netdev: fix the locking for netdev notifications
Subsystems (3)
Jakub Kicinski
maintainerPATCH v2 1/1 netdev: fix the locking for netdev notifications
Kuniyuki reports that the assert for netdev lock fires when there are netdev event listeners (otherwise we skip the netlink event generation).
Correct the locking when coming from the notifier.
The NETDEV_XDP_FEAT_CHANGE notifier is already fully locked, it's the documentation that's incorrect.
---
v2:
- rebase vs net merge which brought in
commit 04efcee6ef8d ("net: hold instance lock during NETDEV_CHANGE")
v1: https://lore.kernel.org/[email protected]
--- Documentation/networking/netdevices.rst | 4 +++- include/linux/netdevice.h | 2 +- include/net/netdev_lock.h | 12 ++++++++++++ net/core/lock_debug.c | 4 +++- net/core/netdev-genl.c | 4 ++++ 5 files changed, 23 insertions(+), 3 deletions(-)
Documentation/networking/netdevices.rst +3 -1
@@ -387,12 +387,14 @@ For device drivers that implement shaping or queue management APIs, some of the notifiers (``enum netdev_cmd``) are running under the netdev instance lock. +The following netdev notifiers are always run under the instance lock: +* ``NETDEV_XDP_FEAT_CHANGE`` + For devices with locked ops, currently only the following notifiers are running under the lock: * ``NETDEV_CHANGE`` * ``NETDEV_REGISTER`` * ``NETDEV_UP`` -* ``NETDEV_XDP_FEAT_CHANGE`` The following notifiers are running without the lock: * ``NETDEV_UNREGISTER``
include/linux/netdevice.h +1 -1
@@ -2520,7 +2520,7 @@ struct net_device {* @net_shaper_hierarchy, @reg_state, @threaded * * Double protects: -* @up, @moving_ns, @nd_net, @xdp_flags +* @up, @moving_ns, @nd_net, @xdp_features * * Double ops protects: * @real_num_rx_queues, @real_num_tx_queues
include/net/netdev_lock.h +12 -0
@@ -48,6 +48,18 @@ static inline void netdev_unlock_ops(struct net_device *dev)netdev_unlock (dev); } +static inline void netdev_lock_ops_to_full (struct net_device *dev) +{ +if (!netdev_need_ops_lock (dev)) +netdev_lock (dev); +} + +static inline void netdev_unlock_full_to_ops (struct net_device *dev) +{ +if (!netdev_need_ops_lock (dev)) +netdev_unlock (dev); +} +static inline void netdev_ops_assert_locked (const struct net_device *dev) {if (netdev_need_ops_lock (dev))
net/core/lock_debug.c +3 -1
@@ -18,10 +18,12 @@ int netdev_debug_event(struct notifier_block *nb, unsigned long event,/* Keep enum and don't add default to trigger -Werror=switch */ switch (cmd) { +case NETDEV_XDP_FEAT_CHANGE : +netdev_assert_locked (dev); + fallthrough;case NETDEV_CHANGE :case NETDEV_REGISTER :case NETDEV_UP : -case NETDEV_XDP_FEAT_CHANGE :netdev_ops_assert_locked (dev); fallthrough;case NETDEV_DOWN :
net/core/netdev-genl.c +4 -0
@@ -963,10 +963,14 @@ static int netdev_genl_netdevice_event(struct notifier_block *nb,switch (event) {case NETDEV_REGISTER : +netdev_lock_ops_to_full (netdev);netdev_genl_dev_notify (netdev,NETDEV_CMD_DEV_ADD_NTF ); +netdev_unlock_full_to_ops (netdev);break ;case NETDEV_UNREGISTER : +netdev_lock (netdev);netdev_genl_dev_notify (netdev,NETDEV_CMD_DEV_DEL_NTF ); +netdev_unlock (netdev);break ;case NETDEV_XDP_FEAT_CHANGE :netdev_genl_dev_notify (netdev,NETDEV_CMD_DEV_CHANGE_NTF );
Signature · 1 line
2.49.0
Quoted 6 lines · Jakub Kicinski
Kuniyuki reports that the assert for netdev lock fires when there are netdev event listeners (otherwise we skip the netlink event generation).
Correct the locking when coming from the notifier.
Quoted 8 lines
The
NETDEV_XDP_FEAT_CHANGEnotifier is already fully locked, it's the documentation that's incorrect.Fixes 99e44f39a8f7 ("netdev: depend on netdev->lock for xdp features")Reported-by syzkaller <[email protected]>Reported-by Kuniyuki Iwashima <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
Quoted 56 lines
---
v2:
- rebase vs net merge which brought in
commit04efcee6ef8d("net: hold instance lock duringNETDEV_CHANGE")
v1: https://lore.kernel.org/[email protected]--- Documentation/networking/netdevices.rst | 4 +++- include/linux/netdevice.h | 2 +- include/net/netdev_lock.h | 12 ++++++++++++ net/core/lock_debug.c | 4 +++- net/core/netdev-genl.c | 4 ++++ 5 files changed, 23 insertions(+), 3 deletions(-)Documentation/networking/netdevices.rst +3
-1@@ -387,12 +387,14 @@ For device drivers that implement shaping or queue management APIs, some of the notifiers (``enum netdev_cmd``) are running under the netdev instance lock. +The following netdev notifiers are always run under the instance lock: +* ``NETDEV_XDP_FEAT_CHANGE`` + For devices with locked ops, currently only the following notifiers are running under the lock: * ``NETDEV_CHANGE`` * ``NETDEV_REGISTER`` * ``NETDEV_UP`` -* ``NETDEV_XDP_FEAT_CHANGE`` The following notifiers are running without the lock: * ``NETDEV_UNREGISTER``include/linux/netdevice.h +1
-1@@ -2520,7 +2520,7 @@ struct net_device {* @net_shaper_hierarchy, @reg_state, @threaded * * Double protects: -* @up, @moving_ns, @nd_net, @xdp_flags +* @up, @moving_ns, @nd_net, @xdp_features * * Double ops protects: * @real_num_rx_queues, @real_num_tx_queues include/net/netdev_lock.h +0
-0
include/net/netdev_lock.h static inline void netdev_unlock_ops(struct net_device *dev) · line 54
Quoted 7 lines
include/net/netdev_lock.h +4
-0@@ -48,6 +48,18 @@ static inline void netdev_unlock_ops(struct net_device *dev)netdev_unlock (dev); } +static inline void netdev_lock_ops_to_full (struct net_device *dev) +{ +if (!netdev_need_ops_lock (dev)) +netdev_lock (dev);
Optional nit: I'm getting lost in all the helpers, I'd add the following here:
else
netdev_ops_assert_locked(dev);
Or maybe even:
if (netdev_need_ops_lock)
netdev_ops_assert_locked
else
netdev_lock
To express the constraints better.
Jakub Kicinski
maintainerQuoted 10 lines · On Mon, 14 Apr 2025 18:11:14 -0700 Stanislav Fomichev wrote:
include/net/netdev_lock.h static inline void netdev_unlock_ops(struct net_device *dev) · line 54
Quoted 4 lines
include/net/netdev_lock.h +4
-0+static inline void netdev_lock_ops_to_full (struct net_device *dev) +{ +if (!netdev_need_ops_lock (dev)) +netdev_lock (dev);Optional nit: I'm getting lost in all the helpers, I'd add the following here:
else
netdev_ops_assert_locked(dev);
Quoted 8 lines
Or maybe even:
if (
netdev_need_ops_lock)
netdev_ops_assert_locked
else
netdev_lockTo express the constraints better.
Hm, yes, I like.