Description of problem:
(gdb) bt
#0  0x00007fe56cb27503 in dht_fsync (frame=0x7fe576d701c4, this=<optimized out>, fd=0x7fe560002cec, datasync=0, xdata=0x0)
    at ../../../../../xlators/cluster/dht/src/dht-inode-read.c:818
#1  0x00007fe56c8db4b8 in wb_fsync (frame=0x7fe576d70118, this=0x7fe56800a2e0, fd=0x7fe560002cec, datasync=0, xdata=0x0)
    at ../../../../../xlators/performance/write-behind/src/write-behind.c:1523
#2  0x00007fe56c6ccb3a in ra_fsync (frame=0x7fe576d7006c, this=0x7fe56800b6b0, fd=0x7fe560002cec, datasync=0, xdata=0x0)
    at ../../../../../xlators/performance/read-ahead/src/read-ahead.c:628
#3  0x00007fe577eefb1d in default_fsync (frame=0x7fe576d7006c, this=0x7fe56800c9e0, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../libglusterfs/src/defaults.c:1817
#4  0x00007fe577eefb1d in default_fsync (frame=0x7fe576d7006c, this=0x7fe56800dde0, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../libglusterfs/src/defaults.c:1817
#5  0x00007fe577eefb1d in default_fsync (frame=0x7fe576d7006c, this=0x7fe56800f0d0, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../libglusterfs/src/defaults.c:1817
#6  0x00007fe577efab3b in default_fsync_resume (frame=0x7fe576d70a80, this=0x7fe568010540, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../libglusterfs/src/defaults.c:1376
#7  0x00007fe577f14fdd in call_resume (stub=0x7fe5767f7dd4) at ../../../libglusterfs/src/call-stub.c:2576
#8  0x00007fe567dfa578 in open_and_resume (this=this@entry=0x7fe568010540, fd=fd@entry=0x7fe560002cec, stub=0x7fe5767f7dd4)
    at ../../../../../xlators/performance/open-behind/src/open-behind.c:241
#9  0x00007fe567dfa922 in ob_fsync (frame=0x7fe576d70a80, this=0x7fe568010540, fd=0x7fe560002cec, flag=<optimized out>, xdata=<optimized out>)
    at ../../../../../xlators/performance/open-behind/src/open-behind.c:498
#10 0x00007fe567becbaf in mdc_fsync (frame=0x7fe576d709d4, this=0x7fe5680118d0, fd=0x7fe560002cec, datasync=0, xdata=0x0)
    at ../../../../../xlators/performance/md-cache/src/md-cache.c:1669
#11 0x00007fe5679d0f76 in io_stats_fsync (frame=0x7fe576d70270, this=0x7fe568012c90, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../../../xlators/debug/io-stats/src/io-stats.c:2207
#12 0x00007fe577eefb1d in default_fsync (frame=0x7fe576d70270, this=0x7fe568014160, fd=0x7fe560002cec, flags=0, xdata=0x0)
    at ../../../libglusterfs/src/defaults.c:1817
#13 0x00007fe577f3040a in syncop_fsync (subvol=subvol@entry=0x7fe568014160, fd=fd@entry=0x7fe560002cec, dataonly=dataonly@entry=0, 
    xdata_in=xdata_in@entry=0x0, xdata_out=xdata_out@entry=0x0) at ../../../libglusterfs/src/syncop.c:2280
#14 0x00007fe56fe0257a in fuse_migrate_fd (this=this@entry=0x1661700, basefd=basefd@entry=0x7fe560002cec, 
    old_subvol=old_subvol@entry=0x7fe568014160, new_subvol=new_subvol@entry=0x7fe560030be0)
    at ../../../../../xlators/mount/fuse/src/fuse-bridge.c:4423
#15 0x00007fe56fe0271d in fuse_handle_opened_fds (this=0x1661700, old_subvol=0x7fe568014160, new_subvol=0x7fe560030be0)
    at ../../../../../xlators/mount/fuse/src/fuse-bridge.c:4508
#16 0x00007fe56fe027c9 in fuse_graph_switch_task (data=<optimized out>) at ../../../../../xlators/mount/fuse/src/fuse-bridge.c:4559
---Type <return> to continue, or q <return> to quit---q
Quit
(gdb) f 0
#0  0x00007fe56cb27503 in dht_fsync (frame=0x7fe576d701c4, this=<optimized out>, fd=0x7fe560002cec, datasync=0, xdata=0x0)
    at ../../../../../xlators/cluster/dht/src/dht-inode-read.c:818
818	        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
(gdb) l
813	        local->call_cnt = 1;
814	        local->rebalance.flags = datasync;
815	
816	        subvol = local->cached_subvol;
817	
818	        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
819	                    fd, datasync, xdata);
820	
821	        return 0;
822	
(gdb) p subvol
$1 = (xlator_t *) 0x0

Version-Release number of selected component (if applicable):
master

How reproducible:
Fairly consistently in some setups

Steps to Reproduce:
1. run ./tests/performance/open-behind.t in a loop. Sometimes Gluster mount crashes after we switch open-behind off. The crash is seen when when "num_graphs" is executed.

2.
3.

Actual results:


Expected results:


Additional info:

RCA: Its a race. If a graph switch happens when we have an fd opened under ".meta" subtree, crash is seen when migrating that fd. meta-fsync is not implemented and hence call is wound to real volume. Other translators have no idea of handling this virtual inode. Fix is to implement fsync in meta. As a complete fix, meta should implement all fops, so that it can check whether an inode belongs to it or not and wind the call down only if it does not belong to itself.