Message-ID: <Y54hHyoUW/tGioLx@mail.google.com>
Date:   Sun, 18 Dec 2022 09:05:51 +1300
From:   Paulo Miguel Almeida <paulo.miguel.almeida.rodenas@...il.com>
To:     Andy Shevchenko <andy.shevchenko@...il.com>
Cc:     Arnd Bergmann <arnd@...db.de>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>,
        Jiri Slaby <jirislaby@...nel.org>,
        Haowen Bai <baihaowen@...zu.com>, linux-kernel@...r.kernel.org,
        linux-hardening@...r.kernel.org,
        paulo.miguel.almeida.rodenas@...il.com
Subject: Re: [PATCH v3] [next] pcmcia: synclink_cs: replace 1-element array
 with flex-array member

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@...il.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.

Powered by blists - more mailing lists