From a00fc1bc24d5654c1b6cedea711027d639185789 Mon Sep 17 00:00:00 2001
From: Vitaliy Filippov <vitalif@yourcmc.ru>
Date: Sat, 12 Jul 2025 01:35:16 +0300
Subject: [PATCH] Fsync after stabilizing

---
 src/blockstore/blockstore_impl.h     |  1 +
 src/blockstore/blockstore_stable.cpp | 26 +++++++++-----------------
 src/blockstore/blockstore_sync.cpp   | 22 ++++++++++++++++------
 3 files changed, 26 insertions(+), 23 deletions(-)

diff --git a/src/blockstore/blockstore_impl.h b/src/blockstore/blockstore_impl.h
index f54a6aa4..8356f683 100644
--- a/src/blockstore/blockstore_impl.h
+++ b/src/blockstore/blockstore_impl.h
@@ -173,6 +173,7 @@ class blockstore_impl_t: public blockstore_i
 
     // Sync
     int continue_sync(blockstore_op_t *op);
+    int do_sync(blockstore_op_t *op, int base_state);
 
     // Stabilize
     int dequeue_stable(blockstore_op_t *op);
diff --git a/src/blockstore/blockstore_stable.cpp b/src/blockstore/blockstore_stable.cpp
index 901e8bcb..797ea8e0 100644
--- a/src/blockstore/blockstore_stable.cpp
+++ b/src/blockstore/blockstore_stable.cpp
@@ -66,28 +66,20 @@ resume_1:
         priv->op_state = 1;
         return 0;
     }
-resume_2:
-    if (!dsk.disable_meta_fsync)
-    {
-        BS_SUBMIT_GET_SQE(sqe, data);
-        io_uring_prep_fsync(sqe, dsk.meta_fd, IORING_FSYNC_DATASYNC);
-        data->iov = { 0 };
-        data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
-        priv->pending_ops++;
-    }
-resume_3:
-    if (priv->pending_ops > 0)
-    {
-        priv->op_state = 3;
-        return 0;
-    }
-resume_4:
     // Mark writes as completed to allow compaction
-    // FIXME: Also mark as fsynced
     for (uint64_t lsn = priv->lsn; lsn <= priv->to_lsn; lsn++)
     {
         heap->mark_lsn_completed(lsn);
     }
+    // Fsync, just because our semantics imply that commit (stabilize) is immediately fsynced
+resume_2:
+resume_3:
+resume_4:
+    int res = do_sync(op, 2);
+    if (res != 2)
+    {
+        return res;
+    }
     // Done. Don't touch op->retval - if anything resulted in ENOENT, return it as is
     FINISH_OP(op);
     return 2;
diff --git a/src/blockstore/blockstore_sync.cpp b/src/blockstore/blockstore_sync.cpp
index e293c0e2..72baf4f1 100644
--- a/src/blockstore/blockstore_sync.cpp
+++ b/src/blockstore/blockstore_sync.cpp
@@ -6,7 +6,21 @@
 
 int blockstore_impl_t::continue_sync(blockstore_op_t *op)
 {
-    int op_state = PRIV(op)->op_state;
+    if (!PRIV(op)->op_state)
+    {
+        op->retval = 0;
+    }
+    int res = do_sync(op, 0);
+    if (res == 2)
+    {
+        FINISH_OP(op);
+    }
+    return res;
+}
+
+int blockstore_impl_t::do_sync(blockstore_op_t *op, int base_state)
+{
+    int op_state = PRIV(op)->op_state - base_state;
     if (op_state == 1) goto resume_1;
     if (op_state == 2) goto resume_2;
     assert(!op_state);
@@ -19,8 +33,6 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
     {
         // We can return immediately because sync is only dequeued after all previous writes
         unsynced_big_write_count = unsynced_small_write_count = 0;
-        op->retval = 0;
-        FINISH_OP(op);
         return 2;
     }
     PRIV(op)->lsn = heap->get_completed_lsn();
@@ -48,12 +60,10 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
 resume_1:
     if (PRIV(op)->pending_ops > 0)
     {
-        PRIV(op)->op_state = 1;
+        PRIV(op)->op_state = base_state+1;
         return 1;
     }
 resume_2:
     heap->mark_lsn_fsynced(PRIV(op)->lsn);
-    op->retval = 0;
-    FINISH_OP(op);
     return 2;
 }