user/sven/linux.git/net/switchdev, branch v6.1.97

net: bridge: switchdev: Skip MDB replays of deferred events on offload

2024-03-01T12:26:35Z

[ Upstream commit dc489f86257cab5056e747344f17a164f63bff4b ] Before this change, generation of the list of MDB events to replay would race against the creation of new group memberships, either from the IGMP/MLD snooping logic or from user configuration. While new memberships are immediately visible to walkers of br->mdb_list, the notification of their existence to switchdev event subscribers is deferred until a later point in time. So if a replay list was generated during a time that overlapped with such a window, it would also contain a replay of the not-yet-delivered event. The driver would thus receive two copies of what the bridge internally considered to be one single event. On destruction of the bridge, only a single membership deletion event was therefore sent. As a consequence of this, drivers which reference count memberships (at least DSA), would be left with orphan groups in their hardware database when the bridge was destroyed. This is only an issue when replaying additions. While deletion events may still be pending on the deferred queue, they will already have been removed from br->mdb_list, so no duplicates can be generated in that scenario. To a user this meant that old group memberships, from a bridge in which a port was previously attached, could be reanimated (in hardware) when the port joined a new bridge, without the new bridge's knowledge. For example, on an mv88e6xxx system, create a snooping bridge and immediately add a port to it: root@infix-06-0b-00:~$ ip link add dev br0 up type bridge mcast_snooping 1 && \ > ip link set dev x3 up master br0 And then destroy the bridge: root@infix-06-0b-00:~$ ip link del dev br0 root@infix-06-0b-00:~$ mvls atu ADDRESS FID STATE Q F 0 1 2 3 4 5 6 7 8 9 a DEV:0 Marvell 88E6393X 33:33:00:00:00:6a 1 static - - 0 . . . . . . . . . . 33:33:ff:87:e4:3f 1 static - - 0 . . . . . . . . . . ff:ff:ff:ff:ff:ff 1 static - - 0 1 2 3 4 5 6 7 8 9 a root@infix-06-0b-00:~$ The two IPv6 groups remain in the hardware database because the port (x3) is notified of the host's membership twice: once via the original event and once via a replay. Since only a single delete notification is sent, the count remains at 1 when the bridge is destroyed. Then add the same port (or another port belonging to the same hardware domain) to a new bridge, this time with snooping disabled: root@infix-06-0b-00:~$ ip link add dev br1 up type bridge mcast_snooping 0 && \ > ip link set dev x3 up master br1 All multicast, including the two IPv6 groups from br0, should now be flooded, according to the policy of br1. But instead the old memberships are still active in the hardware database, causing the switch to only forward traffic to those groups towards the CPU (port 0). Eliminate the race in two steps: 1. Grab the write-side lock of the MDB while generating the replay list. This prevents new memberships from showing up while we are generating the replay list. But it leaves the scenario in which a deferred event was already generated, but not delivered, before we grabbed the lock. Therefore: 2. Make sure that no deferred version of a replay event is already enqueued to the switchdev deferred queue, before adding it to the replay list, when replaying additions. Fixes: 4f2673b3a2b6 ("net: bridge: add helper to replay port and host-joined mdb entries") Signed-off-by: Tobias Waldekranz Reviewed-by: Vladimir Oltean Signed-off-by: David S. Miller Signed-off-by: Sasha Levin

net: rename reference+tracking helpers

2022-06-10T04:52:55Z

Netdev reference helpers have a dev_ prefix for historic reasons. Renaming the old helpers would be too much churn but we can rename the tracking ones which are relatively recent and should be the default for new code. Rename: dev_hold_track() -> netdev_hold() dev_put_track() -> netdev_put() dev_replace_track() -> netdev_ref_replace() Link: https://lore.kernel.org/r/20220608043955.919359-1-kuba@kernel.org Signed-off-by: Jakub Kicinski

net: switchdev: remove lag_mod_cb from switchdev_handle_fdb_event_to_device

2022-02-25T05:31:43Z

When the switchdev_handle_fdb_event_to_device() event replication helper was created, my original thought was that FDB events on LAG interfaces should most likely be special-cased, not just replicated towards all switchdev ports beneath that LAG. So this replication helper currently does not recurse through switchdev lower interfaces of LAG bridge ports, but rather calls the lag_mod_cb() if that was provided. No switchdev driver uses this helper for FDB events on LAG interfaces yet, so that was an assumption which was yet to be tested. It is certainly usable for that purpose, as my RFC series shows: https://patchwork.kernel.org/project/netdevbpf/cover/20220210125201.2859463-1-vladimir.oltean@nxp.com/ however this approach is slightly convoluted because: - the switchdev driver gets a "dev" that isn't its own net device, but rather the LAG net device. It must call switchdev_lower_dev_find(dev) in order to get a handle of any of its own net devices (the ones that pass check_cb). - in order for FDB entries on LAG ports to be correctly refcounted per the number of switchdev ports beneath that LAG, we haven't escaped the need to iterate through the LAG's lower interfaces. Except that is now the responsibility of the switchdev driver, because the replication helper just stopped half-way. So, even though yes, FDB events on LAG bridge ports must be special-cased, in the end it's simpler to let switchdev_handle_fdb_* just iterate through the LAG port's switchdev lowers, and let the switchdev driver figure out that those physical ports are under a LAG. The switchdev_handle_fdb_event_to_device() helper takes a "foreign_dev_check" callback so it can figure out whether @dev can autonomously forward to @foreign_dev. DSA fills this method properly: if the LAG is offloaded by another port in the same tree as @dev, then it isn't foreign. If it is a software LAG, it is foreign - forwarding happens in software. Whether an interface is foreign or not decides whether the replication helper will go through the LAG's switchdev lowers or not. Since the lan966x doesn't properly fill this out, FDB events on software LAG uppers will get called. By changing lan966x_foreign_dev_check(), we can suppress them. Whereas DSA will now start receiving FDB events for its offloaded LAG uppers, so we need to return -EOPNOTSUPP, since we currently don't do the right thing for them. Cc: Horatiu Vultur Signed-off-by: Vladimir Oltean Reviewed-by: Florian Fainelli Signed-off-by: Jakub Kicinski

net: switchdev: avoid infinite recursion from LAG to bridge with port object handler

2022-02-23T12:12:12Z

The logic from switchdev_handle_port_obj_add_foreign() is directly adapted from switchdev_handle_fdb_event_to_device(), which already detects events on foreign interfaces and reoffloads them towards the switchdev neighbors. However, when we have a simple br0 <-> bond0 <-> swp0 topology and the switchdev_handle_port_obj_add_foreign() gets called on bond0, we get stuck into an infinite recursion: 1. bond0 does not pass check_cb(), so we attempt to find switchdev neighbor interfaces. For that, we recursively call __switchdev_handle_port_obj_add() for bond0's bridge, br0. 2. __switchdev_handle_port_obj_add() recurses through br0's lowers, essentially calling __switchdev_handle_port_obj_add() for bond0 3. Go to step 1. This happens because switchdev_handle_fdb_event_to_device() and switchdev_handle_port_obj_add_foreign() are not exactly the same. The FDB event helper special-cases LAG interfaces with its lag_mod_cb(), so this is why we don't end up in an infinite loop - because it doesn't attempt to treat LAG interfaces as potentially foreign bridge ports. The problem is solved by looking ahead through the bridge's lowers to see whether there is any switchdev interface that is foreign to the @dev we are currently processing. This stops the recursion described above at step 1: __switchdev_handle_port_obj_add(bond0) will not create another call to __switchdev_handle_port_obj_add(br0). Going one step upper should only happen when we're starting from a bridge port that has been determined to be "foreign" to the switchdev driver that passes the foreign_dev_check_cb(). Fixes: c4076cdd21f8 ("net: switchdev: introduce switchdev_handle_port_obj_{add,del} for foreign interfaces") Signed-off-by: Vladimir Oltean Signed-off-by: David S. Miller

net: switchdev: introduce switchdev_handle_port_obj_{add,del} for foreign interfaces

2022-02-16T11:21:04Z

The switchdev_handle_port_obj_add() helper is good for replicating a port object on the lower interfaces of @dev, if that object was emitted on a bridge, or on a bridge port that is a LAG. However, drivers that use this helper limit themselves to a box from which they can no longer intercept port objects notified on neighbor ports ("foreign interfaces"). One such driver is DSA, where software bridging with foreign interfaces such as standalone NICs or Wi-Fi APs is an important use case. There, a VLAN installed on a neighbor bridge port roughly corresponds to a forwarding VLAN installed on the DSA switch's CPU port. To support this use case while also making use of the benefits of the switchdev_handle_* replication helper for port objects, introduce a new variant of these functions that crawls through the neighbor ports of @dev, in search of potentially compatible switchdev ports that are interested in the event. The strategy is identical to switchdev_handle_fdb_event_to_device(): if @dev wasn't a switchdev interface, then go one step upper, and recursively call this function on the bridge that this port belongs to. At the next recursion step, __switchdev_handle_port_obj_add() will iterate through the bridge's lower interfaces. Among those, some will be switchdev interfaces, and one will be the original @dev that we came from. To prevent infinite recursion, we must suppress reentry into the original @dev, and just call the @add_cb for the switchdev_interfaces. It looks like this: br0 / | \ / | \ / | \ swp0 swp1 eth0 1. __switchdev_handle_port_obj_add(eth0) -> check_cb(eth0) returns false -> eth0 has no lower interfaces -> eth0's bridge is br0 -> switchdev_lower_dev_find(br0, check_cb, foreign_dev_check_cb)) finds br0 2. __switchdev_handle_port_obj_add(br0) -> check_cb(br0) returns false -> netdev_for_each_lower_dev -> check_cb(swp0) returns true, so we don't skip this interface 3. __switchdev_handle_port_obj_add(swp0) -> check_cb(swp0) returns true, so we call add_cb(swp0) (back to netdev_for_each_lower_dev from 2) -> check_cb(swp1) returns true, so we don't skip this interface 4. __switchdev_handle_port_obj_add(swp1) -> check_cb(swp1) returns true, so we call add_cb(swp1) (back to netdev_for_each_lower_dev from 2) -> check_cb(eth0) returns false, so we skip this interface to avoid infinite recursion Note: eth0 could have been a LAG, and we don't want to suppress the recursion through its lowers if those exist, so when check_cb() returns false, we still call switchdev_lower_dev_find() to estimate whether there's anything worth a recursion beneath that LAG. Using check_cb() and foreign_dev_check_cb(), switchdev_lower_dev_find() not only figures out whether the lowers of the LAG are switchdev, but also whether they actively offload the LAG or not (whether the LAG is "foreign" to the switchdev interface or not). The port_obj_info->orig_dev is preserved across recursive calls, so switchdev drivers still know on which device was this notification originally emitted. Signed-off-by: Vladimir Oltean Signed-off-by: David S. Miller

net: switchdev: rename switchdev_lower_dev_find to switchdev_lower_dev_find_rcu

2022-02-16T11:21:04Z

switchdev_lower_dev_find() assumes RCU read-side critical section calling context, since it uses netdev_walk_all_lower_dev_rcu(). Rename it appropriately, in preparation of adding a similar iterator that assumes writer-side rtnl_mutex protection. Signed-off-by: Vladimir Oltean Signed-off-by: David S. Miller

net/switchdev: use struct_size over open coded arithmetic

2022-02-10T15:37:47Z

Replace zero-length array with flexible-array member and make use of the struct_size() helper in kmalloc(). For example: struct switchdev_deferred_item { ... unsigned long data[]; }; Make use of the struct_size() helper instead of an open-coded version in order to avoid any potential type mistakes. Reported-by: Zeal Robot Signed-off-by: Minghao Chi (CGEL ZTE) Signed-off-by: David S. Miller

net: switchdev: add net device refcount tracker

2021-12-08T04:44:58Z

Signed-off-by: Eric Dumazet Signed-off-by: Jakub Kicinski

net: switchdev: merge switchdev_handle_fdb_{add,del}_to_device

2021-10-27T13:54:02Z

To reduce code churn, the same patch makes multiple changes, since they all touch the same lines: 1. The implementations for these two are identical, just with different function pointers. Reduce duplications and name the function pointers "mod_cb" instead of "add_cb" and "del_cb". Pass the event as argument. 2. Drop the "const" attribute from "orig_dev". If the driver needs to check whether orig_dev belongs to itself and then call_switchdev_notifiers(orig_dev, SWITCHDEV_FDB_OFFLOADED), it can't, because call_switchdev_notifiers takes a non-const struct net_device *. Signed-off-by: Vladimir Oltean Reviewed-by: Ido Schimmel Signed-off-by: David S. Miller

net: make switchdev_bridge_port_{,unoffload} loosely coupled with the bridge

2021-08-04T11:35:07Z

With the introduction of explicit offloading API in switchdev in commit 2f5dc00f7a3e ("net: bridge: switchdev: let drivers inform which bridge ports are offloaded"), we started having Ethernet switch drivers calling directly into a function exported by net/bridge/br_switchdev.c, which is a function exported by the bridge driver. This means that drivers that did not have an explicit dependency on the bridge before, like cpsw and am65-cpsw, now do - otherwise it is not possible to call a symbol exported by a driver that can be built as module unless you are a module too. There was an attempt to solve the dependency issue in the form of commit b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module"). Grygorii Strashko, however, says about it: | In my opinion, the problem is a bit bigger here than just fixing the | build :( | | In case, of ^cpsw the switchdev mode is kinda optional and in many | cases (especially for testing purposes, NFS) the multi-mac mode is | still preferable mode. | | There were no such tight dependency between switchdev drivers and | bridge core before and switchdev serviced as independent, notification | based layer between them, so ^cpsw still can be "Y" and bridge can be | "M". Now for mostly every kernel build configuration the CONFIG_BRIDGE | will need to be set as "Y", or we will have to update drivers to | support build with BRIDGE=n and maintain separate builds for | networking vs non-networking testing. But is this enough? Wouldn't | it cause 'chain reaction' required to add more and more "Y" options | (like CONFIG_VLAN_8021Q)? | | PS. Just to be sure we on the same page - ARM builds will be forced | (with this patch) to have CONFIG_TI_CPSW_SWITCHDEV=m and so all our | automation testing will just fail with omap2plus_defconfig. In the light of this, it would be desirable for some configurations to avoid dependencies between switchdev drivers and the bridge, and have the switchdev mode as completely optional within the driver. Arnd Bergmann also tried to write a patch which better expressed the build time dependency for Ethernet switch drivers where the switchdev support is optional, like cpsw/am65-cpsw, and this made the drivers follow the bridge (compile as module if the bridge is a module) only if the optional switchdev support in the driver was enabled in the first place: https://patchwork.kernel.org/project/netdevbpf/patch/20210802144813.1152762-1-arnd@kernel.org/ but this still did not solve the fact that cpsw and am65-cpsw now must be built as modules when the bridge is a module - it just expressed correctly that optional dependency. But the new behavior is an apparent regression from Grygorii's perspective. So to support the use case where the Ethernet driver is built-in, NET_SWITCHDEV (a bool option) is enabled, and the bridge is a module, we need a framework that can handle the possible absence of the bridge from the running system, i.e. runtime bloatware as opposed to build-time bloatware. Luckily we already have this framework, since switchdev has been using it extensively. Events from the bridge side are transmitted to the driver side using notifier chains - this was originally done so that unrelated drivers could snoop for events emitted by the bridge towards ports that are implemented by other drivers (think of a switch driver with LAG offload that listens for switchdev events on a bonding/team interface that it offloads). There are also events which are transmitted from the driver side to the bridge side, which again are modeled using notifiers. SWITCHDEV_FDB_ADD_TO_BRIDGE is an example of this, and deals with notifying the bridge that a MAC address has been dynamically learned. So there is a precedent we can use for modeling the new framework. The difference compared to SWITCHDEV_FDB_ADD_TO_BRIDGE is that the work that the bridge needs to do when a port becomes offloaded is blocking in its nature: replay VLANs, MDBs etc. The calling context is indeed blocking (we are under rtnl_mutex), but the existing switchdev notification chain that the bridge is subscribed to is only the atomic one. So we need to subscribe the bridge to the blocking switchdev notification chain too. This patch: - keeps the driver-side perception of the switchdev_bridge_port_{,un}offload unchanged - moves the implementation of switchdev_bridge_port_{,un}offload from the bridge module into the switchdev module. - makes everybody that is subscribed to the switchdev blocking notifier chain "hear" offload & unoffload events - makes the bridge driver subscribe and handle those events - moves the bridge driver's handling of those events into 2 new functions called br_switchdev_port_{,un}offload. These functions contain in fact the core of the logic that was previously in switchdev_bridge_port_{,un}offload, just that now we go through an extra indirection layer to reach them. Unlike all the other switchdev notification structures, the structure used to carry the bridge port information, struct switchdev_notifier_brport_info, does not contain a "bool handled". This is because in the current usage pattern, we always know that a switchdev bridge port offloading event will be handled by the bridge, because the switchdev_bridge_port_offload() call was initiated by a NETDEV_CHANGEUPPER event in the first place, where info->upper_dev is a bridge. So if the bridge wasn't loaded, then the CHANGEUPPER event couldn't have happened. Signed-off-by: Vladimir Oltean Tested-by: Grygorii Strashko Signed-off-by: David S. Miller