osdc/Journaler: make sure flush() writes enough data

If _do_flush() fails to write all data because of prezere, _do_flush()
should be called again when prezero finishes. Otherwise flush waiters
may sleep forever.

Fixes: http://tracker.ceph.com/issues/22824
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 94a2076b9de9e8ffb73f3d5c873bdb87a0429e65)

diff --git a/src/osdc/Journaler.cc b/src/osdc/Journaler.cc
index 4217ab91ea1f6c86c18d0f7fbab232e2d18d07d0..94386101a1d34326216d4ed0fbada085c692d2a5 100644 (file)
--- a/src/osdc/Journaler.cc
+++ b/src/osdc/Journaler.cc
@@ -633,19 +633,16 @@ 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 +850,8 @@ 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);
     }
   } else {
     pending_zero.insert(start, len);

diff --git a/src/osdc/Journaler.h b/src/osdc/Journaler.h
index 1c3028045e7f0acc5745de81975a31f5ea8e2339..d65419a652be45bdf675bf799f81ca3306484bb8 100644 (file)
--- a/src/osdc/Journaler.h
+++ b/src/osdc/Journaler.h
@@ -310,7 +310,7 @@ private:
   // protect write_buf from bufferlist _len overflow 
   Throttle write_buf_throttle;
 
-  bool waiting_for_zero;
+  uint64_t waiting_for_zero_pos;
   interval_set<uint64_t> pending_zero;  // non-contig bits we've zeroed
   std::map<uint64_t, uint64_t> pending_safe; // flush_pos -> safe_pos
   // when safe through given offset
@@ -407,7 +407,7 @@ public:
     prezeroing_pos(0), prezero_pos(0), write_pos(0), flush_pos(0),
     safe_pos(0), next_safe_pos(0),
     write_buf_throttle(cct, "write_buf_throttle", UINT_MAX - (UINT_MAX >> 3)),
-    waiting_for_zero(false),
+    waiting_for_zero_pos(0),
     read_pos(0), requested_pos(0), received_pos(0),
     fetch_len(0), temp_fetch_len(0),
     on_readable(0), on_write_error(NULL), called_write_error(false),
@@ -444,7 +444,7 @@ public:
     expire_pos = 0;
     trimming_pos = 0;
     trimmed_pos = 0;
-    waiting_for_zero = false;
+    waiting_for_zero_pos = 0;
   }
 
   // Asynchronous operations