Re: [PATCH v2 3/3] coredump: hand a pidfd to the usermode coredump helper

[Christian Brauner]

On Mon, Apr 14, 2025 at 04:14:50PM +0200, Oleg Nesterov wrote:
> On 04/14, Christian Brauner wrote:
> >
> > -static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
> > +static int umh_coredump_setup(struct subprocess_info *info, struct cred *new)
> >  {
> >  	struct file *files[2];
> >  	struct coredump_params *cp = (struct coredump_params *)info->data;
> >  	int err;
> >
> > +	if (cp->pid) {
> > +		struct file *pidfs_file __free(fput) = NULL;
> > +
> > +		pidfs_file = pidfs_alloc_file(cp->pid, 0);
> > +		if (IS_ERR(pidfs_file))
> > +			return PTR_ERR(pidfs_file);
> > +
> > +		/*
> > +		 * Usermode helpers are childen of either
> > +		 * system_unbound_wq or of kthreadd. So we know that
> > +		 * we're starting off with a clean file descriptor
> > +		 * table. So we should always be able to use
> > +		 * COREDUMP_PIDFD_NUMBER as our file descriptor value.
> > +		 */
> > +		VFS_WARN_ON_ONCE((pidfs_file = fget_raw(COREDUMP_PIDFD_NUMBER)) != NULL);
> > +
> > +		err = replace_fd(COREDUMP_PIDFD_NUMBER, pidfs_file, 0);
> > +		if (err < 0)
> > +			return err;
> 
> Yes, but if replace_fd() succeeds we need to nullify pidfs_file
> to avoid fput from __free(fput) ?

No, since replace_fd() takes its own reference via do_dup2():

replace_fd()
-> do_dup2()
   {
	get_file(file)
	rcu_assign_pointer(fdt->fd[fd], file);
   }

so we always need to call it. I had a comment about this in the previous
patchset so people don't get confused. I can add it back.

Let me know if you're happy with this otherwise.

> 
> And I think in this case __free(fput) doesn't buy too much, but
> up to you.
> 
> Oleg.
>