On Mon, May 05, 2025 at 11:36:15PM +0100, Al Viro wrote: > On Tue, May 06, 2025 at 05:38:02AM +0900, Yunseong Kim wrote: > > The function alloc_fs_context() assumes that current->nsproxy and its > > net_ns field are valid. However, this assumption can be violated in > > cases such as task teardown during do_exit(), where current->nsproxy can > > be NULL or already cleared. > > > > This issue was triggered during stress-ng's kernel-coverage.sh testing, > > Since alloc_fs_context() can be invoked in various contexts — including > > from asynchronous or teardown paths like do_exit() — it's difficult to > > guarantee that its input arguments are always valid. > > > > A follow-up patch will improve the granularity of this fix by moving the > > check closer to the actual mount trigger(e.g., in efivarfs_pm_notify()). > > UGH. > > > diff --git a/fs/fs_context.c b/fs/fs_context.c > > index 582d33e81117..529de43b8b5e 100644 > > --- a/fs/fs_context.c > > +++ b/fs/fs_context.c > > @@ -282,6 +282,9 @@ static struct fs_context *alloc_fs_context(struct file_system_type *fs_type, > > struct fs_context *fc; > > int ret = -ENOMEM; > > > > + if (!current->nsproxy || !current->nsproxy->net_ns) > > + return ERR_PTR(-EINVAL); > > + > > fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL_ACCOUNT); > > if (!fc) > > return ERR_PTR(-ENOMEM); > > That might paper over the oops, but I very much doubt that this will be > a correct fix... Note that in efivarfs_pm_notify() we have other > fun issues when run from such context - have task_work_add() fail in > fput() and if delayed_fput() runs right afterwards and > efivar_init(efivarfs_check_missing, sfi->sb, false); > in there might end up with UAF... We've already accepted a patch that removes the need for vfs_kern_mount() from efivarfs completely.
