| Message ID | 20250114035118.110297-5-kuba@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: use netdev->lock to protect NAPI | expand |
On Tue, Jan 14, 2025 at 4:51 AM Jakub Kicinski <kuba@kernel.org> wrote: > > Some uAPI (netdev netlink) hide net_device's sub-objects while > the interface is down to ensure uniform behavior across drivers. > To remove the rtnl_lock dependency from those uAPIs we need a way > to safely tell if the device is down or up. > > Add an indication of whether device is open or closed, protected > by netdev->lock. The semantics are the same as IFF_UP, but taking > netdev_lock around every write to ->flags would be a lot of code > churn. > > We don't want to blanket the entire open / close path by netdev_lock, > because it will prevent us from applying it to specific structures - > core helpers won't be able to take that lock from any function > called by the drivers on open/close paths. > > So the state of the flag is "pessimistic", as in it may report false > negatives, but never false positives. > > Signed-off-by: Jakub Kicinski <kuba@kernel.org> Reviewed-by: Eric Dumazet <edumazet@google.com>
On Mon, Jan 13, 2025 at 07:51:10PM -0800, Jakub Kicinski wrote: > Some uAPI (netdev netlink) hide net_device's sub-objects while > the interface is down to ensure uniform behavior across drivers. > To remove the rtnl_lock dependency from those uAPIs we need a way > to safely tell if the device is down or up. > > Add an indication of whether device is open or closed, protected > by netdev->lock. The semantics are the same as IFF_UP, but taking > netdev_lock around every write to ->flags would be a lot of code > churn. > > We don't want to blanket the entire open / close path by netdev_lock, > because it will prevent us from applying it to specific structures - > core helpers won't be able to take that lock from any function > called by the drivers on open/close paths. > > So the state of the flag is "pessimistic", as in it may report false > negatives, but never false positives. > > Signed-off-by: Jakub Kicinski <kuba@kernel.org> > --- > include/linux/netdevice.h | 13 ++++++++++++- > net/core/dev.h | 12 ++++++++++++ > net/core/dev.c | 4 ++-- > 3 files changed, 26 insertions(+), 3 deletions(-) Reviewed-by: Joe Damato <jdamato@fastly.com>
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index bdbc5849469c..565dfeb78774 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -2443,12 +2443,23 @@ struct net_device { unsigned long gro_flush_timeout; u32 napi_defer_hard_irqs; + /** + * @up: copy of @state's IFF_UP, but safe to read with just @lock. + * May report false negatives while the device is being opened + * or closed (@lock does not protect .ndo_open, or .ndo_close). + */ + bool up; + /** * @lock: netdev-scope lock, protects a small selection of fields. * Should always be taken using netdev_lock() / netdev_unlock() helpers. * Drivers are free to use it for other protection. * - * Protects: @reg_state, @net_shaper_hierarchy. + * Protects: + * @net_shaper_hierarchy, @reg_state + * Partially protects (readers hold either @lock or rtnl_lock, + * writers must hold both for registered devices): + * @up * Ordering: take after rtnl_lock. */ struct mutex lock; diff --git a/net/core/dev.h b/net/core/dev.h index 25ae732c0775..ef37e2dd44f4 100644 --- a/net/core/dev.h +++ b/net/core/dev.h @@ -128,6 +128,18 @@ void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, void unregister_netdevice_many_notify(struct list_head *head, u32 portid, const struct nlmsghdr *nlh); +static inline void netif_set_up(struct net_device *dev, bool value) +{ + if (value) + dev->flags |= IFF_UP; + else + dev->flags &= ~IFF_UP; + + netdev_lock(dev); + dev->up = value; + netdev_unlock(dev); +} + static inline void netif_set_gso_max_size(struct net_device *dev, unsigned int size) { diff --git a/net/core/dev.c b/net/core/dev.c index 2ded6eedb4cc..1a05ad60b89f 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -1618,7 +1618,7 @@ static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) if (ret) clear_bit(__LINK_STATE_START, &dev->state); else { - dev->flags |= IFF_UP; + netif_set_up(dev, true); dev_set_rx_mode(dev); dev_activate(dev); add_device_randomness(dev->dev_addr, dev->addr_len); @@ -1697,7 +1697,7 @@ static void __dev_close_many(struct list_head *head) if (ops->ndo_stop) ops->ndo_stop(dev); - dev->flags &= ~IFF_UP; + netif_set_up(dev, false); netpoll_poll_enable(dev); } }
Some uAPI (netdev netlink) hide net_device's sub-objects while the interface is down to ensure uniform behavior across drivers. To remove the rtnl_lock dependency from those uAPIs we need a way to safely tell if the device is down or up. Add an indication of whether device is open or closed, protected by netdev->lock. The semantics are the same as IFF_UP, but taking netdev_lock around every write to ->flags would be a lot of code churn. We don't want to blanket the entire open / close path by netdev_lock, because it will prevent us from applying it to specific structures - core helpers won't be able to take that lock from any function called by the drivers on open/close paths. So the state of the flag is "pessimistic", as in it may report false negatives, but never false positives. Signed-off-by: Jakub Kicinski <kuba@kernel.org> --- include/linux/netdevice.h | 13 ++++++++++++- net/core/dev.h | 12 ++++++++++++ net/core/dev.c | 4 ++-- 3 files changed, 26 insertions(+), 3 deletions(-)