From: David Laight <david.laight.linux@gmail.com>
To: Kees Cook <kees@kernel.org>
Cc: linux-hardening@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH next 2/3] fortify: Optimise strnlen()
Date: Fri, 3 Apr 2026 09:50:08 +0100 [thread overview]
Message-ID: <20260403095008.6efbaf11@pumpkin> (raw)
In-Reply-To: <202603311650.A59396A@keescook>
On Tue, 31 Mar 2026 16:51:26 -0700
Kees Cook <kees@kernel.org> wrote:
> On Tue, Mar 31, 2026 at 11:09:14PM +0100, David Laight wrote:
> > Any uses should be replaced by __builtin_strlen().
>
> When I looked at this before, __builtin_strlen() flip to run-time strlen
> on non-constant strings, which is why I had to jump through all the
> hoops to avoid calling it in those cases.
>
Thinks further.
Can you remember anywhere where:
len = __builtin_strlen(x);
if (__builtin_constant_p(len))
...
actually called strlen() for a non-constant string.
I did do some tests and it was always optimised away.
I might try getting all this code to use a renamed strlen() and
then scan the entire kernel for references to strlen() itself.
There might be a small number of valid ones, but I'd expect most
would come from the compiler.
(Or get the compiler to generate 'rep scasb' and look for that.)
I suspect it might be enough to check that both str and str[0]
are constant before calling __builtin_strlen() and then check
the returned length is constant.
All the checks might be needed for:
str = cond ? "four" : "f\0ur";
since the compile might realise that str[0] is always 'f' and
str[4] always 0 - but strlen differs.
However I suspect that __builtin_constant_p(array[index]) currently
requires that both the array and index are constant.
So testing array[0] is equivalent.
Given it needs all the separate paths, writing strscpy with:
if (__builtin_constant_p(src[0]) {
len = __builtin_strlen(src);
if (__builtin_constant_p(len)) {
/* code for constant length */
return xxx;
}
}
/* code for non-constant length */
One thing I did notice is that for:
char src[32];
char dst[32];
void func(void)
{
strscpy(dst, src, 32);
}
it seems to generate a call to strnlen() followed by a call to
strscpy_sized().
That seems wrong, since all three lengths are 32 it should be
safe to just call strscpy_sized().
And having done the strnlen() it ought to use memcpy().
But, really most of that ought to be moved into the called function.
So you want:
int strcpy_sized(char *dst, const char *src, size_t dst_len, size_t src_len);
where the wrapper fills in src_len.
David
next prev parent reply other threads:[~2026-04-03 8:50 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-30 13:20 [PATCH next 0/3] fortify: Minor changes to strlen() and strnlen() david.laight.linux
2026-03-30 13:20 ` [PATCH next 1/3] fortify: replace __compiletime_lessthan() with statically_true() david.laight.linux
2026-03-30 23:50 ` Kees Cook
2026-03-30 13:20 ` [PATCH next 2/3] fortify: Optimise strnlen() david.laight.linux
2026-03-30 23:54 ` Kees Cook
2026-03-31 22:09 ` David Laight
2026-03-31 23:51 ` Kees Cook
2026-04-01 13:48 ` David Laight
2026-04-03 8:50 ` David Laight [this message]
2026-03-31 6:36 ` Kees Cook
2026-03-31 10:14 ` David Laight
2026-03-31 14:55 ` David Laight
2026-03-31 15:56 ` Kees Cook
2026-04-01 0:15 ` kernel test robot
2026-04-03 8:23 ` David Laight
2026-03-30 13:20 ` [PATCH next 3/3] fortify: Simplify strlen() logic david.laight.linux
2026-03-31 6:07 ` Kees Cook
2026-03-31 8:58 ` David Laight
2026-03-31 6:18 ` Kees Cook
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260403095008.6efbaf11@pumpkin \
--to=david.laight.linux@gmail.com \
--cc=kees@kernel.org \
--cc=linux-hardening@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox