All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH net-next v2] net: sparx5: check DT port reg against chip port count
@ 2026-11-16 13:08 Haoran Wang
  2026-11-17 12:59 ` Daniel Machon
  2026-11-18  6:43 ` [PATCH net-next v3] " Haoran Wang
  0 siblings, 2 replies; 3+ messages in thread
From: Haoran Wang @ 2026-11-16 13:08 UTC (permalink / raw)
  To: Daniel Machon, Steen Hegelund, UNGLinuxDriver
  Cc: netdev, linux-arm-kernel, linux-kernel, Andrew Lunn,
	David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni

mchp_sparx5_probe() reads the DT ethernet-port 'reg' property into
portno, which sparx5_create_port() later uses to index the fixed-size
sparx5->ports[] array:

    sparx5->ports[config->portno] = spx5_port;

The value is never validated, so an out-of-range 'reg' (runtime overlay,
corrupted or attacker-controlled DTB, kexec-supplied or bootloader
fixed-up DT) turns probe into an out-of-bounds write.

Skip port nodes with an out-of-range 'reg', as lan966x does in
lan966x_probe_port(). The bound is the per-chip port count from the
match data (sparx5->data->consts->n_ports), so lan969x, which shares
this probe path and has only 30 ports, is bounded correctly as well.

Found by a scripted scan of of_property_read_u32() call sites used to
index fixed-size arrays, followed by manual review. Verified at runtime
on a pristine v7.4-rc2 arm64 QEMU virt machine (inline KASAN,
CONFIG_SPARX5_SWITCH=m): an out-of-tree module supplies only a
malicious DT overlay, the unmodified in-tree driver performs the store.
reg = 15839 yields a KASAN use-after-free, reg = 0x20000000 a wild
write 4 GiB out (paging oops), and reg = 65 silent in-struct corruption
with a later NULL oops. With this patch all three are skipped. The PoC
and full logs are posted in the thread. Not tested on real hardware,
QEMU virt (cortex-a57) and compile testing only.

Fixes: f3cad2611a77 ("net: sparx5: add hostmode with phylink support")
Signed-off-by: Haoran Wang <wanghr2026@mail.zgclab.edu.cn>
---
v1 -> v2:

- Check against sparx5->data->consts->n_ports instead of SPX5_PORTS,
  so that lan969x (30 ports), which goes through the same probe, is
  bounded correctly as well (Daniel Machon).
- Document how the issue was found, the runtime verification and the
  test setup, as requested by the patch bot.

 drivers/net/ethernet/microchip/sparx5/sparx5_main.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c b/b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
index 08a398a..1647067 100644
--- a/a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
+++ b/b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
@@ -845,6 +845,11 @@ static int mchp_sparx5_probe(struct platform_device *pdev)
 			dev_err(sparx5->dev, "port reg property error\n");
 			continue;
 		}
+		if (portno >= sparx5->data->consts->n_ports) {
+			dev_err(sparx5->dev, "port %u: reg out of range\n",
+				portno);
+			continue;
+		}
 		config = &configs[idx];
 		conf = &config->conf;
 		conf->speed = SPEED_UNKNOWN;

^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH net-next v2] net: sparx5: check DT port reg against chip port count
  2026-11-16 13:08 [PATCH net-next v2] net: sparx5: check DT port reg against chip port count Haoran Wang
@ 2026-11-17 12:59 ` Daniel Machon
  2026-11-18  6:43 ` [PATCH net-next v3] " Haoran Wang
  1 sibling, 0 replies; 3+ messages in thread
From: Daniel Machon @ 2026-11-17 12:59 UTC (permalink / raw)
  To: Haoran Wang
  Cc: Steen Hegelund, UNGLinuxDriver, netdev, linux-arm-kernel,
	linux-kernel, Andrew Lunn, David S. Miller, Eric Dumazet,
	Jakub Kicinski, Paolo Abeni

> mchp_sparx5_probe() reads the DT ethernet-port 'reg' property into
> portno, which sparx5_create_port() later uses to index the fixed-size
> sparx5->ports[] array:
> 
>     sparx5->ports[config->portno] = spx5_port;
> 
> The value is never validated, so an out-of-range 'reg' (runtime overlay,
> corrupted or attacker-controlled DTB, kexec-supplied or bootloader
> fixed-up DT) turns probe into an out-of-bounds write.
> 
> Skip port nodes with an out-of-range 'reg', as lan966x does in
> lan966x_probe_port(). The bound is the per-chip port count from the

Actually, lan966x returns -EINVAL and fails. This patch just skips the port,
which is fine.

> match data (sparx5->data->consts->n_ports), so lan969x, which shares
> this probe path and has only 30 ports, is bounded correctly as well.
> 
> Found by a scripted scan of of_property_read_u32() call sites used to
> index fixed-size arrays, followed by manual review. Verified at runtime
> on a pristine v7.4-rc2 arm64 QEMU virt machine (inline KASAN,
> CONFIG_SPARX5_SWITCH=m): an out-of-tree module supplies only a
> malicious DT overlay, the unmodified in-tree driver performs the store.
> reg = 15839 yields a KASAN use-after-free, reg = 0x20000000 a wild
> write 4 GiB out (paging oops), and reg = 65 silent in-struct corruption
> with a later NULL oops. With this patch all three are skipped. The PoC
> and full logs are posted in the thread. Not tested on real hardware,
> QEMU virt (cortex-a57) and compile testing only.
> 
> Fixes: f3cad2611a77 ("net: sparx5: add hostmode with phylink support")
> Signed-off-by: Haoran Wang <wanghr2026@mail.zgclab.edu.cn>

The in-body From: line from v1 is gone, so the author is now
haoranwangsec@gmail.com, which does not match the Signed-off-by.
checkpatch warns about this.

> ---
> v1 -> v2:
> 
> - Check against sparx5->data->consts->n_ports instead of SPX5_PORTS,
>   so that lan969x (30 ports), which goes through the same probe, is
>   bounded correctly as well (Daniel Machon).
> - Document how the issue was found, the runtime verification and the
>   test setup, as requested by the patch bot.
> 
>  drivers/net/ethernet/microchip/sparx5/sparx5_main.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c b/b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
> index 08a398a..1647067 100644
> --- a/a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
> +++ b/b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
> @@ -845,6 +845,11 @@ static int mchp_sparx5_probe(struct platform_device *pdev)
>                         dev_err(sparx5->dev, "port reg property error\n");
>                         continue;
>                 }
> +               if (portno >= sparx5->data->consts->n_ports) {
> +                       dev_err(sparx5->dev, "port %u: reg out of range\n",
> +                               portno);
> +                       continue;
> +               }
>                 config = &configs[idx];
>                 conf = &config->conf;
>                 conf->speed = SPEED_UNKNOWN;

The fix looks correct, however the patch is broken and does not apply.

pw-bot: cr

^ permalink raw reply	[flat|nested] 3+ messages in thread

* [PATCH net-next v3] net: sparx5: check DT port reg against chip port count
  2026-11-16 13:08 [PATCH net-next v2] net: sparx5: check DT port reg against chip port count Haoran Wang
  2026-11-17 12:59 ` Daniel Machon
@ 2026-11-18  6:43 ` Haoran Wang
  1 sibling, 0 replies; 3+ messages in thread
From: Haoran Wang @ 2026-11-18  6:43 UTC (permalink / raw)
  To: Daniel Machon, Steen Hegelund, UNGLinuxDriver
  Cc: netdev, linux-arm-kernel, linux-kernel, Andrew Lunn,
	David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni

From: Haoran Wang <wanghr2026@mail.zgclab.edu.cn>

mchp_sparx5_probe() reads the DT ethernet-port 'reg' property into
portno, which sparx5_create_port() later uses to index the fixed-size
sparx5->ports[] array:

    sparx5->ports[config->portno] = spx5_port;

The value is never validated, so an out-of-range 'reg' (runtime overlay,
corrupted or attacker-controlled DTB, kexec-supplied or bootloader
fixed-up DT) turns probe into an out-of-bounds write.

Skip port nodes with an out-of-range 'reg'.  lan966x_probe_port() has
the same bounds check but fails probe with -EINVAL; skipping here keeps
the remaining valid ports working.  The bound is the per-chip port count
from the match data (sparx5->data->consts->n_ports), so lan969x, which
shares this probe path and has only 30 ports, is bounded correctly as
well.

Found by a scripted scan of of_property_read_u32() call sites used to
index fixed-size arrays, followed by manual review. Verified at runtime
on a pristine v7.4-rc2 arm64 QEMU virt machine (inline KASAN,
CONFIG_SPARX5_SWITCH=m): an out-of-tree module supplies only a
malicious DT overlay, the unmodified in-tree driver performs the store.
reg = 15839 yields a KASAN use-after-free, reg = 0x20000000 a wild
write 4 GiB out (paging oops), and reg = 65 silent in-struct corruption
with a later NULL oops. With this patch all three are skipped. The PoC
and full logs are posted in the thread. Not tested on real hardware,
QEMU virt (cortex-a57) and compile testing only.

Fixes: f3cad2611a77 ("net: sparx5: add hostmode with phylink support")
Signed-off-by: Haoran Wang <wanghr2026@mail.zgclab.edu.cn>
---
v2 -> v3:
- Fix the diff paths (a/a/ -> a/, b/b/ -> b/) so the patch applies
  (Daniel Machon).
- Restore the in-body From: line so the author matches the Signed-off-by
  (checkpatch author/Signed-off-by mismatch).
- Clarify that lan966x fails probe instead of skipping (Daniel Machon).

v1 -> v2:
- Check against sparx5->data->consts->n_ports instead of SPX5_PORTS,
  so that lan969x (30 ports), which goes through the same probe, is
  bounded correctly as well (Daniel Machon).
- Document how the issue was found, the runtime verification and the
  test setup, as requested by the patch bot.

 drivers/net/ethernet/microchip/sparx5/sparx5_main.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
index 08a398adf..164706745 100644
--- a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
+++ b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
@@ -845,6 +845,11 @@ static int mchp_sparx5_probe(struct platform_device *pdev)
 			dev_err(sparx5->dev, "port reg property error\n");
 			continue;
 		}
+		if (portno >= sparx5->data->consts->n_ports) {
+			dev_err(sparx5->dev, "port %u: reg out of range\n",
+				portno);
+			continue;
+		}
 		config = &configs[idx];
 		conf = &config->conf;
 		conf->speed = SPEED_UNKNOWN;



^ permalink raw reply related	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-18  8:22 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 13:08 [PATCH net-next v2] net: sparx5: check DT port reg against chip port count Haoran Wang
2026-11-17 12:59 ` Daniel Machon
2026-11-18  6:43 ` [PATCH net-next v3] " Haoran Wang

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.