This reverts commit ab609f25d19858513919369ff3d9a63c02cd9e2e.
This patch is correct in the sense that we _should_ call device_put() in
case of device_register() failure, but the problem in this code is more
vast.
We need to set bus->state to UNMDIOBUS_REGISTERED before calling
device_register() to correctly release the device in mdiobus_free().
This patch prevents us from doing it, since in case of device_register()
failure put_device() will be called 2 times and it will cause UAF or
something else.
Also, Reported-by: tag in revered commit was wrong, since syzbot
reported different leak in same function.
Link: https://lore.kernel.org/netdev/20210928092657.GI2048@kadam/
Acked-by: Yanfei Xu <[email protected]>
Signed-off-by: Pavel Skripkin <[email protected]>
---
Changes in v3:
CC Yanfei -> Acked-by Yanfei
Changes in v2:
Added this revert
---
drivers/net/phy/mdio_bus.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/drivers/net/phy/mdio_bus.c b/drivers/net/phy/mdio_bus.c
index 6f4b4e5df639..53f034fc2ef7 100644
--- a/drivers/net/phy/mdio_bus.c
+++ b/drivers/net/phy/mdio_bus.c
@@ -537,7 +537,6 @@ int __mdiobus_register(struct mii_bus *bus, struct module *owner)
err = device_register(&bus->dev);
if (err) {
pr_err("mii_bus %s failed to register\n", bus->id);
- put_device(&bus->dev);
return -EINVAL;
}
--
2.33.0
Syzbot reported memory leak in MDIO bus interface, the problem was in
wrong state logic.
MDIOBUS_ALLOCATED indicates 2 states:
1. Bus is only allocated
2. Bus allocated and __mdiobus_register() fails, but
device_register() was called
In case of device_register() has been called we should call put_device()
to correctly free the memory allocated for this device, but mdiobus_free()
calls just kfree(dev) in case of MDIOBUS_ALLOCATED state
To avoid this behaviour we need to set bus->state to MDIOBUS_UNREGISTERED
_before_ calling device_register(), because put_device() should be
called even in case of device_register() failure.
Link: https://lore.kernel.org/netdev/[email protected]/
Fixes: 46abc02175b3 ("phylib: give mdio buses a device tree presence")
Reported-and-tested-by: [email protected]
Reviewed-by: Dan Carpenter <[email protected]>
Signed-off-by: Pavel Skripkin <[email protected]>
---
Changes in v3:
s/==/=/
Added Dan's Reviewed-by tag
Changes in v2:
Removed new state, since MDIOBUS_UNREGISTERED can be used. Also, moved
bus->state initialization _before_ device_register() call, because
Yanfei's patch is reverted in [v2 1/2]
---
drivers/net/phy/mdio_bus.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/net/phy/mdio_bus.c b/drivers/net/phy/mdio_bus.c
index 53f034fc2ef7..a5f910d4a7be 100644
--- a/drivers/net/phy/mdio_bus.c
+++ b/drivers/net/phy/mdio_bus.c
@@ -534,6 +534,13 @@ int __mdiobus_register(struct mii_bus *bus, struct module *owner)
bus->dev.groups = NULL;
dev_set_name(&bus->dev, "%s", bus->id);
+ /* We need to set state to MDIOBUS_UNREGISTERED to correctly realese
+ * the device in mdiobus_free()
+ *
+ * State will be updated later in this function in case of success
+ */
+ bus->state = MDIOBUS_UNREGISTERED;
+
err = device_register(&bus->dev);
if (err) {
pr_err("mii_bus %s failed to register\n", bus->id);
--
2.33.0
On Thu, Sep 30, 2021 at 09:19:54AM +0300, Pavel Skripkin wrote:
> Syzbot reported memory leak in MDIO bus interface, the problem was in
> wrong state logic.
>
> MDIOBUS_ALLOCATED indicates 2 states:
> 1. Bus is only allocated
> 2. Bus allocated and __mdiobus_register() fails, but
> device_register() was called
>
> In case of device_register() has been called we should call put_device()
> to correctly free the memory allocated for this device, but mdiobus_free()
> calls just kfree(dev) in case of MDIOBUS_ALLOCATED state
>
> To avoid this behaviour we need to set bus->state to MDIOBUS_UNREGISTERED
> _before_ calling device_register(), because put_device() should be
> called even in case of device_register() failure.
>
> Link: https://lore.kernel.org/netdev/[email protected]/
> Fixes: 46abc02175b3 ("phylib: give mdio buses a device tree presence")
> Reported-and-tested-by: [email protected]
> Reviewed-by: Dan Carpenter <[email protected]>
> Signed-off-by: Pavel Skripkin <[email protected]>
> ---
>
> Changes in v3:
> s/==/=/
> Added Dan's Reviewed-by tag
>
> Changes in v2:
> Removed new state, since MDIOBUS_UNREGISTERED can be used. Also, moved
> bus->state initialization _before_ device_register() call, because
> Yanfei's patch is reverted in [v2 1/2]
>
> ---
> drivers/net/phy/mdio_bus.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/net/phy/mdio_bus.c b/drivers/net/phy/mdio_bus.c
> index 53f034fc2ef7..a5f910d4a7be 100644
> --- a/drivers/net/phy/mdio_bus.c
> +++ b/drivers/net/phy/mdio_bus.c
> @@ -534,6 +534,13 @@ int __mdiobus_register(struct mii_bus *bus, struct module *owner)
> bus->dev.groups = NULL;
> dev_set_name(&bus->dev, "%s", bus->id);
>
> + /* We need to set state to MDIOBUS_UNREGISTERED to correctly realese
Typo here. s/realese/release/
> + * the device in mdiobus_free()
> + *
> + * State will be updated later in this function in case of success
> + */
> + bus->state = MDIOBUS_UNREGISTERED;
> +
> err = device_register(&bus->dev);
> if (err) {
> pr_err("mii_bus %s failed to register\n", bus->id);