Re: [PATCH] nfc: st-nci: Fix an incorrect skb_buff size in 'st_nci_i2c_read()'

[Christophe JAILLET]

Le 12/08/2019 Ã 05:57, David Miller a ÃcritÂ:
> From: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
> Date: Tue,  6 Aug 2019 16:16:40 +0200
>
> > In 'st_nci_i2c_read()', we allocate a sk_buff with a size of
> > ST_NCI_I2C_MIN_SIZE + len.
> >
> > However, later on, we first 'skb_reserve()' ST_NCI_I2C_MIN_SIZE bytes, then
> > we 'skb_put()' ST_NCI_I2C_MIN_SIZE bytes.
> > Finally, if 'len' is not 0, we 'skb_put()' 'len' bytes.
> >
> > So we use ST_NCI_I2C_MIN_SIZE*2 + len bytes.
> >
> > This is incorrect and should already panic. I guess that it does not occur
> > because of extra memory allocated because of some rounding.
> >
> > Fix it and allocate enough room for the 'skb_reserve()' and the 'skb_put()'
> > calls.
> >
> > Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
> > ---
> > This patch is LIKELY INCORRECT. So think twice to what is the correct
> > solution before applying it.
> > Maybe the skb_reserve should be axed or some other sizes are incorrect.
> > There seems to be an issue, that's all I can say.
> The skb_reserve() should be removed,

I don't fully understand the potential implications, but looks ok to me.
At least, the allocated memory and the size of the used memory would match.

What I don't understand is why is does not BUG_ON with the current code. 
Does my suspected "over allocation" because of rounding/aligment could 
hide the issue?

A Tested-by: by someone who has the corresponding hardware would also be 
useful IMHO.

>   and the second memcpy() should remove
> the " + ST_NCI_I2C_MIN_SIZE".
Hmm, not sure on this one.

The skb is manipulated only with skb_put. So only the tail pointer and 
len are updated. The data pointer remains at the same position, so there 
should effectively be an offset of ST_NCI_I2C_MIN_SIZE for the 2nd memcpy.

Maybe, using skb_put_data would be cleaner here, in order to 
"concatenate" these 2 parts without having to handle by hand the right 
position in the buffer.

If you agree, I'll send a V2.


Thx for the review and comments.

CJ

> This SKB just get sent down to ndlc_recv() so the content returned from I2C
> should places at skb->data to be processed.
>
> Pretty clear this code was never tested.