xref: /linux/net/bridge/br_switchdev.c (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/list.h>
4 #include <linux/netdevice.h>
5 #include <linux/rtnetlink.h>
6 #include <linux/skbuff.h>
7 #include <net/switchdev.h>
8 
9 #include "br_private.h"
10 
11 static struct static_key_false br_switchdev_tx_fwd_offload;
12 
13 static bool nbp_switchdev_can_offload_tx_fwd(const struct net_bridge_port *p,
14 					     const struct sk_buff *skb)
15 {
16 	if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
17 		return false;
18 
19 	return (p->flags & BR_TX_FWD_OFFLOAD) &&
20 	       (p->hwdom != BR_INPUT_SKB_CB(skb)->src_hwdom);
21 }
22 
23 bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb)
24 {
25 	if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload))
26 		return false;
27 
28 	return BR_INPUT_SKB_CB(skb)->tx_fwd_offload;
29 }
30 
31 void br_switchdev_frame_set_offload_fwd_mark(struct sk_buff *skb)
32 {
33 	skb->offload_fwd_mark = br_switchdev_frame_uses_tx_fwd_offload(skb);
34 }
35 
36 /* Mark the frame for TX forwarding offload if this egress port supports it */
37 void nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p,
38 					     struct sk_buff *skb)
39 {
40 	if (nbp_switchdev_can_offload_tx_fwd(p, skb))
41 		BR_INPUT_SKB_CB(skb)->tx_fwd_offload = true;
42 }
43 
44 /* Lazily adds the hwdom of the egress bridge port to the bit mask of hwdoms
45  * that the skb has been already forwarded to, to avoid further cloning to
46  * other ports in the same hwdom by making nbp_switchdev_allowed_egress()
47  * return false.
48  */
49 void nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p,
50 					      struct sk_buff *skb)
51 {
52 	if (nbp_switchdev_can_offload_tx_fwd(p, skb))
53 		set_bit(p->hwdom, &BR_INPUT_SKB_CB(skb)->fwd_hwdoms);
54 }
55 
56 void nbp_switchdev_frame_mark(const struct net_bridge_port *p,
57 			      struct sk_buff *skb)
58 {
59 	if (p->hwdom)
60 		BR_INPUT_SKB_CB(skb)->src_hwdom = p->hwdom;
61 }
62 
63 bool nbp_switchdev_allowed_egress(const struct net_bridge_port *p,
64 				  const struct sk_buff *skb)
65 {
66 	struct br_input_skb_cb *cb = BR_INPUT_SKB_CB(skb);
67 
68 	return !test_bit(p->hwdom, &cb->fwd_hwdoms) &&
69 		(!skb->offload_fwd_mark || cb->src_hwdom != p->hwdom);
70 }
71 
72 /* Flags that can be offloaded to hardware */
73 #define BR_PORT_FLAGS_HW_OFFLOAD (BR_LEARNING | BR_FLOOD | \
74 				  BR_MCAST_FLOOD | BR_BCAST_FLOOD)
75 
76 int br_switchdev_set_port_flag(struct net_bridge_port *p,
77 			       unsigned long flags,
78 			       unsigned long mask,
79 			       struct netlink_ext_ack *extack)
80 {
81 	struct switchdev_attr attr = {
82 		.orig_dev = p->dev,
83 	};
84 	struct switchdev_notifier_port_attr_info info = {
85 		.attr = &attr,
86 	};
87 	int err;
88 
89 	mask &= BR_PORT_FLAGS_HW_OFFLOAD;
90 	if (!mask)
91 		return 0;
92 
93 	attr.id = SWITCHDEV_ATTR_ID_PORT_PRE_BRIDGE_FLAGS;
94 	attr.u.brport_flags.val = flags;
95 	attr.u.brport_flags.mask = mask;
96 
97 	/* We run from atomic context here */
98 	err = call_switchdev_notifiers(SWITCHDEV_PORT_ATTR_SET, p->dev,
99 				       &info.info, extack);
100 	err = notifier_to_errno(err);
101 	if (err == -EOPNOTSUPP)
102 		return 0;
103 
104 	if (err) {
105 		if (extack && !extack->_msg)
106 			NL_SET_ERR_MSG_MOD(extack,
107 					   "bridge flag offload is not supported");
108 		return -EOPNOTSUPP;
109 	}
110 
111 	attr.id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS;
112 	attr.flags = SWITCHDEV_F_DEFER;
113 
114 	err = switchdev_port_attr_set(p->dev, &attr, extack);
115 	if (err) {
116 		if (extack && !extack->_msg)
117 			NL_SET_ERR_MSG_MOD(extack,
118 					   "error setting offload flag on port");
119 		return err;
120 	}
121 
122 	return 0;
123 }
124 
125 void
126 br_switchdev_fdb_notify(struct net_bridge *br,
127 			const struct net_bridge_fdb_entry *fdb, int type)
128 {
129 	const struct net_bridge_port *dst = READ_ONCE(fdb->dst);
130 	struct switchdev_notifier_fdb_info info = {
131 		.addr = fdb->key.addr.addr,
132 		.vid = fdb->key.vlan_id,
133 		.added_by_user = test_bit(BR_FDB_ADDED_BY_USER, &fdb->flags),
134 		.is_local = test_bit(BR_FDB_LOCAL, &fdb->flags),
135 		.offloaded = test_bit(BR_FDB_OFFLOADED, &fdb->flags),
136 	};
137 	struct net_device *dev = (!dst || info.is_local) ? br->dev : dst->dev;
138 
139 	switch (type) {
140 	case RTM_DELNEIGH:
141 		call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_DEVICE,
142 					 dev, &info.info, NULL);
143 		break;
144 	case RTM_NEWNEIGH:
145 		call_switchdev_notifiers(SWITCHDEV_FDB_ADD_TO_DEVICE,
146 					 dev, &info.info, NULL);
147 		break;
148 	}
149 }
150 
151 int br_switchdev_port_vlan_add(struct net_device *dev, u16 vid, u16 flags,
152 			       struct netlink_ext_ack *extack)
153 {
154 	struct switchdev_obj_port_vlan v = {
155 		.obj.orig_dev = dev,
156 		.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
157 		.flags = flags,
158 		.vid = vid,
159 	};
160 
161 	return switchdev_port_obj_add(dev, &v.obj, extack);
162 }
163 
164 int br_switchdev_port_vlan_del(struct net_device *dev, u16 vid)
165 {
166 	struct switchdev_obj_port_vlan v = {
167 		.obj.orig_dev = dev,
168 		.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
169 		.vid = vid,
170 	};
171 
172 	return switchdev_port_obj_del(dev, &v.obj);
173 }
174 
175 static int nbp_switchdev_hwdom_set(struct net_bridge_port *joining)
176 {
177 	struct net_bridge *br = joining->br;
178 	struct net_bridge_port *p;
179 	int hwdom;
180 
181 	/* joining is yet to be added to the port list. */
182 	list_for_each_entry(p, &br->port_list, list) {
183 		if (netdev_phys_item_id_same(&joining->ppid, &p->ppid)) {
184 			joining->hwdom = p->hwdom;
185 			return 0;
186 		}
187 	}
188 
189 	hwdom = find_next_zero_bit(&br->busy_hwdoms, BR_HWDOM_MAX, 1);
190 	if (hwdom >= BR_HWDOM_MAX)
191 		return -EBUSY;
192 
193 	set_bit(hwdom, &br->busy_hwdoms);
194 	joining->hwdom = hwdom;
195 	return 0;
196 }
197 
198 static void nbp_switchdev_hwdom_put(struct net_bridge_port *leaving)
199 {
200 	struct net_bridge *br = leaving->br;
201 	struct net_bridge_port *p;
202 
203 	/* leaving is no longer in the port list. */
204 	list_for_each_entry(p, &br->port_list, list) {
205 		if (p->hwdom == leaving->hwdom)
206 			return;
207 	}
208 
209 	clear_bit(leaving->hwdom, &br->busy_hwdoms);
210 }
211 
212 static int nbp_switchdev_add(struct net_bridge_port *p,
213 			     struct netdev_phys_item_id ppid,
214 			     bool tx_fwd_offload,
215 			     struct netlink_ext_ack *extack)
216 {
217 	int err;
218 
219 	if (p->offload_count) {
220 		/* Prevent unsupported configurations such as a bridge port
221 		 * which is a bonding interface, and the member ports are from
222 		 * different hardware switches.
223 		 */
224 		if (!netdev_phys_item_id_same(&p->ppid, &ppid)) {
225 			NL_SET_ERR_MSG_MOD(extack,
226 					   "Same bridge port cannot be offloaded by two physical switches");
227 			return -EBUSY;
228 		}
229 
230 		/* Tolerate drivers that call switchdev_bridge_port_offload()
231 		 * more than once for the same bridge port, such as when the
232 		 * bridge port is an offloaded bonding/team interface.
233 		 */
234 		p->offload_count++;
235 
236 		return 0;
237 	}
238 
239 	p->ppid = ppid;
240 	p->offload_count = 1;
241 
242 	err = nbp_switchdev_hwdom_set(p);
243 	if (err)
244 		return err;
245 
246 	if (tx_fwd_offload) {
247 		p->flags |= BR_TX_FWD_OFFLOAD;
248 		static_branch_inc(&br_switchdev_tx_fwd_offload);
249 	}
250 
251 	return 0;
252 }
253 
254 static void nbp_switchdev_del(struct net_bridge_port *p)
255 {
256 	if (WARN_ON(!p->offload_count))
257 		return;
258 
259 	p->offload_count--;
260 
261 	if (p->offload_count)
262 		return;
263 
264 	if (p->hwdom)
265 		nbp_switchdev_hwdom_put(p);
266 
267 	if (p->flags & BR_TX_FWD_OFFLOAD) {
268 		p->flags &= ~BR_TX_FWD_OFFLOAD;
269 		static_branch_dec(&br_switchdev_tx_fwd_offload);
270 	}
271 }
272 
273 static int nbp_switchdev_sync_objs(struct net_bridge_port *p, const void *ctx,
274 				   struct notifier_block *atomic_nb,
275 				   struct notifier_block *blocking_nb,
276 				   struct netlink_ext_ack *extack)
277 {
278 	struct net_device *br_dev = p->br->dev;
279 	struct net_device *dev = p->dev;
280 	int err;
281 
282 	err = br_vlan_replay(br_dev, dev, ctx, true, blocking_nb, extack);
283 	if (err && err != -EOPNOTSUPP)
284 		return err;
285 
286 	err = br_mdb_replay(br_dev, dev, ctx, true, blocking_nb, extack);
287 	if (err && err != -EOPNOTSUPP)
288 		return err;
289 
290 	err = br_fdb_replay(br_dev, ctx, true, atomic_nb);
291 	if (err && err != -EOPNOTSUPP)
292 		return err;
293 
294 	return 0;
295 }
296 
297 static void nbp_switchdev_unsync_objs(struct net_bridge_port *p,
298 				      const void *ctx,
299 				      struct notifier_block *atomic_nb,
300 				      struct notifier_block *blocking_nb)
301 {
302 	struct net_device *br_dev = p->br->dev;
303 	struct net_device *dev = p->dev;
304 
305 	br_vlan_replay(br_dev, dev, ctx, false, blocking_nb, NULL);
306 
307 	br_mdb_replay(br_dev, dev, ctx, false, blocking_nb, NULL);
308 
309 	br_fdb_replay(br_dev, ctx, false, atomic_nb);
310 }
311 
312 /* Let the bridge know that this port is offloaded, so that it can assign a
313  * switchdev hardware domain to it.
314  */
315 int br_switchdev_port_offload(struct net_bridge_port *p,
316 			      struct net_device *dev, const void *ctx,
317 			      struct notifier_block *atomic_nb,
318 			      struct notifier_block *blocking_nb,
319 			      bool tx_fwd_offload,
320 			      struct netlink_ext_ack *extack)
321 {
322 	struct netdev_phys_item_id ppid;
323 	int err;
324 
325 	err = dev_get_port_parent_id(dev, &ppid, false);
326 	if (err)
327 		return err;
328 
329 	err = nbp_switchdev_add(p, ppid, tx_fwd_offload, extack);
330 	if (err)
331 		return err;
332 
333 	err = nbp_switchdev_sync_objs(p, ctx, atomic_nb, blocking_nb, extack);
334 	if (err)
335 		goto out_switchdev_del;
336 
337 	return 0;
338 
339 out_switchdev_del:
340 	nbp_switchdev_del(p);
341 
342 	return err;
343 }
344 
345 void br_switchdev_port_unoffload(struct net_bridge_port *p, const void *ctx,
346 				 struct notifier_block *atomic_nb,
347 				 struct notifier_block *blocking_nb)
348 {
349 	nbp_switchdev_unsync_objs(p, ctx, atomic_nb, blocking_nb);
350 
351 	nbp_switchdev_del(p);
352 }
353