librbd: automatically flush IO after blocking write operations

This simplifies other state machines that previously had to flush IO
after blocking write operations.

Fixes: #13913
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

diff --git a/src/librbd/AioImageRequestWQ.cc b/src/librbd/AioImageRequestWQ.cc
index ff9d8e4a3e1fd3870c39fda994db0445986502c1..a888ad6e3dd6c38655a654a04257417ebea40088 100644 (file)
--- a/src/librbd/AioImageRequestWQ.cc
+++ b/src/librbd/AioImageRequestWQ.cc
@@ -165,12 +165,13 @@ void AioImageRequestWQ::block_writes(Context *on_blocked) {
     ++m_write_blockers;
     ldout(cct, 5) << __func__ << ": " << &m_image_ctx << ", "
                   << "num=" << m_write_blockers << dendl;
-    if (m_in_progress_writes > 0) {
+    if (!m_write_blocker_contexts.empty() || m_in_progress_writes > 0) {
       m_write_blocker_contexts.push_back(on_blocked);
       return;
     }
   }
-  on_blocked->complete(0);
+
+  m_image_ctx.op_work_queue->queue(on_blocked);
 }
 
 void AioImageRequestWQ::unblock_writes() {
@@ -230,7 +231,7 @@ void AioImageRequestWQ::process(AioImageRequest *req) {
     req->send();
   }
 
-  Contexts contexts;
+  bool writes_blocked = false;
   {
     Mutex::Locker locker(m_lock);
     if (req->is_write_op()) {
@@ -238,16 +239,17 @@ void AioImageRequestWQ::process(AioImageRequest *req) {
       --m_queued_writes;
 
       assert(m_in_progress_writes > 0);
-      if (--m_in_progress_writes == 0) {
-        contexts.swap(m_write_blocker_contexts);
+      if (--m_in_progress_writes == 0 && !m_write_blocker_contexts.empty()) {
+        writes_blocked = true;
       }
     }
   }
-  delete req;
 
-  for (Contexts::iterator it = contexts.begin(); it != contexts.end(); ++it) {
-    (*it)->complete(0);
+  if (writes_blocked) {
+    RWLock::RLocker owner_locker(m_image_ctx.owner_lock);
+    m_image_ctx.flush(new C_BlockedWrites(this));
   }
+  delete req;
 }
 
 bool AioImageRequestWQ::is_journal_required() const {
@@ -313,4 +315,16 @@ void AioImageRequestWQ::handle_lock_updated(
   }
 }
 
+void AioImageRequestWQ::handle_blocked_writes(int r) {
+  Contexts contexts;
+  {
+    Mutex::Locker locker(m_lock);
+    contexts.swap(m_write_blocker_contexts);
+  }
+
+  for (auto ctx : contexts) {
+    ctx->complete(0);
+  }
+}
+
 } // namespace librbd

diff --git a/src/librbd/AioImageRequestWQ.h b/src/librbd/AioImageRequestWQ.h
index f668c9f03f12111f3c38a86c6e74d6a00839292d..db2757f8c9564f43227b6f3648e01a8805b34d84 100644 (file)
--- a/src/librbd/AioImageRequestWQ.h
+++ b/src/librbd/AioImageRequestWQ.h
@@ -71,6 +71,17 @@ private:
     }
   };
 
+  struct C_BlockedWrites : public Context {
+    AioImageRequestWQ *aio_work_queue;
+    C_BlockedWrites(AioImageRequestWQ *_aio_work_queue)
+      : aio_work_queue(_aio_work_queue) {
+    }
+
+    virtual void finish(int r) {
+      aio_work_queue->handle_blocked_writes(r);
+    }
+  };
+
   ImageCtx &m_image_ctx;
   mutable Mutex m_lock;
   Contexts m_write_blocker_contexts;
@@ -86,6 +97,7 @@ private:
   void queue(AioImageRequest *req);
 
   void handle_lock_updated(ImageWatcher::LockUpdateState state);
+  void handle_blocked_writes(int r);
 };
 
 } // namespace librbd

diff --git a/src/librbd/ImageWatcher.cc b/src/librbd/ImageWatcher.cc
index 30dca71acafc2fa6b693f3b73a3c7339b102b105..75a82bd0f7270a741d9ab01672420b9fb1a28fac 100644 (file)
--- a/src/librbd/ImageWatcher.cc
+++ b/src/librbd/ImageWatcher.cc
@@ -2,6 +2,7 @@
 // vim: ts=8 sw=2 smarttab
 #include "librbd/ImageWatcher.h"
 #include "librbd/AioCompletion.h"
+#include "librbd/AioImageRequestWQ.h"
 #include "librbd/ImageCtx.h"
 #include "librbd/internal.h"
 #include "librbd/ObjectMap.h"
@@ -392,7 +393,6 @@ int ImageWatcher::release_lock()
 
   // ensure all maint operations are canceled
   m_image_ctx.cancel_async_requests();
-  m_image_ctx.flush_async_operations();
 
   int r;
   {
@@ -402,12 +402,8 @@ int ImageWatcher::release_lock()
     // lock is being released
     notify_listeners_updated_lock(LOCK_UPDATE_STATE_RELEASING);
 
-    RWLock::WLocker md_locker(m_image_ctx.md_lock);
-    r = m_image_ctx.flush();
-    if (r < 0) {
-      lderr(cct) << this << " failed to flush: " << cpp_strerror(r) << dendl;
-      goto err_cancel_unlock;
-    }
+    // AioImageRequestWQ will have blocked writes / flushed IO by this point
+    assert(m_image_ctx.aio_work_queue->writes_blocked());
   }
 
   m_image_ctx.owner_lock.get_write();
@@ -430,13 +426,6 @@ int ImageWatcher::release_lock()
   }
 
   return 0;
-
-err_cancel_unlock:
-  m_image_ctx.owner_lock.get_write();
-  if (m_lock_owner_state == LOCK_OWNER_STATE_RELEASING) {
-    m_lock_owner_state = LOCK_OWNER_STATE_LOCKED;
-  }
-  return r;
 }
 
 void ImageWatcher::assert_header_locked(librados::ObjectWriteOperation *op) {

diff --git a/src/librbd/operation/SnapshotCreateRequest.cc b/src/librbd/operation/SnapshotCreateRequest.cc
index a046c0812353f428ba068bf5476c1ee32e4d4a7e..402ce36b48173aeb209baf5202deaa412b1117b9 100644 (file)
--- a/src/librbd/operation/SnapshotCreateRequest.cc
+++ b/src/librbd/operation/SnapshotCreateRequest.cc
@@ -27,9 +27,6 @@ std::ostream& operator<<(std::ostream& os,
   case SnapshotCreateRequest::STATE_SUSPEND_AIO:
     os << "SUSPEND_AIO";
     break;
-  case SnapshotCreateRequest::STATE_FLUSH_AIO:
-    os << "FLUSH_AIO";
-    break;
   case SnapshotCreateRequest::STATE_ALLOCATE_SNAP_ID:
     os << "ALLOCATE_SNAP_ID";
     break;
@@ -86,9 +83,6 @@ bool SnapshotCreateRequest::should_complete(int r) {
     send_suspend_aio();
     break;
   case STATE_SUSPEND_AIO:
-    send_flush_aio();
-    break;
-  case STATE_FLUSH_AIO:
     send_allocate_snap_id();
     break;
   case STATE_ALLOCATE_SNAP_ID:
@@ -156,17 +150,6 @@ void SnapshotCreateRequest::send_suspend_aio() {
   m_image_ctx.aio_work_queue->block_writes(create_async_callback_context());
 }
 
-void SnapshotCreateRequest::send_flush_aio() {
-  assert(m_image_ctx.owner_lock.is_locked());
-
-  CephContext *cct = m_image_ctx.cct;
-  ldout(cct, 5) << this << " " << __func__ << dendl;
-  m_state = STATE_FLUSH_AIO;
-
-  // can issue a re-entrant callback if no IO to flush
-  m_image_ctx.flush(create_async_callback_context());
-}
-
 void SnapshotCreateRequest::send_allocate_snap_id() {
   assert(m_image_ctx.owner_lock.is_locked());
 

diff --git a/src/librbd/operation/SnapshotCreateRequest.h b/src/librbd/operation/SnapshotCreateRequest.h
index 21294423a122cabc082fa578c5f628ca765424a8..5db8441fef8f1fabffdfa0037471e45424a54c09 100644 (file)
--- a/src/librbd/operation/SnapshotCreateRequest.h
+++ b/src/librbd/operation/SnapshotCreateRequest.h
@@ -30,10 +30,7 @@ public:
    *           STATE_SUSPEND_REQUESTS
    *               |
    *               v
-   *           STATE_SUSPEND_AIO
-   *               |
-   *               v
-   *           STATE_FLUSH_AIO * * * * * * * * * * * * * *
+   *           STATE_SUSPEND_AIO * * * * * * * * * * * * *
    *               |                                     *
    *   (retry)     v                                     *
    *   . . . > STATE_ALLOCATE_SNAP_ID  * *               *
@@ -61,7 +58,6 @@ public:
   enum State {
     STATE_SUSPEND_REQUESTS,
     STATE_SUSPEND_AIO,
-    STATE_FLUSH_AIO,
     STATE_ALLOCATE_SNAP_ID,
     STATE_CREATE_SNAP,
     STATE_CREATE_OBJECT_MAP,
@@ -112,7 +108,6 @@ private:
 
   void send_suspend_requests();
   void send_suspend_aio();
-  void send_flush_aio();
   void send_allocate_snap_id();
   void send_create_snap();
   bool send_create_object_map();