hfs: discuss to add offset/length validation in hfs_brec_lenoff

[PATCH 0/1] hfs: discuss to add offset/length validation in hfs_brec_lenoff

Posted by Chenzhi Yang 1 month, 2 weeks ago

From: Yang Chenzhi <yang.chenzhi@vivo.com>

When running syzbot with a crafted HFS/HFS+ disk image containing
invalid record offsets or lengths, the filesystem may hang. For
example, in this case syzbot set the header’s second record offset
to 0x7f00 while node_size is 4096. HFS/HFS+ failed to detect this
fault, which eventually led to a crash.

Since HFS/HFS+ makes heavy use of hfs_brec_lenoff, adding manual
offset/length checks at every call site would be tedious and
error-prone.

Instead, it may be more robust to introduce validation directly
inside hfs_brec_lenoff (or at a similar central point), ensuring
that all callers can safely rely on the returned offset and length
without additional checks.

Yang Chenzhi (1):
  hfs: validate record offset in hfsplus_bmap_alloc

 fs/hfsplus/bnode.c      | 41 ----------------------------------------
 fs/hfsplus/btree.c      |  6 ++++++
 fs/hfsplus/hfsplus_fs.h | 42 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 48 insertions(+), 41 deletions(-)

-- 
2.43.0

Re: [PATCH 0/1] hfs: discuss to add offset/length validation in hfs_brec_lenoff

Posted by Viacheslav Dubeyko 1 month, 2 weeks ago

On Mon, 2025-08-18 at 22:17 +0800, Chenzhi Yang wrote:
> From: Yang Chenzhi <yang.chenzhi@vivo.com>
> 
> When running syzbot with a crafted HFS/HFS+ disk image containing
> invalid record offsets or lengths, the filesystem may hang. For
> example, in this case syzbot set the header’s second record offset
> to 0x7f00 while node_size is 4096. HFS/HFS+ failed to detect this
> fault, which eventually led to a crash.
> 

HFS has 512 bytes b-tree node size.

> Since HFS/HFS+ makes heavy use of hfs_brec_lenoff, adding manual
> offset/length checks at every call site would be tedious and
> error-prone.
> 

You are mentioning HFS here. But you've shared fix only for HFS+. Are you
planning to share the fix for HFS too?

Thanks,
Slava.

> Instead, it may be more robust to introduce validation directly
> inside hfs_brec_lenoff (or at a similar central point), ensuring
> that all callers can safely rely on the returned offset and length
> without additional checks.
> 
> Yang Chenzhi (1):
>   hfs: validate record offset in hfsplus_bmap_alloc
> 
>  fs/hfsplus/bnode.c      | 41 ----------------------------------------
>  fs/hfsplus/btree.c      |  6 ++++++
>  fs/hfsplus/hfsplus_fs.h | 42 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 48 insertions(+), 41 deletions(-)