Description of problem:
I created a DHT volume, the I used postmark to write 100000 of 100k files in /mountpoint/a/b/c.When postmark runinng, I exec force replace-brick and fix-layout.Like this, I would add-bricks many times.Some times, one of the glusterfsds would core dump(not the new brick).This is the bt info of the core:
Core was generated by `/usr/sbin/glusterfsd -s node-1.jdm --volfile-id test.node-1.jdm.glusterfs-w'.
Program terminated with signal 6, Aborted.
#0  0x00007fd41c14b8a5 in raise () from /lib64/libc.so.6
Missing separate debuginfos, use: debuginfo-install glusterfs-3.6.1-0.44.git890879a.el6.x86_64
(gdb) bt
#0  0x00007fd41c14b8a5 in raise () from /lib64/libc.so.6
#1  0x00007fd41c14d085 in abort () from /lib64/libc.so.6
#2  0x00007fd41c144a1e in __assert_fail_base () from /lib64/libc.so.6
#3  0x00007fd41c144ae0 in __assert_fail () from /lib64/libc.so.6
#4  0x00007fd41d0f0512 in __inode_path (inode=0x7fd3ec00dd2c, name=0x7fd3ec00dfb8 "a2", bufp=0x7fd40ae815f8) at inode.c:1175
#5  0x00007fd41d0f08e5 in inode_path (inode=0x7fd3ec00dd2c, name=0x7fd3ec00dfb8 "a2", bufp=0x7fd40ae815f8) at inode.c:1274
#6  0x00007fd40c727ea9 in marker_inode_loc_fill (inode=0x7fd3ec00dd2c, name=0x7fd3ec00dfb8 "a2", loc=0x7fd40ae816d0) at marker.c:108
#7  0x00007fd40c737016 in marker_readdirp_cbk (frame=0x79c27c, cookie=0x79c6ac, this=0x655bc0, op_ret=754, op_errno=0, entries=0x7fd40ae81970, xdata=0x0) at marker.c:2791
#8  0x00007fd41d0dbedb in default_readdirp_cbk (frame=0x79c6ac, cookie=0x7fd3ec00d04c, this=0x651890, op_ret=754, op_errno=0, entries=0x7fd40ae81970, xdata=0x0)
    at defaults.c:1235
#9  0x00007fd40cf7e137 in pl_readdirp_cbk (frame=0x7fd3ec00d04c, cookie=0x7fd3ec00e29c, this=0x650570, op_ret=754, op_errno=0, entries=0x7fd40ae81970, xdata=0x0)
    at posix.c:2120
#10 0x00007fd40d198042 in posix_acl_readdirp_cbk (frame=0x7fd3ec00e29c, cookie=0x7fd3ec00de4c, this=0x64f170, op_ret=754, op_errno=0, entries=0x7fd40ae81970, xdata=0x0)
    at posix-acl.c:1580
#11 0x00007fd40d7de5d3 in posix_do_readdir (frame=0x7fd3ec00de4c, this=0x64ac30, fd=0x76ecbc, size=130944, off=0, whichop=40, dict=0x83468c) at posix.c:5157
#12 0x00007fd40d7de947 in posix_readdirp (frame=0x7fd3ec00de4c, this=0x64ac30, fd=0x76ecbc, size=130944, off=0, dict=0x83468c) at posix.c:5204
#13 0x00007fd41d0e52fa in default_readdirp (frame=0x7fd3ec00de4c, this=0x64c700, fd=0x76ecbc, size=130944, off=0, xdata=0x83468c) at defaults.c:2078
#14 0x00007fd40d1983e4 in posix_acl_readdirp (frame=0x7fd3ec00e29c, this=0x64f170, fd=0x76ecbc, size=130944, offset=0, dict=0x83468c) at posix-acl.c:1614
#15 0x00007fd40cf7e501 in pl_readdirp (frame=0x7fd3ec00d04c, this=0x650570, fd=0x76ecbc, size=130944, offset=0, dict=0x83468c) at posix.c:2150
#16 0x00007fd41d0e21e2 in default_readdirp_resume (frame=0x79c6ac, this=0x651890, fd=0x76ecbc, size=130944, off=0, xdata=0x83468c) at defaults.c:1645
#17 0x00007fd41d0fd1b8 in call_resume_wind (stub=0xf189fc) at call-stub.c:2492
#18 0x00007fd41d104b5c in call_resume (stub=0xf189fc) at call-stub.c:2841
#19 0x00007fd40cd662e7 in iot_worker (data=0x674c60) at io-threads.c:214
#20 0x00007fd41c84d851 in start_thread () from /lib64/libpthread.so.0
#21 0x00007fd41c20190d in clone () from /lib64/libc.so.6
(gdb) 

I also met this problem in other scene, because the assert in _inode_path() as follow:
         if (!inode || uuid_is_null (inode->gfid)) {
   1186                 GF_ASSERT (0);
   1187                 gf_log_callingfn (THIS->name, GF_LOG_WARNING, "invalid inode");
   1188                 return -1;
   1189         }
   1190 

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


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info: