Re: [PATCHv2] mm: do not export const kfree and kstrdup variants

[Christophe JAILLET]

Le 24/09/2024 à 09:02, Christoph Hellwig a écrit :
> On Tue, Sep 24, 2024 at 03:56:53PM +0900, Sergey Senozhatsky wrote:
> > Totally agree with all the points, I haven't looked at how
> > popular that API was before sending out the patch.  Is there
> > some sort of "built time const" but for strings that we, perhaps,
> > can add to kfree_const() (and make kfree_const() always inline)?
> > So that we can turn this
> >
> > 	str = "boom";
> > 	...
> > 	kfree_const(str);
> >
> > into a safe scenario for modules.
>
> Not sure, but even then the API would be horrible as it still would
> not work for constants in other modules than the one calling it.

So, the best is to audit, at least code that can be built as a module 
for str = "boom" pattern, and fix relevant places?


Or, considering that these these _const() versions are only there to 
save a few bytes of memory, wonder if it really worth it?

In other word:
	1) apply something like the patch below
	2) remove the API


At a minimum, I think that devm_kstrdup_const() could be removed.

CJ


diff --git a/mm/util.c b/mm/util.c
index 4f1275023eb7..fd5e98fb6362 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -39,8 +39,7 @@
  */
 void kfree_const(const void *x)
 {
-       if (!is_kernel_rodata((unsigned long)x))
-               kfree(x);
+       kfree(x);
 }
 EXPORT_SYMBOL(kfree_const);

@@ -81,9 +80,6 @@ EXPORT_SYMBOL(kstrdup);
  */
 const char *kstrdup_const(const char *s, gfp_t gfp)
 {
-       if (is_kernel_rodata((unsigned long)s))
-               return s;
-
        return kstrdup(s, gfp);
 }
 EXPORT_SYMBOL(kstrdup_const);