linux-kernel - Re: [PATCH] ubifs: work around high stack usage with clang

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 25 Mar 2019 10:57:00 -0700
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     Arnd Bergmann <arnd@...db.de>
Cc:     Richard Weinberger <richard@....at>,
        Artem Bityutskiy <dedekind1@...il.com>,
        Adrian Hunter <adrian.hunter@...el.com>,
        clang-built-linux@...glegroups.com,
        Nathan Chancellor <natechancellor@...il.com>,
        Eric Biggers <ebiggers@...gle.com>,
        Herbert Xu <herbert@...dor.apana.org.au>,
        Sascha Hauer <s.hauer@...gutronix.de>,
        linux-mtd@...ts.infradead.org, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] ubifs: work around high stack usage with clang

On Sun, Mar 24, 2019 at 12:45 PM Arnd Bergmann <arnd@...db.de> wrote:
>
> Building this file with clang can result in large stack usage as seen from
> this warning:
>
> fs/ubifs/auth.c:78:5: error: stack frame size of 1152 bytes in function 'ubifs_prepare_auth_node'
>
> The problem is that inlining ubifs_hash_calc_hmac() leads to
> two SHASH_DESC_ON_STACK() blocks in the same function, and clang
> for some reason does not reuse the stack space as it should.
>
> Putting the first declaration into a separate basic block avoids
> this problem and reduces the stack allocation to 640 bytes.
>
> Signed-off-by: Arnd Bergmann <arnd@...db.de>

LGTM; in general, it can be useful to always include a comment before
a new scope like:

/* New scope to limit lifetime of large stack allocated hash_desc. */

but folks can always refer to the git blame/log for reasoning as to
why a new scope was added.

I plan on working on a pass in LLVM to improve stack slot reuse
sometime in 2019Q2.  Thanks for this patch.
Reviewed-by: Nick Desaulniers <ndesaulniers@...gle.com>

> ---
>  fs/ubifs/auth.c | 17 ++++++++++-------
>  1 file changed, 10 insertions(+), 7 deletions(-)
>
> diff --git a/fs/ubifs/auth.c b/fs/ubifs/auth.c
> index 5bf5fd08879e..5d3d0b37a908 100644
> --- a/fs/ubifs/auth.c
> +++ b/fs/ubifs/auth.c
> @@ -78,7 +78,6 @@ static int ubifs_hash_calc_hmac(const struct ubifs_info *c, const u8 *hash,
>  int ubifs_prepare_auth_node(struct ubifs_info *c, void *node,
>                              struct shash_desc *inhash)
>  {
> -       SHASH_DESC_ON_STACK(hash_desc, c->hash_tfm);
>         struct ubifs_auth_node *auth = node;
>         u8 *hash;
>         int err;
> @@ -87,13 +86,17 @@ int ubifs_prepare_auth_node(struct ubifs_info *c, void *node,
>         if (!hash)
>                 return -ENOMEM;
>
> -       hash_desc->tfm = c->hash_tfm;
> -       hash_desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
> -       ubifs_shash_copy_state(c, inhash, hash_desc);
> +       {
> +               SHASH_DESC_ON_STACK(hash_desc, c->hash_tfm);
>
> -       err = crypto_shash_final(hash_desc, hash);
> -       if (err)
> -               goto out;
> +               hash_desc->tfm = c->hash_tfm;
> +               hash_desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
> +               ubifs_shash_copy_state(c, inhash, hash_desc);
> +
> +               err = crypto_shash_final(hash_desc, hash);
> +               if (err)
> +                       goto out;
> +       }
>
>         err = ubifs_hash_calc_hmac(c, hash, auth->hmac);
>         if (err)
> --
> 2.20.0
>


-- 
Thanks,
~Nick Desaulniers