Merge tag 'for-5.8-rc4-tag' of git://git.kernel.org/pub/scm/linux/ker…

…nel/git/kdave/linux

Pull btrfs fixes from David Sterba:

 - regression fix of a leak in global block reserve accounting

 - fix a (hard to hit) race of readahead vs releasepage that could lead
   to crash

 - convert all remaining uses of comment fall through annotations to the
   pseudo keyword

 - fix crash when mounting a fuzzed image with -o recovery

* tag 'for-5.8-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: reset tree root pointer after error in init_tree_roots
  btrfs: fix reclaim_size counter leak after stealing from global reserve
  btrfs: fix fatal extent_buffer readahead vs releasepage race
  btrfs: convert comments to fallthrough annotations

fs/btrfs/ctree.c

@@ -1196,7 +1196,7 @@ __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
 		switch (tm->op) {
 		case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
 			BUG_ON(tm->slot < n);
-			/* Fallthrough */
+			fallthrough;
 		case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
 		case MOD_LOG_KEY_REMOVE:
 			btrfs_set_node_key(eb, &tm->key, tm->slot);

fs/btrfs/disk-io.c

@@ -2593,10 +2593,12 @@ static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
 		    !extent_buffer_uptodate(tree_root->node)) {
 			handle_error = true;
 
-			if (IS_ERR(tree_root->node))
+			if (IS_ERR(tree_root->node)) {
 				ret = PTR_ERR(tree_root->node);
-			else if (!extent_buffer_uptodate(tree_root->node))
+				tree_root->node = NULL;
+			} else if (!extent_buffer_uptodate(tree_root->node)) {
 				ret = -EUCLEAN;
+			}
 
 			btrfs_warn(fs_info, "failed to read tree root");
 			continue;

fs/btrfs/extent_io.c

@@ -5058,25 +5058,28 @@ struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
 static void check_buffer_tree_ref(struct extent_buffer *eb)
 {
 	int refs;
-	/* the ref bit is tricky.  We have to make sure it is set
-	 * if we have the buffer dirty.   Otherwise the
-	 * code to free a buffer can end up dropping a dirty
-	 * page
+	/*
+	 * The TREE_REF bit is first set when the extent_buffer is added
+	 * to the radix tree. It is also reset, if unset, when a new reference
+	 * is created by find_extent_buffer.
 	 *
-	 * Once the ref bit is set, it won't go away while the
-	 * buffer is dirty or in writeback, and it also won't
-	 * go away while we have the reference count on the
-	 * eb bumped.
+	 * It is only cleared in two cases: freeing the last non-tree
+	 * reference to the extent_buffer when its STALE bit is set or
+	 * calling releasepage when the tree reference is the only reference.
 	 *
-	 * We can't just set the ref bit without bumping the
-	 * ref on the eb because free_extent_buffer might
-	 * see the ref bit and try to clear it.  If this happens
-	 * free_extent_buffer might end up dropping our original
-	 * ref by mistake and freeing the page before we are able
-	 * to add one more ref.
+	 * In both cases, care is taken to ensure that the extent_buffer's
+	 * pages are not under io. However, releasepage can be concurrently
+	 * called with creating new references, which is prone to race
+	 * conditions between the calls to check_buffer_tree_ref in those
+	 * codepaths and clearing TREE_REF in try_release_extent_buffer.
 	 *
-	 * So bump the ref count first, then set the bit.  If someone
-	 * beat us to it, drop the ref we added.
+	 * The actual lifetime of the extent_buffer in the radix tree is
+	 * adequately protected by the refcount, but the TREE_REF bit and
+	 * its corresponding reference are not. To protect against this
+	 * class of races, we call check_buffer_tree_ref from the codepaths
+	 * which trigger io after they set eb->io_pages. Note that once io is
+	 * initiated, TREE_REF can no longer be cleared, so that is the
+	 * moment at which any such race is best fixed.
 	 */
 	refs = atomic_read(&eb->refs);
 	if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
@@ -5527,6 +5530,11 @@ int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
 	clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 	eb->read_mirror = 0;
 	atomic_set(&eb->io_pages, num_reads);
+	/*
+	 * It is possible for releasepage to clear the TREE_REF bit before we
+	 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
+	 */
+	check_buffer_tree_ref(eb);
 	for (i = 0; i < num_pages; i++) {
 		page = eb->pages[i];
 

fs/btrfs/ref-verify.c

@@ -509,7 +509,7 @@ static int process_leaf(struct btrfs_root *root,
 		switch (key.type) {
 		case BTRFS_EXTENT_ITEM_KEY:
 			*num_bytes = key.offset;
-			/* fall through */
+			fallthrough;
 		case BTRFS_METADATA_ITEM_KEY:
 			*bytenr = key.objectid;
 			ret = process_extent_item(fs_info, path, &key, i,

fs/btrfs/space-info.c

@@ -879,8 +879,8 @@ static bool steal_from_global_rsv(struct btrfs_fs_info *fs_info,
 		return false;
 	}
 	global_rsv->reserved -= ticket->bytes;
+	remove_ticket(space_info, ticket);
 	ticket->bytes = 0;
-	list_del_init(&ticket->list);
 	wake_up(&ticket->wait);
 	space_info->tickets_id++;
 	if (global_rsv->reserved < global_rsv->size)

fs/btrfs/super.c

@@ -523,7 +523,7 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 		case Opt_compress_force:
 		case Opt_compress_force_type:
 			compress_force = true;
-			/* Fallthrough */
+			fallthrough;
 		case Opt_compress:
 		case Opt_compress_type:
 			saved_compress_type = btrfs_test_opt(info,
@@ -622,7 +622,7 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 			btrfs_set_opt(info->mount_opt, NOSSD);
 			btrfs_clear_and_info(info, SSD,
 					     "not using ssd optimizations");
-			/* Fallthrough */
+			fallthrough;
 		case Opt_nossd_spread:
 			btrfs_clear_and_info(info, SSD_SPREAD,
 					     "not using spread ssd allocation scheme");
@@ -793,7 +793,7 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 		case Opt_recovery:
 			btrfs_warn(info,
 				   "'recovery' is deprecated, use 'usebackuproot' instead");
-			/* fall through */
+			fallthrough;
 		case Opt_usebackuproot:
 			btrfs_info(info,
 				   "trying to use backup root at mount time");

fs/btrfs/volumes.h

@@ -408,7 +408,7 @@ static inline enum btrfs_map_op btrfs_op(struct bio *bio)
 		return BTRFS_MAP_WRITE;
 	default:
 		WARN_ON_ONCE(1);
-		/* fall through */
+		fallthrough;
 	case REQ_OP_READ:
 		return BTRFS_MAP_READ;
 	}