niu_get_and_validate_port() obtains parent->num_ports from hardware
(nr64(ESPC_NUM_PORTS_MACS) & ESPC_NUM_PORTS_MACS_VAL), an 8-bit value that
can range from 0 to 255, with no upper-bound check.  The driver's per-port
arrays (rxchan_per_port[], txchan_per_port[], rdc_group_cfg[]) are only
NIU_MAX_PORTS (4) entries, so niu_divide_channels() and
niu_divide_rdc_groups() iterate num_ports times and write out of bounds
whenever num_ports is greater than 4.

Reject num_ports > NIU_MAX_PORTS with -EINVAL so the probe aborts before any
of these arrays are written.

Covers fuzzing bug IDs 321, 322, 323 and 324 (same root cause).

Signed-off-by: Yang Zi <2959243019@qq.com>
---
diff --git a/drivers/net/ethernet/sun/niu.c b/drivers/net/ethernet/sun/niu.c
index 54dd7281191d..d33c87489bd9 100644
--- a/drivers/net/ethernet/sun/niu.c
+++ b/drivers/net/ethernet/sun/niu.c
@@ -8586,6 +8586,14 @@ static int niu_get_and_validate_port(struct niu *np)
         }
     }
 
+    /* num_ports may come from an 8-bit hardware field (ESPC_NUM_PORTS_MACS),
+     * so validate it against the fixed per-port array sizes (NIU_MAX_PORTS)
+     * to avoid out-of-bounds accesses in niu_divide_channels() and
+     * niu_divide_rdc_groups().
+     */
+    if (parent->num_ports > NIU_MAX_PORTS)
+        return -EINVAL;
+
     if (np->port >= parent->num_ports)
         return -ENODEV;
 

On Tue, 25 Aug 2026 17:42:42 +0800 Yang Zi wrote:
> niu_get_and_validate_port() obtains parent->num_ports from hardware
> (nr64(ESPC_NUM_PORTS_MACS) & ESPC_NUM_PORTS_MACS_VAL), an 8-bit value that
> can range from 0 to 255, with no upper-bound check.  The driver's per-port
> arrays (rxchan_per_port[], txchan_per_port[], rdc_group_cfg[]) are only
> NIU_MAX_PORTS (4) entries, so niu_divide_channels() and
> niu_divide_rdc_groups() iterate num_ports times and write out of bounds
> whenever num_ports is greater than 4.
> 
> Reject num_ports > NIU_MAX_PORTS with -EINVAL so the probe aborts before any
> of these arrays are written.
> 
> Covers fuzzing bug IDs 321, 322, 323 and 324 (same root cause).
> 
> Signed-off-by: Yang Zi <2959243019@qq.com>

this is a waste of time, please don't send any more patches to linux
networking this year