linux-kernel - Re: [PATCH 3/4] mm: Add free()

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

Message-ID: <20180323143435.GB5624@bombadil.infradead.org>
Date:   Fri, 23 Mar 2018 07:34:35 -0700
From:   Matthew Wilcox <willy@...radead.org>
To:     Rasmus Villemoes <linux@...musvillemoes.dk>
Cc:     linux-mm@...ck.org, Kirill Tkhai <ktkhai@...tuozzo.com>,
        Matthew Wilcox <mawilcox@...rosoft.com>,
        linux-kernel@...r.kernel.org,
        "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
Subject: Re: [PATCH 3/4] mm: Add free()

On Fri, Mar 23, 2018 at 09:04:10AM +0100, Rasmus Villemoes wrote:
> On 2018-03-22 20:58, Matthew Wilcox wrote:
> > From: Matthew Wilcox <mawilcox@...rosoft.com>
> > 
> > free() can free many different kinds of memory.
> 
> I'd be a bit worried about using that name. gcc very much knows about
> the C standard's definition of that function, as can be seen on
> godbolt.org by compiling
> 
> void free(const void *);
> void f(void)
> {
>     free((void*)0);
> }
> 
> with -O2 -Wall -Wextra -c. Anything from 4.6 onwards simply compiles this to
> 
> f:
>  repz retq
> 
> And sure, your free() implementation obviously also has that property,
> but I'm worried that they might one day decide to warn about the
> prototype mismatch (actually, I'm surprised it doesn't warn now, given
> that it obviously pretends to know what free() function I'm calling...),
> or make some crazy optimization that will break stuff in very subtle ways.
> 
> Also, we probably don't want people starting to use free() (or whatever
> name is chosen) if they do know the kind of memory they're freeing?
> Maybe it should not be advertised that widely (i.e., in kernel.h).

All that you've said I see as an advantage, not a disadvantage.
Maybe I should change the prototype to match the userspace
free(), although gcc is deliberately lax about the constness of
function arguments when determining compatibility with builtins.
See match_builtin_function_types() if you're really curious.

gcc already does some nice optimisations around free().  For example, it
can eliminate dead stores:

#include <stdlib.h>

void f(char *foo)
{
	foo[1] = 3;
	free(foo);
}

becomes:

0000000000000000 <f>:
   0:	e9 00 00 00 00       	jmpq   5 <f+0x5>
			1: R_X86_64_PLT32	free-0x4

You can see more things it knows about free() by grepping for
BUILT_IN_FREE.  

I absolutely do want to see people using free() instead of kfree()
if it's not important that the memory was kmalloced.  I wouldn't go
through and change existing code, but I do want to see
#define malloc(x) kvmalloc((x), GFP_KERNEL)