From: Sage Weil <sage@redhat.com>
Date: Wed, 26 Feb 2020 00:08:22 +0000 (-0600)
Subject: mon/Monitor: fail forwarded tell commands
X-Git-Tag: v15.1.1~204^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=e6e3fa84d7ef1b4e31954cc787227977a301d6b8;p=ceph.git

mon/Monitor: fail forwarded tell commands

If we have a mix of pre-octopus and octopus+ mons, the old mon may forward
a tell command.  If we try to process it via the asok/tell queue, we will
assert out when we try to send a reply message (see AnonConnection
impl).  Instead, just reply with an error, and a hint that the command
is failing becuase the cluster is mid-upgrade.

Fixes: https://tracker.ceph.com/issues/44297
Signed-off-by: Sage Weil <sage@redhat.com>
---

diff --git a/src/mon/Monitor.cc b/src/mon/Monitor.cc
index 49acde967c5c..13fc886ab15a 100644
--- a/src/mon/Monitor.cc
+++ b/src/mon/Monitor.cc
@@ -3235,6 +3235,19 @@ void Monitor::handle_command(MonOpRequestRef op)
        prefix == "smart" ||
        prefix == "mon_status" ||
        prefix == "heap")) {
+    if (m->get_connection()->get_messenger() == 0) {
+      // Prior to octopus, monitors might forward these messages
+      // around. that was broken at baseline, and if we try to process
+      // this message now, it will assert out when we try to send a
+      // message in reply from the asok/tell worker (see
+      // AnonConnection).  Just reply with an error.
+      dout(5) << __func__ << " failing forwarded command from a (presumably) "
+	      << "pre-octopus peer" << dendl;
+      reply_command(
+	op, -EBUSY,
+	"failing forwarded tell command in mixed-version mon cluster", 0);
+      return;
+    }
     dout(5) << __func__ << " passing command to tell/asok" << dendl;
     cct->get_admin_socket()->queue_tell_command(m);
     return;