr
wind_to=subvol->fops->getxattr
unwind_to=dht_getxattr_cbk

In afr_getxattr we intialize local structure and collect return value in op_ret.
AFR_LOCAL_INIT which intializes the local structure returns values less than zero (infact -ve of the corresponding errno happened) upon some failure. But we will unwind only if op_ret is -1. thus it leads to lost frames and application hanging.