Re: [PATCH net] nfc: nci: validate packet length when parsing NCI 2.x RF interfaces

On Thu, Jun 11, 2026 at 09:27:16AM -0700, Zijing Yin wrote:
> nci_core_init_rsp_packet_v2() parses the variable-length list of
> supported RF interfaces carried in an NCI 2.x CORE_INIT_RSP without ever
> validating the controller-supplied lengths against the size of the
> received packet.
> 
> Each list entry is a (RF interface, RF extension count, RF extensions[])
> tuple. The loop walks the list using the per-entry extension count
> (rf_extension_cnt, up to 255) taken straight from the packet, so a
> malformed CORE_INIT_RSP can advance the read pointer far past the end of
> the skb data buffer. The stored interface count is clamped to
> NCI_MAX_SUPPORTED_RF_INTERFACES so the write side is bounded, but the
> read side runs off the end of the buffer.
> 
> A malformed CORE_INIT_RSP from the controller, also reachable from user
> space through the virtual NCI device (CONFIG_NFC_VIRTUAL_NCI) once the
> device has entered NCI 2.x mode, therefore makes the parser read past the
> end of the response buffer while walking the interface list, copying the
> out-of-bounds bytes into ndev->supported_rf_interfaces[].
> 
> Reject responses shorter than the fixed part of the structure, and make
> sure each interface entry and its extension bytes lie within the received
> packet before dereferencing them. A truncated or malformed list is
> treated as a syntax error, which fails the CORE_INIT request instead of
> reading out of bounds.
> 
> Fixes: bcd684aace34 ("net/nfc/nci: Support NCI 2.x initial sequence")
> Cc: [email protected]
> Signed-off-by: Zijing Yin <[email protected]>

Reviewed-by: Simon Horman <[email protected]>

FTR, there is an AI-generated review of this patch available on sashiko.dev
However, I believe that the issue flagged there can be considered in the
context of possible follow-up rather than effecting the progress of this
patch.