* [PATCH net 1/2] net: phy: restore timestamping state on probe failure
2026-11-18 10:33 [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure Kimberly Morris
@ 2026-11-18 10:34 ` Kimberly Morris
2026-11-18 10:34 ` [PATCH net 2/2] net: phy: broadcom: release PTP clock with PHY devres Kimberly Morris
2026-11-18 10:38 ` [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure netdev-bot+sinfo
2 siblings, 0 replies; 5+ messages in thread
From: Kimberly Morris @ 2026-11-18 10:34 UTC (permalink / raw)
To: andrew, hkallweit1, linux, florian.fainelli
Cc: bcm-kernel-feedback-list, davem, edumazet, kuba, pabeni,
richardcochran, netdev, linux-kernel, gregkh, w
A PHY driver can install a devm-backed mii_timestamper and select it as
the default before its probe sequence has completed. If a later probe
step fails, the driver core releases the driver's devres, but the
phy_device retains both timestamping fields.
If the generic PHY driver later attaches, skb_defer_rx_timestamp() can
dereference the freed mii_timestamper. This also affects non-PTP packets
because the receive path tests hwtstamp_set before classifying the
packet.
Save the incoming mii_ts and default_timestamp state and restore it on
every probe error path. Restoring the previous values also preserves a
timestamper installed externally before the probe.
On an x86_64 KASAN kernel, a KUnit reproducer using a real mii_bus,
phy_device and phy_driver triggered a slab use-after-free after the
failed probe, driver-core devres cleanup and generic-PHY fallback. The
same test passes after this change.
Fixes: 39db6be781cd ("net: phy: broadcom: Add PTP support for some
Broadcom PHYs.")
Cc: stable@vger.kernel.org
Signed-off-by: Kimberly Morris <gy770616@gmail.com>
---
drivers/net/phy/phy_device.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
index b39e945..4017568 100644
--- a/drivers/net/phy/phy_device.c
+++ b/drivers/net/phy/phy_device.c
@@ -3812,8 +3812,10 @@ EXPORT_SYMBOL_GPL(fwnode_get_phy_node);
static int phy_probe(struct device *dev)
{
struct phy_device *phydev = to_phy_device(dev);
+ struct mii_timestamper *mii_ts = phydev->mii_ts;
struct device_driver *drv = phydev->mdio.dev.driver;
struct phy_driver *phydrv = to_phy_driver(drv);
+ bool default_timestamp = phydev->default_timestamp;
int err = 0;
phydev->drv = phydrv;
@@ -3951,6 +3953,9 @@ static int phy_probe(struct device *dev)
phydev->drv->remove(phydev);
out_reset:
+ phydev->mii_ts = mii_ts;
+ phydev->default_timestamp = default_timestamp;
+
/* Re-assert the reset signal on error */
phy_device_reset(phydev, 1);
--
2.55.0.windows.3
On Wed, 18 Nov 2026 05:33:45 -0500, Kimberly Morris <gy770616@gmail.com> wrote:
> Hello,
>
> Following Greg's request for an applyable patch, this series addresses the
> automatic PHY probe-failure lifetime path from the earlier report. It does
> not rely on the root-only sysfs driver-unbind trigger.
>
> A PHY driver can publish a devres-backed mii_timestamper before its probe
> has completed. If a later probe step fails, driver-core cleanup frees the
> owner while the phy_device can retain the published timestamping state. A
> later generic-PHY fallback can then reach the freed object from
> skb_defer_rx_timestamp(). Broadcom BCM54210E/BCM54213PE also leaves a
> registered PHC backed by the same freed storage.
>
> Patch 1 restores the timestamping state that existed before phy_probe().
> Patch 2 adds the missing managed Broadcom PHC and timestamper cleanup.
>
> I reproduced the first issue on an x86_64 KASAN kernel with a KUnit test
> that registers a real mii_bus, phy_device and phy_driver, follows the real
> failed-probe and devres cleanup paths, attaches the generic PHY driver, and
> passes a non-PTP skb to skb_defer_rx_timestamp(). Before the fix KASAN
> reported a slab use-after-free; after the fix the test passed without a
> KASAN report.
>
> Tested on:
>
> - upstream master 20eef29b7c8465ac98d7dd6d6f2ac07a8d8056ca
> - net/main 2a324a58a084c5cfd7e5a2f55e0786bee170a4a5
>
> The submitted series is generated against current upstream master
> 63fdb511c27afcebce797e3db580b3534ddfa639. I have not tested it on physical
> BCM54213PE hardware, and I am not claiming an attacker-controlled probe
> failure, privilege escalation, or code execution.
>
> Please let me know if the rollback belongs in individual PHY drivers rather
> than phylib core.
>
> Thank you.
>
> Kimberly Morris (2):
> net: phy: restore timestamping state on probe failure
> net: phy: broadcom: release PTP clock with PHY devres
>
> drivers/net/phy/bcm-phy-ptp.c | 19 +++++++++++++++++++
> drivers/net/phy/phy_device.c | 5 +++++
> 2 files changed, 24 insertions(+)
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH net 2/2] net: phy: broadcom: release PTP clock with PHY devres
2026-11-18 10:33 [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure Kimberly Morris
2026-11-18 10:34 ` [PATCH net 1/2] net: phy: restore timestamping state on " Kimberly Morris
@ 2026-11-18 10:34 ` Kimberly Morris
2026-11-18 10:38 ` [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure netdev-bot+sinfo
2 siblings, 0 replies; 5+ messages in thread
From: Kimberly Morris @ 2026-11-18 10:34 UTC (permalink / raw)
To: andrew, hkallweit1, linux, florian.fainelli
Cc: bcm-kernel-feedback-list, davem, edumazet, kuba, pabeni,
richardcochran, netdev, linux-kernel, gregkh, w
bcm_ptp_probe() registers a PHC whose ptp_clock_info and MII timestamper
are embedded in devres-managed private data. The enclosing PHY probe can
still fail while acquiring the wakeup GPIO, requesting its IRQ or
enabling wakeup. Driver-core cleanup then frees the private data without
unregistering the PHC, leaving pointers to freed memory in both the PTP
core and the phy_device.
Register a managed cleanup action after publishing the timestamper. The
action removes the driver-owned timestamper, unregisters the PHC and
purges the transmit queue before the private data is released. The same
action also supplies the missing cleanup for normal driver removal.
Fixes: 39db6be781cd ("net: phy: broadcom: Add PTP support for some
Broadcom PHYs.")
Cc: stable@vger.kernel.org
Signed-off-by: Kimberly Morris <gy770616@gmail.com>
---
drivers/net/phy/bcm-phy-ptp.c | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)
diff --git a/drivers/net/phy/bcm-phy-ptp.c b/drivers/net/phy/bcm-phy-ptp.c
index 443917c..085cc31 100644
--- a/drivers/net/phy/bcm-phy-ptp.c
+++ b/drivers/net/phy/bcm-phy-ptp.c
@@ -917,10 +917,25 @@ static void bcm_ptp_init(struct bcm_ptp_private *priv)
priv->phydev->mii_ts = &priv->mii_ts;
}
+static void bcm_ptp_remove(void *data)
+{
+ struct bcm_ptp_private *priv = data;
+ struct phy_device *phydev = priv->phydev;
+
+ if (phydev->mii_ts == &priv->mii_ts) {
+ phydev->mii_ts = NULL;
+ phydev->default_timestamp = false;
+ }
+
+ ptp_clock_unregister(priv->ptp_clock);
+ skb_queue_purge(&priv->tx_queue);
+}
+
struct bcm_ptp_private *bcm_ptp_probe(struct phy_device *phydev)
{
struct bcm_ptp_private *priv;
struct ptp_clock *clock;
+ int err;
switch (BRCM_PHY_MODEL(phydev)) {
case PHY_ID_BCM54210E:
@@ -949,6 +964,10 @@ struct bcm_ptp_private *bcm_ptp_probe(struct
phy_device *phydev)
priv->phydev = phydev;
bcm_ptp_init(priv);
+ err = devm_add_action_or_reset(&phydev->mdio.dev, bcm_ptp_remove, priv);
+ if (err)
+ return ERR_PTR(err);
+
return priv;
}
EXPORT_SYMBOL_GPL(bcm_ptp_probe);
--
2.55.0.windows.3
On Wed, 18 Nov 2026 05:33:45 -0500, Kimberly Morris <gy770616@gmail.com> wrote:
> Hello,
>
> Following Greg's request for an applyable patch, this series addresses the
> automatic PHY probe-failure lifetime path from the earlier report. It does
> not rely on the root-only sysfs driver-unbind trigger.
>
> A PHY driver can publish a devres-backed mii_timestamper before its probe
> has completed. If a later probe step fails, driver-core cleanup frees the
> owner while the phy_device can retain the published timestamping state. A
> later generic-PHY fallback can then reach the freed object from
> skb_defer_rx_timestamp(). Broadcom BCM54210E/BCM54213PE also leaves a
> registered PHC backed by the same freed storage.
>
> Patch 1 restores the timestamping state that existed before phy_probe().
> Patch 2 adds the missing managed Broadcom PHC and timestamper cleanup.
>
> I reproduced the first issue on an x86_64 KASAN kernel with a KUnit test
> that registers a real mii_bus, phy_device and phy_driver, follows the real
> failed-probe and devres cleanup paths, attaches the generic PHY driver, and
> passes a non-PTP skb to skb_defer_rx_timestamp(). Before the fix KASAN
> reported a slab use-after-free; after the fix the test passed without a
> KASAN report.
>
> Tested on:
>
> - upstream master 20eef29b7c8465ac98d7dd6d6f2ac07a8d8056ca
> - net/main 2a324a58a084c5cfd7e5a2f55e0786bee170a4a5
>
> The submitted series is generated against current upstream master
> 63fdb511c27afcebce797e3db580b3534ddfa639. I have not tested it on physical
> BCM54213PE hardware, and I am not claiming an attacker-controlled probe
> failure, privilege escalation, or code execution.
>
> Please let me know if the rollback belongs in individual PHY drivers rather
> than phylib core.
>
> Thank you.
>
> Kimberly Morris (2):
> net: phy: restore timestamping state on probe failure
> net: phy: broadcom: release PTP clock with PHY devres
>
> drivers/net/phy/bcm-phy-ptp.c | 19 +++++++++++++++++++
> drivers/net/phy/phy_device.c | 5 +++++
> 2 files changed, 24 insertions(+)
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure
2026-11-18 10:33 [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure Kimberly Morris
2026-11-18 10:34 ` [PATCH net 1/2] net: phy: restore timestamping state on " Kimberly Morris
2026-11-18 10:34 ` [PATCH net 2/2] net: phy: broadcom: release PTP clock with PHY devres Kimberly Morris
@ 2026-11-18 10:38 ` netdev-bot+sinfo
2026-11-18 10:57 ` Kimberly Morris
2 siblings, 1 reply; 5+ messages in thread
From: netdev-bot+sinfo @ 2026-11-18 10:38 UTC (permalink / raw)
To: Kimberly Morris
Cc: andrew, hkallweit1, linux, florian.fainelli,
bcm-kernel-feedback-list, davem, edumazet, kuba, pabeni,
richardcochran, netdev, linux-kernel, gregkh, w
Hi!
This is an automated message. This series looks like a fix, but its
commit messages seem to be missing some information:
- How the issue was discovered, e.g. hit in production, hit during
development, syzbot report, manual code inspection, LLM or static
analysis tool scan.
Please do not repost the series just to address the above. Instead,
reply to this email with the missing information, so that reviewers
can take it into account.
*IMPORTANT*: If you do repost (for other reasons) - please include
the relevant info in the cover letter or commit messages.
--
The evaluation is done by an LLM so it may be wrong, if you think
that is the case please reply and explain.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure
2026-11-18 10:38 ` [PATCH net 0/2] net: phy: fix timestamping lifetime after probe failure netdev-bot+sinfo
@ 2026-11-18 10:57 ` Kimberly Morris
0 siblings, 0 replies; 5+ messages in thread
From: Kimberly Morris @ 2026-11-18 10:57 UTC (permalink / raw)
To: netdev-bot+sinfo
Cc: andrew, hkallweit1, linux, florian.fainelli,
bcm-kernel-feedback-list, davem, edumazet, kuba, pabeni,
richardcochran, netdev, linux-kernel, gregkh, w
> - How the issue was discovered, e.g. hit in production, hit during
> development, syzbot report, manual code inspection, LLM or static
> analysis tool scan.
The issue was found during a targeted source audit of PHY hardware
timestamping lifetime and probe-error paths. The source audit was assisted
by an LLM; it was not a production report or a syzbot finding.
The resulting hypothesis was then validated against the current Linux
source with a focused KUnit/KASAN reproducer. The test registers a real
mii_bus, phy_device and phy_driver, follows phylib's probe failure and the
driver core's devres cleanup, attaches the generic PHY driver, and calls
skb_defer_rx_timestamp() with a non-PTP skb. KASAN reported the
use-after-free before the fix, and the same test passed after the fix.
The concrete BCM54210E/BCM54213PE publication-before-failure ordering and
the missing PHC cleanup were also checked directly in the driver source.
No physical hardware reproduction has been completed.
On Wed, 18 Nov 2026 10:38:42 +0000, netdev-bot+sinfo@kernel.org wrote:
> Hi!
>
> This is an automated message. This series looks like a fix, but its
> commit messages seem to be missing some information:
>
> - How the issue was discovered, e.g. hit in production, hit during
> development, syzbot report, manual code inspection, LLM or static
> analysis tool scan.
>
> Please do not repost the series just to address the above. Instead,
> reply to this email with the missing information, so that reviewers
> can take it into account.
>
> *IMPORTANT*: If you do repost (for other reasons) - please include
> the relevant info in the cover letter or commit messages.
> --
> The evaluation is done by an LLM so it may be wrong, if you think
> that is the case please reply and explain.
^ permalink raw reply [flat|nested] 5+ messages in thread