Thread (2 messages) 2 messages, 2 authors, 6h ago
HOTtoday

[PATCH net] fsl/fman: validate the port id in fman_set_port_params()

From: Weigang He <hidden>
Date: 2026-09-26 05:40:08
Also in: lkml
Subsystem: networking drivers, the rest · Maintainers: Andrew Lunn, "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Linus Torvalds

fman_set_port_params() uses the port id, which comes from the device
tree "cell-index" property, to index fixed-size tables without
validating it:

	set_port_liodn(fman, port_id, fman->liodn_base[port_id],
		       fman->liodn_offset[port_id]);
	...
	mac_id = hw_port_id_to_sw_port_id(fman->state->rev_info.major, port_id);
	if (port_params->max_frame_length >= fman->state->mac_mfl[mac_id]) {
		fman->state->port_mfl[mac_id] = port_params->max_frame_length;

liodn_base[] and liodn_offset[] have FMAN_LIODN_TBL (64) entries, and
set_port_liodn() also accesses fmbm_spliodn[port_id - 1]. mac_mfl[] and
port_mfl[] have MAX_NUM_OF_MACS (10) entries, but
hw_port_id_to_sw_port_id() maps, for example, RX port id 0x12 to 10 and
TX port id 0x3f to 23. fman_port_probe() truncates the property to a u8
and does not otherwise check it, so a device tree with a bad cell-index
can cause out-of-bounds accesses.

Reject a port id of 0, a port id outside the LIODN table, and a port id
that does not map to a MAC, before any state or register is changed. A
cell-index above 0xff that aliases a valid id after the u8 truncation
is not caught here.

Found by static analysis tool CodeQL.

Fixes: 414fd46e7762 ("fsl/fman: Add FMan support")
Assisted-by: LLM codeql
Signed-off-by: Weigang He <redacted>
---

Notes:
    Compile-tested only (ARCH=arm64 and ARCH=x86_64 allmodconfig, W=1). Not
    tested on hardware, and there is no reproducer.
    
    The CodeQL query behind this report was synthesized with LLM assistance,
    and the fix and changelog were drafted with LLM assistance; I have
    reviewed them.

 drivers/net/ethernet/freescale/fman/fman.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/freescale/fman/fman.c b/drivers/net/ethernet/freescale/fman/fman.c
index 46cc28895e565..e25050189b1f3 100644
--- a/drivers/net/ethernet/freescale/fman/fman.c
+++ b/drivers/net/ethernet/freescale/fman/fman.c
@@ -2103,6 +2103,18 @@ int fman_set_port_params(struct fman *fman,
 	unsigned long flags;
 	u8 port_id = port_params->port_id, mac_id;
 
+	/*
+	 * port_id comes from the device tree: reject ids outside the LIODN
+	 * table and ids that do not map to a MAC.
+	 */
+	mac_id = hw_port_id_to_sw_port_id(fman->state->rev_info.major, port_id);
+	if (port_id == 0 || port_id >= FMAN_LIODN_TBL ||
+	    mac_id >= MAX_NUM_OF_MACS) {
+		dev_err(fman->dev, "%s: invalid port id %u\n", __func__,
+			port_id);
+		return -EINVAL;
+	}
+
 	spin_lock_irqsave(&fman->spinlock, flags);
 
 	err = set_num_of_tasks(fman, port_params->port_id,
@@ -2170,8 +2182,6 @@ int fman_set_port_params(struct fman *fman,
 	if (fman->state->rev_info.major < 6)
 		set_port_order_restoration(fman->fpm_regs, port_id);
 
-	mac_id = hw_port_id_to_sw_port_id(fman->state->rev_info.major, port_id);
-
 	if (port_params->max_frame_length >= fman->state->mac_mfl[mac_id]) {
 		fman->state->port_mfl[mac_id] = port_params->max_frame_length;
 	} else {
base-commit: 165768bb70265b5c38cf0b73fafd75be235f8b14
-- 
2.43.0
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help