From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Date: Sat, 6 Mar 2021 14:31:03 +0000 (-0500)
Subject: btrfs: tree mod log vs zoned fix 2 from fdmanana
X-Git-Url: http://git.hungrycats.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fheads%2Ftree-mod-test;p=linux

btrfs: tree mod log vs zoned fix 2 from fdmanana

Great!

Ok, so that seems to confirm what I suspected and made me run into
other sorts of weirdness during logical ino calls (returning
unexpected results).
I haven't hit the BUG_ON() as you do, but if this is indeed caused by
allowing to reuse unwritten extent buffers in the same transaction,
it's no wonder that BUG_ON() and many other weird issues happen.

Can you now try the following version?

https://pastebin.com/raw/5VHjzdn6

Leave it for at least as many days as you tested the previous patch,
hell, even a week or more if you can.

Thanks, much appreciated.
---

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index b5f24bb879713..d840a626e01d2 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3336,6 +3336,7 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 
 	if (last_ref && btrfs_header_generation(buf) == trans->transid) {
 		struct btrfs_block_group *cache;
+               bool have_mod_log_users;
 
 		if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
 			ret = check_ref_cleanup(trans, buf->start);
@@ -3353,7 +3354,19 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			goto out;
 		}
 
-		if (btrfs_is_zoned(fs_info)) {
+               /*
+                * If there are tree mod log users, we may have recorded mog log
+                * operations on this extent buffer, so we can not add back this
+                * extent buffer to the free space cache/tree, because if it gets
+                * allocated and changed while there are mod log users, replaying
+                * the mod log operations will result in all sorts of unexpected
+                * results.
+                */
+               read_lock(&fs_info->tree_mod_log_lock);
+               have_mod_log_users = !list_empty(&fs_info->tree_mod_seq_list);
+               read_unlock(&fs_info->tree_mod_log_lock);
+
+               if (have_mod_log_users || btrfs_is_zoned(fs_info)) {
 			btrfs_redirty_list_add(trans->transaction, buf);
 			pin_down_extent(trans, cache, buf->start, buf->len, 1);
 			btrfs_put_block_group(cache);