[PATCH] builtin: implement __builtin_strlen() for constants

[PATCH] builtin: implement __builtin_strlen() for constants

[Dan Carpenter]

People are adding compile time asserts to check whether strings are
the expected length.  In GCC and Clang strlen("foo") is expanded at
compile time so this works, but in Sparse it triggers a "bad constant
expression" warning.  Implement expand_strlen() to handle string
literals.

Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Link: https://lore.kernel.org/all/CANiDSCsBAq3Yx4ybarUb_1NkQ-bvfXvWqb-DfqXatkiYJFZWiQ@mail.gmail.com/
---
 builtin.c | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/builtin.c b/builtin.c
index 3a29c3aec8a6..9e8fa5964b37 100644
--- a/builtin.c
+++ b/builtin.c
@@ -596,6 +596,28 @@ static struct symbol_op object_size_op = {
 	.expand = expand_object_size,
 };
 
+static int expand_strlen(struct expression *expr, int cost)
+{
+	struct expression *arg = first_expression(expr->args);
+
+	if (!arg)
+		return UNSAFE;
+	if (arg->type == EXPR_SYMBOL)
+		arg = arg->symbol->initializer;
+	if (!arg || arg->type != EXPR_STRING || !arg->string->length)
+		return UNSAFE;
+
+	expr->flags |= CEF_SET_ICE;
+	expr->type = EXPR_VALUE;
+	expr->value = arg->string->length - 1;
+	expr->taint = 0;
+	return 0;
+}
+
+static struct symbol_op strlen_op = {
+	.expand = expand_strlen,
+};
+
 /*
  * Builtin functions
  */
@@ -775,7 +797,7 @@ static const struct builtin_fn builtins_common[] = {
 	{ "__builtin_strcpy", &string_ctype, 0, { &string_ctype, &const_string_ctype }},
 	{ "__builtin_strcspn", size_t_ctype, 0, { &const_string_ctype, &const_string_ctype }},
 	{ "__builtin_strdup", &string_ctype, 0, { &const_string_ctype }},
-	{ "__builtin_strlen", size_t_ctype, 0, { &const_string_ctype }},
+	{ "__builtin_strlen", size_t_ctype, 0, { &const_string_ctype }, .op = &strlen_op},
 	{ "__builtin_strncasecmp", &int_ctype, 0, { &const_string_ctype, &const_string_ctype, size_t_ctype }},
 	{ "__builtin_strncat", &string_ctype, 0, { &string_ctype, &const_string_ctype, size_t_ctype }},
 	{ "__builtin_strncmp", &int_ctype, 0, { &const_string_ctype, &const_string_ctype, size_t_ctype }},
-- 
2.51.0

Re: [PATCH] builtin: implement __builtin_strlen() for constants

[Linus Torvalds]

On Wed, 15 Oct 2025 at 06:09, Dan Carpenter <dan.carpenter@linaro.org> wrote:
>
> People are adding compile time asserts to check whether strings are
> the expected length.  In GCC and Clang strlen("foo") is expanded at
> compile time so this works, but in Sparse it triggers a "bad constant
> expression" warning.  Implement expand_strlen() to handle string
> literals.

Ack.

Except it's not quite right.

Try this:

    int i(void)
    {
        return __builtin_strlen("hello\0hi");
    }

and you'll see that it returns 8, even though the correct string length is 5.

So you should add a

    #include <string.h>

at the top, and do something like

-       expr->value = arg->string->length - 1;
+       expr->value = strlen(arg->string->data);

in there instead, because constant strings can have embedded NUL characters.

           Linus

Re: [PATCH] builtin: implement __builtin_strlen() for constants

[Dan Carpenter]

On Wed, Oct 15, 2025 at 08:13:51AM -0700, Linus Torvalds wrote:
> So you should add a
> 
>     #include <string.h>
> 
> at the top, and do something like
> 
> -       expr->value = arg->string->length - 1;
> +       expr->value = strlen(arg->string->data);
> 
> in there instead, because constant strings can have embedded NUL characters.
> 
>            Linus

Ah.  Thanks.

regards,
dan carpenter

Re: [PATCH] builtin: implement __builtin_strlen() for constants

[Chris Li]

Thanks,

Will apply with Linus' fix up.

Chris

On Thu, Oct 16, 2025 at 1:29 AM Dan Carpenter <dan.carpenter@linaro.org> wrote:
>
> On Wed, Oct 15, 2025 at 08:13:51AM -0700, Linus Torvalds wrote:
> > So you should add a
> >
> >     #include <string.h>
> >
> > at the top, and do something like
> >
> > -       expr->value = arg->string->length - 1;
> > +       expr->value = strlen(arg->string->data);
> >
> > in there instead, because constant strings can have embedded NUL characters.
> >
> >            Linus
>
> Ah.  Thanks.
>
> regards,
> dan carpenter
>