Merge tag 'vfio-v4.11-rc1' of git://github.com/awilliam/linux-vfio
[linux-2.6-block.git] / net / bridge / br_if.c
1 /*
2  *      Userspace interface
3  *      Linux ethernet bridge
4  *
5  *      Authors:
6  *      Lennert Buytenhek               <buytenh@gnu.org>
7  *
8  *      This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  */
13
14 #include <linux/kernel.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
17 #include <linux/netpoll.h>
18 #include <linux/ethtool.h>
19 #include <linux/if_arp.h>
20 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <linux/rtnetlink.h>
23 #include <linux/if_ether.h>
24 #include <linux/slab.h>
25 #include <net/sock.h>
26 #include <linux/if_vlan.h>
27 #include <net/switchdev.h>
28
29 #include "br_private.h"
30
31 /*
32  * Determine initial path cost based on speed.
33  * using recommendations from 802.1d standard
34  *
35  * Since driver might sleep need to not be holding any locks.
36  */
37 static int port_cost(struct net_device *dev)
38 {
39         struct ethtool_link_ksettings ecmd;
40
41         if (!__ethtool_get_link_ksettings(dev, &ecmd)) {
42                 switch (ecmd.base.speed) {
43                 case SPEED_10000:
44                         return 2;
45                 case SPEED_1000:
46                         return 4;
47                 case SPEED_100:
48                         return 19;
49                 case SPEED_10:
50                         return 100;
51                 }
52         }
53
54         /* Old silly heuristics based on name */
55         if (!strncmp(dev->name, "lec", 3))
56                 return 7;
57
58         if (!strncmp(dev->name, "plip", 4))
59                 return 2500;
60
61         return 100;     /* assume old 10Mbps */
62 }
63
64
65 /* Check for port carrier transitions. */
66 void br_port_carrier_check(struct net_bridge_port *p)
67 {
68         struct net_device *dev = p->dev;
69         struct net_bridge *br = p->br;
70
71         if (!(p->flags & BR_ADMIN_COST) &&
72             netif_running(dev) && netif_oper_up(dev))
73                 p->path_cost = port_cost(dev);
74
75         if (!netif_running(br->dev))
76                 return;
77
78         spin_lock_bh(&br->lock);
79         if (netif_running(dev) && netif_oper_up(dev)) {
80                 if (p->state == BR_STATE_DISABLED)
81                         br_stp_enable_port(p);
82         } else {
83                 if (p->state != BR_STATE_DISABLED)
84                         br_stp_disable_port(p);
85         }
86         spin_unlock_bh(&br->lock);
87 }
88
89 static void br_port_set_promisc(struct net_bridge_port *p)
90 {
91         int err = 0;
92
93         if (br_promisc_port(p))
94                 return;
95
96         err = dev_set_promiscuity(p->dev, 1);
97         if (err)
98                 return;
99
100         br_fdb_unsync_static(p->br, p);
101         p->flags |= BR_PROMISC;
102 }
103
104 static void br_port_clear_promisc(struct net_bridge_port *p)
105 {
106         int err;
107
108         /* Check if the port is already non-promisc or if it doesn't
109          * support UNICAST filtering.  Without unicast filtering support
110          * we'll end up re-enabling promisc mode anyway, so just check for
111          * it here.
112          */
113         if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT))
114                 return;
115
116         /* Since we'll be clearing the promisc mode, program the port
117          * first so that we don't have interruption in traffic.
118          */
119         err = br_fdb_sync_static(p->br, p);
120         if (err)
121                 return;
122
123         dev_set_promiscuity(p->dev, -1);
124         p->flags &= ~BR_PROMISC;
125 }
126
127 /* When a port is added or removed or when certain port flags
128  * change, this function is called to automatically manage
129  * promiscuity setting of all the bridge ports.  We are always called
130  * under RTNL so can skip using rcu primitives.
131  */
132 void br_manage_promisc(struct net_bridge *br)
133 {
134         struct net_bridge_port *p;
135         bool set_all = false;
136
137         /* If vlan filtering is disabled or bridge interface is placed
138          * into promiscuous mode, place all ports in promiscuous mode.
139          */
140         if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br))
141                 set_all = true;
142
143         list_for_each_entry(p, &br->port_list, list) {
144                 if (set_all) {
145                         br_port_set_promisc(p);
146                 } else {
147                         /* If the number of auto-ports is <= 1, then all other
148                          * ports will have their output configuration
149                          * statically specified through fdbs.  Since ingress
150                          * on the auto-port becomes forwarding/egress to other
151                          * ports and egress configuration is statically known,
152                          * we can say that ingress configuration of the
153                          * auto-port is also statically known.
154                          * This lets us disable promiscuous mode and write
155                          * this config to hw.
156                          */
157                         if (br->auto_cnt == 0 ||
158                             (br->auto_cnt == 1 && br_auto_port(p)))
159                                 br_port_clear_promisc(p);
160                         else
161                                 br_port_set_promisc(p);
162                 }
163         }
164 }
165
166 static void nbp_update_port_count(struct net_bridge *br)
167 {
168         struct net_bridge_port *p;
169         u32 cnt = 0;
170
171         list_for_each_entry(p, &br->port_list, list) {
172                 if (br_auto_port(p))
173                         cnt++;
174         }
175         if (br->auto_cnt != cnt) {
176                 br->auto_cnt = cnt;
177                 br_manage_promisc(br);
178         }
179 }
180
181 static void nbp_delete_promisc(struct net_bridge_port *p)
182 {
183         /* If port is currently promiscuous, unset promiscuity.
184          * Otherwise, it is a static port so remove all addresses
185          * from it.
186          */
187         dev_set_allmulti(p->dev, -1);
188         if (br_promisc_port(p))
189                 dev_set_promiscuity(p->dev, -1);
190         else
191                 br_fdb_unsync_static(p->br, p);
192 }
193
194 static void release_nbp(struct kobject *kobj)
195 {
196         struct net_bridge_port *p
197                 = container_of(kobj, struct net_bridge_port, kobj);
198         kfree(p);
199 }
200
201 static struct kobj_type brport_ktype = {
202 #ifdef CONFIG_SYSFS
203         .sysfs_ops = &brport_sysfs_ops,
204 #endif
205         .release = release_nbp,
206 };
207
208 static void destroy_nbp(struct net_bridge_port *p)
209 {
210         struct net_device *dev = p->dev;
211
212         p->br = NULL;
213         p->dev = NULL;
214         dev_put(dev);
215
216         kobject_put(&p->kobj);
217 }
218
219 static void destroy_nbp_rcu(struct rcu_head *head)
220 {
221         struct net_bridge_port *p =
222                         container_of(head, struct net_bridge_port, rcu);
223         destroy_nbp(p);
224 }
225
226 static unsigned get_max_headroom(struct net_bridge *br)
227 {
228         unsigned max_headroom = 0;
229         struct net_bridge_port *p;
230
231         list_for_each_entry(p, &br->port_list, list) {
232                 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev);
233
234                 if (dev_headroom > max_headroom)
235                         max_headroom = dev_headroom;
236         }
237
238         return max_headroom;
239 }
240
241 static void update_headroom(struct net_bridge *br, int new_hr)
242 {
243         struct net_bridge_port *p;
244
245         list_for_each_entry(p, &br->port_list, list)
246                 netdev_set_rx_headroom(p->dev, new_hr);
247
248         br->dev->needed_headroom = new_hr;
249 }
250
251 /* Delete port(interface) from bridge is done in two steps.
252  * via RCU. First step, marks device as down. That deletes
253  * all the timers and stops new packets from flowing through.
254  *
255  * Final cleanup doesn't occur until after all CPU's finished
256  * processing packets.
257  *
258  * Protected from multiple admin operations by RTNL mutex
259  */
260 static void del_nbp(struct net_bridge_port *p)
261 {
262         struct net_bridge *br = p->br;
263         struct net_device *dev = p->dev;
264
265         sysfs_remove_link(br->ifobj, p->dev->name);
266
267         nbp_delete_promisc(p);
268
269         spin_lock_bh(&br->lock);
270         br_stp_disable_port(p);
271         spin_unlock_bh(&br->lock);
272
273         br_ifinfo_notify(RTM_DELLINK, p);
274
275         list_del_rcu(&p->list);
276         if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom)
277                 update_headroom(br, get_max_headroom(br));
278         netdev_reset_rx_headroom(dev);
279
280         nbp_vlan_flush(p);
281         br_fdb_delete_by_port(br, p, 0, 1);
282         switchdev_deferred_process();
283
284         nbp_update_port_count(br);
285
286         netdev_upper_dev_unlink(dev, br->dev);
287
288         dev->priv_flags &= ~IFF_BRIDGE_PORT;
289
290         netdev_rx_handler_unregister(dev);
291
292         br_multicast_del_port(p);
293
294         kobject_uevent(&p->kobj, KOBJ_REMOVE);
295         kobject_del(&p->kobj);
296
297         br_netpoll_disable(p);
298
299         call_rcu(&p->rcu, destroy_nbp_rcu);
300 }
301
302 /* Delete bridge device */
303 void br_dev_delete(struct net_device *dev, struct list_head *head)
304 {
305         struct net_bridge *br = netdev_priv(dev);
306         struct net_bridge_port *p, *n;
307
308         list_for_each_entry_safe(p, n, &br->port_list, list) {
309                 del_nbp(p);
310         }
311
312         br_fdb_delete_by_port(br, NULL, 0, 1);
313
314         br_vlan_flush(br);
315         br_multicast_dev_del(br);
316         cancel_delayed_work_sync(&br->gc_work);
317
318         br_sysfs_delbr(br->dev);
319         unregister_netdevice_queue(br->dev, head);
320 }
321
322 /* find an available port number */
323 static int find_portno(struct net_bridge *br)
324 {
325         int index;
326         struct net_bridge_port *p;
327         unsigned long *inuse;
328
329         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
330                         GFP_KERNEL);
331         if (!inuse)
332                 return -ENOMEM;
333
334         set_bit(0, inuse);      /* zero is reserved */
335         list_for_each_entry(p, &br->port_list, list) {
336                 set_bit(p->port_no, inuse);
337         }
338         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
339         kfree(inuse);
340
341         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
342 }
343
344 /* called with RTNL but without bridge lock */
345 static struct net_bridge_port *new_nbp(struct net_bridge *br,
346                                        struct net_device *dev)
347 {
348         struct net_bridge_port *p;
349         int index, err;
350
351         index = find_portno(br);
352         if (index < 0)
353                 return ERR_PTR(index);
354
355         p = kzalloc(sizeof(*p), GFP_KERNEL);
356         if (p == NULL)
357                 return ERR_PTR(-ENOMEM);
358
359         p->br = br;
360         dev_hold(dev);
361         p->dev = dev;
362         p->path_cost = port_cost(dev);
363         p->priority = 0x8000 >> BR_PORT_BITS;
364         p->port_no = index;
365         p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD;
366         br_init_port(p);
367         br_set_state(p, BR_STATE_DISABLED);
368         br_stp_port_timer_init(p);
369         err = br_multicast_add_port(p);
370         if (err) {
371                 dev_put(dev);
372                 kfree(p);
373                 p = ERR_PTR(err);
374         }
375
376         return p;
377 }
378
379 int br_add_bridge(struct net *net, const char *name)
380 {
381         struct net_device *dev;
382         int res;
383
384         dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN,
385                            br_dev_setup);
386
387         if (!dev)
388                 return -ENOMEM;
389
390         dev_net_set(dev, net);
391         dev->rtnl_link_ops = &br_link_ops;
392
393         res = register_netdev(dev);
394         if (res)
395                 free_netdev(dev);
396         return res;
397 }
398
399 int br_del_bridge(struct net *net, const char *name)
400 {
401         struct net_device *dev;
402         int ret = 0;
403
404         rtnl_lock();
405         dev = __dev_get_by_name(net, name);
406         if (dev == NULL)
407                 ret =  -ENXIO;  /* Could not find device */
408
409         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
410                 /* Attempt to delete non bridge device! */
411                 ret = -EPERM;
412         }
413
414         else if (dev->flags & IFF_UP) {
415                 /* Not shutdown yet. */
416                 ret = -EBUSY;
417         }
418
419         else
420                 br_dev_delete(dev, NULL);
421
422         rtnl_unlock();
423         return ret;
424 }
425
426 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
427 int br_min_mtu(const struct net_bridge *br)
428 {
429         const struct net_bridge_port *p;
430         int mtu = 0;
431
432         ASSERT_RTNL();
433
434         if (list_empty(&br->port_list))
435                 mtu = ETH_DATA_LEN;
436         else {
437                 list_for_each_entry(p, &br->port_list, list) {
438                         if (!mtu  || p->dev->mtu < mtu)
439                                 mtu = p->dev->mtu;
440                 }
441         }
442         return mtu;
443 }
444
445 static void br_set_gso_limits(struct net_bridge *br)
446 {
447         unsigned int gso_max_size = GSO_MAX_SIZE;
448         u16 gso_max_segs = GSO_MAX_SEGS;
449         const struct net_bridge_port *p;
450
451         list_for_each_entry(p, &br->port_list, list) {
452                 gso_max_size = min(gso_max_size, p->dev->gso_max_size);
453                 gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs);
454         }
455         br->dev->gso_max_size = gso_max_size;
456         br->dev->gso_max_segs = gso_max_segs;
457 }
458
459 /*
460  * Recomputes features using slave's features
461  */
462 netdev_features_t br_features_recompute(struct net_bridge *br,
463         netdev_features_t features)
464 {
465         struct net_bridge_port *p;
466         netdev_features_t mask;
467
468         if (list_empty(&br->port_list))
469                 return features;
470
471         mask = features;
472         features &= ~NETIF_F_ONE_FOR_ALL;
473
474         list_for_each_entry(p, &br->port_list, list) {
475                 features = netdev_increment_features(features,
476                                                      p->dev->features, mask);
477         }
478         features = netdev_add_tso_features(features, mask);
479
480         return features;
481 }
482
483 /* called with RTNL */
484 int br_add_if(struct net_bridge *br, struct net_device *dev)
485 {
486         struct net_bridge_port *p;
487         int err = 0;
488         unsigned br_hr, dev_hr;
489         bool changed_addr;
490
491         /* Don't allow bridging non-ethernet like devices, or DSA-enabled
492          * master network devices since the bridge layer rx_handler prevents
493          * the DSA fake ethertype handler to be invoked, so we do not strip off
494          * the DSA switch tag protocol header and the bridge layer just return
495          * RX_HANDLER_CONSUMED, stopping RX processing for these frames.
496          */
497         if ((dev->flags & IFF_LOOPBACK) ||
498             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
499             !is_valid_ether_addr(dev->dev_addr) ||
500             netdev_uses_dsa(dev))
501                 return -EINVAL;
502
503         /* No bridging of bridges */
504         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
505                 return -ELOOP;
506
507         /* Device is already being bridged */
508         if (br_port_exists(dev))
509                 return -EBUSY;
510
511         /* No bridging devices that dislike that (e.g. wireless) */
512         if (dev->priv_flags & IFF_DONT_BRIDGE)
513                 return -EOPNOTSUPP;
514
515         p = new_nbp(br, dev);
516         if (IS_ERR(p))
517                 return PTR_ERR(p);
518
519         call_netdevice_notifiers(NETDEV_JOIN, dev);
520
521         err = dev_set_allmulti(dev, 1);
522         if (err)
523                 goto put_back;
524
525         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
526                                    SYSFS_BRIDGE_PORT_ATTR);
527         if (err)
528                 goto err1;
529
530         err = br_sysfs_addif(p);
531         if (err)
532                 goto err2;
533
534         err = br_netpoll_enable(p);
535         if (err)
536                 goto err3;
537
538         err = netdev_rx_handler_register(dev, br_handle_frame, p);
539         if (err)
540                 goto err4;
541
542         dev->priv_flags |= IFF_BRIDGE_PORT;
543
544         err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL);
545         if (err)
546                 goto err5;
547
548         err = nbp_switchdev_mark_set(p);
549         if (err)
550                 goto err6;
551
552         dev_disable_lro(dev);
553
554         list_add_rcu(&p->list, &br->port_list);
555
556         nbp_update_port_count(br);
557
558         netdev_update_features(br->dev);
559
560         br_hr = br->dev->needed_headroom;
561         dev_hr = netdev_get_fwd_headroom(dev);
562         if (br_hr < dev_hr)
563                 update_headroom(br, dev_hr);
564         else
565                 netdev_set_rx_headroom(dev, br_hr);
566
567         if (br_fdb_insert(br, p, dev->dev_addr, 0))
568                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
569
570         err = nbp_vlan_init(p);
571         if (err) {
572                 netdev_err(dev, "failed to initialize vlan filtering on this port\n");
573                 goto err7;
574         }
575
576         spin_lock_bh(&br->lock);
577         changed_addr = br_stp_recalculate_bridge_id(br);
578
579         if (netif_running(dev) && netif_oper_up(dev) &&
580             (br->dev->flags & IFF_UP))
581                 br_stp_enable_port(p);
582         spin_unlock_bh(&br->lock);
583
584         br_ifinfo_notify(RTM_NEWLINK, p);
585
586         if (changed_addr)
587                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
588
589         dev_set_mtu(br->dev, br_min_mtu(br));
590         br_set_gso_limits(br);
591
592         kobject_uevent(&p->kobj, KOBJ_ADD);
593
594         return 0;
595
596 err7:
597         list_del_rcu(&p->list);
598         br_fdb_delete_by_port(br, p, 0, 1);
599         nbp_update_port_count(br);
600 err6:
601         netdev_upper_dev_unlink(dev, br->dev);
602 err5:
603         dev->priv_flags &= ~IFF_BRIDGE_PORT;
604         netdev_rx_handler_unregister(dev);
605 err4:
606         br_netpoll_disable(p);
607 err3:
608         sysfs_remove_link(br->ifobj, p->dev->name);
609 err2:
610         kobject_put(&p->kobj);
611         p = NULL; /* kobject_put frees */
612 err1:
613         dev_set_allmulti(dev, -1);
614 put_back:
615         dev_put(dev);
616         kfree(p);
617         return err;
618 }
619
620 /* called with RTNL */
621 int br_del_if(struct net_bridge *br, struct net_device *dev)
622 {
623         struct net_bridge_port *p;
624         bool changed_addr;
625
626         p = br_port_get_rtnl(dev);
627         if (!p || p->br != br)
628                 return -EINVAL;
629
630         /* Since more than one interface can be attached to a bridge,
631          * there still maybe an alternate path for netconsole to use;
632          * therefore there is no reason for a NETDEV_RELEASE event.
633          */
634         del_nbp(p);
635
636         dev_set_mtu(br->dev, br_min_mtu(br));
637         br_set_gso_limits(br);
638
639         spin_lock_bh(&br->lock);
640         changed_addr = br_stp_recalculate_bridge_id(br);
641         spin_unlock_bh(&br->lock);
642
643         if (changed_addr)
644                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
645
646         netdev_update_features(br->dev);
647
648         return 0;
649 }
650
651 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
652 {
653         struct net_bridge *br = p->br;
654
655         if (mask & BR_AUTO_MASK)
656                 nbp_update_port_count(br);
657 }