Re: [PATCH v3] hfsplus: validate B-tree record offset table

[Viacheslav Dubeyko <slava@dubeyko.com>]

On Sun, 2026-07-12 at 14:09 +0800, Jiaming Zhang wrote:
> A crafted HFS+ image can contain a corrupted B-tree node. The node
> descriptor
> may contain a record count that does not fit in the node, and record
> offsets may
> be unordered, unaligned, outside the node, or point into the offset
> table
> itself.
> 
> Several B-tree helpers consume these on-disk fields before validating
> them:
> hfs_bnode_dump() can walk past the offset table when num_recs is
> corrupted,
> hfs_brec_lenoff() can produce an underflowed length or a record range
> that
> overlaps the offset table. This can make the unlink/writeback path
> repeatedly
> call hfs_bnode_read_u16() with invalid offsets while holding the HFS+
> B-tree
> lock, producing a flood of "requested invalid offset" messages. Other
> writeback
> workers then block on tree->tree_lock and the system reports tasks
> hung in
> hfsplus_write_inode().
> 
> Validate num_recs against the node size before walking the record
> offset table.
> Reject record ranges that are unordered, unaligned, outside the node,
> or
> overlapping the offset table. Reject invalid record indexes before
> reading their
> offset entries, and avoid decrementing an already-zero leaf_count.
> 
> Closes:
> https://lore.kernel.org/lkml/CANypQFb_2TqKGrztAXj5m0_v+QChxXDnQVeifzV8J25Vuju10Q@mail.gmail.com/
> Assisted-by: Codex:gpt-5.5-xhigh
> Signed-off-by: Jiaming Zhang <r772577952@gmail.com>
> ---
> Changes in v3:
> - Drop the keylen == len check.
> - Drop the explicit zero-record check in __hfs_brec_find().
> - Move find cursor reset into hfs_find_reset() and call it from
> hfs_find_init() and hfs_brec_find().
> - Rename helper-local variables as suggested.
> 
>  fs/hfsplus/bfind.c      | 13 ++++----
>  fs/hfsplus/bnode.c      | 16 ++++++---
>  fs/hfsplus/brec.c       | 35 +++++++++++++-------
>  fs/hfsplus/hfsplus_fs.h | 72
> +++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 113 insertions(+), 23 deletions(-)
> 
> diff --git a/fs/hfsplus/bfind.c b/fs/hfsplus/bfind.c
> index 9a55fa6d5294..a5391ff07c70 100644
> --- a/fs/hfsplus/bfind.c
> +++ b/fs/hfsplus/bfind.c
> @@ -18,6 +18,7 @@ int hfs_find_init(struct hfs_btree *tree, struct
> hfs_find_data *fd)
>  
>  	fd->tree = tree;
>  	fd->bnode = NULL;
> +	hfs_find_reset(fd);

The hfs_find_reset() sounds weird. We are trying to initialize the
result section of the b-tree search data structure. So,
hfs_find_result_init() sounds like more proper name, for my taste.

>  	ptr = kzalloc(tree->max_key_len * 2 + 4, GFP_KERNEL);
>  	if (!ptr)
>  		return -ENOMEM;
> @@ -106,12 +107,14 @@ int __hfs_brec_find(struct hfs_bnode *bnode,
> struct hfs_find_data *fd,
>  	u16 off, len, keylen;
>  	int rec;
>  	int b, e;
> -	int res;
> +	int res = -ENOENT;
>  
>  	BUG_ON(!rec_found);
> +	if (!hfs_bnode_num_recs_valid(bnode))
> +		goto fail;

Are you sure that we don't need to call the hfs_find_reset() here?
Because, __hfs_brec_find() could be called not only from
hfs_brec_find() context.

> +
>  	b = 0;
>  	e = bnode->num_recs - 1;
> -	res = -ENOENT;
>  	do {
>  		rec = (e + b) / 2;
>  		len = hfs_brec_lenoff(bnode, rec, &off);
> @@ -158,11 +161,7 @@ int hfs_brec_find(struct hfs_find_data *fd,
> search_strategy_t do_key_compare)
>  	__be32 data;
>  	int height, res;
>  
> -	fd->record = -1;
> -	fd->keyoffset = -1;
> -	fd->keylength = -1;
> -	fd->entryoffset = -1;
> -	fd->entrylength = -1;
> +	hfs_find_reset(fd);

Ditto.

>  
>  	tree = fd->tree;
>  	if (fd->bnode)
> diff --git a/fs/hfsplus/bnode.c b/fs/hfsplus/bnode.c
> index d088fb7eb0df..f185c012d090 100644
> --- a/fs/hfsplus/bnode.c
> +++ b/fs/hfsplus/bnode.c
> @@ -352,15 +352,22 @@ void hfs_bnode_dump(struct hfs_bnode *node)
>  	struct hfs_bnode_desc desc;
>  	__be32 cnid;
>  	int i, off, key_off;
> +	u16 num_recs;
>  
>  	hfs_dbg("node %d\n", node->this);
>  	hfs_bnode_read(node, &desc, 0, sizeof(desc));
> +	num_recs = node->num_recs;
>  	hfs_dbg("next %d, prev %d, type %d, height %d, num_recs
> %d\n",
>  		be32_to_cpu(desc.next), be32_to_cpu(desc.prev),
>  		desc.type, desc.height, be16_to_cpu(desc.num_recs));
>  
> +	if (!hfs_bnode_num_recs_valid(node)) {
> +		hfs_dbg("invalid num_recs %u\n", num_recs);
> +		return;
> +	}
> +
>  	off = node->tree->node_size - 2;
> -	for (i = be16_to_cpu(desc.num_recs); i >= 0; off -= 2, i--)
> {
> +	for (i = num_recs; i >= 0; off -= 2, i--) {
>  		key_off = hfs_bnode_read_u16(node, off);
>  		hfs_dbg(" key_off %d", key_off);
>  		if (i && node->type == HFS_NODE_INDEX) {
> @@ -579,6 +586,9 @@ struct hfs_bnode *hfs_bnode_find(struct hfs_btree
> *tree, u32 num)
>  		goto node_error;
>  	}
>  
> +	if (!hfs_bnode_num_recs_valid(node))
> +		goto node_error;

We assign the num_recs value here [1]:

	node->prev = be32_to_cpu(desc->prev);
	node->next = be32_to_cpu(desc->next);
	node->num_recs = be16_to_cpu(desc->num_recs);
	node->type = desc->type;
	node->height = desc->height;

I think that it is better place for the check.

> +
>  	rec_off = tree->node_size - 2;
>  	off = hfs_bnode_read_u16(node, rec_off);
>  	if (off != sizeof(struct hfs_bnode_desc))
> @@ -586,9 +596,7 @@ struct hfs_bnode *hfs_bnode_find(struct hfs_btree
> *tree, u32 num)
>  	for (i = 1; i <= node->num_recs; off = next_off, i++) {
>  		rec_off -= 2;
>  		next_off = hfs_bnode_read_u16(node, rec_off);
> -		if (next_off <= off ||
> -		    next_off > tree->node_size ||
> -		    next_off & 1)
> +		if (!hfs_brec_range_valid(node, off, next_off,
> rec_off))

Please, see my comment below.

>  			goto node_error;
>  		entry_size = next_off - off;
>  		if (node->type != HFS_NODE_INDEX &&
> diff --git a/fs/hfsplus/brec.c b/fs/hfsplus/brec.c
> index e3df89284079..3112c3bcf9cf 100644
> --- a/fs/hfsplus/brec.c
> +++ b/fs/hfsplus/brec.c
> @@ -20,35 +20,41 @@ static int hfs_btree_inc_height(struct hfs_btree
> *);
>  u16 hfs_brec_lenoff(struct hfs_bnode *node, u16 rec, u16 *off)
>  {
>  	__be16 retval[2];
> -	u16 dataoff;
> +	u16 data_off;
> +	u16 next_off;
>  
> -	dataoff = node->tree->node_size - (rec + 2) * 2;
> -	hfs_bnode_read(node, retval, dataoff, 4);
> +	if (!hfs_brec_record_valid(node, rec)) {
> +		*off = 0;

Are you sure that assigning 0 is correct? We will assign fd->keyoffset
= off. What if fd->keyoffset is used in the case of error? Maybe
U16_MAX will be better?

> +		return 0;

Ditto. Are you sure that 0 is a safe value? Maybe U16_MAX will be
better?

> +	}
> +
> +	data_off = node->tree->node_size - (rec + 2) * 2;
> +	hfs_bnode_read(node, retval, data_off, 4);
>  	*off = be16_to_cpu(retval[1]);
> -	return be16_to_cpu(retval[0]) - *off;
> +	next_off = be16_to_cpu(retval[0]);
> +	if (!hfs_brec_range_valid(node, *off, next_off, data_off))
> +		return 0;

Ditto.

> +	return next_off - *off;
>  }
>  
>  /* Get the length of the key from a keyed record */
>  u16 hfs_brec_keylen(struct hfs_bnode *node, u16 rec)
>  {
> -	u16 retval, recoff;
> +	u16 retval, recoff, len;
>  
>  	if (node->type != HFS_NODE_INDEX && node->type !=
> HFS_NODE_LEAF)
>  		return 0;
> +	if (!hfs_brec_record_valid(node, rec))
> +		return 0;

Ditto.

>  
>  	if ((node->type == HFS_NODE_INDEX) &&
>  	   !(node->tree->attributes & HFS_TREE_VARIDXKEYS) &&
>  	   (node->tree->cnid != HFSPLUS_ATTR_CNID)) {
>  		retval = node->tree->max_key_len + 2;
>  	} else {
> -		recoff = hfs_bnode_read_u16(node,
> -			node->tree->node_size - (rec + 1) * 2);
> -		if (!recoff)
> -			return 0;
> -		if (recoff > node->tree->node_size - 2) {
> -			pr_err("recoff %d too large\n", recoff);
> +		len = hfs_brec_lenoff(node, rec, &recoff);

Ditto.

> +		if (len == 0)
>  			return 0;
> -		}
>  
>  		retval = hfs_bnode_read_u16(node, recoff) + 2;
>  		if (retval > node->tree->max_key_len + 2) {
> @@ -185,10 +191,15 @@ int hfs_brec_remove(struct hfs_find_data *fd)
>  	tree = fd->tree;
>  	node = fd->bnode;
>  again:
> +	if (!hfs_brec_record_valid(node, fd->record))
> +		return -EIO;

We check the fd->record. If this value is incorrect, then it's internal
logic error but not -EIO error code.

> +
>  	rec_off = tree->node_size - (fd->record + 2) * 2;
>  	end_off = tree->node_size - (node->num_recs + 1) * 2;
>  
>  	if (node->type == HFS_NODE_LEAF) {
> +		if (tree->leaf_count == 0)
> +			return -EIO;

Ditto.

>  		tree->leaf_count--;
>  		mark_inode_dirty(tree->inode);
>  	}
> diff --git a/fs/hfsplus/hfsplus_fs.h b/fs/hfsplus/hfsplus_fs.h
> index ec04b82ad927..d87d55a35d25 100644
> --- a/fs/hfsplus/hfsplus_fs.h
> +++ b/fs/hfsplus/hfsplus_fs.h
> @@ -587,6 +587,78 @@ bool is_bnode_offset_valid(struct hfs_bnode
> *node, u32 off)
>  	return is_valid;
>  }
>  
> +static inline
> +bool hfs_bnode_num_recs_valid(struct hfs_bnode *node)
> +{
> +	u32 node_size;
> +	u32 table_size;
> +	u32 area_size;
> +	u32 rec_size = sizeof(__be16);
> +	u32 desc_size = sizeof(struct hfs_bnode_desc);
> +
> +	if (!node || !node->tree)
> +		return false;
> +
> +	node_size = node->tree->node_size;
> +	if (node_size < desc_size)
> +		return false;
> +
> +	area_size = node_size - desc_size;
> +	table_size = ((u32)node->num_recs + 1) * rec_size;
> +
> +	return table_size <= area_size;
> +}
> +
> +static inline
> +bool hfs_brec_record_valid(struct hfs_bnode *node, int record)
> +{
> +	if (!hfs_bnode_num_recs_valid(node))
> +		return false;
> +	if (record < 0)
> +		return false;
> +
> +	return record < node->num_recs;
> +}
> +
> +static inline
> +bool hfs_brec_range_valid(struct hfs_bnode *node, u16 off, u16
> next_off,
> +			  u16 rec_off)

We are checking not range but neighboring items. So, I cannot agree
with hfs_brec_range_valid() name because there are no range here in the
check.

> +{
> +	u32 table_size;
> +	u32 table_start;
> +	u32 rec_size = sizeof(__be16);
> +	u32 desc_size = sizeof(struct hfs_bnode_desc);
> +
> +	if (!node || !node->tree)
> +		return false;
> +
> +	if (off < desc_size || (off & 1))
> +		return false;
> +
> +	if (next_off <= off ||
> +		next_off > node->tree->node_size ||
> +		next_off > rec_off ||
> +		(next_off & 1))
> +		return false;
> +
> +	table_size = ((u32)node->num_recs + 1) * rec_size;
> +	table_start = node->tree->node_size - table_size;
> +	if (next_off > table_start)
> +		return false;
> +
> +	return true;
> +}
> +
> +static inline
> +void hfs_find_reset(struct hfs_find_data *fd)

Please, see my comment above.

Thanks,
Slava.

> +{
> +	fd->record = -1;
> +	fd->keyoffset = -1;
> +	fd->keylength = -1;
> +	fd->entryoffset = -1;
> +	fd->entrylength = -1;
> +}
> +
>  static inline
>  u32 check_and_correct_requested_length(struct hfs_bnode *node, u32
> off, u32 len)
>  {

[1]
https://elixir.bootlin.com/linux/v7.2-rc3/source/fs/hfsplus/bnode.c#L559