Merge remote-tracking branch 'origin/hammer'

src/librbd/AioRequest.cc

@@ -259,8 +259,6 @@ namespace librbd {
       m_state = LIBRBD_AIO_WRITE_GUARD;
       m_write.assert_exists();
       ldout(m_ictx->cct, 20) << __func__ << " guarding write" << dendl;
-    } else {
-      m_state = LIBRBD_AIO_WRITE_FLAT;
     }
   }
 
@@ -480,6 +478,7 @@ namespace librbd {
     ldout(m_ictx->cct, 20) << "send_write " << this << " " << m_oid << " "
 			   << m_object_off << "~" << m_object_len << dendl;
 
+    m_state = LIBRBD_AIO_WRITE_FLAT;
     guard_write();
     add_write_ops(&m_write);
     assert(m_write.size() != 0);

src/librbd/ImageWatcher.cc

@@ -131,13 +131,22 @@ bool ImageWatcher::has_pending_aio_operations() {
 }
 
 void ImageWatcher::flush_aio_operations() {
+  C_SaferCond *ctx = new C_SaferCond();
+  flush_aio_operations(ctx);
+  ctx->wait();
+}
+
+void ImageWatcher::flush_aio_operations(Context *ctx) {
   Mutex::Locker l(m_aio_request_lock);
-  while (m_retrying_aio_requests || !m_aio_requests.empty()) {
+  if (!m_retrying_aio_requests && m_aio_requests.empty()) {
-    ldout(m_image_ctx.cct, 20)  << "flushing aio operations: "
+    ctx->complete(0);
-				<< "retrying=" << m_retrying_aio_requests << ","
+    return;
-				<< "count=" << m_aio_requests.size() << dendl;
-    m_aio_request_cond.Wait(m_aio_request_lock);
   }
+
+  ldout(m_image_ctx.cct, 20) << "pending flush: " << ctx << " "
+			     << "retrying=" << m_retrying_aio_requests << ", "
+			     << "count=" << m_aio_requests.size() << dendl;
+  m_aio_flush_contexts.push_back(ctx);
 }
 
 int ImageWatcher::try_lock() {
@@ -545,23 +554,13 @@ void ImageWatcher::retry_aio_requests() {
 
   Mutex::Locker l(m_aio_request_lock);
   m_retrying_aio_requests = false;
-  m_aio_request_cond.Signal();
+  while (!m_aio_flush_contexts.empty()) {
-}
+    Context *flush_ctx = m_aio_flush_contexts.front();
+    m_aio_flush_contexts.pop_front();
 
-void ImageWatcher::cancel_aio_requests(int result) {
+    ldout(m_image_ctx.cct, 20) << "completed flush: " << flush_ctx << dendl;
-  Mutex::Locker l(m_aio_request_lock);
+    flush_ctx->complete(0);
-  for (std::vector<AioRequest>::iterator iter = m_aio_requests.begin();
-       iter != m_aio_requests.end(); ++iter) {
-    AioCompletion *c = iter->second;
-    c->get();
-    c->lock.Lock();
-    c->rval = result;
-    c->lock.Unlock();
-    c->finish_adding_requests(m_image_ctx.cct);
-    c->put();
   }
-  m_aio_requests.clear();
-  m_aio_request_cond.Signal();
 }
 
 void ImageWatcher::cancel_async_requests(int result) {

src/librbd/ImageWatcher.h

@@ -3,9 +3,9 @@
 #ifndef CEPH_LIBRBD_IMAGE_WATCHER_H
 #define CEPH_LIBRBD_IMAGE_WATCHER_H
 
-#include "common/Cond.h"
 #include "common/Mutex.h"
 #include "common/RWLock.h"
+#include "include/Context.h"
 #include "include/rados/librados.hpp"
 #include "include/rbd/librbd.hpp"
 #include <set>
@@ -16,7 +16,6 @@
 #include "include/assert.h"
 
 class entity_name_t;
-class Context;
 class Finisher;
 class SafeTimer;
 
@@ -60,6 +59,7 @@ namespace librbd {
 
     bool has_pending_aio_operations();
     void flush_aio_operations();
+    void flush_aio_operations(Context *ctx);
 
     int try_lock();
     int request_lock(const boost::function<int(AioCompletion*)>& restart_op,
@@ -170,7 +170,7 @@ namespace librbd {
     std::set<RemoteAsyncRequest> m_async_progress;
 
     Mutex m_aio_request_lock;
-    Cond m_aio_request_cond;
+    std::list<Context *> m_aio_flush_contexts;
     std::vector<AioRequest> m_aio_requests;
     bool m_retrying_aio_requests;
     Context *m_retry_aio_context;
@@ -191,7 +191,6 @@ namespace librbd {
     void finalize_retry_aio_requests();
     void retry_aio_requests();
 
-    void cancel_aio_requests(int result);
     void cancel_async_requests(int result);
 
     uint64_t encode_async_request(bufferlist &bl);

src/librbd/internal.cc

@@ -3103,9 +3103,16 @@ reprotect_and_return_err:
     ictx->user_flushed();
 
     c->get();
-    c->add_request();
     c->init_time(ictx, AIO_TYPE_FLUSH);
+
+    if (ictx->image_watcher != NULL) {
+      C_AioWrite *flush_ctx = new C_AioWrite(cct, c);
+      c->add_request();
+      ictx->image_watcher->flush_aio_operations(flush_ctx);
+    }
+
     C_AioWrite *req_comp = new C_AioWrite(cct, c);
+    c->add_request();
     if (ictx->object_cacher) {
       ictx->flush_cache_aio(req_comp);
     } else {