Commit a9363909 authored Dec 14, 2017 by NeilBrown Committed by Greg Kroah-Hartman Dec 19, 2017

staging: lustre: lnet: Fix recent breakage from list_for_each conversion

Commit 8e55b6fd ("staging: lustre: lnet: replace list_for_each
with list_for_each_entry") was intended to be an idempotent change,
but actually broke the behavior of ksocknal_add_peer() causing mounts to fail.
The fact that it caused an existing "route2 = NULL;" to become
redundant could have been a clue. The fact that the loop body
set the new loop variable to NULL might also have been a clue

The original code relied on "route2" being NULL if nothing was found.
The new code would always set route2 to a non-NULL value if the list
was empty, and would likely crash if the list was not empty.

Restore correct functionality by using code-flow rather the value of
"route2" to determine whether to use on old route, or to add a new one.

Fixes: 8e55b6fd ("staging: lustre: lnet: replace list_for_each with list_for_each_entry")
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

parent 1291a0d5

drivers/staging/lustre/lnet/klnds/socklnd/socklnd.c

+10 −13

Original line number	Diff line number	Diff line
		@@ -487,21 +487,18 @@ ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ipaddr,
		ksocknal_nid2peerlist(id.nid));
		}

		route2 = NULL;
		list_for_each_entry(route2, &peer->ksnp_routes, ksnr_list) {
		if (route2->ksnr_ipaddr == ipaddr)
		break;

		route2 = NULL;
		}
		if (!route2) {
		ksocknal_add_route_locked(peer, route);
		route->ksnr_share_count++;
		} else {
		if (route2->ksnr_ipaddr == ipaddr) {
		/* Route already exists, use the old one */
		ksocknal_route_decref(route);
		route2->ksnr_share_count++;
		goto out;
		}

		}
		/* Route doesn't already exist, add the new one */
		ksocknal_add_route_locked(peer, route);
		route->ksnr_share_count++;
		out:
		write_unlock_bh(&ksocknal_data.ksnd_global_lock);

		return 0;