[v3,next] pcmcia: synclink_cs: replace 1-element array with flex-array member

  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. While at it, fix an edge case which could cause
rx_buf_count to be 0 when max_frame_size was set to the maximum
allowed value (65535).

It's worth mentioning that struct RXBUF was allocating 1 byte "too much"
for what is required (ignoring bytes added by padding).

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>
---
Changelog:

- v3:
  fix size calculation mistakes using overflow.h macros: (Req: Andy
  Shevchenko, Kees Cook)
  add notes struct RXBUF size (Kees Cook)

- v2: removed changes to how the size of RXBUF was calculated. I
  changed my mind after thinking about the existing padding in the
  struct. Happy to discuss it if anyone sees it differently.

- v1: https://lore.kernel.org/lkml/Y5mMWEtHWKOiPVU+@mail.google.com/
---
 drivers/char/pcmcia/synclink_cs.c | 33 +++++++++++++++++++------------
 1 file changed, 20 insertions(+), 13 deletions(-)
  

On Sat, Dec 17, 2022 at 11:59:06AM +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. While at it, fix an edge case which could cause
> rx_buf_count to be 0 when max_frame_size was set to the maximum
> allowed value (65535).
> 
> It's worth mentioning that struct RXBUF was allocating 1 byte "too much"
> for what is required (ignoring bytes added by padding).

What was the result of using __packed to make sure there wasn't a sizing
error?

-Kees

> 
> 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>
> ---
> Changelog:
> 
> - v3:
>   fix size calculation mistakes using overflow.h macros: (Req: Andy
>   Shevchenko, Kees Cook)
>   add notes struct RXBUF size (Kees Cook)
> 
> - v2: removed changes to how the size of RXBUF was calculated. I
>   changed my mind after thinking about the existing padding in the
>   struct. Happy to discuss it if anyone sees it differently.
> 
> - v1: https://lore.kernel.org/lkml/Y5mMWEtHWKOiPVU+@mail.google.com/
> ---
>  drivers/char/pcmcia/synclink_cs.c | 33 +++++++++++++++++++------------
>  1 file changed, 20 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/char/pcmcia/synclink_cs.c b/drivers/char/pcmcia/synclink_cs.c
> index b2735be81ab2..eee6772a0978 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 */
> @@ -229,12 +229,18 @@ typedef struct _mgslpc_info {
>  } MGSLPC_INFO;
>  
>  #define MGSLPC_MAGIC 0x5402
> +#define MGSLPC_MAX_FRAME_SIZE 65535
> +#define MGSLPC_MIN_FRAME_SIZE 4096
>  
>  /*
>   * The size of the serial xmit buffer is 1 page, or 4096 bytes
>   */
>  #define TXBUFSIZE 4096
>  
> +/*
> + * RXBUF accommodates at least 1 buffer (header+data) of MGSLPC_MAX_FRAME_SIZE
> + */
> +#define RXBUF_MAX_SIZE (sizeof(RXBUF) + MGSLPC_MAX_FRAME_SIZE)
>  
>  #define CHA     0x00   /* channel A offset */
>  #define CHB     0x40   /* channel B offset */
> @@ -529,7 +535,7 @@ static int mgslpc_probe(struct pcmcia_device *link)
>  	tty_port_init(&info->port);
>  	info->port.ops = &mgslpc_port_ops;
>  	INIT_WORK(&info->task, bh_handler);
> -	info->max_frame_size = 4096;
> +	info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
>  	init_waitqueue_head(&info->status_event_wait_q);
>  	init_waitqueue_head(&info->event_wait_q);
>  	spin_lock_init(&info->lock);
> @@ -2611,19 +2617,20 @@ 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;
> +	if (check_add_overflow(sizeof(RXBUF), info->max_frame_size, &info->rx_buf_size))
> +		return -EINVAL;
>  
> -	/* calculate total allocation size for 8 buffers */
> -	info->rx_buf_total_size = info->rx_buf_size * 8;
> +	/* try to alloc as many buffers that can fit within RXBUF_MAX_SIZE (up to 8) */
> +	if (check_mul_overflow(info->rx_buf_size, 8, &info->rx_buf_total_size))
> +		return -EINVAL;
>  
> -	/* limit total allocated memory */
> -	if (info->rx_buf_total_size > 0x10000)
> -		info->rx_buf_total_size = 0x10000;
> +	if (info->rx_buf_total_size > RXBUF_MAX_SIZE)
> +		info->rx_buf_total_size = RXBUF_MAX_SIZE;
>  
>  	/* calculate number of buffers */
>  	info->rx_buf_count = info->rx_buf_total_size / info->rx_buf_size;
>  
> -	info->rx_buf = kmalloc(info->rx_buf_total_size, GFP_KERNEL);
> +	info->rx_buf = kcalloc(info->rx_buf_count, info->rx_buf_size, GFP_KERNEL);
>  	if (info->rx_buf == NULL)
>  		return -ENOMEM;
>  
> @@ -2695,10 +2702,10 @@ static int mgslpc_add_device(MGSLPC_INFO *info)
>  		current_dev->next_device = info;
>  	}
>  
> -	if (info->max_frame_size < 4096)
> -		info->max_frame_size = 4096;
> -	else if (info->max_frame_size > 65535)
> -		info->max_frame_size = 65535;
> +	if (info->max_frame_size < MGSLPC_MIN_FRAME_SIZE)
> +		info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
> +	else if (info->max_frame_size > MGSLPC_MAX_FRAME_SIZE)
> +		info->max_frame_size = MGSLPC_MAX_FRAME_SIZE;
>  
>  	printk("SyncLink PC Card %s:IO=%04X IRQ=%d\n",
>  		info->device_name, info->io_base, info->irq_level);
> -- 
> 2.38.1
>
  

On Fri, Dec 16, 2022 at 03:42:47PM -0800, Kees Cook wrote:
> On Sat, Dec 17, 2022 at 11:59:06AM +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. While at it, fix an edge case which could cause
> > rx_buf_count to be 0 when max_frame_size was set to the maximum
> > allowed value (65535).
> > 
> > It's worth mentioning that struct RXBUF was allocating 1 byte "too much"
> > for what is required (ignoring bytes added by padding).
> 
> What was the result of using __packed to make sure there wasn't a sizing
> error?
> 
> -Kees
> 

With or without __packed__ attribute, sufficient space would be
allocated which is good :-)

In both cases there is still some "extra space" (1 byte on __packed and
4 bytes on non-packed) but that should be negligible. OTOH, if I'm asked
to cull those bytes I am happy to do it too.

pahole -C RXBUF non-packed/drivers/char/pcmcia/synclink_cs.o
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;

pahole -C RXBUF packed/drivers/char/pcmcia/synclink_cs.o
typedef struct {
	int                        count;                /*     0     4 */
	unsigned char              status;               /*     4     1 */
	char                       data[];               /*     5     0 */

	/* size: 5, cachelines: 1, members: 3 */
	/* last cacheline: 5 bytes */
} __attribute__((__packed__)) RXBUF;

- Paulo A.

> > 
> > 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>
> > ---
> > Changelog:
> > 
> > - v3:
> >   fix size calculation mistakes using overflow.h macros: (Req: Andy
> >   Shevchenko, Kees Cook)
> >   add notes struct RXBUF size (Kees Cook)
> > 
> > - v2: removed changes to how the size of RXBUF was calculated. I
> >   changed my mind after thinking about the existing padding in the
> >   struct. Happy to discuss it if anyone sees it differently.
> > 
> > - v1: https://lore.kernel.org/lkml/Y5mMWEtHWKOiPVU+@mail.google.com/
> > ---
> >  drivers/char/pcmcia/synclink_cs.c | 33 +++++++++++++++++++------------
> >  1 file changed, 20 insertions(+), 13 deletions(-)
> > 
> > diff --git a/drivers/char/pcmcia/synclink_cs.c b/drivers/char/pcmcia/synclink_cs.c
> > index b2735be81ab2..eee6772a0978 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 */
> > @@ -229,12 +229,18 @@ typedef struct _mgslpc_info {
> >  } MGSLPC_INFO;
> >  
> >  #define MGSLPC_MAGIC 0x5402
> > +#define MGSLPC_MAX_FRAME_SIZE 65535
> > +#define MGSLPC_MIN_FRAME_SIZE 4096
> >  
> >  /*
> >   * The size of the serial xmit buffer is 1 page, or 4096 bytes
> >   */
> >  #define TXBUFSIZE 4096
> >  
> > +/*
> > + * RXBUF accommodates at least 1 buffer (header+data) of MGSLPC_MAX_FRAME_SIZE
> > + */
> > +#define RXBUF_MAX_SIZE (sizeof(RXBUF) + MGSLPC_MAX_FRAME_SIZE)
> >  
> >  #define CHA     0x00   /* channel A offset */
> >  #define CHB     0x40   /* channel B offset */
> > @@ -529,7 +535,7 @@ static int mgslpc_probe(struct pcmcia_device *link)
> >  	tty_port_init(&info->port);
> >  	info->port.ops = &mgslpc_port_ops;
> >  	INIT_WORK(&info->task, bh_handler);
> > -	info->max_frame_size = 4096;
> > +	info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
> >  	init_waitqueue_head(&info->status_event_wait_q);
> >  	init_waitqueue_head(&info->event_wait_q);
> >  	spin_lock_init(&info->lock);
> > @@ -2611,19 +2617,20 @@ 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;
> > +	if (check_add_overflow(sizeof(RXBUF), info->max_frame_size, &info->rx_buf_size))
> > +		return -EINVAL;
> >  
> > -	/* calculate total allocation size for 8 buffers */
> > -	info->rx_buf_total_size = info->rx_buf_size * 8;
> > +	/* try to alloc as many buffers that can fit within RXBUF_MAX_SIZE (up to 8) */
> > +	if (check_mul_overflow(info->rx_buf_size, 8, &info->rx_buf_total_size))
> > +		return -EINVAL;
> >  
> > -	/* limit total allocated memory */
> > -	if (info->rx_buf_total_size > 0x10000)
> > -		info->rx_buf_total_size = 0x10000;
> > +	if (info->rx_buf_total_size > RXBUF_MAX_SIZE)
> > +		info->rx_buf_total_size = RXBUF_MAX_SIZE;
> >  
> >  	/* calculate number of buffers */
> >  	info->rx_buf_count = info->rx_buf_total_size / info->rx_buf_size;
> >  
> > -	info->rx_buf = kmalloc(info->rx_buf_total_size, GFP_KERNEL);
> > +	info->rx_buf = kcalloc(info->rx_buf_count, info->rx_buf_size, GFP_KERNEL);
> >  	if (info->rx_buf == NULL)
> >  		return -ENOMEM;
> >  
> > @@ -2695,10 +2702,10 @@ static int mgslpc_add_device(MGSLPC_INFO *info)
> >  		current_dev->next_device = info;
> >  	}
> >  
> > -	if (info->max_frame_size < 4096)
> > -		info->max_frame_size = 4096;
> > -	else if (info->max_frame_size > 65535)
> > -		info->max_frame_size = 65535;
> > +	if (info->max_frame_size < MGSLPC_MIN_FRAME_SIZE)
> > +		info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
> > +	else if (info->max_frame_size > MGSLPC_MAX_FRAME_SIZE)
> > +		info->max_frame_size = MGSLPC_MAX_FRAME_SIZE;
> >  
> >  	printk("SyncLink PC Card %s:IO=%04X IRQ=%d\n",
> >  		info->device_name, info->io_base, info->irq_level);
> > -- 
> > 2.38.1
> > 
> 
> -- 
> Kees Cook
  

On Sat, Dec 17, 2022 at 12:59 AM Paulo Miguel Almeida
<paulo.miguel.almeida.rodenas@gmail.com> 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. While at it, fix an edge case which could cause
> rx_buf_count to be 0 when max_frame_size was set to the maximum
> allowed value (65535).
>
> It's worth mentioning that struct RXBUF was allocating 1 byte "too much"
> for what is required (ignoring bytes added by padding).
>
> 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].

...

>  static int rx_alloc_buffers(MGSLPC_INFO *info)
>  {
>         /* each buffer has header and data */
> -       info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> +       if (check_add_overflow(sizeof(RXBUF), info->max_frame_size, &info->rx_buf_size))
> +               return -EINVAL;
>
> -       /* calculate total allocation size for 8 buffers */
> -       info->rx_buf_total_size = info->rx_buf_size * 8;

> +       /* try to alloc as many buffers that can fit within RXBUF_MAX_SIZE (up to 8) */
> +       if (check_mul_overflow(info->rx_buf_size, 8, &info->rx_buf_total_size))
> +               return -EINVAL;

This check is implied by kcalloc(). But to make it effective we
probably need to get a count first.

> -       /* limit total allocated memory */
> -       if (info->rx_buf_total_size > 0x10000)
> -               info->rx_buf_total_size = 0x10000;
> +       if (info->rx_buf_total_size > RXBUF_MAX_SIZE)
> +               info->rx_buf_total_size = RXBUF_MAX_SIZE;

If max_frame_size > 8192 - sizeof(RXBUF), we bump into this condition...

>         /* calculate number of buffers */
>         info->rx_buf_count = info->rx_buf_total_size / info->rx_buf_size;

...which means that rx_buf_count < 8...
(and if max_frame_size > RXBUF_MAX_SIZE - sizeof(RXBUF), count becomes
0, I don't know if below clamp_val() is the only place to guarantee
that)

> -       info->rx_buf = kmalloc(info->rx_buf_total_size, GFP_KERNEL);
> +       info->rx_buf = kcalloc(info->rx_buf_count, info->rx_buf_size, GFP_KERNEL);

...hence rx_buf size will be less than rx_buf_total_size.

That is probably not an issue per se, but I'm wondering if the
(bigger) value of rx_buf_total_size is the problem further in the
code.

>         if (info->rx_buf == NULL)
>                 return -ENOMEM;

Maybe something like

static int rx_alloc_buffers(MGSLPC_INFO *info)
{
    /* Prevent count from being 0 */
    if (->max_frame_size > MAX_FRAME_SIZE)
        return -EINVAL;
    ...
   count = ...;
   ...
   rx_total_size = ...
   rx_buf = kcalloc(...);

Then you don't need to check overflow with check_add_overflow() and
check_mul_overflow() will be inside the kcalloc.

...

> -       if (info->max_frame_size < 4096)
> -               info->max_frame_size = 4096;
> -       else if (info->max_frame_size > 65535)
> -               info->max_frame_size = 65535;
> +       if (info->max_frame_size < MGSLPC_MIN_FRAME_SIZE)
> +               info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
> +       else if (info->max_frame_size > MGSLPC_MAX_FRAME_SIZE)
> +               info->max_frame_size = MGSLPC_MAX_FRAME_SIZE;

You can use clamp_val() macro here.
  

On Sat, Dec 17, 2022 at 01:43:40PM +0200, Andy Shevchenko wrote:
> On Sat, Dec 17, 2022 at 12:59 AM Paulo Miguel Almeida
> <paulo.miguel.almeida.rodenas@gmail.com> 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. While at it, fix an edge case which could cause
> > rx_buf_count to be 0 when max_frame_size was set to the maximum
> > allowed value (65535).
> >
> > It's worth mentioning that struct RXBUF was allocating 1 byte "too much"
> > for what is required (ignoring bytes added by padding).
> >
> > 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].
> 
> ...
> 
> >  static int rx_alloc_buffers(MGSLPC_INFO *info)
> >  {
> >         /* each buffer has header and data */
> > -       info->rx_buf_size = sizeof(RXBUF) + info->max_frame_size;
> > +       if (check_add_overflow(sizeof(RXBUF), info->max_frame_size, &info->rx_buf_size))
> > +               return -EINVAL;
> >
> > -       /* calculate total allocation size for 8 buffers */
> > -       info->rx_buf_total_size = info->rx_buf_size * 8;
> 
> > +       /* try to alloc as many buffers that can fit within RXBUF_MAX_SIZE (up to 8) */
> > +       if (check_mul_overflow(info->rx_buf_size, 8, &info->rx_buf_total_size))
> > +               return -EINVAL;
> 
> This check is implied by kcalloc(). But to make it effective we
> probably need to get a count first.
> 
> > -       /* limit total allocated memory */
> > -       if (info->rx_buf_total_size > 0x10000)
> > -               info->rx_buf_total_size = 0x10000;
> > +       if (info->rx_buf_total_size > RXBUF_MAX_SIZE)
> > +               info->rx_buf_total_size = RXBUF_MAX_SIZE;
> 
> If max_frame_size > 8192 - sizeof(RXBUF), we bump into this condition...
> 
> >         /* calculate number of buffers */
> >         info->rx_buf_count = info->rx_buf_total_size / info->rx_buf_size;
> 
> ...which means that rx_buf_count < 8...

that's correct. My reading of what the original author intended is the
following:

- rx_buf_count can be < 8 if max_frame_size needs to be > 8192 so that
  userspace tools don't need to collate the different packets together
  then again, SyncLink_CS supports a variety of protocols.

- the more circular buffers, the better, but it looks perfectly acceptable
  to have 1 big rx_buf (max_frame_size possible) if the communication is
  orchestrated nicely (which part sends what and when) especially for
  RS-232-based communications.


> (and if max_frame_size > RXBUF_MAX_SIZE - sizeof(RXBUF), count becomes
> 0, I don't know if below clamp_val() is the only place to guarantee
> that)
> 

I can confirm that the clamp_val() below is the only place that
guarantees the max_frame_size isn't greater than RXBUF_MAX_SIZE. That
happens at the device probing stage: 

( mgslpc_probe > mgslpc_add_device > clamp_val-like routine )

As max_frame_size can only be set as a module parameter and no other way
is exposed to userspace to tweak that afterwards, my 2 cents is that 
clamp_val() routine should be fine as rx_buf_count will always be > 0 
after this fix.

> > -       info->rx_buf = kmalloc(info->rx_buf_total_size, GFP_KERNEL);
> > +       info->rx_buf = kcalloc(info->rx_buf_count, info->rx_buf_size, GFP_KERNEL);
> 
> ...hence rx_buf size will be less than rx_buf_total_size.
> 
> That is probably not an issue per se, but I'm wondering if the
> (bigger) value of rx_buf_total_size is the problem further in the
> code.
> 

rx_buf_total_size isn't used outside of this function so it
could be a local variable IMO.. so I would say that this wouldn't be a
problem.

I had noticed that rx_buf_total_size could be moved into a local
variable before but I thought that removing it from MGSLPC struct
should be part of a separate patch instead.

> >         if (info->rx_buf == NULL)
> >                 return -ENOMEM;
> 
> Maybe something like
> 
> static int rx_alloc_buffers(MGSLPC_INFO *info)
> {
>     /* Prevent count from being 0 */
>     if (->max_frame_size > MAX_FRAME_SIZE)
>         return -EINVAL;

This boils down to whether having the clamp_val() on the probe method is
sufficient in your point of view. You make the final call on this :-)

>     ...
>    count = ...;
>    ...
>    rx_total_size = ...
>    rx_buf = kcalloc(...);
> 
> Then you don't need to check overflow with check_add_overflow() and
> check_mul_overflow() will be inside the kcalloc.
> 

check_mul_overflow point -> agreed.

check_add_overflow -> similar suggestion as my previous point, if the
clamp_val on probe is sufficient for you, I would say that we don't need
it as of now too. But if you still think that we need it, I'm flexible
with that too.

> ...
> 
> > -       if (info->max_frame_size < 4096)
> > -               info->max_frame_size = 4096;
> > -       else if (info->max_frame_size > 65535)
> > -               info->max_frame_size = 65535;
> > +       if (info->max_frame_size < MGSLPC_MIN_FRAME_SIZE)
> > +               info->max_frame_size = MGSLPC_MIN_FRAME_SIZE;
> > +       else if (info->max_frame_size > MGSLPC_MAX_FRAME_SIZE)
> > +               info->max_frame_size = MGSLPC_MAX_FRAME_SIZE;
> 
> You can use clamp_val() macro here.
> 

Nice, I didn't know about this macro. I will make that change for v4.

All really nice points you've made Andy, I'm learning heaps of new
things with this patch :-)

thanks!

- Paulo A.