From: Xuehan Xu <xuxuehan@qianxin.com>
Date: Sat, 30 May 2026 02:58:27 +0000 (+0800)
Subject: crimson/os/seastore: invalidate the transaction if it has written
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fheads%2Fdev-lbc;p=ceph-ci.git

crimson/os/seastore: invalidate the transaction if it has written
lba/backref nodes to the disk when the rewrite transaction tries to
merge those nodes' content

Otherwise, the result of the merging won't be persisted to the disk

Signed-off-by: Xuehan Xu <xuxuehan@qianxin.com>
---

diff --git a/src/crimson/os/seastore/cache.cc b/src/crimson/os/seastore/cache.cc
index 8d724278f53..b57b6feed8f 100644
--- a/src/crimson/os/seastore/cache.cc
+++ b/src/crimson/os/seastore/cache.cc
@@ -1029,20 +1029,19 @@ void Cache::commit_replace_extent(
 
 }
 
-void Cache::invalidate_extent(
-    Transaction& t,
-    CachedExtent& extent)
+void Cache::invalidate_trans(
+  Transaction &t,
+  CachedExtent &extent,
+  bool backref_lba_written_only)
 {
-  if (!extent.may_conflict()) {
-    assert(extent.read_transactions.empty());
-    extent.set_invalid(t);
-    return;
-  }
-
   LOG_PREFIX(Cache::invalidate_extent);
   bool do_conflict_log = true;
   std::vector<Transaction*> invalidated_trans;
   for (auto &&i: extent.read_transactions) {
+    if (likely(backref_lba_written_only &&
+               !i.t->backref_lba_ool_written)) {
+      continue;
+    }
     if (!i.t->conflicted) {
       if (do_conflict_log) {
         SUBDEBUGT(seastore_t, "conflict begin -- {}", t, extent);
@@ -1060,6 +1059,20 @@ void Cache::invalidate_extent(
     trans->retired_set.clear();
     trans->views.clear();
   }
+}
+
+void Cache::invalidate_extent(
+    Transaction& t,
+    CachedExtent& extent)
+{
+  if (!extent.may_conflict()) {
+    assert(extent.read_transactions.empty());
+    extent.set_invalid(t);
+    return;
+  }
+
+  invalidate_trans(t, extent, false);
+
   extent.set_invalid(t);
 }
 
@@ -1443,6 +1456,9 @@ record_t Cache::prepare_record(
     if (should_use_no_conflict_publish(t.get_src(), i->get_type())) {
       i->new_committer(t);
       i->committer->block_trans(t);
+      if (is_lba_backref_node(i->get_type())) {
+        invalidate_trans(t, *i->prior_instance, true);
+      }
     }
     assert(i->is_exist_mutation_pending() ||
 	   i->prior_instance);
@@ -1724,6 +1740,9 @@ record_t Cache::prepare_record(
       assert(i->committer);
       auto &committer = *i->committer;
       committer.block_trans(t);
+      if (is_lba_backref_node(i->get_type())) {
+        invalidate_trans(t, *i->prior_instance, true);
+      }
       i->get_prior_instance()->set_io_wait(
         CachedExtent::extent_state_t::CLEAN, should_use_no_conflict_publish(t.get_src(), i->get_type()));
     }
@@ -1759,6 +1778,9 @@ record_t Cache::prepare_record(
       i->get_prior_instance()->committer = i->committer;
       auto &committer = *i->committer;
       committer.block_trans(t);
+      if (is_lba_backref_node(i->get_type())) {
+        invalidate_trans(t, *i->prior_instance, true);
+      }
       i->get_prior_instance()->set_io_wait(
         CachedExtent::extent_state_t::CLEAN, true);
     }
diff --git a/src/crimson/os/seastore/cache.h b/src/crimson/os/seastore/cache.h
index e88157c4e45..1baca40efb3 100644
--- a/src/crimson/os/seastore/cache.h
+++ b/src/crimson/os/seastore/cache.h
@@ -1993,6 +1993,10 @@ private:
 void stage_visibility_handoff(Transaction& t,
                               CachedExtentRef next,
                               CachedExtentRef prev);
+  void invalidate_trans(
+    Transaction &t,
+    CachedExtent &extent,
+    bool backref_lba_written_only);
 
   /// Invalidate extent and mark affected transactions
   void invalidate_extent(Transaction& t, CachedExtent& extent);
diff --git a/src/crimson/os/seastore/transaction.h b/src/crimson/os/seastore/transaction.h
index 4abcf240c71..11c72c5dedf 100644
--- a/src/crimson/os/seastore/transaction.h
+++ b/src/crimson/os/seastore/transaction.h
@@ -552,6 +552,7 @@ public:
     rewrite_stats = {};
     conflicted = false;
     blocked_by.clear();
+    backref_lba_ool_written = false;
     assert(backref_entries.empty());
     if (!has_reset) {
       has_reset = true;
@@ -685,6 +686,7 @@ public:
   // when the rewrite transactions that modifies the extent are
   // committing.
   std::set<transaction_id_t> blocked_by;;
+  bool backref_lba_ool_written = false;
 
 private:
   friend class Cache;
diff --git a/src/crimson/os/seastore/transaction_manager.cc b/src/crimson/os/seastore/transaction_manager.cc
index 97d8d15f779..7018721269d 100644
--- a/src/crimson/os/seastore/transaction_manager.cc
+++ b/src/crimson/os/seastore/transaction_manager.cc
@@ -733,6 +733,13 @@ TransactionManager::do_submit_transaction(
 
   auto num_extents = allocated_extents.size();
   SUBTRACET(seastore_t, "process {} allocated extents", tref, num_extents);
+  if (epm->get_main_backend_type() == backend_type_t::RANDOM_BLOCK) {
+    tref.backref_lba_ool_written = true;
+  }
+  while (!tref.blocked_by.empty()) {
+    co_await trans_intr::make_interruptible(seastar::yield());
+  }
+
   co_await epm->write_preallocated_ool_extents(tref, allocated_extents);
 
   SUBTRACET(seastore_t, "entering prepare", tref);
@@ -740,9 +747,6 @@ TransactionManager::do_submit_transaction(
     tref.get_handle().enter(write_pipeline.prepare)
   );
 
-  while (!tref.blocked_by.empty()) {
-    co_await trans_intr::make_interruptible(seastar::yield());
-  }
   if (trim_alloc_to && *trim_alloc_to != JOURNAL_SEQ_NULL) {
     SUBTRACET(seastore_t, "trim backref_bufs to {}", tref, *trim_alloc_to);
     cache->trim_backref_bufs(*trim_alloc_to);