Bug 1395510

+++ This bug was initially created as a clone of Bug #1395261 +++

+++ This bug was initially created as a clone of Bug #1394752 +++

On my systemic setup, 
I am seeing lot of error messages on my clients as below


[2016-11-14 02:43:50.274000] E [snapview-client.c:283:gf_svc_lookup_cbk] 0-sysvol-snapview-client: Lookup failed on normal graph with error Transport endpoint is not connected
[2016-11-14 02:43:50.275390] E [dht-helper.c:1666:dht_inode_ctx_time_update] (-->/usr/lib64/glusterfs/3.8.4/xlator/cluster/replicate.so(+0x5d75c) [0x7f2a4ee4175c] -->/usr/lib64/glusterfs/3.8.4/xlator/cluster/distribute.so(+0x4623c) [0x7f2a4eba023c] -->/usr/lib64/glusterfs/3.8.4/xlator/cluster/distribute.so(+0x99b0) [0x7f2a4eb639b0] ) 0-sysvol-dht: invalid argument: inode [Invalid argument]


I see these message repeating about every 20 min in bulk

--- Additional comment from Nithya Balachandran on 2016-11-15 09:40:06 EST ---

Steps to see the issue:

1. Create a pure dist volume with 2 bricks.
2. Fuse mount the volume.
3. Create a directory dir1 and cd into it.
4. Kill one of the brick processes (kill -9)
5. Try to create directories inside dir1


The mount log has the messages:

[2016-11-15 14:06:39.738234] E [dht-helper.c:1666:dht_inode_ctx_time_update] (-->/usr/local/lib/glusterfs/3.10dev/xlator/protocol/c    lient.so(+0x2f750) [0x7f1fa58a8750] -->/usr/local/lib/glusterfs/3.10dev/xlator/cluster/distribute.so(+0x39ffe) [0x7f1fa5604ffe] -->    /usr/local/lib/glusterfs/3.10dev/xlator/cluster/distribute.so(+0xde27) [0x7f1fa55d8e27] ) 0-time-dht: invalid argument: inode [Invalid argument]




RCA:

The function dht_lookup_dir_cbk () does not check if the lookup succeeded on at least one subvolume before attempting to set the inode ctx.


        this_call_cnt = dht_frame_return (frame);

        if (is_last_call (this_call_cnt)) {
                if (local->need_selfheal) {
                        local->need_selfheal = 0;
                        dht_lookup_everywhere (frame, this, &local->loc);
                        return 0;
                }

                if (local->op_ret == 0) {
                        ret = dht_layout_normalize (this, &local->loc, layout);

                        if (ret != 0) {
                                gf_msg_debug (this->name, 0,
                                              "fixing assignment on %s",
                                              local->loc.path);
                                goto selfheal;
                        }

                        dht_layout_set (this, local->inode, layout);
                }

                dht_inode_ctx_time_update (local->inode, this,
                                           &local->stbuf, 1); <-- local->inode is NULL here as the directory was not found on any brick.

--- Additional comment from Worker Ant on 2016-11-15 10:15:55 EST ---

REVIEW: http://review.gluster.org/15847 (cluster/dht: Check for null inode) posted (#1) for review on master by N Balachandran (nbalacha)

--- Additional comment from Worker Ant on 2016-11-15 22:42:37 EST ---

COMMIT: http://review.gluster.org/15847 committed in master by Atin Mukherjee (amukherj) 
------
commit 8313d53accaa22feb14d284fb91245be0a32e16e
Author: N Balachandran <nbalacha>
Date:   Tue Nov 15 20:40:08 2016 +0530

    cluster/dht: Check for null inode
    
    Check for NULL inode before attempting to
    set dht inode ctx.
    
    Change-Id: I7693c18445f138221d8417df5e95b118cedb818a
    BUG: 1395261
    Signed-off-by: N Balachandran <nbalacha>
    Reviewed-on: http://review.gluster.org/15847
    Smoke: Gluster Build System <jenkins.org>
    Reviewed-by: Shyamsundar Ranganathan <srangana>
    NetBSD-regression: NetBSD Build System <jenkins.org>
    CentOS-regression: Gluster Build System <jenkins.org>
    Reviewed-by: Atin Mukherjee <amukherj>