update source to 12.2.11

[ceph.git] / ceph / src / osdc / Journaler.cc

diff --git a/ceph/src/osdc/Journaler.cc b/ceph/src/osdc/Journaler.cc
index 4217ab91ea1f6c86c18d0f7fbab232e2d18d07d0..0b5663b81fa7daffddc5e4bc5aff1645c9c5d00c 100644 (file)
--- a/ceph/src/osdc/Journaler.cc
+++ b/ceph/src/osdc/Journaler.cc
@@ -532,11 +532,12 @@ void Journaler::_finish_flush(int r, uint64_t start, ceph::real_time stamp)
   // adjust safe_pos
   auto it = pending_safe.find(start);
   assert(it != pending_safe.end());
+  uint64_t min_next_safe_pos = pending_safe.begin()->second;
   pending_safe.erase(it);
   if (pending_safe.empty())
     safe_pos = next_safe_pos;
   else
-    safe_pos = pending_safe.begin()->second;
+    safe_pos = min_next_safe_pos;
 
   ldout(cct, 10) << "_finish_flush safe from " << start
                 << ", pending_safe " << pending_safe
@@ -633,19 +634,16 @@ void Journaler::_do_flush(unsigned amount)
       ldout(cct, 10) << "_do_flush wanted to do " << flush_pos << "~" << len
                     << " already too close to prezero_pos " << prezero_pos
                     << ", zeroing first" << dendl;
-      waiting_for_zero = true;
+      waiting_for_zero_pos = flush_pos + len;
       return;
     }
     if (static_cast<uint64_t>(newlen) < len) {
       ldout(cct, 10) << "_do_flush wanted to do " << flush_pos << "~" << len
                     << " but hit prezero_pos " << prezero_pos
                     << ", will do " << flush_pos << "~" << newlen << dendl;
+      waiting_for_zero_pos = flush_pos + len;
       len = newlen;
-    } else {
-      waiting_for_zero = false;
     }
-  } else {
-    waiting_for_zero = false;
   }
   ldout(cct, 10) << "_do_flush flushing " << flush_pos << "~" << len << dendl;
 
@@ -853,8 +851,15 @@ void Journaler::_finish_prezero(int r, uint64_t start, uint64_t len)
       pending_zero.erase(b);
     }
 
-    if (waiting_for_zero) {
-      _do_flush();
+    if (waiting_for_zero_pos > flush_pos) {
+      _do_flush(waiting_for_zero_pos - flush_pos);
+    }
+
+    if (prezero_pos == prezeroing_pos &&
+       !waitfor_prezero.empty()) {
+      list<Context*> ls;
+      ls.swap(waitfor_prezero);
+      finish_contexts(cct, ls, 0);
     }
   } else {
     pending_zero.insert(start, len);
@@ -865,6 +870,17 @@ void Journaler::_finish_prezero(int r, uint64_t start, uint64_t len)
                 << dendl;
 }
 
+void Journaler::wait_for_prezero(Context *onfinish)
+{
+  assert(onfinish);
+  lock_guard l(lock);
+
+  if (prezero_pos == prezeroing_pos) {
+    finisher->queue(onfinish, 0);
+    return;
+  }
+  waitfor_prezero.push_back(wrap_finisher(onfinish));
+}
 
 
 /***************** READING *******************/
@@ -1252,6 +1268,13 @@ bool Journaler::try_read_entry(bufferlist& bl)
 
   // prefetch?
   _prefetch();
+
+  // If bufferlist consists of discontiguous memory, decoding types whose
+  // denc_traits needs contiguous memory is inefficient. The bufferlist may
+  // get copied to temporary memory multiple times (copy_shallow() in
+  // src/include/denc.h actually does deep copy)
+  if (bl.get_num_buffers() > 1)
+    bl.rebuild();
   return true;
 }