Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 14 Dec 2022 12:26:01 -0800
From:   Kees Cook <keescook@chromium.org>
To:     Paulo Miguel Almeida <paulo.miguel.almeida.rodenas@gmail.com>
Cc:     Arnd Bergmann <arnd@arndb.de>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Ilpo =?iso-8859-1?Q?J=E4rvinen?= <ilpo.jarvinen@linux.intel.com>,
        Jiri Slaby <jirislaby@kernel.org>,
        Haowen Bai <baihaowen@meizu.com>,
        Andy Shevchenko <andy.shevchenko@gmail.com>,
        linux-kernel@vger.kernel.org, linux-hardening@vger.kernel.org
Subject: Re: [PATCH] [next] pcmcia: synclink_cs: replace 1-element array with
 flex-array member
Message-ID: <202212141224.0D30B9E@keescook>
References: <Y5mMWEtHWKOiPVU+@mail.google.com>
 <202212141124.736E3DE2A8@keescook>
 <Y5otilaHc6HBPCAF@mail.google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y5otilaHc6HBPCAF@mail.google.com>
Precedence: bulk

On Thu, Dec 15, 2022 at 09:09:46AM +1300, Paulo Miguel Almeida wrote:
> On Wed, Dec 14, 2022 at 11:29:37AM -0800, Kees Cook wrote:
> > On Wed, Dec 14, 2022 at 09:42:00PM +1300, Paulo Miguel Almeida wrote:
> > > One-element arrays are deprecated, and we are replacing them with
> > > flexible array members instead. So, replace one-element array with
> > > flexible-array member in struct RXBUF and refactor the rest of the code
> > > accordingly.
> > > 
> > > It's worth mentioning that doing a build before/after this patch
> > > results in no binary output differences.
> > > 
> > > This helps with the ongoing efforts to tighten the FORTIFY_SOURCE
> > > routines on memcpy() and help us make progress towards globally
> > > enabling -fstrict-flex-arrays=3 [1].
> > > 
> > > Link: https://github.com/KSPP/linux/issues/79
> > > Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101836 [1]
> > > 
> > > Signed-off-by: Paulo Miguel Almeida <paulo.miguel.almeida.rodenas@gmail.com>
> > > ---
> > >  drivers/char/pcmcia/synclink_cs.c | 5 +++--
> > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/char/pcmcia/synclink_cs.c b/drivers/char/pcmcia/synclink_cs.c
> > > index b2735be81ab2..1ab2d552f498 100644
> > > --- a/drivers/char/pcmcia/synclink_cs.c
> > > +++ b/drivers/char/pcmcia/synclink_cs.c
> > > @@ -105,7 +105,7 @@ static MGSL_PARAMS default_params = {
> > >  typedef struct {
> > >  	int count;
> > >  	unsigned char status;
> > > -	char data[1];
> > > +	char data[];
> > >  } RXBUF;
> > >  
> > >  /* The queue of BH actions to be performed */
> > > @@ -2611,7 +2611,8 @@ static int mgslpc_proc_show(struct seq_file *m, void *v)
> > >  static int rx_alloc_buffers(MGSLPC_INFO *info)
> > >  {
> > >  	/* each buffer has header and data */
> > > -	info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> > > +	info->rx_buf_size = max(offsetof(typeof(RXBUF), data) + 1, sizeof(RXBUF))
> > > +		+ info->max_frame_size;
> > 
> > It seems like there is an existing size bug here, and likely should be
> > fixed separately?
> > 
> > i.e. this was already allocating 1 byte "too much". I'd expect this
> > first:
> > 
> > -	info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> > +	info->rx_buf_size = sizeof(RXBUF) - 1 + info->max_frame_size;
> > 
> > and then the next patch:
> > 
> > -	char data[1];
> > +	char data[];
> > ...
> > -	info->rx_buf_size = sizeof(RXBUF) - 1 + info->max_frame_size;
> > +	info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> > 
> > The above would induce a binary output change, and the second would not.
> > 
> > Though this results in what you had for the v2 patch (but I can't
> > believe it had no binary changes...)
> > 
> > -- 
> > Kees Cook
> 
> Hi Kees, Hi Andy, Thanks for taking the time to review this patch.
> 
> As both of you had similar points, I will reply them here.
> 
> The reasons why it had no binary changes was because of the combination
> of this 2 things:
> 
> 1) Existing padding - so sizeof(RXBUF) returned 8 bytes in both cases.
> 
> pahole -C RXBUF gcc/before/drivers/char/pcmcia/synclink_cs.ko
> typedef struct {
> 	int                        count;                /*     0     4 */
> 	unsigned char              status;               /*     4     1 */
> 	char                       data[1];              /*     5     1 */
> 
> 	/* size: 8, cachelines: 1, members: 3 */
> 	/* padding: 2 */
> 	/* last cacheline: 8 bytes */
> } RXBUF;
> 
> pahole -C RXBUF gcc/after/drivers/char/pcmcia/synclink_cs.ko
> typedef struct {
> 	int                        count;                /*     0     4 */
> 	unsigned char              status;               /*     4     1 */
> 	char                       data[];               /*     5     0 */
> 
> 	/* size: 8, cachelines: 1, members: 3 */
> 	/* padding: 3 */
> 	/* last cacheline: 8 bytes */
> } RXBUF;

Ah-ha, now I see.

> 
> 2) RXBUF (as implemented now) is just  like a pair of lenses from which a
> developer can have access to one of the circular buffers in MGSLPC_INFO
> struct called 'rx_buf'. 
> 
> 2611 static int rx_alloc_buffers(MGSLPC_INFO *info)
> 2612 {
> 2613         /* each buffer has header and data */
> 2614         info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> 2615
> 2616         /* calculate total allocation size for 8 buffers */
> 2617         info->rx_buf_total_size = info->rx_buf_size * 8;
> 2618
> 2619         /* limit total allocated memory */
> 2620         if (info->rx_buf_total_size > 0x10000)
> 2621                 info->rx_buf_total_size = 0x10000;
> 2622
> 2623         /* calculate number of buffers */
> 2624         info->rx_buf_count = info->rx_buf_total_size / info->rx_buf_size;
> 2625
> 2626         info->rx_buf = kmalloc(info->rx_buf_total_size, GFP_KERNEL);
> 
> To be honest, char data[_1_] in RXBUF was never required to be there.
> The code base seems to make sure that it doesn't run past its limits by
> keeping track of size buffer on MGSLPC_INFO->rx_buf_size (and sometimes
> RXBUF->count)
> 
> (Addressing one point made by Andy about using of of the macros in
> overflow.h)
> 	struct_size(buf, data, 1) would return 9 bytes which could
> 	potentially break the existing driver as it produces binary
> 	changes.

Yeah, I think your v2 is fine. Perhaps explicitly repeat the notes about
struct size padding in a v3 commit log?

-- 
Kees Cook