Re: [PATCH 16/24] link_path_walk: kill the recursion

From: Linus Torvalds <torvalds@linux-foundation.org>
To: Al Viro <viro@zeniv.linux.org.uk>
Cc: Neil Brown <neilb@suse.de>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [PATCH 16/24] link_path_walk: kill the recursion
Date: Mon, 20 Apr 2015 14:04:53 -0700	[thread overview]
Message-ID: <CA+55aFwP_zR8ETgmX-AfpSzG+B6+FZnaJ+_DFiOmKxEAd5=tkg@mail.gmail.com> (raw)
In-Reply-To: <1429553588-24764-16-git-send-email-viro@ZenIV.linux.org.uk>

On Mon, Apr 20, 2015 at 11:13 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>   */
>  static int link_path_walk(const char *name, struct nameidata *nd)
>  {
> +       struct saved {
> +               struct path link;
> +               void *cookie;
> +               const char *name;
> +       } stack[MAX_NESTED_LINKS], *last = stack + nd->depth - 1;
>         struct path next;
>         int err;

I was going to complain about this, and suggest that you move it to
the nameidata, but I see that you did that later.

That said, you then introduce a stack-allocated "struct saved stack[]"
in path_mountpoint[] instead, *and* nameidata is saved on stack, so
this all ends up being very stack-intensive anyway.

I might have missed some patch here, but would it be possible to just
allocate a single per-thread nameidata, and just leave it at that?
Because allocating that thing on the stack when it contains what is
now one kilobyte of array data is *not* acceptable.

Other than that, my quick scan through this looked fine. And maybe
that quick scan missed where you already did that too.

                           Linus