Re: [PATCH] nfc: llcp: fix integer underflow and missing bounds checks in TLV parsing

On Tue, Jun 02, 2026 at 01:58:55PM +0100, David Laight wrote:
> On Mon, 25 May 2026 22:24:27 +0200
> Doruk Tan Ozturk <[email protected]> wrote:
> 
> > Multiple out-of-bounds read vulnerabilities exist in the NFC LLCP TLV
> > parsers:
> > 
> > 1. In nfc_llcp_recv_snl(), when an SDREQ TLV has length == 0,
> >    service_name_len = length - 1 underflows to SIZE_MAX (size_t is
> >    unsigned). The subsequent strncmp() and nfc_llcp_sock_from_sn()
> >    calls then read unbounded kernel heap memory.
> > 
> > 2. All LLCP TLV parsing loops (nfc_llcp_recv_snl, nfc_llcp_connect_sn,
> >    nfc_llcp_parse_gb_tlv, nfc_llcp_parse_connection_tlv) read tlv[0]
> >    and tlv[1] without first verifying that at least 2 bytes remain in
> >    the buffer.
> > 
> > A nearby malicious NFC device can trigger these without authentication --
> > LLCP link activation happens automatically after NFC-DEP.
> > 
> > Fix by adding a minimum length check before the subtraction in the
> > SDREQ case, and adding bounds validation at the top of each TLV loop
> > iteration.
> 
> Why not fix it properly so that it doesn't read beyond the end of the skb.
> 1) Change offset and tlv_len to be 32bit - no point trying to do 16bit mathx.
> 2) Worry about non-linear skb - perhaps just process skb_header_len() bytes.
> 3) Allow for very short frames.
> 4) Don't read beyond the end of the skb data.
> 
> Something like:
> 	tlv = skb->data + LLCP_HEADER_SIZE;
> 	tlv_end = skb_tail_pointer(skb);
> 	for (; tlv + 2 < tlv_end; tlv += 2 + length) {
> 		type = tlv[0];
> 		length = tlv[1];
> 		if (tlv + 2 + length > tlv_end)
> 			break;
> 
> Then as well as the check for length >= 1 in SDREQ (even 1 doesn't make sense),
> the SDRES path is missing a check for length >= 2.

[ just moved [email protected] to Bcc since this is public discussion, please
  don't re-add it ]

Thanks,
Willy