Merge tag 'for-6.17-fix-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs fix from David Sterba:
 "A single btrfs commit. It fixes a problem that people started to hit
  since 6.15.3 during log replay (e.g. after a crash).

  The bug is old but got more likely to happen since commit
  5e85262 ("btrfs: fix fsync of files with no hard links not
  persisting deletion") got backported to stable (6.15 only)"

* tag 'for-6.17-fix-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: fix log tree replay failure due to file with 0 links and extents

Author: torvalds

fs/btrfs/tree-log.c

@@ -321,8 +321,7 @@ struct walk_control {
 
 	/*
 	 * Ignore any items from the inode currently being processed. Needs
-	 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
-	 * the LOG_WALK_REPLAY_INODES stage.
+	 * to be set every time we find a BTRFS_INODE_ITEM_KEY.
 	 */
 	bool ignore_cur_inode;
 
@@ -2465,32 +2464,45 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
 
 	nritems = btrfs_header_nritems(eb);
 	for (i = 0; i < nritems; i++) {
-		btrfs_item_key_to_cpu(eb, &key, i);
+		struct btrfs_inode_item *inode_item;
 
-		/* inode keys are done during the first stage */
-		if (key.type == BTRFS_INODE_ITEM_KEY &&
-		    wc->stage == LOG_WALK_REPLAY_INODES) {
-			struct btrfs_inode_item *inode_item;
-			u32 mode;
+		btrfs_item_key_to_cpu(eb, &key, i);
 
-			inode_item = btrfs_item_ptr(eb, i,
-					    struct btrfs_inode_item);
+		if (key.type == BTRFS_INODE_ITEM_KEY) {
+			inode_item = btrfs_item_ptr(eb, i, struct btrfs_inode_item);
 			/*
-			 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
-			 * and never got linked before the fsync, skip it, as
-			 * replaying it is pointless since it would be deleted
-			 * later. We skip logging tmpfiles, but it's always
-			 * possible we are replaying a log created with a kernel
-			 * that used to log tmpfiles.
+			 * An inode with no links is either:
+			 *
+			 * 1) A tmpfile (O_TMPFILE) that got fsync'ed and never
+			 *    got linked before the fsync, skip it, as replaying
+			 *    it is pointless since it would be deleted later.
+			 *    We skip logging tmpfiles, but it's always possible
+			 *    we are replaying a log created with a kernel that
+			 *    used to log tmpfiles;
+			 *
+			 * 2) A non-tmpfile which got its last link deleted
+			 *    while holding an open fd on it and later got
+			 *    fsynced through that fd. We always log the
+			 *    parent inodes when inode->last_unlink_trans is
+			 *    set to the current transaction, so ignore all the
+			 *    inode items for this inode. We will delete the
+			 *    inode when processing the parent directory with
+			 *    replay_dir_deletes().
 			 */
 			if (btrfs_inode_nlink(eb, inode_item) == 0) {
 				wc->ignore_cur_inode = true;
 				continue;
 			} else {
 				wc->ignore_cur_inode = false;
 			}
-			ret = replay_xattr_deletes(wc->trans, root, log,
-						   path, key.objectid);
+		}
+
+		/* Inode keys are done during the first stage. */
+		if (key.type == BTRFS_INODE_ITEM_KEY &&
+		    wc->stage == LOG_WALK_REPLAY_INODES) {
+			u32 mode;
+
+			ret = replay_xattr_deletes(wc->trans, root, log, path, key.objectid);
 			if (ret)
 				break;
 			mode = btrfs_inode_mode(eb, inode_item);