md/raid5-cache: clear MD_SB_CHANGE_PENDING before flushing stripes (c9020e64) · Commits · 戴 / test

In recovery, if we process too much data, raid5-cache may set MD_SB_CHANGE_PENDING, which causes spinning in handle_stripe(). Fix this issue by clearing the bit before flushing data only stripes. This issue was initially discussed in [1]. [1] https://www.spinics.net/lists/raid/msg64409.html Signed-off-by:

Song Liu <songliubraving@fb.com>

drivers/md/raid5-cache.c

+7 −0

Original line number	Diff line number	Diff line
		@@ -2430,10 +2430,15 @@ static void r5c_recovery_flush_data_only_stripes(struct r5l_log *log,
		struct mddev *mddev = log->rdev->mddev;
		struct r5conf *conf = mddev->private;
		struct stripe_head sh, next;
		bool cleared_pending = false;

		if (ctx->data_only_stripes == 0)
		return;

		if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
		cleared_pending = true;
		clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
		}
		log->r5c_journal_mode = R5C_JOURNAL_MODE_WRITE_BACK;

		list_for_each_entry_safe(sh, next, &ctx->cached_list, lru) {
		@@ -2448,6 +2453,8 @@ static void r5c_recovery_flush_data_only_stripes(struct r5l_log *log,
		atomic_read(&conf->active_stripes) == 0);

		log->r5c_journal_mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
		if (cleared_pending)
		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
		}

		static int r5l_recovery_log(struct r5l_log *log)

Admin message