From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Wed, 16 Oct 2024 07:36:40 +0000 (-0400)
Subject: bcachefs: fix restart handling in bch2_alloc_write_key()
X-Git-Tag: v6.12~97^2~31
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=d8b50597748d6d960c936c89b972516c5e39ee96;p=platform%2Fkernel%2Flinux-amlogic.git

bcachefs: fix restart handling in bch2_alloc_write_key()

This is ugly:

We may discover in alloc_write_key that the data type we calculated is
wrong, because BCH_DATA_need_discard is checked/set elsewhere, and the
disk accounting counters we calculated need to be updated.

But bch2_alloc_key_to_dev_counters(..., BTREE_TRIGGER_gc) is not safe
w.r.t. transaction restarts, so we need to propagate the fixup back to
our gc state in case we take a transaction restart.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---

diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 94bbd8505582..0ca3feeb42c8 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -820,12 +820,22 @@ static int bch2_alloc_write_key(struct btree_trans *trans,
 	 * fix that here:
 	 */
 	alloc_data_type_set(&gc, gc.data_type);
-
 	if (gc.data_type != old_gc.data_type ||
 	    gc.dirty_sectors != old_gc.dirty_sectors) {
 		ret = bch2_alloc_key_to_dev_counters(trans, ca, &old_gc, &gc, BTREE_TRIGGER_gc);
 		if (ret)
 			return ret;
+
+		/*
+		 * Ugly: alloc_key_to_dev_counters(..., BTREE_TRIGGER_gc) is not
+		 * safe w.r.t. transaction restarts, so fixup the gc_bucket so
+		 * we don't run it twice:
+		 */
+		percpu_down_read(&c->mark_lock);
+		struct bucket *gc_m = gc_bucket(ca, iter->pos.offset);
+		gc_m->data_type = gc.data_type;
+		gc_m->dirty_sectors = gc.dirty_sectors;
+		percpu_up_read(&c->mark_lock);
 	}
 
 	if (fsck_err_on(new.data_type != gc.data_type,