bcachefs: Shutdown path improvements

We're seeing occasional firings of the assertion in the key cache
shutdown code that nr_dirty == 0, which means we must sometimes be doing
transaction commits after we've gone read only.

Cleanups & changes:
 - BCH_FS_ALLOC_CLEAN renamed to BCH_FS_CLEAN_SHUTDOWN
 - new helper bch2_btree_interior_updates_flush(), which returns true if
   it had to wait
 - bch2_btree_flush_writes() now also returns true if there were btree
   writes in flight
 - __bch2_fs_read_only now checks if btree writes were in flight in the
   shutdown loop: btree write completion does a transaction update, to
   update the pointer in the parent node
 - assert that !BCH_FS_CLEAN_SHUTDOWN in __bch2_trans_commit

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/bcachefs.h

@@ -494,7 +494,7 @@ struct bch_dev {
 
 enum {
 	/* startup: */
-	BCH_FS_ALLOC_CLEAN,
+	BCH_FS_CLEAN_SHUTDOWN,
 	BCH_FS_INITIAL_GC_DONE,
 	BCH_FS_INITIAL_GC_UNFIXED,
 	BCH_FS_TOPOLOGY_REPAIR_DONE,

fs/bcachefs/btree_gc.c

@@ -1751,9 +1751,7 @@ int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only)
 
 	down_write(&c->gc_lock);
 
-	/* flush interior btree updates: */
-	closure_wait_event(&c->btree_interior_update_wait,
-			   !bch2_btree_interior_updates_nr_pending(c));
+	bch2_btree_interior_updates_flush(c);
 
 	ret   = bch2_gc_start(c, metadata_only) ?:
 		bch2_gc_alloc_start(c, metadata_only) ?:

fs/bcachefs/btree_io.c

@@ -2099,29 +2099,33 @@ void bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 	}
 }
 
-static void __bch2_btree_flush_all(struct bch_fs *c, unsigned flag)
+static bool __bch2_btree_flush_all(struct bch_fs *c, unsigned flag)
 {
 	struct bucket_table *tbl;
 	struct rhash_head *pos;
 	struct btree *b;
 	unsigned i;
+	bool ret = false;
 restart:
 	rcu_read_lock();
 	for_each_cached_btree(b, c, tbl, i, pos)
 		if (test_bit(flag, &b->flags)) {
 			rcu_read_unlock();
 			wait_on_bit_io(&b->flags, flag, TASK_UNINTERRUPTIBLE);
+			ret = true;
 			goto restart;
 		}
 	rcu_read_unlock();
+
+	return ret;
 }
 
-void bch2_btree_flush_all_reads(struct bch_fs *c)
+bool bch2_btree_flush_all_reads(struct bch_fs *c)
 {
-	__bch2_btree_flush_all(c, BTREE_NODE_read_in_flight);
+	return __bch2_btree_flush_all(c, BTREE_NODE_read_in_flight);
 }
 
-void bch2_btree_flush_all_writes(struct bch_fs *c)
+bool bch2_btree_flush_all_writes(struct bch_fs *c)
 {
-	__bch2_btree_flush_all(c, BTREE_NODE_write_in_flight);
+	return __bch2_btree_flush_all(c, BTREE_NODE_write_in_flight);
 }

fs/bcachefs/btree_io.h

@@ -152,8 +152,8 @@ static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 	bch2_btree_node_write(c, b, lock_held, BTREE_WRITE_ONLY_IF_NEED);
 }
 
-void bch2_btree_flush_all_reads(struct bch_fs *);
-void bch2_btree_flush_all_writes(struct bch_fs *);
+bool bch2_btree_flush_all_reads(struct bch_fs *);
+bool bch2_btree_flush_all_writes(struct bch_fs *);
 
 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 				  unsigned version, unsigned big_endian,

fs/bcachefs/btree_update_interior.c

@@ -2175,19 +2175,27 @@ void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c)
 	mutex_unlock(&c->btree_interior_update_lock);
 }
 
-size_t bch2_btree_interior_updates_nr_pending(struct bch_fs *c)
+static bool bch2_btree_interior_updates_pending(struct bch_fs *c)
 {
-	size_t ret = 0;
-	struct list_head *i;
+	bool ret;
 
 	mutex_lock(&c->btree_interior_update_lock);
-	list_for_each(i, &c->btree_interior_update_list)
-		ret++;
+	ret = !list_empty(&c->btree_interior_update_list);
 	mutex_unlock(&c->btree_interior_update_lock);
 
 	return ret;
 }
 
+bool bch2_btree_interior_updates_flush(struct bch_fs *c)
+{
+	bool ret = bch2_btree_interior_updates_pending(c);
+
+	if (ret)
+		closure_wait_event(&c->btree_interior_update_wait,
+				   !bch2_btree_interior_updates_pending(c));
+	return ret;
+}
+
 void bch2_journal_entries_to_btree_roots(struct bch_fs *c, struct jset *jset)
 {
 	struct btree_root *r;

fs/bcachefs/btree_update_interior.h

@@ -309,7 +309,7 @@ static inline bool bch2_btree_node_insert_fits(struct bch_fs *c,
 
 void bch2_btree_updates_to_text(struct printbuf *, struct bch_fs *);
 
-size_t bch2_btree_interior_updates_nr_pending(struct bch_fs *);
+bool bch2_btree_interior_updates_flush(struct bch_fs *);
 
 void bch2_journal_entries_to_btree_roots(struct bch_fs *, struct jset *);
 struct jset_entry *bch2_btree_roots_to_journal_entries(struct bch_fs *,

fs/bcachefs/btree_update_leaf.c

@@ -1117,6 +1117,8 @@ int __bch2_trans_commit(struct btree_trans *trans)
 			goto out_reset;
 	}
 
+	EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
+
 	memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
 
 	trans->journal_u64s		= trans->extra_journal_entries.nr;

fs/bcachefs/migrate.c

@@ -175,10 +175,7 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 			goto err;
 	}
 
-	/* flush relevant btree updates */
-	closure_wait_event(&c->btree_interior_update_wait,
-			   !bch2_btree_interior_updates_nr_pending(c));
-
+	bch2_btree_interior_updates_flush(c);
 	ret = 0;
 err:
 	bch2_trans_exit(&trans);

fs/bcachefs/move.c

@@ -942,9 +942,7 @@ static int bch2_move_btree(struct bch_fs *c,
 	if (ret)
 		bch_err(c, "error %i in bch2_move_btree", ret);
 
-	/* flush relevant btree updates */
-	closure_wait_event(&c->btree_interior_update_wait,
-			   !bch2_btree_interior_updates_nr_pending(c));
+	bch2_btree_interior_updates_flush(c);
 
 	progress_list_del(c, stats);
 	return ret;

fs/bcachefs/super.c

@@ -195,57 +195,33 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 {
 	struct bch_dev *ca;
 	unsigned i, clean_passes = 0;
+	u64 seq = 0;
 
 	bch2_rebalance_stop(c);
 	bch2_copygc_stop(c);
 	bch2_gc_thread_stop(c);
 
-	/*
-	 * Flush journal before stopping allocators, because flushing journal
-	 * blacklist entries involves allocating new btree nodes:
-	 */
-	bch2_journal_flush_all_pins(&c->journal);
-
 	bch_verbose(c, "flushing journal and stopping allocators");
 
-	bch2_journal_flush_all_pins(&c->journal);
-
 	do {
 		clean_passes++;
 
-		if (bch2_journal_flush_all_pins(&c->journal))
-			clean_passes = 0;
-
-		/*
-		 * In flight interior btree updates will generate more journal
-		 * updates and btree updates (alloc btree):
-		 */
-		if (bch2_btree_interior_updates_nr_pending(c)) {
-			closure_wait_event(&c->btree_interior_update_wait,
-					   !bch2_btree_interior_updates_nr_pending(c));
+		if (bch2_btree_interior_updates_flush(c) ||
+		    bch2_journal_flush_all_pins(&c->journal) ||
+		    bch2_btree_flush_all_writes(c) ||
+		    seq != atomic64_read(&c->journal.seq)) {
+			seq = atomic64_read(&c->journal.seq);
 			clean_passes = 0;
 		}
-		flush_work(&c->btree_interior_update_work);
-
-		if (bch2_journal_flush_all_pins(&c->journal))
-			clean_passes = 0;
 	} while (clean_passes < 2);
+
 	bch_verbose(c, "flushing journal and stopping allocators complete");
 
-	set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
-
-	closure_wait_event(&c->btree_interior_update_wait,
-			   !bch2_btree_interior_updates_nr_pending(c));
-	flush_work(&c->btree_interior_update_work);
-
+	if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
+	    !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
+		set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 	bch2_fs_journal_stop(&c->journal);
 
-	/*
-	 * the journal kicks off btree writes via reclaim - wait for in flight
-	 * writes after stopping journal:
-	 */
-	bch2_btree_flush_all_writes(c);
-
 	/*
 	 * After stopping journal:
 	 */
@@ -304,7 +280,7 @@ void bch2_fs_read_only(struct bch_fs *c)
 	    !test_bit(BCH_FS_ERROR, &c->flags) &&
 	    !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 	    test_bit(BCH_FS_STARTED, &c->flags) &&
-	    test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
+	    test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 	    !c->opts.norecovery) {
 		bch_verbose(c, "marking filesystem clean");
 		bch2_fs_mark_clean(c);
@@ -395,7 +371,7 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 	if (ret)
 		goto err;
 
-	clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
+	clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 
 	for_each_rw_member(ca, c, i)
 		bch2_dev_allocator_add(c, ca);