Re: [PATCH bpf] bpf, lsm: Mark backing_file_alloc/free LSM hooks as untrusted
From: Matt Bobrowski
Date: Mon Nov 16 2026 - 06:19:05 EST
On Wed, Nov 11, 2026 at 09:35:52PM +0100, Sergei Nizovtsev wrote:
> The backing_file_alloc and backing_file_free LSM hooks get the backing
> file as "struct file *backing_file", but its f_path is not populated
> while they run:
>
> - security_backing_file_alloc() is called from init_backing_file() in
> alloc_empty_backing_file(), before vfs_open() sets f_path, so
> backing_file->f_path.dentry is NULL;
> - security_backing_file_free() can run for a backing file whose open
> failed and whose f_path was already cleared.
>
> Neither hook is listed in untrusted_lsm_hooks, so bpf_lsm_is_trusted()
> lets BPF LSM programs use the argument as a trusted pointer. A program
> attached to lsm/backing_file_alloc can pass &backing_file->f_path to the
> trusted-args kfunc bpf_path_d_path(), and d_path() dereferences the NULL
> dentry on the first open of a file on overlayfs:
>
> extern int bpf_path_d_path(const struct path *path, char *buf,
> size_t sz) __ksym;
> char buf[256];
>
> SEC("lsm/backing_file_alloc")
> int BPF_PROG(poc, struct file *backing_file,
> const struct file *user_file)
> {
> bpf_path_d_path(&backing_file->f_path, buf, sizeof(buf));
> return 0;
> }
>
> # bpftool prog load poc.bpf.o /sys/fs/bpf/poc autoattach
> # cat <any file on an overlayfs mount>
>
> Oops: general protection fault, probably for non-canonical address 0xdffffc000000000c: 0000 [#1] SMP KASAN NOPTI
> KASAN: null-ptr-deref in range [0x0000000000000060-0x0000000000000067]
> RIP: 0010:d_path+0xb4/0x810
> Call Trace:
> bpf_path_d_path+0x26/0x70
> bpf_prog_f8f2e241d4ff7217_poc+0x33/0x37
> bpf_trampoline_6442521065+0x61/0xf9
> security_backing_file_alloc+0x78/0xe0
> alloc_empty_backing_file+0xf0/0x170
> backing_file_open+0xfd/0x290
> ovl_open_realfile+0x1ed/0x270
> ovl_open+0x1a9/0x2f0
> do_dentry_open+0x404/0x1280
> vfs_open_consume+0xba/0x4b0
> path_openat+0xc4f/0x1f60
>
> Add both hooks to untrusted_lsm_hooks, next to file_alloc_security and
> file_free_security, which are listed there for the same reason. The
> program above is then rejected at load time with "R1 must be referenced
> or trusted".
I literally had a patch queued up and was just about to send it out for
review. Thank you for this.
> Fixes: 6af36aeb147a ("lsm: add backing_file LSM hooks")
> Cc: stable@xxxxxxxxxxxxxxx
> Assisted-by: Claude:claude-fable-5-2
> Signed-off-by: Sergei Nizovtsev <snizovtsev@xxxxxxxxxxxxxx>
> ---
> Tested on bpf.git master (475a0cfd02f1) under virtme-ng with KASAN: the
> reproducer oopses without the patch and fails to load with it. Only the
> backing_file_alloc case was reproduced; backing_file_free is from code
> reading.
>
> kernel/bpf/bpf_lsm.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/kernel/bpf/bpf_lsm.c b/kernel/bpf/bpf_lsm.c
> index f58dce888ff4a..f39a3766f2245 100644
> --- a/kernel/bpf/bpf_lsm.c
> +++ b/kernel/bpf/bpf_lsm.c
> @@ -411,6 +411,8 @@ BTF_ID(func, bpf_lsm_bpf_map_free)
> BTF_ID(func, bpf_lsm_bpf_prog_free)
> BTF_ID(func, bpf_lsm_file_alloc_security)
> BTF_ID(func, bpf_lsm_file_free_security)
> +BTF_ID(func, bpf_lsm_backing_file_alloc)
> +BTF_ID(func, bpf_lsm_backing_file_free)
> #ifdef CONFIG_SECURITY_NETWORK
> BTF_ID(func, bpf_lsm_sk_alloc_security)
> BTF_ID(func, bpf_lsm_sk_free_security)
> --
> 2.56.0
>