diff --git a/src/blockstore.h b/src/blockstore.h
index 413d1166..265372c5 100644
--- a/src/blockstore.h
+++ b/src/blockstore.h
@@ -107,7 +107,7 @@ Input:
 - buf = pre-allocated obj_ver_id array <len> units long
 
 Output:
-- retval = 0 or negative error number (-EINVAL, -ENOENT if no such version or -EBUSY if not synced)
+- retval = 0 or negative error number (-ENOENT if no such version for stabilize)
 
 ## BS_OP_SYNC_STAB_ALL
 
diff --git a/src/blockstore_impl.cpp b/src/blockstore_impl.cpp
index 2f37de91..32952740 100644
--- a/src/blockstore_impl.cpp
+++ b/src/blockstore_impl.cpp
@@ -171,7 +171,7 @@ void blockstore_impl_t::loop()
                     // Can't submit SYNC before previous writes
                     continue;
                 }
-                wr_st = continue_sync(op, false);
+                wr_st = continue_sync(op);
                 if (wr_st != 2)
                 {
                     has_writes = wr_st > 0 ? 1 : 2;
@@ -371,13 +371,18 @@ void blockstore_impl_t::enqueue_op(blockstore_op_t *op)
         ringloop->set_immediate([op]() { std::function<void (blockstore_op_t*)>(op->callback)(op); });
         return;
     }
+    init_op(op);
+    submit_queue.push_back(op);
+    ringloop->wakeup();
+}
+
+void blockstore_impl_t::init_op(blockstore_op_t *op)
+{
     // Call constructor without allocating memory. We'll call destructor before returning op back
     new ((void*)op->private_data) blockstore_op_private_t;
     PRIV(op)->wait_for = 0;
     PRIV(op)->op_state = 0;
     PRIV(op)->pending_ops = 0;
-    submit_queue.push_back(op);
-    ringloop->wakeup();
 }
 
 static bool replace_stable(object_id oid, uint64_t version, int search_start, int search_end, obj_ver_id* list)
diff --git a/src/blockstore_impl.h b/src/blockstore_impl.h
index 6541fa22..c7d7d6ef 100644
--- a/src/blockstore_impl.h
+++ b/src/blockstore_impl.h
@@ -216,6 +216,11 @@ struct pool_shard_settings_t
     uint32_t pg_stripe_size;
 };
 
+#define STAB_SPLIT_DONE 1
+#define STAB_SPLIT_WAIT 2
+#define STAB_SPLIT_SYNC 3
+#define STAB_SPLIT_TODO 4
+
 class blockstore_impl_t
 {
     blockstore_disk_t dsk;
@@ -298,6 +303,7 @@ class blockstore_impl_t
     blockstore_init_journal* journal_init_reader;
 
     void check_wait(blockstore_op_t *op);
+    void init_op(blockstore_op_t *op);
 
     // Read
     int dequeue_read(blockstore_op_t *read_op);
@@ -317,7 +323,7 @@ class blockstore_impl_t
     void handle_write_event(ring_data_t *data, blockstore_op_t *op);
 
     // Sync
-    int continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync);
+    int continue_sync(blockstore_op_t *op);
     void ack_sync(blockstore_op_t *op);
 
     // Stabilize
@@ -325,6 +331,8 @@ class blockstore_impl_t
     int continue_stable(blockstore_op_t *op);
     void mark_stable(const obj_ver_id & ov, bool forget_dirty = false);
     void stabilize_object(object_id oid, uint64_t max_ver);
+    blockstore_op_t* selective_sync(blockstore_op_t *op);
+    int split_stab_op(blockstore_op_t *op, std::function<int(obj_ver_id v)> decider);
 
     // Rollback
     int dequeue_rollback(blockstore_op_t *op);
diff --git a/src/blockstore_rollback.cpp b/src/blockstore_rollback.cpp
index c5199ac8..d0db8a99 100644
--- a/src/blockstore_rollback.cpp
+++ b/src/blockstore_rollback.cpp
@@ -9,48 +9,39 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
     {
         return continue_rollback(op);
     }
-    obj_ver_id *v, *nv;
-    int i, todo = op->len;
-    for (i = 0, v = (obj_ver_id*)op->buf, nv = (obj_ver_id*)op->buf; i < op->len; i++, v++, nv++)
+    int r = split_stab_op(op, [this](obj_ver_id ov)
     {
-        if (nv != v)
-        {
-            *nv = *v;
-        }
         // Check that there are some versions greater than v->version (which may be zero),
         // check that they're unstable, synced, and not currently written to
         auto dirty_it = dirty_db.lower_bound((obj_ver_id){
-            .oid = v->oid,
+            .oid = ov.oid,
             .version = UINT64_MAX,
         });
         if (dirty_it == dirty_db.begin())
         {
-skip_ov:
             // Already rolled back, skip this object version
-            todo--;
-            nv--;
-            continue;
+            return STAB_SPLIT_DONE;
         }
         else
         {
             dirty_it--;
-            if (dirty_it->first.oid != v->oid || dirty_it->first.version < v->version)
+            if (dirty_it->first.oid != ov.oid || dirty_it->first.version < ov.version)
             {
-                goto skip_ov;
+                // Already rolled back, skip this object version
+                return STAB_SPLIT_DONE;
             }
-            while (dirty_it->first.oid == v->oid && dirty_it->first.version > v->version)
+            while (dirty_it->first.oid == ov.oid && dirty_it->first.version > ov.version)
             {
                 if (IS_IN_FLIGHT(dirty_it->second.state))
                 {
                     // Object write is still in progress. Wait until the write request completes
-                    return 0;
+                    return STAB_SPLIT_WAIT;
                 }
                 else if (!IS_SYNCED(dirty_it->second.state) ||
                     IS_STABLE(dirty_it->second.state))
                 {
-                    op->retval = -EBUSY;
-                    FINISH_OP(op);
-                    return 2;
+                    // Sync the object
+                    return STAB_SPLIT_SYNC;
                 }
                 if (dirty_it == dirty_db.begin())
                 {
@@ -58,19 +49,16 @@ skip_ov:
                 }
                 dirty_it--;
             }
+            return STAB_SPLIT_TODO;
         }
-    }
-    op->len = todo;
-    if (!todo)
+    });
+    if (r != 1)
     {
-        // Already rolled back
-        op->retval = 0;
-        FINISH_OP(op);
-        return 2;
+        return r;
     }
     // Check journal space
     blockstore_journal_check_t space_check(this);
-    if (!space_check.check_available(op, todo, sizeof(journal_entry_rollback), 0))
+    if (!space_check.check_available(op, op->len, sizeof(journal_entry_rollback), 0))
     {
         return 0;
     }
@@ -78,7 +66,8 @@ skip_ov:
     BS_SUBMIT_CHECK_SQES(space_check.sectors_to_write);
     // Prepare and submit journal entries
     int s = 0;
-    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    auto v = (obj_ver_id*)op->buf;
+    for (int i = 0; i < op->len; i++, v++)
     {
         if (!journal.entry_fits(sizeof(journal_entry_rollback)) &&
             journal.sector_info[journal.cur_sector].dirty)
diff --git a/src/blockstore_stable.cpp b/src/blockstore_stable.cpp
index 34be29ec..4b6f52c1 100644
--- a/src/blockstore_stable.cpp
+++ b/src/blockstore_stable.cpp
@@ -41,60 +41,309 @@
 // 4) after a while it takes his synced object list and sends stabilize requests
 //    to peers and to its own blockstore, thus freeing the old version
 
-int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
+struct ver_vector_t
 {
-    if (PRIV(op)->op_state)
+    obj_ver_id *items = NULL;
+    uint64_t alloc = 0, size = 0;
+};
+
+static void init_versions(ver_vector_t & vec, obj_ver_id *start, obj_ver_id *end, uint64_t len)
+{
+    if (!vec.items)
     {
-        return continue_stable(op);
+        vec.alloc = len;
+        vec.items = (obj_ver_id*)malloc_or_die(sizeof(obj_ver_id) * vec.alloc);
+        for (auto sv = start; sv < end; sv++)
+        {
+            vec.items[vec.size++] = *sv;
+        }
     }
+}
+
+static void append_version(ver_vector_t & vec, obj_ver_id ov)
+{
+    if (vec.size >= vec.alloc)
+    {
+        vec.alloc = !vec.alloc ? 4 : vec.alloc*2;
+        vec.items = (obj_ver_id*)realloc_or_die(vec.items, sizeof(obj_ver_id) * vec.alloc);
+    }
+    vec.items[vec.size++] = ov;
+}
+
+static bool check_unsynced(std::vector<obj_ver_id> & check, obj_ver_id ov, std::vector<obj_ver_id> & to, int *count)
+{
+    bool found = false;
+    int j = 0, k = 0;
+    while (j < check.size())
+    {
+        if (check[j] == ov)
+            found = true;
+        if (check[j].oid == ov.oid && check[j].version <= ov.version)
+        {
+            to.push_back(check[j++]);
+            if (count)
+                (*count)--;
+        }
+        else
+            check[k++] = check[j++];
+    }
+    check.resize(k);
+    return found;
+}
+
+blockstore_op_t* blockstore_impl_t::selective_sync(blockstore_op_t *op)
+{
+    unsynced_big_write_count -= unsynced_big_writes.size();
+    unsynced_big_writes.swap(PRIV(op)->sync_big_writes);
+    unsynced_big_write_count += unsynced_big_writes.size();
+    unsynced_small_writes.swap(PRIV(op)->sync_small_writes);
+    // Create a sync operation, insert into the end of the queue
+    // And move ourselves into the end too!
+    // Rather hacky but that's what we need...
+    blockstore_op_t *sync_op = new blockstore_op_t;
+    sync_op->opcode = BS_OP_SYNC;
+    sync_op->buf = NULL;
+    sync_op->callback = [this](blockstore_op_t *sync_op)
+    {
+        delete sync_op;
+    };
+    init_op(sync_op);
+    int sync_res = continue_sync(sync_op);
+    if (sync_res != 2)
+    {
+        // Put SYNC into the queue if it's not finished yet
+        submit_queue.push_back(sync_op);
+    }
+    // Restore unsynced_writes
+    unsynced_small_writes.swap(PRIV(op)->sync_small_writes);
+    unsynced_big_write_count -= unsynced_big_writes.size();
+    unsynced_big_writes.swap(PRIV(op)->sync_big_writes);
+    unsynced_big_write_count += unsynced_big_writes.size();
+    if (sync_res == 2)
+    {
+        // Sync is immediately completed
+        return NULL;
+    }
+    return sync_op;
+}
+
+// Returns: 2 = stop processing and dequeue, 0 = stop processing and do not dequeue, 1 = proceed with op itself
+int blockstore_impl_t::split_stab_op(blockstore_op_t *op, std::function<int(obj_ver_id v)> decider)
+{
+    bool add_sync = false;
+    ver_vector_t good_vers, bad_vers;
     obj_ver_id* v;
     int i, todo = 0;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
-        auto dirty_it = dirty_db.find(*v);
-        if (dirty_it == dirty_db.end())
+        int action = decider(*v);
+        if (action < 0)
         {
-            auto & clean_db = clean_db_shard(v->oid);
-            auto clean_it = clean_db.find(v->oid);
-            if (clean_it == clean_db.end() || clean_it->second.version < v->version)
+            // Rollback changes
+            for (auto & ov: PRIV(op)->sync_big_writes)
             {
-                // No such object version
-                op->retval = -ENOENT;
-                FINISH_OP(op);
-                return 2;
+                unsynced_big_writes.push_back(ov);
+                unsynced_big_write_count++;
             }
-            else
+            for (auto & ov: PRIV(op)->sync_small_writes)
             {
-                // Already stable
+                unsynced_small_writes.push_back(ov);
             }
-        }
-        else if (IS_IN_FLIGHT(dirty_it->second.state))
-        {
-            // Object write is still in progress. Wait until the write request completes
-            return 0;
-        }
-        else if (!IS_SYNCED(dirty_it->second.state))
-        {
-            // Object not synced yet. Caller must sync it first
-            op->retval = -EBUSY;
+            free(good_vers.items);
+            good_vers.items = NULL;
+            free(bad_vers.items);
+            bad_vers.items = NULL;
+            // Error
+            op->retval = action;
             FINISH_OP(op);
             return 2;
         }
-        else if (!IS_STABLE(dirty_it->second.state))
+        else if (action == STAB_SPLIT_DONE)
         {
+            // Already done
+            init_versions(good_vers, (obj_ver_id*)op->buf, v, op->len);
+        }
+        else if (action == STAB_SPLIT_WAIT)
+        {
+            // Already in progress, we just have to wait until it finishes
+            init_versions(good_vers, (obj_ver_id*)op->buf, v, op->len);
+            append_version(bad_vers, *v);
+        }
+        else if (action == STAB_SPLIT_SYNC)
+        {
+            // Needs a SYNC, we have to send a SYNC if not already in progress
+            //
+            // If the object is not present in unsynced_(big|small)_writes then
+            // it's currently being synced. If it's present then we can initiate
+            // its sync ourselves.
+            init_versions(good_vers, (obj_ver_id*)op->buf, v, op->len);
+            append_version(bad_vers, *v);
+            if (!add_sync)
+            {
+                PRIV(op)->sync_big_writes.clear();
+                PRIV(op)->sync_small_writes.clear();
+                add_sync = true;
+            }
+            check_unsynced(unsynced_small_writes, *v, PRIV(op)->sync_small_writes, NULL);
+            check_unsynced(unsynced_big_writes, *v, PRIV(op)->sync_big_writes, &unsynced_big_write_count);
+        }
+        else /* if (action == STAB_SPLIT_TODO) */
+        {
+            if (good_vers.items)
+            {
+                // If we're selecting versions then append it
+                // Main idea is that 99% of the time all versions passed to BS_OP_STABLE are synced
+                // And we don't want to select/allocate anything in that optimistic case
+                append_version(good_vers, *v);
+            }
             todo++;
         }
     }
-    if (!todo)
+    // In a pessimistic scenario, an operation may be split into 3:
+    // - Stabilize synced entries
+    // - Sync unsynced entries
+    // - Continue for unsynced entries after sync
+    add_sync = add_sync && (PRIV(op)->sync_big_writes.size() || PRIV(op)->sync_small_writes.size());
+    if (!todo && !bad_vers.size)
     {
         // Already stable
         op->retval = 0;
         FINISH_OP(op);
         return 2;
     }
+    op->retval = 0;
+    if (!todo && !add_sync)
+    {
+        // Only wait for inflight writes or current in-progress syncs
+        return 0;
+    }
+    blockstore_op_t *sync_op = NULL, *split_stab_op = NULL;
+    if (add_sync)
+    {
+        // Initiate a selective sync for PRIV(op)->sync_(big|small)_writes
+        sync_op = selective_sync(op);
+    }
+    if (bad_vers.size)
+    {
+        // Split part of the request into a separate operation
+        split_stab_op = new blockstore_op_t;
+        split_stab_op->opcode = op->opcode;
+        split_stab_op->buf = bad_vers.items;
+        split_stab_op->len = bad_vers.size;
+        init_op(split_stab_op);
+        submit_queue.push_back(split_stab_op);
+    }
+    if (sync_op || split_stab_op || good_vers.items)
+    {
+        void *orig_buf = op->buf;
+        if (good_vers.items)
+        {
+            op->buf = good_vers.items;
+            op->len = good_vers.size;
+        }
+        // Make a wrapped callback
+        int *split_op_counter = (int*)malloc_or_die(sizeof(int));
+        *split_op_counter = (sync_op ? 1 : 0) + (split_stab_op ? 1 : 0) + (todo ? 1 : 0);
+        auto cb = [this, op, good_items = good_vers.items,
+            bad_items = bad_vers.items, split_op_counter,
+            orig_buf, real_cb = op->callback](blockstore_op_t *split_op)
+        {
+            if (split_op->retval != 0)
+                op->retval = split_op->retval;
+            (*split_op_counter)--;
+            assert((*split_op_counter) >= 0);
+            if (op != split_op)
+                delete split_op;
+            if (!*split_op_counter)
+            {
+                free(good_items);
+                free(bad_items);
+                free(split_op_counter);
+                op->buf = orig_buf;
+                real_cb(op);
+            }
+        };
+        if (sync_op)
+        {
+            sync_op->callback = cb;
+        }
+        if (split_stab_op)
+        {
+            split_stab_op->callback = cb;
+        }
+        op->callback = cb;
+    }
+    if (!todo)
+    {
+        // All work is postponed
+        op->callback = NULL;
+        return 2;
+    }
+    return 1;
+}
+
+int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state)
+    {
+        return continue_stable(op);
+    }
+    int r = split_stab_op(op, [this](obj_ver_id ov)
+    {
+        auto dirty_it = dirty_db.find(ov);
+        if (dirty_it == dirty_db.end())
+        {
+            auto & clean_db = clean_db_shard(ov.oid);
+            auto clean_it = clean_db.find(ov.oid);
+            if (clean_it == clean_db.end() || clean_it->second.version < ov.version)
+            {
+                // No such object version
+                printf("Error: %lx:%lx v%lu not found while stabilizing\n", ov.oid.inode, ov.oid.stripe, ov.version);
+                return -ENOENT;
+            }
+            else
+            {
+                // Already stable
+                return STAB_SPLIT_DONE;
+            }
+        }
+        else if (IS_IN_FLIGHT(dirty_it->second.state))
+        {
+            // Object write is still in progress. Wait until the write request completes
+            return STAB_SPLIT_WAIT;
+        }
+        else if (!IS_SYNCED(dirty_it->second.state))
+        {
+            // Object not synced yet - sync it
+            // In previous versions we returned EBUSY here and required
+            // the caller (OSD) to issue a global sync first. But a global sync
+            // waits for all writes in the queue including inflight writes. And
+            // inflight writes may themselves be blocked by unstable writes being
+            // still present in the journal and not flushed away from it.
+            // So we must sync specific objects here.
+            //
+            // Even more, we have to process "stabilize" request in parts. That is,
+            // we must stabilize all objects which are already synced. Otherwise
+            // they may block objects which are NOT synced yet.
+            return STAB_SPLIT_SYNC;
+        }
+        else if (IS_STABLE(dirty_it->second.state))
+        {
+            // Already stable
+            return STAB_SPLIT_DONE;
+        }
+        else
+        {
+            return STAB_SPLIT_TODO;
+        }
+    });
+    if (r != 1)
+    {
+        return r;
+    }
     // Check journal space
     blockstore_journal_check_t space_check(this);
-    if (!space_check.check_available(op, todo, sizeof(journal_entry_stable), 0))
+    if (!space_check.check_available(op, op->len, sizeof(journal_entry_stable), 0))
     {
         return 0;
     }
@@ -102,9 +351,9 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     BS_SUBMIT_CHECK_SQES(space_check.sectors_to_write);
     // Prepare and submit journal entries
     int s = 0;
-    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    auto v = (obj_ver_id*)op->buf;
+    for (int i = 0; i < op->len; i++, v++)
     {
-        // FIXME: Only stabilize versions that aren't stable yet
         if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
             journal.sector_info[journal.cur_sector].dirty)
         {
diff --git a/src/blockstore_sync.cpp b/src/blockstore_sync.cpp
index ddacff16..85706091 100644
--- a/src/blockstore_sync.cpp
+++ b/src/blockstore_sync.cpp
@@ -12,7 +12,7 @@
 #define SYNC_JOURNAL_SYNC_SENT 7
 #define SYNC_DONE 8
 
-int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync)
+int blockstore_impl_t::continue_sync(blockstore_op_t *op)
 {
     if (immediate_commit == IMMEDIATE_ALL)
     {
@@ -145,7 +145,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
             PRIV(op)->op_state = SYNC_DONE;
         }
     }
-    if (PRIV(op)->op_state == SYNC_DONE && !queue_has_in_progress_sync)
+    if (PRIV(op)->op_state == SYNC_DONE)
     {
         ack_sync(op);
         return 2;
diff --git a/src/object_id.h b/src/object_id.h
index 0b46c1ea..ca768111 100644
--- a/src/object_id.h
+++ b/src/object_id.h
@@ -39,6 +39,11 @@ struct __attribute__((__packed__)) obj_ver_id
     uint64_t version;
 };
 
+inline bool operator == (const obj_ver_id & a, const obj_ver_id & b)
+{
+    return a.oid == b.oid && a.version == b.version;
+}
+
 inline bool operator < (const obj_ver_id & a, const obj_ver_id & b)
 {
     return a.oid < b.oid || a.oid == b.oid && a.version < b.version;