From: Kautilya Tripathi <kautilya.tripathi@ibm.com>
Date: Tue, 7 Jul 2026 04:05:20 +0000 (+0530)
Subject: crimson/osd: discard stale primary backfill scan after backfill finishes
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=7b214cba9a04048d6353e1fe29be244e781ca433;p=ceph.git

crimson/osd: discard stale primary backfill scan after backfill finishes

do_request_primary_scan() is an async coroutine.  Backfill can complete
(and on_pg_clean() can destroy backfill_state) while the coroutine is
suspended in scan_for_backfill_primary(), e.g. waiting on an object
context load.  Resuming and delivering PrimaryScanned then dereferences
a null backfill_state.

Discard the scan result if backfill_state is gone; the backfill has
already finished successfully.

Fixes: https://tracker.ceph.com/issues/77623
Signed-off-by: Kautilya Tripathi <kautilya.tripathi@ibm.com>
---

diff --git a/src/crimson/osd/pg_recovery.cc b/src/crimson/osd/pg_recovery.cc
index 7b252e212cf..678464dd5c8 100644
--- a/src/crimson/osd/pg_recovery.cc
+++ b/src/crimson/osd/pg_recovery.cc
@@ -542,6 +542,13 @@ PGRecovery::do_request_primary_scan(hobject_t begin)
       local_conf()->osd_backfill_scan_min,
       local_conf()->osd_backfill_scan_max,
       pg->get_peering_state().get_backfill_targets());
+    if (!backfill_state) {
+      // Backfill may finish (and tear down backfill_state) while this
+      // coroutine is suspended in scan_for_backfill_primary().
+      DEBUGDPP("primary scan result discarded, backfill finished",
+               *pg->get_dpp());
+      co_return;
+    }
     using BackfillState = crimson::osd::BackfillState;
     backfill_state->process_event(
       BackfillState::PrimaryScanned{ std::move(bi) }.intrusive_from_this());