From: Sage Weil <sage@redhat.com>
Date: Sun, 18 Feb 2018 02:27:30 +0000 (-0600)
Subject: osd: fix pg removal vs _process race
X-Git-Tag: v13.1.0~390^2~54
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=3a0b197cd1682f2fc68b1b962d577ec44cba5944;p=ceph.git

osd: fix pg removal vs _process race

Signed-off-by: Sage Weil <sage@redhat.com>
---

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 85464d29a1622..32cf018d25a94 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -9593,10 +9593,22 @@ void OSD::ShardedOpWQ::_process(uint32_t thread_index, heartbeat_handle_d *hb)
   sdata->sdata_op_ordering_lock.Lock();
 
   auto q = sdata->pg_slots.find(token);
-  assert(q != sdata->pg_slots.end());
+  if (q == sdata->pg_slots.end()) {
+    // this can happen if we race with pg removal.
+    dout(20) << __func__ << " slot " << token << " no longer there" << dendl;
+    pg->unlock();
+    sdata->sdata_op_ordering_lock.Unlock();
+    return;
+  }
   auto *slot = q->second.get();
   --slot->num_running;
 
+  if (pg && !slot->pg) {
+    // this can happen if we race with pg removal.
+    dout(20) << __func__ << " slot " << token << " no longer attached" << dendl;
+    pg->unlock();
+    pg = nullptr;
+  }
   if (slot->to_process.empty()) {
     // raced with wake_pg_waiters or consume_map
     dout(20) << __func__ << " " << token